The Penrose Singularity Theorem

The Penrose theorem converts a quasi-local sign condition—a compact surface from which both future null families converge—into a global failure of future null completeness. The decisive contradiction is topological. If every future null geodesic were complete, focusing would make the future horismos of the trapped surface compact; global hyperbolicity would make it an achronal hypersurface; and projecting it onto a noncompact Cauchy surface would force that surface to be compact.

Required background. Congruences and Optical Scalars supplies null expansion and focusing; Einstein Equations relates matter assumptions to null convergence. Helpful background. Lorentzian Metrics and Causal Cones fixes future null directions; Geodesics and Jacobi Fields supplies affine completeness and focal points; Causality and Global Structure supplies Cauchy surfaces and achronal boundaries; Curvature fixes Ricci contraction.

Precise smooth theorem

Let $(M,g)$ be a connected, time-oriented, smooth spacetime of dimension $n\geq3$ and without boundary. Assume:

1. Global causality and topology: $M$ has a noncompact Cauchy hypersurface $\Sigma$. In particular, $M$ is globally hyperbolic and chronological.
2. Curvature: the null convergence condition holds, $$R_{ab}k^ak^b\geq0$$ for every null vector $k$.
3. Initial geometry: $M$ contains a closed, achronal, future trapped surface $P$: a compact spacelike codimension-two submanifold without boundary whose two future null expansions are strictly negative everywhere.

Then $(M,g)$ is future null geodesically incomplete. At least one future-inextendible null geodesic has finite affine length.

This modern classical form and its proof architecture are stated in Steinbauer 2023, Theorem 3.9, pp. 16–18. Penrose’s original four-dimensional statement and compactness argument appear in Penrose 1965, pp. 57–59. Some refinements weaken or reorganize achronality and regularity assumptions; the proof here keeps the displayed form fixed.

“Closed” means compact without boundary, not merely topologically closed in $M$. “Future trapped” is a strict two-sided statement: both future null expansions are negative. The normalization of the two null normals changes the numerical expansions by positive factors but not their signs. No generic curvature condition is required because strict negative expansion is already supplied on $P$.

Assume future null completeness

The proof is indirect. Suppose every future-inextendible affinely parametrized null geodesic is defined on $[0,\infty)$. Choose a smooth future timelike vector field $T$ and normalize every future null normal $k$ to $P$ by

$$g(k,T)=-1.$$

There are two such directions at each point of $P$. Because $P$ is compact, their normalized bundle $K$ is compact. The two expansion functions are continuous and strictly negative, so a uniform constant $c>0$ exists with

$$\theta_k(0)\leq-c$$

for every $k\in K$.

Along each orthogonal null geodesic, the congruence is twist-free before a focal point. Null convergence and the Raychaudhuri equation give

$$\frac{d\theta}{d\lambda} \leq-\frac{\theta^2}{n-2}.$$

Hence a focal point to $P$ occurs by affine parameter at most

$$T_*=\frac{n-2}{c},$$

provided the generator exists that long. The assumed future null completeness supplies precisely that existence. The analytic step is the focusing lemma proved on Focusing and Conjugate-Point Lemmas.

Compactness of the future horismos

Define the future horismos

$$E^+(P)=J^+(P)\setminus I^+(P).$$

In a globally hyperbolic spacetime, $J^+(P)$ is closed for compact $P$, and

$$E^+(P)=\partial J^+(P)=\partial I^+(P).$$

It is an achronal boundary. Every point $q\in E^+(P)\setminus P$ is reached by a future null geodesic leaving $P$ orthogonally, with no focal point to $P$ before $q$. If the geodesic were not orthogonal, or if a focal point occurred first, a variation would produce a timelike curve from $P$ to $q$, placing $q$ in $I^+(P)$ rather than $E^+(P)$.

The uniform focal bound therefore confines every boundary generator to $0\leq\lambda\leq T_*$. Consequently,

$$E^+(P)\subset \exp^\perp\!\bigl(K\times[0,T_*]\bigr).$$

The set on the right is compact as the continuous image of a compact set. Since $E^+(P)$ is closed, it too is compact. This is the exact role of assumed completeness: without it, some normal generator can terminate at finite affine parameter before the compactness construction finishes, which is already the theorem’s conclusion.

Focusing has not shown that the null geodesics end at their focal points. They can continue into $I^+(P)$. What becomes compact is the portion that can still generate the achronal boundary. Confusing those two statements is the most common false shortcut in the proof.

Projection to the Cauchy surface

Rescale $T$ by a positive function so that its integral curves are complete. Each maximal integral curve is an inextendible timelike curve and therefore intersects the Cauchy surface $\Sigma$ exactly once. Define

$$\rho:M\longrightarrow\Sigma$$

by following the $T$-orbit through a point to that intersection.

The restriction $\rho\rvert_{E^+(P)}$ is injective: two points of the achronal set $E^+(P)$ cannot lie on the same timelike orbit. It is continuous. Because both $E^+(P)$ and $\Sigma$ are topological hypersurfaces of dimension $n-1$, invariance of domain makes its image open in $\Sigma$. Compactness of $E^+(P)$ makes the image compact, hence closed in the Hausdorff manifold $\Sigma$.

The image is nonempty. Since $M$ is connected, its Cauchy surface is connected; an open-and-closed nonempty subset must be all of $\Sigma$. Thus

$$\rho(E^+(P))=\Sigma.$$

But the continuous image of compact $E^+(P)$ is compact, so $\Sigma$ would be compact. This contradicts the hypothesis that $\Sigma$ is noncompact. Future null completeness must therefore fail. This projection and invariance-of-domain step—not Raychaudhuri alone—is where the noncompact-Cauchy hypothesis enters Penrose 1965, pp. 58–59.

Necessity of the hypotheses

Each assumption blocks a recognizable escape.

Strict trappedness starts focusing. A marginal surface with one zero expansion does not give the same uniform finite bound without an additional genericity or stability input.

Null convergence sustains focusing. If $R_{ab}k^ak^b<0$ strongly enough, the Ricci term can defocus the bundle. The null energy condition implies null convergence in Einstein gravity because the trace and cosmological terms vanish when contracted with $k^ak^b$, but the theorem itself is geometric.

Global hyperbolicity controls the boundary. It supplies closed causal futures, maximizing null generators, and a Cauchy projection. Chronology alone is insufficient for this proof.

Noncompactness creates the topological contradiction. Global de Sitter spacetime has compact Cauchy surfaces and is null complete. In its contracting half, suitable compact spheres can be future trapped while $R_{ab}k^ak^b=0$. There is no contradiction because the projected compact achronal boundary can cover a compact $S^3$ slice. Thus noncompactness is not cosmetic shorthand for an “isolated system”; it is a used theorem hypothesis.

Schwarzschild interpretation and limits

In the maximally extended Schwarzschild spacetime, a round sphere with $r<2GM$ in the black-hole region is future trapped, the vacuum equations give $R_{ab}k^ak^b=0$, and a Cauchy surface has noncompact topology. The theorem diagnoses future null incompleteness. It does not use the event horizon as a hypothesis, and it does not prove that the incomplete generator ends at the $r=0$ curvature singularity. Those stronger identifications come from analyzing the exact solution.

More generally, the theorem proves neither event-horizon formation nor visibility properties. A trapped surface is quasi-local, whereas an event horizon refers to future infinity. The theorem also does not prove curvature blowup, $C^0$-inextendibility, timelike incompleteness, or incompleteness of every null geodesic. Its strict conclusion is the existence of at least one future-incomplete null geodesic, as defined on Timelike, Null, and Geodesic Completeness.

The classical proof uses pointwise curvature, unique null geodesics, focal points, and the normal exponential map. Low-regularity Penrose theorems at $C^{1,1}$ and $C^1$ replace these tools with regularization and distributional convergence conditions. They have additional technical hypotheses and are presented separately in Steinbauer 2023, §§4.5–4.6, pp. 27–34.

Exercise

Suppose every hypothesis of the theorem holds except that the Cauchy surface is compact. Follow the proof under assumed null completeness and identify the strongest conclusion that still follows and the exact step that no longer contradicts anything.

Solution

Strict trappedness, compactness of $P$, null convergence, and null completeness still give a uniform focal bound. Global hyperbolicity still makes $E^+(P)$ a closed achronal boundary generated by focal-point-free null segments, so $E^+(P)$ is compact. Projection along a complete timelike flow still makes its image in the connected Cauchy surface open and closed, hence all of $\Sigma$.

The final statement is then that $\Sigma$ is the continuous image of compact $E^+(P)$ and is therefore compact. That agrees with the modified hypothesis rather than contradicting it. Thus the proof yields a compact future horismos and a surjective projection, but it yields no incompleteness. Global de Sitter spacetime shows that this is a genuine limitation, not merely a gap in the argument.

References

• Penrose, Roger, “Gravitational Collapse and Space-Time Singularities”, Physical Review Letters 14, pp. 57–59, 1965.
• Steinbauer, Roland, “The Singularity Theorems of General Relativity and Their Low Regularity Extensions”, Jahresbericht der Deutschen Mathematiker-Vereinigung 125, Theorem 3.9 and proof, pp. 16–18, 2023.