The Hawking-Penrose Theorem

The Hawking–Penrose theorem removes the Cauchy-surface assumptions of the separate Hawking and Penrose results. In exchange, it imposes a generic tidal-curvature condition on every causal geodesic and uses a deeper causal lemma. Its conclusion is deliberately disjunctive: the spacetime is timelike or null geodesically incomplete. The classical statement does not generally select which causal type or which time direction fails.

Required background. Congruences and Optical Scalars supplies causal focusing; Einstein Equations relates matter assumptions to Ricci convergence. Helpful background. Lorentzian Metrics and Causal Cones fixes chronology and causal type; Geodesics and Jacobi Fields supplies completeness and conjugate points; Causality and Global Structure supplies trapped sets and limit curves; Curvature supplies the transverse tidal operator.

Precise classical statement

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

1. Chronology: $M$ contains no closed timelike curve.
2. Causal convergence: $$R_{ab}v^av^b\geq0$$ for every timelike vector $v$, and hence by continuity for every causal vector. This is the geometric timelike-convergence or strong-energy input.
3. Genericity: along every inextendible timelike or null geodesic $\gamma$, there is a point at which the transverse tidal operator $$[X]\longmapsto [R(X,\dot\gamma)\dot\gamma]$$ is not identically zero on $\dot\gamma^\perp$ in the timelike case or on $\dot\gamma^\perp/\operatorname{span}\{\dot\gamma\}$ in the null case.
4. One initial alternative: $M$ contains at least one of the following:
   • a compact achronal set without edge;
   • a closed future- or past-trapped codimension-two surface;
   • a trapped point, meaning that every null geodesic in one chosen time direction from the point eventually acquires negative expansion.

Then $(M,g)$ is causal geodesically incomplete: at least one inextendible timelike or null geodesic has a finite affine or proper-parameter endpoint.

This formulation isolates the three alternatives in the four-dimensional classical result. Hawking and Penrose state the causal lemma and derive the theorem in Hawking and Penrose 1970, pp. 538–544. A modern $n$-dimensional presentation, with an additional higher-codimension trapped-submanifold alternative and its curvature trace condition, is Steinbauer 2023, Theorem 3.11, pp. 18–20. The extra alternative is a later extension, not silently part of the original theorem.

Chronology is much weaker than global hyperbolicity: no Cauchy surface, compact-diamond condition, or asymptotic region is assumed. Genericity is correspondingly stronger than the assumptions of the preceding theorems. Penrose starts with strictly negative null expansions; Hawking starts with uniform timelike convergence of a slice. Hawking–Penrose requires every complete causal geodesic to encounter genuine transverse tidal curvature.

The causal Hawking–Penrose lemma

The global core can be stated without curvature. The following three properties cannot all hold in one spacetime:

1. the spacetime is chronological;
2. every inextendible causal geodesic contains a pair of conjugate points;
3. the spacetime contains a trapped set $A$, meaning a closed achronal set whose future or past horismos is compact.

This is the Hawking–Penrose causal lemma. Its full proof is a limit-curve and maximizing argument, not a one-line consequence of Raychaudhuri. Compact horismos confines the relevant causal boundary. Chronology prevents a timelike recurrence that would dispose of the boundary by a closed timelike loop. Limit curves and maximizing segments then produce an inextendible causal geodesic that must remain free of a conjugate pair, contradicting property 2. The precise statement and causal construction occupy the theorem and lemmas on Hawking and Penrose 1970, pp. 531–544.

The lemma explains how global hyperbolicity can be dropped. The proof no longer projects a compact horismos onto a Cauchy surface, as Penrose does, nor maximizes distance from a Cauchy slice, as Hawking does. Instead, chronology, trapped-set compactness, and the universal conjugate-pair property are themselves incompatible. This gain costs the generic condition needed to make the conjugate-pair property universal.

Completeness plus genericity gives conjugate pairs

Assume for contradiction that every timelike and null geodesic is complete. Fix any inextendible causal geodesic $\gamma$. Ricci convergence makes the expansion of every twist-free Lagrange Jacobi family nonincreasing:

$$\dot\theta =-\frac{\theta^2}{d} -\operatorname{tr}(\sigma^2) -R_{ab}\dot\gamma^a\dot\gamma^b \leq0.$$

If $\theta$ becomes negative in the future, the elementary focusing lemma produces a future conjugate point. If it is positive, reverse the parameter and obtain a past conjugate point. The difficult case is a family whose expansion never departs from zero on a half-geodesic.

In that case Raychaudhuri forces the shear and Ricci contraction to vanish there. The full matrix Riccati equation

$$\dot B+B^2+\mathcal R=0$$

then forces the transverse tidal operator $\mathcal R[X]=[R(X,\dot\gamma)\dot\gamma]$ to vanish, contradicting genericity at the designated point. A careful choice of Jacobi families on both sides of that point yields two conjugate points along $\gamma$. The complete matrix argument, which is stronger than the negative-expansion estimate, is summarized in Steinbauer 2023, Proposition 3.12, pp. 19–20.

Thus convergence and genericity, under assumed causal completeness, establish property 2 of the causal lemma for every inextendible causal geodesic. Genericity alone is not a positivity condition and convergence alone permits a perfectly parallel bundle. Their conjunction excludes persistent tidal degeneracy.

Each initial alternative supplies a trapped set

The three initial alternatives feed property 3 in different ways.

Compact achronal edgeless set. Such a set is a compact topological hypersurface—a compact partial Cauchy surface. Its appropriate horismos is compact, so it directly provides the trapped-set structure required by the lemma. This is the cosmological alternative, but it does not assume that the set is a global Cauchy surface.

Closed trapped surface. Its strict null expansions and causal convergence make every complete orthogonal null generator focal within a bounded affine interval. Generators with focal points cannot remain on the horismos. Compact initial normal data then confine the horismos to a compact exponential image. Some additional causal bookkeeping is needed because the trapped surface itself need not be globally achronal; under the contradiction assumptions, chronology, convergence, genericity, and null completeness supply the required stronger causal control.

Trapped point. If every null ray from a point in one time direction eventually has negative expansion, compactness of the normalized celestial sphere and an appropriate uniformity argument again compactify the relevant horismos, unless a null ray is already incomplete.

In every branch, either incompleteness has already appeared or a trapped set exists. Under the assumption of complete causal geodesics, the latter must occur. The causal lemma now contradicts chronology plus the conjugate-pair property, completing the proof.

Why every hypothesis matters

The Einstein static universe is an instructive genericity check. It is chronological, geodesically complete, and has compact achronal Cauchy slices. Its Ricci tensor is nonnegative on timelike directions for the standard matter-supported solution, but comoving timelike geodesics can fail to encounter a nontrivial transverse tidal operator. The generic condition blocks this complete, finely balanced case.

If chronology is removed, a closed timelike curve can coexist with causal focusing while invalidating the causal lemma. The theorem does not prove chronology protection. If convergence is removed, negative Ricci contraction can counter shear and expansion. If all three initial alternatives are absent, generic focusing along individual complete geodesics need not produce a compact trapped causal boundary.

The “compact achronal set without edge” alternative is topological, not synonymous with a compact spacelike Cauchy surface. The trapped-surface alternative is quasi-local but still requires compactness. A marginally trapped surface with a zero expansion does not automatically satisfy the strict alternative. The trapped-point condition is also stronger than observing lensing or convergence along a few light rays: it quantifies over all null directions from the point.

Direction, causal type, and regularity limits

The theorem’s conclusion must remain disjunctive. It does not say “future null incomplete” merely because one displayed alternative is future trapped, nor does it say “past timelike incomplete” merely because the compact-set alternative is cosmological. Directional strengthenings exist under corresponding oriented hypotheses, but the classical combined statement above guarantees only that some causal geodesic—timelike or null, in at least one direction—is incomplete.

It also does not prove curvature blowup, a singular endpoint, an event horizon, or inextendibility in any regularity class. Genericity says that tidal curvature is nonzero somewhere on each causal geodesic, not that it diverges. The exact logical limits are collected on What Singularity Theorems Conclude.

At $C^{1,1}$ regularity, generalized Ricci and genericity conditions can be handled by regularization. At $C^1$, curvature is distributional, geodesics may branch, and modern Hawking–Penrose results impose a maximal causal nonbranching condition in addition to weak convergence and genericity formulations. These are separate results, not a change of font in the classical theorem; see Steinbauer 2023, §§4.7–4.10, pp. 33–40.

Exercise

For each change below, identify which part of the proof fails: (a) keep all assumptions but delete genericity; (b) keep genericity and convergence but delete the initial alternatives; (c) replace chronology by the statement that one neighborhood is causal.

Solution

(a) Completeness plus convergence need not force a conjugate pair on every causal geodesic. A bundle can remain expansion-free, shear-free, and tidally trivial. Property 2 of the causal lemma is therefore unavailable.

(b) The analytic argument may still give conjugate pairs under assumed completeness, but there is no reason for a compact trapped horismos to exist. Property 3 of the causal lemma is unavailable.

(c) Local causality does not exclude a closed timelike curve elsewhere in $M$. The causal lemma requires global chronology, so property 1 is unavailable. None of the three weakened sets of assumptions yields the Hawking–Penrose contradiction.

References

• Hawking, Stephen W., and Roger Penrose, “The Singularities of Gravitational Collapse and Cosmology”, Proceedings of the Royal Society A 314, pp. 529–548, 1970.
• Steinbauer, Roland, “The Singularity Theorems of General Relativity and Their Low Regularity Extensions”, Jahresbericht der Deutschen Mathematiker-Vereinigung 125, Theorem 3.11 and proof architecture, pp. 18–20, 2023.