Hawking Cosmological Singularity Theorems

Hawking’s cosmological theorem replaces a trapped codimension-two surface by expansion or contraction of an entire spacelike Cauchy slice. If a compact universe is uniformly expanding now and timelike Ricci curvature remains focusing, then past-directed normal geodesics acquire focal points within a uniform proper-time bound. Global hyperbolicity simultaneously provides focal-point-free maximizers from the slice to every earlier event. Past timelike completeness would make these two statements collide.

Required background. Congruences and Optical Scalars supplies timelike expansion and Raychaudhuri; Einstein Equations relates matter assumptions to timelike Ricci convergence. Helpful background. Lorentzian Metrics and Causal Cones fixes time orientation; Geodesics and Jacobi Fields supplies proper-time completeness and focal points; Causality and Global Structure supplies Cauchy developments; Curvature fixes Ricci contraction.

Past-directed cosmological theorem

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

1. Cauchy geometry: $M$ contains a compact smooth spacelike Cauchy hypersurface $S$.
2. Timelike convergence: $$R_{ab}v^av^b\geq0$$ for every timelike vector $v$.
3. Uniform present expansion: if $u$ is the future-directed unit normal to $S$, its expansion satisfies $$\Theta=\nabla_a u^a\geq\Theta_0>0$$ everywhere on $S$.

Then $(M,g)$ is past timelike geodesically incomplete: at least one past-inextendible timelike geodesic has finite proper time.

The time-reversed form says that if the future-normal congruence is uniformly converging, $\Theta\leq-\Theta_0<0$, then the spacetime is future timelike incomplete. Steinbauer states this contracting version with compact $S$ and positive “future convergence” in Steinbauer 2023, Theorem 3.8, pp. 15–16. Compactness is used to turn a continuous strict sign into a uniform positive minimum; one can instead formulate variants by assuming the uniform bound directly under suitable global hypotheses.

This separation matters in applications. Compactness of $S$ is a topological statement and does not itself imply contraction or expansion. Conversely, a noncompact slice may satisfy a uniform lower expansion bound, but one must use a theorem formulated for that setting rather than cite the compact-slice statement verbatim. In the version proved here, compactness also keeps the initial normal data globally controlled without adding falloff conditions at spatial infinity. It is not the same topological input as in the Penrose theorem: Hawking uses a compact cosmological slice, whereas Penrose needs a noncompact slice for its projection contradiction.

The theorem is often described using the strong energy condition, but its geometric proof uses timelike convergence. With $\Lambda=0$, Einstein’s equation translates the usual matter strong energy condition into this Ricci inequality. A positive cosmological constant contributes with the defocusing sign for unit timelike vectors, so “matter satisfies the strong energy condition” is not by itself a substitute for the displayed curvature hypothesis when $\Lambda\ne0$.

Reverse the normal congruence

To study the past, set $w=-u$, the past-directed unit normal. Parameterize its geodesics by increasing proper time $\tau$ into the past. Reversing the tangent reverses the expansion, so their initial expansion is

$$\theta(0)=-\Theta\leq-\Theta_0.$$

The normal congruence is twist-free. In the adopted signature and curvature convention, timelike Raychaudhuri gives

$$\frac{d\theta}{d\tau} =-\frac{\theta^2}{n-1} -\sigma_{ab}\sigma^{ab} -R_{ab}w^aw^b \leq-\frac{\theta^2}{n-1}.$$

Integration shows that every such past normal geodesic either is incomplete before

$$\tau_*=\frac{n-1}{\Theta_0}$$

or develops a focal point to $S$ no later than $\tau_*$. The focal statement is not that curvature becomes infinite. It means that the normal exponential map loses rank; segments continuing beyond that focal point can no longer maximize Lorentzian distance from $S$. The precise normal-focusing and maximizing lemmas appear in Steinbauer 2023, Propositions 3.4 and 3.7, pp. 13–16.

If the strict expansion is known only at one point or tends to zero along a noncompact slice, no single $\tau_*$ controls all normal geodesics. This is why a pointwise sign on a compact $S$, or an explicitly uniform bound, is a substantive part of the theorem.

The maximizing contradiction

Because $S$ is Cauchy, the spacetime is globally hyperbolic and its past Cauchy horizon is empty. For every event $p\in I^-(S)$, compact-diamond arguments give a timelike geodesic from $p$ to $S$ that maximizes proper time. First variation makes this maximizer orthogonal to $S$, and maximality prohibits a focal point before $S$.

Suppose, for contradiction, that all past-directed timelike geodesics are complete. Choose any past-directed unit normal geodesic $\gamma$ from $S$ and a point $p=\gamma(\tau)$ with $\tau>\tau_*$. Its segment has proper time $\tau$, so the Lorentzian distance from $p$ to $S$ obeys

$$d(p,S)\geq\tau>\tau_*.$$

Global hyperbolicity supplies a maximizing timelike geodesic from $p$ to $S$ of length $d(p,S)$. Read from $S$ into the past, it is a normal geodesic with no focal point before $p$. But every complete normal geodesic must become focal by $\tau_*$. This is the contradiction. Therefore at least one past-directed timelike geodesic is incomplete.

Equivalently, the proof establishes a bound on past time separation:

$$d(p,S)\leq\frac{n-1}{\Theta_0} \qquad (p\in I^-(S)).$$

Past completeness would generate events along a unit-speed timelike geodesic whose distance from $S$ exceeds that bound. In fact, the bound says more than the theorem’s minimal existential conclusion: every past-directed timelike curve starting on $S$ has total proper duration at most $(n-1)/\Theta_0$ within this spacetime. Each finite segment has length at most the corresponding time separation; taking the supremum over segments gives the same bound for an entire past history. In particular, every past-directed timelike geodesic starting on $S$ is incomplete. This bounds elapsed proper time to the chosen slice, not the location or nature of a common past endpoint. The causal input is the existence of global maximizers; Raychaudhuri supplies only their finite maximizing lifetime.

The stronger proper-duration conclusion is stated explicitly in Wald 1984, Theorem 9.5.1 and proof, p. 237, with a uniform expansion bound in place of compactness. This proof route is developed in the original cosmological series and its causal refinements; a primary historical source is Hawking 1967, pp. 187–201. The fixed compact-Cauchy formulation above follows the modern account rather than merging all variants carrying Hawking’s name.

FLRW sign and scale check

For an $n$-dimensional Friedmann–Lemaître–Robertson–Walker metric

$$ds^2=-dt^2+a^2(t)d\Sigma_{n-1,K}^2,$$

the future comoving unit normal is $u=\partial_t$, and

$$\Theta=(n-1)H, \qquad H=\frac{\dot a}{a}.$$

If a compact constant-$t$ slice has $H(t_0)\geq H_0>0$, then $\Theta_0=(n-1)H_0$ and the theorem’s bound is

$$\tau_*\leq H_0^{-1}.$$

Let $d\Sigma_{n-1,K}^2$ have constant sectional curvature $K$. The displayed comoving contraction is

$$R_{ab}u^au^b=-(n-1)\frac{\ddot a}{a}\geq0,$$

but it is necessary rather than sufficient for the theorem’s hypothesis, which quantifies over every timelike direction. Choose a unit spatial vector $e\perp u$ and write an arbitrary unit timelike vector as $v=\gamma(u+\beta e)$, with $\lvert\beta\rvert\lt1$. Setting

$$q=\frac{\ddot a}{a}, \qquad X=H^2+\frac{K}{a^2},$$

isotropy gives

$$R_{ab}v^av^b = \gamma^2\left[ -(n-1)q +\beta^2\bigl(q+(n-2)X\bigr) \right].$$

Consequently, timelike convergence in every direction is equivalent to

$$\frac{\ddot a}{a}\leq0, \qquad H^2+\frac{K}{a^2}-\frac{\ddot a}{a}\geq0.$$

The first inequality is the comoving component; the second is the null, or infinitely tilted timelike, limit. For an Einstein–perfect-fluid model with $\Lambda=0$, they are respectively equivalent to

$$(n-3)\rho+(n-1)p\geq0, \qquad \rho+p\geq0.$$

With $\Lambda\ne0$, the first matter inequality becomes

$$8\pi G\bigl[(n-3)\rho+(n-1)p\bigr]\geq2\Lambda,$$

while the second is unchanged. Thus $\ddot a\leq0$ supports the elementary comoving concavity check, but by itself does not verify Hawking’s hypothesis. As a direct counterexample, $K=-1$ and $a(t)=A+bt$ with $A>0$ and $0<b<1$ give $R_{ab}u^au^b=0$ while $X-q=(b^2-1)/a^2<0$, so sufficiently tilted timelike directions violate convergence. The same distinction is tabulated in Ling 2020, Proposition 3.9 and Appendix B, pp. 397–398 and 425–427.

The theorem itself does not assume homogeneity or isotropy; its uniform expansion and Ricci inequalities replace them.

Global de Sitter spacetime is the essential check on the curvature hypothesis. With de Sitter radius $\ell$, its global scale factor is $a(t)=\ell\cosh(t/\ell)$ and its Hubble function is $H(t)=\ell^{-1}\tanh(t/\ell)$. Its compact slices expand for $t\gt0$, but

$$R_{ab}u^au^b=-\frac{n-1}{\ell^2}\lt0.$$

The curvature radius is constant whereas the global-slicing Hubble function is not; substituting $H(t)^2$ for $\ell^{-2}$ would give the wrong Ricci contraction. De Sitter violates timelike convergence and is geodesically complete. The Einstein static universe supplies the complementary check: timelike convergence may hold, but $\Theta=0$, so no negative past-directed expansion starts the focusing estimate.

What “cosmological singularity” does not mean

The stated conclusion guarantees past timelike incompleteness. The proof additionally bounds the proper duration of every past timelike history from $S$, as shown above. It does not identify those histories with the lives of actual observers, locate a common Big Bang point, establish curvature blowup, or show null incompleteness. Additional analysis of an FLRW model may establish some of these properties; they are not consequences of the proper-time bound alone.

Nor does it prove inextendibility. A Milne-type cosmological region can be past incomplete while embedding into a larger smooth Minkowski spacetime; incompleteness was measured in the restricted spacetime. The relation between completeness and extension regularity belongs to Spacetime Extensions and Maximality.

No genericity condition is needed here because uniform expansion gives negative expansion immediately in the past direction. No trapped surface is required. Conversely, positive average expansion along one observer is not the displayed hypersurface condition; the Borde–Guth–Vilenkin family of results uses different hypotheses and conclusions and should not be labeled the Hawking theorem.

Finally, the classical argument is a $C^2$-level statement. $C^{1,1}$ and $C^1$ Hawking theorems exist, using distributional Ricci inequalities, regularized metrics, and carefully defined completeness when geodesic uniqueness can fail. Their hypotheses are detailed separately in Steinbauer 2023, §§4.1–4.6, pp. 21–32.

Exercise

In four dimensions, a compact Cauchy slice has future expansion $\Theta\geq6/T$ everywhere, and timelike convergence holds. What past proper-time bound follows? Which conclusion would be invalid if the measured expansion were positive only in a small neighborhood on the slice?

Solution

Here $n-1=3$ and $\Theta_0=6/T$, so

$$\tau_*=\frac{3}{6/T}=\frac{T}{2}.$$

Past-directed normal geodesics either end earlier or become focal to the slice by proper time $T/2$. Global maximality then contradicts universal past timelike completeness, so at least one past timelike geodesic is incomplete.

If expansion were positive only locally, normal geodesics starting elsewhere might have zero or negative future expansion and no uniform negative past expansion. The proof could not bound all maximizing geodesics from earlier events to the slice. A local measurement therefore cannot replace the global hypersurface inequality.

References

• Hawking, Stephen W., “The Occurrence of Singularities in Cosmology. III. Causality and Singularities”, Proceedings of the Royal Society A 300, pp. 187–201, 1967.
• Ling, Eric. “The Big Bang Is a Coordinate Singularity for $k=-1$ Inflationary FLRW Spacetimes.” Foundations of Physics 50 (2020): 385–428. DOI.
• Steinbauer, Roland, “The Singularity Theorems of General Relativity and Their Low Regularity Extensions”, Jahresbericht der Deutschen Mathematiker-Vereinigung 125, Theorem 3.8 and proof, pp. 15–16, 2023.
• Wald, Robert M., General Relativity, University of Chicago Press, Chicago, 1984.