The Topological Censorship Theorem

Topological censorship is a theorem about causal accessibility, not the absence of topology. Under suitable asymptotic, causal, and focusing assumptions, a causal curve sent from infinity and received at infinity cannot represent a new homotopy class in the exterior. The proof lifts such a curve to a covering spacetime and turns it into forbidden communication between distinct copies of infinity.

Required background. Exterior Algebra and Differential Forms supplies pullbacks and boundary orientation; Hypersurface Normals and Tangential Projectors supplies codimension-two cuts and normal expansions.

Helpful background. Pseudo-Riemannian Metrics and Signature supplies conformal causal geometry; Affine and Variational Geodesics supplies focal points; Causality and Global Structure supplies global hyperbolicity and conformal infinity; Timelike Congruences and Spatial Projection prepares Raychaudhuri focusing.

A precise asymptotically flat theorem

Because the original 1993 proof has a known gap, the principal result on this page is a later theorem with explicit conditions. Let $(M,g)$ be a smooth, connected, time-oriented Lorentzian spacetime of dimension $n\geq3$, without physical manifold boundary. Assume it has a conformal completion $(\widetilde M,\widetilde g,\Omega)$ such that

$$\widetilde g=\Omega^2g, \qquad \Omega>0\text{ on }M, \qquad \Omega=0, d\Omega\ne0\text{ on }\mathscr I,$$

and the boundary consists of connected null hypersurfaces $\mathscr I^-$ and $\mathscr I^+$. The physical metric may be taken $C^2$ and the unphysical metric $C^1$ up to the boundary; smooth metrics suffice. Because $\Omega>0$ in the physical region and $\Omega=0$ at infinity, decreasing $\Omega$ points outward toward $\mathscr I$, whereas increasing $\Omega$ points inward into the physical spacetime. This orientation will matter in both the hypothesis and the focusing step.

Here are the remaining hypotheses, before the conclusion.

1. The physical spacetime $(M,g)$ is globally hyperbolic.
2. If $D^+=I^-(\mathscr I^+)\cap M$, then $\widetilde D^+=D^+\cup\mathscr I^+$ is globally hyperbolic in the completed causal space. This is the condition that supplies causal compactness at the null boundary and prevents the proof from losing curves through spatial infinity.
3. There is a simply connected neighborhood $U$ of $\mathscr I^-\cup\mathscr I^+$.
4. A neighborhood of $\mathscr I^-$ is foliated by smooth acausal hypersurfaces $S_\tau$ that are spacelike up to the boundary. For every $\tau$ and all sufficiently small $\epsilon\geq0$, the cuts

$$S_{\tau,\epsilon}=S_\tau\cap\{\Omega=\epsilon\}$$

are nonempty, smooth, connected, compact codimension-two submanifolds. 5. For each such cut, every null normal geodesic initially directed toward decreasing $\Omega$—outward toward conformal infinity—remains in $U$. This is an asymptotic-confinement assumption on the outward family, not the future-inner-trapped family used for focusing. 6. The null convergence condition holds:

$$R_{ab}k^ak^b\geq0$$

for every null $k$. If Einstein’s equation is imposed, this is equivalent to $T_{ab}k^ak^b\geq0$ because the trace and cosmological-constant terms vanish on a null vector. No Einstein equation is otherwise required.

No spacetime-orientability hypothesis is needed, beyond time orientation and the two-sided asymptotic cuts already stated. The conformal boundary is part of the theorem’s completed causal setting, so this is not an illicit application of a boundaryless theorem.

Topological censorship. Under these assumptions, the domain of outer communications

$$\mathcal D=I^+(\mathscr I^- )\cap I^-(\mathscr I^+)$$

is simply connected Chruściel–Galloway 2019, Theorem 3.1 and Remarks 3.2–3.4, pp. 4–5. Therefore any two curves in $\mathcal D$ with the same endpoints are homotopic relative to their endpoints. In particular, a causal curve from past to future infinity is deformable to a reference curve contained in the simply connected asymptotic neighborhood $U$.

There is an averaged variant. Pointwise null convergence can be replaced, with the same global hypotheses, by the condition that every relevant future half-geodesic $\gamma$ starting near $\mathscr I^-$ obey

$$\int_0^\infty R_{ab}\dot\gamma^a\dot\gamma^b\,ds\geq0.$$

The direction and starting region are essential: an average on unspecified complete two-sided geodesics is not the statement used by this proof Chruściel–Galloway 2019, discussion following Remark 3.4, pp. 5–6.

Proof architecture

The argument has five steps.

1. Unwrap the topology. Suppose the nonempty $\mathcal D$ is not simply connected and pass to the universal cover of its completion by null infinity. The simply connected neighborhood $U$ lifts to distinct copies $U_\alpha$. We first rule out any causal connection from $\mathscr I^-_1$ to $\mathscr I^+_2$ with $1\ne2$; connectedness will then finish the topological argument. A nontrivial loop is not automatically a causal curve, so these are separate steps.

2. Insert an asymptotic cut. Choose $S_{\tau,\epsilon}$ near the point where that connection leaves $\mathscr I^-_1$. For small $\epsilon$, the future null expansion of the inward family—initially directed toward increasing $\Omega$—is negative. This is the future-inner-trapped property of the cut, not a declaration that both null expansions are negative.

3. Find an achronal generator. Global hyperbolicity of the physical and completed regions, together with the fact that the future of a compact set cannot contain all of $\mathscr I^+_2$, provides a generator of $\dot J^+(S_{\tau,\epsilon})$ reaching toward the second future infinity. It begins orthogonally at the cut and has no focal point strictly before its endpoint while it remains on the achronal boundary.

4. Select the inner family. In the lifted asymptotic copy, the decreasing-$\Omega$ family points outward toward its conformal infinity and remains inside the lifted neighborhood $U_1$ by hypothesis 5. The putative generator must meet $\partial U_1$ on its way to a distinct future-infinity copy, so it cannot belong to that confined outward family. It is therefore forced onto the increasing-$\Omega$ inward family, whose expansion is negative by step 2. Alternative sufficient conditions exist, but some such global control is needed.

5. Focus and finish with connectedness. Raychaudhuri evolution produces a focal point along the complete generator. Beyond it the generator enters $I^+(S_{\tau,\epsilon})$, contradicting achronality. Different lifted infinities therefore cannot communicate. Every point of the lifted domain lies between some past and future infinity, which must now carry the same label. The open sets $I^+(\mathscr I^-_\alpha)\cap I^-(\mathscr I^+_\alpha)$ consequently cover that connected domain and are pairwise disjoint. There can be only one, contradicting the multiple sheets of a nontrivial universal cover. Thus $\mathcal D$ is simply connected Chruściel–Galloway 2019, proof of Theorem 3.1, pp. 5–7.

The local focusing step alone does not prove the theorem. Hypothesis 2 supplies the boundary generator, while hypothesis 5 confines the outward branch and thereby selects the inward branch to which focusing applies.

The 1993 formulation and its erratum

Friedman, Schleich, and Witt formulated the result in four-dimensional asymptotically flat spacetimes whose conformal boundary has components $\mathscr I^\pm\simeq S^2\times\mathbb R$ with complete null generators. They assumed global hyperbolicity and ANEC,

$$\int_{-\infty}^{\infty}T_{ab}k^ak^b\,d\lambda\geq0,$$

on every inextendible affinely parametrized null geodesic, and claimed that every causal curve from $\mathscr I^-$ to $\mathscr I^+$ is deformable relative to infinity to a reference timelike curve in a simply connected neighborhood of infinity Friedman–Schleich–Witt 1993, Theorem 1 and proof, pp. 1486–1489.

Two later corrections must be distinguished.

The published erratum concerns a secondary passive-censorship claim. The original article also attributed a result implying that every $K(\pi,1)$ prime factor of a Cauchy surface lay behind an apparent horizon. Burnett constructed a counterexample. The authors explained that they had conflated future and past apparent horizons: the factor could lie in a white-hole region and be passively visible. They withdrew that secondary claim while stating that the active Theorem 1 was unaffected Friedman–Schleich–Witt 1995, erratum, p. 1872; see also Burnett 1995, pp. 6856–6859.

A separate gap affects the proof of the active formulation. Krasnikov observed that the argument treats a certain nonempty subset $A$ of a connected component of lifted $\mathscr I^+$ as closed, then concludes it cannot also be open. Connectedness only says a clopen subset is empty or the whole component; $A$ could equal all of $\mathscr I^+_\alpha$. The missing exclusion is precisely what is needed to locate a boundary point and hence an achronal null generator. This is the spatial-infinity-avoidance problem Krasnikov 2013, §2, pp. 54–56.

The modern theorem above should therefore be read as a precise sufficient result, not as a silent restatement of the original hypotheses. Chruściel and Galloway explicitly add completed-region global hyperbolicity and control of the relevant null normals, and list alternative conditions that can replace the latter Chruściel–Galloway 2019, Introduction and Proposition 3.6, pp. 1–2 and 7–9.

What the theorem does not censor

Maximally extended Schwarzschild contains an Einstein–Rosen bridge on a Cauchy slice. It does not provide a causal trip from one infinity through the bridge to another: the throat pinches off causally, exactly as the theorem permits. The theorem neither removes the bridge nor proves the existence of its event horizon.

The $\mathbb{RP}^3$ geon similarly has nontrivial spatial topology. Signals originating at a past white-hole singularity can reveal that topology passively, but they do not define a causal probe sent from $\mathscr I^-$ and returned to $\mathscr I^+$ in a forbidden homotopy class.

A standard traversable Morris–Thorne throat evades the focusing premise by violating null convergence, and typically ANEC, along the threading null congruence. A flat spatial quotient can retain visible loops while failing the prescribed asymptotic neighborhood. Anti-de Sitter infinity is timelike and requires a boundary-capable version with its own hypotheses. None of these is a counterexample to the theorem actually stated.

The conclusion applies to $\mathcal D$, not to the black-hole interior or the whole spacetime. It does not classify higher homotopy groups, prove spherical horizon topology, prohibit topology change, or establish cosmic censorship.

Exercises

Assume all hypotheses of the precise theorem and suppose, for contradiction, that a causal curve from $\mathscr I^-$ to $\mathscr I^+$ is not fixed-endpoint homotopic to a curve in $U$. Explain why its lift must connect distinct asymptotic copies and identify the two distinct reasons that the corresponding null generator cannot remain on $\dot J^+(S_{\tau,\epsilon})$ forever.

Solution

Choose the lift of the past endpoint in one copy $U_1$. Fixed-endpoint homotopy classes are classified by endpoints of lifted paths. A reference curve in $U$ lifts from $U_1^-$ to $U_1^+$; the nonhomotopic curve must therefore end at $U_2^+$ for a different sheet.

Global hyperbolicity of the completed future region and avoidance of spatial infinity provide a point on the boundary of the future of a compact asymptotic cut. The resulting curve is an orthogonal null generator of $\dot J^+(S_{\tau,\epsilon})$. The decreasing-$\Omega$ normals point outward and remain in the chosen lifted asymptotic neighborhood. Because the generator must leave that neighborhood to reach a distinct copy of future infinity, it is forced instead onto the increasing-$\Omega$ inward family, whose initial expansion is negative. Null convergence, or the stated future half-geodesic averaged condition, then forces a focal point. A generator after a focal point lies in $I^+(S_{\tau,\epsilon})$, not on its boundary. Thus the global causal assumptions first produce and classify the generator, while focusing separately destroys its achronal-boundary status. Both ingredients are necessary.

References

• Burnett, Gregory A., “Counterexample to the Passive Topological Censorship of $K(\pi,1)$ Prime Factors”, Physical Review D 52, pp. 6856–6859, 1995. Open PDF.
• Chruściel, Piotr T., and Gregory J. Galloway, “Roads to Topological Censorship”, preprint UWThPh-2019-17, 2019.
• Friedman, John L., Kristin Schleich, and Donald M. Witt, “Topological Censorship”, Physical Review Letters 71, pp. 1486–1489, 1993; erratum, volume 75, p. 1872, 1995. Open PDF.
• Krasnikov, Sergey V., “‘Topological Censorship’ Is Not Proven”, Gravitation and Cosmology 19, no. 1, pp. 54–56, 2013. Open PDF.