Maximal Cauchy Development and Determinism

Einstein initial data do determine a largest spacetime region in which the initial hypersurface remains Cauchy. That region is the maximal globally hyperbolic development (MGHD), and it is unique as a geometry even though coordinate descriptions are not unique. This is the precise deterministic statement available before either cosmic-censorship conjecture is invoked. It does not say that the MGHD is inextendible as a Lorentzian manifold, or that an extension beyond a Cauchy horizon is unique.

Required background. The Einstein Equation supplies the field equations and their constraint content. Helpful background. Pseudo-Riemannian Metrics and Signature supplies Lorentzian geometry; Affine and Variational Geodesics supplies completeness language; Causality and Global Structure supplies Cauchy surfaces and domains of dependence; Formulations of General Relativity separates constraints, evolution, and gauge; Exact Solutions of Einstein’s Equations supplies the horizon examples; PDEs and Green Functions for Gravitation supplies hyperbolic well-posedness.

Initial data and developments

Let $\Sigma$ be a three-manifold. Vacuum initial data consist of a Riemannian metric $h_{ij}$ and a symmetric tensor $K_{ij}$. With future unit normal $n^a$, this page chooses the extrinsic-curvature convention

$$K_{ij}=-h_i{}^a h_j{}^b\nabla_a n_b =-\frac12\mathcal L_n h_{ij},$$

where $n^a$ is the future unit normal. With matter fields included, their initial positions and momenta must also be prescribed. The data are not arbitrary: they satisfy the Hamiltonian and momentum constraints

$${}^{(3)}R+K^2-K_{ij}K^{ij}=16\pi G\rho+2\Lambda,$$ $$D_jK^j{}_i-D_iK=8\pi G j_i.$$

Here $K=h^{ij}K_{ij}$, $\rho=T_{ab}n^a n^b$, and $j_i=-h_i{}^aT_{ab}n^b$. A particular Einstein–matter system may impose further constraints, such as the Maxwell Gauss laws. “Admissible initial data” must therefore specify the matter equations, constraint class, differentiability, topology of $\Sigma$, and asymptotic or boundary conditions. The vacuum theorem cannot simply be transferred to a matter model whose evolution is not locally well posed.

A development of these data is a time-oriented spacetime $(M,g)$, together with the evolved matter fields and an embedding $\iota:\Sigma\to M$, such that:

• the field equations hold in the chosen regularity class;
• $\iota^*g=h$ and the embedded hypersurface has second fundamental form $K$;
• the embedded matter data agree with the prescribed data; and
• $\iota(\Sigma)$ is a Cauchy surface of $(M,g)$.

The final condition is decisive. It says that every inextendible causal history in $M$ meets the data surface exactly once. Hence every point of the development lies in the region controlled by those data. A solution region in which $\Sigma$ is merely spacelike is not necessarily a Cauchy development.

Two developments represent the same physical evolution when there is an isometry between them that fixes the embedded initial data. This quotient by data-fixing isometry removes coordinate gauge. It does not erase genuinely different continuations beyond a predictability boundary.

The maximal-development theorem

Fix an Einstein or Einstein–matter Cauchy problem for which local geometric existence and local geometric uniqueness have been proved at a stated differentiability. Order its globally hyperbolic developments by data-fixing isometric embedding. The Choquet-Bruhat–Geroch result is then:

Maximal globally hyperbolic development theorem. Every admissible initial data set has a globally hyperbolic development $(M_{\max},g_{\max})$ into which every other globally hyperbolic development of the same data embeds isometrically while fixing the data. Any two such maximal developments are related by a data-fixing isometry.

This is Theorem 3 of Choquet-Bruhat and Geroch 1969, pp. 332–334. Their definitions of initial data, development, and extension appear immediately beforehand in Choquet-Bruhat and Geroch 1969, pp. 330–332. In the original paper the explicit equations are source-free, while the argument is stated to apply to systems—such as suitable fluid or electromagnetic couplings—whose local Cauchy problem has the requisite existence and uniqueness. Thus the global theorem packages a local well-posedness theorem; it does not manufacture one for arbitrary matter.

The adjective “maximal” has a categorical meaning: maximal among developments in which the original data surface remains Cauchy. It is stronger than merely having no larger member along one chosen chain, because every other development embeds into $M_{\max}$. It is weaker than “inextendible as a spacetime of regularity $C^k$,” because a larger spacetime may exist in which $\Sigma$ is no longer Cauchy.

Why the theorem is global

Local hyperbolic theory gives two ingredients. First, at least one development exists near $\Sigma$. Second, any two developments contain a common neighborhood of $\Sigma$ on which a data-fixing isometry identifies them. The global problem is to combine all compatible local pieces without creating a non-Hausdorff quotient or branch-dependent evolution.

Choquet-Bruhat and Geroch first use local uniqueness to show that compatible identifications between two developments coincide wherever both are defined. Zorn’s lemma then produces a maximal common development. They also order all developments and form upper bounds for chains. The remaining subtlety is Hausdorffness when a candidate maximal development is glued to another one along their maximal common part. If a pair of boundary points failed to separate, a small spacelike surface through such a point would have locally unique developments on both sides; this would enlarge the alleged maximal common part, a contradiction. The gluing is therefore a spacetime, and maximality forces it to coincide with the candidate. This proof architecture, including the Hausdorff step, occupies Choquet-Bruhat and Geroch 1969, pp. 332–334.

The proof is “global” in the order-and-gluing sense. It does not prove that curvature stays bounded for all time, that geodesics are complete, or that no black hole forms. An MGHD may end at a crushing singularity, at a weak null singularity, or at a regular-looking Cauchy horizon. Those are dynamical and regularity questions.

What is deterministic

Within the MGHD, the initial data determine the following objects up to isometry:

• the manifold region reached while $\Sigma$ remains Cauchy;
• the metric and matter fields there;
• all causal, curvature, horizon, and asymptotic structures intrinsic to that region; and
• the MGHD obtained from any other Cauchy surface lying inside the same development.

The last statement is the theorem’s Corollary 1: changing to another Cauchy slice of $M_{\max}$ does not change the maximal development Choquet-Bruhat and Geroch 1969, pp. 334–335. This is a geometric semigroup property of evolution, not a preferred-foliation claim.

Coordinate functions are not determined. A lapse, shift, harmonic coordinate system, or double-null gauge may be chosen differently without changing the spacetime. Nor does the theorem assert continuous dependence in every imaginable norm. Stability requires a specified topology on the constraint set and a well-posedness estimate for the chosen formulation.

Where determinism can stop

Suppose $M_{\max}$ embeds properly into a larger spacetime $(\widetilde M,\widetilde g)$ that solves the same Einstein–matter equations in the development’s regularity class. The image of $\Sigma$ cannot remain Cauchy there; otherwise $\widetilde M$ would be a larger globally hyperbolic development, contradicting maximality. Such an extension introduces causal histories not determined by $\Sigma$, with the loss of its domain of dependence organized by a Cauchy horizon.

For an arbitrary Lorentzian metric extension, the same argument is unavailable: the extension might fail the field equations or the stipulated matter regularity, so it need not be another development even if the initial surface remains Cauchy. This is why metric inextendibility is a stronger demand than inextendibility within a chosen solution class. The exact inner-horizon examples below do satisfy the same classical field equations across the horizon and illustrate the genuine Cauchy-horizon case.

Reissner–Nordström and Kerr make the distinction concrete. For appropriate complete two-ended initial data, their MGHDs terminate at inner Cauchy horizons, while the exact analytic metrics admit extensions beyond them. The extensions are not globally hyperbolic developments of the original data. Data on $\Sigma$ alone do not select all events in the added region, and different extensions need not be isometric. This is the determinism problem addressed by Strong Cosmic Censorship: are MGHDs from generic data inextendible in a specified regularity class?

Even a continuous extension and a $C^2$ extension are different propositions. Divergent curvature can obstruct $C^2$ continuation while leaving a continuous metric extension possible. Conversely, $C^0$-inextendibility blocks every stronger metric extension. The maximal-development theorem chooses none of these thresholds.

Weak cosmic censorship asks a different question: whether singular behavior arising from generic asymptotically flat collapse is visible from future null infinity. It concerns the exterior and asymptotic observers, not uniqueness beyond every possible Cauchy horizon. Isenberg explicitly stresses that weak and strong cosmic censorship do not imply one another in Isenberg 2015, §3, pp. 6–7.

Limits of the theorem

The result must not be overextended in four common ways.

Regularity is part of the theorem. “Smooth” in the 1969 paper means the differentiability required by its local Theorems 1 and 2. Modern low-regularity Einstein systems require their own local existence and geometric uniqueness statements.

Matter is part of the theorem. A stress tensor satisfying an energy condition is not enough. The coupled evolution equations and their constraints must form a well-posed Cauchy problem.

Boundaries change the problem. Anti-de Sitter-type evolution has a timelike conformal boundary and requires boundary data. It is an initial-boundary value problem, not automatically a Cauchy development from data on $\Sigma$ alone.

Maximal is not geodesically complete. The theorem guarantees a largest predictable development, not infinite affine length or absence of curvature blowup. Minkowski space has both properties; black-hole and cosmological developments need not.

Formulation diagnostic

Problem. A smooth, constraint-satisfying vacuum data set has an MGHD $M$. The metric extends smoothly through a null boundary to two nonisometric spacetimes $\widetilde M_1$ and $\widetilde M_2$. In each extension there are causal curves that never meet the original data surface. Has the Choquet-Bruhat–Geroch uniqueness theorem failed? Which additional claim would rule out both extensions?

Solution

No. The theorem compares globally hyperbolic developments in which the original surface is Cauchy. The causal curves avoiding that surface show that neither $\widetilde M_1$ nor $\widetilde M_2$ belongs to this category. The theorem says that both contain the same MGHD, not that their added regions agree. A strong-cosmic-censorship statement asserting generic smooth—or, more strongly, $C^0$—inextendibility of the MGHD would rule out such extensions. The formulation must still name the data class and topology behind “generic.”

Next compare the observer-dependent notions on What Counts as a Naked Singularity? before formulating weak censorship.

References

• Choquet-Bruhat, Yvonne, and Robert Geroch. “Global Aspects of the Cauchy Problem in General Relativity.” Communications in Mathematical Physics 14 (1969): 329–335. https://doi.org/10.1007/BF01645389.
• Isenberg, James. “On Strong Cosmic Censorship.” Surveys in Differential Geometry 20 (2015): 17–36. arXiv:1505.06390.