Global Hyperbolicity

Global hyperbolicity is the strongest of the standard causality conditions, and the one that turns general relativity into a deterministic field theory. A spacetime earns the name when there is a single instant of time — a Cauchy surface $\Sigma$ — that every inextendible causal curve crosses exactly once. Give the fields and their normal derivatives on that one slice and, in principle, the entire past and future of the universe is fixed: nothing sneaks in from a singularity, from infinity, or around a closed timelike loop to spoil the prediction. It is the geometric statement that the world runs on initial data. The condition wears two faces that a deep theorem proves identical. The analytic face, in the streamlined Bernal–Sánchez form, asks only that the spacetime be causal and that every causal diamond $J^{+}(p)\cap J^{-}(q)$ be compact — a "no leaks" criterion forbidding causal curves that run off to a missing edge in finite parameter. The geometric face simply asks for a Cauchy surface. Geroch's 1970 theorem welds the two together and adds a bonus: a globally hyperbolic spacetime splits smoothly as $\mathbb{R}\times\Sigma$ , foliated by Cauchy slices of a global time function. The physical dividend is the initial value formulation. Cast in harmonic gauge the Einstein equations become quasilinear wave equations, and Choquet-Bruhat and Geroch proved that admissible data on $\Sigma$ possess a unique maximal globally hyperbolic development. Global hyperbolicity is exactly the arena on which that determinism holds; beyond it lie Cauchy horizons where prediction fails. The same condition underwrites quantum field theory on curved spacetime, the singularity theorems, and the very statement of cosmic censorship.

Key takeaways

A spacetime $(M,g)$ is globally hyperbolic when it is causal (no closed causal curves) and every causal diamond $J^{+}(p)\cap J^{-}(q)$ is compact.
Equivalently — the deep theorem — $M$ is globally hyperbolic if and only if it admits a Cauchy surface: a closed achronal set whose domain of dependence is all of $M$ .
Geroch's theorem then splits the spacetime smoothly, $M\cong\mathbb{R}\times\Sigma$ , foliated by the Cauchy level sets of a global time function $t$ .
It is the precise condition for a well-posed initial value problem: data on $\Sigma$ determine the solution of every hyperbolic field equation throughout $M$ , uniquely and with finite propagation speed.
It sits at the top of the causality hierarchy and forbids interior Cauchy horizons and naked singularities; being singular (Schwarzschild) or geodesically complete (anti-de Sitter) is a separate question.
Examples: Minkowski, FRW, maximally extended Schwarzschild up to $r=0$ . Non-examples: anti-de Sitter, the analytically extended Reissner–Nordström and Kerr interiors, and Gödel's universe.
Key idea: global hyperbolicity = causality + compact causal diamonds, equivalent to the existence of a Cauchy surface and to a smooth splitting $M\cong\mathbb{R}\times\Sigma$ — the geometric encoding of deterministic, well-posed physics, where data on one spacelike slice fix everything.

The core insight

Global hyperbolicity = causality + compact causal diamonds, which is equivalent to the existence of a Cauchy surface and to a smooth splitting $M\cong\mathbb{R}\times\Sigma$ . It is the geometric encoding of deterministic, well-posed physics: data on one spacelike slice fix the entire spacetime.

Where this lives today

Global hyperbolicity is the silent hypothesis under nearly every quantitative frontier in gravity. Numerical relativity — the source of the waveform templates that let LIGO–Virgo–KAGRA detect black-hole mergers — is a direct implementation of the initial value formulation: codes lay constrained Cauchy data on a spacelike slice and march it forward with the reduced Einstein equations in a hyperbolic gauge, and the whole scheme is well-posed precisely because the development is globally hyperbolic. Quantum field theory on curved spacetime rests on the same footing: only there does the wave operator have unique advanced and retarded Green's functions whose difference fixes the canonical commutation relations, and the Hadamard condition selecting physical states is a statement about that causal propagator. The sharpest open problem is strong cosmic censorship — the conjecture that generic reasonable initial data have an inextendible maximal globally hyperbolic development, so that the Cauchy horizons of Reissner–Nordström and Kerr are artifacts of exact symmetry. Recent work on the blue-shift instability of those inner horizons has turned the conjecture into a quantitative programme, still unsettled. Determinism in general relativity is not a given — it is a theorem that must be re-proved at the edge of every black hole.

The physical picture

Picture the history of the universe as a deck of photographs, one per instant. If there is a single card — one spacelike slice $\Sigma$ — from which, knowing the fields and their rates of change, you could reconstruct every other card without any further input, then the spacetime is globally hyperbolic. That card is a Cauchy surface, and the whole deck is the foliation $\mathbb{R}\times\Sigma$ . Every inextendible causal curve — every possible worldline of light or matter, traced maximally in both directions — punches through the deck crossing each card exactly once, so no signal can ever appear that was not already registered on $\Sigma$ .

Figure

The globally hyperbolic splitting M ≅ ℝ × Σ: horizontal Cauchy surfaces Σ_t stack along a time function t, and every inextendible causal curve threads the whole stack meeting each slice exactly once, so data on one Σ fix the fields on all of them through the metric g = −N² dt² + h_t

The compact-diamond condition is the "no leaks" clause spelled out. The causal diamond $J^{+}(p)\cap J^{-}(q)$ collects every event $p$ can influence that can in turn influence $q$ . If that region is compact — closed and bounded, nothing escaping to infinity or to a missing edge — then a causal curve caught between $p$ and $q$ cannot wander off in finite parameter to somewhere the data on $\Sigma$ never controlled. That closes every loophole: no naked singularity emitting unpredictable signals, no timelike infinity feeding in fresh boundary data forever. Where global hyperbolicity fails, one loophole is open. Anti-de Sitter is the cleanest case: its conformal boundary is timelike and light reaches it in finite time, so no single slice can ever be Cauchy.

Figure

Anti-de Sitter is not globally hyperbolic: its conformal boundary 𝓘 at r → ∞ is timelike and an outgoing null ray reaches it in the finite coordinate time Δt = πℓ/2 — the integral of dr/(1 + r²/ℓ²) from 0 to ∞ — so influence enters from 𝓘 across a Cauchy horizon and no spacelike slice Σ has domain of dependence equal to the whole spacetime

The whole subject is conformal. A rescaling $g\to\Omega^{2}g$ with $\Omega>0$ leaves every light cone where it was, so global hyperbolicity — a statement about causal curves alone — is untouched. That is why it can be read straight off a Penrose–Carter diagram: draw a candidate slice and ask whether every $45^{\circ}$ causal line crosses it once. Minkowski passes; anti-de Sitter, with its timelike edge, does not.

The core equations

The modern definition is the cleanest starting point. In the Bernal–Sánchez formulation only two hypotheses are needed — that the spacetime be causal, and that its causal diamonds be compact — and everything else about global hyperbolicity follows.

\text{(GH)}\iff (M,g)\ \text{causal}\ \text{and}\ J^{+}(p)\cap J^{-}(q)\ \text{compact}\ \ \forall\,p,q\in M \qquad (1)

where:

$M$ — the spacetime manifold; $g$ — Lorentzian metric, signature $(-,+,+,+)$ .
$J^{+}(p)$ — causal future of $p$ : events reachable from $p$ by future-directed causal curves; $J^{-}(q)$ — causal past of $q$ .
causal — no closed causal curve passes through any point.

Reading: compactness of the diamond is a "no leaks" clause — a causal curve trapped between $p$ and $q$ cannot escape to infinity or to a missing edge in finite parameter. The geometric characterization replaces the analytic hypotheses with a single object: a slice that all of physics can be posed on. A Cauchy surface is a set no causal curve can dodge.

\Sigma\ \text{Cauchy surface}\iff \Sigma\ \text{closed, achronal, and}\ D(\Sigma)=M \qquad (2)

where:

$\Sigma$ — a candidate hypersurface; achronal — no two of its points are joined by a timelike curve.
$D(\Sigma)=D^{+}(\Sigma)\cup D^{-}(\Sigma)$ — the total domain of dependence.
$D^{+}(\Sigma)$ — future domain of dependence: points all of whose past-inextendible causal curves meet $\Sigma$ .

Reading: $D(\Sigma)=M$ says the data on $\Sigma$ reach everywhere — no event escapes their causal grip. The bridge between the two faces is Geroch's equivalence theorem. It is the reason one may use "globally hyperbolic" and "admits a Cauchy surface" interchangeably.

(M,g)\ \text{globally hyperbolic} \iff M\ \text{admits a Cauchy surface} \qquad (3)

where:

$\iff$ — logical equivalence: the analytic condition (1) and the geometric condition (2) pick out the identical class of spacetimes.

Reading: compact diamonds force a Cauchy surface to exist, and the existence of one forces the diamonds to be compact. The theorem carries a structural bonus. A globally hyperbolic spacetime is not merely slice-able; it is a smooth product, with a lapse and a running spatial metric.

M\cong \mathbb{R}\times\Sigma,\qquad g=-N^{2}\,\mathrm{d}t^{2}+h_{t} \qquad (4)

where:

$\mathbb{R}$ — the time factor, parametrized by a smooth temporal function $t$ ; $\Sigma$ — a Cauchy surface, and every level set is diffeomorphic to it.
$N>0$ — the lapse, proper time per unit $t$ along the normal; $h_{t}$ — the induced Riemannian metric on $\Sigma_{t}$ .

Reading: there is a genuine cosmic "time" $t$ whose constant- $t$ surfaces are all Cauchy — determinism has a foliation to run on. This is what determinism buys concretely. A hyperbolic field equation with data on $\Sigma$ has one and only one solution on all of $M$ , the scalar wave equation being the prototype.

\square_{g}\,\phi = 0,\quad \phi|_{\Sigma}=\phi_{0},\ \ n^{\mu}\nabla_{\mu}\phi|_{\Sigma}=\pi_{0}\ \Rightarrow\ \phi\ \text{unique on}\ M \qquad (5)

where:

$\square_{g}=g^{\mu\nu}\nabla_{\mu}\nabla_{\nu}$ — the curved-space d'Alembertian; $\phi$ — a scalar field, the prototype of any hyperbolic field.
$\phi_{0},\pi_{0}$ — initial data: field and normal derivative on $\Sigma$ ; $n^{\mu}$ — future-pointing unit normal.

Reading: two functions on one slice determine the field forever — the mathematical meaning of "well-posed". For gravity itself the data are geometric, and they are not free: the Gauss–Codazzi relations impose constraints that any Cauchy data must satisfy before evolution can begin.

{}^{(3)}R+K^{2}-K_{ij}K^{ij}=16\pi\rho,\qquad D_{j}\!\left(K^{ij}-h^{ij}K\right)=8\pi j^{i} \qquad (6)

where:

${}^{(3)}R$ — Ricci scalar of the induced metric $h_{ij}$ ; $K_{ij}$ — extrinsic curvature of $\Sigma$ ; $K=h^{ij}K_{ij}$ its trace.
$\rho,\,j^{i}$ — energy and momentum density seen by the normal observer; $D_{j}$ — intrinsic covariant derivative. Units $G=c=1$ .

Reading: the Hamiltonian and momentum constraints are the " $00$ " and " $0i$ " Einstein equations, fixing which $(\Sigma,h_{ij},K_{ij})$ are admissible. Global hyperbolicity is the summit of a graded ladder of causality conditions, each strictly stronger than the last.

\text{glob.\ hyp.}\Rightarrow\text{causally simple}\Rightarrow\text{stably causal}\Rightarrow\text{strongly causal}\Rightarrow\text{distinguishing}\Rightarrow\text{causal}\Rightarrow\text{chronological} \qquad (7)

where:

$\Rightarrow$ — logical implication; a spacetime satisfying the left condition satisfies every one to its right, but not conversely.

Reading: global hyperbolicity is the strongest reasonable demand — everything weaker leaves some door to acausality or unpredictability ajar. Finally, the whole hierarchy — global hyperbolicity included — is conformally invariant, which is what makes causal diagrams legitimate diagnostic tools.

g\to\Omega^{2}g,\ \ \Omega>0\ \Rightarrow\ \text{global hyperbolicity preserved} \qquad (8)

where:

$\Omega(x)$ — a smooth strictly positive conformal factor.

Reading: rescaling cannot flip the sign of $g_{ab}V^{a}V^{b}$ , so light cones — and every causal relation built from them — survive untouched.

Derivation AGeroch's theorem and the splitting into a product spacetime

The goal is to build, from nothing but compact causal diamonds, a continuous global time function whose level sets are Cauchy surfaces, and so prove that a globally hyperbolic spacetime is a topological product $M\cong\mathbb{R}\times\Sigma$ . The construction is startlingly economical: it uses only spacetime volumes measured toward past and future, and the single fact that moving forward enlarges your past while shrinking your future. Surjectivity of the time function, achronality of its level sets, and the product structure all fall out of compactness.

Step 1. Because $M$ is a manifold it carries a smooth measure; rescale it so the total volume is finite, $\mu(M)<\infty$ , and define the past and future volumes $\theta^{-}(p)=\mu\!\left(J^{-}(p)\right)$ and $\theta^{+}(p)=\mu\!\left(J^{+}(p)\right)$ . Why: a finite total volume makes these two numbers finite and comparable at every event, which is all the raw material the proof needs.

Step 2. Along any future-directed causal curve the past region only grows and the future region only shrinks, so $\theta^{-}$ strictly increases and $\theta^{+}$ strictly decreases toward the future. Why: strict monotonicity is what will make the time function genuinely increasing along every causal curve, not merely non-decreasing.

Step 3. Causality together with compact diamonds implies strong causality (the Bernal–Sánchez sharpening), and strong causality makes $\theta^{\pm}$ continuous: a small displacement of $p$ changes the volumes continuously because no causal curve re-enters a small neighbourhood it has left. Why: continuity is required before the level sets can be honest hypersurfaces rather than fractured sets.

Step 4. Define the time function $t(p)=\ln\!\big(\theta^{-}(p)/\theta^{+}(p)\big)$ . It is continuous, and by Step 2 strictly increasing along every future-directed causal curve — a global time function. Why: the logarithm of the ratio combines the two monotone volumes into one number that grows without cancellation.

Step 5. Compactness of the diamonds forces $\theta^{-}\to0$ in the far past and $\theta^{+}\to0$ in the far future along any inextendible causal curve, so $t$ runs over all of $(-\infty,\infty)$ and each such curve meets each level set exactly once. Why: this surjectivity plus single-crossing is precisely the defining property of a Cauchy surface.

Step 6. Each level set $\Sigma_{a}=t^{-1}(a)$ is therefore a closed achronal set crossed once by every inextendible causal curve — a Cauchy surface — and the flow lines of a complete future-directed timelike vector field $V$ with $V(t)>0$ carry one slice onto another, giving the homeomorphism $M\cong\mathbb{R}\times\Sigma$ . Why: the flow supplies the explicit product coordinates, one factor time, one factor the fixed slice topology.

Step 7. Bernal and Sánchez later upgraded $t$ to a smooth temporal function with spacelike level sets, promoting the topological product to the smooth metric splitting $g=-N^{2}\mathrm{d}t^{2}+h_{t}$ . Why: the smooth version is what the initial value formulation actually runs on, since it needs differentiable slices and a lapse.

Go deeper · Derivation B · 2 minWell-posedness of the wave equation on a globally hyperbolic spacetime

Derivation BWell-posedness of the wave equation on a globally hyperbolic spacetime

The goal is to show that global hyperbolicity is exactly what makes the scalar wave equation deterministic: data on a Cauchy surface yield a unique solution, propagated at finite speed within the light cones. The engine is a conserved positive energy built from the field's stress tensor and the foliation's timelike normal; a slab between two slices gives uniqueness, a compact diamond gives finite propagation speed. Compactness enters at the decisive step, which is why the theorem lives on globally hyperbolic spacetimes and nowhere weaker.

Step 1. For a solution of $\square_{g}\phi=0$ form the stress tensor $T_{\mu\nu}=\partial_{\mu}\phi\,\partial_{\nu}\phi-\tfrac12 g_{\mu\nu}(\partial\phi)^{2}$ , which is conserved, $\nabla^{\mu}T_{\mu\nu}=0$ . Why: a conserved tensor is what Gauss's theorem can turn into a statement about boundary fluxes between slices.

Step 2. Contract with the foliation's future timelike vector $\xi^{\nu}=N n^{\nu}$ to build the energy current $P^{\mu}=-T^{\mu}{}_{\nu}\xi^{\nu}$ ; its divergence $\nabla_{\mu}P^{\mu}=-T^{\mu\nu}\nabla_{\mu}\xi_{\nu}$ is controlled entirely by the geometry. Why: choosing $\xi$ timelike makes the flux through a spacelike slice a positive-definite energy rather than an indefinite quantity.

Step 3. Integrating $P^{\mu}$ over the slab between $\Sigma_{t_{1}}$ and $\Sigma_{t_{2}}$ , the boundary flux is the positive energy $E(t)=\int_{\Sigma_{t}}\!\big(\tfrac12(n\!\cdot\!\partial\phi)^{2}+\tfrac12 h^{ij}\partial_{i}\phi\,\partial_{j}\phi\big)\sqrt{h}\,\mathrm{d}^{3}x\ge0$ . Why: positivity of $E$ is the crux — a non-negative quantity that vanishes only when the field and its gradient vanish.

Step 4. The bulk term is bounded by a constant times $E$ , so Grönwall's inequality gives $\mathrm{d}E/\mathrm{d}t\le C E$ and hence $E(t)\le E(t_{0})e^{C(t-t_{0})}$ . Why: this exponential bound prevents energy from being created out of nothing away from the data.

Step 5. If $\phi_{0}=\pi_{0}=0$ on $\Sigma$ then $E(t_{0})=0$ , so $E(t)=0$ for all $t$ , forcing $\partial_{\mu}\phi=0$ throughout $D(\Sigma)=M$ : the solution with zero data is zero, so solutions are unique. Why: uniqueness for the difference of two solutions is uniqueness itself, by linearity.

Step 6. Repeat the estimate on a truncated diamond $J^{+}(p)\cap J^{-}(q)$ — compact by global hyperbolicity — to conclude that $\phi(q)$ depends only on the data in $J^{-}(q)\cap\Sigma$ : information travels no faster than light. Why: compactness bounds the region and closes the energy argument, which is exactly where the hypothesis is indispensable.

Step 7. Existence is supplied by the symmetric-hyperbolic theory of Leray and Choquet-Bruhat, which returns a smooth solution attaining the data; the same machinery in harmonic gauge yields the maximal globally hyperbolic development of the Einstein equations. Why: uniqueness plus existence is well-posedness, the property that certifies a deterministic theory.

Go deeper · Derivation C · 1 minWhy the Einstein equations are hyperbolic in harmonic gauge

Derivation CWhy the Einstein equations are hyperbolic in harmonic gauge

The goal is to see why gravity itself has a well-posed Cauchy problem — why "the metric is determined by initial data" is a theorem, not a hope. The obstacle is gauge: the Einstein equations are generally covariant, so they cannot fix the metric until coordinates are pinned down. Harmonic (wave) coordinates do exactly that, and in them the principal part of the Ricci tensor collapses to a d'Alembertian, turning Einstein's equations into quasilinear wave equations to which the previous derivation's well-posedness applies.

Step 1. Impose harmonic coordinates, $\square_{g}x^{\mu}=0$ , equivalently the gauge condition $g^{\alpha\beta}\Gamma^{\mu}_{\alpha\beta}=0$ . Why: four coordinate conditions remove the four-fold diffeomorphism freedom that otherwise prevents the metric from being determined uniquely.

Step 2. In this gauge the Ricci tensor's second-derivative part reduces to $R_{\mu\nu}=-\tfrac12 g^{\alpha\beta}\partial_{\alpha}\partial_{\beta}g_{\mu\nu}+Q_{\mu\nu}(g,\partial g)$ , with $Q$ containing only the metric and its first derivatives. Why: the leading operator is now the scalar wave operator acting component-by-component on $g_{\mu\nu}$ — the signature of a hyperbolic system.

Step 3. The field equations $R_{\mu\nu}=8\pi\!\left(T_{\mu\nu}-\tfrac12 g_{\mu\nu}T\right)$ therefore become $\square_{g}g_{\mu\nu}=2Q_{\mu\nu}-16\pi(T_{\mu\nu}-\tfrac12 g_{\mu\nu}T)$ : ten quasilinear wave equations for the ten metric components. Why: this "reduced" system is genuinely hyperbolic, so local existence and uniqueness follow from the standard theory.

Step 4. The contracted Bianchi identity $\nabla^{\mu}G_{\mu\nu}=0$ guarantees the harmonic gauge and the constraints (6) are preserved by the evolution once they hold on $\Sigma$ , so data satisfying the constraints evolve consistently. Why: without this propagation the reduced equations could drift away from the true Einstein equations; the Bianchi identity is what keeps them equivalent.

Step 5. Patching the local developments, Choquet-Bruhat and Geroch proved a unique maximal globally hyperbolic development exists, on which $\Sigma$ is a Cauchy surface. Why: this is the precise sense in which general relativity is deterministic — and global hyperbolicity is the exact extent of that determinism.

Limiting cases

Case	Result	Meaning
Minkowski space	$\eta_{\mu\nu}=\mathrm{diag}(-1,1,1,1)$	Globally hyperbolic; any $t=\text{const}$ hyperplane is a Cauchy surface and every diamond is compact.
FRW cosmology	$\mathrm{d}s^{2}=-\mathrm{d}t^{2}+a(t)^{2}\mathrm{d}\Sigma_{k}^{2}$	Globally hyperbolic; constant- $t$ slices are Cauchy, whether the space is flat, open, or closed.
Schwarzschild (maximal)	$J^{+}(p)\cap J^{-}(q)$ compact	Globally hyperbolic up to the spacelike $r=0$ singularity — singular but predictable; the singularity is a boundary, not an interior Cauchy horizon.
Anti-de Sitter	$\mathrm{d}s^{2}=-(1+r^{2}/\ell^{2})\mathrm{d}t^{2}+\cdots$	NOT globally hyperbolic: the timelike conformal boundary $\mathscr{I}$ is reached by null rays in finite time $\pi\ell/2$ ; no Cauchy surface without boundary conditions.
Reissner–Nordström / Kerr interior	$r=r_{-}$ (inner horizon)	The maximal analytic extension is NOT globally hyperbolic past the inner Cauchy horizon $r_{-}$ ; prediction fails there — the arena of strong cosmic censorship.
Gödel universe	closed timelike curves exist	Violates even causality, so cannot be globally hyperbolic; no consistent initial value formulation exists at all.

Common trapGlobal hyperbolicity is not the same as being singularity-free or geodesically complete.

These properties are independent, and conflating them is the classic error. Maximally extended Schwarzschild is globally hyperbolic yet thoroughly singular: causal diamonds between regular events are compact and Kruskal $T=0$ is a Cauchy surface, even though geodesics run into $r=0$ . Determinism is about whether data fix the future, not about whether worldlines can be extended forever. Conversely, anti-de Sitter is geodesically complete and has no closed causal curves, yet it fails to be globally hyperbolic, because its timelike boundary at infinity lets fresh information stream in — the volume that ought to be a compact diamond leaks out to $\mathscr{I}$ . And note that compactness of $J^{+}(p)\cap J^{-}(q)$ is the load-bearing hypothesis: strong causality by itself is far weaker and does not deliver a Cauchy surface. Predictability, completeness, and regularity are three separate questions.

Worked examples

1 · The Cauchy horizon of a charged black hole (fully worked)

A Reissner–Nordström black hole has horizons where $g^{rr}=1-2M/r+Q^{2}/r^{2}=0$ . Take charge-to-mass ratio $Q/M=0.6$ in geometric units $G=c=1$ . Locate both horizons, identify which one is a Cauchy horizon, and convert it to kilometres for a solar-mass hole.

Step 1. Solve the horizon equation. Multiplying by $r^{2}$ gives $r^{2}-2Mr+Q^{2}=0$ , so $r_{\pm}=M\pm\sqrt{M^{2}-Q^{2}}$ . With $Q=0.6M$ , the discriminant is $M^{2}-0.36M^{2}=0.64M^{2}$ , whose root is $0.8M$ . Hence $r_{+}=1.8M$ (event horizon) and $r_{-}=0.2M$ (inner horizon).

Step 2. Identify the Cauchy horizon. A partial Cauchy surface reaching from spatial infinity has a future domain of dependence bounded by the inner horizon: the maximal globally hyperbolic development of the exterior data terminates at $r=r_{-}=0.2M$ . Beyond it the analytic extension continues, but the metric there is no longer fixed by the data — that surface is the Cauchy horizon, and it is where global hyperbolicity fails.

Step 3. Put in numbers for a solar mass. In geometric units $GM_{\odot}/c^{2}=1.477\ \mathrm{km}$ , so $r_{+}=1.8\times1.477=2.66\ \mathrm{km}$ and $r_{-}=0.2\times1.477=0.295\ \mathrm{km}$ . The predictable region ends less than $300$ metres from the centre.

Step 4. Read the limits. As $Q\to M$ the horizons merge, $r_{\pm}\to M$ (extremal), and the Cauchy horizon survives; for $Q>M$ the discriminant goes negative and the singularity is naked. The astrophysical analogue is Kerr, with spin for charge: $r_{\pm}=M\pm\sqrt{M^{2}-a^{2}}$ . In every case the inner horizon $r_{-}$ marks the edge of determinism; whether real collapse produces it is the content of strong cosmic censorship.

2 · Light reaching the anti-de Sitter boundary in finite time (fully worked)

Anti-de Sitter space has $\mathrm{d}s^{2}=-(1+r^{2}/\ell^{2})\mathrm{d}t^{2}+(1+r^{2}/\ell^{2})^{-1}\mathrm{d}r^{2}+r^{2}\mathrm{d}\Omega^{2}$ . Show that a radial light ray from the centre reaches $r=\infty$ in finite coordinate time, and contrast with Minkowski. This is why AdS has no Cauchy surface.

Step 1. Set $\mathrm{d}s^{2}=0$ for a radial ray ( $\mathrm{d}\Omega=0$ ): $(1+r^{2}/\ell^{2})\mathrm{d}t^{2}=(1+r^{2}/\ell^{2})^{-1}\mathrm{d}r^{2}$ , so $\mathrm{d}t=\mathrm{d}r/(1+r^{2}/\ell^{2})$ . The lapse factor cancels once, leaving a convergent integrand.

Step 2. Integrate out to infinity. $\displaystyle \Delta t=\int_{0}^{\infty}\frac{\mathrm{d}r}{1+r^{2}/\ell^{2}}=\ell\Big[\arctan\frac{r}{\ell}\Big]_{0}^{\infty}=\ell\cdot\frac{\pi}{2}$ , so $\Delta t=\pi\ell/2$ , a finite time (numerically $1.571\,\ell$ ). The affine parameter, by contrast, runs to infinity: $\dot r=E$ is constant along the ray, so AdS is null geodesically complete.

Step 3. Contrast with Minkowski, where $\mathrm{d}t=\mathrm{d}r$ and $\int_{0}^{\infty}\mathrm{d}r=\infty$ : light takes infinite coordinate time to reach infinity, whose conformal boundary is therefore null, not timelike. The finite versus infinite result is the whole difference: in Minkowski a $t=\text{const}$ slice is Cauchy, in AdS none is.

Step 4. Interpret. Because $\mathscr{I}$ is timelike and reached in finite time, an observer can receive signals from it, so data on any spacelike $\Sigma$ leave part of the future undetermined until boundary conditions are supplied at $\mathscr{I}$ . AdS is globally hyperbolic only after one adjoins boundary conditions at infinity — exactly the move that AdS/CFT and quantum field theory in AdS are forced to make.

Physics of the portrait

The modern theory of global hyperbolicity is largely one mathematician's translation of a physicist's slogan — "the world runs on initial data" — into airtight geometry. Before 1970 the equivalence of compact diamonds, Cauchy surfaces, and product structure was folklore; afterward it was a theorem with a two-page proof.

Origins · 1 minRobert Geroch and how this was discovered

Robert Geroch

1942– · the global geometry of the Cauchy problem

Robert Geroch, a student of John Wheeler at Princeton, did more than anyone to give the global structure of spacetime its rigorous modern shape. His 1970 paper "Domain of Dependence" proved the theorem this article is built around: a spacetime is globally hyperbolic if and only if it admits a Cauchy surface, and it then splits as $\mathbb{R}\times\Sigma$ foliated by Cauchy slices of a global time function. The proof's engine — a time function manufactured purely from past and future spacetime volumes — was a piece of mathematical audacity, converting a statement about causal curves into one about measures. A year earlier, with Yvonne Choquet-Bruhat, he had settled the companion analytic question, establishing that admissible data for the Einstein equations possess a unique maximal globally hyperbolic development — the theorem that makes general relativity a deterministic field theory. His reach ran across the subject: the Geroch–Held–Penrose formalism for null congruences, the Geroch group of hidden symmetries in reduced gravity, work on singularities and topology change, and lectures on mathematical physics prized for their clarity. His signature was to take a physical intuition about prediction and horizons and expose the exact geometric hypothesis that makes it true — which is precisely what global hyperbolicity is.

History

1952

— Choquet-Bruhat proves local well-posedness of the Einstein evolution problem

1953

— Leray introduces the term global hyperbolicity for hyperbolic differential operators

1969

— Choquet-Bruhat and Geroch establish the unique maximal globally hyperbolic development

1970

— Geroch proves global hyperbolicity is equivalent to a Cauchy surface and gives the splitting M ≅ ℝ × Σ

2005

— Bernal and Sánchez upgrade Geroch's splitting to a smooth spacelike foliation

Going further

Texts. Wald, General Relativity (1984), Ch. 8 and §10.1–10.2 · Hawking & Ellis, The Large Scale Structure of Space-Time (1973), §6.6 and §7.4 · Misner, Thorne & Wheeler, Gravitation (1973), §21.13 · Carroll, Spacetime and Geometry (2004), Appendix on causality and Ch. 5 · Schutz, A First Course in General Relativity (2nd ed., 2009), Ch. 10 · Hartle, Gravity (2003), Ch. 12 · Geroch, "Domain of Dependence", J. Math. Phys. 11, 437 (1970) · Choquet-Bruhat & Geroch, Commun. Math. Phys. 14, 329 (1969) · Bernal & Sánchez, Commun. Math. Phys. 257, 43 (2005) and Lett. Math. Phys. 77, 183 (2006). Key results to keep in mind: the volume time function $t=\ln(\theta^{-}/\theta^{+})$ increases along causal curves and has Cauchy level sets; the energy estimate gives uniqueness and finite propagation on compact diamonds; the harmonic-gauge reduction reads $R_{\mu\nu}=-\tfrac12\square_{g}g_{\mu\nu}+Q_{\mu\nu}$ ; the Reissner–Nordström horizons are $r_{\pm}=M\pm\sqrt{M^{2}-Q^{2}}$ , giving $r_{+}=1.8M$ and $r_{-}=0.2M$ at $Q/M=0.6$ , with $r_{-}\approx0.295$ km for a solar mass; the AdS radial-null integral is $\int_{0}^{\infty}\mathrm{d}r/(1+r^{2}/\ell^{2})=\pi\ell/2$ ; Minkowski, FRW, and Schwarzschild (up to $r=0$ ) are globally hyperbolic, while AdS, the RN/Kerr interior past $r_{-}$ , and Gödel are not. The $16\pi$ and $8\pi$ in the constraints use $G=c=1$ .

Prerequisites. Causal Structure of Spacetime · Chronological and Causal Futures and Pasts · Achronal Sets and Edges · Domain of Dependence and Cauchy Horizons · Cauchy Surfaces · Causality Conditions Hierarchy. Leads to. Penrose–Carter Conformal Diagrams · Asymptotic Flatness · Conformal Infinity and Asymptotic Structure · Null Infinity (Scri).

Ask the tutor: Want to walk through how a Cauchy horizon actually forms inside a charged black hole, or how Geroch's volume-time-function proof turns compact diamonds into a global clock?

What is the quickest intuitive statement of global hyperbolicity?

It is the condition that a spacetime runs on initial data. Concretely: there is one spacelike slice $\Sigma$ — a Cauchy surface — such that every possible history of light or matter, traced maximally forward and backward, crosses $\Sigma$ exactly once. Because nothing can dodge the slice, knowing the fields and their rates of change on $\Sigma$ determines them everywhere else. The equivalent technical phrasing is "causal, with compact causal diamonds", but the Cauchy-surface picture is what to carry in your head: one photograph of the universe, plus the laws, gives every other photograph.

How can compactness of causal diamonds possibly be equivalent to having a Cauchy surface?

That equivalence is Geroch's theorem, and the bridge is a time function built from spacetime volumes. Give the manifold a finite total measure and, at each event $p$ , record the volume of its past $J^{-}(p)$ and its future $J^{+}(p)$ . Moving forward in time grows the past and shrinks the future, so $t=\ln(\text{past}/\text{future})$ increases along every causal curve. Compactness of the diamonds is exactly what forces those volumes to sweep from zero to the whole measure along any inextendible curve, so $t$ ranges over all of $\mathbb{R}$ and each of its level sets is met once by every such curve — the definition of a Cauchy surface. Compactness is the hypothesis that makes the construction work.

Does a singularity stop a spacetime from being globally hyperbolic?

No — singularity and predictability are different questions. Maximally extended Schwarzschild is globally hyperbolic all the way up to $r=0$ : the causal diamonds between regular events are compact, and the Kruskal $T=0$ surface is Cauchy, so data on it determine the entire spacetime. The $r=0$ singularity is a boundary of the manifold, not an interior point from which unpredictable signals emerge. What does break global hyperbolicity is a Cauchy horizon — an interior surface, like the inner horizon of a charged or rotating black hole, beyond which the metric is no longer fixed by the data. Geodesic incompleteness threatens completeness; only a Cauchy horizon threatens determinism.

Why is global hyperbolicity a conformally invariant property?

Because it is defined entirely through the causal structure — which curves are causal, which sets are achronal, which diamonds are compact — and a conformal rescaling $g\to\Omega^{2}g$ with $\Omega>0$ leaves the light cones untouched. The sign of $g_{ab}V^{a}V^{b}$ cannot flip when you multiply by a positive factor, so timelike stays timelike and null stays null; every causal curve of $g$ is a causal curve of $\Omega^{2}g$ and vice versa. Both the compact-diamond condition and the existence of a Cauchy surface are statements about these causal relations alone. This invariance is exactly what makes Penrose–Carter diagrams a legitimate diagnostic: you can compress an infinite spacetime to a finite picture and still read global hyperbolicity off it.

What is the "maximal globally hyperbolic development" and why "maximal"?

Given initial data $(\Sigma,h_{ij},K_{ij})$ satisfying the constraints, Choquet-Bruhat and Geroch proved there is a spacetime — the maximal globally hyperbolic development — that (i) solves the Einstein equations, (ii) is globally hyperbolic with $\Sigma$ as a Cauchy surface, and (iii) contains every other such development. "Maximal" means it is the largest region the data can uniquely determine: you cannot extend it further while keeping both the field equations and global hyperbolicity. Its boundary, if the development is extendible, is a Cauchy horizon. Strong cosmic censorship is the conjecture that for generic data this maximal development is inextendible, so determinism never actually runs out inside a real black hole.

Where does global hyperbolicity actually get used in practice?

Everywhere the initial value formulation is run. Numerical relativity evolves constrained Cauchy data with hyperbolic (harmonic or BSSN) equations to produce merger waveforms — well-posed precisely because the development is globally hyperbolic. Quantum field theory on curved spacetime needs it so the wave operator has unique retarded and advanced Green's functions and the field algebra is well defined; anti-de Sitter, lacking it, forces boundary conditions at infinity and becomes the home of AdS/CFT. The singularity theorems, the positive mass theorem, and black-hole uniqueness all assume it or a partial-Cauchy-surface version. It is less a topic than a standing hypothesis — the fine print that lets general relativity predict at all.

Global Hyperbolicity

Where this lives today

The physical picture

The core equations

Limiting cases

Worked examples

1 · The Cauchy horizon of a charged black hole (fully worked)

2 · Light reaching the anti-de Sitter boundary in finite time (fully worked)

Physics of the portrait

Going further

Before this

Where this leads