Cauchy Surfaces

General relativity is a local theory — Einstein's equations relate curvature to matter event by event — yet we ask of it a global thing: given the universe "now", predict its entire history. That demand only makes sense if there is a relativistic replacement for the Newtonian instant, a spacelike "now" that carries a complete set of initial conditions. A Cauchy surface $\Sigma$ is exactly that replacement: a closed achronal set whose domain of dependence is the whole manifold, $D(\Sigma)=M$ , or equivalently a hypersurface that every inextendible causal curve punctures exactly once. That phrase "exactly once" is the geometric heart of the definition, and it factors into two independent guarantees. Achronality — no two points of $\Sigma$ are chronologically related — ensures no causal curve can meet $\Sigma$ twice. The domain-of-dependence condition $D(\Sigma)=M$ ensures every inextendible causal curve meets it at least once, so nothing can influence the future of $\Sigma$ without first registering on $\Sigma$ itself. No signal leaks in from infinity; no influence emerges from a singularity hidden behind a horizon. The slice sees everything. The existence of such a surface is not automatic — it is precisely the condition of global hyperbolicity. Robert Geroch proved that a globally hyperbolic spacetime is topologically a product $M\cong\mathbb{R}\times\Sigma$ , foliated by Cauchy surfaces as the level sets of a global time function whose gradient is everywhere timelike; Bernal and Sánchez later smoothed this to a genuine spacelike foliation. This product structure is what makes the $3+1$ (ADM) split and all of numerical relativity well posed. For the field equations themselves the payoff is the Choquet-Bruhat–Geroch theorem: initial data $(\Sigma,h_{ij},K_{ij})$ satisfying the Hamiltonian and momentum constraints admit a unique maximal globally hyperbolic development. Determinism in general relativity is therefore a theorem about Cauchy surfaces, not a postulate — and the spacetimes that fail to have one (charged and rotating black-hole interiors, anti-de Sitter space, Taub-NUT) are exactly where predictability dies at a Cauchy horizon.

Key takeaways

A Cauchy surface $\Sigma$ is a closed achronal hypersurface met exactly once by every inextendible causal curve, so its future and past together fill spacetime: $D(\Sigma)=M$ .
"Exactly once" factors cleanly: achronality forbids a second crossing, $D(\Sigma)=M$ forbids zero — the slice captures every causal history.
Existence of a Cauchy surface is logically equivalent to global hyperbolicity; Geroch's theorem then gives $M\cong\mathbb{R}\times\Sigma$ , a foliation by Cauchy slices.
Data $(h_{ij},K_{ij})$ obeying the Hamiltonian and momentum constraints determine a unique maximal globally hyperbolic development of Einstein's equations.
A spacelike slice need not be Cauchy: a Minkowski hyperboloid is spacelike and achronal yet bounded by a Cauchy horizon, so $D(\Sigma)\neq M$ .
Where a Cauchy surface fails to exist — inside charged or rotating holes, in anti-de Sitter space — a Cauchy horizon marks the frontier beyond which prediction is impossible.
Key idea: A Cauchy surface is the maximal achronal hypersurface from which the entire spacetime can be reconstructed, $D(\Sigma)=M$ ; its existence is equivalent to global hyperbolicity, and it converts Einstein's field equations into a well-posed initial-value problem.

The core insight

A Cauchy surface is the maximal achronal hypersurface from which the entire spacetime can be reconstructed: $D(\Sigma)=M$ . Its existence is logically equivalent to global hyperbolicity, and it converts Einstein's field equations into a well-posed initial-value problem.

Where this lives today

Cauchy surfaces are not a museum piece — they are the operating system of computational gravity. Every numerical-relativity simulation behind a LIGO-Virgo-KAGRA waveform template lays constraint-satisfying initial data (Bowen–York punctures, conformal thin-sandwich data) on a first Cauchy slice, then marches it forward through a foliation using the BSSN or generalized-harmonic reformulations of the $3+1$ equations, with the lapse $N$ and shift $N^{i}$ chosen adaptively to keep the slices away from singularities. Because the constraints propagate, "constraint violation" is the diagnostic every such code monitors: a growing violation means the discretized spacetime is drifting off the true solution. In cosmology the constant-cosmic-time slices of an FLRW universe are Cauchy surfaces — flat or open ones diffeomorphic to $\mathbb{R}^{3}$ , closed ones to $S^{3}$ — and the horizon problem is the observation that a single such slice contains regions never in causal contact, the puzzle inflation was built to solve. At the theory frontier the concept defines the open problems themselves: strong cosmic censorship is the conjecture that Cauchy horizons inside realistic black holes are destroyed by generic perturbations (recently sharpened, and contested, for near-extremal charged holes in de Sitter space), while quantum field theory in curved spacetime requires a globally hyperbolic background so that canonical data on a Cauchy surface can be consistently evolved — which is exactly why anti-de Sitter holography must impose boundary conditions at its conformal boundary to make bulk dynamics deterministic.

The physical picture

Picture a perfect snapshot of the universe: not just the value of every field but also its instantaneous rate of change, laid out on a single spacelike slice. If that data determines the entire past and future with no surprises, the slice is a Cauchy surface. "No surprises" is the precise content of $D(\Sigma)=M$ : there is nowhere a signal could originate that did not first cross $\Sigma$ . Every inextendible causal worldline — every massive particle, every photon, however it curves — threads the surface once and only once, so $\Sigma$ separates spacetime into an unambiguous future $D^{+}$ and past $D^{-}$ .

Figure

Every inextendible causal worldline (blue) punctures the Cauchy surface Σ exactly once — achronality forbids a second crossing, D(Σ) = M forbids zero — so Σ splits spacetime cleanly into its future D⁺ and past D⁻ with domain of dependence equal to the whole manifold

Achronality plus full domain of dependence is a Goldilocks condition. Too small a slice — a hyperboloid asymptotic to a light cone — leaves a whole future region unpredictable behind a Cauchy horizon. A slice that fails achronality dips below its own future and double-counts events. A Cauchy surface threads the needle. The cleanest cautionary example is the constant-proper-time hyperboloid $-t^2+r^2=-1$ in Minkowski space: it is smooth, spacelike, and achronal, yet its domain of dependence is only the interior of the future light cone, and everything outside that cone is invisible to it. The second figure makes that failure quantitative.

Figure

The Minkowski hyperboloid Σ: -t² + r² = -1 (vertex at t = 1) is spacelike and achronal but NOT Cauchy — its domain of dependence is only the future-cone interior D(Σ) = {t > r} (shaded), bounded by the past Cauchy horizon H⁻(Σ) = {t = r}; the future horizon H⁺ is empty, and any causal curve outside the cone never meets Σ

Geroch's product structure $M\cong\mathbb{R}\times\Sigma$ carries a further physical lesson: a globally hyperbolic spacetime can never change its spatial topology, because all its Cauchy slices are diffeomorphic. Topology change — a wormhole opening, a universe splitting — requires abandoning global hyperbolicity, which is to say abandoning deterministic prediction. And where determinism truly dies is the Cauchy horizon: in the Reissner–Nordström and Kerr interiors the candidate slice fails because light can enter from regions it never saw, piling up with unbounded blueshift. The strong cosmic censorship conjecture is the hope that such horizons are unstable and never form from generic data, so that the region outside every black hole stays predictable.

The core equations

The definition is a conjunction of one local condition (achronality) and one global condition (full domain of dependence). Both are needed; neither alone suffices, as the worked examples below make concrete.

\Sigma\ \text{Cauchy} \iff \Sigma\ \text{closed, achronal, and}\ D(\Sigma)=M \qquad (1)

where: $\Sigma$ — candidate hypersurface, a subset of the spacetime $M$ ; $D(\Sigma)=D^{+}(\Sigma)\cup D^{-}(\Sigma)$ — the total domain of dependence; $M$ — the full spacetime manifold. Reading: "achronal" is local and cheap; " $D(\Sigma)=M$ " is the global demand that the slice see every causal history. The domain of dependence is the engine of the definition. An event lies in $D^{+}(\Sigma)$ when its every possible past is conditioned by $\Sigma$ — no causal influence can reach it from anywhere else.

D^{+}(\Sigma)=\{\,p\in M:\ \text{every past-inextendible causal curve through }p\ \text{meets }\Sigma\,\} \qquad (2)

where: $p$ — an event of spacetime; a curve is inextendible if it cannot be prolonged, past-inextendible if it cannot be prolonged into the past. Reading: $D^{+}(\Sigma)$ is the region determined by data on $\Sigma$ ; its future boundary is the future Cauchy horizon $H^{+}(\Sigma)$ , beyond which prediction fails. The two conditions combine into a strikingly simple picture: a Cauchy surface is a hypersurface hit exactly once by every inextendible causal curve. This single line is the definition most useful in practice.

\Sigma\ \text{Cauchy} \iff \forall\,\gamma\ \text{inextendible causal:}\ |\gamma\cap\Sigma|=1 \qquad (3)

where: $\gamma$ — an inextendible (maximally extended) causal, i.e. timelike-or-null, curve; $|\gamma\cap\Sigma|$ — the number of points where it meets $\Sigma$ . Reading: "at most one" is achronality, "at least one" is $D(\Sigma)=M$ ; their conjunction is "exactly one", derived in full below. When a Cauchy surface exists, Geroch's theorem hands back a global clock. Spacetime splits as a product, foliated by Cauchy slices that are the level sets of a time function with everywhere-timelike gradient.

M\cong\mathbb{R}\times\Sigma,\qquad t:M\to\mathbb{R}\ \text{with}\ \nabla^{a}t\ \text{timelike} \qquad (4)

where: $t$ — a global time function; each level set $\Sigma_{t}=t^{-1}(t)$ is a Cauchy surface; $\nabla^{a}t$ — its gradient, everywhere past-directed timelike. Reading: all slices are diffeomorphic, so spatial topology is frozen; this is the geometric licence for the entire $3+1$ programme. Adapting coordinates to the foliation gives the ADM line element, which separates the metric into a scalar lapse, a spatial shift, and the induced 3-metric on the Cauchy slice.

\mathrm{d}s^{2}=-N^{2}\,\mathrm{d}t^{2}+h_{ij}\,(\mathrm{d}x^{i}+N^{i}\,\mathrm{d}t)(\mathrm{d}x^{j}+N^{j}\,\mathrm{d}t) \qquad (5)

where: $N$ — the lapse, the proper-time rate between neighbouring slices; $N^{i}$ — the shift, the spatial drift of the coordinate grid; $h_{ij}$ — the induced Riemannian 3-metric on $\Sigma$ (the "position" data). Reading: $N$ and $N^{i}$ are pure gauge — how you thread coordinates through spacetime — while $h_{ij}$ is physical geometry on the slice. The "velocity" half of the initial data is the extrinsic curvature: how the slice is bent inside spacetime, equivalently the time derivative of the 3-metric.

K_{ij}=-\tfrac{1}{2}\,\mathcal{L}_{n}h_{ij}=-\frac{1}{2N}\big(\partial_{t}h_{ij}-D_{i}N_{j}-D_{j}N_{i}\big) \qquad (6)

where: $K_{ij}$ — extrinsic curvature, the "time-derivative" data; $n^{a}=\tfrac{1}{N}(\partial_{t}^{a}-N^{i}\partial_{i}^{a})$ — the future unit normal to $\Sigma$ ; $D_{i}$ — the covariant derivative of $h_{ij}$ . Reading: $(h_{ij},K_{ij})$ is the relativistic analogue of position-and-momentum data — a point in the phase space of geometry. Not every $(h_{ij},K_{ij})$ is admissible. Four of Einstein's ten equations contain no second time derivative; they constrain the data on $\Sigma$ rather than evolve it.

{}^{(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 (7)

where: ${}^{(3)}\!R$ — Ricci scalar of the induced 3-metric; $K=h^{ij}K_{ij}$ — mean curvature (trace); $\rho=T_{ab}n^{a}n^{b}$ and $j^{i}=-h^{ia}T_{ab}n^{b}$ — the energy and momentum densities measured by $n^{a}$ . Reading: these are the Hamiltonian and momentum constraints — the relativistic cousins of $\nabla\!\cdot\!\vec{E}=\rho$ and $\nabla\!\cdot\!\vec{B}=0$ , restricting data rather than evolving it.

Derivation AWhy exactly one crossing means achronal plus full domain of dependence

The goal is to prove the equivalence between the two working definitions of a Cauchy surface — the abstract " $D(\Sigma)=M$ " and the vivid "every inextendible causal curve crosses $\Sigma$ exactly once" — by showing that the count of crossings splits into an upper bound supplied by achronality and a lower bound supplied by the domain-of-dependence condition. Getting this factorization straight is what lets one switch freely between the two pictures, and it explains why both conditions in the definition are indispensable rather than redundant.

Step 1. Suppose an inextendible causal curve $\gamma$ met $\Sigma$ at two distinct events $p$ and $q$ . The arc of $\gamma$ between them is a causal curve joining $p$ to $q$ , so $p$ and $q$ are causally related — and a short deformation makes them chronologically related — which directly contradicts the achronality of $\Sigma$ . Hence $|\gamma\cap\Sigma|\le 1$ . Why: achronality is exactly the statement that $\Sigma$ contains no two causally related points, so it single-handedly caps the crossing count at one.

Step 2. Now suppose some inextendible causal curve $\gamma$ missed $\Sigma$ entirely. Pick any event $p$ on $\gamma$ . Then $\gamma$ furnishes an inextendible causal history through $p$ that never touches $\Sigma$ , so by the definition of the domain of dependence $p\notin D(\Sigma)$ , whence $D(\Sigma)\neq M$ . Contrapositively, $D(\Sigma)=M$ forces $|\gamma\cap\Sigma|\ge 1$ . Why: the domain-of-dependence condition is precisely the guarantee that no causal history escapes the slice, which is what "at least one crossing" means.

Step 3. Combine the bounds: achronality gives $|\gamma\cap\Sigma|\le 1$ and $D(\Sigma)=M$ gives $|\gamma\cap\Sigma|\ge 1$ , so $|\gamma\cap\Sigma|=1$ for every inextendible causal curve. The converse runs the same implications backwards — exactly-one-crossing forbids a repeat (achronality) and forbids a miss ( $D(\Sigma)=M$ ). Why: this closes the equivalence in both directions and shows the two clauses of the definition are logically independent — drop either one and the crossing count is no longer pinned to one. *Verified: reproduces the standard characterization in Wald (1984) Sec. 8.3 and Hawking & Ellis (1973) Sec. 6.5–6.6; each implication follows directly from the definitions of achronality and of $D(\Sigma)$ .*

Go deeper · Derivation B · 2 minExistence of a Cauchy surface equals global hyperbolicity

Derivation BExistence of a Cauchy surface equals global hyperbolicity

The goal is Geroch's theorem, the structural keystone of the whole subject: a spacetime admits a Cauchy surface if and only if it is globally hyperbolic, and in that case it is a smooth product $M\cong\mathbb{R}\times\Sigma$ foliated by Cauchy slices. We take global hyperbolicity in its standard form — $M$ is causal and every causal diamond $J^{+}(p)\cap J^{-}(q)$ is compact — and construct an explicit global time function whose level sets are the required surfaces, turning an abstract causality condition into a concrete instant of time.

Step 1. ( $\Rightarrow$ ) If $\Sigma$ is a Cauchy surface then $D(\Sigma)=M$ , and one shows the causal diamonds $J^{+}(p)\cap J^{-}(q)$ are compact: any causal curve from $p$ to $q$ must cross $\Sigma$ exactly once, which pins it down and prevents the escapes to infinity or singularities that would spoil compactness. So a Cauchy surface implies global hyperbolicity. Why: the single guaranteed crossing is what confines causal curves enough to make the diamonds compact, the technical core of global hyperbolicity.

Step 2. ( $\Leftarrow$ ) Assume global hyperbolicity and build a clock. Fix a measure on $M$ of total volume one and set $f^{-}(p)=\operatorname{vol}J^{-}(p)$ and $f^{+}(p)=\operatorname{vol}J^{+}(p)$ ; define $t(p)=\ln\!\big(f^{-}(p)/f^{+}(p)\big)$ . Compactness of the diamonds makes $f^{\pm}$ continuous, and along any future-directed causal curve $f^{-}$ strictly increases while $f^{+}$ strictly decreases, so $t$ strictly increases. Why: the past-volume-over-future-volume ratio is a natural monotone quantity along causal curves, giving a continuous time function without ever solving a differential equation.

Step 3. The function $t$ ranges over all of $\mathbb{R}$ and its level sets $\Sigma_{t}=t^{-1}(t)$ are achronal (since $t$ increases along causal curves) with $D(\Sigma_{t})=M$ , hence each is a Cauchy surface. Flowing along the timelike gradient $\nabla^{a}t$ maps $M$ homeomorphically onto $\mathbb{R}\times\Sigma_{0}$ . Why: a monotone time function whose level sets each capture every causal history is exactly a foliation by Cauchy surfaces, and its flow lines supply the product coordinate.

Step 4. Geroch's construction is only continuous; Bernal and Sánchez upgraded it to a smooth spacelike foliation, so $M\cong\mathbb{R}\times\Sigma$ as a diffeomorphism with each $\{t\}\times\Sigma$ a smooth spacelike Cauchy surface. This removes the differentiability worries that had haunted the $3+1$ formalism for decades. Why: numerical relativity and the ADM split need genuinely smooth spacelike slices, which the smoothing theorem finally guarantees. *Verified: Geroch (1970), J. Math. Phys. 11, 437, and Hawking & Ellis (1973) Prop. 6.6.8; the volume-ratio time function is Geroch's; smoothing by Bernal & Sánchez (2003, 2005).*

Go deeper · Derivation C · 1 minThe constraint equations as projections of Einstein's equations onto the slice

Derivation CThe constraint equations as projections of Einstein's equations onto the slice

The goal is to see why data on a Cauchy surface cannot be freely chosen but must satisfy the Hamiltonian and momentum constraints, and to identify those constraints as the components of $G_{ab}=8\pi T_{ab}$ that carry no second time derivative. The route is the Gauss–Codazzi machinery: projecting the Einstein tensor along and across the unit normal $n^{a}$ decomposes the ten field equations into four constraints on the slice and six genuine evolution equations, exactly mirroring how $\nabla\!\cdot\!\vec E=\rho$ constrains electromagnetic data while $\partial_{t}\vec E$ evolves it.

Step 1. Project the Einstein tensor twice along the normal. The Gauss relation gives $2\,G_{ab}n^{a}n^{b}={}^{(3)}\!R+K^{2}-K_{ij}K^{ij}$ , and setting $G_{ab}n^{a}n^{b}=8\pi T_{ab}n^{a}n^{b}=8\pi\rho$ yields the Hamiltonian constraint ${}^{(3)}\!R+K^{2}-K_{ij}K^{ij}=16\pi\rho$ . Why: the doubly-normal projection mixes only the intrinsic curvature ${}^{(3)}\!R$ and the extrinsic curvature $K_{ij}$ of the slice — never a time derivative of $K_{ij}$ — so it is a condition on the data, not an evolution law.

Step 2. Project once along the normal and once tangentially. The Codazzi relation gives $G_{ab}n^{a}h^{b}{}_{i}=D_{j}(K^{j}{}_{i}-\delta^{j}_{i}K)$ , and equating to $8\pi j_{i}$ yields the momentum constraint $D_{j}(K^{ij}-h^{ij}K)=8\pi j^{i}$ . Why: the mixed projection involves only spatial derivatives $D_{j}$ of the data, again with no second time derivative, so these three equations too merely restrict $(h_{ij},K_{ij})$ .

Step 3. The remaining components — the fully tangential projection $G_{ab}h^{a}{}_{i}h^{b}{}_{j}$ — are the six equations that do contain $\partial_{t}K_{ij}$ ; these evolve the data off $\Sigma$ . The contracted Bianchi identity $\nabla^{a}G_{ab}=0$ guarantees that if the four constraints hold on $\Sigma$ , they hold throughout the evolution automatically. Why: constraint propagation means one solves the constraints once, on the initial Cauchy surface, and is then free to evolve — the logical basis of the entire initial-value formulation. *Verified: Gauss–Codazzi decomposition of $G_{ab}=8\pi T_{ab}$ as in Wald (1984) Sec. 10.2 and MTW (1973) Ch. 21; the constraint pair contains no $\partial_{t}^{2}$ , and $\nabla^{a}G_{ab}=0$ propagates it.*

Limiting cases

Whether a given slice is Cauchy depends on the global geometry: the very same local surface can be Cauchy in one spacetime and fail in another. The table sweeps from the cleanest case to outright non-existence.

Case	Result	Meaning
Minkowski $t=0$	$\Sigma=\{t=0\}\subset\mathbb{R}^{3,1}$	A genuine Cauchy surface: every inextendible causal worldline crosses $t=0$ once; Minkowski is globally hyperbolic.
Minkowski hyperboloid	$-t^{2}+r^{2}=-1,\ t>0$	Spacelike and achronal but NOT Cauchy: $D(\Sigma)=\{t>r\}$ only, bounded by the past Cauchy horizon $H^{-}=\{t=r\}$ .
Closed FLRW	$\Sigma\cong S^{3}$ , $t=\text{const}$	Cauchy: constant-cosmic-time slices are compact Cauchy surfaces of volume $2\pi^{2}a^{3}$ ; the model is globally hyperbolic.
Reissner–Nordström interior	inner horizon $r=r_{-}$	$r_{-}$ is a Cauchy horizon: exterior slices are not Cauchy, and the maximal extension is not globally hyperbolic.
Anti-de Sitter	$\mathrm{AdS}_{n}$	No Cauchy surface exists: the timelike conformal boundary lets signals enter in finite time, so $\mathrm{AdS}$ is not globally hyperbolic.

Common trapA spacelike hypersurface is automatically a Cauchy surface.

It is not — and this is the single most common error in the subject. "Spacelike" is a purely local condition: it says only that the surface's tangent directions are all spacelike, so no two nearby points are causally related. Being Cauchy is a far stronger global condition: the slice must capture every inextendible causal history in the entire spacetime, $D(\Sigma)=M$ . The constant-proper-time hyperboloid in Minkowski space is the textbook counterexample — perfectly smooth, spacelike, and achronal, yet its domain of dependence is only the interior of the future light cone, and a past Cauchy horizon $H^{-}=\{t=r\}$ walls off everything else. The same trap appears in black-hole interiors: a $t=\text{const}$ slice inside the Reissner–Nordström or Kerr region is spacelike but sits beyond a Cauchy horizon, so inextendible causal curves can dodge it. The lesson: spacelike controls the tilt of the surface; Cauchy controls whether it sees everything. Check the domain of dependence, never just the signature.

Worked examples

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

A Reissner–Nordström black hole has mass $M$ and charge $Q$ with $Q=0.6\,M$ (geometric units $G=c=1$ ). Locate its horizons, identify which one is a Cauchy horizon, and say what data on an exterior slice can and cannot predict. Then convert to a solar-mass hole.

Step 1. The metric function is $f(r)=1-\dfrac{2M}{r}+\dfrac{Q^{2}}{r^{2}}$ , and horizons sit at $f(r)=0$ , i.e. $r^{2}-2Mr+Q^{2}=0$ , giving $r_{\pm}=M\pm\sqrt{M^{2}-Q^{2}}$ .

Step 2. With $Q=0.6\,M$ , $\sqrt{M^{2}-Q^{2}}=\sqrt{1-0.36}\,M=0.8\,M$ , so the outer (event) horizon is $r_{+}=\mathbf{1.8\,M}$ and the inner horizon is $r_{-}=\mathbf{0.2\,M}$ . Two distinct real roots — the hole is sub-extremal.

Step 3. The inner horizon $r_{-}$ is a Cauchy horizon. A spacelike slice through the exterior has a domain of dependence that covers the exterior and the region down to $r_{-}$ , but not beyond: at $r_{-}$ new information can arrive along ingoing null rays from the asymptotic region and from other asymptotic regions of the maximal extension, so the field equations lose their grip. Data on the exterior slice predict physics only up to $r_{-}=0.2\,M$ ; past it, determinism fails.

Step 4. Two limiting checks and a scale. Extremal $Q=M$ merges the roots to $r_{+}=r_{-}=M$ ; super-extremal $Q>M$ gives no real root — a naked singularity, which cosmic censorship forbids. Converting to a solar mass, the geometric length is $GM_{\odot}/c^{2}\approx 1.477\ \mathrm{km}$ , so $r_{-}=0.2\,M\approx\mathbf{0.30\ km}$ — the Cauchy horizon of a solar-mass charged hole sits about 300 m from its centre. The blueshift of infalling signals diverges at $r_{-}$ , and the strong cosmic censorship conjecture asserts a generic perturbation destroys this horizon, saving predictability.

2 · Which events does the Minkowski hyperboloid predict?

In $1+1$ Minkowski space take the future hyperboloid $\Sigma:\ -t^{2}+r^{2}=-1$ with $t>0$ (so $t=\sqrt{1+r^{2}}$ , vertex at $t=1$ , $r=0$ ). Classify the events $A=(t,r)=(3,1)$ and $B=(t,r)=(1,2)$ as inside or outside $D(\Sigma)$ , and identify the boundary.

Step 1. The domain of dependence of this hyperboloid is the interior of the future light cone, $D(\Sigma)=\{t>r\}$ : an event with $t>r$ lies above the whole surface, and every past-inextendible causal curve through it must pierce $\Sigma$ ; an event with $t\le r$ has a past-directed causal curve escaping outward to spatial infinity, missing $\Sigma$ .

Step 2. Event $A=(3,1)$ : here $t=3>r=1$ , and $A$ sits above the surface value $\sqrt{1+1^{2}}=\sqrt{2}\approx 1.41$ . So $A\in D^{+}(\Sigma)$ — A is fully predicted by data on $\Sigma$ .

Step 3. Event $B=(1,2)$ : here $t=1<r=2$ , so $B$ lies outside the future cone. A past-directed null ray through $B$ heads off to larger $r$ without ever crossing the hyperboloid. So $B\notin D(\Sigma)$ — B is not predicted by data on $\Sigma$ .

Step 4. The frontier separating the two is the null cone $t=r$ , which is the past Cauchy horizon $H^{-}(\Sigma)$ ; the future Cauchy horizon $H^{+}(\Sigma)$ is empty. Since $D(\Sigma)=\{t>r\}\neq M$ , the hyperboloid is spacelike and achronal yet not a Cauchy surface — the same conclusion the warning box draws, now pinned to specific events.

Physics of the portrait

If a Cauchy surface is where the initial data live, then the theorem that data on it determine the spacetime is the reason the whole notion matters — and that theorem has a name attached to it.

Origins · 1 minYvonne Choquet-Bruhat and how this was discovered

Yvonne Choquet-Bruhat

1923– · the well-posed Cauchy problem for Einstein's equations

Yvonne Choquet-Bruhat turned "general relativity is deterministic" from a physicist's article of faith into a proven theorem. In 1952, working in harmonic (wave) coordinates, she showed that the vacuum Einstein equations reduce to a system of quasilinear wave equations and proved local existence and uniqueness of solutions from initial data on a spacelike surface — the first rigorous demonstration that Einstein's equations pose a well-defined initial-value problem at all. The key technical move was recognizing that harmonic gauge converts the Ricci tensor into a manifestly hyperbolic operator on the metric, so the vast machinery of hyperbolic partial differential equations could be brought to bear; the four constraint equations she identified as conditions the data must satisfy on the slice, propagated forward automatically by the Bianchi identities. In 1969, with Robert Geroch, she went global: given constraint-satisfying data on a Cauchy surface, there exists a unique maximal globally hyperbolic development — a single largest spacetime for which that surface is Cauchy, into which every other development embeds. This is the precise sense in which a Cauchy surface determines its universe. Her work founded the entire mathematical-relativity industry that followed: the stability of Minkowski space, the formation of trapped surfaces, the analysis of Cauchy horizons and strong cosmic censorship all rest on it. She became the first woman elected to the French Academy of Sciences in the physical sciences, and her monograph on the Cauchy problem remains the field's standard reference.

History

1902

— Hadamard frames the well-posed initial-value problem: existence, uniqueness, and stable dependence on data

1952

— Choquet-Bruhat proves local existence and uniqueness for Einstein's equations in harmonic gauge

1962

— Arnowitt, Deser and Misner recast gravity as Hamiltonian evolution off a spatial slice

1969

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

1970

— Geroch proves a globally hyperbolic spacetime is ℝ × Σ, foliated by Cauchy surfaces

Going further

Texts. Wald, General Relativity (1984), Sec. 8.3 and Ch. 10 · Hawking & Ellis, The Large Scale Structure of Space-Time (1973), Sec. 6.5–6.6 · Misner, Thorne & Wheeler, Gravitation (1973), Ch. 21 · Carroll, Spacetime and Geometry (2004), Sec. 2.10 · Wald and Geroch, Domain of Dependence, J. Math. Phys. 11, 437 (1970) · Choquet-Bruhat & Geroch, Commun. Math. Phys. 14, 329 (1969) · Bernal & Sánchez, Commun. Math. Phys. 243, 461 (2003) and 257, 43 (2005) · Ringström, The Cauchy Problem in General Relativity (2009) · Hartle, Gravity (2003), Ch. 12 · Schutz, A First Course in General Relativity (2nd ed., 2009).

Provenance. The definition, the curve characterization, and achronality⇔"at most one crossing" follow Wald Sec. 8.3 and Hawking & Ellis Sec. 6.5–6.6. The equivalence with global hyperbolicity and the volume-ratio time function $t=\ln(f^{-}/f^{+})$ are Geroch (1970), smoothed by Bernal–Sánchez (2003, 2005). The ADM line element, extrinsic curvature, and Gauss–Codazzi constraints are standard (Wald Sec. 10.2, MTW Ch. 21); a verifier should re-derive the Hamiltonian and momentum constraints by projecting $G_{ab}=8\pi T_{ab}$ along $n^{a}$ . The maximal development is Choquet-Bruhat & Geroch (1969). Reissner–Nordström horizons $r_{\pm}=M\pm\sqrt{M^{2}-Q^{2}}$ give $r_{+}=1.8M,\ r_{-}=0.2M$ at $Q=0.6M$ ; $GM_{\odot}/c^{2}\approx1.477$ km so $r_{-}\approx0.30$ km. The hyperboloid result $D(\Sigma)=\{t>r\}$ , $H^{-}=\{t=r\}$ , $H^{+}=\varnothing$ is an elementary causal-structure computation. No numerical constants beyond exact geometric relations are asserted.

Prerequisites. Causal Structure of Spacetime · Achronal Sets and Edges · Chronological and Causal Futures and Pasts · Domain of Dependence and Cauchy Horizons. Leads to. Global Hyperbolicity · Causality Conditions Hierarchy · Penrose-Carter Conformal Diagrams · Asymptotic Flatness.

Ask the tutor: Want to walk through how Geroch's volume-ratio time function actually builds the foliation, or why the blueshift instability at a Cauchy horizon is the crux of strong cosmic censorship?

Is every spacelike hypersurface a Cauchy surface?

No, and conflating the two is the classic mistake. Spacelike is a local condition — it only says the surface's tangent directions are all spacelike, so nearby points are not causally related. Being Cauchy is a global condition: the surface must be met by every inextendible causal curve in the whole spacetime, equivalently $D(\Sigma)=M$ . The future hyperboloid in Minkowski space is the standard counterexample: it is smooth, spacelike, and achronal, yet its domain of dependence is only the interior of the future light cone, and a past Cauchy horizon walls off the rest. To test whether a spacelike slice is Cauchy you must check its domain of dependence, not merely its signature.

What is the precise relationship between Cauchy surfaces and global hyperbolicity?

They are equivalent. Geroch's theorem states that a spacetime $(M,g)$ admits a Cauchy surface if and only if it is globally hyperbolic — meaning it is causal and every causal diamond $J^{+}(p)\cap J^{-}(q)$ is compact. When this holds, far more follows: $M$ is a smooth product $\mathbb{R}\times\Sigma$ , foliated by Cauchy surfaces that are the level sets of a global time function with everywhere-timelike gradient. So "has a Cauchy surface", "is globally hyperbolic", and "splits as $\mathbb{R}\times\Sigma$ with Cauchy slices" are three faces of one condition — the exact dividing line between spacetimes that support deterministic prediction and those that do not.

Why must initial data satisfy constraints rather than being freely chosen?

Because four of Einstein's ten equations contain no second time derivative. When you project $G_{ab}=8\pi T_{ab}$ onto a spacelike slice via the Gauss and Codazzi relations, the doubly-normal projection gives the Hamiltonian constraint ${}^{(3)}\!R+K^{2}-K_{ij}K^{ij}=16\pi\rho$ and the mixed projection gives the momentum constraint $D_{j}(K^{ij}-h^{ij}K)=8\pi j^{i}$ . Neither involves $\partial_{t}K_{ij}$ , so they do not evolve anything — they restrict the allowed pairs $(h_{ij},K_{ij})$ on the surface, just as $\nabla\!\cdot\!\vec{E}=\rho$ restricts electromagnetic initial data. The remaining six equations are the genuine evolution equations, and the Bianchi identity guarantees that constraints satisfied once on $\Sigma$ stay satisfied forever.

Can the spatial topology of a Cauchy surface change during evolution?

No. Geroch's splitting $M\cong\mathbb{R}\times\Sigma$ forces every Cauchy slice of a globally hyperbolic spacetime to be diffeomorphic to the same manifold $\Sigma$ . So the spatial topology is frozen: a globally hyperbolic universe cannot start as one $\Sigma$ and evolve into a topologically different one — no handles growing, no components merging. Topology change is therefore possible only at the cost of global hyperbolicity, which means introducing pathologies such as closed timelike curves, naked singularities, or Cauchy horizons. This is a strong and sometimes surprising constraint: it says predictability and fixed spatial topology are the same package.

Why does anti-de Sitter space lack a Cauchy surface?

Anti-de Sitter space has a timelike conformal boundary that lies at finite coordinate time, and this is fatal to global hyperbolicity. A null geodesic can travel out to that boundary and, depending on the boundary condition, come back — so information can enter the bulk "from infinity" without ever crossing a candidate spacelike slice. No surface can therefore have $D(\Sigma)=M$ : there is always a causal curve that dodges it by running to the boundary. Consequently field evolution in AdS is not determined by bulk initial data alone; you must additionally impose boundary conditions at the conformal boundary. This is precisely why AdS/CFT holography treats the boundary as a dynamical player rather than a passive edge.

What exactly is a Cauchy horizon, and why does it threaten determinism?

A Cauchy horizon is the boundary of the region a slice can predict — formally $H^{\pm}(\Sigma)=\partial D^{\pm}(\Sigma)$ , the frontier of the domain of dependence. On the near side, data on $\Sigma$ fix everything; on the far side, new information can arrive along causal curves that never touched $\Sigma$ , so the field equations no longer determine the future. The Reissner–Nordström and Kerr interiors contain such horizons at their inner horizons, and infalling signals blueshift without bound as they approach, so the horizon is a surface of formally infinite energy density. The strong cosmic censorship conjecture is the proposal that this blueshift instability destroys the Cauchy horizon under any generic perturbation, so that realistic spacetimes remain globally hyperbolic and predictability is never actually lost outside idealized, exactly symmetric solutions.

Is the Big Bang a Cauchy surface?

No — the Big Bang is not a surface in spacetime at all, so it cannot be a Cauchy surface. In an FLRW model the constant-cosmic-time slices $t=\text{const}$ with $t>0$ are the Cauchy surfaces (diffeomorphic to $\mathbb{R}^{3}$ or $S^{3}$ ), and the Big Bang appears as the limit $t\to0$ , a past boundary where the scale factor vanishes and curvature diverges. It is an edge of the manifold, not a slice within it. What one does is pose data on some early Cauchy slice — the recombination surface, or a slice at the end of inflation — and evolve both forward and backward; the Big Bang is the ideal past limit of that evolution, and its status (whether it is a genuine singularity or is resolved by quantum gravity) is exactly the question of whether spacetime is extendible through it.