03.02.17 · differential-geometry / manifolds

Lorentzian Hopf-Rinow and global hyperbolicity (introductory pseudo-Riemannian geometry)

shipped3 tiersLean: none

Anchor (Master): O'Neill, Semi-Riemannian Geometry, Ch. 14; Sternberg, Curvature in Mathematics and Physics, Ch. 11; Beem-Ehrlich-Easley, Global Lorentzian Geometry, 2nd ed. (CRC, 1996), Ch. 3 and 6

Intuition Beginner

On an ordinary curved surface — the kind you measure with a ruler — there is a clean and famous fact. If you can keep walking forever in every direction without falling off an edge, then any two points are joined by a shortest path, and "shortest path forever" and "no edges" and "closed-and-bounded means you can't escape to infinity" all turn out to mean the same thing. This is the Hopf-Rinow theorem. It is the backbone of Riemannian geometry: completeness gives you geodesics that genuinely minimise distance.

Spacetime is different. In relativity the "ruler" is the metric of special relativity, and it has a minus sign in front of the time direction. That minus sign changes everything. There is no honest distance function any more: the separation between two events can be zero even when they are far apart, because a light ray connects them. So the very thing Hopf-Rinow is about — a distance you minimise — does not exist in the same form.

Two more surprises follow. First, a timelike path between two events does not get shorter as you straighten it; it gets longer. The straightest worldline between two events — a free-falling geodesic — records the most proper time, not the least. This is the geometry behind the twin paradox: the twin who travels and turns around ages less, because the bent path is shorter in proper time. Second, a spacetime can be complete — you can keep falling forever — and yet two events that ought to be causally connected may have no geodesic joining them at all.

Relativists found the right replacement for completeness. It is called global hyperbolicity. A spacetime is globally hyperbolic when its causal structure is well-behaved in two senses: you cannot travel into your own past, and the set of events that can both be influenced by one event and influence another is compact — it does not leak away to infinity or to a missing edge. On such spacetimes the good half of Hopf-Rinow comes back, in the form relativity needs.

The replacement theorem is the Avez-Seifert theorem. On a globally hyperbolic spacetime, any two events where one can causally reach the other are joined by a causal geodesic that maximises proper time. This is the genuine Lorentzian Hopf-Rinow. It is the reason fields evolve predictably from initial data, the reason the singularity theorems can locate longest worldlines that must hit a singularity, and the reason curved-spacetime quantum field theory has a place to stand.

Visual Beginner

The cleanest picture contrasts the two geometries side by side. On the left, a Riemannian surface: two points and a single curve between them, drawn short and taut, the shortest path. Straightening the curve shrinks its length. On the right, a Lorentzian diagram: time runs up, space runs sideways, and at each event a light cone opens upward. Two events are stacked so that the upper one sits inside the future cone of the lower one. Between them, several causal worldlines are drawn — but now the straight timelike geodesic is the longest in proper time, and the wiggly accelerated paths are shorter.

The inset captures the failure mode. Anti-de Sitter spacetime is drawn as a vertical strip with a wall on each side — a timelike boundary at infinity. A light ray reaches that wall in finite time. Because of the wall, the region that can both be reached from one event and reach another can run off toward the boundary and fail to be compact. So the Avez-Seifert conclusion fails there: this is exactly the spacetime where completeness is not enough and global hyperbolicity is what you actually need.

Worked example Beginner

Take flat spacetime in two dimensions with coordinates $(t, x)$ and the rule that the squared separation of a small step $(d t, d x)$ is $- d t^{2} + d x^{2}$ (light speed set to $1$ ). Compare two worldlines from the event $p = (0, 0)$ to the event $q = (2, 0)$ — same place in space, two ticks later in time.

Worldline A is the straight, motionless one: $x = 0$ , with $t$ running from $0$ to $2$ . The proper time it records is the total of $d t^{2} - d x^{2}$ along the path. With $d x = 0$ everywhere, each step contributes $d t^{2} = d t$ , so the proper time is exactly $2$ .

Worldline B goes out and comes back: from $(0, 0)$ travel to $(1, 0.5)$ , then to $(2, 0)$ . On the first leg $d t = 1$ and $d x = 0.5$ , so the squared step is $1 - 0.25 = 0.75$ and the proper time of the leg is $0.75 \approx 0.866$ . The second leg is the mirror image with the same proper time $\approx 0.866$ . The total for worldline B is about $1.73$ .

So the bent path B records about $1.73$ ticks of proper time, while the straight path A records exactly $2$ . The straight one is longer. Add more zig-zags and the proper time only shrinks further, toward zero as the path approaches the speed of light.

What this tells us: in Lorentzian geometry the geodesic does not minimise — it maximises proper time among nearby causal worldlines. The number $2$ is the largest, not the smallest. A Riemannian intuition that "the geodesic is the short way" is exactly inverted. This single sign flip is the reason Hopf-Rinow cannot be copied over from Riemannian geometry without being rebuilt, and it is the content the Avez-Seifert theorem restores in the correct direction.

Check your understanding Beginner

Exercise (easy, multiple choice).

In the Riemannian Hopf-Rinow theorem, geodesic completeness is equivalent to several other conditions. Which one is among them?

A. Every geodesic is a straight line B. Closed and bounded subsets are compact, and any two points are joined by a minimising geodesic C. The curvature is constant D. The space is flat

Hint

Hopf-Rinow ties together completeness as a metric space, geodesic completeness, and a Heine-Borel-type property, and then concludes that minimising geodesics exist between any two points.

Answer

B. Closed and bounded subsets are compact, and any two points are joined by a minimising geodesic.

The Riemannian Hopf-Rinow theorem states the equivalence of geodesic completeness, metric completeness, and the Heine-Borel property (closed bounded sets are compact), and from any of these draws the conclusion that any two points are joined by a length-minimising geodesic. Options A, C, and D describe special metrics, not the general completeness statement.

Formal definition Intermediate+

A Lorentzian manifold is a pair $(M, g)$ with $M$ a smooth connected $n$ -manifold ( $n \geq 2$ ) and $g$ a smooth symmetric non-degenerate $(0, 2)$ -tensor field of signature $(-, +, \dots, +)$ . This unit fixes the mostly-plus convention, following O'Neill and Sternberg ^{[Sternberg Ch. 11]}, and states it explicitly because the opposite mostly-minus convention flips the sign of $g (v, v)$ throughout. The Levi-Civita connection $\nabla$ and the Riemann curvature are defined exactly as in the Riemannian case; the formulas do not see the signature.

A non-zero $v \in T_{p} M$ is timelike if $g (v, v) < 0$ , **null** if $g (v, v) = 0$ , **spacelike** if $g (v, v) > 0$ , and causal if timelike or null. A piecewise-smooth curve is timelike (null, causal, spacelike) when its velocity is so at every regular point. A time orientation is a continuous choice of one of the two components of the cone of timelike vectors at each point, equivalently a smooth nowhere-zero timelike vector field $T$ ; the chosen component is future-directed. All spacetimes here are time-oriented.

For a future-directed causal curve $γ : [a, b] \to M$ the proper time (Lorentzian arc length) is $$ L(\gamma) = \int_a^b \sqrt{-,g(\dot\gamma, \dot\gamma)}; ds, $$ defined because $g (\overset{γ}{˙}, \overset{γ}{˙}) \leq 0$ along a causal curve. The Lorentzian distance (time separation) is $$ d(p,q) = \sup{, L(\gamma) : \gamma \text{ a future-directed causal curve from } p \text{ to } q ,}, $$ with $d (p, q) = 0$ when no such curve exists. This is a supremum, not an infimum, and it is not a metric: it fails symmetry, satisfies a reversed triangle inequality $d (p, r) \geq d (p, q) + d (q, r)$ for causally ordered $p \leq q \leq r$ , and can be infinite or zero between distinct points.

The chronological future is $I^{+} (p) = {q : \exists future-directed timelike curve p \to q}$ and the causal future is $J^{+} (p) = {q : \exists future-directed causal curve p \to q} \cup {p}$ ; the pasts $I^{-}, J^{-}$ reverse the time orientation. One writes $p ≪ q$ for $q \in I^{+} (p)$ (chronological precedence) and $p \leq q$ for $q \in J^{+} (p)$ (causal precedence). The set $I^{+} (p)$ is open; $J^{+} (p)$ need not be closed.

A spacetime is strongly causal at $p$ if every neighbourhood of $p$ contains a smaller causally convex neighbourhood — one no causal curve re-enters after leaving. It is strongly causal if this holds everywhere; this forbids closed and almost-closed causal curves.

Definition (geodesic completeness). $(M, g)$ is geodesically complete if every geodesic extends to all real values of its affine parameter. One distinguishes timelike, null, and spacelike completeness; in indefinite signature these are independent (a spacetime can be null-complete but timelike-incomplete).

Definition (global hyperbolicity). A time-oriented Lorentzian manifold $(M, g)$ is globally hyperbolic if it is strongly causal and, for every $p, q \in M$ , the causal diamond $J^{+} (p) \cap J^{-} (q)$ is compact.

A Cauchy hypersurface is a subset $Σ \subset M$ met exactly once by every inextendible timelike curve. Existence of a Cauchy hypersurface is equivalent to global hyperbolicity (Geroch).

Counterexamples to common slips

Completeness does not imply global hyperbolicity. Anti-de Sitter spacetime is geodesically complete (every geodesic extends for all affine parameter) yet not globally hyperbolic, because null geodesics reach the timelike conformal boundary in finite affine parameter and the causal diamonds fail to be compact.
Global hyperbolicity does not imply geodesic completeness. The Schwarzschild exterior is globally hyperbolic, but radial timelike geodesics reach the horizon in finite proper time and leave the region. The two properties are logically independent in Lorentzian signature, unlike the tight Riemannian equivalence.
Causally related points need not be joined by a geodesic in a merely complete spacetime. Geroch's example of a complete spacetime with a point removed from its causal interior shows two timelike-related events with no causal geodesic between them. Completeness alone is not the Lorentzian Hopf-Rinow hypothesis; global hyperbolicity is.
The Lorentzian distance is upper-semicontinuous, not continuous. Unlike the Riemannian distance, $d (p, \cdot)$ can jump downward; it is continuous precisely on globally hyperbolic spacetimes, which is one reason that hypothesis is the right one.

Key theorem with proof Intermediate+

The Riemannian Hopf-Rinow theorem and its Lorentzian replacement are best stated together so the contrast is exact.

Theorem (Riemannian Hopf-Rinow, 1931). Let $(N, h)$ be a connected Riemannian manifold with distance function $ρ$ . The following are equivalent: (i) $(N, ρ)$ is complete as a metric space; (ii) $(N, h)$ is geodesically complete; (iii) closed and $ρ$ -bounded subsets of $N$ are compact. Any of these implies that any two points of $N$ are joined by a minimising geodesic ^{[Hopf-Rinow 1931]}.

This theorem is the engine of Riemannian global geometry ^{[O'Neill Ch. 5]}. Its proof rests on the fact that the Riemannian distance is a genuine metric: balls are the right compact objects, and minimising sequences of curves have uniformly bounded length, so Arzelà-Ascoli extracts a convergent limit curve.

In Lorentzian signature each pillar of this proof collapses. The Lorentzian distance $d$ is a supremum and not a metric; there is no Heine-Borel notion driven by $d$ ; and timelike geodesics maximise proper time, so "minimising sequence" must become "maximising sequence." The correct replacement keeps the conclusion — existence of a geodesic realising the extremal arc length — but changes the hypothesis from completeness to global hyperbolicity.

Theorem (Avez-Seifert; the Lorentzian Hopf-Rinow). Let $(M, g)$ be a globally hyperbolic spacetime and let $p \leq q$ with $p \neq = q$ . Then there exists a future-directed causal geodesic $γ$ from $p$ to $q$ whose proper time equals $d (p, q)$ , i.e. $γ$ maximises proper time among all future-directed causal curves from $p$ to $q$ . The function $d$ is finite and continuous on $M \times M$ ^{[Avez 1963]} ^{[Seifert 1967]}.

Proof. Fix $p \leq q$ . The space of future-directed causal curves from $p$ to $q$ , taken up to reparametrisation, carries the $C^{0}$ topology of uniform convergence. Each such curve lies in the causal diamond $J^{+} (p) \cap J^{-} (q)$ , which is compact by global hyperbolicity. This compactness is what replaces the Riemannian bounded-length step.

Take a sequence $γ_{k}$ of future-directed causal curves from $p$ to $q$ with $L (γ_{k}) \to d (p, q)$ (a maximising sequence; $d (p, q) < \infty$ because the diamond is compact and $L$ is bounded on it). Parametrise each $γ_{k}$ by an auxiliary complete Riemannian arc length so the curves are uniformly Lipschitz and confined to the compact diamond. By the limit curve lemma — the Lorentzian replacement for Arzelà-Ascoli, valid on a strongly causal spacetime — a subsequence converges uniformly to a future-directed causal limit curve $γ_{\infty}$ from $p$ to $q$ ^{[O'Neill Ch. 14]} ^{[Hawking-Ellis 1973]}.

Lorentzian arc length is upper semicontinuous under $C^{0}$ convergence of causal curves: a small perturbation can only shorten proper time, never lengthen it beyond the limit. Hence $$ L(\gamma_\infty) ;\ge; \limsup_k L(\gamma_k) ;=; d(p,q). $$ Since $γ_{\infty}$ is itself a causal curve from $p$ to $q$ , the reverse inequality $L (γ_{\infty}) \leq d (p, q)$ holds by definition of the supremum. Therefore $L (γ_{\infty}) = d (p, q)$ : the limit curve realises the time separation.

A causal curve that realises the time separation is a geodesic. If $γ_{\infty}$ had a corner or failed the geodesic equation on any subinterval, a fixed-endpoint variation in the timelike interior would increase proper time — by the first-variation formula, whose Lorentzian sign is opposite to the Riemannian one — contradicting maximality. Where $γ_{\infty}$ is timelike it is a timelike geodesic; a maximal causal curve that is null on a subinterval is a null geodesic with no conjugate points before $q$ . Reparametrising by proper time (timelike case) or an affine parameter (null case) gives the asserted geodesic.

Finiteness and continuity of $d$ follow from the same compactness: $d (p, q) < \infty$ because $L$ is bounded on the compact diamond, and continuity is upper semicontinuity (always available) together with lower semicontinuity, which global hyperbolicity supplies through compactness of the diamonds varying continuously with their endpoints. $■$

Bridge. The Avez-Seifert theorem builds toward 13.09.01, where global hyperbolicity becomes the standing hypothesis under which the Klein-Gordon Cauchy problem is well-posed, and appears again in 13.02.02, where the maximising property of timelike geodesics is sharpened into the second-variation index form that the singularity theorems exploit. The mechanism behind the bridge is the substitution of one compactness source for another: the Riemannian theorem draws compactness from bounded-and-closed balls in a genuine metric, while the Lorentzian theorem draws it from compactness of causal diamonds, a global causal hypothesis with no metric-space analogue. This same compactness underlies the diffeomorphic splitting $M ≅ R \times Σ$ of 13.09.01 and the finite-speed-of-propagation Green's functions of the wave operator, so the geometric statement proved here is the entry point to the entire analytic theory of fields on curved backgrounds, and it recurs whenever a longest-worldline or maximal-development argument is needed in mathematical relativity.

Exercises Intermediate+

Exercise 3 (medium, symbolic).

Prove that the Lorentzian distance satisfies the reverse triangle inequality: if $p \leq q \leq r$ then $d (p, r) \geq d (p, q) + d (q, r)$ .

Hint

Concatenate a near-maximal causal curve from $p$ to $q$ with one from $q$ to $r$ , and compare its length to the supremum defining $d (p, r)$ .

Answer

Let $ε > 0$ . By definition of the supremum there are future-directed causal curves $α$ from $p$ to $q$ with $L (α) > d (p, q) - ε /2$ and $β$ from $q$ to $r$ with $L (β) > d (q, r) - ε /2$ . The concatenation $α * β$ is a future-directed causal curve from $p$ to $r$ (piecewise smooth, causal on each piece), and $L (α * β) = L (α) + L (β) > d (p, q) + d (q, r) - ε$ . Since $d (p, r)$ is the supremum over all such curves, $d (p, r) \geq L (α * β) > d (p, q) + d (q, r) - ε$ . As $ε > 0$ was arbitrary, $d (p, r) \geq d (p, q) + d (q, r)$ . The inequality direction is reversed from the Riemannian case because $d$ is a supremum of lengths, not an infimum of distances.

Exercise 4 (medium, symbolic).

Show that the chronological future $I^{+} (p)$ is open, while the causal future $J^{+} (p)$ need not be closed. Give the standard example for the second claim.

Hint

Timelike-ness is an open condition on tangent vectors. For non-closedness, delete a point from Minkowski space and look at a boundary null ray.

Answer

If $q \in I^{+} (p)$ there is a future-directed timelike curve $γ$ from $p$ to $q$ . In a chart near $q$ the future-timelike cone is an open condition on velocity, so the endpoint of a small perturbation of the final segment of $γ$ still receives a future-directed timelike curve from $p$ ; a neighbourhood of $q$ lies in $I^{+} (p)$ , so $I^{+} (p)$ is open. For $J^{+} (p)$ : in Minkowski space $R^{1, 2}$ remove the point $z = (1, 1, 0)$ , which lies on the boundary of the light cone of $p = (0, 0, 0)$ . The null ray through $z$ now has its portion beyond $z$ missing from $J^{+} (p)$ , but points just inside the cone past $z$ are still in $J^{+} (p)$ and accumulate on the missing null ray. So $J^{+} (p)$ is not closed. Intersecting with $J^{-} (q)$ under global hyperbolicity restores compactness, hence closedness of the diamond.

Exercise 5 (medium, symbolic).

Verify directly in Minkowski space $R^{1, 1}$ that the straight timelike geodesic from $p = (0, 0)$ to $q = (T, 0)$ , $T > 0$ , maximises proper time, by computing $L$ for the family of broken worldlines through $(T /2, a)$ and maximising over $a$ .

Hint

Each leg has $Δ t = T /2$ and $∣Δ x ∣ = ∣ a ∣$ . Write $L (a)$ and find its maximum over $a \in (- T /2, T /2)$ .

Answer

The broken worldline through $(T /2, a)$ has two legs each with $Δ t = T /2$ , $∣Δ x ∣ = ∣ a ∣$ , so $$ L(a) = 2\sqrt{(T/2)^2 - a^2} = \sqrt{T^2 - 4a^2}. $$ This is defined and causal for $∣ a ∣ < T /2$ , and $L (a)$ is strictly decreasing in $∣ a ∣$ , with maximum at $a = 0$ where $L (0) = T$ . The maximiser $a = 0$ is exactly the straight geodesic, and its proper time $T$ equals $d (p, q)$ . Any deflection $a \neq = 0$ strictly decreases proper time, confirming the geodesic maximises.

Exercise 6 (hard, symbolic).

Anti-de Sitter $AdS_{2}$ in conformal coordinates has metric $g = sec^{2} x (- d t^{2} + d x^{2})$ for $x \in (- π /2, π /2)$ . Show it is geodesically complete for timelike geodesics yet not globally hyperbolic, and explain why Avez-Seifert can fail for an interior point and a point in its causal future near the boundary.

Hint

Null geodesics follow $d t = \pm d x$ and reach $x = \pm π /2$ in finite affine parameter (the integral $\int cos^{2} x d x$ converges). Use this to argue the causal diamond can be non-compact.

Answer

Timelike geodesics in $AdS_{2}$ oscillate in $x$ and never reach the boundary; reparametrised by proper time they extend for all parameter values, so the spacetime is timelike-geodesically complete. A right-moving null geodesic follows $t = x + const$ , and its affine parameter is $λ \propto \int^{x} cos^{2} (x^{'}) d x^{'}$ , which converges as $x \to π / 2^{-}$ : the null geodesic reaches the conformal boundary in finite affine parameter. Because causal curves can run out to the timelike boundary, for an interior point $p$ and a point $q$ to its future placed so that $J^{+} (p) \cap J^{-} (q)$ stretches toward the boundary, the diamond is not compact — global hyperbolicity fails. The maximising-sequence argument of Avez-Seifert then has no compact arena: a maximising sequence of causal curves can escape toward the boundary with no limit curve in $M$ , so a maximal geodesic between such $p$ and $q$ need not exist. Field theory on $AdS$ correspondingly requires boundary conditions imposed by hand.

Exercise 7 (hard, symbolic).

Show that a Lorentzian manifold admitting a closed timelike curve cannot be globally hyperbolic, and that on such a spacetime the Lorentzian distance $d$ is identically $+ \infty$ on a chronology-violating region.

Hint

Global hyperbolicity requires strong causality, which forbids closed causal curves. For the distance, traverse the closed timelike curve repeatedly.

Answer

A globally hyperbolic spacetime is strongly causal by definition, and strong causality forbids closed (and almost-closed) causal curves. A closed timelike curve is a closed causal curve, so its presence violates strong causality and hence global hyperbolicity. For the distance: let $γ$ be a closed timelike curve through $p$ with proper time $τ_{0} > 0$ per loop. Then for any positive integer $N$ , traversing $γ$ $N$ times is a future-directed timelike curve from $p$ to $p$ of proper time $N τ_{0}$ . Since $d (p, p)$ is the supremum of proper times of such curves, $d (p, p) \geq N τ_{0}$ for all $N$ , so $d (p, p) = + \infty$ . The same holds for any point in the chronology-violating region. Gödel's rotating cosmology is the classical example carrying such curves.

Lean formalization Intermediate+

Mathlib has the Riemannian half of this unit. Geodesics as length-minimisers, the exponential map, metric-space completeness, and a Hopf-Rinow statement (via ProperSpace / geodesic_space infrastructure on a complete connected Riemannian manifold) are present. The Lorentzian half is entirely absent: there is no indefinite-signature metric structure, no time-orientation line bundle, no causal classification of tangent vectors, no $I^{\pm}$ / $J^{\pm}$ , no causality conditions, and no length-maximising causal geodesic — the Lorentzian arc length being a supremum makes the metric-space scaffold that powers Mathlib's Riemannian Hopf-Rinow inapplicable.

lean_status: none records this gap; no Lean module ships. the Mathlib gap analysis names the missing infrastructure, of which the indefinite-signature metric plus the time-orientation bundle is load-bearing; the limit-curve compactness argument that replaces Arzelà-Ascoli would build on it. Tyler attests intermediate-tier correctness; the master-tier index-form and the Avez-Seifert limit-curve argument are flagged for external differential-geometry review.

Advanced results Master

Three developments take the Avez-Seifert theorem to the depth O'Neill Ch. 14 and Beem-Ehrlich-Easley require, and locate it within the causal-completeness theory of Lorentzian manifolds.

The index form and the maximality of timelike geodesics. Along a timelike geodesic $γ$ parametrised by proper time, the second variation of the proper-time functional under a fixed-endpoint variation with orthogonal variation field $V$ is governed by the Lorentzian index form $$ I(V, V) = \int \big( g(V', V') - g(R(V, \dot\gamma)\dot\gamma,, V) \big), ds, $$ where $V^{'}$ denotes covariant differentiation and $R$ the curvature ^{[O'Neill Ch. 14]}. The sign placement differs from the Riemannian case precisely because timelike geodesics maximise: $γ$ is locally maximising exactly when $I$ is negative semidefinite on orthogonal variation fields vanishing at the endpoints, and the failure of maximality past the first conjugate point is the timelike Morse-index theorem. A causal geodesic with a conjugate point in its interior is no longer maximising; the Avez-Seifert maximal geodesic therefore has no conjugate point before its endpoint, a fact the Hawking-Penrose singularity theorems convert into a contradiction when curvature focusing forces a conjugate point too early, proving timelike geodesic incompleteness.

The causal ladder and where global hyperbolicity sits. Causality conditions on a Lorentzian manifold form a strictly nested ladder ^{[Hawking-Ellis 1973]}: chronology (no closed timelike curves) $\supset$ causality (no closed causal curves) $\supset$ distinguishing $\supset$ strong causality $\supset$ stable causality (existence of a global time function) $\supset$ global hyperbolicity, the top of the ladder. Each strict inclusion is realised by an explicit spacetime: Gödel violates chronology; Reissner-Nordström past the Cauchy horizon violates strong causality; anti-de Sitter and punctured Minkowski are stably causal without being globally hyperbolic. Sánchez and Bernal proved that "strongly causal" may be relaxed to "causal" in the definition of global hyperbolicity without enlarging the class — the compactness of causal diamonds promotes causality to strong causality. The position at the top of the ladder is exactly what the Avez-Seifert proof needs: stable causality supplies the limit-curve topology, and diamond-compactness supplies the maximising arena.

Finiteness, continuity, and the Geroch splitting. On a globally hyperbolic spacetime the Lorentzian distance $d$ is finite and continuous, in contrast to its upper-semicontinuous-only behaviour in general ^{[O'Neill Ch. 14]}. Geroch's theorem turns global hyperbolicity into a Cauchy hypersurface and a topological product $M \approx R \times Σ$ ^{[Geroch 1970]}; Bernal and Sánchez upgraded this to a smooth diffeomorphism with the metric splitting $g = - β^{2} d t^{2} + h_{t}$ , each slice a smooth spacelike Cauchy hypersurface ^{[Bernal-Sánchez 2005]}. The continuity of $d$ and the smooth splitting are the two analytic dividends of the same compactness hypothesis: continuity makes the supremum in Avez-Seifert attained, and the splitting makes the wave operator strictly hyperbolic in adapted coordinates. The maximal geodesic of Avez-Seifert and the foliating Cauchy hypersurfaces of Geroch-Bernal-Sánchez are two faces of the single geometric fact that compact causal diamonds tame the indefinite-signature geometry.

Synthesis. The Riemannian Hopf-Rinow theorem and the Lorentzian Avez-Seifert theorem are the same theorem read through opposite signatures, with completeness replaced by global hyperbolicity and minimisation replaced by maximisation. This builds toward 13.09.01, where global hyperbolicity is the standing hypothesis of curved-spacetime field theory, and the continuity of $d$ proved here is what makes the Cauchy problem's domains of dependence behave. It appears again in 13.02.02, where the index form converts the maximising property into the focusing estimates of the singularity theorems, and in 03.02.06, where the Riemannian Hopf-Rinow apparatus this unit contrasts against supplies the Cartan-Hadamard and Bonnet-Myers comparison results for definite signature. The deciding structural difference is that Riemannian compactness flows from a genuine distance metric while Lorentzian compactness must be imposed as a causal hypothesis; once imposed, the maximising geodesic, the continuous time separation, and the Geroch-Bernal-Sánchez splitting all follow, and the indefinite-signature geometry acquires the predictive content that flat special relativity had by fiat. The pattern propagates through the singularity theorems, the positive-mass and Penrose-inequality programmes, and the cosmic-censorship conjecture, every one of which is a statement about whether physically realistic spacetimes stay within the globally hyperbolic class where this theorem holds.

Full proof set Master

Proposition (upper semicontinuity of Lorentzian arc length). Let $(M, g)$ be a strongly causal spacetime and let $γ_{k} \to γ_{\infty}$ be a $C^{0}$ -convergent sequence of future-directed causal curves with common endpoints $p, q$ , each contained in a fixed compact set $K \subset M$ . Then $L (γ_{\infty}) \geq lim sup_{k} L (γ_{k})$ .

Proof. Cover $K$ by finitely many convex normal neighbourhoods $U_{1}, \dots, U_{m}$ on each of which the Lorentzian distance $d_{U_{i}}$ of the convex chart is continuous and, for causally related points within $U_{i}$ , realised by the unique connecting geodesic segment. Subdivide each $γ_{k}$ and $γ_{\infty}$ at a common finite set of parameter values $0 = s_{0} < \dots < s_{N} = 1$ chosen so that for large $k$ each subarc $γ_{k} ∣_{[s_{j - 1}, s_{j}]}$ and $γ_{\infty} ∣_{[s_{j - 1}, s_{j}]}$ lies in a single $U_{i (j)}$ . On a convex neighbourhood the proper time of a causal curve between two points is at most the convex distance $d_{U}$ of its endpoints, with equality for the geodesic; thus $L (γ_{k} ∣_{[s_{j - 1}, s_{j}]}) \leq d_{U_{i (j)}} (γ_{k} (s_{j - 1}), γ_{k} (s_{j}))$ . Summing over $j$ , taking $k \to \infty$ , and using continuity of each $d_{U_{i (j)}}$ together with $γ_{k} \to γ_{\infty}$ uniformly, $$ \limsup_k L(\gamma_k) ;\le; \sum_j d_{U_{i(j)}}(\gamma_\infty(s_{j-1}), \gamma_\infty(s_j)) ;=; \sum_j L(\gamma_\infty|{[s{j-1},s_j]}) ;=; L(\gamma_\infty), $$ where the middle equality holds because $γ_{\infty}$ restricted to each subarc is itself a causal curve in $U_{i (j)}$ whose length is bounded by the convex distance, and the global subdivision can be refined so the bound is saturated in the limit. This is the semicontinuity claim. $□$

Proposition (limit curve lemma on a globally hyperbolic spacetime). Let $(M, g)$ be globally hyperbolic and $p \leq q$ . Any sequence of future-directed causal curves from $p$ to $q$ has a subsequence converging uniformly to a future-directed causal curve from $p$ to $q$ .

Proof. Each curve lies in the compact diamond $D = J^{+} (p) \cap J^{-} (q)$ . Equip $M$ with an auxiliary complete Riemannian metric $h$ and reparametrise every curve by its $h$ -arc length scaled to $[0, 1]$ ; the reparametrised curves are $1$ -Lipschitz into the compact set $D$ for the $h$ -distance, after normalising total $h$ -length, and have a uniform $h$ -length bound because $D$ is compact and causal curves in a strongly causal spacetime have $h$ -length bounded by a constant times the $h$ -diameter of $D$ . By the Arzelà-Ascoli theorem for maps into the compact metric space $(D, d_{h})$ , a subsequence converges uniformly to a Lipschitz curve $γ_{\infty} : [0, 1] \to D$ with $γ_{\infty} (0) = p$ , $γ_{\infty} (1) = q$ . That $γ_{\infty}$ is future-directed causal follows from the closedness of the causal cone field: at almost every parameter the velocity is a limit of future-directed causal velocities of the $γ_{k}$ , hence future-directed causal. $□$

Proposition (a maximal causal curve is a geodesic). Let $(M, g)$ be a spacetime and $γ$ a future-directed causal curve from $p$ to $q$ with $L (γ) = d (p, q)$ . Then $γ$ is, up to reparametrisation, an unbroken causal geodesic.

Proof. Suppose $γ$ is not a geodesic on some subinterval, or has a corner at an interior point $r$ . Work in a convex normal neighbourhood $U$ of $r$ . If $γ$ is timelike near $r$ , the proper time between two points of $U$ joined by a causal curve is strictly less than the convex distance $d_{U}$ unless the curve is the connecting geodesic; replacing the subarc of $γ$ in $U$ by that geodesic strictly increases proper time while keeping the curve causal with the same endpoints, contradicting $L (γ) = d (p, q)$ . If $γ$ is null near $r$ and not a null geodesic, then it can be deformed to a timelike curve between the same endpoints with strictly positive proper time, again contradicting maximality among causal curves once embedded in the global comparison. Hence $γ$ has no corners and satisfies the geodesic equation on every subinterval, so it is an unbroken causal geodesic; reparametrising by proper time (timelike case) or an affine parameter (null case) completes the claim. $□$

These three propositions assemble into the Avez-Seifert theorem: the limit curve lemma extracts $γ_{\infty}$ from a maximising sequence, upper semicontinuity forces $L (γ_{\infty}) = d (p, q)$ , and the maximal-implies-geodesic proposition identifies $γ_{\infty}$ as the asserted maximising geodesic.

Connections Master

Sectional, Ricci, scalar curvature 03.02.05 supplies the curvature operator $R$ that enters the Lorentzian index form $I (V, V) = \int (g (V^{'}, V^{'}) - g (R (V, \overset{γ}{˙}) \overset{γ}{˙}, V)) d s$ . The same Riemann tensor that measures geodesic deviation in Riemannian signature controls, with the opposite variational sign, the focusing of timelike geodesics that the singularity theorems exploit; the curvature definitions are signature-blind even though their geometric consequences invert.
Constant-curvature spaces and Killing-Hopf 03.02.06 carries the Riemannian Hopf-Rinow theorem against which this unit is the deliberate Lorentzian foil: there completeness yields minimising geodesics through Cartan-Hadamard and Bonnet-Myers comparison, here global hyperbolicity yields maximising geodesics through Avez-Seifert. The contrast between the definite-signature equivalence (completeness ⇔ Heine-Borel ⇔ minimising connectivity) and its indefinite-signature breakdown is the conceptual content of the present unit.
Globally hyperbolic Lorentzian manifolds 13.09.01 is the downstream GR-cosmology unit that takes global hyperbolicity as a standing hypothesis and develops the Geroch and Bernal-Sánchez splitting in full, then uses it to pose the Klein-Gordon Cauchy problem and the Hadamard-state programme. The Avez-Seifert maximal-geodesic theorem proved here is the differential-geometric core that unit assumes; the continuity of the Lorentzian distance established here is what makes its domains of dependence behave.
Geodesics and parallel transport 13.02.02 provides the geodesic equation and the first- and second-variation formulae specialised to spacetime; this unit supplies the sign reversal that makes timelike geodesics maximise proper time, and the index-form analysis that the singularity-theorem focusing arguments build on. The proper-time maximisation is the variational statement that the twin paradox makes concrete.

Historical & philosophical context Master

The Riemannian completeness theorem is due to Heinz Hopf and Willi Rinow, whose 1931 Commentarii Mathematici Helvetici paper ^{[Hopf-Rinow 1931]} established the equivalence of metric completeness, geodesic completeness, and the Heine-Borel property on a connected Riemannian manifold, with the corollary that any two points are joined by a minimising geodesic. The theorem became the standard hypothesis of global Riemannian geometry.

The Lorentzian analogue developed two decades later, driven by general relativity rather than pure geometry. Jean Leray's 1953 mimeographed Princeton lecture notes ^{[Leray 1953]} introduced the notion later named global hyperbolicity in the study of hyperbolic partial differential equations on a Lorentzian manifold. André Avez, in a 1963 Annales de l'Institut Fourier paper ^{[Avez 1963]}, proved that on a globally hyperbolic spacetime any two causally related points are joined by a maximising causal geodesic — the genuine Lorentzian Hopf-Rinow. Hans-Jürgen Seifert gave an independent proof of the connectivity statement in 1967 ^{[Seifert 1967]}, and the result is now standard as the Avez-Seifert theorem. Robert Geroch's 1970 Journal of Mathematical Physics paper ^{[Geroch 1970]} recast global hyperbolicity as the existence of a Cauchy hypersurface and a topological splitting, and Stephen Hawking and George Ellis's 1973 monograph ^{[Hawking-Ellis 1973]} consolidated the causal-structure theory and used the maximal-geodesic existence to prove the singularity theorems.

The textbook synthesis is due to Barrett O'Neill, whose 1983 Semi-Riemannian Geometry gave the unified pseudo-Riemannian treatment in which the Riemannian and Lorentzian theorems sit side by side, and to Shlomo Sternberg's Curvature in Mathematics and Physics, Ch. 11 ^{[Sternberg Ch. 11]}, which presents the breakdown of Hopf-Rinow and the causal structure for a mathematically literate physics audience. The smoothness of the Geroch splitting remained an assumption until Antonio Bernal and Miguel Sánchez proved it a theorem in 2003 and 2005 ^{[Bernal-Sánchez 2005]}, producing the smooth metric splitting $g = - β^{2} d t^{2} + h_{t}$ that the modern curved-spacetime field-theory literature relies on.

Bibliography Master

Primary literature:

Hopf, H. & Rinow, W., "Ueber den Begriff der vollständigen differentialgeometrischen Fläche", Comment. Math. Helv. 3 (1931), 209-225. [The Riemannian completeness theorem.]
Leray, J., Hyperbolic differential equations, Institute for Advanced Study, Princeton (1953, mimeographed lecture notes). [Originating notion of global hyperbolicity.]
Avez, A., "Essais de géométrie riemannienne hyperbolique globale. Applications à la relativité générale", Ann. Inst. Fourier (Grenoble) 13 (1963), 105-190. [Maximising causal geodesic on a globally hyperbolic spacetime.]
Seifert, H.-J., "Global connectivity by timelike geodesics", Z. Naturforsch. 22a (1967), 1356-1360. [Independent proof of the Avez-Seifert connectivity theorem.]
Geroch, R., "Domain of dependence", J. Math. Phys. 11 (1970), 437-449. [Global hyperbolicity ⇔ Cauchy surface and topological splitting.]
Bernal, A. N. & Sánchez, M., "Smoothness of time functions and the metric splitting of globally hyperbolic spacetimes", Comm. Math. Phys. 257 (2005), 43-50. [Smooth metric splitting $g = - β^{2} d t^{2} + h_{t}$ .]
Gödel, K., "An example of a new type of cosmological solutions of Einstein's field equations of gravitation", Rev. Mod. Phys. 21 (1949), 447-450. [Closed-timelike-curve cosmological solution; chronology-violating counterexample.]

Singularity-theorem literature:

Penrose, R., "Gravitational collapse and space-time singularities", Phys. Rev. Lett. 14 (1965), 57-59.
Hawking, S. W. & Penrose, R., "The singularities of gravitational collapse and cosmology", Proc. R. Soc. A 314 (1970), 529-548.

Modern monographs and textbooks:

O'Neill, B., Semi-Riemannian Geometry with Applications to Relativity (Academic Press, 1983). [Unified pseudo-Riemannian treatment; Ch. 5 completeness, Ch. 14 causality and the Avez-Seifert theorem.]
Sternberg, S., Curvature in Mathematics and Physics (Dover, 2012). [Ch. 11 develops Lorentzian geometry and the breakdown of Hopf-Rinow for a physics-literate audience.]
Hawking, S. W. & Ellis, G. F. R., The Large Scale Structure of Space-Time (Cambridge University Press, 1973). [Causal structure and the singularity theorems; Ch. 6.]
Beem, J. K., Ehrlich, P. E. & Easley, K. L., Global Lorentzian Geometry, 2nd ed. (CRC Press, 1996). [Comprehensive monograph on Lorentzian completeness, the limit curve theorem, and the Avez-Seifert theory; Ch. 3 and 6.]
Wald, R. M., General Relativity (University of Chicago Press, 1984). [Ch. 8 covers causal structure and global hyperbolicity at the modern level.]
Minguzzi, E. & Sánchez, M., "The causal hierarchy of spacetimes", in Recent Developments in Pseudo-Riemannian Geometry (eds. D. V. Alekseevsky and H. Baum, EMS, 2008), 299-358. [Comprehensive modern review of causality conditions.]

Prerequisites

03.02.05
03.02.06

Tier anchors

beginner: Schutz, A First Course in General Relativity, Ch. 2-3 (light cones, causal structure); Penrose, The Road to Reality, Ch. 17
intermediate: O'Neill, Semi-Riemannian Geometry with Applications to Relativity (Academic, 1983), Ch. 5 (geodesics, completeness) and Ch. 14 (causality); Sternberg, Curvature in Mathematics and Physics (Dover, 2012), Ch. 11
master: O'Neill, Semi-Riemannian Geometry, Ch. 14; Sternberg, Curvature in Mathematics and Physics, Ch. 11; Beem-Ehrlich-Easley, Global Lorentzian Geometry, 2nd ed. (CRC, 1996), Ch. 3 and 6

References

Sternberg, S. — Curvature in Mathematics and Physics (Dover, 2012) · Ch. 11 (Lorentzian geometry; the breakdown of Hopf-Rinow and the causal structure of spacetime)
O'Neill, B. — Semi-Riemannian Geometry with Applications to Relativity (Academic Press, 1983) · Ch. 5 (geodesics and completeness), Ch. 6 (special relativity), Ch. 14 (causality in Lorentz manifolds; the Avez-Seifert theorem, Thm. 14.20-14.22)
Hopf, H. & Rinow, W. — 'Ueber den Begriff der vollständigen differentialgeometrischen Fläche' · Comment. Math. Helv. 3 (1931) 209-225 (the Riemannian completeness theorem)
Geroch, R. — 'Domain of dependence', J. Math. Phys. 11 (1970) 437 · Theorem 11 (global hyperbolicity ⇔ existence of a Cauchy surface; the topological splitting)
Avez, A. — 'Essais de géométrie riemannienne hyperbolique globale. Applications à la relativité générale' · Ann. Inst. Fourier 13 (1963) 105-190 (maximising causal geodesic in a globally hyperbolic spacetime)
Seifert, H.-J. — 'Global connectivity by timelike geodesics' · Z. Naturforsch. 22a (1967) 1356-1360 (independent proof of the Avez-Seifert connectivity theorem)
Bernal, A. N. & Sánchez, M. — 'Smoothness of time functions and the metric splitting of globally hyperbolic spacetimes', Comm. Math. Phys. 257 (2005) 43 · Theorem 1.1 (smooth metric splitting M ≅ ℝ × Σ with g = -β² dt² + h_t; diffeomorphic refinement of Geroch)
Hawking, S. W. & Ellis, G. F. R. — The Large Scale Structure of Space-Time (Cambridge, 1973) · Ch. 6 (causal structure); §6.7 (the existence of maximal geodesics and global hyperbolicity)
Leray, J. — 'Hyperbolic differential equations', Institute for Advanced Study, Princeton (1953, mimeographed lecture notes) · Originating notion of global hyperbolicity for hyperbolic systems on a Lorentzian manifold

Estimated time

beginner: 18m
intermediate: 42m
master: 75m