Achronal Sets and Edges

Take the everyday idea of "an instant of time" and strip away everything that assumes a clock. What is left is a surface that no observer can pass through twice — a slice you cross once, going forward, and never meet again. In a general curved spacetime there is no preferred time function to define such a slice, so relativity replaces "constant-time surface" with a purely causal notion: an achronal set, a set of events no two of which are joined by a future-directed timelike curve. This one condition, written $I^{+}(S)\cap S=\varnothing$ , is the geometric backbone of the initial-value problem, because it guarantees that each timelike worldline meets $S$ at most once — exactly what one needs to lay down Cauchy data without redundancy. The subtlety lives at the rim. A set can be achronal everywhere yet fail to be a proper surface at certain limit points, where worldlines sneak around it instead of through it. Penrose's edge construction isolates precisely these leaks: $p$ is an edge point when every neighbourhood of $p$ contains a timelike curve from past to future of $p$ that dodges $S$ entirely. The clean theorem — an edgeless closed achronal set is a $C^{0}$ hypersurface without boundary — is the rigorous form of the slogan edgeless equals good slice. Remove the edge and you have a membrane fit to carry initial data; leave it, and predictability leaks out through the hole. These ideas interlock with the whole causal-structure cluster: the boundary $\partial I^{+}(S)$ of a future is itself the prototypical achronal hypersurface, woven from null geodesics that, traced into the past, end on an edge or run on forever — the geometric engine of the Penrose and Hawking singularity theorems. Achronality, edges, and null generators form the trio that organises the large-scale causal geometry of every spacetime.

Key takeaways

An achronal set $S$ contains no two chronologically related events: $I^{+}(S)\cap S=\varnothing$ , so every timelike worldline meets it at most once.
Achronality is strictly weaker than acausality (which forbids null connections too) — a piece of a light cone is achronal but not acausal, showing achronal is not the same as spacelike.
The edge of $S$ is the set of points near which $S$ leaks: every neighbourhood holds a timelike curve from past to future of $p$ that misses $S$ entirely.
An edgeless closed achronal set is a closed embedded $C^{0}$ hypersurface without boundary — precisely the structure a Cauchy surface must have.
Where $S$ has an edge it anchors the Cauchy horizons $H^{\pm}(S)$ that bound the domain of dependence $D(S)$ ; an edgeless $S$ can still carry a horizon, one ruled by past-inextendible null rays.
Boundaries of futures, $\partial I^{+}(S)$ , are achronal $C^{0}$ hypersurfaces ruled by null geodesic generators that, traced into the past, end on the edge or are past-inextendible — the lemma that drives the singularity theorems.
Key idea: an achronal set is a one-way membrane for timelike worldlines, and its edge is exactly where the membrane develops holes; remove the edge and you obtain a genuine $C^{0}$ hypersurface fit to carry Cauchy data.

The core insight

An achronal set is a one-way membrane for timelike worldlines, and its edge is exactly the set of points where the membrane develops holes; remove the edge and you obtain a genuine $C^{0}$ spacelike-or-null hypersurface fit to carry Cauchy data.

Where this lives today

Achronal slices are the working substrate of numerical relativity. Every simulation behind a LIGO-Virgo-KAGRA waveform template evolves Einstein's equations forward from data laid on an achronal Cauchy surface, in the ADM or BSSN formulation, with moving-puncture gauges that keep the slice achronal as it wraps around forming black holes; the true event horizon — an achronal null boundary $\partial I^{-}(\mathscr{I}^{+})$ — is then reconstructed by shooting null generators backward from the settled remnant, precisely the generators of the boundary-is-achronal lemma. The edge concept is equally alive in mathematical relativity: the Cauchy horizon of a rotating black hole is where prediction from an exterior achronal slice ends, and the strong cosmic censorship conjecture asks whether that horizon is always singular enough to forbid extension. The 2017 work of Dafermos and Luk on the Kerr interior showed it is generically a weak null singularity — the metric stays continuous ( $C^{0}$ -extendible) but not differentiable across it — a result stated in exactly the language of achronal boundaries and their generators. Causal set theory, meanwhile, takes the article's deepest fact — that causal order plus a counting measure recovers the metric — and proposes a discrete achronal-antichain structure as the substance of spacetime itself.

The physical picture

Picture a soap film stretched across spacetime so that no timelike worldline pierces it twice. A massive particle's history is a timelike curve, strictly inside the light cones, and achronality demands that once such a curve crosses the film going forward it can never return. That is what "one-way membrane" means: no timelike worldline meets the surface a second time. The film is allowed to touch the light cone — to have null pieces — because a null connection is not a chronological one; what it may never do is be penetrated twice by a strictly timelike curve. This is why achronality is a global statement about the whole causal web, not a local statement about the surface's shape.

To test whether a boundary point $p$ is an edge point, shrink a tiny box around $p$ and ask a single question: can I draw a timelike curve from just below $p$ to just above $p$ , staying inside the box, while completely missing $S$ ? At an ordinary point the answer is no — the surface cleanly separates the box into a past side and a future side, and every short timelike curve from below to above must puncture it. At an edge point the answer is yes for arbitrarily small boxes: the surface has frayed away, and worldlines slip around it. The most consequential example is the flat disk with a rim: the disk is achronal, but its bounding circle is an edge, and from that circle a Cauchy horizon springs upward, capping off exactly the region the disk can predict.

Figure

The open unit disk S at t = 0 has edge r = 1; its domain of dependence D(S) is a finite diamond capped by Cauchy horizons t = 1 − r that rise from the edge to the apex at t = 1

The most important achronal sets arise on their own: the boundary $\partial I^{+}(S)$ of any future is always achronal, always closed, and away from its edge is a $C^{0}$ hypersurface stitched from null geodesic generators. Traced into the past those generators have no endpoints inside the boundary — they run back to the edge or are past-inextendible — and that fact, married to the Raychaudhuri focusing of null rays, forces geodesic incompleteness in the singularity theorems. Achronality is not a technicality; it is the hinge on which the global theorems of general relativity turn.

The core equations

The whole subject rests on one definition, stated in the language of chronological futures. A set is achronal exactly when its own future never reaches back into it, so that no internal pair of events is timelike related.

S\ \text{achronal}\iff I^{+}(S)\cap S=\varnothing \iff \forall\,p,q\in S:\ q\notin I^{+}(p) \qquad (1)

where: $S$ — a subset of the spacetime manifold $(M,g)$ ; $I^{+}(p)$ — the chronological future of $p$ , the events reachable by a future-directed timelike curve; $I^{+}(S)=\bigcup_{p\in S}I^{+}(p)$ — the chronological future of the whole set. Reading: "no event of $S$ lies in the timelike future of another event of $S$ " — the surface is a one-way membrane for timelike worldlines. Acausality is the stronger sibling. It forbids even null connections, so it is the condition to demand when light rays, not just massive particles, must not relink the surface.

S\ \text{acausal}\iff \forall\,p,q\in S:\ q\ne p\Rightarrow q\notin J^{+}(p) \qquad (2)

where: $J^{+}(p)$ — the causal future of $p$ , reachable by a future-directed timelike or null curve. Reading: since $I^{+}(p)\subset J^{+}(p)$ , acausal implies achronal, but a piece of a light cone is achronal and not acausal — the two are genuinely different conditions. Penrose's edge picks out the points where $S$ stops being a proper membrane. It is defined by the existence, in every neighbourhood, of a timelike curve that threads from past to future of $p$ while missing $S$ .

p\in\mathrm{edge}(S)\iff p\in\bar S\ \text{and}\ \forall\,\mathcal{U}\ni p\ \ \exists\,\gamma\subset\mathcal{U}\ \text{timelike from}\ I^{-}(p,\mathcal{U})\ \text{to}\ I^{+}(p,\mathcal{U}),\ \gamma\cap S=\varnothing \qquad (3)

where: $\bar S$ — the topological closure of $S$ ; $\mathcal{U}$ — an open neighbourhood of $p$ ; $I^{\pm}(p,\mathcal{U})$ — the chronological future/past of $p$ computed inside $\mathcal{U}$ ; $\gamma$ — a timelike curve from below $p$ to above $p$ avoiding $S$ . Reading: a hole persists at every scale — the surface has frayed away at $p$ . The structural payoff is the edgeless theorem: kill the edge and an achronal set is automatically a topological hypersurface, the very object a Cauchy surface must be.

\mathrm{edge}(S)=\varnothing\ \text{and}\ S\ \text{closed achronal}\ \Longrightarrow\ S\ \text{is a } C^{0}\ \text{embedded hypersurface without boundary} \qquad (4)

where: $C^{0}$ — continuous (locally Lipschitz) regularity, not necessarily smooth. Reading: edgeless equals good slice; the surface can be locally written as a continuous graph $t=f(x^{i})$ over a spatial coordinate patch. The boundary of any future is the canonical achronal hypersurface, and its achronality is automatic — it needs no hypothesis on $S$ at all.

\partial I^{+}(S)\ \text{is a closed achronal } C^{0}\ \text{hypersurface, ruled by null geodesic generators} \qquad (5)

where: $\partial I^{+}(S)$ — the topological boundary of the open set $I^{+}(S)$ . Reading: because $I^{+}(S)$ is open, no two of its boundary points can be timelike related, and the frontier between "reachable" and "not" can only be woven from light rays. Achronal sets feed the initial-value problem through the domain of dependence and its Cauchy horizons — the region $S$ can predict, and the boundary where prediction stops.

D(S)=D^{+}(S)\cup D^{-}(S),\qquad H^{\pm}(S)=\overline{D^{\pm}(S)}\setminus I^{\mp}\!\big(D^{\pm}(S)\big) \qquad (6)

where: $D^{+}(S)$ — the future domain of dependence, the events whose every past-inextendible causal curve meets $S$ ; $H^{+}(S)$ — the future Cauchy horizon, the future boundary of $D^{+}(S)$ . Reading: data on $S$ determine physics throughout $D(S)$ and nowhere beyond $H^{\pm}(S)$ . Where $S$ has an edge the horizons are anchored to it: a generator traced into the past either reaches $\mathrm{edge}(S)$ or is past-inextendible, so the edge is where predictability begins to leak whenever there is one. This closes the loop from achronality to Cauchy development.

\mathrm{edge}(S)=\varnothing\ \text{and}\ D(S)=M\ \Longleftrightarrow\ S\ \text{is a Cauchy surface} \qquad (7)

where: $M$ — the full spacetime manifold. Reading: a Cauchy surface is exactly a closed achronal set that is edgeless and whose domain of dependence is everything; drop $D(S)=M$ and $S$ is only a partial Cauchy surface, with a horizon whose generators spring from its edge when the edge is non-empty, and are past-inextendible when it is not.

Derivation ARemoving the edge upgrades an achronal set to a topological hypersurface

The goal is to prove the structural fact on which the notion of a Cauchy surface rests: a closed achronal set with empty edge is a closed embedded $C^{0}$ hypersurface without boundary — near each point it is the graph of a continuous function of the spatial coordinates. The engine is one physical idea: thread the neighbourhood with timelike worldlines, use achronality so each thread meets $S$ at most once, and use edgelessness so it meets $S$ at least once. Continuity then follows because any jump in the graph would hide a forbidden timelike link.

Step 1. Fix a point $p\in S$ and choose a convex normal neighbourhood $\mathcal{U}$ small enough that its causal structure is a mild deformation of Minkowski's, with a coordinate $\partial_{0}$ that is everywhere timelike; the integral curves of $\partial_{0}$ are timelike worldlines threading $\mathcal{U}$ from bottom to top. Why: a convex normal neighbourhood is where local causal reasoning is clean — every pair of its points is joined by a unique geodesic, so "which thread hits $S$ where" is well defined.

Step 2. By achronality, no timelike curve meets $S$ twice, so each $\partial_{0}$ -thread intersects $S$ in at most one point. Recording that point defines a partial function $t=f(x^{i})$ from the spatial coordinates to the time coordinate — the height at which the thread punctures $S$ , wherever it does. Why: this converts the set $S$ into the graph of a function, the object we want, but so far only a partial one — defined only where threads happen to hit.

Step 3. Because $p\notin\mathrm{edge}(S)$ , shrinking $\mathcal{U}$ makes every thread running from $I^{-}(p,\mathcal{U})$ to $I^{+}(p,\mathcal{U})$ cross $S$ ; if some thread missed $S$ , it would be exactly the timelike curve from past to future of $p$ avoiding $S$ that the edge definition forbids. So $f$ is defined on a full coordinate ball around the spatial projection of $p$ . Why: edgelessness is precisely the hypothesis that plugs the holes in the domain of $f$ , turning the partial graph into a total one.

Step 4. The function $f$ is continuous. If it jumped between two nearby spatial points, the two corresponding surface events would sit at different heights on almost-parallel threads, and a short timelike curve could run from the lower one up to a point timelike-above the higher one and back down through $S$ — reconnecting two events of $S$ by a timelike curve, which achronality forbids. Why: continuity is not assumed; it is squeezed out of achronality, which is why the surface is $C^{0}$ but need not be smooth.

Step 5. Near $p$ the set $S$ is therefore the graph $\{(f(x^{i}),x^{i})\}$ , an embedded $C^{0}$ hypersurface without boundary; since $p\in S$ was arbitrary, $S$ is globally a closed $C^{0}$ hypersurface. Why: "no boundary" is the whole point — an edge would have appeared exactly as a place where the graph terminated, and we assumed there is none.

Go deeper · Derivation B · 2 minThe boundary of a future is achronal and ruled by null geodesics

Derivation BThe boundary of a future is achronal and ruled by null geodesics

The aim is to establish the two properties of $\partial I^{+}(S)$ that make it the workhorse of the singularity theorems: it is achronal, and away from $S$ it is generated by null geodesics with no future endpoint in the boundary. The first is a short openness argument; the second is the classic "round the corner" reasoning, run twice — once to force nullity, once to forbid corners. The upshot is a surface of light rays that, traced into the past, run back to the edge or never end at all, exactly what the focusing arguments exploit to prove geodesic incompleteness.

Step 1. Suppose two boundary points $p,q\in\partial I^{+}(S)$ were timelike related, $q\in I^{+}(p)$ . Since $I^{+}(p)$ is open and contains $q$ , an entire neighbourhood of $q$ lies in $I^{+}(p)\subset I^{+}(S)$ , so $q$ is interior to $I^{+}(S)$ — contradicting $q\in\partial I^{+}(S)$ . Hence no such pair exists and $\partial I^{+}(S)$ is achronal. Why: openness of chronological futures is the entire mechanism — a boundary point simply cannot have chronological room around it, so timelike relations between boundary points are impossible.

Step 2. Take $x\in\partial I^{+}(S)$ with $x\notin\bar S$ . There are points $q_{n}\in I^{+}(S)$ with $q_{n}\to x$ , each joined to $S$ by a future-directed timelike curve; by the limit curve lemma these accumulate on a past-directed causal curve $\lambda$ from $x$ that stays in $\overline{I^{+}(S)}$ . Why: the limit of "just barely reachable" timelike curves is the causal curve that traces the frontier itself, so $\lambda$ lies exactly on the boundary.

Step 3. The curve $\lambda$ must be null. If any segment of it were timelike, then $x$ would inherit a timelike curve from $S$ and lie in the open set $I^{+}(S)$ , contradicting $x\in\partial I^{+}(S)$ . So every tangent of $\lambda$ is null. Why: a boundary point sits on the razor's edge between reachable and not — a single timelike step would push it strictly inside.

Step 4. The null curve $\lambda$ is a geodesic. A null curve with a corner can always be smoothed into a strictly timelike curve across the corner (the standard rounding argument), which would again drop $x$ into $I^{+}(S)$ . Having no corners, $\lambda$ is a null geodesic — a generator of the boundary. Why: geodesy is forced by the same "no timelike shortcut" principle, now applied to kinks instead of segments.

Step 5. Traced into the past, a generator stays in $\partial I^{+}(S)$ until it reaches $\overline{S}$ — at a point of $\mathrm{edge}(S)$ when $S$ is achronal — or else runs on past-inextendibly; it has no past endpoint in the interior of the boundary, because such an endpoint would admit a deformation into $I^{+}(S)$ . To the future a generator may leave the boundary, and at a caustic it must. That pairing is the premise the Penrose focusing argument needs. Why: it means a generator that meets a conjugate point (forced by Raychaudhuri under an energy condition) must leave the boundary there or be incomplete — the mechanism behind the singularity theorem.

Go deeper · Derivation C · 2 minThe edge anchors the Cauchy horizon

Derivation CThe edge anchors the Cauchy horizon

The goal is to show why the edge of a partial Cauchy surface is the physical location where predictability from $S$ runs out: the Cauchy horizon $H^{+}(S)$ is an achronal null surface whose generators, followed into the past, end on $\mathrm{edge}(S)$ when $S$ has one. This ties three of the cluster's objects together — the achronal set, its edge, and the domain of dependence — and explains the flat-disk picture, where a horizon rises from the rim to cap the region the disk determines. The argument reuses the null-generator machinery of the previous derivation, applied now to the boundary of $D^{+}(S)$ .

Step 1. Let $S$ be closed and achronal with $\mathrm{edge}(S)\ne\varnothing$ — a partial Cauchy surface such as the open disk, so that $D(S)\ne M$ — and consider a point $x\in H^{+}(S)=\overline{D^{+}(S)}\setminus I^{-}(D^{+}(S))$ . By construction $x$ is a future limit of the determined region but is not itself in the timelike past of it, so it sits on an achronal boundary. Why: the Cauchy horizon is a boundary of a causal region, so it inherits the same achronal-boundary structure as $\partial I^{+}$ .

Step 2. The same limit-curve and no-corner arguments as before show $H^{+}(S)$ is achronal and, where it is not on $\overline{S}$ , is generated by null geodesics with no past endpoint in the horizon. Each generator is a past-inextendible null ray lying in the horizon. Why: horizons are frontiers of prediction, and frontiers of causal reachability are always ruled by light rays — the geometry does not care whether the region is a future or a domain of dependence.

Step 3. Follow a generator into the past. It cannot enter $D^{+}(S)$ (there it would be determined, not on the horizon), and it cannot cross $S$ transversally (it would then be inside the domain of dependence). So if the generator terminates at all it does so where $S$ itself fails to block causal curves — a point of $\mathrm{edge}(S)$ ; otherwise it has no past endpoint and is simply past-inextendible in the horizon. Why: the edge is by definition where timelike and null curves slip around $S$ , so it is exactly where a horizon generator can begin — but an edgeless $S$ leaves the other branch open.

Step 4. Hence $\overline{H^{+}(S)}$ and $\overline{H^{-}(S)}$ share the edge as their common anchor: $\mathrm{edge}(S)\subset\overline{H^{+}(S)}\cap\overline{H^{-}(S)}$ . When $\mathrm{edge}(S)=\varnothing$ no generator can have a past endpoint, so any horizon that survives is ruled entirely by past-inextendible null rays — as for the hyperboloid of Worked Example 2 — and only when additionally $D(S)=M$ is the surface a full Cauchy surface. Why: this is the precise sense in which "no edge and full domain" equals "Cauchy surface" — the edge is the seed of every failure of global prediction. *The flat-disk Cauchy horizon $t=1-|x|$ rising from the rim $|x|=1$ is the standard illustration of this construction.*

Limiting cases

Case	Result	Meaning
Spacelike slice in Minkowski	$S=\{t=0\}\subset\mathbb{R}^{1,3}$	Edgeless, acausal, smooth spacelike hypersurface; a Cauchy surface with $D(S)=\mathbb{R}^{1,3}$ .
Null cone piece	$S\subset\partial I^{+}(p)$	Achronal but not acausal (null-related to $p$ ) and not spacelike — the standard proof that achronal is not the same as spacelike.
Open disk with rim	$S=\{t=0,\ r<1\}$	Achronal, but the circle $r=1$ is its edge; $D(S)$ is the finite diamond $t=1-\|x\|$ and the Cauchy horizon rises from the rim.
Hyperboloid in Minkowski	$S=\{t^{2}-r^{2}=1,\ t>0\}$	Edgeless achronal spacelike surface, yet not a Cauchy surface: null rays asymptotic to the light cone never meet it, so $D(S)\ne M$ .

Common trapAchronal, acausal, and spacelike are the same condition

They are three different conditions and conflating them is the classic error. Achronal forbids only timelike links between points of $S$ ; acausal forbids timelike and null links, so it is strictly stronger — a slice of a light cone is achronal but not acausal. Spacelike is a local metric property (the induced metric is positive-definite at each point), whereas achronality is a global causal property of how the surface sits in the whole spacetime. The two come apart in both directions: an achronal set may be partly null and nowhere spacelike (a light-cone piece), and a locally spacelike surface can fail to be achronal if a long timelike curve reconnects two of its points — as wherever the time direction closes up, for instance Minkowski space identified under $t\sim t+T$ , or anywhere closed timelike curves exist. Only under extra global assumptions do the notions coincide.

Worked examples

1 · The domain of dependence of a flat disk (fully worked)

In two-dimensional Minkowski space with $c=1$ and signature $(-,+)$ , take the open unit "disk" $S=\{t=0,\ |x|<1\}$ . Find its edge, its domain of dependence, the equation of its Cauchy horizon, and decide whether the events $(t,x)=(0.5,0.3)$ and $(0.5,0.7)$ are determined by data on $S$ .

Step 1. The edge. $S$ is a flat segment on $t=0$ ; every interior point is an ordinary achronal point, but the two limit points $x=\pm1$ are edge points, because at $x=1$ a short timelike curve can loop from just below to just above while slipping around the end of the segment. So $\mathrm{edge}(S)=\{(0,\pm1)\}$ — the two rim points, and $S$ is achronal but not edgeless.

Step 2. The horizon. A future Cauchy-horizon generator is a null ray anchored to the edge. From the rim point $(0,1)$ the inward-going future null ray is $x=1-t$ ; from $(0,-1)$ it is $x=-1+t$ . These meet at the apex $(t,x)=(1,0)$ , so the future Cauchy horizon is $H^{+}(S):\ t=1-|x|$ , and by time reversal $H^{-}(S):\ t=-(1-|x|)$ . The horizons are 45° null lines rising from the rim to an apex at $t=1$ .

Step 3. The domain of dependence is the region they enclose: $D(S)=\{|x|+|t|< 1\}$ , a diamond of coordinate "height" $2$ from $t=-1$ to $t=+1$ . Any past-inextendible causal curve through an interior point of this diamond is trapped by the 45° walls into crossing $S$ ; outside it, a causal curve can escape through the edge without ever meeting $S$ . $D(S)$ is the finite diamond $|x|+|t|< 1$ , not the whole plane.

Step 4. Test the two events. For $(0.5,0.3)$ : the horizon height at $x=0.3$ is $1-0.3=0.7$ , and $0.5<0.7$ , so the event is below $H^{+}$ , inside $D^{+}(S)$ — determined by data on $S$ . For $(0.5,0.7)$ : the height at $x=0.7$ is $1-0.7=0.3$ , and $0.5>0.3$ , so it lies above $H^{+}$ , outside $D(S)$ — not determined: a past-directed causal curve can run from it out through the rim to past null infinity without ever touching $S$ . The lesson is sharp: the edge at $|x|=1$ is what caps prediction at the line $t=1-|x|$ .

2 · An edgeless achronal surface that is still not a Cauchy surface

Show that the upper unit hyperboloid $S=\{t=\sqrt{1+x^{2}},\ t>0\}$ in two-dimensional Minkowski space is achronal, spacelike, and edgeless, yet fails to be a Cauchy surface by exhibiting an inextendible causal curve that never meets it.

Step 1. Achronal and spacelike. Any two points of $S$ satisfy $t^{2}-x^{2}=1$ , and a short computation shows the interval between distinct points of $S$ is spacelike, so no two are timelike related — $S$ is achronal, and its induced metric is positive-definite, so it is a smooth spacelike surface. It is also edgeless: it is a complete surface with no rim, extending to arbitrarily large $|x|$ .

Step 2. The domain of dependence is not everything. Consider the event $(t,x)=(0,5)$ , which lies far outside the future light cone of the origin (it has $|x|>t$ ). It is not in $D^{+}(S)$ or $D^{-}(S)$ , and to prove it we produce a future-inextendible causal curve through it that misses $S$ .

Step 3. Take the future-directed null ray $x(t)=5+t$ for $t\ge0$ . Setting it equal to the surface requires $\sqrt{1+(5+t)^{2}}=t$ , i.e. $1+(5+t)^{2}=t^{2}$ , which reduces to $26+10t=0$ , giving $t=-2.6$ . That solution lies in the excluded past, so for all $t\ge0$ the ray never reaches $S$ — it climbs forever in the region $x>t$ , staying below the hyperboloid because $\sqrt{1+x^{2}}>x>t$ .

Step 4. Interpretation. Because this inextendible causal curve threads to future null infinity without crossing $S$ , data on $S$ cannot determine physics along it, so $D(S)\ne M$ and $S$ is not a Cauchy surface. Contrast a static worldline $x=5$ , which does cross $S$ at $t=\sqrt{26}\approx 5.10$ — some curves meet it, but a Cauchy surface must be met by every inextendible causal curve. The moral: edgeless plus achronal plus spacelike is necessary but not sufficient; one also needs $D(S)=M$ .

Physics of the portrait

The modern theory of achronal sets and their edges is very nearly one person's invention. Before the 1960s, "a surface of the present" and "the region from which nothing can influence infinity" were physical pictures without rigorous definitions; it was the geometrisation of causality — conformal infinity, the achronal boundary, the edge — that turned them into theorems with hypotheses one can check.

Origins · 1 minRoger Penrose and how this was discovered

Roger Penrose

1931– · global causal structure, achronal boundaries, and the edge construction

Roger Penrose rebuilt general relativity around topology and global causal analysis in a field dominated by explicit metrics and local computation. In the early 1960s he introduced conformal infinity, rescaling the metric so the infinite reaches of spacetime become a finite ideal boundary, and proved that because the rescaling is positive in the interior it preserves every causal relation. Out of this program came the objects this article formalises: the chronological and causal futures, the achronal set, the precise edge construction that isolates where such a set fails to be a hypersurface, and the theorem that an edgeless closed achronal set is a $C^{0}$ hypersurface. His 1965 singularity theorem showed these tools had teeth: assuming a closed trapped surface, an energy condition, and a non-compact Cauchy surface, he proved the achronal boundary $\partial I^{+}$ is ruled by null geodesics with no future endpoints, and that focusing forces one to be incomplete — a singularity deduced from causal hypotheses alone, with no symmetry assumed. He codified the apparatus in his 1972 monograph, stated the cosmic censorship conjecture, and shared the 2020 Nobel Prize in Physics for showing black-hole formation is a robust prediction of the theory. Almost every named object in this cluster — the achronal boundary, the edge, the null-generator lemma — descends from his hand.

History

1959

— Arnowitt, Deser, and Misner pose gravitational initial data on a spacelike slice

1965

— Penrose proves the first singularity theorem using an achronal boundary ruled by null generators

1968

— Penrose defines the edge of an achronal set in his Battelle lectures on the structure of spacetime

1970

— Geroch shows a globally hyperbolic spacetime is foliated by Cauchy surfaces

1972

— Penrose codifies achronal sets and edges in Techniques of Differential Topology in Relativity

Going further

Texts. Wald, General Relativity (1984), Chapter 8, especially Section 8.3 · Hawking and Ellis, The Large Scale Structure of Space-Time (1973), Sections 6.3, 6.5, 6.6 · Misner, Thorne and Wheeler, Gravitation (1973), Chapter 34 · Carroll, Spacetime and Geometry (2004), Chapter 9 · Penrose, Techniques of Differential Topology in Relativity (1972) · Geroch, Domain of Dependence, J. Math. Phys. 11, 437 (1970). Key checkpoints for the results in this article: acausal implies achronal via $I^{+}(p)\subset J^{+}(p)$ , with the light-cone piece as the counterexample to the converse; the $C^{0}$ -hypersurface theorem holds for closed edgeless achronal sets; for the unit disk, $\mathrm{edge}(S)=\{(0,\pm1)\}$ , $H^{+}(S):t=1-|x|$ , apex at $t=1$ , $D(S)=\{|x|+|t|<1\}$ , with $(0.5,0.3)$ inside and $(0.5,0.7)$ outside; for the hyperboloid, the null ray $x=5+t$ from $(0,5)$ never meets $S$ (would-meet at $t=-2.6$ ), while the static line $x=5$ meets it at $t=\sqrt{26}\approx5.10$ . All statements are topological or exact; no approximate physical constants are asserted.

Prerequisites. Causal Structure of Spacetime · Chronological and Causal Futures and Pasts · Null Geodesics and Light Cones. Leads to. Domain of Dependence and Cauchy Horizons · Cauchy Surfaces · Global Hyperbolicity · Causality Conditions Hierarchy.

Ask the tutor: Want to see how the edge of a partial Cauchy surface grows into a full Cauchy horizon, or how the null-generator lemma feeds directly into Penrose's singularity theorem?

What is the difference between achronal, acausal, and spacelike?

Achronal forbids timelike connections: no two events of $S$ are joined by a future-directed timelike curve. Acausal is stronger — it also forbids null connections, so even light rays may not relink the surface; a slice of a light cone is achronal but not acausal. Spacelike is different in kind: a local metric statement that the induced metric is positive-definite at each point, saying nothing directly about how the surface sits in the global causal web. A spacelike surface can fail to be achronal (if the spacetime wraps a timelike curve back onto it), and an achronal set can fail to be spacelike (a light-cone piece is partly null). For the initial-value problem, achronal is the condition that matters, because it guarantees "at most one crossing" per worldline.

Why does removing the edge turn an achronal set into a hypersurface?

Because the edge is precisely the obstruction. Thread a small neighbourhood with timelike worldlines. Achronality ensures each thread hits the surface at most once, so the surface is the graph of a partial function $t=f(x^{i})$ . The only way this graph fails to be a full, continuous surface is for some thread to miss the set entirely — and a thread that runs from past to future of a point while avoiding the set is exactly what the edge definition detects. So "no edge" means "no missing threads," which makes $f$ defined on a whole coordinate patch, and achronality forces $f$ to be continuous (a jump would hide a forbidden timelike link). The result is a $C^{0}$ graph: an embedded hypersurface without boundary.

Is every spacelike hypersurface achronal?

Not in general. Spacelike is a local, pointwise condition — the induced metric is positive-definite — constraining only the surface's immediate neighbourhood. Achronality is global: it forbids any timelike curve, however far it roams, from reconnecting two points of the surface. A surface can be impeccably spacelike everywhere yet be pierced twice by a timelike curve that leaves it, loops through a distant region, and returns — this happens when the time direction is periodic, and dramatically in any spacetime with closed timelike curves. (Compactness of space alone does not do it: the slice $t=\text{const}$ of a flat spatial torus is achronal, indeed a Cauchy surface.) The two notions coincide only when extra global hypotheses (such as the surface being a Cauchy surface of a globally hyperbolic spacetime) rule out such long-range reconnections.

Why is ∂I⁺(S) always achronal even when S itself is not?

Because achronality of the boundary is forced by the openness of chronological futures, not by any property of $S$ . Suppose two boundary points $p$ and $q$ of $I^{+}(S)$ were timelike related, with $q$ in the future of $p$ . The open set $I^{+}(p)$ would contain $q$ ; but then an entire neighbourhood of $q$ would lie in $I^{+}(p)$ , hence in $I^{+}(S)$ , making $q$ an interior point — contradicting $q$ being on the boundary. So no two boundary points can be timelike related, whatever $S$ is. This is why the boundary of a future is a reliable achronal hypersurface even when the seed set $S$ is a ragged, non-achronal blob: taking the boundary of its future cleans it up automatically.

What does an edge point look like, physically?

At an ordinary point, a small enough neighbourhood is cleanly cut in two — a past side and a future side — and every short timelike curve from below to above must puncture the surface, which behaves like a local wall. An edge point is where that wall has frayed to nothing: no matter how small a box you draw, you can find a timelike curve that starts below the point, ends above it, and slips entirely around the surface. The rim of a flat disk is the cleanest example — right at the edge circle a worldline loops around the end of the disk from bottom to top. That is why edges are exactly where the surface stops being a hypersurface-without-boundary, and why they seed Cauchy horizons.

Why isn't the mass hyperboloid a Cauchy surface, even though it is edgeless?

Because edgeless and achronal guarantees "at most one crossing," but a Cauchy surface needs "exactly one crossing for every inextendible causal curve," and the hyperboloid fails the second. The surface $t=\sqrt{1+x^{2}}$ is asymptotic to the future light cone $t=|x|$ , so a causal curve that hugs the cone from outside — for instance the null ray $x=5+t$ from the event $(0,5)$ — climbs forever in the region $x>t$ and never crosses it. Its domain of dependence is therefore only the interior of the future light cone of the origin, so data on the hyperboloid leave the rest of spacetime undetermined. Edgelessness removes the horizons a rim would create, but it cannot force the surface to catch curves escaping to null infinity.