Trapped Surfaces

A trapped surface is the moment gravity stops being a force you can fight and becomes a direction you cannot avoid. Deep inside a collapsing star, spacetime curves so violently that even an outward flash of light is dragged into smaller area — the geometry has tilted the light cones fully inward. To make this precise, take a closed spacelike two-surface $S$ and equip it with its two future-directed null normals: the outgoing $k_+$ and ingoing $k_-$ . Measure how the area of each null congruence changes as it leaves $S$ ; those fractional rates are the null expansions $\theta_\pm$ . In flat space an outgoing wavefront spreads ( $\theta_+>0$ ) and an ingoing one focuses ( $\theta_-<0$ ). A surface is trapped exactly when $\theta_+<0$ and $\theta_-<0$ everywhere on $S$ : every light ray leaving it, either way, is converging. The genius of the definition is that it is local, geometric, and coordinate-free. It uses only the induced metric on $S$ and how $S$ sits in the spacetime — no time slicing, no waiting for null infinity, no symmetry assumption. That is what let Roger Penrose replace the delicate exact-symmetry collapse models of the 1930s with a robust theorem in 1965: a trapped surface, plus the null energy condition and a non-compact Cauchy surface, forces an incomplete null geodesic — a singularity — without ever solving Einstein's equations. The boundary case $\theta_+=0$ (with $\theta_-<0$ ) is a marginally outer trapped surface, whose outermost instance on a slice is the apparent horizon — the quasi-local stand-in for the event horizon that every numerical-relativity code hunts for.

Key takeaways

A trapped surface is a closed spacelike 2-surface $S$ on which both null expansions are negative: $\theta_+<0$ and $\theta_-<0$ . Even the outgoing light converges.
The expansion $\theta=\dfrac{1}{\delta A}\dfrac{\mathrm d(\delta A)}{\mathrm d\lambda}$ is the fractional rate of change of the surface's area element along a null normal; trapped means area shrinks in every light-like direction.
The null Raychaudhuri equation forces any converging congruence to reach a caustic within affine parameter $\lambda\le 2/|\theta_0|$ — the engine behind the Penrose singularity theorem.
For Schwarzschild in ingoing coordinates, $\theta_+=(1-2M/r)/r$ and $\theta_-=-2/r$ : the outgoing expansion changes sign exactly at the horizon $r=2M$ , and every sphere with $r<2M$ is trapped.
Trapped surfaces are quasi-local and observer-independent, making them — not the teleological event horizon — the workhorse of black-hole formation, apparent-horizon finders, and cosmic censorship.
Key idea: Project the derivative of each null normal along $S$ to get the two expansions $\theta_\pm=q^{ab}\nabla_a(k_\pm)_b$ ; both negative means light cannot reach larger area in any direction, Raychaudhuri's equation forces caustics in finite affine time, and global causal arguments turn that focusing into geodesic incompleteness — a singularity.

The core insight

Project the gradient of a closed spacelike 2-surface's two null normals onto the surface to obtain the expansions $\theta_\pm=q^{ab}\nabla_a(k_\pm)_b$ ; when both are negative, every light ray leaving $S$ is converging, Raychaudhuri's equation guarantees those congruences hit caustics in finite affine time, and global causal structure converts that focusing into an incomplete geodesic — a singularity.

Where this lives today

Trapped surfaces are the live diagnostic of every numerical black-hole simulation. Codes modelling a binary merger run, on each time slice, an apparent-horizon finder that solves $\theta_+=0$ for the outermost marginally trapped surface; the area and multipole moments of that MOTS, via the isolated- and dynamical-horizon formalism, yield the mass and spin of the remnant compared against the ringdown of events like GW150914. On the mathematical frontier, Demetrios Christodoulou's 2009 theorem showed that focused gravitational waves can create a trapped surface out of regular vacuum data — trapping is not seeded only by matter. That sharpens the central open problem of classical relativity: weak cosmic censorship, the unproven conjecture that the singularities these surfaces guarantee are always cloaked behind an event horizon, so no distant observer sees a naked one. Quasi-local horizon thermodynamics — a first law and area-increase law for dynamical horizons, and the entropy $S_{\text{BH}}=A/4$ of a MOTS — is where these objects meet black-hole thermodynamics.

The physical picture

Imagine standing on a spherical balloon and firing a flash of light both outward and inward. Normally the outward flash swells into an ever-larger sphere and the inward flash collapses — outgoing area grows, ingoing shrinks. Now drop the balloon deep inside a black hole. The curvature is so severe that the outward flash also loses ground: the sphere it forms an instant later is smaller than the one you stood on. When both flashes shrink, no light-like direction increases area — the surface, and everything it encloses, is trapped and doomed to a singularity. The first figure makes the definition a picture: three snapshots of the outgoing and ingoing wavefronts just after they leave $S$ , sorting all closed surfaces into untrapped, marginal, and trapped.

Figure

A closed 2-surface S flashes light both ways. Untrapped: the outgoing wavefront grows while the ingoing one shrinks. Marginal: the outgoing wavefront is frozen at the same area — a marginally outer trapped surface. Trapped: both wavefronts shrink, so no light escapes to larger area.

Raychaudhuri's equation makes the trapped condition catastrophic rather than descriptive. Once $\theta$ is negative anywhere, it is forced to $-\infty$ in finite affine parameter — a caustic. A trapped surface gives both congruences a head start of $\theta<0$ , so caustics on both sides are unavoidable, and Penrose's topology upgrades "unavoidable caustic" to "incomplete geodesic". The second figure grounds this in Schwarzschild, where the outgoing expansion changes sign exactly at the horizon.

Figure

Schwarzschild null expansions of round spheres in ingoing Eddington-Finkelstein form with M = 1, so the horizon sits at r = 2M = 2. The outgoing expansion θ₊ = (1 − 2M/r)/r crosses zero at r = 2M and is negative inside; the ingoing θ₋ = −2/r is negative everywhere. Both negative inside r = 2M means every sphere there is trapped.

Quasi-locality is the practical payoff. A code evolving a binary merger cannot "see" future null infinity to locate the true event horizon while it runs, so instead it solves $\theta_+=0$ on each time slice for the outermost marginally trapped surface — the apparent horizon, whose area and shape give the remnant's mass and spin. The trapped surface is not an abstraction; it is the number the machine computes.

The core equations

Everything begins with the expansion of a null normal — the fractional rate at which the area element of $S$ grows or shrinks as the surface is carried along its light rays. Projecting the gradient of the normal onto $S$ strips away the longitudinal and rotational parts.

\theta_\pm \;=\; q^{ab}\,\nabla_a (k_\pm)_b \;=\; h^{ab}\,\nabla_a (k_\pm)_b \qquad (1)

where: $S$ — closed (compact, $\partial S=\varnothing$ ) spacelike 2-surface; $k^a_\pm$ — the two future-directed null fields normal to $S$ , $+$ outgoing and $-$ ingoing; $q_{ab}=h_{ab}$ — induced positive-definite metric on $S$ , equivalently the transverse projector $g_{ab}+k^+_a k^-_b+k^-_a k^+_b$ with $k_+\!\cdot k_-=-1$ ; $\nabla_a$ — spacetime connection. Reading: $\theta_+>0$ means outgoing light gains area; $\theta_+<0$ means even outgoing light loses it. The definition is then a single sign condition on both normals — a pointwise, not an average, statement.

\text{Trapped surface}:\qquad \theta_+<0 \;\text{ and }\; \theta_-<0 \ \text{ everywhere on } S \qquad (2)

where: $\theta_\pm$ — the expansions from (1); "everywhere on $S$ " — holding at every point of the compact surface. If instead $\theta_+=0$ with $\theta_-<0$ the surface is marginally outer trapped (a MOTS); if both are positive it is anti-trapped (a white hole). Why negativity is a life sentence, not a passing phase, is the content of the Raychaudhuri equation.

\frac{\mathrm d\theta}{\mathrm d\lambda} \;=\; -\tfrac{1}{2}\,\theta^2 \;-\; \sigma_{ab}\sigma^{ab} \;+\; \omega_{ab}\omega^{ab} \;-\; R_{ab}\,k^a k^b \qquad (3)

where: $\lambda$ — affine parameter; $\theta$ — expansion of the congruence; $\sigma_{ab}$ — shear, with $\sigma_{ab}\sigma^{ab}\ge0$ ; $\omega_{ab}$ — twist, vanishing for surface-orthogonal congruences; $R_{ab}k^ak^b$ — Ricci focusing term, $\ge0$ under the null energy condition via Einstein's equations. Reading: every term except the twist pushes $\mathrm d\theta/\mathrm d\lambda$ negative, so a surface-orthogonal congruence in ordinary matter can only focus faster. The individual signs of $\theta_\pm$ depend on an arbitrary boost of the null frame, but their product does not — so the product carries the coordinate-free classification.

\theta_+\,\theta_- \;>\;0 \iff \text{trapped or anti-trapped},\qquad \theta_+\,\theta_-\;<\;0 \iff \text{untrapped} \qquad (4)

where: $\theta_+\theta_-$ — the product of the two expansions, invariant under the boost $k_+\to e^{\alpha}k_+,\ k_-\to e^{-\alpha}k_-$ that preserves $k_+\!\cdot k_-=-1$ and sends $\theta_\pm\to e^{\pm\alpha}\theta_\pm$ . Reading: the product's sign is a scalar on $S$ ; trapped is distinguished from anti-trapped by both factors being individually negative. These become explicit numbers in Schwarzschild. In ingoing coordinates, where $\mathrm ds^2=-(1-2M/r)\,\mathrm dv^2+2\,\mathrm dv\,\mathrm dr+r^2\,\mathrm d\Omega^2$ is regular at the horizon, the two expansions of a round sphere are elementary functions of the areal radius.

\theta_+ \;=\; \frac{1}{r}\Big(1-\frac{2M}{r}\Big),\qquad \theta_- \;=\; -\,\frac{2}{r},\qquad \theta_+\theta_- \;=\; -\,\frac{2}{r^2}\Big(1-\frac{2M}{r}\Big) \qquad (5)

where: $r$ — areal radius; $M$ — mass in geometrized units $G=c=1$ . Reading: $\theta_-<0$ for all $r>0$ ; $\theta_+>0$ for $r>2M$ , $0$ at $r=2M$ , $<0$ for $r<2M$ . The product is negative (untrapped) outside, zero at the horizon, positive (trapped) inside — the classifier (4) with a formula. (The boost-symmetric exterior form $\theta_\pm=\pm\frac{\sqrt2}{r}\sqrt{1-2M/r}$ is valid for $r>2M$ ; the EF form is regular for all $r$ .) The horizon is the borderline surface, and its dynamical generalization is what numerical relativity tracks.

\text{MOTS}:\quad \theta_+=0,\ \ \theta_-<0;\qquad \text{apparent horizon} = \text{outermost MOTS on a Cauchy slice} \qquad (6)

where: $\theta_+=0$ — outgoing rays momentarily neither expand nor contract; "outermost" — the MOTS enclosing all others on a slice. Reading: for static Schwarzschild the MOTS sits at $r=2M$ and equals the event horizon; in dynamical collapse it lies strictly inside the event horizon.

Derivation AWhy the projected divergence is the fractional rate of area change

The goal of this derivation is to justify the physical reading of equation (1): to show from first principles that the abstract quantity $q^{ab}\nabla_a(k_+)_b$ is exactly the fractional rate at which the area element of $S$ changes when every point is dragged along the outgoing null normal $k_+$ . Only then does the trapped condition $\theta_+<0$ acquire its meaning — that outgoing light loses area — on which the edifice of focusing rests. We carry the area element through a Lie drag along $k_+$ .

Step 1. Write the area element of $S$ in a chart $x^A$ intrinsic to the surface as $\delta A=\sqrt{\det q}\ \mathrm d^2x$ , where $\det q$ is the determinant of the induced two-metric $q_{AB}$ . Why: all of "how big is $S$ " sits in $\sqrt{\det q}$ ; tracking its change tracks the area change, coordinate-free.

Step 2. Drag the surface an affine step along $k_+$ and use the Jacobi determinant identity, $\mathcal L_{k_+}\!\sqrt{\det q}=\tfrac12\,q^{ab}\big(\mathcal L_{k_+}q_{ab}\big)\sqrt{\det q}$ . Why: the variation of a determinant is its trace against the inverse metric ( $\delta\ln\det q=q^{ab}\,\delta q_{ab}$ ) — where area becomes a trace over $S$ .

Step 3. Recognize the projected Lie derivative as the projected symmetric gradient of the normal, $\tfrac12\,q^{ab}\mathcal L_{k_+}q_{ab}=q^{ab}\nabla_a(k_+)_b=\theta_+$ . Why: the trace against $q^{ab}$ discards the trace-free shear and antisymmetric twist, leaving exactly the expansion.

Step 4. Combine the steps into $\theta_+=\dfrac{1}{\delta A}\dfrac{\mathrm d(\delta A)}{\mathrm d\lambda}$ . Why: $\theta_+$ is literally the fractional rate of area change, so $\theta_+<0$ means outgoing light loses area, and a trapped surface is one where area shrinks along both normals.

Go deeper · Derivation B · 1 minFocusing: a converging null congruence must reach a caustic in finite affine time

Derivation BFocusing: a converging null congruence must reach a caustic in finite affine time

The goal here is to prove the statement that turns the trapped condition into a catastrophe: any surface-orthogonal null congruence whose expansion is negative at some point, in matter obeying the null energy condition, is driven to $\theta\to-\infty$ within a bounded affine parameter set by that initial expansion. This focusing theorem is the differential-geometric fact the Penrose singularity theorem converts, by global causal arguments, into an incomplete geodesic. We integrate an inequality from Raychaudhuri's equation (3).

Step 1. For a surface-orthogonal congruence the twist vanishes ( $\omega_{ab}=0$ ), the shear term is non-negative ( $\sigma_{ab}\sigma^{ab}\ge0$ ), and the null energy condition gives $R_{ab}k^ak^b\ge0$ via Einstein's equations. Dropping the two non-negative terms turns (3) into $\dfrac{\mathrm d\theta}{\mathrm d\lambda}\le-\tfrac12\theta^2$ . Why: every discarded term only made $\mathrm d\theta/\mathrm d\lambda$ more negative, so the true expansion focuses at least as fast as the bound.

Step 2. Change variables to $u=1/\theta$ . Then $\dfrac{\mathrm du}{\mathrm d\lambda}=-\dfrac{1}{\theta^2}\dfrac{\mathrm d\theta}{\mathrm d\lambda}\ge\tfrac12$ , so integrating from $\theta_0=\theta(0)<0$ gives $\dfrac{1}{\theta(\lambda)}\ge\dfrac{1}{\theta_0}+\dfrac{\lambda}{2}$ . Why: the substitution linearizes the Riccati inequality — the quadratic term becomes a constant slope, reducing it to a straight line crossing zero.

Step 3. Since $\theta_0<0$ , the right-hand side starts negative and rises with slope $\tfrac12$ , reaching zero at $\lambda=2/|\theta_0|$ . For $1/\theta$ to stay above it, $\theta$ must diverge to $-\infty$ no later than there: a caustic forms within $\lambda\le 2/|\theta_0|$ . Why: a caustic is where neighbouring rays cross and the congruence leaves the boundary of the future — the seed the singularity theorem needs. A trapped surface gives both congruences $\theta_0<0$ , so caustics on both sides are guaranteed.

Go deeper · Derivation C · 1 minThe Schwarzschild expansions and the sign change at the horizon

Derivation CThe Schwarzschild expansions and the sign change at the horizon

The goal of this derivation is to produce the explicit formulas of equation (5) — to compute, for a round sphere at fixed radius in Schwarzschild, both null expansions and see the outgoing one pass through zero precisely at $r=2M$ . It shows in a solvable case that the horizon is exactly the marginally trapped surface and the interior genuinely trapped. We use ingoing Eddington–Finkelstein coordinates so that nothing is singular at the horizon.

Step 1. Take $\mathrm ds^2=-f\,\mathrm dv^2+2\,\mathrm dv\,\mathrm dr+r^2\,\mathrm d\Omega^2$ with $f=1-2M/r$ , and choose the ingoing null normal $k_-^a=-(\partial_r)^a$ . It is null ( $g_{rr}=0$ ) and affinely geodesic, giving $\theta_-=\dfrac{2}{r}\,k_-^r=-\dfrac{2}{r}$ . Why: for any radial null field the transverse expansion of the round spheres is $\dfrac{2}{r}$ times its radial component — the area $4\pi r^2$ changes only through $r$ .

Step 2. Choose the outgoing normal $k_+^a=(\partial_v)^a+\tfrac{f}{2}(\partial_r)^a$ , fixed by $k_+\!\cdot k_-=-1$ . With $g_{vr}=1$ , $g_{rr}=0$ one checks $k_+\!\cdot k_-=g_{vr}k_+^v k_-^r=-1$ , so it holds and $k_+$ is null. Why: $k_+\!\cdot k_-=-1$ fixes the boost freedom and makes the two expansions comparable.

Step 3. Compute the outgoing expansion the same way, $\theta_+=\dfrac{2}{r}\,k_+^r=\dfrac{2}{r}\cdot\dfrac{f}{2}=\dfrac{f}{r}=\dfrac{1}{r}\Big(1-\dfrac{2M}{r}\Big)$ . Why: the radial component of $k_+$ is $f/2$ , so the outgoing area change is set by $f$ — the horizon physics is in that single factor.

Step 4. Read off the signs. $\theta_-=-2/r<0$ always; $\theta_+>0$ for $r>2M$ , $=0$ at $r=2M$ , $<0$ for $r<2M$ ; hence $\theta_+\theta_-=-\dfrac{2f}{r^2}$ is negative outside (untrapped), zero at the horizon (MOTS), positive inside (trapped). Why: the invariant classifier (4) in a concrete metric — the Schwarzschild event horizon is the marginally outer trapped surface, and every sphere it encloses is trapped.

Limiting cases

The expansions $\theta_\pm$ sort every closed spacelike 2-surface into four mutually exclusive types (normalize $k_+\!\cdot k_-=-1$ with $k_+$ outgoing).

Case	Result	Meaning
Untrapped (normal)	$\theta_+>0,\ \theta_-<0$	Generic weak field: outgoing light expands, ingoing focuses. Any sphere in flat space, and Schwarzschild spheres at $r>2M$ .
Marginally outer trapped (MOTS)	$\theta_+=0,\ \theta_-<0$	The borderline: outgoing rays momentarily frozen. The outermost MOTS on a slice is the apparent horizon; for Schwarzschild it sits at $r=2M$ .
Trapped	$\theta_+<0,\ \theta_-<0$	Both normals converge. Penrose's hypothesis; lies inside the apparent horizon and defines the trapped region ( $r<2M$ in Schwarzschild).
Anti-trapped (white hole / expanding)	$\theta_+>0,\ \theta_->0$	Both diverge — the time-reverse of trapped. Behind a white hole's past horizon and in the expanding universe near the Big Bang.

Common trapTrapped is not the same as "inside the event horizon", and the individual signs of θ± are convention

Three confusions to disarm. (1) Quasi-local versus global. The event horizon is teleological — defined by the entire future of the spacetime, locatable only after the fact. A trapped surface and the apparent horizon are quasi-local, computable from one slice. In a dynamical collapse the apparent horizon lies strictly inside the event horizon, and the trapped region strictly inside that; they coincide only in stationary spacetimes like Schwarzschild. (2) The signs are frame-dependent, the product is not. $\theta_+$ and $\theta_-$ separately depend on the arbitrary boost $k_+\to e^{\alpha}k_+$ , $k_-\to e^{-\alpha}k_-$ ; the invariant content is the product $\theta_+\theta_-$ plus the requirement that both be simultaneously negative. (3) Trapped is foliation-independent; the apparent horizon is not. Whether a given $S$ is trapped depends only on $S$ and its normals. But the apparent horizon — the outermost MOTS — depends on the slice, and different foliations can place it differently.

Worked examples

1 · Classifying Schwarzschild spheres, and the horizon of the Sun

Using $\theta_+=(1-2M/r)/r$ and $\theta_-=-2/r$ , classify the spheres at $r=3M$ , $r=2M$ , and $r=M$ , then give the physical horizon radius for a one-solar-mass black hole.

Step 1. At $r=3M$ : $\theta_+=\dfrac{1}{3M}\big(1-\tfrac{2}{3}\big)=\dfrac{1}{9M}>0$ and $\theta_-=-\dfrac{2}{3M}<0$ , so $\theta_+\theta_-=-\dfrac{2}{27M^2}<0$ : this sphere is untrapped — outgoing light still escapes to larger area.

Step 2. At $r=2M$ : $\theta_+=\dfrac{1}{2M}\big(1-1\big)=0$ and $\theta_-=-\dfrac{1}{M}<0$ : a marginally outer trapped surface, and being outermost it is the apparent horizon, coincident here with the event horizon.

Step 3. At $r=M$ : $\theta_+=\dfrac{1}{M}\big(1-2\big)=-\dfrac{1}{M}<0$ and $\theta_-=-\dfrac{2}{M}<0$ . Both negative, so the sphere is fully trapped — even the outgoing flash is dragged inward, and $\theta_+\theta_-=+\dfrac{2}{M^2}>0$ confirms it.

Step 4. Restore units for a solar-mass hole: $r_s=2M=\dfrac{2GM_\odot}{c^2}=\dfrac{2(6.674\times10^{-11})(1.989\times10^{30})}{(2.998\times10^8)^2}\ \text{m}=\mathbf{2.95\ km}$ . Every sphere inside $2.95$ km is trapped; the sphere exactly at $2.95$ km is the marginally trapped horizon; every sphere outside is untrapped.

2 · How fast a caustic forms once a surface is trapped

A sub-horizon sphere has outgoing expansion $\theta_0=-0.5\ \text{km}^{-1}$ at the moment it becomes trapped. Using the focusing bound, find the maximum affine parameter before the outgoing congruence reaches a caustic, and compare it to the horizon scale.

Step 1. The focusing theorem gives $\dfrac{1}{\theta(\lambda)}\ge\dfrac{1}{\theta_0}+\dfrac{\lambda}{2}$ with $\theta_0<0$ ; the congruence reaches $\theta\to-\infty$ no later than $\lambda_{\max}=\dfrac{2}{|\theta_0|}$ .

Step 2. Substitute: $\lambda_{\max}=\dfrac{2}{0.5\ \text{km}^{-1}}=\mathbf{4\ km}$ of affine parameter (for a null geodesic, units of length in the chosen normalization).

Step 3. Compare to scale. A horizon is a few kilometres, so the caustic forms within roughly one horizon crossing — the focusing is a prompt collapse, not a slow leak. The ingoing congruence, starting more negative, caustics even sooner. Both caustics, inside a globally hyperbolic spacetime with a non-compact Cauchy surface, are what Penrose converts into an incomplete geodesic: a singularity.

Physics of the portrait

The trapped surface turned black holes from a curiosity of one exact solution into a generic prediction of general relativity — through one geometer who thought in pictures.

Origins · 1 minRoger Penrose and how this was discovered

Roger Penrose

b. 1931 · general relativity, gravitational collapse, and the singularity theorems

Roger Penrose was trained as a pure mathematician, and it shows in the trapped surface: a topological idea smuggled into physics. Through the early 1960s the study of gravitational collapse was stuck. The Oppenheimer–Snyder model of 1939 showed a perfectly spherical dust star collapsing to a singularity, but everyone suspected that was an artefact of exact spherical symmetry — surely a realistic, lumpy star would miss the centre and bounce. Penrose broke the deadlock in 1965 with a two-page paper in Physical Review Letters that introduced the closed trapped surface and proved that its mere existence, with the null energy condition and a non-compact Cauchy surface, forces an incomplete null geodesic. The argument used no symmetry and never solved Einstein's equations; it ran on the causal structure of spacetime — futures of sets, achronal boundaries, global hyperbolicity — machinery he and Stephen Hawking developed into the general singularity theorems of 1965–1970. Because the trapped surface is quasi-local and stable under perturbation, it defeated the "symmetry artefact" objection: perturb the star and the surface, once formed, is still there. Penrose went on to invent the Penrose–Carter conformal diagrams, formulate the cosmic censorship conjectures, and receive the 2020 Nobel Prize in Physics for showing that black-hole formation is a robust consequence of gravity.

History

1939

— Oppenheimer and Snyder compute the continued gravitational collapse of a spherical dust star

1965

— Penrose introduces the closed trapped surface and proves the first singularity theorem

1970

— Hawking and Penrose prove the general singularity theorem for cosmology and collapse

1973

— Hawking and Ellis systematize trapped surfaces and horizons in The Large Scale Structure of Space-Time

2009

— Christodoulou proves the dynamical formation of trapped surfaces from regular initial data

2020

— Penrose receives the Nobel Prize in Physics for the singularity-theorem work

Going further

Texts. Wald, General Relativity (1984), §9.2 (congruences and focusing), §9.5 and §12.2 (the Penrose theorem and trapped surfaces) · Misner, Thorne & Wheeler, Gravitation (1973), §34.5–34.6 · Hawking & Ellis, The Large Scale Structure of Space-Time (1973), §8.2, §9.1 · Poisson, A Relativist's Toolkit (2004), §2.4–2.5 (congruences) and Ch. 13 (horizons) · Carroll, Spacetime and Geometry (2004), §5.8, §9.5 · Penrose, "Gravitational Collapse and Space-Time Singularities", Phys. Rev. Lett. 14, 57 (1965). Key results to keep in mind: the Schwarzschild expansions in ingoing Eddington–Finkelstein coordinates are $\theta_+=(1-2M/r)/r$ and $\theta_-=-2/r$ , with the outgoing sign change at $r=2M$ ; the null Raychaudhuri equation yields the focusing bound $\theta\to-\infty$ within $\lambda\le 2/|\theta_0|$ via the substitution $u=1/\theta$ ; the Penrose (1965) incompleteness argument requires a trapped surface, the null energy condition, global hyperbolicity, and a non-compact Cauchy surface; the boost $k_\pm\to e^{\pm\alpha}k_\pm$ sends $\theta_\pm\to e^{\pm\alpha}\theta_\pm$ , leaving the product $\theta_+\theta_-$ invariant. The horizon radius for one solar mass, $2M=2GM_\odot/c^2=2.95$ km, follows directly from those constants.

Prerequisites. Null Geodesics and Light Cones · Causal Structure of Spacetime · Global Hyperbolicity · Achronal Sets and Edges. Leads to. Penrose–Carter Conformal Diagrams · Domain of Dependence and Cauchy Horizons · Asymptotic Flatness.

Ask the tutor: Ask me to walk through Penrose's 1965 argument line by line, or to compute the null expansions for a Reissner–Nordström or Vaidya black hole and watch the trapped region appear.

Why does it take both expansions negative — why is θ₋ < 0 not already enough?

For any closed spacelike 2-surface, even in flat space, the ingoing congruence focuses: $\theta_-<0$ is the generic, harmless behaviour of light aimed inward — every ordinary sphere has it. The pathology of strong gravity is that even the outgoing rays converge, $\theta_+<0$ , so light cannot reach larger area in any direction. Only when both hold does the focusing theorem trap both congruences into caustics and Penrose's topology force a singularity. Requiring just $\theta_-<0$ would call every sphere in the universe "trapped", which is why the definition insists on the simultaneous negativity that flat space never supplies.

Is a trapped surface the same thing as being inside the event horizon?

Not in general, and the difference is the whole reason the concept exists. The event horizon is teleological: defined by the entire future of the spacetime, so you cannot locate it until everything that will ever fall in has. A trapped surface is quasi-local — testable on a single slice. In static Schwarzschild the trapped region, apparent horizon, and event horizon line up at $r=2M$ . But in a dynamical collapse or merger they separate: the trapped region sits inside the apparent horizon, which sits inside the event horizon, because the event horizon grows in anticipation of matter not yet arrived. Numerical relativity tracks the apparent horizon precisely because the event horizon is invisible in real time.

Do the actual numbers θ₊ and θ₋ mean anything, if a boost can rescale them?

Their individual magnitudes carry no invariant meaning: rescaling $k_+\to e^{\alpha}k_+$ , $k_-\to e^{-\alpha}k_-$ (preserving $k_+\!\cdot k_-=-1$ ) sends $\theta_\pm\to e^{\pm\alpha}\theta_\pm$ , so you can inflate one and deflate the other. What survives is the sign of each expansion and the product $\theta_+\theta_-$ , both untouched by the boost. That is why classifying a surface as trapped, marginal, or untrapped is coordinate-free even though a value like " $\theta_+=0.1\ \text{km}^{-1}$ " is not. Explicit numbers belong to a specific frame; read the physics from the signs and product.

How can Penrose prove a singularity without solving Einstein's equations?

Einstein's equations enter in exactly one place: the null energy condition. Through the field equations $T_{ab}k^ak^b\ge0$ becomes $R_{ab}k^ak^b\ge0$ , making the Ricci term in Raychaudhuri's equation focus rather than defocus light. Everything else is causal topology. The focusing theorem guarantees caustics; global hyperbolicity and the compactness of the trapped surface make the boundary of its future compact; but a non-compact Cauchy surface forces that boundary to be non-compact — a contradiction, unless some boundary generator is incomplete, running off the edge of spacetime in finite affine parameter. Penrose needs only that gravity focuses light and that causal structure is well-behaved, which is why the result is so robust.

What is a MOTS, and why does numerical relativity care so much about it?

A marginally outer trapped surface (MOTS) is the borderline case $\theta_+=0$ with $\theta_-<0$ : outgoing light is momentarily frozen, neither gaining nor losing area. The outermost MOTS on a slice is the apparent horizon. Numerical relativity cares because it cannot find the true event horizon during a simulation — that needs the entire future. Solving $\theta_+=0$ , by contrast, is an elliptic equation for a surface shape using only current-slice data, so an apparent-horizon finder locates the black hole every few time steps. Its area gives the mass, its geometry the spin, and those feed the gravitational-wave templates matched against detector data.

What does the trapped surface have to do with cosmic censorship?

A trapped surface guarantees a singularity but says nothing about whether it is hidden. Weak cosmic censorship is the conjecture that it always is — that generic collapse from regular data produces singularities cloaked behind an event horizon, never a "naked" one visible to a distant observer. Christodoulou's 2009 theorem sharpened the question by proving trapped surfaces form dynamically from regular vacuum data, so trapping is genuinely generic, not an artefact of special matter or symmetry. Whether the resulting singularities are always dressed remains one of the great unsolved problems of classical general relativity, with the trapped surface at its centre.