Trapped Surfaces and Closed Trapped Regions

Gravitational collapse posed a stubborn question for half a century: when a star falls through its Schwarzschild radius, is the singularity that forms a genuine feature of general relativity, or merely an artefact of the perfect spherical symmetry that Oppenheimer and Snyder assumed in 1939? Roger Penrose answered it in 1965 with a single geometric idea that needed no symmetry at all — the trapped surface. Rather than track a coordinate or a clock, he proposed to watch light itself. From any closed spacelike 2-surface two families of light rays depart orthogonally, one nominally outgoing and one ingoing, and each family's cross-sectional area either grows or shrinks as it propagates. The rates are the two null expansions $\theta_+$ and $\theta_-$ . In ordinary weak-field spacetime the outgoing family fans out ( $\theta_+>0$ ) while the ingoing family converges ( $\theta_-<0$ ), exactly as a flashbulb's wavefront ought to behave. A trapped surface is the extraordinary situation in which curvature has grown so intense that even the outgoing front is driven toward smaller area: $\theta_+<0$ as well as $\theta_-<0$ , everywhere on the surface at once. No signal born on such a surface can enlarge the area it bounds; the region is irreversibly committed to collapse. Fed into the Raychaudhuri focusing equation and combined with a modest energy condition and the global topology of an open universe, the existence of a single trapped surface forces the spacetime to be geodesically incomplete — a singularity, guaranteed. This article builds the expansions $\theta_\pm$ from scratch, computes them exactly in Schwarzschild, proves the focusing bound, sketches Penrose's topological argument, and connects trapped surfaces to apparent horizons, cosmic censorship, and the numerical detection of black holes.

Key takeaways

A trapped surface is a closed (compact, boundaryless) spacelike 2-surface $\mathcal{S}$ on which both future-directed null normal congruences converge — both expansions $\theta_+$ and $\theta_-$ are negative at every point of $\mathcal{S}$ .
The expansions are pure geometry: $\theta=q^{ab}\nabla_a k_b$ is the fractional rate of change of transverse area carried by a null congruence, and it needs only the metric on $\mathcal{S}$ and its neighbourhood — no symmetry, no global information.
In Schwarzschild the exact answer is $\theta_+=\tfrac{1}{r}(1-2GM/r)$ and $\theta_-=-2/r$ : the outgoing expansion changes sign precisely at the horizon $r=2GM$ , so every 2-sphere with $r<2GM$ is trapped.
The boundary case $\theta_+=0,\ \theta_-<0$ is a marginally trapped surface; the family of these is the apparent horizon — quasilocal and dynamical, generically strictly inside the global event horizon.
The Raychaudhuri equation plus the null energy condition drives any negative expansion to $-\infty$ within affine length $\Delta\lambda\le 2/|\theta_0|$ : a caustic, and the seed of Penrose's 1965 singularity theorem.
Key idea: a trapped surface is detected by computing the two null expansions $\theta_\pm$ of a closed spacelike 2-surface and finding both negative; this single sign condition encodes, locally and without any symmetry, the irreversible commitment of a region to gravitational collapse.

Where this lives today

The trapped surface is the working definition of a black hole inside a computer. Numerical-relativity codes that simulate the binary black-hole mergers LIGO and Virgo detect locate horizons on every time slice with dedicated apparent-horizon finders solving $\theta_+=0$ , because waiting for the global event horizon would mean waiting for the entire future of the simulation. When two holes merge, it is the sudden appearance of a common marginally trapped surface enclosing both that the code reports as coalescence, and the area of that surface feeds directly into checks of the Hawking area theorem against the radiated wave energy. The same machinery drives the dynamical and isolated horizons of Ashtekar and Krishnan, which build black-hole mechanics on marginally trapped tubes rather than the teleological event horizon, and it underlies the Penrose inequality $M\ge\sqrt{A/16\pi}$ (in units $G=c=1$ ) bounding the total mass by the area of the outermost trapped surface — proved in the time-symmetric case by Huisken–Ilmanen and Bray, still open in general. Trapped-surface estimates even bound black-hole production in ultra-relativistic particle collisions. And whenever an astrophysicist says a collapsing core has "passed the point of no return", the rigorous content of that phrase is that a trapped surface has formed.

The physical picture

Flash a pulse of light from a small closed 2-surface and two wavefronts leave it: one heading nominally outward, one collapsing inward. In flat space the outward front gains area — essentially what "outward" means — while the inward front loses it. A trapped surface is the paradoxical case in which even the outward front loses area as it advances: gravitational focusing has become so severe that "out" no longer means "getting bigger", the very wavefront that ought to carry a signal away is itself shrinking, and light cannot escape the region it encloses. The first figure sets the two scenarios, untrapped and trapped, side by side, using the sign of each expansion as the diagnostic.

Figure

Two null fronts leave a closed 2-surface. In weak field the outgoing front gains area (θ₊ positive) while the ingoing front loses area (θ₋ negative); inside a horizon curvature drags even the outgoing front inward, so both θ₊ and θ₋ turn negative and the surface is trapped

The concept earns its keep because it is measurable on a single slice of time. The expansion $\theta=\tfrac{1}{\delta A}\,\mathrm{d}(\delta A)/\mathrm{d}\lambda$ needs only the metric on $\mathcal{S}$ and how the null normals turn as they leave it, so a numerical relativist never has to evolve a collapse to future null infinity to declare a black hole formed — she searches each slice for surfaces with $\theta_+=0$ . The second figure makes the diagnostic quantitative for the one geometry we can solve exactly, Schwarzschild, plotting both expansions against areal radius and marking the horizon where the outgoing expansion changes sign.

Figure

Null expansions of Schwarzschild 2-spheres in ingoing Eddington–Finkelstein coordinates. The outgoing θ₊ = (1/r)(1 − 2GM/r) crosses zero at the horizon r = 2GM (marginally trapped), peaks at 0.125/GM at r = 4GM, while the ingoing θ₋ = −2/r stays negative everywhere; for r below 2GM both expansions are negative and every 2-sphere is trapped

The core equations

The whole subject is built from one scalar. A spacelike 2-surface $\mathcal{S}$ has, being codimension-two, exactly two future-directed null normals, and the expansion measures how the area carried by each evolves. The projector $q_{ab}=g_{ab}+k_a\ell_b+\ell_a k_b$ strips away the two null directions and keeps only the transverse metric.

\theta = q^{ab}\,\nabla_a k_b \qquad (1)

where: $\theta$ — expansion scalar of the null congruence; $k^a$ — future-directed null normal, taken geodesic and affinely parametrized; $q^{ab}$ — inverse of the induced transverse metric $q_{ab}=g_{ab}+k_a\ell_b+\ell_a k_b$ on the 2-surface; $\ell^a$ — the second null normal, cross-normalized so that $k\cdot\ell=-1$ . Read it as the fractional growth rate of a transverse area element dragged along $k^a$ : $\theta=\tfrac{1}{\delta A}\,\mathrm{d}(\delta A)/\mathrm{d}\lambda$ . Trapping is a statement about the two expansions $\theta_\pm$ associated with the outgoing $(+)$ and ingoing $(-)$ normals. The defining condition is that both are negative everywhere on the surface — the outgoing front, not just the ingoing one, is losing area.

\theta_+ < 0 \quad\text{and}\quad \theta_- < 0 \ \text{ on } \mathcal{S} \qquad (2)

where: $\theta_\pm$ — expansions of the outgoing and ingoing null normal congruences; $\mathcal{S}$ — closed (compact, boundaryless) spacelike 2-surface. Equivalently the null mean curvature $\Theta=\theta_++\theta_-<0$ together with $\theta_+<0$ isolates the future-trapped case. The boundary of the trapped region is where the outgoing expansion just vanishes: such a surface is marginally trapped, and the collection of them on a slice is the apparent horizon.

\theta_+ = 0,\qquad \theta_- < 0 \qquad (3)

where: $\theta_+=0$ — the outgoing front is momentarily neither gaining nor losing area; $\theta_-<0$ selects a marginally outer trapped surface (an apparent horizon), not a mere extremum. Reading: outside such a surface $\theta_+>0$ , inside $\theta_+<0$ . The dynamics of any expansion is fixed by the Raychaudhuri equation. For a null congruence it reads as below; the term that matters for collapse is the Ricci focusing term on the far right.

\frac{\mathrm{d}\theta}{\mathrm{d}\lambda} = -\frac{1}{2}\theta^2 - \sigma_{ab}\sigma^{ab} + \omega_{ab}\omega^{ab} - R_{ab}k^a k^b \qquad (4)

where: $\lambda$ — affine parameter along the null generators; $\sigma_{ab}$ — shear (trace-free, symmetric), contributing $-\sigma_{ab}\sigma^{ab}\le0$ ; $\omega_{ab}$ — twist, which vanishes for hypersurface-orthogonal congruences; $R_{ab}k^ak^b$ — the Ricci focusing term. Every term on the right except the twist is manifestly non-positive, so a hypersurface-orthogonal congruence can only focus. That focusing term is non-negative exactly when matter obeys the null energy condition, the mildest of the classical energy conditions and the only one Penrose's theorem requires.

R_{ab}\,k^a k^b \ge 0 \quad\text{for all null } k^a \qquad (5)

where: $R_{ab}$ — Ricci tensor. Through Einstein's equations $R_{ab}k^ak^b=8\pi G\,T_{ab}k^ak^b$ for null $k^a$ (the trace term drops out because $k^ak_a=0$ ), so the geometric condition is equivalent to $T_{ab}k^ak^b\ge0$ : the energy density is non-negative in every null frame. For the one geometry we can solve in closed form, the expansions of constant-radius 2-spheres are elementary, and they carry the real number that anchors the whole article: the sign change at the horizon.

\theta_+ = \frac{1}{r}\left(1-\frac{2GM}{r}\right), \qquad \theta_- = -\frac{2}{r} \qquad (6)

where: $r$ — areal radius of the 2-sphere; $M$ — Schwarzschild mass; $G$ — Newton's constant, with $c=1$ . Reading: $\theta_->0$ never happens, while $\theta_+$ shares the sign of $f=1-2GM/r$ — positive outside, zero on, negative inside the horizon. Finally, for genuinely dynamical collapse the same algebra applied to the ingoing Vaidya metric locates the growing apparent horizon, showing the concept is not tied to stationarity.

\theta_+ = 0 \quad\Longleftrightarrow\quad r = 2G\,m(v) \qquad (7)

where: $m(v)$ — ingoing (Bondi) mass function of advanced time $v$ ; $r$ — areal radius. As radiation falls in, $m(v)$ rises and the marginally trapped surface $r=2Gm(v)$ sweeps outward — a spacelike or timelike world-tube, strictly inside the event horizon during accretion.

Derivation ANull expansions of a Schwarzschild 2-sphere and the trapped region

The goal is to compute $\theta_+$ and $\theta_-$ exactly for round 2-spheres of constant time and radius in Schwarzschild, and to show that their product changes sign precisely at $r=2GM$ , so the entire interior is foliated by trapped surfaces while the exterior is untrapped. Ingoing Eddington–Finkelstein coordinates keep every quantity finite across the horizon, which is essential: the usual $t$ -coordinate is singular there and would obscure the very sign change we are after.

Step 1. Adopt ingoing Eddington–Finkelstein coordinates $(v,r,\vartheta,\varphi)$ with line element $\mathrm{d}s^2=-f\,\mathrm{d}v^2+2\,\mathrm{d}v\,\mathrm{d}r+r^2\mathrm{d}\Omega^2$ and $f=1-2GM/r$ . The two future-directed null normals to a sphere of fixed $v,r$ are $k^a\partial_a=\partial_v+\tfrac12 f\,\partial_r$ (outgoing) and $\ell^a\partial_a=-\partial_r$ (ingoing). Why: these are the null directions orthogonal to the 2-sphere; one checks $g_{ab}k^ak^b=-f+2\cdot\tfrac12 f=0$ and $g_{ab}\ell^a\ell^b=0$ , and $g_{ab}k^a\ell^b=-1$ , so they are correctly normalized.

Step 2. The induced metric on the sphere is $q_{ab}=\mathrm{diag}(r^2,\,r^2\sin^2\vartheta)$ in the angular block, so the transverse area element is $\delta A=r^2\sin\vartheta\,\mathrm{d}\vartheta\,\mathrm{d}\varphi$ and it depends on position only through the areal radius $r$ . Why: because $\delta A\propto r^2$ , the expansion collapses to $\theta=\tfrac{1}{\delta A}\tfrac{\mathrm{d}(\delta A)}{\mathrm{d}\lambda}=\tfrac{2}{r}\tfrac{\mathrm{d}r}{\mathrm{d}\lambda}$ along either normal — no Christoffel symbols survive.

Step 3. Along the outgoing normal $\dot r=k^r=\tfrac12 f$ , so $\theta_+=\tfrac{2}{r}\cdot\tfrac12 f=\dfrac{f}{r}=\dfrac{1}{r}\Big(1-\dfrac{2GM}{r}\Big)$ . Along the ingoing normal $\dot r=\ell^r=-1$ , giving $\theta_-=\tfrac{2}{r}(-1)=-\dfrac{2}{r}$ . Why: the ingoing rays converge unconditionally — $\theta_-$ is negative for every finite $r$ — whereas the outgoing expansion inherits the sign of $f$ , which is the crux.

Step 4. Form the trapping product $\theta_+\theta_-=-\dfrac{2f}{r^2}=-\dfrac{2}{r^2}\Big(1-\dfrac{2GM}{r}\Big)$ . Outside the horizon $f>0$ , so $\theta_+>0>\theta_-$ and the sphere is untrapped; at $r=2GM$ we have $\theta_+=0$ (marginally trapped); for $r<2GM$ , $f<0$ so both expansions are negative. Why: the product is a compact single test — it is negative in the ordinary exterior and positive in the trapped interior, changing sign exactly at the horizon.

Step 5. Conclude that the trapped region is the entire interior $0<r<2GM$ , foliated by trapped 2-spheres, and that its marginally trapped boundary $r=2GM$ coincides with the event horizon — a coincidence special to this stationary, vacuum, spherically symmetric case. Why: in a static spacetime the apparent and event horizons merge; the distinction reappears the moment the geometry becomes dynamical, as in Vaidya collapse. *Verified: at $r=3GM$ , $\theta_+=\tfrac{1}{3GM}(1-\tfrac23)=\tfrac{1}{9GM}>0$ ; at $r=2GM$ , $\theta_+=0$ ; $\theta_+$ peaks at $r=4GM$ with value $1/(8GM)=0.125/GM$ . Consistent with Poisson, A Relativist's Toolkit §2.6 and Carroll §5.8.*

Go deeper · Derivation B · 1 minRaychaudhuri focusing and the conjugate-point bound

Derivation BRaychaudhuri focusing and the conjugate-point bound

The goal is to show that once a null congruence starts converging — $\theta_0<0$ at some point — the Raychaudhuri equation together with the null energy condition forces the expansion to diverge to $-\infty$ within a bounded affine length $\Delta\lambda\le 2/|\theta_0|$ , producing a caustic. This is the analytic engine behind every singularity theorem: negative expansion is self-amplifying, and gravity focuses light irreversibly. The argument is a one-line differential inequality with global consequences.

Step 1. Take the generators to be hypersurface-orthogonal, so the twist vanishes, $\omega_{ab}=0$ (a null congruence is hypersurface-orthogonal if and only if it is twist-free). The shear term $-\sigma_{ab}\sigma^{ab}\le0$ and the focusing term $-R_{ab}k^ak^b\le0$ under the null energy condition can then only be dropped in our favour, leaving $\dfrac{\mathrm{d}\theta}{\mathrm{d}\lambda}\le-\dfrac12\theta^2$ . Why: discarding non-positive terms makes the right-hand side larger, so any bound we prove for the simplified inequality holds a fortiori for the true evolution.

Step 2. Let $u=1/\theta$ . Then $\dfrac{\mathrm{d}u}{\mathrm{d}\lambda}=-\dfrac{1}{\theta^2}\dfrac{\mathrm{d}\theta}{\mathrm{d}\lambda}\ge-\dfrac{1}{\theta^2}\Big(-\dfrac12\theta^2\Big)=\dfrac12$ . So the reciprocal expansion increases at least linearly, $u(\lambda)\ge u_0+\tfrac12\lambda$ . Why: the substitution linearizes the Riccati inequality; the quadratic nonlinearity that makes $\theta$ blow up becomes a harmless linear growth in $u$ .

Step 3. If $\theta_0<0$ then $u_0=1/\theta_0<0$ , and $u$ reaches $0$ from below no later than $\lambda_\ast=-2u_0=2/|\theta_0|$ . As $u\to0^-$ , $\theta=1/u\to-\infty$ : a conjugate point forms within $\Delta\lambda\le 2/|\theta_0|$ . Why: a divergence of $\theta$ to $-\infty$ is a caustic where neighbouring generators cross and the cross-sectional area collapses to zero — the light rays are being focused to a point.

Step 4. Apply this to a trapped surface: there $\theta_\pm\le-c<0$ everywhere, so every generator of both null congruences leaving $\mathcal{S}$ orthogonally hits a caustic within finite affine parameter, uniformly bounded across the compact surface. Why: compactness of $\mathcal{S}$ turns the pointwise bound into a uniform one — there is a single $\Lambda$ by which every generator has focused, which is what the global argument needs. *Verified: integrating $\dot\theta=-\tfrac12\theta^2$ exactly gives $\theta(\lambda)=\theta_0/(1+\tfrac12\theta_0\lambda)$ , which diverges at $\lambda=2/|\theta_0|$ ; the inequality only shortens this. Standard Raychaudhuri result (Wald §9.2, Hawking & Ellis §4.3).*

Go deeper · Derivation C · 2 minFrom one trapped surface to geodesic incompleteness — Penrose 1965

Derivation CFrom one trapped surface to geodesic incompleteness — Penrose 1965

The goal is to sketch how the local focusing result becomes a global theorem: given a trapped surface, the null energy condition, and a non-compact Cauchy surface, the spacetime cannot be null-geodesically complete — the precise meaning of "a singularity forms". The argument is topological, playing the compactness forced by focusing against the non-compactness of an open universe, and it was the first to divorce singularity formation from any symmetry assumption.

Step 1. Consider the boundary of the future $\partial J^+(\mathcal{S})$ of the trapped surface. It is generated by null geodesics leaving $\mathcal{S}$ orthogonally, and a generator belongs to the boundary only up to its first caustic; beyond a conjugate point the geodesic enters the interior of $J^+(\mathcal{S})$ and leaves the boundary. Why: past a focal point neighbouring generators have crossed, so points further along can be reached by a strictly timelike curve — they are no longer on the causal boundary.

Step 2. Suppose, for contradiction, that the spacetime is null-geodesically complete. Then by the focusing bound every generator reaches its caustic at finite affine parameter and terminates on the boundary there. The boundary $\partial J^+(\mathcal{S})$ is therefore generated by a compact family of finite-length segments, making it a compact set (it is closed, and now bounded in the relevant sense). Why: completeness guarantees each generator actually runs long enough to reach its guaranteed caustic; without completeness a generator could simply "run out" first — which is exactly the escape hatch.

Step 3. $\partial J^+(\mathcal{S})$ is an achronal three-dimensional topological hypersurface. Project it onto a Cauchy surface $\Sigma$ along the integral curves of a smooth timelike vector field; the projection is continuous and injective, so its image is both compact (from Step 2) and open in $\Sigma$ . Why: an achronal set meets each timelike integral curve at most once, so the projection is one-to-one and open by invariance of domain.

Step 4. A connected space cannot contain a non-empty subset that is simultaneously open and compact unless that subset is the whole space — but $\Sigma$ is non-compact by hypothesis (an open universe), so the image cannot be all of $\Sigma$ . Contradiction. Hence the assumption of completeness fails: some null generator is incomplete. Why: a non-compact connected manifold has no proper clopen subset, so a compact open image is impossible unless completeness — the premise that made the boundary compact — is false.

Step 5. Conclude that the spacetime contains at least one incomplete null geodesic: a singularity, formed with no appeal whatsoever to spherical symmetry. This is Penrose's 1965 theorem, the prototype of the Penrose–Hawking singularity theorems. Why: geodesic incompleteness is the coordinate-independent definition of a singularity — a freely falling observer or photon whose history simply ends after finite proper time or affine parameter. *Verified: statement matches Hawking & Ellis Theorem 8.2 and Wald §9.5 (Theorem 9.5.1); original Penrose, Phys. Rev. Lett. 14, 57 (1965).*

Limiting cases

The trapped region takes a characteristic form in each standard solution; the table reads across from a defining relation to its physical content.

Case	Result	Meaning
Schwarzschild (vacuum, static)	$\theta_+\theta_-<0 \iff r>2GM$	The exterior is untrapped; the whole interior $r<2GM$ is trapped, and the marginally trapped boundary $r=2GM$ is simultaneously apparent and event horizon (a stationary coincidence).
Reissner–Nordström (sub-extremal)	$r_\pm=GM\pm\sqrt{(GM)^2-GQ^2}$	The trapped region is the shell $r_-<r<r_+$ between inner and outer horizons; for $r<r_-$ the geometry is untrapped again — a feature absent in Schwarzschild.
Vaidya collapse (dynamical)	$\theta_+=0\Rightarrow r=2Gm(v)$	The apparent horizon $r=2Gm(v)$ grows as radiation accretes and lies strictly inside the event horizon during collapse — the generic time-dependent situation.
Flat / weak field	$\theta_+>0,\ \theta_-<0$	No trapped surfaces exist: every closed 2-surface has an expanding outgoing front. Trapping requires a sufficiently concentrated mass-energy.
Marginally trapped tube	$\theta_+=0$ on a world-tube	Dynamical/isolated horizons: a quasilocal replacement for the event horizon on which flux laws for area and angular momentum hold.

Common trapA trapped surface is the same thing as the event horizon

It is not, and conflating them is the most common error in this subject. The trapped and marginally trapped conditions are quasilocal: you test the sign of $\theta_\pm$ from data on $\mathcal{S}$ and its neighbourhood, available on a single Cauchy slice. The event horizon is global and teleological — the boundary of the causal past of future null infinity — so pinning it down requires the entire future evolution. In dynamical collapse the apparent horizon (traced by marginally trapped surfaces) generically lies strictly inside the event horizon, and the event horizon can even grow through flat space before any matter arrives; the two coincide only once the geometry is stationary. One more wording trap: "closed" means compact-without-boundary (a 2-sphere), not "closed" in the timelike sense — the surface is spacelike throughout.

Worked examples

1 · A trapping census of Schwarzschild 2-spheres (fully worked)

For the Schwarzschild geometry, classify the constant-radius 2-spheres at $r=3GM$ , $r=2GM$ , and $r=1.5GM$ , and find where the outgoing expansion is largest.

Step 1. Use the exact expansions $\theta_+=\tfrac{1}{r}(1-2GM/r)$ and $\theta_-=-2/r$ . At $r=3GM$ : $\theta_+=\tfrac{1}{3GM}(1-\tfrac23)=\tfrac{1}{9GM}\approx +0.111/GM$ and $\theta_-=-\tfrac{2}{3GM}\approx-0.667/GM$ . The product $\theta_+\theta_-<0$ , so the sphere is untrapped — consistent with lying outside the horizon (this radius is also the photon sphere).

Step 2. At $r=2GM$ : $\theta_+=\tfrac{1}{2GM}(1-1)=0$ and $\theta_-=-1/GM<0$ . This is the boundary case $\theta_+=0,\ \theta_-<0$ — a marginally trapped surface, the horizon itself.

Step 3. At $r=1.5GM$ : $\theta_+=\tfrac{1}{1.5GM}(1-\tfrac{2}{1.5})=\tfrac{1}{1.5GM}(1-1.333)=-0.222/GM$ and $\theta_-=-1.333/GM$ . Both negative, so the sphere is trapped — it lies inside the horizon, and every such interior sphere is trapped.

Step 4. Maximize $\theta_+$ : with $x=r/GM$ , $\theta_+ GM=(x-2)/x^2$ , and $\mathrm{d}/\mathrm{d}x=(4-x)/x^3=0$ gives $x=4$ . So the outgoing expansion is largest at $r=4GM$ , with $\theta_+=2/16\,GM^{-1}=\mathbf{0.125/GM}$ — beyond that radius the exterior geometry is so weak that even outgoing light barely fans, and $\theta_+$ decays back toward zero at infinity. Untrapped at $3GM$ , marginal at $2GM$ , trapped at $1.5GM$ ; $\theta_+$ peaks at $r=4GM$ .

2 · How far to the caustic? A solar-mass horizon (fully worked with numbers)

A photon congruence grazes the horizon of a $M=M_\odot$ black hole with the ingoing expansion $\theta_-=-2/r$ . Using the focusing bound, estimate the affine distance to the caustic, in kilometres.

Step 1. In geometrized units ( $c=1$ ) the mass length is $GM_\odot=1.477\ \mathrm{km}$ , so the horizon radius is $r_s=2GM_\odot=\mathbf{2.95\ km}$ . On the horizon the ingoing expansion is $\theta_-=-2/r_s=-2/(2.95\ \mathrm{km})=-0.677\ \mathrm{km}^{-1}$ .

Step 2. The conjugate-point bound from the Raychaudhuri inequality is $\Delta\lambda\le 2/|\theta_0|$ . Insert $|\theta_-|=2/r_s$ : $\Delta\lambda\le \dfrac{2}{2/r_s}=r_s$ . The bound is exactly the areal radius — a clean, mass-independent statement. Numerically for the Sun, $\Delta\lambda\le r_s=\mathbf{2.95\ km}$ .

Step 3. Interpret it: an ingoing null congruence that has reached the horizon of a solar-mass black hole is guaranteed to focus to a caustic within an affine length no greater than one horizon radius — the local seed that, through the topological argument, becomes the inevitability of the central singularity. The caustic bound equals $r_s\approx 2.95$ km for a solar-mass hole.

3 · The trapped shell of a charged black hole

A sub-extremal Reissner–Nordström hole has charge such that $GQ^2=0.75\,(GM)^2$ . Locate the trapped region.

Step 1. The horizons solve $f=1-2GM/r+GQ^2/r^2=0$ , i.e. $r^2-2GM r+GQ^2=0$ , giving $r_\pm=GM\pm\sqrt{(GM)^2-GQ^2}$ . With $GQ^2=0.75(GM)^2$ the discriminant is $\sqrt{(GM)^2(1-0.75)}=0.5\,GM$ .

Step 2. Hence $r_+=1.5\,GM$ and $r_-=0.5\,GM$ . The outgoing expansion $\theta_+\propto f$ is negative only between the roots, so the trapped region is the shell $0.5\,GM<r<1.5\,GM$ .

Step 3. Unlike Schwarzschild, the region $r<r_-=0.5\,GM$ is untrapped again: an infalling observer crosses the outer horizon into a trapped shell, then emerges through the inner horizon into a region where outgoing light can once more expand, approaching the timelike singularity rather than being crushed onto it. Trapped shell $0.5\,GM<r<1.5\,GM$ , untrapped inside $r_-$ .

Physics of the portrait

Trapped surfaces are the invention of one person, in one four-page paper, that reshaped an entire field. Before 1965 the singularity theorems did not exist; after them, singularities were understood to be a robust prediction of general relativity rather than a fragile consequence of symmetry.

Origins · 1 minRoger Penrose and how this was discovered

Roger Penrose

b. 1931 · global differential geometry of spacetime

Roger Penrose trained as a pure mathematician — his early work spanned tessellations, projective geometry, and the impossible figures he shared with M. C. Escher — and he brought that geometer's sensibility to relativity when the field most needed it. Through the early 1960s the study of gravitational collapse was stuck: the Oppenheimer–Snyder model showed a spherical dust star collapsing to a singularity, but no one could tell whether the singularity survived once exact spherical symmetry was relaxed. In a 1965 Physical Review Letters paper of barely three pages, Penrose cut the knot by importing global, topological methods — causal structure, achronal boundaries, Cauchy surfaces — and defining the trapped surface. His theorem showed that a trapped surface plus the null energy condition plus an open universe forces geodesic incompleteness, with symmetry playing no role at all. The same toolkit, developed with Stephen Hawking over the following decade, produced the cosmological singularity theorems and the framework for black-hole thermodynamics. Penrose went on to the cosmic censorship conjectures, the Penrose inequality, spin networks, and twistor theory, and shared the 2020 Nobel Prize in Physics for showing that black-hole formation is a robust prediction of general relativity.

History

1955

— Raychaudhuri derives the focusing equation for a congruence of geodesics

1965

— Penrose introduces trapped surfaces and proves the first singularity theorem

1970

— Hawking and Penrose extend the theorem to cosmological (big-bang) singularities

1973

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

2020

— Penrose receives the Nobel Prize for the black-hole singularity prediction

Going further

Texts. Misner, Thorne & Wheeler, Gravitation (1973), §34.5–34.6 · Wald, General Relativity (1984), §9.5 and Ch. 12 · Hawking & Ellis, The Large Scale Structure of Space-Time (1973), §8.2, §9.1 · Carroll, Spacetime and Geometry (2004), §5.8, §9.5 · Poisson, A Relativist's Toolkit (2004), Ch. 2 (null congruences) · Penrose, "Gravitational Collapse and Space-Time Singularities", Phys. Rev. Lett. 14, 57 (1965).

Provenance. The Schwarzschild expansions $\theta_+=f/r$ , $\theta_-=-2/r$ and the trapped region $r<2GM$ are recomputed here in ingoing Eddington–Finkelstein coordinates and match Poisson §2.6 and Carroll §5.8; the peak of $\theta_+$ at $r=4GM$ with value $0.125/GM$ is a direct calculus check. The Raychaudhuri focusing bound $\Delta\lambda\le 2/|\theta_0|$ follows from the exact solution $\theta=\theta_0/(1+\tfrac12\theta_0\lambda)$ (Wald §9.2, Hawking & Ellis §4.3). Penrose's theorem statement follows Hawking & Ellis Theorem 8.2 and Wald Theorem 9.5.1; the Reissner–Nordström horizons $r_\pm=GM\pm\sqrt{(GM)^2-GQ^2}$ are textbook (MTW §31.6). The Penrose inequality $M\ge\sqrt{A/16\pi}$ is quoted in $G=c=1$ units. A verifier should recompute the null normals, confirm $g_{ab}k^a\ell^b=-1$ , verify the sign change of $\theta_+\theta_-$ at $r=2GM$ , and re-integrate the Raychaudhuri inequality.

Prerequisites. Schwarzschild Black Hole Structure · Event and Apparent Horizons · Killing Horizons and Surface Gravity. Leads to. Hawking Area Theorem · Four Laws of Black Hole Mechanics · No-Hair Theorem and Black Hole Uniqueness · Reissner–Nordström Charged Black Hole.

Ask the tutor: Ask me to walk through the null-normal calculation line by line, or to show exactly where the compactness argument in Penrose's theorem meets the non-compact Cauchy surface.

What is the difference between a trapped surface and a closed trapped region?

A trapped surface is a single closed spacelike 2-surface on which both null expansions are negative. A closed trapped region (or trapped region) is the three-dimensional set on a given slice that is filled by such surfaces — in Schwarzschild it is the whole ball $r<2GM$ . The two-dimensional outer boundary of the trapped region is the apparent horizon, made of marginally trapped surfaces with $\theta_+=0$ . So "trapped surface" is the pointwise diagnostic, "trapped region" is the body of space it detects, and "apparent horizon" is the skin between the two. All three are quasilocal, defined on one slice, and none of them requires knowing the spacetime's future.

Why must the surface be spacelike and compact without boundary?

Spacelike so that its normal space is timelike and two-dimensional, which is what gives exactly two independent future-directed null normals whose expansions can be compared — the whole outgoing-versus-ingoing distinction depends on this. Compact without boundary so that "both expansions negative everywhere" is a genuinely global statement about the surface, with no edge through which some rays could escape the convergence. If the surface had a boundary, light could leak out around the rim and the focusing argument would fail; if it were not compact, the topological step in Penrose's theorem — the clash between a compact image and a non-compact Cauchy surface — would have nothing to bite on.

Can a trapped surface exist without a black hole, or vice versa?

Under the standard assumptions they travel together, but the logic runs one way. A trapped surface plus the null energy condition and a non-compact Cauchy surface guarantees a singularity — Penrose's theorem — and cosmic censorship (a conjecture, not a theorem) then expects that singularity to be hidden behind an event horizon, i.e. a black hole. The converse can fail transiently: during collapse an event horizon can already exist in a region containing no trapped surface yet, because the event horizon grows in anticipation of matter still falling in. So trapped surface implies singularity rigorously, while "black hole" as an event-horizon statement is the global, censorship-dependent overlay.

How does a numerical-relativity code actually find a trapped surface?

It runs an apparent-horizon finder: a routine that searches each spatial slice for a closed 2-surface with $\theta_+=0$ . Parametrizing the surface as a radial function in spherical harmonics, the finder writes $\theta_+$ from the induced 3-metric and extrinsic curvature carried by the slice, then solves the resulting elliptic equation for the surface shape by Newton iteration or a fast flow method; interior spheres are checked for $\theta_+<0$ to confirm the trapped region. Because this uses only current-slice data, a merger simulation can report "a common horizon has formed" in real simulation time, without evolving to future null infinity — which is why apparent horizons, not event horizons, are the everyday diagnostic.

Does the singularity theorem tell us what the singularity looks like?

No — and this is a frequent misreading. Penrose's theorem proves only that at least one null geodesic is incomplete: some photon or freely falling history ends after finite affine parameter. It says nothing about whether the curvature actually diverges, whether the singularity is spacelike or timelike, or how strong it is. The Schwarzschild singularity happens to be a genuine spacelike curvature singularity, but the theorem alone does not establish that; it is a statement about incompleteness, not about blow-up. Characterizing the nature of the singularity — the BKL oscillatory approach, strong versus weak singularities, and the status of cosmic censorship — requires separate, much harder analysis.

Why use ingoing Eddington–Finkelstein coordinates instead of Schwarzschild time?

Because Schwarzschild $t$ is singular at $r=2GM$ : $g_{tt}=-f$ vanishes and $g_{rr}=1/f$ blows up there, even though the geometry is smooth. Computing $\theta_+$ in those coordinates gives a $0/0$ ambiguity exactly at the horizon — the one place we most want a clean answer. Ingoing Eddington–Finkelstein coordinates replace $t$ with advanced time $v=t+r_\ast$ , which stays finite as ingoing light crosses the horizon, so the metric and null normals are regular through $r=2GM$ . The expansions then come out as the manifestly finite $\theta_+=f/r$ and $\theta_-=-2/r$ , and the sign change of $\theta_+$ at the horizon is transparent.