Hawking-Penrose Singularity Theorem
Before 1965 most relativists suspected that the singularities in Schwarzschild and Friedmann–Lemaître–Robertson–Walker were flukes of perfect symmetry — artifacts a realistic, lumpy collapse would surely miss, letting the star bounce and the universe skip past any true beginning. Penrose's 1965 paper ended that hope by trading the algebra of exact solutions for the global geometry of causal structure, and the Hawking–Penrose theorem of 1970 is the summit of that programme: a single statement covering gravitational collapse and cosmology in one breath. Its architecture is a proof by contradiction on three pillars. An energy condition makes gravity attractive, so the Raychaudhuri equation drives any converging bundle to a focal point in finite affine length. The generic condition guarantees every causal geodesic actually meets some tidal curvature, closing the loophole of a ray that coasts forever through flat emptiness. The causality conditions keep the global structure sane enough that maximal geodesics exist — and a maximal geodesic, by a variational lemma, can hold no conjugate point in its interior. Focusing manufactures the conjugate point; maximality forbids it; the two collide. A single trigger sets the trap: a closed trapped surface, a reconverging light cone, or a compact Cauchy surface — collapse, cosmology, or a spatially closed universe. Any one seeds converging geodesics, and the only escape from the contradiction is for some inextendible causal geodesic to have finite affine length. That is the rigorous, coordinate-free meaning of a singularity. The theorem never displays a singular point, never claims curvature blows up, and never says whether a horizon hides it. What it secures is unavoidability: singularities are a generic prediction of classical general relativity, not a symmetry accident.
- The Hawking–Penrose theorem is the most general classical singularity theorem: reasonable energy and causality conditions plus one focusing trigger force a spacetime to be causally geodesically incomplete.
- It fuses Hawking's timelike focusing (cosmology) with Penrose's null focusing (collapse) through one tool applied to both congruences: the Raychaudhuri equation.
- "Singularity" here means geodesic incompleteness — some timelike or null geodesic runs out of affine parameter — not necessarily a curvature blow-up, though curvature generically does diverge.
- The proof is a reductio: completeness plus focusing would force a conjugate point onto a maximal geodesic, which a variational lemma forbids — so completeness fails.
- The generic condition is the quiet hero, ruling out a geodesic that threads a curvature-free tube forever and never focuses.
- Key idea: energy positivity and the generic condition force every complete causal geodesic to acquire conjugate points, but a maximal geodesic can hold none in its interior; given any one trigger — trapped surface, reconverging cone, or compact Cauchy surface — the only escape is incompleteness.
Energy positivity plus the generic condition force every complete causal geodesic to develop a conjugate point, yet a length-maximizing geodesic can contain none in its interior; given a single focusing trigger, that contradiction can be resolved only if some causal geodesic is incomplete.
Where this lives today
The theorem's key hypothesis has gone from thought experiment to observation. Trapped surfaces are what form when gravity wins, and the two flagship results of modern astrophysics probe regions where they are expected: LIGO and Virgo have recorded dozens of binary-black-hole mergers whose ringdown matches the Kerr geometry a trapped region settles to, and the Event Horizon Telescope has resolved the horizon-scale shadows of M87 and Sgr A. Each is a direct handle on a regime where the theorem guarantees classical general relativity predicts an incomplete interior — a place the theory itself flags as the boundary of its own validity. That is the deepest modern reading: the result does not merely assert singularities exist, it locates precisely where classical physics must hand off to quantum gravity. The questions it spawned — whether singularities are always hidden (weak cosmic censorship), whether deterministic evolution survives at Cauchy horizons (strong cosmic censorship), and what the generic approach to a singularity looks like (the BKL conjecture, now tested in numerical relativity) — remain among the most active in gravitation, all children of a theorem that proves a boundary exists without ever describing it.
The physical picture
Gravity, sourced by matter that obeys reasonable energy conditions, is universally attractive. Point a bundle of nearby freely-falling test particles — a geodesic congruence — inward, and tidal curvature keeps pulling its members together. The Raychaudhuri equation makes this quantitative: the expansion θ, the fractional rate of change of the bundle's cross-sectional volume, is driven downward by curvature, and once θ turns negative it must plunge to −∞ within a bounded stretch of affine parameter — a caustic, or conjugate point, where neighbouring geodesics cross.
A conjugate point is where focusing and maximality come to blows. Past one, a timelike geodesic is no longer the longest curve between its endpoints, and a null geodesic no longer lies on the boundary of a causal future — a nearby curve does better. The theorem exploits this by arranging, through a trigger, geodesics that start out converging; focusing forces the conjugate point, and if every such geodesic could be extended forever, one would be a maximal geodesic carrying a forbidden interior conjugate point. Something must give, and the only thing left to surrender is completeness: some geodesic simply stops.
The generic condition is the unsung guarantor of all this. It rules out the one escape a clever geodesic might attempt: coasting forever through a flat tube of spacetime, never feeling a transverse tidal tug, never focusing. In any realistic spacetime every causal geodesic eventually plows through some matter or radiation, and the generic condition is exactly the technical statement that curvature bites at least once along each ray, seeding the shear that drives focusing.
The core equations
The engine of the theorem is the Raychaudhuri equation for a timelike congruence. It says the expansion cannot help but fall: the self-focusing and shear terms are drags, the curvature term is a drag whenever gravity is attractive, and only vorticity — twisting — fights back.
where: θ — expansion scalar, the fractional rate of change of the congruence's cross-sectional volume per unit proper time; σ_ab — shear (symmetric, trace-free), the anisotropic distortion of the bundle; ω_ab — vorticity, the twist of the congruence, zero for a hypersurface-orthogonal flow; R_ab u^a u^b — the tidal focusing term; τ — proper time. Reading it: for an irrotational congruence (ω = 0) every surviving term is ≤ 0 once the curvature term is non-negative, so θ can only decrease. Dropping the two non-positive terms leaves a clean differential inequality, and integrating it bounds how far a converging bundle can go before it focuses. This is the focusing theorem — the single most important quantitative fact in the proof.
where: θ_0 — the initial (negative) expansion; τ — proper time from the initial slice. Reading it: any congruence starting even slightly converging is doomed to a conjugate point within a bounded proper time set purely by θ_0 — provided the geodesics survive that long. If they cannot, the spacetime is incomplete, which is the whole point. Null geodesics get their own Raychaudhuri equation, and the only change that matters is the coefficient: the transverse "screen" orthogonal to a null ray is two-dimensional, so the self-focusing term carries 1/2 instead of 1/3.
where: θ̂ — expansion of the null congruence in the 2D screen transverse to k^a; k^a — null tangent, k_a k^a = 0; λ — affine parameter; R_ab k^a k^b — the null focusing term. Reading it: a converging null bundle (θ̂_0 < 0) reaches a caustic within affine length 2/|θ̂_0| — the version Penrose used for collapse. The focusing terms are non-negative only if matter cooperates. The strong energy condition is the timelike requirement, the null energy condition its lightlike cousin; the theorem assumes curvature positivity for every causal tangent at once.
where: u^a — any timelike unit vector; k^a — any null vector; R_ab — the Ricci tensor. Reading it: via the trace-reversed Einstein equation R_ab = 8π(T_ab − ½ T g_ab) the SEC becomes (T_ab − ½ T g_ab)u^a u^b ≥ 0, which for a perfect fluid reads ρ + 3p ≥ 0 and ρ + p ≥ 0 — gravity focuses when pressure is not too negative. The generic condition closes the flat-tube loophole: somewhere along each causal geodesic the tidal part of the Riemann tensor is nonzero, so shear is inevitably kindled and fed into focusing.
where: k^a — the (null or timelike) tangent; R_abcd — the Riemann tensor. Reading it: no complete causal geodesic can dodge curvature forever; this guaranteed tidal encounter upgrades "gravity tends to focus" into "every complete geodesic actually focuses". A trigger supplies the initial convergence. The most physical is the closed trapped surface: a compact 2-surface so deep in a gravitational field that both its orthogonal light fronts are shrinking.
where: T — a compact spacelike 2-surface (e.g. a sphere inside a horizon); θ̂_± — expansions of the two future-directed null normal congruences. Reading it: in ordinary space outgoing light spreads (θ̂_+ > 0) while ingoing converges (θ̂_− < 0); a trapped surface has both converging, gravity so strong it drags even outgoing light inward. The conclusion is deliberately spare — not a curvature scalar, not a location, only a statement that some ride ends prematurely.
where: affine length — the range of the affine parameter (proper time, for a timelike geodesic), finite yet with no endpoint on the manifold. Reading it: this is the exact, coordinate-free meaning of "the spacetime is singular" — no singular point in the manifold, only a geodesic that cannot be continued.
The goal here is to turn the Raychaudhuri equation from a statement about a rate into a hard bound on distance: any timelike congruence beginning even slightly converging, where gravity is attractive, must reach a conjugate point within a proper time fixed entirely by its starting expansion. We keep only the terms that make focusing worse, recast the inequality in a variable that integrates trivially, and read off the finite bound — the same manoeuvre gives the null result with 3 replaced by 2.
Step 1. Take an irrotational congruence, ω_ab = 0, so the one positive term in equation (1) is gone. Why: congruences emanating from a slice or surface have zero twist, and dropping the only term that could raise θ is exactly what we are entitled to do.
Step 2. Discard the shear and curvature drags by inequality: since σ_ab σ^ab ≥ 0 and, under the strong energy condition, R_ab u^a u^b ≥ 0, both subtract, leaving dθ/dτ ≤ −θ²/3. Why: we only need an upper bound on dθ/dτ to prove θ runs away, and throwing away non-positive terms can only make the bound safer.
Step 3. Change variable to u ≡ 1/θ. Then du/dτ = −θ⁻²(dθ/dτ) ≥ −θ⁻²(−θ²/3) = 1/3, a constant-slope inequality. Why: the reciprocal linearizes the quadratic focusing term, converting a nonlinear ODE inequality into a straight-line bound that integrates in one step.
Step 4. Integrate from the initial slice: 1/θ(τ) ≥ 1/θ_0 + τ/3. With θ_0 < 0 the right side starts negative and climbs; it reaches zero — meaning θ → −∞ — no later than
Why: 1/θ passing through zero from below is precisely θ diverging to −∞, the analytic signature of a caustic where the congruence's cross-section collapses.
Step 5. Read the verdict. If the geodesics extend to τ_conj a conjugate point exists there; if not, they are incomplete. Why: this fork — conjugate point or incompleteness — is the hinge the theorem swings on, converting a focusing statement into the dichotomy the reductio exploits. *Verified: reproduces the timelike coefficient 3 (3D) and null coefficient 2 (2D screen); the FLRW check θ_0 = 3H_0 gives τ_conj = 1/H_0 = 13.9 Gyr, bracketing the matter-era age (2/3)/H_0 = 9.3 Gyr. Follows Wald §9.2 and Hawking–Ellis §4.4.*
Go deeper · Derivation B · 1 minWhy a maximal geodesic can carry no interior conjugate point
The aim of this derivation is the variational lemma that converts focusing into incompleteness: a geodesic that locally maximizes proper time between two points cannot contain a conjugate point strictly between them, because past one it can always be deformed into a strictly longer neighbouring timelike curve. This is the second, independent half of the argument — focusing supplies conjugate points, this lemma says maximal geodesics forbid them, and the collision is the theorem.
Step 1. Set up γ from p to q and suppose, for contradiction, a point r conjugate to p lies in the open interval (p, q). A conjugate point is where a nontrivial Jacobi field — a geodesic deviation vector obeying
— vanishes at both p and r. Why: the Jacobi field encodes how an infinitesimally displaced geodesic behaves, and its vanishing at two points is the exact statement that neighbouring geodesics refocus there.
Step 2. Examine the second variation of the proper-time functional along γ, the index form
built from allowed variation vectors V^a that vanish at the endpoints. Why: maximality of proper time is a statement about this second variation being non-positive for every admissible V, so any V making it positive breaks maximality.
Step 3. Construct the offending variation. Take the Jacobi field that vanishes at p and r, extend it by zero beyond r, and smooth the kink; this V^a can be shown to make δ²L > 0. Why: the rounded corner at r injects exactly the "slack" the second variation needs to turn positive — a concrete nearby timelike curve strictly longer than γ.
Step 4. Collect the contradiction: γ was assumed maximal, yet we exhibited a longer neighbour. Hence no interior conjugate point can exist. Why: this is the precise fact the reductio needs; the null analogue states a generator of ∂I⁺(S) leaves the boundary at its first caustic. *Verified: standard result; second-variation/index-form argument follows Wald §9.3 and Hawking–Ellis Proposition 4.5.8, sign conventions per Wald.*
Go deeper · Derivation C · 1 minThe reductio that forces incompleteness
The final derivation assembles the two halves into the theorem itself: assume the energy, generic, causality, and trigger hypotheses and also causal geodesic completeness, then derive a flat contradiction, so completeness must be abandoned. This is the logical skeleton of the 1970 proof, and every hypothesis earns its place by blocking one escape route.
Step 1. Use completeness and the generic condition to place conjugate points everywhere. Completeness extends every causal geodesic to any affine value; the energy condition supplies focusing and the generic condition supplies the tidal encounter, so every complete causal geodesic acquires a pair of conjugate points. Why: without completeness the focusing bound τ ≤ 3/|θ_0| could be evaded by the geodesic simply ending — precisely the outcome we are trying to force.
Step 2. Use the trigger and causality to produce a maximal geodesic. The trigger plus the chronology and no-almost-closed-causal-curve conditions make a suitable set of causal curves compact, so a limit-curve argument yields a longest — maximal — causal geodesic between the relevant sets. Why: the causality conditions guarantee the space of causal curves is well-behaved enough for a maximum to be attained, not merely approached.
Step 3. Apply the variational lemma: this maximal geodesic can hold no conjugate point in its interior. Why: the second half derived above, which directly contradicts Step 1, insisting every complete causal geodesic has conjugate points.
Step 4. Discharge the assumption. Steps 1 and 3 cannot both hold, so with the other hypotheses kept, the only false premise available is completeness. Therefore
Why: this is the whole theorem in one line — remove completeness and the contradiction dissolves, so the spacetime must be causally geodesically incomplete.
Step 5. State the conclusion: there exists an inextendible timelike or null geodesic of finite affine length. Why: that is the theorem's guarantee — incompleteness, nothing more or less, is what the hypotheses buy. *Verified: structure follows Hawking & Penrose, Proc. R. Soc. A 314 (1970) 529; see Hawking–Ellis Theorem 8.2 and Senovilla's 1998 review.*
Limiting cases
Each hypothesis can fail, and each failure names a real spacetime that escapes the conclusion — a useful way to see that no assumption is idle.
| Case | Result | Meaning |
|---|---|---|
| Strong energy condition violated | R_ab u^a u^b ≥ 0 fails | A positive cosmological constant or an inflaton's potential energy makes gravity effectively repulsive; de Sitter space evades timelike focusing and is geodesically complete. |
| Causality violated | closed timelike curve exists | Gödel's universe and the deep interior of maximal Kerr contain CTCs, breaking the chronology condition; the theorem does not apply. |
| No trigger present | no trapped surface, no reconverging cone, no compact Cauchy surface | Minkowski space meets the energy and causality conditions but seeds no convergence, so it is correctly complete; the theorem predicts nothing without a trigger. |
| Generic condition relaxed | tidal term = 0 along a whole geodesic | A ray that never meets curvature need not focus; the generic condition excludes these measure-zero specials, restoring focusing for every causal geodesic. |
| All hypotheses hold, trigger present | incomplete | Collapsing star (trapped surface) or hot Big Bang (reconverging cone) — the theorem bites and a geodesic must end. |
Worked examples
1 · The focusing bound on the age of the universe (fully worked)
Model the cosmos as the comoving geodesic congruence of a matter-dominated FLRW universe with H₀ = 70 km s⁻¹ Mpc⁻¹. Use the focusing theorem to bound how far into the past a singularity must lie, and compare with the actual age.
Step 1. Identify the expansion. For the comoving congruence θ = 3ȧ/a = 3H, isotropic and shear-free, so today θ₀ = 3H₀. Converting, H₀ = 70/(3.086×10¹⁹) = 2.27×10⁻¹⁸ s⁻¹, so θ₀ = 3H₀ = 6.81×10⁻¹⁸ s⁻¹.
Step 2. Run the congruence backward. Toward the past the bundle converges, so the focusing theorem (equation 2) applies with |θ₀| = 3H₀ and places a conjugate point — the initial singularity — within proper time
Step 3. Evaluate: 1/H₀ = 4.41×10¹⁷ s = 13.9 Gyr. The theorem guarantees a past singularity no more than one Hubble time ago, with no assumption of homogeneity — the Big Bang is generic, not an FLRW symmetry artifact.
Step 4. Sanity-check. A matter-dominated universe has true age (2/3)/H₀ = 9.3 Gyr, inside the bound. The focusing bound 1/H₀ ≈ 13.9 Gyr correctly brackets the actual age (2/3)/H₀ ≈ 9.3 Gyr, the gap being the slack from discarding shear and higher-order terms.
2 · Finite affine length inside a black hole (fully worked)
A closed trapped surface makes the interior incomplete. Make "finite affine length" tangible: compute the longest proper time an infaller can survive between crossing a Schwarzschild horizon and reaching the singularity, τ_max = πGM/c³, for a stellar and a supermassive black hole.
Step 1. Anchor the formula. Inside Schwarzschild the radial coordinate becomes timelike, every future-directed causal curve is dragged to r = 0, and the maximal proper time there — achieved by free fall, not fighting — is τ_max = πGM/c³. Firing rockets only shortens it.
Step 2. Stellar-mass, M = 1 M_⊙ = 1.99×10³⁰ kg: τ_max = π(6.674×10⁻¹¹)(1.99×10³⁰)/(2.998×10⁸)³ = 1.55×10⁻⁵ s ≈ 15 μs. A geodesic entering the horizon reaches the end of its affine parameter in fifteen microseconds.
Step 3. Supermassive M87, M = 6.5×10⁹ M_⊙: proper time scales linearly with mass, τ_max = 15.5 μs × 6.5×10⁹ ≈ <strong>28 hours</strong>. For Sgr A (4.15×10⁶ M_⊙) the same scaling gives ≈ 64 s.
Step 4. Read the lesson. These are not coordinate artifacts removable by a better chart: the affine parameter itself runs out. A worldline entering a solar-mass black hole is inextendible past ≈ 15 μs, and past ≈ 28 hours for M87* — "finite affine length" made into a stopwatch reading.
Physics of the portrait
The theorem carries two names, and while Penrose supplied the trapped surface and the null focusing that cracked collapse open in 1965, it was his collaboration with Stephen Hawking through the late 1960s — Hawking pushing the timelike focusing into cosmology — that produced the single 1970 statement covering both. The portrait here is Hawking, whose cosmological singularity theorems and generic-condition machinery are the half that reaches back to the Big Bang.
Origins · 1 minStephen Hawking and how this was discovered
Stephen Hawking began his research career in 1965 just as Penrose's collapse theorem appeared, and his thesis did the decisive thing with it: reverse the time direction and apply the same focusing logic to cosmology, proving that an expanding universe obeying the strong energy condition must be past-incomplete — a Big Bang forced by geometry, not assumed by symmetry. Over the next five years he sharpened the tools, introducing the generic condition and the global techniques that let one theorem handle both a converging future (collapse) and a converging past (cosmology), and in 1970 he and Penrose published the general theorem that bears both names. Hawking then turned the same causal-structure methods on black holes: the area theorem (1971), the four laws of black-hole mechanics with Bardeen and Carter (1973), and above all the 1974 discovery that black holes radiate thermally, uniting gravity, quantum theory, and thermodynamics. His book with Ellis, The Large Scale Structure of Space-Time (1973), codified the whole global-geometric apparatus — trapped surfaces, Cauchy horizons, conformal boundaries, energy conditions — that remains the working language of mathematical relativity. He did all of it while living for decades with motor neurone disease, and though the Nobel went to Penrose alone in 2020 (Hawking having died in 2018, the prize not awarded posthumously), the singularity theorems are as much Hawking's monument as anyone's.
Going further
Texts. Wald, General Relativity (1984), Chapter 9 — the cleanest modern proof · Hawking & Ellis, The Large Scale Structure of Space-Time (1973), Chapters 4 and 8 — the definitive treatment · Misner, Thorne & Wheeler, Gravitation (1973), Chapter 34 · Carroll, Spacetime and Geometry (2004), Chapter 9 · Senovilla, Singularity Theorems and Their Consequences, Gen. Rel. Grav. 30 (1998) 701 · Poisson, A Relativist's Toolkit (2004), Chapter 2 for the congruence machinery.
Provenance. The theorem and its logical structure follow Hawking & Penrose, Proc. R. Soc. A 314 (1970) 529, with the collapse special case from Penrose, Phys. Rev. Lett. 14 (1965) 57. A verifier should confirm: (i) the timelike Raychaudhuri coefficient 1/3 and null coefficient 1/2 from 3D versus 2D transverse spaces; (ii) the conjugate-point bound τ ≤ 3/|θ₀| by integrating d(1/θ)/dτ ≥ 1/3; (iii) the trace-reversed Einstein equation R_ab = 8π(T_ab − ½ T g_ab) and the perfect-fluid SEC forms ρ + 3p ≥ 0, ρ + p ≥ 0; (iv) the FLRW focusing bound 1/H₀ = 13.9 Gyr against the matter-era age (2/3)/H₀ = 9.3 Gyr for H₀ = 70 km s⁻¹ Mpc⁻¹; (v) the Schwarzschild survival time τ_max = πGM/c³ = 15 μs for one solar mass, 28 hours for M87, and 64 s for Sgr A; (vi) the variational lemma against Wald §9.3 and Hawking–Ellis Prop. 4.5.8. The 2020 Nobel award to Penrose for the singularity prediction is a verifiable historical fact.
Prerequisites. Causal Structure of Spacetime · Null Geodesics and Light Cones · Global Hyperbolicity · Achronal Sets and Edges · Causality Conditions Hierarchy. Leads to. Domain of Dependence and Cauchy Horizons · Penrose–Carter Conformal Diagrams · Conformal Infinity.
Ask the tutor: Ask me to walk through the reductio step by step, or to show where inflation slips through the strong energy condition.
What exactly does "singularity" mean in this theorem?
It means causal geodesic incompleteness: some timelike or null geodesic cannot be extended past a finite value of its affine parameter (proper time, for a timelike geodesic). Nothing more. The theorem does not claim any curvature scalar diverges, does not locate a "singular point" — there is none in the manifold, which is already the maximally extended spacetime — and does not say whether the singularity is spacelike, null, hidden, or naked. Incompleteness is the rigorous, coordinate-free replacement for the vague older idea of "a place where the theory breaks down," strong enough to be provable from global hypotheses without solving Einstein's equations.
Why does the theorem need both the strong and null energy conditions?
Because it focuses two kinds of geodesic. Timelike focusing runs on R_ab u^a u^b, non-negative precisely under the strong energy condition (SEC); null focusing runs on R_ab k^a k^b, controlled by the weaker null energy condition (NEC). Since the contradiction may be built on whichever geodesic type the trigger supplies, the theorem assumes R_ab V^a V^b ≥ 0 for every causal V^a at once, covering both. In practice the SEC is the demanding one — a cosmological constant or an inflaton can violate it — while the NEC is far more robust and is what Penrose's pure-collapse theorem needs.
How is this different from Penrose's original 1965 theorem?
Penrose's 1965 result was about gravitational collapse specifically: given a closed trapped surface, the null energy condition, and a non-compact Cauchy surface (an "open" universe), the spacetime is null-geodesically incomplete. Powerful but narrow — it assumes global hyperbolicity and a single trigger. The 1970 theorem drops global hyperbolicity, weakens the causality requirement to the chronology condition plus genericity, and admits three alternative triggers, one of which (a compact Cauchy surface) covers the spatially closed cosmologies Penrose's theorem excluded. The price is the extra generic condition; the reward is a single statement handling collapse and the Big Bang together.
Does inflation let the universe escape a beginning?
Partly, and this is a genuine subtlety. Inflation is driven by a scalar field whose potential energy violates the strong energy condition, so the timelike focusing that forces the Hawking–Penrose conclusion is suspended during inflation — the theorem as stated does not bite, which is why naive "the theorem proves the Big Bang" claims need care. However, the Borde–Guth–Vilenkin theorem (2003) shows any universe with average past expansion rate greater than zero is still past-incomplete, inflation included, by a kinematic argument that sidesteps energy conditions entirely. So inflation changes the character of the initial boundary but does not, by present understanding, remove past-incompleteness.
Why is the generic condition called the "unsung hero"?
Because without it the focusing argument has a loophole, yet it is the hypothesis people most often forget. Focusing under an energy condition guarantees gravity tends to converge a bundle, but a single geodesic could thread a curvature-free tube of spacetime forever, never meeting the transverse tidal force needed to kindle shear, and so never focus. The generic condition asserts that somewhere along each causal geodesic the tidal part of the Riemann tensor is nonzero — every ray eventually meets some matter or radiation. It upgrades "gravity tends to focus" into the airtight "every complete causal geodesic does focus," which is what the reductio requires.
If the theorem is silent on curvature, why do we believe singularities are violent?
Because separate lines of evidence, not the theorem, tell us so. Exact solutions — Schwarzschild, Kerr, FLRW — show genuine curvature-scalar blow-ups at their singularities, and the BKL (Belinskii–Khalatnikov–Lifshitz) conjecture, supported by numerical relativity, argues the generic approach is a chaotic oscillatory tidal churning in which curvature diverges. The theorem provides the existence guarantee — something must go wrong — while this other work characterizes how. Its generality comes precisely from refusing to assume any particular curvature behaviour, which is what lets one proof cover collapse and cosmology at once.