The Proof in an Hour

The last milestone of the Path is to explain the whole proof to someone else in an hour. It is a real test. A talk forces choices: what to show, what to state without proof, what to leave out, and where the audience will stop you. If the guide's seams have held, a one-hour version of the proof can be written from the guide alone, with one figure per slide and every claim checkable in a chapter. This chapter is that talk: twelve slides, a timing for each, three sentences of script, the figure to show, and the questions an audience is likely to ask, with answers.

By the end of this chapter you will be able to:

• give a one-hour talk on the proof of the Poincaré conjecture, with a figure on every slide;
• say for each slide which chapter supports it, and what it states without proof;
• answer the common questions about the proof, including the hostile ones;
• shorten the talk to ten minutes, or adapt it to an audience that knows only calculus.

6.1Planning a route

6.2The talk

The talk is sixty minutes: fifty for the slides and ten for questions, with the slide times adding to fifty. Slide figures are drawn from the chapters named, where their captions explain them.

slide	minutes	title	show	chapters
1	3	The question	the lasso test; $S^3$ as two balls	7A.9 The Poincaré Conjecture, Precisely
2	4	The idea: a heat equation for shape	the shrinking sphere	11A.1 The Equation and Its First Solutions
3	4	What the flow does to curvature	the curvature ODE portrait	11A.4 Maximum Principles under Ricci Flow, 11A.5 Hamilton–Ivey Pinching
4	4	Singularities	the neckpinch	11B.4 Singularities
5	3	Hamilton's program and its gaps	the four gaps of 2002	11B.5 Hamilton’s Program in 2002
6	5	Perelman's entropy	$\mathcal W$ for Gaussians	12A.3 The 𝓦-Entropy
7	4	No collapse, no cigar	$\mathcal W$ against volume	12A.4 κ-Noncollapsing
8	5	The models: necks and caps	catalogue of $\kappa$-solutions	12B.1 κ-Solutions–12B.3 The Canonical Neighbourhood Theorem
9	5	Surgery	surgery on a horn	12B.4 Surgery, 12B.5 Ricci Flow with Surgery for All Time
10	5	Extinction	the width bound	12C.2 Finite Extinction
11	4	Reading off $S^3$	the logic chain	12C.1 Reading Off the Topology, 12C.3 The Poincaré Conjecture, Assembled
12	4	Geometrization and after	the fates of pieces	12C.4 Geometrization, 12C.5 After Perelman

Figure 6.1 shows the timing, and Figure 6.2, the ten steps of 12C.3 The Poincaré Conjecture, Assembled, is the map to keep on screen during questions.

6.2.1The script

Slide 1. The question (3 minutes). A closed 3-manifold is a finite universe without boundary; it is simply connected if every loop in it can be shrunk to a point. Poincaré asked in 1904 whether every such space is the 3-sphere, the space made of two solid balls glued along their boundary spheres. In two dimensions the answer is a classical yes, and in three it took a century and a new method.

Slide 2. The idea (4 minutes). Hamilton's idea, in 1982, was to put any metric on the manifold and let it evolve by the Ricci flow, $\partial_tg = -2\operatorname{Ric}$, a heat equation for the shape. A round sphere shrinks homothetically, with $r^2 = r_0^2 - 4t$, and vanishes in finite time. The hope is that any metric on a simply connected manifold flows towards something round, so that the manifold was a sphere all along.

Slide 3. What the flow does to curvature (4 minutes). Curvature evolves by a reaction–diffusion equation, so the maximum principle controls it, and in dimension three Hamilton and Ivey showed that wherever curvature is large, it is nearly positive. Hamilton proved in 1982 that positive Ricci curvature flows to a round metric. In general, though, the flow does not simply round things off.

Slide 4. Singularities (4 minutes). Where a manifold has a thin neck, the neck can pinch off in finite time while the rest stays smooth: a neckpinch. At a singularity, curvature blows up, and to understand it you zoom in, rescaling so that the curvature is $1$ and taking a limit. The limits are the possible shapes of a singularity, and the proof depends on knowing them.

Slide 5. Hamilton's program and its gaps (3 minutes). By the late 1990s Hamilton had the flow, its estimates, the Harnack inequality, compactness and a surgery procedure in four dimensions. Four things were missing: control on how thin things can get, a list of singularity models, control of surgery, and a route from the flow to topology. Perelman's three preprints of 2002–03 supplied all four.

Slide 6. Perelman's entropy (5 minutes). Perelman found that the Ricci flow is a gradient flow, and built an entropy $\mathcal W$ that only increases along it and is constant exactly on shrinking solitons. On flat space, the statement that $\mathcal W \geq 0$ is the Gaussian logarithmic Sobolev inequality: the Gaussian is the equality case. This monotone quantity is the analytic heart of the proof.

Slide 7. No collapse, no cigar (4 minutes). A ball that is very thin for its curvature supports a test function with very negative entropy, while monotonicity bounds the entropy below by its value at time zero. So no region ever becomes thin at its own scale: the flow is noncollapsed. That rules out the cigar, the shape Hamilton could not exclude, and makes limits of zoomed-in flows exist.

Slide 8. The models (5 minutes). The zoomed-in limits are $\kappa$-solutions: ancient, nonnegatively curved and noncollapsed, such as shrinking spheres, cylinders and the Bryant soliton. Perelman showed they form a compact family in which every point lies in a neck, a cap or a small closed piece. So in any three-dimensional Ricci flow, every point of high curvature looks like a neck or a cap: the canonical neighbourhood theorem.

Slide 9. Surgery (5 minutes). At a singular time, cut each thin horn along a neck, throw the horn away, glue in a standard cap, and continue. Perelman chose the parameters so that the canonical neighbourhoods and noncollapsing survive each surgery, and each surgery removes a definite volume, so surgeries cannot pile up. The flow with surgery runs for all time.

Slide 10. Extinction (5 minutes). On a simply connected manifold, there is a nontrivial family of 2-spheres sweeping it out, and the area of the largest sphere in the best such family is positive. Under the flow, that area drops at a rate fixed by Gauss–Bonnet, $W' \leq -4\pi + \frac{3}{4(t + C)}W$, so it would become negative in finite time. Therefore the flow must end first: the manifold becomes extinct.

Slide 11. Reading off $S^3$ (4 minutes). Every surgery is undone by a connected sum, and every piece that was removed or vanished is a spherical space form or $S^2\times S^1$. So the manifold is a connected sum of such pieces, and its fundamental group is the free product of theirs. Simple connectivity leaves only $S^3$: the manifold was the 3-sphere.

Slide 12. Geometrization and after (4 minutes). For a general 3-manifold, the flow with surgery may run forever, and in the long run it splits the manifold into hyperbolic pieces and graph manifolds: Thurston's geometrization conjecture, also proved by Perelman. Since then, the flow through singularities has been made canonical, the singularity models classified, and the methods carried to higher dimensions. The Poincaré conjecture was the first problem the Ricci flow solved completely, but not the last.

6.3Questions from the audience

These are the questions most often asked after such a talk, with answers drawn from the guide.

"Why should a heat equation tell you anything about topology?" Because the flow is determined by the metric, and the metric lives on the manifold, so whatever the flow does has to be compatible with the topology. If the flow ends with the manifold vanishing in finite time, having only removed understood pieces, the topology is forced. The heat equation is the means; the topology is read from what it can and cannot do (12C.1 Reading Off the Topology).

"Isn't surgery cheating? You changed the manifold." Surgery changes it in a recorded way: every surgery is undone by a connected sum, so the original manifold is reconstructed from the pieces (12B.4 Surgery). It is the prime decomposition carried out geometrically.

"How do you know nothing worse than necks and caps can happen?" That is the canonical neighbourhood theorem: noncollapsing makes limits exist, Hamilton–Ivey makes them nonnegatively curved, and Perelman's classification of $\kappa$-solutions shows that every point of one lies in a neck, a cap or a small closed piece (12B.3 The Canonical Neighbourhood Theorem).

"Where did you use that the manifold is simply connected?" Only at the end: to make the sweepouts by spheres nontrivial, so that extinction is forced, and to discard every summand but $S^3$. Everything before applies to every closed 3-manifold (12C.3 The Poincaré Conjecture, Assembled).

"Why doesn't this work in dimension four?" Several steps use dimension three: the Hamilton–Ivey pinching estimate, the classification of singularity models, and necks of the form $S^2\times\mathbb{R}$ whose central spheres are the spheres of the prime decomposition. In dimension four the singularities are far less understood, and the smooth four-dimensional Poincaré conjecture is open; the topological version was proved by Freedman by other methods (12C.5 After Perelman).

"How do we know the proof is correct?" Three independent groups, Kleiner and Lott, Morgan and Tian, and Cao and Zhu, wrote complete expositions between 2003 and 2006, filling in every step; the community accepted the proof by 2006; and two decades of work have built on it, including results Perelman conjectured and others then proved (12C.5 After Perelman).

"Did Perelman need geometrization to prove the Poincaré conjecture?" No. For a simply connected manifold the flow becomes extinct, so the long-time analysis and the collapsing theorem are not needed (12C.3 The Poincaré Conjecture, Assembled).

"Why did Perelman decline the prizes?" He declined both the Fields Medal (2006) and the Millennium Prize (2010). His reasons are his own, and this guide does not speculate about them.

6.4History

Perelman explained his work in person in a series of lectures in the United States in April 2003, between his second and third preprints; Colding and Minicozzi's paper records a question he asked at a dinner in New York that month (12C.2 Finite Extinction). Since then the proof has been taught in courses and summer schools, written up in the expositions and in surveys such as Morgan's, and explained in many one-hour talks of the kind this chapter scripts.

6.5Exercises

Solution

One choice: slides 2 (the flow), 7 (noncollapsing, as the representative of Perelman's analysis), 9 (surgery) and 11 (reading off $S^3$). Bridges: after 2, "where the flow develops singularities, Perelman showed they look like necks and caps, because nothing collapses"; after 7, "so one can cut along the necks"; after 9, "on a simply connected manifold the flow eventually dies out, because an area shrinks at a fixed rate"; 11 concludes. The choice keeps the three acts (Hamilton's flow, Perelman's analysis, the topology) and the single most important new idea.

Solution

Slide 1: replace "closed 3-manifold" by "a finite universe without edges" and "simply connected" by the lasso test, with the 2-sphere and the torus as examples. Slide 2: the shrinking round sphere, computed with the ODE $\frac{d}{dt}r^2 = -4$. Slide 6: the Gaussian and the heat equation, which they know from calculus. Slide 10: the shrinking circle, whose enclosed area drops at rate $2\pi$ under curve shortening. Slides 8 and 9: pictures only, a dumbbell pinching and being cut.

Solution

Perelman's entropy is a scale-dependent log-Sobolev functional for the manifold, and along the Ricci flow its minimum over test functions only increases. On a region that is very thin compared with its curvature, a cut-off function makes this functional as negative as you like, because the region's volume is small compared with Euclidean balls of the same radius. Since the minimum cannot drop below its value at time zero, no such region ever forms: the flow is noncollapsed.

Solution

Every step has been written out in full, independently, by three groups of experts (Kleiner–Lott, Morgan–Tian, Cao–Zhu), and the earlier ingredients, Hamilton's theorems and the classical topology, have been in use for decades. The proof has also been built on for twenty years, by people who would have found a gap if there were one; that is the same standard of evidence by which any long proof in mathematics is accepted.

Solution

There is no model answer: this exercise is the milestone itself. If you ran over, the usual culprits are slides 6 and 8, where the temptation is to prove too much; state the result, show the figure, and move on. If a question stumped you, the chapter column of the table says where to look.