03.02.24 · differential-geometry / manifolds

The generalised Poincaré conjecture in high dimensions

shipped3 tiersLean: none

Anchor (Master): Milnor Lectures on the h-Cobordism Theorem §§6-9; Smale Generalized Poincaré's conjecture (1961); Kervaire-Milnor Groups of homotopy spheres I (1963); Freedman The topology of four-dimensional manifolds (1982); Morgan-Tian Ricci Flow and the Poincaré Conjecture (2007)

Intuition Beginner

A sphere can be split cleanly into two bowls. Take an ordinary ball-shaped surface, slice it along its equator, and you hold an upper cap and a lower cap, each a curved disc, joined along a circle. Run this in reverse and the recipe for a sphere becomes plain: take two discs and glue them along their rims. The same recipe works one dimension up and beyond — a higher sphere is two higher discs glued along their common rim.

Now suppose someone hands you a shape that passes every loop-and-hole test a sphere passes: it has no holes, every loop on it shrinks to a point, and it matches the sphere in all the bending-and-stretching bookkeeping. Such a shape is called a homotopy sphere. The question is whether it really is a sphere, or only a clever impostor that fools every test. In high enough dimensions the answer is that it must be a sphere, at least as far as continuous reshaping can tell.

The proof reuses the slab result from the previous unit. Poke two small disc-shaped holes in the suspect shape and look at what is left. That leftover is a slab between two rims, and because the suspect passed all the tests, the slab is interchangeable at its two rims — exactly the setup the slab theorem needs. In high dimensions the slab is then a plain thickened rim with nothing hidden inside. So the suspect shape is just two discs with a plain collar between them: two discs glued along a rim, the recipe for a sphere.

There is one piece of fine print. The gluing that closes up the two discs can be a little crooked. Continuous reshaping can always straighten it, so the shape is a sphere to the eye. A perfectly smooth straightening can fail, and that failure is the story of exotic spheres.

Visual Beginner

A round sphere is drawn on the left, sliced along its equator into a top cap and a bottom cap, with a curved arrow showing the two caps pulling apart into two separate discs that share a circular rim. The message of the left panel: a sphere is two discs glued along a rim.

On the right, a lumpy blob labelled "homotopy sphere" has two small discs removed, leaving two round holes. The leftover middle is shaded as a slab between the two hole-rims, with upward flow arrows linking the two rims to show they correspond. A bracket points from this slab to the words "product collar," and a final arrow reassembles the two discs plus the straightened collar back into a clean round sphere.

The takeaway from the picture: punch two disc holes in a high-dimensional homotopy sphere, the leftover slab is a plain product collar, and so the whole shape is two discs joined by a collar — a sphere up to continuous reshaping.

Worked example Beginner

Building the ordinary 2-sphere from two discs. Take the lower half of a globe and the upper half. Each half is a curved disc, and the two rims are the same equator circle. Glue them rim to rim by matching each point of one rim to the same-longitude point of the other, and the globe closes up into a full sphere. This is the recipe in its plainest form: two discs, one rim, glued straight across. The collar in between — a thin band around the equator — is just the equator circle times a short width, a plain product band.

A crooked gluing still gives a sphere. Glue the two rims with a twist: match each rim point not to its own longitude but to one rotated by some angle. The closed-up shape is still a sphere, because you can slowly unwind the twist as you sweep from the rim inward across a disc, smoothing the crookedness away. The twist made the gluing crooked, yet a sphere came out all the same.

What this tells us. The high-dimensional theorem says exactly this picture holds in five dimensions and up: a homotopy sphere is two discs joined by a plain product collar, glued by a rim map that may be crooked. Continuous unwinding straightens the crookedness, so the shape is a sphere to the eye. The bridge from the globe to the theorem is the slab result: it is what guarantees the middle really is a plain product collar and not something with hidden structure.

Check your understanding Beginner

Formal definition Intermediate+

Throughout, $Σ^{n}$ is a closed smooth manifold of dimension $n$ . We write $D^{n}$ for the standard closed $n$ -disc and $S^{n - 1} = \partial D^{n}$ for its boundary sphere.

Definition (homotopy sphere). A closed smooth $n$ -manifold $Σ^{n}$ is a homotopy $n$ -sphere when it is homotopy equivalent to $S^{n}$ . For $n \geq 2$ this is the same as asking that $Σ$ be simply connected with the integral homology of the sphere, $H_{*} (Σ; Z) = H_{*} (S^{n}; Z)$ : simple connectivity plus the homology condition force a homotopy equivalence by the Hurewicz and Whitehead theorems.

Definition (twisted sphere). Fix an orientation-preserving diffeomorphism $g : S^{n - 1} \to S^{n - 1}$ . The twisted sphere $Σ_{g}$ is the smooth manifold obtained from two copies $D_{0}^{n}, D_{1}^{n}$ of the disc by gluing their boundaries along $g$ : $Σ_{g} = D_{0}^{n} \cup_{g} D_{1}^{n}$ , where a point $x \in \partial D_{0}^{n} = S^{n - 1}$ is identified with $g (x) \in \partial D_{1}^{n} = S^{n - 1}$ . When $g$ is the identity, $Σ_{g} = S^{n}$ .

Smale's theorem (statement). Let $Σ^{n}$ be a smooth homotopy $n$ -sphere with $n \geq 5$ . Then $Σ^{n}$ is homeomorphic to $S^{n}$ ; indeed it is a twisted sphere $Σ_{g}$ for some orientation-preserving diffeomorphism $g$ of $S^{n - 1}$ , and every twisted sphere is homeomorphic to $S^{n}$ . ^{[Smale 1961]}

The topological versus smooth distinction. The conclusion is homeomorphic, not diffeomorphic. A twisted sphere $Σ_{g}$ is homeomorphic to $S^{n}$ for every $g$ , but $Σ_{g}$ is diffeomorphic to $S^{n}$ exactly when the gluing $g$ extends to a diffeomorphism of the disc $D^{n}$ — and that extension can fail. The classes of twisted spheres up to orientation-preserving diffeomorphism form the group $Θ_{n}$ of homotopy $n$ -spheres under connected sum, computed by Kervaire and Milnor ^{[KM 1963]}; $Θ_{n} = 0$ would say every homotopy $n$ -sphere is the standard smooth sphere, while $Θ_{n} \neq = 0$ records exotic spheres. The first exotic example is $Θ_{7} = Z /28$ , the seven-dimensional case of 03.06.17.

Counterexamples to common slips

The hypothesis is $n \geq 5$ , not $n \geq 6$ . The h-cobordism theorem of 03.02.23 needs the cobordism to have dimension $\geq 6$ , but the cobordism here is the twice-punctured $Σ^{n}$ of dimension $n$ , so $n \geq 6$ is what the punctured argument uses directly; the value $n = 5$ is recovered by a dual argument or by the engulfing methods of Stallings and Zeeman. The corollary's clean floor is $n \geq 5$ .
"Homeomorphic to $S^{n}$ " does not say "diffeomorphic to $S^{n}$ ." Conflating the two erases the entire theory of exotic spheres. Smale's theorem is silent on the smooth classification; that is the content of $Θ_{n}$ .
A homotopy sphere is not assumed to be a twisted sphere at the start — that it is one is part of the conclusion, supplied by the product structure of the punctured h-cobordism.

Key theorem with proof Intermediate+

We deduce Smale's theorem from the h-cobordism theorem of 03.02.23, whose statement we use as a black box: a compact simply connected h-cobordism of dimension $\geq 6$ is a product.

Theorem (high-dimensional Poincaré conjecture, smooth category, LHC §8). A smooth homotopy $n$ -sphere $Σ^{n}$ with $n \geq 6$ is a twisted sphere, hence homeomorphic to $S^{n}$ . ^{[Smale 1961]}

Proof. Choose two disjoint smoothly embedded closed $n$ -discs $D_{0}, D_{1} \subset Σ$ ; such discs exist because $Σ$ is a smooth manifold and any point has a disc neighbourhood, and two distinct points have disjoint ones. Set $W = Σ ∖ (\overset{˚}{D}_{0} \cup \overset{˚}{D}_{1}),$ the complement of the two open discs. Then $W$ is a compact smooth $n$ -manifold with boundary $\partial W = V_{0} ⊔ V_{1}$ , where $V_{0} = \partial D_{0}$ and $V_{1} = \partial D_{1}$ are each diffeomorphic to $S^{n - 1}$ . So $(W; V_{0}, V_{1})$ is a cobordism of dimension $n$ .

Step 1 — $W$ is simply connected. Since $n \geq 3$ , removing two points (and a fortiori two open discs) from the simply connected $Σ$ leaves a simply connected complement: $π_{1} (W) = π_{1} (Σ) = 1$ , because a loop in $W$ is a loop in $Σ$ , where it bounds a disc, and that disc can be pushed off the two removed $n$ -discs in codimension $\geq 3$ .

Step 2 — the inclusions are homology isomorphisms. Compute $H_{*} (W)$ by Mayer-Vietoris for $Σ = W \cup (D_{0} ⊔ D_{1})$ , with intersection $W \cap (D_{0} ⊔ D_{1})$ a collar neighbourhood of $V_{0} ⊔ V_{1}$ , homotopy equivalent to $S^{n - 1} ⊔ S^{n - 1}$ . The discs are contractible, so the sequence reads $\dots \to H_{k} (S^{n - 1} ⊔ S^{n - 1}) \to H_{k} (W) \oplus H_{k} (pt ⊔ pt) \to H_{k} (Σ) \to \dots .$ Feeding in $H_{*} (Σ) = H_{*} (S^{n})$ and $H_{*} (S^{n - 1} ⊔ S^{n - 1})$ shows $H_{*} (W, V_{0}) = 0$ and $H_{*} (W, V_{1}) = 0$ for $1 \leq n - 1$ , i.e. the boundary spheres carry all the homology of $W$ . With Step 1 and the relative Hurewicz theorem, the inclusions $V_{0} ↪ W$ and $V_{1} ↪ W$ are homotopy equivalences. Thus $(W; V_{0}, V_{1})$ is an h-cobordism.

Step 3 — apply the h-cobordism theorem. Here $dim W = n \geq 6$ and $W, V_{0}, V_{1}$ are simply connected ( $V_{i} ≅ S^{n - 1}$ with $n - 1 \geq 5$ ). The h-cobordism theorem of 03.02.23 gives a diffeomorphism $W ≅ V_{0} \times [0, 1] ≅ S^{n - 1} \times [0, 1],$ a product collar carrying $V_{0}$ to $S^{n - 1} \times {0}$ and $V_{1}$ to $S^{n - 1} \times {1}$ .

Step 4 — reassemble into a twisted sphere. Now $Σ = D_{0} \cup_{ϕ} (S^{n - 1} \times [0, 1]) \cup_{ψ} D_{1}$ , two discs capped onto the ends of the product collar by the boundary identifications $ϕ, ψ$ . Absorbing the collar $S^{n - 1} \times [0, 1]$ into the disc $D_{1}$ (a collar is a thickened boundary and merges smoothly into the disc) presents $Σ$ as $D_{0} \cup_{g} D_{1}$ for the orientation-preserving diffeomorphism $g = ψ \circ ϕ^{- 1}$ of $S^{n - 1}$ . Hence $Σ = Σ_{g}$ is a twisted sphere.

Step 5 — every twisted sphere is homeomorphic to $S^{n}$ . It remains to show $Σ_{g} ≅ S^{n}$ topologically. Define the Alexander trick extension of $g$ : the radial coning $\overset{g}{^} : D^{n} \to D^{n}, \overset{g}{^} (0) = 0, \overset{g}{^} (t x) = t g (x) (0 < t \leq 1, x \in S^{n - 1}) .$ This $\overset{g}{^}$ is a homeomorphism of $D^{n}$ restricting to $g$ on the boundary. Using $\overset{g}{^}$ to identify the gluing $g$ with the identity gluing produces a homeomorphism $Σ_{g} ≅ D^{n} \cup_{id} D^{n} = S^{n}$ . $□$

Bridge. Putting these together, the high-dimensional Poincaré conjecture is the h-cobordism theorem of 03.02.23 read off a punctured sphere: removing two discs turns a homotopy sphere into an h-cobordism, and the product structure of that h-cobordism is exactly the statement that the sphere is two discs joined by a collar. The foundational reason the argument runs only in high dimensions is the same dimension floor as in 03.02.22 — the Whitney disc that empties the handle complex needs five ambient dimensions — so this theorem builds toward nothing weaker than its prerequisite and inherits that prerequisite's exact range. The central insight is that the homotopy-theoretic hypothesis (homotopy sphere) is converted into a smooth product (the collar) and then a topological gluing (the Alexander trick), and the gap between the smooth product and the topological gluing is exactly where the smooth classification hides: the coning $\overset{g}{^}$ is a homeomorphism but generically not a diffeomorphism at the cone point. This is exactly the passage that identifies the topological type of $Σ$ with $S^{n}$ while leaving its smooth type free, and it appears again in 03.06.17, where the residual smooth freedom is measured by a characteristic number. The bridge is the twisted sphere: a homotopy sphere generalises the standard sphere only up to the crookedness of a single boundary gluing, and continuous coning straightens that crookedness while smooth coning need not.

Exercises Intermediate+

Exercise (medium, short-answer). Verify directly that the Alexander coning $\hat g(tx) = t\, g(x)$, $\hat g(0) = 0$, is a homeomorphism of $D^n$ that restricts to $g$ on $S^{n-1}$.

Hint

Write down the inverse using $g^{- 1}$ , and check continuity at the origin via the bound $∣ \overset{g}{^} (v) ∣ = ∣ v ∣$ .

Answer

Argument. Each non-zero $v \in D^{n}$ has a unique polar form $v = t x$ with $t = ∣ v ∣ \in (0, 1]$ and $x = v /∣ v ∣ \in S^{n - 1}$ , and $\overset{g}{^} (v) = t g (x)$ has $∣ \overset{g}{^} (v) ∣ = t = ∣ v ∣$ since $g$ maps $S^{n - 1}$ to itself. The map $\overset{ˇ}{h} (t y) = t g^{- 1} (y)$ , $\overset{ˇ}{h} (0) = 0$ , satisfies $\overset{ˇ}{h} \circ \overset{g}{^} = id$ and $\overset{g}{^} \circ \overset{ˇ}{h} = id$ , so $\overset{g}{^}$ is a bijection with continuous inverse $\overset{ˇ}{h}$ . Continuity at $0$ follows from $∣ \overset{g}{^} (v) ∣ = ∣ v ∣ \to 0$ as $v \to 0$ . On the boundary $t = 1$ , $\overset{g}{^} (x) = g (x)$ . Hence $\overset{g}{^}$ is a homeomorphism of $D^{n}$ extending $g$ . (It is generally not differentiable at $0$ , since $g$ need not be the restriction of a linear map.)

Exercise (hard, short-answer). Run the Mayer-Vietoris computation of Step 2 to show $H_*(W, V_0) = 0$ for a homotopy $n$-sphere, $n \ge 2$. State the groups in each degree.

Hint

$W$ deformation retracts to a complex with $H_{*} (W) = H_{*} (S^{n - 1})$ ; compare with $V_{0} = S^{n - 1}$ .

Answer

Argument. From $Σ = W \cup (D_{0} ⊔ D_{1})$ with overlap $≃ S^{n - 1} ⊔ S^{n - 1}$ , the reduced Mayer-Vietoris sequence and $\tilde{H}_{*} (Σ) = \tilde{H}_{*} (S^{n})$ (so $Z$ in degree $n$ , $0$ else) give $\tilde{H}_{k} (W) = Z$ for $k = n - 1$ and $0$ otherwise: $W$ has the homology of a single $(n - 1)$ -sphere. The inclusion $V_{0} = S^{n - 1} ↪ W$ induces an isomorphism on $H_{n - 1}$ (both are $Z$ , generated by the boundary sphere, which is non-zero in $W$ because capping it with $D_{0}$ would otherwise kill the top class of $Σ$ ). The long exact sequence of the pair then gives $H_{k} (W, V_{0}) = 0$ in every degree, so the inclusion is a homology isomorphism; with simple connectivity it is a homotopy equivalence.

Advanced results Master

The smooth classification gap and $Θ_{n}$ . Smale's corollary settles the topological type of every high-dimensional homotopy sphere and leaves the smooth type open. The set of orientation-preserving diffeomorphism classes of homotopy $n$ -spheres forms an abelian group $Θ_{n}$ under connected sum, with the standard $S^{n}$ as identity; Kervaire and Milnor ^{[KM 1963]} identified $Θ_{n}$ for $n \geq 5$ as a finite group fitting into an exact sequence involving the stable homotopy groups of spheres and the image of the $J$ -homomorphism. The first non-zero case is $Θ_{7} = Z /28$ , the exotic seven-spheres of 03.06.17; $Θ_{5} = Θ_{6} = 0$ , so in those two dimensions the smooth Poincaré statement holds outright. The twisted-sphere presentation of Smale's proof identifies the class $[Σ] \in Θ_{n}$ with the isotopy class of the gluing diffeomorphism $g$ modulo those extending over the disc, identifying $Θ_{n}$ with $π_{0} Diff^{+} (S^{n - 1}) / π_{0} Diff^{+} (D^{n}, \partial)$ — the obstruction to coning $g$ smoothly.

Dimension four: the topological theorem and the smooth gap. Freedman ^{[Freedman 1982]} proved the topological four-dimensional Poincaré conjecture: a closed topological $4$ -manifold homotopy equivalent to $S^{4}$ is homeomorphic to $S^{4}$ . His method supplied a topologically embedded Whitney disc — a Casson handle — exactly where the smooth Whitney trick of 03.02.22 fails, since a smooth $2$ -disc cannot be embedded in the borderline ambient dimension $4$ . The smooth four-dimensional Poincaré conjecture — whether every smooth homotopy $4$ -sphere is diffeomorphic to $S^{4}$ — remains open as of 2026, and Donaldson-theoretic obstructions show the smooth and topological categories genuinely diverge in dimension four.

Dimension three: Ricci flow. The original Poincaré conjecture — that a closed simply connected $3$ -manifold is homeomorphic to $S^{3}$ — was settled by Perelman ^{[Perelman 2002]} through Hamilton's Ricci flow with surgery, an analytic method orthogonal to the handle calculus of this chapter: one evolves the metric by $\partial_{t} g = - 2 Ric (g)$ , surgically excising singularities, until the manifold rounds to a constant-curvature sphere. In dimension three smoothness and topology coincide, so the theorem is simultaneously topological and smooth.

Synthesis. Putting these together, the generalised Poincaré conjecture is one statement read in four different categories of dimension, and the bridge in each is the same question — does a homotopy sphere round into the standard sphere — answered by a different machine. The foundational reason the high-dimensional case is the easiest is exactly the Whitney trick of 03.02.22: five or more ambient dimensions give the membrane room, and this is precisely the room that dimension four denies, where Freedman's topological Casson handle and Donaldson's smooth obstruction identify the topological category with the homotopy type while leaving the smooth category genuinely larger. This is exactly the divergence that the high-dimensional theorem hides inside the Alexander trick, where the coning of $g$ is continuous but generically not smooth; the smooth residue is the group $Θ_{n}$ of 03.06.17, so the central insight is that Smale's theorem identifies the topological type with $S^{n}$ and reroutes the entire smooth classification into the homotopy of diffeomorphism groups. The three-dimensional case, governed by Ricci flow rather than handles, generalises the rounding intuition analytically and closes the one dimension where the handle calculus has no purchase. This pattern recurs across the surgery-theoretic classification: the homotopy type is the easy invariant, and the smooth structure is the obstruction group bolted on top.

Full proof set Master

Proposition (the Alexander trick). Let $g : S^{n - 1} \to S^{n - 1}$ be a homeomorphism. The radial coning $\overset{g}{^} : D^{n} \to D^{n}$ defined by $\overset{g}{^} (0) = 0$ and $\overset{g}{^} (v) = ∣ v ∣ g (v /∣ v ∣)$ for $v \neq = 0$ is a homeomorphism of $D^{n}$ restricting to $g$ on the boundary. Consequently $D^{n} \cup_{g} D^{n}$ is homeomorphic to $S^{n} = D^{n} \cup_{id} D^{n}$ .

Proof. For $v \neq = 0$ write $v = t x$ with $t = ∣ v ∣ \in (0, 1]$ , $x = v /∣ v ∣$ . Then $\overset{g}{^} (v) = t g (x)$ and $∣ \overset{g}{^} (v) ∣ = t ∣ g (x) ∣ = t = ∣ v ∣$ , since $g$ preserves $S^{n - 1}$ . So $\overset{g}{^}$ maps $D^{n}$ to $D^{n}$ and the sphere of radius $t$ to itself. The candidate inverse is $\hat{h} (w) = ∣ w ∣ g^{- 1} (w /∣ w ∣)$ , $\hat{h} (0) = 0$ ; the composite $\hat{h} (\overset{g}{^} (t x)) = t g^{- 1} (g (x)) = t x$ and symmetrically $\overset{g}{^} \circ \hat{h} = id$ , so $\overset{g}{^}$ is a bijection. Continuity of $\overset{g}{^}$ away from $0$ is clear from the formula; at $0$ the bound $∣ \overset{g}{^} (v) ∣ = ∣ v ∣$ gives $\overset{g}{^} (v) \to 0 = \overset{g}{^} (0)$ , and the same bound applied to $\hat{h}$ gives continuity of the inverse at $0$ . Thus $\overset{g}{^}$ is a homeomorphism with $\overset{g}{^} ∣_{S^{n - 1}} = g$ . Now $id_{D^{n}} \cup \overset{g}{^} : D^{n} \cup_{id} D^{n} \to D^{n} \cup_{g} D^{n}$ is well-defined and a homeomorphism, since on the shared boundary the two pieces agree: the first disc maps by the identity, the second by $\overset{g}{^}$ , and the gluing condition $id (x) \mapsto \overset{g}{^} (x) = g (x)$ matches the gluing $g$ . Hence $S^{n} ≅ D^{n} \cup_{g} D^{n}$ . $□$

Proposition (a twisted sphere is smooth and a homotopy sphere). For any orientation-preserving diffeomorphism $g$ of $S^{n - 1}$ , the twisted sphere $Σ_{g} = D_{0}^{n} \cup_{g} D_{1}^{n}$ carries a smooth structure making it a closed smooth homotopy $n$ -sphere.

Proof. The gluing $g$ is a diffeomorphism of the boundary, so $Σ_{g}$ inherits a smooth structure away from the seam, and a collar $S^{n - 1} \times (- ε, ε)$ across the seam is smoothed using $g$ to identify the two collar halves (the standard collaring theorem for gluing manifolds along diffeomorphic boundaries). The result is a closed smooth $n$ -manifold. To see it is a homotopy sphere, cover $Σ_{g}$ by the two discs $U_{0}, U_{1}$ (slightly enlarged to open sets); each is contractible and their intersection deformation retracts to the seam $S^{n - 1}$ . The Mayer-Vietoris sequence then reads $\tilde{H}_{k} (Σ_{g}) ≅ \tilde{H}_{k - 1} (S^{n - 1})$ in the relevant range, giving $\tilde{H}_{k} (Σ_{g}) = Z$ for $k = n$ and $0$ otherwise, the homology of $S^{n}$ . Van Kampen with two simply connected pieces (each disc) and connected intersection ( $S^{n - 1}$ , connected for $n \geq 2$ ) gives $π_{1} (Σ_{g}) = 1$ . A simply connected closed manifold with the homology of $S^{n}$ is a homotopy sphere by Hurewicz and Whitehead. $□$

Proposition ( $Θ_{5} = Θ_{6} = 0$ gives the smooth statement in those dimensions). Every smooth homotopy $5$ -sphere is diffeomorphic to $S^{5}$ , and every smooth homotopy $6$ -sphere is diffeomorphic to $S^{6}$ .

Proof. By the twisted-sphere presentation, a smooth homotopy $n$ -sphere $Σ$ is $Σ_{g}$ for some $g \in Diff^{+} (S^{n - 1})$ , and its diffeomorphism class is the class $[Σ] \in Θ_{n} = π_{0} Diff^{+} (S^{n - 1}) / (those extending over D^{n})$ . The Kervaire-Milnor computation ^{[KM 1963]} gives $Θ_{5} = 0$ and $Θ_{6} = 0$ : both groups vanish because the relevant stable stems and Kervaire invariants vanish there. With the receiving group zero, $[Σ] = 0$ , so $g$ extends to a diffeomorphism of the disc and $Σ_{g} ≅ S^{n}$ by the preceding proposition's extension argument. Hence every homotopy $5$ - or $6$ -sphere is the standard smooth sphere. $□$

These three propositions complete the corollary: the Alexander trick supplies the homeomorphism for every $n$ , the twisted-sphere proposition confirms the objects produced are genuine smooth homotopy spheres, and the $Θ_{n}$ computation pins down the dimensions in which "homeomorphic" can be upgraded to "diffeomorphic" without obstruction.

Connections Master

The engine of this unit is the h-cobordism theorem of 03.02.23: the twice-punctured homotopy sphere is the h-cobordism whose product structure forces the two-disc presentation, and every hypothesis Smale's theorem needs — simple connectivity, dimension $\geq 6$ , the homotopy-equivalence of the boundary inclusions — is checked precisely to feed that theorem its inputs, so this unit is the first and most famous corollary of the one before it.

The dimension floor and the smooth-versus-topological split both trace to the Whitney trick of 03.02.22: the five-ambient-dimension requirement that this corollary inherits as $n \geq 5$ is the room a smooth Whitney disc needs, and the failure of that room in dimension four is exactly why Freedman's topological Casson handle succeeds where no smooth disc embeds, marking the boundary between the categories along the same geometric fact about discs that governs handle cancellation.

The smooth residue this theorem cannot see is the exotic-sphere theory of 03.06.17: Smale's homeomorphism leaves the diffeomorphism class free, and the obstruction to smoothing the Alexander coning is the class $[Σ] \in Θ_{n}$ whose first non-zero value $Θ_{7} = Z /28$ that unit computes by characteristic numbers; this unit supplies the topological half of the statement, and 03.06.17 supplies the smooth half.

The homology and homotopy bookkeeping that certify the twice-punctured sphere is an h-cobordism — Mayer-Vietoris, the Hurewicz and Whitehead theorems converting a homology isomorphism on a simply connected space into a homotopy equivalence — are the constructions of 03.12.16 (Poincaré duality, which pins the top homology class used in Step 2) and the cellular and singular homology machinery they rest on, so the verification of the h-cobordism hypothesis is where this differential-topology corollary draws on algebraic topology.

Historical & philosophical context Master

Stephen Smale announced the generalised Poincaré conjecture in dimensions five and above in Generalized Poincaré's conjecture in dimensions greater than four (Annals of Mathematics 74, 1961, 391–406) ^{[Smale 1961]}, with the supporting handle machinery in On the structure of manifolds (American Journal of Mathematics 84, 1962, 387–399) ^{[Smale 1962]}. The result reversed the expected order of difficulty: the Poincaré conjecture, open since Henri Poincaré's 1904 question, fell first in the high dimensions where geometric intuition is weakest, because exactly there the Whitney disc has the room it needs. John Milnor's Lectures on the h-Cobordism Theorem (Princeton Mathematical Notes, 1965), notes by Larry Siebenmann and John Sondow, organised the proof into its canonical form and presents the Poincaré corollary in §8 as the immediate payoff of the product theorem. The same year as Smale's papers, John Stallings and Erik Christopher Zeeman gave independent piecewise-linear proofs by engulfing, and the $n = 5$ case was secured by these methods and by the dual h-cobordism.

The smooth refinement was already visible: Milnor's 1956 exotic seven-spheres ^{[KM 1963]} showed that "homeomorphic to $S^{n}$ " could not be strengthened to "diffeomorphic," and Kervaire and Milnor's Groups of homotopy spheres I (Annals of Mathematics 77, 1963, 504–537) ^{[KM 1963]} computed the obstruction group $Θ_{n}$ . The two low-dimensional cases were settled by entirely different machinery decades later: Michael Freedman's The topology of four-dimensional manifolds (Journal of Differential Geometry 17, 1982, 357–453) ^{[Freedman 1982]} resolved the topological four-dimensional case using Casson handles, and Grigori Perelman's 2002–2003 preprints ^{[Perelman 2002]} settled the three-dimensional case by Ricci flow with surgery, completing Richard Hamilton's programme. Smale, Freedman, and Perelman were each recognised with the Fields Medal for this circle of results, Perelman declining the award in 2006.

Bibliography Master

@article{smale1961poincare,
  author  = {Smale, Stephen},
  title   = {Generalized {P}oincar{\'e}'s conjecture in dimensions greater than four},
  journal = {Annals of Mathematics},
  volume  = {74},
  number  = {2},
  pages   = {391--406},
  year    = {1961}
}

@article{smale1962structure,
  author  = {Smale, Stephen},
  title   = {On the structure of manifolds},
  journal = {American Journal of Mathematics},
  volume  = {84},
  number  = {3},
  pages   = {387--399},
  year    = {1962}
}

@book{milnor1965hcobordism,
  author    = {Milnor, John W.},
  title     = {Lectures on the h-Cobordism Theorem},
  series    = {Princeton Mathematical Notes},
  publisher = {Princeton University Press},
  year      = {1965},
  note      = {Notes by L. Siebenmann and J. Sondow}
}

@article{kervairemilnor1963,
  author  = {Kervaire, Michel A. and Milnor, John W.},
  title   = {Groups of homotopy spheres. {I}},
  journal = {Annals of Mathematics},
  volume  = {77},
  number  = {3},
  pages   = {504--537},
  year    = {1963}
}

@article{freedman1982,
  author  = {Freedman, Michael H.},
  title   = {The topology of four-dimensional manifolds},
  journal = {Journal of Differential Geometry},
  volume  = {17},
  number  = {3},
  pages   = {357--453},
  year    = {1982}
}

@misc{perelman2002entropy,
  author = {Perelman, Grisha},
  title  = {The entropy formula for the {R}icci flow and its geometric applications},
  year   = {2002},
  note   = {arXiv:math/0211159}
}

@misc{perelman2003surgery,
  author = {Perelman, Grisha},
  title  = {{R}icci flow with surgery on three-manifolds},
  year   = {2003},
  note   = {arXiv:math/0303109}
}

@book{morgantian2007,
  author    = {Morgan, John and Tian, Gang},
  title     = {Ricci Flow and the {P}oincar{\'e} Conjecture},
  series    = {Clay Mathematics Monographs},
  volume    = {3},
  publisher = {American Mathematical Society},
  year      = {2007}
}

@book{kosinski1993,
  author    = {Kosinski, Antoni A.},
  title     = {Differential Manifolds},
  series    = {Pure and Applied Mathematics},
  volume    = {138},
  publisher = {Academic Press},
  year      = {1993}
}

Prerequisites

03.02.23
03.02.22

Tier anchors

beginner: Milnor Lectures on the h-Cobordism Theorem §8 (the Poincaré corollary, informal); Matsumoto An Introduction to Morse Theory Ch. 3 (assembling a sphere from two discs)
intermediate: Milnor Lectures on the h-Cobordism Theorem §8; Smale Generalized Poincaré's conjecture in dimensions greater than four (1961); Kosinski Differential Manifolds Ch. X
master: Milnor Lectures on the h-Cobordism Theorem §§6-9; Smale Generalized Poincaré's conjecture (1961); Kervaire-Milnor Groups of homotopy spheres I (1963); Freedman The topology of four-dimensional manifolds (1982); Morgan-Tian Ricci Flow and the Poincaré Conjecture (2007)

References

03-modern-geometry/Bott-LecturesOnMorseTheory.pdf · Gradient-like vector fields, handle cancellation, and the reduction of a handle decomposition to a product (the engine behind the Poincaré corollary)
Milnor — Lectures on the h-Cobordism Theorem (Princeton Mathematical Notes, 1965) · §8: the h-cobordism theorem, its corollaries, and the generalised Poincaré conjecture (notes by L. Siebenmann and J. Sondow)
Smale — Generalized Poincaré's conjecture in dimensions greater than four (1961) · Annals of Mathematics 74, pp. 391--406; a smooth homotopy n-sphere with n >= 5 is homeomorphic to S^n
Smale — On the structure of manifolds (1962) · American Journal of Mathematics 84, pp. 387--399; the h-cobordism theorem from which the Poincaré corollary is read off
Kervaire and Milnor — Groups of homotopy spheres I (1963) · Annals of Mathematics 77, pp. 504--537; the group Theta_n of smooth structures on the homotopy n-sphere and the smooth/topological gap
Freedman — The topology of four-dimensional manifolds (1982) · Journal of Differential Geometry 17, pp. 357--453; the topological four-dimensional Poincaré conjecture via Casson handles
Perelman — The entropy formula for the Ricci flow and its geometric applications (2002); Ricci flow with surgery on three-manifolds (2003) · arXiv:math/0211159 and arXiv:math/0303109; the three-dimensional Poincaré conjecture via Ricci flow with surgery

Estimated time

beginner: 15m
intermediate: 38m
master: 72m