03.02.23 · differential-geometry / manifolds

The h-cobordism theorem

shipped3 tiersLean: none

Anchor (Master): Milnor Lectures on the h-Cobordism Theorem §§6-9; Smale On the structure of manifolds (1962); Smale Generalized Poincaré's conjecture (1961); Barden-Mazur-Stallings s-cobordism; Milnor Whitehead torsion (1966)

Intuition Beginner

Imagine a solid slab whose bottom face and top face are two surfaces. Suppose every loop you could draw on the bottom face can be slid up through the slab and matched with a loop on the top, and every loop on the top slides down to match one on the bottom, so the two faces are interchangeable as far as bending and stretching can tell. The question is whether the slab is really just one of the faces thickened — the face times a short interval, like a sheet of paper given a little depth — with nothing hidden inside.

The answer, in high enough dimensions, is yes. If the two faces are interchangeable in this stretchy sense, and if neither face has any hidden loops of its own that fail to close up, and if there is enough room — five or more dimensions across each face — then the slab is exactly the bottom face thickened. A surprising consequence follows at once: the bottom face and the top face are the same shape, not merely interchangeable up to stretching, but genuinely identical as smooth shapes.

The proof is a tidying operation. Earlier units showed how to build the slab by gluing on handles, one for each spot where a height function on the slab levels off, and how to sort those handles neatly by size. Here we finish the job: we show that the interchangeability of the two faces forces every handle to come in a cancelling pair, and we peel the pairs off one by one. When the last pair is gone, no handles remain, and a slab with no handles is just a thickened face.

The room condition is the catch. The peeling move needs a small membrane with space to slide, and that space appears only in five or more dimensions across a face. In low dimensions the membrane can be forced to cross itself, the move stalls, and the clean conclusion can fail.

Visual Beginner

A thick slab is drawn standing on its bottom face, with its top face above. Inside, a height function is shown by faint level lines, and at two spots the lines pinch into little summits and saddles where handles were glued on. The two faces are connected by a family of upward arrows, suggesting that points on the bottom flow up to points on the top and back, the two faces matching.

The takeaway from the picture: when the two faces correspond perfectly and there is enough room, every internal handle pairs off and cancels, and the slab collapses in our understanding to the bottom face times a short interval — a plain product, with straight flow lines and nothing hidden inside.

Worked example Beginner

A cylinder between two circles, thickened up a dimension. Picture the simplest product: a flat ring (an annulus) sitting between its inner circle and its outer circle. The annulus is the inner circle times a short interval — you walk straight outward from inner to outer along radial lines. The two boundary circles correspond perfectly through these radial lines, and the annulus hides nothing. This is the picture the theorem promises in general, only here it is plain to see by eye.

Why the high-dimensional statement is not plain. Now imagine a slab between two five-dimensional faces, built by gluing on a handle and then, higher up, a second handle of the next size that seems to undo it. By eye you cannot see whether the slab is a product. The two faces might correspond perfectly through some flow, yet the handles sit in the way. The theorem says: check the correspondence by a bookkeeping count of how the handles link, find the count forces them to cancel, peel them, and conclude the slab was a product all along.

What this tells us. The annulus shows the conclusion; the high-dimensional slab shows why the conclusion needs proof. The bridge between them is the handle bookkeeping: a perfect correspondence of the two faces, translated into the language of handles, says the handles cancel in pairs, and once they are gone the slab is a thickened face exactly like the annulus.

Check your understanding Beginner

Formal definition Intermediate+

Throughout, $W$ is a compact smooth manifold of dimension $m$ with boundary the disjoint union $\partial W = V_{0} ⊔ V_{1}$ of two closed manifolds.

Definition (cobordism, h-cobordism). The triple $(W; V_{0}, V_{1})$ is a cobordism from $V_{0}$ to $V_{1}$ . It is an h-cobordism when both inclusions $V_{0} ↪ W$ and $V_{1} ↪ W$ are homotopy equivalences. Equivalently, the relative homology and homotopy of the pairs vanish: $H_{*} (W, V_{0}) = 0$ and $H_{*} (W, V_{1}) = 0$ , and similarly for $π_{*}$ once basepoints are fixed.

Definition (product cobordism). The cobordism is a product when there is a diffeomorphism $W ≅ V_{0} \times [0, 1]$ carrying $V_{0}$ to $V_{0} \times {0}$ and $V_{1}$ to $V_{0} \times {1}$ . A product cobordism is an h-cobordism whose two boundary inclusions are even homotopy equivalences in the strong sense of deformation retractions, and it forces $V_{0} ≅ V_{1}$ as smooth manifolds.

The statement to be proved. Let $(W; V_{0}, V_{1})$ be a compact h-cobordism with $W$ , $V_{0}$ , $V_{1}$ simply connected and $dim W = m \geq 6$ . Then $(W; V_{0}, V_{1})$ is a product: $W ≅ V_{0} \times [0, 1]$ , and in particular $V_{0} ≅ V_{1}$ . This is the h-cobordism theorem ^{[Smale 1962]}.

The combinatorial encoding. By 03.02.21 choose a self-indexing Morse function $f : W \to [0, m]$ with a Morse-Smale gradient-like field $ξ$ : critical points of index $λ$ lie on the level $f^{- 1} (λ + \frac{1}{2})$ , and ascending and descending manifolds meet transversally. The handles assemble into a free based chain complex $C_{*} = C_{*} (f, ξ)$ , with $C_{λ}$ free abelian on the index- $λ$ critical points and boundary map $\partial p = \sum_{q} [p : q] q$ given by signed trajectory counts. By 03.02.21 this complex computes the relative homology: $H_{λ} (C_{*}) = H_{λ} (W, V_{0})$ . Simple connectivity makes $C_{*}$ a complex of free $Z$ -modules, with no $Z [π_{1}]$ decoration; the h-cobordism hypothesis makes it acyclic, $H_{*} (C_{*}) = 0$ .

Counterexamples to common slips

The hypothesis is on $W$ , $V_{0}$ , $V_{1}$ being simply connected and on $dim W \geq 6$ . Dropping simple connectivity is not a harmless relaxation: there are h-cobordisms with $π_{1} \neq = 1$ that are not products, distinguished by their Whitehead torsion (the $s$ -cobordism refinement below). Dropping the dimension bound is fatal in dimension $4$ , where the smooth theorem is false.
An h-cobordism asks for homotopy equivalence of the inclusions, a far weaker input than the product conclusion. The content of the theorem is precisely that, in high dimensions and simply connected, this weak input is upgraded to a diffeomorphism.
" $V_{0} ≅ V_{1}$ " is a corollary, not the theorem. The theorem is the stronger product statement; two manifolds can be diffeomorphic without an a priori product cobordism between them being given.

Key theorem with proof Intermediate+

We assemble the cancellation machinery of 03.02.22 into the full theorem. The engine is the Strong Cancellation Theorem of 03.02.22: in a simply connected level of dimension $\geq 5$ , a $λ$ -handle and $(λ + 1)$ -handle whose belt and attaching spheres have algebraic intersection number $\pm 1$ cancel.

Theorem (h-cobordism theorem, LHC §§6-8). Let $(W; V_{0}, V_{1})$ be a compact h-cobordism with $W, V_{0}, V_{1}$ simply connected and $m = dim W \geq 6$ . Then $W ≅ V_{0} \times [0, 1]$ . ^{[Smale 1962]}

Proof. Fix a self-indexing Morse-Smale pair $(f, ξ)$ as above, with based free chain complex $C_{*}$ computing $H_{*} (W, V_{0}) = 0$ ; so $C_{*}$ is acyclic.

Step 1 — normalise to middle indices. By the handle-trading argument of 03.02.21 and LHC §7, the index- $0$ and index- $1$ handles are eliminated. Simple connectivity of $V_{0}$ and $W$ lets each $1$ -handle be cancelled against an added $2$ - $3$ pair or traded past, and a connectivity count removes the $0$ -handles; dually the top two indices $m - 1, m$ are removed. This needs the middle range $[2, m - 2]$ to be non-empty and to absorb the traded handles, which is where $m \geq 6$ rather than $m \geq 5$ is consumed. After Step 1 the only handles have index $λ$ with $2 \leq λ \leq m - 2$ , and $C_{*}$ is still acyclic.

Step 2 — make a boundary entry $\pm 1$ . Pick the lowest index $λ$ with $C_{λ} \neq = 0$ . Acyclicity gives an exact sequence, so $\partial : C_{λ + 1} \to C_{λ}$ is surjective onto $ker (\partial : C_{λ} \to C_{λ - 1}) = C_{λ}$ (the lower boundary is zero by minimality). Since $C_{λ}$ is free and $\partial$ is surjective, after a change of basis — realised geometrically by handle slides, the Smale moves that add a multiple of one handle's attaching sphere to another's by sliding over it — some basis element $e \in C_{λ + 1}$ maps to a basis element $\partial e = \pm f$ in $C_{λ}$ , with $f$ a $λ$ -handle. Handle slides are isotopies of the gradient-like field and change neither $W$ nor the diffeomorphism type; they only recompute the boundary matrix by elementary column and row operations.

Step 3 — cancel the pair. The entry $\partial e = \pm f$ says the belt sphere of the $λ$ -handle $f$ and the attaching sphere of the $(λ + 1)$ -handle $e$ have algebraic intersection number $\pm 1$ in the middle level $f^{- 1} (λ + \frac{1}{2})$ , a level of dimension $m - 1 \geq 5$ , simply connected because $W$ is simply connected and the level is obtained from $V_{0}$ by surgeries that do not destroy simple connectivity in this range. The Strong Cancellation Theorem of 03.02.22 applies: the Whitney trick isotopes the spheres to meet in a single geometric point, and the First Cancellation Theorem deletes both handles, modifying $ξ$ to a gradient-like field with two fewer critical points.

Step 4 — induct. Deleting the pair lowers the total number of handles by $2$ and preserves acyclicity of the (smaller) complex. Repeat Steps 2-3 until $C_{*} = 0$ . A Morse function with no critical points has, by the no-critical-points criterion of 03.02.20, a gradient-like flow giving $W$ a product structure: the flow carries $V_{0}$ diffeomorphically across to $V_{1}$ , giving $W ≅ V_{0} \times [0, 1]$ . $□$

Bridge. Putting these together, the h-cobordism theorem is the global completion of the local cancellation of 03.02.22: the acyclic chain complex of 03.02.21 is emptied one cancelling pair at a time, and the foundational reason this terminates is that acyclicity plus freeness over $Z$ guarantees a $\pm 1$ boundary entry at every stage, while simple connectivity and dimension $\geq 5$ guarantee that entry can be cashed in geometrically. This is exactly the bridge from algebra to geometry that 03.02.22 set up — an algebraic $\pm 1$ is converted to a single geometric intersection, which is a cancelled handle pair. The same reduction-of-an-acyclic-complex pattern appears again in the $s$ -cobordism refinement below, where the obstruction to emptying the complex by elementary moves is no longer automatically zero but is measured by Whitehead torsion; and it builds toward the high-dimensional Poincaré conjecture, where the h-cobordism is the one between a punctured homotopy sphere and a disc. The central insight is that the homotopy hypothesis (acyclicity) and the geometric room (dimension $\geq 5$ ) together identify the algebraic vanishing of $C_{*}$ with the geometric emptiness of the handle decomposition.

Exercises Intermediate+

Exercise (hard, short-answer). Deduce from the h-cobordism theorem that a smooth manifold $\Sigma^n$ ($n \ge 5$) which is a homotopy $n$-sphere is homeomorphic to $S^n$. Outline the construction of the relevant h-cobordism.

Hint

Remove two disjoint open discs from $Σ$ ; the complement is an h-cobordism between two $(n - 1)$ -spheres.

Answer

Argument. Remove the interiors of two disjoint smooth $n$ -discs $D_{0}, D_{1} \subset Σ$ . The complement $W = Σ ∖ (\overset{˚}{D}_{0} \cup \overset{˚}{D}_{1})$ is a cobordism between $V_{0} = \partial D_{0} ≅ S^{n - 1}$ and $V_{1} = \partial D_{1} ≅ S^{n - 1}$ , of dimension $n \geq 5$ . Since $Σ$ is a homotopy sphere, $W$ is simply connected and the inclusions of the boundary spheres are homotopy equivalences, so $W$ is an h-cobordism with $dim W = n \geq 5$ ; for $n \geq 6$ the theorem gives $W ≅ S^{n - 1} \times [0, 1]$ . (For $n = 5$ one argues separately or invokes the dual.) Thus $Σ$ is two discs glued along a product collar of an $(n - 1)$ -sphere, i.e. a twisted double of $D^{n}$ , which is homeomorphic — by the Alexander trick on the gluing diffeomorphism — to $S^{n}$ . This is the high-dimensional Poincaré conjecture, treated in full in 03.02.24.

Advanced results Master

The high-dimensional Poincaré conjecture. Smale's original 1961 application ^{[Smale 1961]} is the corollary worked above: a smooth homotopy $n$ -sphere with $n \geq 5$ is homeomorphic (indeed PL-homeomorphic) to $S^{n}$ . The h-cobordism between two boundary spheres of a punctured homotopy sphere is a product, so the homotopy sphere is a twisted double of a disc, hence a topological sphere. The conclusion is homeomorphic, not diffeomorphic: the gluing diffeomorphism of $S^{n - 1}$ need not extend smoothly over the disc, and the failure to extend is exactly Milnor's exotic spheres — distinct smooth structures on the topological $S^{7}$ , detected in the companion homotopy-sphere unit. The h-cobordism theorem is the engine that upgrades a cohomological or homotopical match to a homeomorphism with $S^{n}$ ; it is silent on the finer smooth classification, which the Kervaire-Milnor surgery-theoretic computation of the group $Θ_{n}$ supplies.

The $s$ -cobordism theorem and Whitehead torsion. When $π_{1} (V_{0}) = π \neq = 1$ the based chain complex $C_{*}$ lives over the group ring $Z [π]$ , intersection numbers take values in $Z [π]$ , and a Whitney disc exists only when the obstructing loop is null-homotopic — which now requires the relevant boundary entry to be a unit $\pm g$ ( $g \in π$ ), not merely $\pm 1$ . The acyclic based $Z [π]$ -complex $C_{*}$ carries an invariant under the elementary moves (handle slides, addition and deletion of cancelling pairs): its Whitehead torsion $τ (W, V_{0}) \in Wh (π)$ , an element of the Whitehead group $Wh (π) = K_{1} (Z [π]) / (\pm π)$ from the algebraic $K$ -theory of the group ring ^{[Milnor 1966]}. The $s$ -cobordism theorem of Barden, Mazur, and Stallings ^{[BMS 1963-65]} states: a compact h-cobordism $(W; V_{0}, V_{1})$ with $dim W \geq 6$ is a product iff $τ (W, V_{0}) = 0$ . The full Whitehead torsion obstruction is the subject of 03.08.20; here we record that the simply connected h-cobordism theorem is the special case $Wh (1) = 0$ , where the torsion vanishes for every h-cobordism, recovering Smale's product conclusion with no extra hypothesis.

The dimension floor and its boundary. The smooth theorem requires $dim W \geq 6$ ; the topological h-cobordism theorem in dimension $5$ (so $W$ five-dimensional, boundaries four-dimensional) was obtained by Freedman in the topological category, using Casson handles to supply a topologically embedded Whitney disc where no smooth one exists. The smooth statement fails for four-dimensional boundaries: Donaldson's gauge theory exhibits smooth h-cobordisms between non-diffeomorphic simply connected $4$ -manifolds. So the single geometric fact — a $2$ -disc needs five ambient dimensions to embed — marks the boundary between the dimensions where the theorem reigns and the four-dimensional world where smooth and topological diverge.

Synthesis. The h-cobordism theorem is the central insight that converts a homotopy-theoretic hypothesis into a smooth classification result, and it is the foundational reason high-dimensional differential topology is governed by algebra. Putting these together: the acyclic Morse-Smale complex of 03.02.21 is dual to the cellular complex of 03.12.13, the Whitney trick of 03.02.22 identifies an algebraic $\pm 1$ boundary entry with a geometric handle cancellation, and the global induction empties the complex; this is exactly the passage from the homotopy type of the pair $(W, V_{0})$ to its diffeomorphism type. The same machine generalises in two directions, and the bridge in each is the same: replacing $Z$ by $Z [π_{1}]$ promotes the vanishing obstruction to the Whitehead torsion of 03.08.20 and yields the $s$ -cobordism theorem, while replacing "product" by "controlled normal map" promotes the cancellation to the surgery exact sequence of Browder, Novikov, Sullivan, and Wall. The theorem is dual, through Poincaré duality 03.12.16, to itself under reversing the cobordism, and this self-duality is what lets the low-handle and high-handle eliminations be run as mirror images. It is the load-bearing base case of the entire surgery-theoretic classification of high-dimensional manifolds.

Full proof set Master

Proposition (a handle-free h-cobordism is a product). Let $(W; V_{0}, V_{1})$ admit a Morse function with no critical points and a gradient-like field $ξ$ . Then $W ≅ V_{0} \times [0, 1]$ .

Proof. With no critical points, $f : W \to [0, c]$ has $\nabla$ -like field $ξ$ nowhere zero, with $f$ increasing strictly along every $ξ$ -trajectory and $df (ξ) > 0$ . Rescale $ξ$ to $η = ξ / df (ξ)$ , so $df (η) = 1$ : along each trajectory $f$ increases at unit rate. Every trajectory then starts on $V_{0} = f^{- 1} (0)$ and ends on $V_{1} = f^{- 1} (c)$ after parameter length $c$ , with no trajectory trapped at an interior critical point (there are none). The map $Φ : V_{0} \times [0, c] \to W$ sending $(x, t)$ to the time- $t$ flow of $η$ from $x$ is smooth, injective (distinct start points or times give distinct images since $f \circ Φ (x, t) = t$ and trajectories partition $W$ ), and surjective (every point lies on the trajectory through its $η$ -flow back to $V_{0}$ ). Its differential is invertible — the $V_{0}$ -directions map to the corresponding directions on each level and the $[0, c]$ -direction to $η \neq = 0$ transverse to the levels — so $Φ$ is a diffeomorphism by the inverse function theorem applied globally via properness. Reparametrising $[0, c] ≅ [0, 1]$ gives $W ≅ V_{0} \times [0, 1]$ . $□$

Proposition (acyclic free $Z$ -complex of finite type cancels to zero by elementary moves). Let $C_ $b e a f ini t ec hain co m pl e x o f f ini t e l y g e n er a t e df r ee$ \mathbb{Z} $- m o d u l es w i t h$ H_*(C_*) = 0 $. T h e n$ C_* $i sr e d u ce d t o t h ez er oco m pl e x b y a f ini t ese q u e n ceo f e l e m e n t a r y m o v es : c han g eo f ba s i s in e a c h$ C_\lambda $, an d r e m o v a l o f a p ai r o f ba s i se l e m e n t s$ e \in C_{\lambda+1}, f \in C_\lambda $w i t h$ \partial e = \pm f + (\text{terms in other basis elements}) $a f t er aba s i sc han g e mak in g$ \partial e = \pm f$.*

Proof. Induct on the total rank $\sum_{λ} rank C_{λ}$ . If all ranks are zero there is nothing to do. Otherwise let $λ$ be minimal with $C_{λ} \neq = 0$ . Since $C_{λ - 1}$ receives the zero map from below (minimality) the lowest boundary $\partial : C_{λ} \to C_{λ - 1}$ has image a free summand, and acyclicity forces $C_{λ} = ker (\partial_{λ}) = im (\partial_{λ + 1})$ ; in particular $\partial_{λ} = 0$ and $\partial_{λ + 1} : C_{λ + 1} ↠ C_{λ}$ is onto. By the Smith normal form of the matrix of $\partial_{λ + 1}$ over $Z$ , a change of basis in $C_{λ + 1}$ and $C_{λ}$ makes this matrix have a single $1$ in some row and zero elsewhere in that row and column — possible because the cokernel $C_{λ} / im = 0$ has all invariant factors equal to $1$ . That picks out $e \in C_{λ + 1}, f \in C_{λ}$ with $\partial e = f$ and $f$ not hit otherwise. Delete the pair $(e, f)$ : the quotient complex $C_{*}^{'}$ has $\partial^{'}$ induced, is still a complex (the deleted $f$ was a free summand killed by $\partial e$ , and no other differential involved it after the basis change), and remains acyclic by the long exact sequence of the deletion, since we removed an acyclic two-term summand $Z ⟨ e ⟩ ≅ Z ⟨ f ⟩$ . The total rank dropped by $2$ ; induct. $□$

This proposition is the algebraic skeleton of the h-cobordism proof: each algebraic deletion of a pair $(e, f)$ is realised geometrically by the Strong Cancellation Theorem of 03.02.22, provided the ambient level is simply connected and of dimension $\geq 5$ , and each change of basis is realised by handle slides.

Proposition (the Whitehead obstruction vanishes in the simply connected case). If $π = π_{1} (V_{0}) = 1$ then $Wh (π) = 0$ , so every h-cobordism in this case has $τ = 0$ and the $s$ -cobordism theorem reduces to Smale's.

Proof. For $π = 1$ the group ring is $Z [π] = Z$ . Then $K_{1} (Z)$ is computed from invertible matrices over $Z$ up to stabilisation and elementary operations: $K_{1} (Z) = Z^{\times} = {\pm 1}$ , generated by the determinant, since every invertible integer matrix is a product of elementary matrices times a diagonal $diag (\pm 1, 1, \dots, 1)$ (integer Gaussian elimination, the Euclidean algorithm on entries). The Whitehead group quotients out the units $\pm π = {\pm 1}$ : $Wh (1) = K_{1} (Z) / {\pm 1} = {\pm 1} / {\pm 1} = 0$ . With the receiving group zero, the torsion $τ (W, V_{0}) \in Wh (1)$ is forced to vanish for every simply connected h-cobordism, and the $s$ -cobordism criterion " $τ = 0$ " is automatic. $□$

Connections Master

The h-cobordism theorem completes the handle programme begun in 03.02.20 and sorted in 03.02.21: the self-indexing Morse function and its Morse-Smale gradient field there produce the acyclic free chain complex that this unit empties, and the no-critical-points criterion of 03.02.20 is what turns an empty complex into the product structure; without the rearrangement of 03.02.21 there would be no chain complex to call acyclic, and without the elementary-cobordism dictionary of 03.02.20 there would be no flow to give $W$ its product structure at the end.

The geometric step inside each induction is the Whitney trick and Strong Cancellation Theorem of 03.02.22: every $\pm 1$ boundary entry produced by the algebra of this unit is converted there into a single geometric intersection and a deleted handle pair, and the dimension floor $m - 1 \geq 5$ and simple-connectivity hypotheses of this unit are exactly the inputs that the Whitney Lemma of 03.02.22 requires, so the two units are the algebraic and geometric halves of one argument.

The non-simply-connected refinement reroutes the obstruction into algebraic $K$ -theory: when $π_{1} \neq = 1$ the vanishing that this unit obtains for free becomes the Whitehead torsion $τ \in Wh (π)$ studied in 03.08.20, and the $s$ -cobordism theorem says a product structure exists iff $τ = 0$ ; the simply connected case proved here is the special instance $Wh (1) = 0$ , so 03.08.20 is the exact measure of how far this theorem's hypothesis can be relaxed.

The acyclic complex this unit reduces is the smooth realisation of the cellular chain complex of 03.12.13, and the algebraic intersection numbers it manipulates are the homological pairing made geometric through Poincaré duality 03.12.16; the self-duality of the h-cobordism under reversing $f \mapsto m - f$ , which lets the low-handle and high-handle eliminations mirror each other, is precisely Poincaré-Lefschetz duality of the pair $(W, V_{0})$ against $(W, V_{1})$ , identifying $H_{λ} (W, V_{0})$ with $H_{m - λ} (W, V_{1})$ .

The analytic foundation of the trajectory counts and the transversality of ascending and descending manifolds is the gradient-flow Morse-Smale theory of 03.15.01: the boundary entries $[p : q]$ this unit reads off the chain complex are signed counts of isolated trajectories there, and the broken-trajectory compactness that gives $\partial^{2} = 0$ — without which "acyclic complex" would be meaningless — is imported from that unit.

Historical & philosophical context Master

Stephen Smale proved the h-cobordism theorem and, as its first great consequence, the generalised Poincaré conjecture in dimensions five and above, in two papers of the early 1960s: Generalized Poincaré's conjecture in dimensions greater than four (Annals of Mathematics 74, 1961, 391–406) ^{[Smale 1961]} and On the structure of manifolds (American Journal of Mathematics 84, 1962, 387–399) ^{[Smale 1962]}. The achievement was startling: the Poincaré conjecture, open since 1904 and synonymous with the hardest problems in topology, fell first in the high dimensions where intuition is weakest, because precisely there the Whitney disc has room to embed. Smale's insight was to recast the topology of a cobordism as the combinatorics of a handle decomposition and then to cancel handles with the Whitney trick — converting a classification problem into linear algebra over $Z$ . John Milnor's Lectures on the h-Cobordism Theorem (Princeton Mathematical Notes, 1965), with notes by Larry Siebenmann and John Sondow, gave the proof its canonical pedagogical form, organised around the First and Strong Cancellation Theorems and the reduction of an acyclic Morse-Smale complex; the book is the standard entry to the subject and the acknowledged prerequisite for surgery theory.

The non-simply-connected story unfolded in parallel. The $s$ -cobordism theorem — that an h-cobordism is a product iff its Whitehead torsion vanishes — was obtained independently by Dennis Barden, Barry Mazur, and John Stallings around 1963–1965 ^{[BMS 1963-65]}, building on J. H. C. Whitehead's 1939–1950 theory of simple homotopy types. Milnor's survey Whitehead torsion (Bulletin of the AMS 72, 1966, 358–426) ^{[Milnor 1966]} organised the algebraic $K$ -theory of the group ring into the working tool $Wh (π) = K_{1} (Z [π]) / (\pm π)$ that measures the obstruction. Philosophically, the h-cobordism theorem marks the moment when high-dimensional manifold topology became algebraic: the diffeomorphism type of a cobordism is read from the vanishing of a homology group, and the residual subtlety in the non-simply-connected case is an element of a $K$ -group. The four-dimensional boundary, where Freedman's topological methods and Donaldson's smooth obstructions later showed the two categories splitting apart, stands as the permanent reminder that the whole edifice rests on a single geometric fact about discs and dimensions.

Bibliography Master

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

@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{milnor1966whitehead,
  author  = {Milnor, John W.},
  title   = {Whitehead torsion},
  journal = {Bulletin of the American Mathematical Society},
  volume  = {72},
  number  = {3},
  pages   = {358--426},
  year    = {1966}
}

@phdthesis{barden1963structure,
  author = {Barden, Dennis},
  title  = {The structure of manifolds},
  school = {University of Cambridge},
  year   = {1963}
}

@article{mazur1963relative,
  author  = {Mazur, Barry},
  title   = {Relative neighborhoods and the theorems of {S}male},
  journal = {Annals of Mathematics},
  volume  = {77},
  number  = {2},
  pages   = {232--249},
  year    = {1963}
}

@article{stallings1965whitehead,
  author  = {Stallings, John R.},
  title   = {Whitehead torsion of free products},
  journal = {Annals of Mathematics},
  volume  = {82},
  number  = {2},
  pages   = {354--363},
  year    = {1965}
}

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

Prerequisites

03.02.22
03.02.21
03.02.20

Tier anchors

beginner: Milnor Lectures on the h-Cobordism Theorem §8 informal; Matsumoto An Introduction to Morse Theory Ch. 3 (assembling cancellations)
intermediate: Milnor Lectures on the h-Cobordism Theorem §§6-8 (the h-cobordism theorem assembled); Kosinski Differential Manifolds Ch. IX
master: Milnor Lectures on the h-Cobordism Theorem §§6-9; Smale On the structure of manifolds (1962); Smale Generalized Poincaré's conjecture (1961); Barden-Mazur-Stallings s-cobordism; Milnor Whitehead torsion (1966)

References

03-modern-geometry/Bott-LecturesOnMorseTheory.pdf · Handle cancellation, gradient-like vector fields, and the reduction of a handle decomposition to a product
Milnor — Lectures on the h-Cobordism Theorem (Princeton Mathematical Notes, 1965) · §§6-8: cancellation in the middle dimensions, elimination of low and high handles, the h-cobordism theorem and its corollaries (notes by L. Siebenmann and J. Sondow)
Smale — On the structure of manifolds (1962) · American Journal of Mathematics 84, pp. 387--399; the h-cobordism theorem and the handlebody calculus
Smale — Generalized Poincaré's conjecture in dimensions greater than four (1961) · Annals of Mathematics 74, pp. 391--406; the high-dimensional Poincaré conjecture as a corollary of handle cancellation
Barden — The structure of manifolds (Cambridge thesis, 1963); Mazur — Relative neighborhoods and the theorems of Smale (1963); Stallings — Whitehead torsion of free products (1965) · The s-cobordism theorem: an h-cobordism is a product iff its Whitehead torsion vanishes
Milnor — Whitehead torsion (1966) · Bulletin of the AMS 72, pp. 358--426; the Whitehead group Wh(pi) and the torsion of an acyclic based chain complex over Z[pi]

Estimated time

beginner: 18m
intermediate: 47m
master: 88m