03.15.08 · modern-geometry / morse-homology

The Morse Homology Theorem

shipped3 tiersLean: none

Anchor (Master): Schwarz *Morse Homology* Part II Ch. 6; Thom 1949; Milnor *Morse Theory* §3; Eilenberg-Steenrod 1952

Intuition Beginner

For the last several units you built a bookkeeping machine out of a height function: one slot per resting point, sorted by index, with a boundary rule that counts the signed streams between adjacent shelves. You also learned that this machine has a homology, and that the homology does not change when you tilt or reshape the height function. The natural worry is whether this homemade machine measures anything real, or whether it is just an internal curiosity of gradient flows.

This unit answers that worry. The homology of the Morse machine is the same as the ordinary homology of the underlying shape — the count of holes that any other method would report. A doughnut has one piece, two loops, one cavity; the Morse machine for any height function on the doughnut reports exactly those numbers. The streams between resting points have been measuring the topology all along.

There are two honest ways to see why. One way reshapes the height function so its sublevel sets grow one cell at a time, turning the manifold into a tidy block structure whose own homology is visibly the Morse homology. The other way checks that the Morse machine obeys the same handful of structural rules that pin down ordinary homology uniquely, so the two must agree. Either way, the conclusion is that the height function was never arbitrary: its critical points and their connecting streams encode the shape itself.

Visual Beginner

Alt text: A two-panel figure. The left panel shows an upright torus whose height function has a bottom, two saddles, and a top; each critical point's descending region is shaded as a cell, and the four cells tile the surface. The right panel places the Morse complex beside the cellular chain complex of that tiling, joined by a bracket reading "same boundary numbers" and an equals sign captioned "Morse homology = ordinary homology." The picture conveys that the descending cells of the height function form a cell structure whose own homology is the Morse homology, which is the ordinary homology of the shape.

Worked example Beginner

Take complex projective space in its smallest interesting size, the surface of complex dimension one, which is an ordinary sphere. A standard height function on it has just two resting points: a bottom of index $0$ and a top of index $2$ , with nothing of index $1$ in between. The Morse machine has one slot on shelf $0$ , no slots on shelf $1$ , and one slot on shelf $2$ .

Now run the boundary rule. With no slots on the middle shelf, every boundary count is a sum over an empty collection, so the rule always returns nothing. The machine has shelves of size $1$ , $0$ , $1$ and a boundary rule that does nothing at all.

Reading off the homology is then immediate: shelf $0$ keeps its one slot, shelf $1$ keeps its zero slots, shelf $2$ keeps its one slot. The homology is $1$ , $0$ , $1$ .

Compare this to the ordinary homology of a sphere, computed by any standard method: one piece, no loops, one cavity, that is $1$ , $0$ , $1$ . The two answers match exactly. The height function with two resting points has reported the topology of the sphere correctly.

What this tells us: the larger projective spaces work the same way. The standard height function on the projective space of complex dimension $n$ has one resting point in each even index $0, 2, 4, \dots, 2 n$ and none in odd indices. No two of these shelves are adjacent, so the boundary rule does nothing, and the Morse homology is one copy of the integers in every even degree up to $2 n$ . That is precisely the known homology of these spaces — the machine recovers it for free.

Check your understanding Beginner

Question 2 (easy, multiple-choice). On a sphere with a two-resting-point height function (a bottom of index 0 and a top of index 2), why is every boundary count zero?

Because the two resting points are at different heights.
Because there are no resting points on the middle shelf to connect them.
Because the sphere has no holes.
Because the streams cancel in opposite-sign pairs.

Hint

The boundary rule only counts streams between adjacent shelves; what sits on shelf 1 here?

Answer

B. With no index-1 resting points, the boundary rule sums over an empty collection and returns nothing. Feedback if correct: exactly — the missing middle shelf forces every count to be empty. Feedback if wrong: the rule counts streams to the shelf one below; since shelf 1 is empty, there is nothing to count, so every boundary is zero.

Formal definition Intermediate+

Fix a closed Riemannian manifold $(M^{n}, g)$ and a Morse-Smale pair $(f, g)$ with Morse homology $H M_{*} (f, g) = H_{*} (C_{*} (f), \partial)$ as constructed in 03.15.06, invariant under the pair by the continuation isomorphisms of 03.15.07. The object to be compared is the singular homology $H_{*} (M; Z)$ of 03.12.11 and, through the cellular comparison, the cellular homology $H_{*}^{cell} (M; Z)$ of 03.12.13.

A Morse function $f$ is self-indexing when $f (x) = μ (x)$ for every critical point $x$ ; equivalently, the critical values are exactly the indices $0, 1, \dots, n$ , with all index- $k$ critical points on the level $f = k$ . By Smale's theorem ^{[Smale 1961]}, every Morse-Smale pair on a closed manifold can be deformed, through Morse-Smale pairs, to one whose function is self-indexing, without changing $H M_{*}$ by the invariance of 03.15.07.

For a self-indexing $f$ , the closed sublevel sets $M^{k + 1/2} = f^{- 1} (- \infty, k + \frac{1}{2}]$ form an increasing filtration $$ \varnothing = M^{-1/2} \subseteq M^{1/2} \subseteq \cdots \subseteq M^{n+1/2} = M, $$ and passing the level $k$ attaches one cell of dimension $k$ — the descending cell $D^{u} (x) = W^{u} (x) \cap M^{k + 1/2}$ — for each index- $k$ critical point $x$ , by the handle-attachment theorem of 03.02.31. This realises $M$ as a CW complex with one $k$ -cell per index- $k$ critical point, the cell decomposition by descending (unstable) manifolds of Thom ^{[Thom 1949]}.

The cellular chain group of this CW structure is $C_{k}^{cell} (M) = H_{k} (M^{k + 1/2}, M^{k - 1/2}) ≅ ⨁_{μ (x) = k} Z$ , free abelian on the $k$ -cells, hence on the index- $k$ critical points — abstractly the same group as the Morse chain group $C_{k} (f)$ . The comparison theorem asserts that the cellular boundary $d_{k} : C_{k}^{cell} \to C_{k - 1}^{cell}$ , computed by the degrees of the attaching maps, coincides with the Morse boundary $\partial_{k}$ , computed by the signed flow-line counts $n (x, y)$ .

A homology theory $h_{*}$ on the category of pairs of spaces is an ordinary homology theory when it satisfies the Eilenberg-Steenrod axioms of 03.12.15: homotopy invariance, the long exact sequence of a pair, excision, additivity, and the dimension axiom $h_{k} (pt) = 0$ for $k \neq = 0$ . Two ordinary homology theories with the same coefficient group $h_{0} (pt)$ are naturally isomorphic on CW pairs.

Counterexamples to common slips

The isomorphism $H M_{*} (f, g) ≅ H_{*} (M)$ is canonical (independent of $(f, g)$ ) only after the continuation identification of 03.15.07 is invoked; for a fixed pair the comparison isomorphism is constructed, but its independence of the pair is a separate input.
The descending cells $D^{u} (x)$ give a CW structure only when $f$ is self-indexing (or, more weakly, when the unstable and stable manifolds are arranged so that the closure of each cell meets only lower cells). For a general Morse-Smale pair the closure of an unstable manifold can fail to be a subcomplex, and the naive cell decomposition need not be CW.
The cellular-boundary degree $d_{k} (x) = \sum_{y} [x : y] y$ and the Morse count $\partial_{k} (x) = \sum_{y} n (x, y) y$ agree as integers, but only because the same coherent orientation is used to orient the cells and to sign the flow lines; a mismatch of conventions changes both by the same diagonal sign and leaves the homology unchanged.
Naturality is in $M$ , not in the pair: a smooth map $φ : M \to N$ induces $φ_{*}$ on singular homology, and the comparison isomorphisms intertwine it with a map on Morse homology, but there is in general no Morse function on $M$ carried to one on $N$ by $φ$ , so the naturality is mediated through singular homology.

Key theorem with proof Intermediate+

The theorem the book is named for closes the chapter: the homemade Morse machine computes ordinary homology ^{[Schwarz Part II Ch. 6]}.

Theorem (Morse Homology Theorem). Let $(M^{n}, g)$ be a closed Riemannian manifold and $(f, g)$ a Morse-Smale pair. There is an isomorphism of graded abelian groups $$ HM_k(f,g) ;\cong; H_k(M;\mathbb{Z}) \qquad (k = 0, 1, \dots, n), $$ canonical (independent of the pair $(f, g)$ via the continuation isomorphisms of 03.15.07) and natural in $M$ .

Proof (cellular route). By Smale's theorem ^{[Smale 1961]} deform $(f, g)$ , through Morse-Smale pairs, to a self-indexing pair; the invariance of 03.15.07 identifies the two Morse homologies, so it suffices to treat the self-indexing case. The sublevel filtration $M^{k + 1/2}$ then attaches exactly the index- $k$ critical cells at level $k$ (handle attachment, 03.02.31), giving a CW structure on $M$ with one $k$ -cell $e_{x}$ per index- $k$ critical point $x$ , the descending-cell decomposition of Thom ^{[Thom 1949]}.

Identify the cellular chain group $C_{k}^{cell} (M) = H_{k} (M^{k + 1/2}, M^{k - 1/2}) ≅ ⨁_{μ (x) = k} Z ⟨ e_{x} ⟩$ with the Morse chain group $C_{k} (f) = ⨁_{μ (x) = k} Z ⟨ x ⟩$ by $e_{x} \mapsto x$ . It remains to match the boundary maps. The cellular boundary degree is the incidence number $[e_{x} : e_{y}]$ for $μ (x) = k$ , $μ (y) = k - 1$ , equal to the degree of the composite $$ S^{k-1} = \partial e_x \xrightarrow{\ \text{attach}\ } M^{k-1/2} \longrightarrow M^{k-1/2}/M^{k-3/2} = \bigvee_y S^{k-1}y \xrightarrow{\ \text{collapse}\ } S^{k-1}y, $$ the last map collapsing all wedge summands but $e_{y}$ . Schwarz's lemma (Part II Ch. 6) computes this degree as the signed count of the negative-gradient trajectories from $x$ to $y$ : the attaching sphere of $e_{x}$ meets the ascending sphere of $y$ transversally in $W^{u} (x) \cap W^{s} (y) \cap {f = k - 1/2}$ , a finite set in bijection with $M (x, y)$ , and the local degree at each intersection point is the characteristic sign $n (u)$ of the corresponding flow line $u$ . Therefore $$ [e_x : e_y] = \sum{u\in\widehat{\mathcal{M}}(x,y)} n(u) = n(x,y), $$ so the cellular boundary $d_{k}$ equals the Morse boundary $\partial_{k}$ under the identification $e_{x} \mapsto x$ . The two chain complexes are equal, hence so are their homologies: $$ HM_k(f,g) = H_k(C(f),\partial) = H_k(C_^{\mathrm{cell}}(M), d) = H_k^{\mathrm{cell}}(M) \cong H_k(M;\mathbb{Z}), $$ the last isomorphism being the agreement of cellular and singular homology of 03.12.13. Naturality in $M$ is the naturality of the cellular-to-singular comparison. $□$

The signs require the coherent orientation of 03.15.05 to be used simultaneously for the cells and the flow lines; a change of orientation conjugates both boundary operators by the same diagonal matrix.

Bridge. This theorem builds toward the cohomological and dualised refinements that appear again in 03.15.09 (Morse cohomology and the cup product) and 03.15.10 (Poincaré duality via flow reversal), where the same complex is read in the opposite grading. The foundational reason the Morse machine computes ordinary homology is that its boundary operator is a cellular boundary in disguise: counting gradient trajectories between adjacent indices is exactly computing the attaching-map degrees of the descending-cell CW structure, and this is the central insight that identifies the analytic count with the topological one. Putting these together, the entire trajectory-moduli apparatus — transversality, compactness, gluing, orientations — is revealed as a long computation of one classical invariant, the cellular chain complex; the bridge is that the signed flow-line count $n (x, y)$ and the incidence number $[e_{x} : e_{y}]$ are the same integer. This is dual to the cohomological story, and the same identification appears again in 05.08.02 (Floer homology), where the loop-space analogue of this theorem is replaced by an invariance statement because no finite cell structure is available.

Exercises Intermediate+

Exercise 3 (medium, symbolic). Let $(f,g)$ be a Morse-Smale pair on a closed $M^n$. Using the Morse Homology Theorem, prove the equality of Euler characteristics $\sum_x (-1)^{\mu(x)} = \chi(M)$, where the left sum is over critical points.

Hint

The graded Euler characteristic of a finite free complex equals that of its homology; apply the theorem to identify the homology.

Answer

Rubric. Full credit: $\sum_{x} (- 1)^{μ (x)} = \sum_{k} (- 1)^{k} rank C_{k} (f)$ since $rank C_{k}$ counts index- $k$ critical points. For any finite complex of finitely generated free abelian groups, the alternating sum of chain ranks equals the alternating sum of homology ranks. By the Morse Homology Theorem $H M_{k} ≅ H_{k} (M; Z)$ , so $\sum_{k} (- 1)^{k} rank H M_{k} = \sum_{k} (- 1)^{k} b_{k} (M) = χ (M)$ . Combining gives $\sum_{x} (- 1)^{μ (x)} = χ (M)$ , the Poincaré-Hopf theorem for the gradient field.

Exercise 4 (medium, short-answer). Explain why the descending cells of a self-indexing Morse function give a CW structure with one $k$-cell per index-$k$ critical point, and why self-indexing is needed for the cell-closure condition.

Hint

Use handle attachment at each critical level; the closure of a $k$ -cell must meet only cells of dimension $< k$ .

Answer

Rubric. Full credit: passing the critical level $k$ attaches a $k$ -handle, homotopy-equivalent to a $k$ -cell, for each index- $k$ critical point (handle attachment, 03.02.31); the unstable manifold $W^{u} (x)$ intersected with the sublevel set is the open $k$ -cell $D^{u} (x)$ . For this to be a CW structure, the closure of each $k$ -cell must lie in the $k$ -skeleton, i.e. meet only cells of dimension $\leq k$ . Self-indexing $f (x) = μ (x)$ forces every index- $< k$ critical point onto a strictly lower level, so the boundary of $D^{u} (x)$ (a limit of trajectories descending to lower critical points) lands in the union of lower-index cells. Without self-indexing, an index- $k$ unstable manifold could limit onto another index- $k$ critical point at the same level, violating the skeletal-closure condition.

Exercise 5 (medium, symbolic). For the real projective plane $\mathbb{RP}^2$, a Morse function has critical points of indices $0, 1, 2$ (one each), and the index-$1$ to index-$0$ count is $n(s,b) = 0$ while the index-$2$ to index-$1$ count is $n(t,s) = \pm 2$. Compute $HM_*(\mathbb{RP}^2;\mathbb{Z})$ and confirm it matches the known integer homology.

Hint

With $C_{2} = C_{1} = C_{0} = Z$ and $\partial_{2} = (\pm 2)$ , $\partial_{1} = 0$ , compute kernels and images.

Answer

Rubric. Full credit: $C_{0} = C_{1} = C_{2} = Z$ . The boundary $\partial_{2} : Z \to Z$ is multiplication by $\pm 2$ and $\partial_{1} = 0$ . Then $H M_{0} = ker \partial_{0} / im \partial_{1} = Z /0 = Z$ ; $H M_{1} = ker \partial_{1} / im \partial_{2} = Z /2 Z = Z /2$ ; $H M_{2} = ker \partial_{2} /0 = 0$ (since $\times 2$ is injective). So $H M_{*} = (Z, Z /2, 0)$ , matching $H_{*} (RP^{2}; Z)$ . Here the cellular incidence number $\pm 2$ is the degree of the attaching map of the $2$ -cell, the same $\pm 2$ counted by the two oppositely-or-equally signed flow lines, illustrating the cellular comparison directly.

Exercise 6 (hard, short-answer). Sketch the axiomatic (Eilenberg-Steenrod) proof that $HM_*$ equals singular homology, identifying which axiom each ingredient of the Morse construction supplies, and where the relative groups $HM_*(M, A)$ come from.

Hint

Build relative Morse homology from a Morse function adapted to a sublevel-set pair, then verify homotopy invariance, exactness, excision, additivity, and the dimension axiom.

Answer

Rubric. Full credit identifies: (homotopy invariance and well-definedness) from the continuation isomorphisms 03.15.07, making $H M_{*}$ a functor independent of the pair; (long exact sequence of a pair) from a Morse function with $A = M^{a}$ a regular sublevel set, where the critical points of $f$ above level $a$ generate a relative complex $C_{*} (f; M, A)$ and the short exact sequence of complexes $0 \to C_{*} (A) \to C_{*} (M) \to C_{*} (M, A) \to 0$ yields the long exact sequence; (excision) from the locality of gradient trajectories — flow lines counted in $C_{*} (M, A)$ lie in $M ∖ A$ , so removing an interior set away from the relevant critical points does not change the count; (additivity) from the direct-sum decomposition of the complex over connected components; (dimension axiom) from $H M_{*} (pt) = Z$ in degree $0$ , since a point carries the unique constant Morse function with one index- $0$ critical point. By the uniqueness theorem 03.12.15, any homology theory satisfying these axioms with $H M_{0} (pt) = Z$ agrees with singular homology on CW pairs, hence $H M_{*} ≅ H_{*} (-; Z)$ .

Exercise 7 (hard, short-answer). Two Morse-Smale pairs $(f_0,g_0)$ and $(f_1,g_1)$ on the same $M$ give isomorphic Morse homologies by continuation. Show that the cellular-comparison isomorphism $HM_*(f_i,g_i) \cong H_*(M)$ is compatible with the continuation isomorphism $HM_*(f_0,g_0)\cong HM_*(f_1,g_1)$, so that the identification with $H_*(M)$ is independent of the pair.

Hint

Both routes land in the fixed group $H_{*} (M)$ ; show the triangle of isomorphisms commutes by reducing each pair to a common self-indexing refinement.

Answer

Rubric. Full credit: let $Φ : H M_{*} (f_{0}, g_{0}) \to H M_{*} (f_{1}, g_{1})$ be the continuation isomorphism and $c_{i} : H M_{*} (f_{i}, g_{i}) \to H_{*} (M)$ the cellular-comparison isomorphisms. The claim is $c_{1} \circ Φ = c_{0}$ . Reduce both pairs to self-indexing form by Smale's theorem; the continuation map between a pair and its self-indexing deformation is induced by a monotone homotopy, and on the cellular side this homotopy is realised by a cellular map of the induced CW structures inducing the identity on $H_{*} (M)$ (since the CW structures are different cell decompositions of the same $M$ and the comparison to singular homology is canonical). Chasing the resulting commutative diagram of cellular and continuation maps gives $c_{1} \circ Φ = c_{0}$ . Hence the composite $H M_{*} (f, g) c H_{*} (M)$ is the same for every pair, and $H M_{*} (M) := H M_{*} (f, g)$ is canonically $H_{*} (M)$ . Full credit requires the explicit triangle and the reduction to a common self-indexing model; partial credit for stating the triangle without justifying commutativity.

Lean formalization Intermediate+

Mathlib carries singular homology, cellular homology via its CWComplex API, chain complexes and their homology (HomologicalComplex), and the comparison of chain-homotopic maps on homology, so the algebraic statement that two equal chain complexes have equal homology is available. What is missing is the Morse homology functor and the geometric identification of its differential with a cellular boundary, so this unit ships at lean_status: none. The pseudo-Lean below indicates the intended shape; it does not compile because morseHomology, selfIndexing, and descendingCWStructure are not defined in Mathlib.

-- Illustrative only; not wired into the Lean build (lean_status: none).

-- A self-indexing Morse-Smale pair: f(x) = index f x on critical points.
structure SelfIndexing (g) (f) [MorseSmale g f] : Prop where
  value_eq_index : ∀ x, IsCrit f x → f x = (index f x : ℝ)

-- The descending-cell CW structure of a self-indexing pair: one k-cell
-- per index-k critical point (handle attachment).
noncomputable def descendingCWStructure (g) (f) [MorseSmale g f]
    (hs : SelfIndexing g f) : CWComplex M := sorry

-- The Morse complex equals the cellular complex of the descending CW
-- structure: same generators, and ∂(Morse) = d(cellular) via the
-- attaching-degree = signed flow count identity.
theorem morseComplex_eq_cellular (o : CoherentOrientation g f)
    (hs : SelfIndexing g f) (k : ℕ) :
    morseBoundary o k = cellularBoundary (descendingCWStructure g f hs) k :=
  sorry  -- [e_x : e_y] = n(x,y): attaching-map degree = signed trajectory count

-- The Morse Homology Theorem: HM_* ≅ H_*(M; ℤ), natural in M.
theorem morseHomology_iso_singularHomology (g) (f) [MorseSmale g f]
    (o : CoherentOrientation g f) (k : ℕ) :
    morseHomology o k ≃+ singularHomology M k :=
  sorry  -- cellular route via morseComplex_eq_cellular, or ES-axiom route

The proof gap is substantive. Mathlib needs the Morse homology functor (already named as the gap for 03.15.02-03.15.06), Smale's self-indexing theorem, the descending-cell CW structure with its handle-attachment construction, and either (a) the attaching-degree-equals-signed-count lemma feeding morseComplex_eq_cellular, or (b) a verification that morseHomology satisfies the Eilenberg-Steenrod axioms feeding the uniqueness theorem. Each is a separate formalisation target; the algebraic homology engine and the singular/cellular comparison already exist.

Advanced results Master

The theorem admits two complete proofs, and each carries its own corollaries: the cellular route exposes the chain-level coincidence, while the axiomatic route exposes the universal property.

The axiomatic proof (Schwarz's internal route). Continuation 03.15.07 makes $H M_{*}$ a functor on smooth manifolds, independent of the Morse-Smale pair. Schwarz constructs a relative theory $H M_{*} (M, A)$ for $A$ a compact sublevel set, fitting into the long exact sequence of the pair; verifies homotopy invariance from continuation under homotopies of the function; verifies excision from the locality of gradient flow lines, which never leave the region between the relevant critical levels; and computes $H M_{*} (pt) = Z$ concentrated in degree zero from the unique constant Morse function on a point. By the Eilenberg-Steenrod uniqueness theorem 03.12.15, a homology theory satisfying these axioms with integer coefficient group is naturally isomorphic to singular homology on CW pairs ^{[Eilenberg-Steenrod 1952]}. This proof never mentions cells: it identifies $H M_{*}$ with $H_{*}$ by their shared universal property, the route Schwarz takes precisely to keep the construction self-contained and to make the analytic Morse theory a stand-alone homology theory.

The cellular proof (Thom-Smale-Milnor comparison). Self-index the function (Smale ^{[Smale 1961]}); the descending cells give a CW structure (Thom ^{[Thom 1949]}, handle attachment Milnor ^{[Milnor 1963]}); the cellular boundary degrees equal the signed flow counts; the complexes coincide. This proof is concrete and computational and is the one that makes the Poincaré-Hopf and Morse-inequality corollaries immediate.

Naturality and the functor on the Morse groupoid. The collection of Morse-Smale pairs on a fixed $M$ , with continuation isomorphisms as morphisms, forms a groupoid on which $H M_{*}$ is a functor to graded abelian groups; the Morse Homology Theorem identifies this functor's value with the constant $H_{*} (M; Z)$ . A smooth map $φ : M \to N$ is not a morphism of this groupoid, but the comparison isomorphisms intertwine $φ_{*}$ on singular homology with an induced map on Morse homology, so $H M_{*}$ is natural in $M$ through the singular comparison.

The Morse inequalities recovered. The existence of the complex $(C_{*} (f), \partial)$ with $H_{*} ≅ H_{*} (M)$ gives the weak inequalities $c_{k} \geq b_{k} (M)$ and the strong inequalities $M_{t} - P_{t} = (1 + t) Q_{t}$ with $Q_{t} \geq 0$ , exactly as in 03.02.31, now as a consequence of the chain-level theorem rather than of sublevel-set induction. The two derivations agree because the underlying complex is the same.

Field and twisted coefficients. Tensoring the integral Morse complex with a field $F$ computes $H M_{*} (f, g; F) ≅ H_{*} (M; F)$ by the universal coefficient theorem 03.12.18 applied to the free complex $C_{*} (f)$ . Over $Z /2$ the coherent orientation is unnecessary and the differential counts flow lines without signs, recovering $H_{*} (M; Z /2)$ . Twisting by a representation $π_{1} (M) \to GL (V)$ gives Morse homology with local coefficients, equal to singular homology with the same local system.

Synthesis. The foundational reason the Morse machine computes ordinary homology is that its boundary operator is a cellular boundary read through the dictionary "signed trajectory count $=$ attaching-map degree," and this is exactly the identification that turns the analytic apparatus of the chapter into a computation of a classical invariant. Putting these together, the cellular route and the axiomatic route are two faces of one fact: the cellular route exhibits the chain-level coincidence $n (x, y) = [e_{x} : e_{y}]$ , while the axiomatic route exhibits $H M_{*}$ as a homology theory pinned down by its universal property, and the bridge between them is that the descending-cell CW structure is the geometric witness of the universal property's uniqueness. The central insight is that gradient flow was computing the homology of $M$ all along, so the Poincaré-Hopf identity, the Morse inequalities, and the universal-coefficient behaviour are corollaries of one theorem rather than separate facts. This same pattern appears again in 05.08.02 (Floer homology) and 03.07.23 (instanton Floer homology), where no finite cell structure exists and the cellular route is unavailable: there only the axiomatic-style invariance survives, which is why Schwarz designed the finite-dimensional theorem — where both routes work — as the rehearsal that makes the infinite-dimensional invariance statements intelligible.

Full proof set Master

The Key theorem section proves the cellular route in full. The propositions below supply the cellular-comparison lemma in detail and the axiomatic-route ingredients.

Proposition (attaching degree equals signed count). Let $(f, g)$ be a self-indexing Morse-Smale pair with a fixed coherent orientation, and let $x, y$ be critical points with $μ (x) = k$ , $μ (y) = k - 1$ . In the descending-cell CW structure, the cellular incidence number satisfies $[e_{x} : e_{y}] = n (x, y)$ .

Proof. The attaching map of the $k$ -cell $e_{x}$ is $a_{x} : S^{k - 1} = \partial D^{u} (x) \to M^{k - 1/2}$ , the restriction of the unstable manifold's boundary sphere. The incidence number $[e_{x} : e_{y}]$ is the degree of the composite of $a_{x}$ with the quotient $M^{k - 1/2} \to M^{k - 1/2} / M^{k - 3/2}$ followed by the collapse onto the $y$ -summand sphere. By transversality 03.15.02, the descending sphere $\partial D^{u} (x)$ meets the ascending sphere $S^{a} (y) = W^{s} (y) \cap {f = k - 1/2}$ in a finite set, and each intersection point lies on a unique negative-gradient trajectory from $x$ to $y$ , giving a bijection with $M (x, y)$ . Near each intersection point the collapse map is a local homeomorphism, and the local degree is $+ 1$ or $- 1$ according to whether the orientation of $\partial D^{u} (x)$ induced by the coherent orientation of $W^{u} (x)$ agrees with the orientation of the target sphere; this local sign is, by the definition of the coherent orientation 03.15.05, the characteristic sign $n (u)$ of the corresponding flow line. Summing the local degrees gives $[e_{x} : e_{y}] = \sum_{u \in M (x, y)} n (u) = n (x, y)$ . $■$

Proposition (the descending cells form a CW structure). For a self-indexing Morse-Smale pair on a closed $M^{n}$ , the descending manifolds ${D^{u} (x)}$ are the open cells of a CW structure on $M$ with one $k$ -cell per index- $k$ critical point.

Proof. By handle attachment 03.02.31, $M^{k + 1/2}$ is obtained from $M^{k - 1/2}$ by attaching one $k$ -handle per index- $k$ critical point, and each handle deformation-retracts to the descending cell $D^{u} (x) = W^{u} (x) \cap M^{k + 1/2}$ , an open $k$ -disk. The union $⋃_{x} D^{u} (x)$ is all of $M$ because every point flows down to a unique critical point under the negative gradient (the unstable manifolds partition $M$ ). The closure of $D^{u} (x)$ is the image of a characteristic map $\overline{D^{k}} \to M$ whose boundary lands in $M^{k - 1/2}$ ; self-indexing forces every critical point reached by the boundary trajectories to have index $< k$ and to lie on a strictly lower level, so $\overline{D^{u} (x)} ∖ D^{u} (x)$ is contained in the union of cells of dimension $< k$ . The closure-finiteness and weak-topology conditions hold because $M$ is compact with finitely many cells. Hence ${D^{u} (x)}$ is a CW structure. $■$

Proposition (excision for relative Morse homology). Let $A = M^{a}$ be a regular sublevel set of a Morse-Smale pair $(f, g)$ , and let $U \subseteq int (A)$ be open with $\overline{U}$ disjoint from $W^{u} (x) \cap W^{s} (y)$ for every pair of critical points $x \in / A$ , $y$ . Then the inclusion induces an isomorphism $HM_(M\setminus U, A\setminus U) \cong HM_*(M, A)$.*

Proof. The relative complex $C_{*} (f; M, A)$ is generated by the critical points of $f$ lying above level $a$ , with differential counting the flow lines between them that remain above level $a$ (lines descending into $A$ are quotiented out). A negative-gradient trajectory between two critical points above $a$ lies in $⋃_{x, y} W^{u} (x) \cap W^{s} (y)$ for those critical points, and by hypothesis avoids $\overline{U}$ . Removing $U$ therefore deletes no critical point above $a$ and alters no counted flow line, so the relative complexes of $(M, A)$ and $(M ∖ U, A ∖ U)$ are identical, and their homologies agree. $■$

Proposition (dimension axiom). Morse homology of a one-point space is $Z$ in degree zero and $0$ otherwise.

Proof. A point carries the unique Morse function, the constant, with a single critical point of index $0$ and no others. The Morse complex is $Z$ in degree zero with zero differential, so $H M_{0} (pt) = Z$ and $H M_{k} (pt) = 0$ for $k \neq = 0$ . $■$

The remaining axioms — homotopy invariance and the long exact sequence of a pair — are supplied directly by the continuation theory of 03.15.07 and the short exact sequence $0 \to C_{*} (A) \to C_{*} (M) \to C_{*} (M, A) \to 0$ of the sublevel filtration; together with the four propositions above and the uniqueness theorem 03.12.15, they complete the axiomatic proof. The strong Morse inequalities $M_{t} - P_{t} = (1 + t) Q_{t}$ follow from the $(1 + t)$ -divisibility of the difference of counting and Poincaré polynomials for any finite free complex, applied to $(C_{*} (f), \partial)$ with $H_{*} = H_{*} (M)$ by the theorem ^{[Schwarz Part II Ch. 6]}.

Connections Master

03.15.06 (the Morse complex and $\partial^{2} = 0$ ) is the object whose homology this theorem identifies. That unit builds the complex $(C_{*} (f), \partial)$ and proves the differential squares to zero; the present theorem supplies the missing half of its meaning, that the resulting homology is the manifold's own. Without 03.15.06 there is no complex to compare; without this unit the complex is an uninterpreted invariant of the gradient flow.

03.15.07 (continuation maps and invariance of $H M_{*}$ ) supplies the canonicity. The Morse Homology Theorem identifies $H M_{*} (f, g)$ with $H_{*} (M)$ for each pair, and the continuation isomorphisms of 03.15.07 make this identification independent of the pair, so the notation $H M_{*} (M)$ is licensed. In the axiomatic proof, continuation is exactly the homotopy-invariance axiom and the functoriality that turns $H M_{*}$ into a homology theory at all.

03.12.13 (cellular homology) is the comparison target of the cellular route. The descending cells of a self-indexing Morse function give a CW structure, and the cellular boundary degrees equal the signed flow-line counts, so the Morse complex is the cellular chain complex of this structure. The agreement of cellular and singular homology then delivers the theorem; cellular homology is the bridge between the analytic Morse count and singular topology.

03.12.15 (the Eilenberg-Steenrod axioms) is the comparison engine of the axiomatic route. Schwarz's internal proof verifies that $H M_{*}$ satisfies the seven axioms and concludes by the uniqueness theorem that it equals singular homology, without ever invoking cells. This is the route that makes Morse homology a self-contained ordinary homology theory.

03.02.31 (handle attachment, CW type, Morse inequalities) is the classical predecessor: Milnor computes the homotopy type of $M$ by attaching one handle per critical point, recovering the Morse inequalities by sublevel-set induction. The Morse Homology Theorem recovers the same inequalities from the chain complex, and the handle-attachment theorem is precisely what builds the CW structure used in the cellular proof.

05.08.02 (Floer homology) and 03.07.23 (instanton Floer homology) are the infinite-dimensional descendants. There no finite cell structure exists, so the cellular route is unavailable and only an invariance statement — the analogue of continuation, not of the Morse Homology Theorem — can be proved; Schwarz designed the finite-dimensional theorem, where both routes succeed, as the rehearsal that explains what the Floer invariance statements are approximating.

Historical & philosophical context Master

The descending-cell decomposition at the heart of the cellular proof is due to René Thom, Sur une partition en cellules associée à une fonction sur une variété (C.R. Acad. Sci. Paris 228, 1949, 973–975) ^{[Thom 1949]}, who observed that the unstable manifolds of a Morse function's gradient flow partition the manifold into open cells of dimensions equal to the Morse indices. That a self-indexing function can always be arranged, making the cell decomposition genuinely CW, was established by Stephen Smale, On gradient dynamical systems (Annals of Mathematics 74, 1961, 199–206) ^{[Smale 1961]}, in the same circle of ideas that produced his proof of the high-dimensional Poincaré conjecture. John Milnor's Morse Theory (Annals of Mathematics Studies 51, Princeton, 1963) ^{[Milnor 1963]} gave the handle-attachment account of the homotopy type and the Morse inequalities, the classical route the cellular proof formalises at chain level.

The axiomatic characterisation of ordinary homology that drives Schwarz's internal proof is the uniqueness theorem of Samuel Eilenberg and Norman Steenrod, Foundations of Algebraic Topology (Princeton, 1952) ^{[Eilenberg-Steenrod 1952]}: a homology theory on CW pairs satisfying the seven axioms is determined up to natural isomorphism by its coefficient group on a point. Edward Witten's Supersymmetry and Morse theory (Journal of Differential Geometry 17, 1982, 661–692) and Andreas Floer's late-1980s programme reframed the complex as an instanton count and exported it to infinite dimensions, but neither carried out the finite-dimensional construction with all analytic details. Matthias Schwarz's Morse Homology (Progress in Mathematics 111, Birkhäuser, 1993) ^{[Schwarz Part II Ch. 6]}, from his 1992 ETH Zürich dissertation under Eduard Zehnder and Dietmar Salamon, supplied that construction and proved the eponymous theorem by the axiomatic route, deliberately avoiding the cellular comparison so that Morse homology would stand as an independently constructed ordinary homology theory rather than a corollary of Milnor's handle calculus.

Bibliography Master

@book{Schwarz1993,
  author    = {Schwarz, Matthias},
  title     = {Morse Homology},
  series    = {Progress in Mathematics},
  volume    = {111},
  publisher = {Birkh\"auser Verlag, Basel},
  year      = {1993}
}

@article{Thom1949,
  author  = {Thom, Ren\'e},
  title   = {Sur une partition en cellules associ\'ee \`a une fonction sur une vari\'et\'e},
  journal = {Comptes Rendus de l'Acad\'emie des Sciences, Paris},
  volume  = {228},
  pages   = {973--975},
  year    = {1949}
}

@article{Smale1961,
  author  = {Smale, Stephen},
  title   = {On gradient dynamical systems},
  journal = {Annals of Mathematics},
  volume  = {74},
  number  = {1},
  pages   = {199--206},
  year    = {1961}
}

@book{Milnor1963,
  author    = {Milnor, John W.},
  title     = {Morse Theory},
  series    = {Annals of Mathematics Studies},
  volume    = {51},
  publisher = {Princeton University Press},
  year      = {1963}
}

@book{EilenbergSteenrod1952,
  author    = {Eilenberg, Samuel and Steenrod, Norman E.},
  title     = {Foundations of Algebraic Topology},
  publisher = {Princeton University Press},
  year      = {1952}
}

@article{Witten1982,
  author  = {Witten, Edward},
  title   = {Supersymmetry and {M}orse theory},
  journal = {Journal of Differential Geometry},
  volume  = {17},
  number  = {4},
  pages   = {661--692},
  year    = {1982}
}

@book{AudinDamian2014,
  author    = {Audin, Mich\`ele and Damian, Mihai},
  title     = {Morse Theory and Floer Homology},
  series    = {Universitext},
  publisher = {Springer-Verlag, London},
  year      = {2014}
}

@book{BanyagaHurtubise2004,
  author    = {Banyaga, Augustin and Hurtubise, David},
  title     = {Lectures on Morse Homology},
  series    = {Kluwer Texts in the Mathematical Sciences},
  volume    = {29},
  publisher = {Kluwer Academic Publishers, Dordrecht},
  year      = {2004}
}

Prerequisites

03.15.06
03.15.07
03.12.13
03.12.15

Tier anchors

beginner: Audin-Damian *Morse Theory and Floer Homology* §4.9 (Morse homology equals singular homology, informal); Banyaga-Hurtubise *Lectures on Morse Homology* Ch. 9
intermediate: Schwarz *Morse Homology* Part II Ch. 6; Audin-Damian §4.9; Banyaga-Hurtubise Ch. 9-10
master: Schwarz *Morse Homology* Part II Ch. 6; Thom 1949; Milnor *Morse Theory* §3; Eilenberg-Steenrod 1952

References

03-modern-geometry/Bott-LecturesOnMorseTheory.pdf · The descending-cell decomposition and the comparison of the Morse picture with the homology of the manifold (background)
Schwarz — Morse Homology (Progress in Mathematics 111, Birkhäuser 1993) · Part II, Ch. 6 (the Morse Homology Theorem: $HM_*(f,g) \cong H_*(M;\mathbb{Z})$ via the axiomatic uniqueness of an ordinary homology theory)
Thom — Sur une partition en cellules associée à une fonction sur une variété · C.R. Acad. Sci. Paris 228 (1949), pp. 973-975 (the partition of $M$ into descending cells, one of dimension $\mu(x)$ per critical point)
Milnor — Morse Theory · Annals of Mathematics Studies 51, Princeton 1963, §3 (the handle/CW homotopy type and the cellular comparison route)
Eilenberg-Steenrod — Foundations of Algebraic Topology · Princeton 1952, Ch. I (the seven axioms) and Ch. III (uniqueness: a homology theory is determined by its coefficient group on a point)
Smale — On gradient dynamical systems · Annals of Mathematics 74 (1961), pp. 199-206 (existence of self-indexing Morse functions for the Morse-Smale condition)

Estimated time

beginner: 18m
intermediate: 48m
master: 88m