03.15.01 · modern-geometry / morse-homology

Gradient flow, stable/unstable manifolds, and the Morse-Smale condition

shipped3 tiersLean: none

Anchor (Master): Schwarz Morse Homology Part I §2-3; Smale 1961 (On gradient dynamical systems); Audin-Damian §3; Banyaga-Hurtubise Ch. 6

Intuition Beginner

Stand on a smooth hilly landscape and let go of a ball. It rolls straight downhill, always taking the steepest descent. Trace the whole path it follows and you get a single curve along the surface. Do this from every starting point and the whole landscape fills up with downhill paths. That family of steepest-descent paths is the gradient flow of the height function.

Each path has to end somewhere. A ball rolling down can only come to rest at a level spot — a valley bottom, a mountain pass, or, if it starts in just the right place, the very top. So every downhill path is heading toward one of the special level points you met when studying Morse functions. Watching where paths come from and where they go to turns a static landscape into a story about motion.

This motion is the bridge between the local picture (what one level spot looks like up close) and the global picture (how the whole shape is glued together). The level points are the actors; the downhill paths between them are the plot.

Visual Beginner

Picture the standing doughnut (a torus on its edge) with its height function. It has four level spots: a bottom (a bowl), two middle saddles, and a top (an inverted bowl). Now draw the downhill paths.

Look at one saddle. Some starting points flow into it; they form a curve. Other points start at the saddle and flow away from it; they also form a curve. Collecting "everything that flows into a level spot" and "everything that flows out of it" is the whole idea of this unit. The sizes of those collections record the index you already know.

Worked example Beginner

Take the round sphere with its height function. There are two level spots: the south pole (bottom) and the north pole (top). Release a ball anywhere except exactly the north pole, and it rolls down to the south pole.

So the collection of starting points that flow down to the south pole is almost the entire sphere — every point except the single north pole. That is a two-dimensional collection. The south pole is a bowl, index $0$ .

Now run the clock backward, so balls roll uphill instead. Starting anywhere except the south pole, a ball rolls up to the north pole. The collection of points that flow up to the north pole is again almost the whole sphere. The north pole is an inverted bowl, index $2$ .

What this tells us: at the index- $0$ bottom, the "flows into it" collection is big (dimension $2$ ) and the "flows out of it" collection is tiny (just the point itself, dimension $0$ ). At the index- $2$ top it is the reverse. The size of the "flows out" collection equals the index. That single rule — out-flow size equals index — is the engine of everything that follows.

Check your understanding Beginner

Formal definition Intermediate+

Let $(M, g)$ be a closed (compact, boundaryless) Riemannian manifold of dimension $n$ and let $f \in C^{\infty} (M)$ be a Morse function, so every critical point is nondegenerate with a well-defined Morse index. The metric $g$ converts the differential $df$ into a vector field. The gradient $\nabla f \in X (M)$ is the unique smooth vector field with $g (\nabla f, X) = df (X)$ for every vector field $X$ ; in local coordinates $(\nabla f)^{i} = g^{ij} \partial_{j} f$ , where $(g^{ij})$ is the inverse metric matrix. The negative gradient flow is the flow $φ_{t} : M \to M$ of $- \nabla f$ , the one-parameter family of diffeomorphisms solving

\frac{d}{d t} φ_{t} (x) = - \nabla f (φ_{t} (x)), φ_{0} (x) = x .

Because $M$ is compact the field is complete, so $φ_{t}$ is defined for all $t \in R$ . Along any trajectory $γ (t) = φ_{t} (x)$ the function decreases monotonically:

\frac{d}{d t} f (γ (t)) = df (\overset{γ}{˙}) = g (\nabla f, - \nabla f) = - ∣\nabla f (γ (t)) ∣_{g}^{2} \leq 0,

with equality at a parameter value $t$ exactly when $γ (t)$ is a critical point. The set of critical points of $f$ is denoted $Crit (f)$ ; on a closed manifold a Morse function has finitely many of them.

For a critical point $p \in Crit (f)$ , define the stable and unstable manifolds

W^{s} (p) = {x \in M : t \to + \infty lim φ_{t} (x) = p}, W^{u} (p) = {x \in M : t \to - \infty lim φ_{t} (x) = p} .

A point lies in $W^{s} (p)$ when its forward trajectory limits onto $p$ , and in $W^{u} (p)$ when its backward trajectory does — equivalently, when its forward $(- \nabla f)$ -trajectory emanates from $p$ . The unstable manifold of $p$ is the stable manifold of $p$ for the reversed flow, that is for $+ \nabla f$ , so the two notions are interchanged by $f \mapsto - f$ .

The structural fact, established below, is that each $W^{s} (p)$ and $W^{u} (p)$ is an embedded (open) submanifold diffeomorphic to a Euclidean cell, with

dim W^{u} (p) = λ (p), dim W^{s} (p) = n - λ (p),

where $λ (p)$ is the Morse index. Every point of $M$ lies on exactly one full trajectory, and that trajectory limits onto a critical point in each time direction (forward and backward), so

M = p \in Crit (f) ⨆ W^{u} (p) = p \in Crit (f) ⨆ W^{s} (p)

are two partitions of $M$ into cells — the unstable (descending) and stable (ascending) decompositions.

The pair $(f, g)$ is said to satisfy the Morse-Smale condition when, for every pair of critical points $p, q$ , the unstable manifold of $p$ meets the stable manifold of $q$ transversally:

W^{u} (p) ⋔ W^{s} (q) for all p, q \in Crit (f) .

When this holds, $(f, g)$ is called a Morse-Smale pair. Transversality forces the intersection $W^{u} (p) \cap W^{s} (q)$ to be a submanifold of dimension $dim W^{u} (p) + dim W^{s} (q) - n = λ (p) - λ (q)$ . The negative gradient field is tangent to this intersection and nowhere zero on it (away from $p, q$ ), so the additive group $R$ acts freely by the flow, and the moduli space of flow lines

M (p, q) = (W^{u} (p) \cap W^{s} (q)) / R

is a manifold of dimension $λ (p) - λ (q) - 1$ whenever $p \neq = q$ .

Counterexamples to common slips

The stable/unstable splitting requires nondegeneracy. For $f (x, y) = x^{2}$ on a region with a whole line of critical points, no isolated rest point exists, the linearisation is not hyperbolic, and " $W^{u} (p)$ " is not a cell of dimension equal to an index.
The Morse-Smale condition is a property of the pair $(f, g)$ , not of $f$ alone. The standard upright torus with the round metric is not Morse-Smale: the two saddles sit at the same critical level and are joined by a degenerate sheet of flow lines. A small tilt of the embedding (equivalently a small metric change) restores transversality. Morse-Smale is generic in $g$ , not automatic.
Transversality of $W^{u} (p)$ and $W^{s} (q)$ can hold vacuously: if $λ (p) < λ (q)$ the expected intersection dimension $λ (p) - λ (q)$ is negative, so transversality means the intersection is empty. There are then no flow lines from $p$ down to $q$ — descent cannot raise the index.
The dimension count $dim W^{u} (p) = λ (p)$ uses the negative gradient flow. For the positive gradient flow the roles of stable and unstable swap, and the unstable manifold of $p$ has dimension $n - λ (p)$ . Fixing the sign convention once (here: $- \nabla f$ , $W^{u}$ of dimension $λ$ ) is mandatory.

Key theorem with proof Intermediate+

The two structural inputs are the realisation of critical points as hyperbolic rest points and the resulting cell structure of the descending manifolds. The treatment follows Schwarz ^{[Schwarz Part I §1-3]} and Bott ^{[fasttrack-texts Gradient flow of a Morse function]}.

Theorem (local stable manifold; descending-cell dimension). Let $p$ be a nondegenerate critical point of $f$ with Morse index $λ$ , for the negative gradient flow of $(M, g)$ . Then $p$ is a hyperbolic rest point of $- \nabla f$ : the linearisation of $- \nabla f$ at $p$ is a symmetric (with respect to $g_{p}$ ) endomorphism of $T_{p} M$ with no zero eigenvalue, having a $λ$ -dimensional negative eigenspace $E^{u}$ and an $(n - λ)$ -dimensional positive eigenspace $E^{s}$ . There are local stable and unstable manifolds $W_{loc}^{s} (p)$ , $W_{loc}^{u} (p)$ through $p$ , smoothly embedded discs tangent at $p$ to $E^{s}$ and $E^{u}$ respectively, with $dim W_{loc}^{u} (p) = λ$ and $dim W_{loc}^{s} (p) = n - λ$ . The global manifolds are their flow-out,

W^{u} (p) = t \in R ⋃ φ_{t} (W_{loc}^{u} (p)), W^{s} (p) = t \in R ⋃ φ_{t} (W_{loc}^{s} (p)),

each an injectively immersed submanifold diffeomorphic to $R^{λ}$ , respectively $R^{n - λ}$ .

Proof. Work in a chart centred at $p$ in which, by the Morse lemma, $f = f (p) - \frac{1}{2} \sum_{i \leq λ} x_{i}^{2} + \frac{1}{2} \sum_{i > λ} x_{i}^{2}$ (the factor $\frac{1}{2}$ rescales the normal form and is harmless). Linearise $- \nabla f$ at $p$ . The gradient is $\nabla f = g^{- 1} df$ , and at the critical point the derivative of $X \mapsto \nabla f$ in the direction $v$ is $g_{p}^{- 1} H_{p} f (v, \cdot)$ , where $H_{p} f$ is the Hessian bilinear form; the metric-dependent correction to the connection drops because $d f_{p} = 0$ . Hence the linearisation of $- \nabla f$ at $p$ is $L = - g_{p}^{- 1} H_{p} f$ , viewed as an endomorphism of $T_{p} M$ . Since $H_{p} f$ and $g_{p}$ are symmetric and $g_{p}$ is positive definite, $L$ is self-adjoint for the inner product $g_{p}$ and is conjugate to the symmetric matrix $- g_{p}^{- 1/2} H_{p} f g_{p}^{- 1/2}$ . Its eigenvalues are real and nonzero (the Hessian is nondegenerate). The negative eigenvalues of $H_{p} f$ — there are $λ$ of them, by the definition of the index — give positive eigenvalues of $L$ , hence directions in which $- \nabla f$ points away from $p$ : the unstable eigenspace $E^{u}$ , $dim E^{u} = λ$ . The remaining $n - λ$ eigenvalues of $L$ are negative, giving the stable eigenspace $E^{s}$ . No eigenvalue is zero, so $p$ is hyperbolic.

The Hadamard-Perron stable manifold theorem for a hyperbolic rest point of a smooth flow now applies: there exist local invariant manifolds $W_{loc}^{s} (p)$ and $W_{loc}^{u} (p)$ , smoothly embedded, tangent at $p$ to $E^{s}$ and $E^{u}$ , characterised as the points whose forward (respectively backward) trajectory stays in a neighbourhood of $p$ and converges to $p$ . Their dimensions are $dim E^{s} = n - λ$ and $dim E^{u} = λ$ .

To globalise, note that the local unstable manifold is invariant and that any point $x$ with $lim_{t \to - \infty} φ_{t} (x) = p$ enters $W_{loc}^{u} (p)$ for $t$ sufficiently negative, hence $x = φ_{t_{0}} (y)$ for some $y \in W_{loc}^{u} (p)$ and some $t_{0}$ . Therefore $W^{u} (p) = ⋃_{t} φ_{t} (W_{loc}^{u} (p))$ . Each $φ_{t}$ is a diffeomorphism and the union is an increasing nested family of embedded $λ$ -discs glued along the flow, so the flow-out is an injectively immersed copy of $R^{λ}$ : the map $E^{u} \times R \to W^{u} (p)$ built from the local chart and the flow is a bijective immersion, and the convergence at $p$ caps it off to a cell. The same argument with the reversed flow gives $W^{s} (p) ≅ R^{n - λ}$ . Compactness of $M$ ensures completeness of the flow, so no trajectory escapes to infinity and every trajectory limits onto critical points in both time directions, giving the decompositions $M = ⨆_{p} W^{u} (p) = ⨆_{p} W^{s} (p)$ . $■$

Bridge. This cell structure builds toward 03.15.02, where the intersection $W^{u} (p) \cap W^{s} (q)$ is recast as the zero set of a Fredholm section whose index reproduces $λ (p) - λ (q)$ , sharpening the dimension count proved here into an analytic transversality theorem. It appears again in 03.15.06, where the index-difference-one moduli $M (p, q)$ become the matrix entries of the Morse boundary operator $\partial$ , and the index-difference-two moduli, compactified by broken trajectories, force $\partial^{2} = 0$ . The unstable decomposition $M = ⨆_{p} W^{u} (p)$ is, for a Morse-Smale pair, a CW decomposition refining the handle picture of 03.02.31: each descending cell $W^{u} (p)$ is the open $λ (p)$ -cell whose closure attaches along lower cells, which is the dynamical realisation of the attaching maps that handle attachment supplies abstractly. And the genericity of the Morse-Smale condition, proved by Sard-Smale below, is the exact finite-dimensional rehearsal for the transversality arguments that make every Floer homology well-defined.

Exercises Intermediate+

Exercise (medium, short-answer). Explain why, for a Morse-Smale pair, there are no flow lines from a critical point $p$ to a critical point $q$ with $\lambda(q) \ge \lambda(p)$ (and $q \ne p$).

Hint

Consider the expected dimension $λ (p) - λ (q)$ of $W^{u} (p) \cap W^{s} (q)$ . What does transversality say when this number is $\leq 0$ ?

Answer

Rubric. Under Morse-Smale, $W^{u} (p) ⋔ W^{s} (q)$ forces $dim (W^{u} (p) \cap W^{s} (q)) = λ (p) - λ (q)$ when nonempty. If $λ (q) > λ (p)$ this number is negative, which is impossible for a nonempty intersection, so the intersection is empty: no flow lines. If $λ (q) = λ (p)$ and $q \neq = p$ , the intersection would be $0$ -dimensional and flow-invariant; but $R$ acts freely on a genuine flow line, forcing dimension $\geq 1$ , a contradiction unless the intersection is empty. So a nonconstant gradient trajectory strictly decreases the index.

Exercise (hard, short-answer). Show that the linearisation of $-\nabla f$ at a critical point $p$ is, in $g_p$-orthonormal coordinates, the symmetric matrix $-H$, where $H$ is the Hessian matrix of $f$ at $p$. Conclude that $p$ is a hyperbolic rest point with $\dim E^u = \lambda(p)$.

Hint

Choose normal coordinates so that $g_{p} = Id$ and the Christoffel symbols vanish at $p$ ; then $\nabla f = g^{- 1} df$ linearises to $g_{p}^{- 1} H = H$ at $p$ .

Answer

Rubric. In $g$ -normal coordinates at $p$ one has $g_{ij} (p) = δ_{ij}$ and $\partial_{k} g_{ij} (p) = 0$ . Then $(\nabla f)^{i} = g^{ij} \partial_{j} f$ , and its derivative at $p$ is $\partial_{k} (\nabla f)^{i} (p) = g^{ij} (p) \partial_{k} \partial_{j} f (p) + (\partial_{k} g^{ij}) (p) \partial_{j} f (p)$ . The second term vanishes because $\partial_{j} f (p) = 0$ , leaving $δ^{ij} \partial_{k} \partial_{j} f (p) = H_{k i}$ . So the linearisation of $- \nabla f$ is $- H$ , symmetric with real eigenvalues. The Hessian has $λ$ negative eigenvalues, giving $- H$ exactly $λ$ positive eigenvalues: a $λ$ -dimensional expanding (unstable) eigenspace and an $(n - λ)$ -dimensional contracting one. No eigenvalue is zero since $H$ is invertible, so $p$ is hyperbolic and $dim E^{u} = λ$ .

Exercise (hard, short-answer). The upright torus with the round metric and height function is not Morse-Smale: its two index-$1$ saddles $q_1, q_2$ sit at the same critical level. Explain qualitatively what goes wrong with the transversality $W^u(q_1) \pitchfork W^s(q_2)$, and why a generic small perturbation of the metric fixes it.

Hint

$W^{u} (q_{1})$ and $W^{s} (q_{2})$ are both $1$ -dimensional inside the $2$ -dimensional torus; for a flow line from $q_{1}$ to $q_{2}$ the height would have to strictly decrease, but the saddles are at equal height.

Answer

Rubric. A flow line from $q_{1}$ to $q_{2}$ requires $f (q_{1}) > f (q_{2})$ , but the two saddles are at equal height, so there can be no honest trajectory between them; the apparent intersection is the pair of horizontal circles meeting along a degenerate locus rather than transversally. More precisely, the descending and ascending arcs coincide along whole curves instead of meeting cleanly in isolated points of the right dimension. Tilting the torus slightly (or equivalently perturbing $g$ ) breaks the height-symmetry of the two saddles, separating their critical levels; the unstable arc of the upper saddle then crosses the stable arc of the lower one transversally in finitely many points, and Sard-Smale guarantees this transversal behaviour for a generic metric.

Lean formalization Intermediate+

Mathlib has Riemannian metrics, smooth vector fields, and the existence-uniqueness theory of their flows, but no gradient-Morse-dynamics layer, so this unit ships at lean_status: none. The statements below indicate the shape of a Mathlib contribution; they do not compile against current Mathlib because gradient, negGradFlow, IsHyperbolicCritical, stableManifold, and MorseSmale are not defined.

-- Intended Mathlib-facing statements (not yet in Mathlib).

-- The metric gradient: g (gradient g f x) v = mfderiv f x v.
def gradient (g : RiemannianMetric I M) (f : M → ℝ) : Π x, TangentSpace I x := ...

-- The negative gradient flow as the flow of -gradient.
def negGradFlow (g) (f) : ℝ → M → M := flowOf (fun x => - gradient g f x)

-- f decreases along the flow: derivative = -‖gradient‖².
theorem negGradFlow_monotone (g) (f) (x) (t) :
    deriv (fun s => f (negGradFlow g f s x)) t
      = - ‖gradient g f (negGradFlow g f t x)‖_[g]^2 := by sorry

-- Stable / unstable manifolds of a critical point.
def stableManifold (g) (f) (p : M) : Set M :=
  { x | Filter.Tendsto (fun t => negGradFlow g f t x) atTop (nhds p) }
def unstableManifold (g) (f) (p : M) : Set M :=
  { x | Filter.Tendsto (fun t => negGradFlow g f t x) atBot (nhds p) }

-- Dimension of the unstable manifold equals the Morse index.
theorem dim_unstable_eq_index
    (g) (f) (p) (hp : IsNondegenerateCriticalPoint f p) :
    Manifold.dim (unstableManifold g f p) = morseIndex f p hp := by sorry

-- Morse-Smale transversality, and the moduli dimension count.
def MorseSmale (g) (f) : Prop :=
  ∀ p q, Transverse (unstableManifold g f p) (stableManifold g f q)

theorem dim_intersection (g) (f) (h : MorseSmale g f) (p q) :
    Manifold.dim (unstableManifold g f p ∩ stableManifold g f q)
      = morseIndex f p _ - morseIndex f q _ := by sorry

The proof obligation behind dim_unstable_eq_index is the local stable manifold theorem for the hyperbolic rest point $p$ , whose linearisation is $- g_{p}^{- 1} H_{p} f$ ; behind MorseSmale genericity is the Sard-Smale theorem on a universal moduli over the Banach manifold of metrics. The Sard-Smale (Fredholm) version of Sard's theorem is the concrete contribution target the Mathlib gap analysis names.

Advanced results Master

The cell structure of the descending manifolds is the dynamical face of the handle decomposition. For a Morse-Smale pair $(f, g)$ the closure of an unstable cell satisfies $\overline{W^{u} (p)} \subseteq ⋃_{λ (q) \leq λ (p)} W^{u} (q)$ , so the unstable decomposition $M = ⨆_{p} W^{u} (p)$ is a CW decomposition with one $λ$ -cell per index- $λ$ critical point — Thom's 1949 observation ^{[Thom 1949]}. The attaching map of the cell $W^{u} (p)$ records which lower-index unstable cells the broken-trajectory boundary of $W^{u} (p)$ runs into, the same data that handle attachment encodes through framed embeddings of attaching spheres. The Morse-Smale hypothesis is precisely what makes the closure relation hold; without it $\overline{W^{u} (p)}$ can spiral and fail to be a subcomplex.

Smale isolated the transversality condition in 1961 ^{[Smale 1961]} and proved that it can always be achieved. Theorem (genericity of Morse-Smale, Smale-Sard-Smale form). For a fixed Morse function $f$ on a closed manifold $M$ , the set of Riemannian metrics $g$ for which $(f, g)$ is Morse-Smale is residual (a countable intersection of open dense sets, hence dense) in the space of $C^{k}$ metrics, $2 \leq k \leq \infty$ , with the $C^{k}$ topology. The proof realises the intersections $W^{u} (p) \cap W^{s} (q)$ as the zero set of a section of a Banach bundle over a path space, parametrised by the metric; the universal section is shown to be transverse to the zero section by an explicit perturbation; and the Sard-Smale theorem — the Fredholm-operator analogue of Sard, asserting that the regular values of a $C^{k}$ Fredholm map between separable Banach manifolds are residual provided $k$ exceeds the Fredholm index — extracts a residual set of metrics for which the parametrised section is transverse, which is the Morse-Smale condition. This is the finite-dimensional template for the transversality theorems of Floer theory: the perturbation space there is a space of almost-complex structures or Hamiltonians, and the same Sard-Smale argument delivers a generic well-defined complex.

The intersection dimension count refines further once the moduli space is unparametrised. When $λ (p) - λ (q) = 1$ , the manifold $M (p, q)$ is $0$ -dimensional, and a compactness argument (the broken-trajectory limits of the next unit) shows it is finite — these isolated flow lines, counted with signs from coherent orientations, are the entries of the Morse differential. When $λ (p) - λ (q) = 2$ , $M (p, q)$ is a $1$ -manifold whose ends are exactly the once-broken trajectories $p \to r \to q$ with $λ (r) = λ (p) - 1$ ; the count of those ends, being the boundary count of a compact $1$ -manifold, vanishes mod the orientation signs, and this is the geometric origin of $\partial^{2} = 0$ .

The whole construction is invariant under flow reversal. Replacing $f$ by $- f$ exchanges $\nabla f \leftrightarrow - \nabla f$ and hence $W^{u} (p; f) = W^{s} (p; - f)$ , $W^{s} (p; f) = W^{u} (p; - f)$ , while the index transforms as $λ_{- f} (p) = n - λ_{f} (p)$ (the Hessian negates, swapping its positive and negative eigenspaces). The Morse-Smale condition is symmetric in $p, q$ under this swap, so $(f, g)$ is Morse-Smale if and only if $(- f, g)$ is, and the flow-reversal symmetry is the dynamical seed of Poincaré duality between the Morse complexes of $f$ and $- f$ .

Synthesis. The gradient flow converts the local invariants of 03.02.30 into a global decomposition and a moduli problem. The Morse index becomes the dimension of the descending cell $W^{u} (p)$ , tying the linear-algebra count of negative Hessian eigenvalues to the dimension of an honest submanifold; the same index reappears in 03.15.02 as the Fredholm index $λ (p) - λ (q)$ of the linearised flow operator, in 03.15.06 as the homological grading of the Morse complex, and in 03.02.31 as the dimension of the attached handle. The Morse-Smale transversality condition is the genericity hypothesis that makes the moduli spaces $M (p, q)$ manifolds of the predicted dimension, and the Sard-Smale argument that secures it is the rehearsal every Floer theory replays in infinite dimensions. Flow reversal $f \mapsto - f$ interchanges stable and unstable manifolds and is the geometric source of Poincaré duality. Thom's descending-cell decomposition and Smale's transversality condition together are the precise sense in which "watching the landscape flow downhill" recovers the manifold's cell structure.

Full proof set Master

The local stable manifold theorem and the descending-cell dimension are proved in full in the Key theorem section. The remaining Master claims are recorded here.

Hyperbolicity and the eigenspace count. At a critical point $p$ the linearisation of $- \nabla f$ is $L = - g_{p}^{- 1} H_{p} f$ , self-adjoint for $g_{p}$ (since $g_{p}^{1/2} L g_{p}^{- 1/2} = - g_{p}^{- 1/2} H_{p} f g_{p}^{- 1/2}$ is symmetric), with eigenvalues the negatives of the generalised eigenvalues of $H_{p} f$ relative to $g_{p}$ . By Sylvester's law of inertia the count of negative eigenvalues of $H_{p} f$ is $λ (p)$ regardless of the positive-definite $g_{p}$ used, so $L$ has exactly $λ (p)$ positive eigenvalues (unstable) and $n - λ (p)$ negative ones (stable), and none zero. Hyperbolicity and the dimensions $dim E^{u} = λ$ , $dim E^{s} = n - λ$ follow. $■$

Partition of $M$ by cells. Each $x \in M$ lies on a unique maximal trajectory $t \mapsto φ_{t} (x)$ . Since $f$ is strictly decreasing along nonconstant trajectories and $M$ is compact with $f$ bounded, $f (φ_{t} (x))$ converges as $t \to \pm \infty$ ; the limit set is contained in $Crit (f)$ (a level where $\nabla f = 0$ ), and because critical points are isolated the trajectory converges to a single critical point in each direction. Hence $x$ lies in exactly one $W^{u} (p)$ (by its backward limit) and exactly one $W^{s} (q)$ (by its forward limit), giving the two partitions $M = ⨆_{p} W^{u} (p) = ⨆_{q} W^{s} (q)$ . $■$

Moduli dimension under Morse-Smale. With $W^{u} (p) ⋔ W^{s} (q)$ , the intersection $I = W^{u} (p) \cap W^{s} (q)$ is a submanifold of dimension $dim W^{u} (p) + dim W^{s} (q) - n = λ (p) + (n - λ (q)) - n = λ (p) - λ (q)$ . The negative gradient field is tangent to $I$ (both factors are flow-invariant) and nonvanishing on $I ∖ {p, q}$ , so it generates a free, proper $R$ -action there, and the quotient $M (p, q) = I / R$ is a manifold of dimension $λ (p) - λ (q) - 1$ for $p \neq = q$ . $■$

Genericity (Sard-Smale), proof outline. Fix $f$ . Over the separable Banach manifold $G$ of $C^{k}$ metrics, form the universal trajectory space $M (p, q; G) = {(γ, g) : γ \in M_{g} (p, q)}$ as the zero set of a section $F (γ, g) = \overset{γ}{˙} + \nabla_{g} f (γ)$ of a Banach bundle over (path space) $\times G$ . The vertical differential $D F_{(γ, g)}$ is surjective because varying $g$ alone already spans the cokernel of the $γ$ -linearisation (an explicit metric variation supported near a regular point of $γ$ moves $\nabla_{g} f$ in any prescribed direction transverse to $\overset{γ}{˙}$ ). So the universal moduli is a Banach manifold and the projection $π : M (p, q; G) \to G$ is a $C^{k}$ Fredholm map of index $λ (p) - λ (q)$ . The Sard-Smale theorem gives a residual set of regular values $g$ ; for such $g$ the fibre $M_{g} (p, q)$ is cut out transversally, i.e. $W_{g}^{u} (p) ⋔ W_{g}^{s} (q)$ . Intersecting the (countably many) residual sets over all pairs $(p, q)$ — a finite set, so a finite intersection — yields a residual set of Morse-Smale metrics. $■$

Connections Master

Morse functions and the index 03.02.30 are the direct upstream input: the index $λ (p)$ defined there as the number of negative Hessian eigenvalues is shown here to equal $dim W^{u} (p)$ , the dimension of the descending cell of the gradient flow. The Morse lemma's normal form is what makes the critical point hyperbolic for $- \nabla f$ , and the isolatedness of nondegenerate critical points is what makes every trajectory converge to a single critical point in each time direction; without those two facts the stable/unstable decomposition would not exist.

Handle attachment and CW type 03.02.31 is the topological counterpart of the dynamical decomposition. There, passing an index- $λ$ critical level attaches a $λ$ -handle to the sublevel set; here, the unstable cell $W^{u} (p)$ is the open $λ$ -cell of the resulting CW structure, and for a Morse-Smale pair the descending decomposition $M = ⨆_{p} W^{u} (p)$ realises that CW structure explicitly, with attaching maps read off from broken gradient trajectories. The handle picture proves the cell structure exists abstractly; the gradient flow exhibits it dynamically, and the two agree because both are governed by the same index.

Trajectory spaces and the Fredholm/transversality theorem 03.15.02 is the immediate successor: it promotes the intersection $W^{u} (p) \cap W^{s} (q)$ studied here into the zero set of a Fredholm section over a Banach path space, whose index is the dimension count $λ (p) - λ (q)$ established here, and recasts the Morse-Smale genericity proved here via Sard-Smale into the transversality theorem that makes the moduli $M (p, q)$ smooth manifolds of the predicted dimension. The finite count of isolated flow lines in the index-difference-one case becomes, in 03.15.06, the Morse differential, and the index-difference-two moduli force $\partial^{2} = 0$ .

Historical & philosophical context Master

René Thom announced in 1949 (Sur une partition en cellules associée à une fonction sur une variété, Comptes Rendus de l'Académie des Sciences 228, 973–975) that the descending manifolds of a gradient-like flow partition a manifold into cells, one per critical point of dimension equal to the index — the first appearance of the gradient-flow route to the cell structure that Morse's 1925 critical-point theory had obtained by sublevel-set arguments. The systematic study of the gradient flow as a dynamical system, and the transversality condition now bearing his name, is due to Stephen Smale, who introduced the condition $W^{u} (p) ⋔ W^{s} (q)$ in On gradient dynamical systems (Annals of Mathematics 74, 1961, 199–206) and used it in his proof of the generalised Poincaré conjecture in dimensions $\geq 5$ and the $h$ -cobordism theorem.

Milnor's Morse Theory (Annals of Mathematics Studies 51, 1963) carried the gradient flow only as far as the deformation-retraction arguments of Part I §3, deferring the trajectory-counting programme; the modern synthesis — gradient flow, stable/unstable manifolds, the Morse-Smale condition, and the trajectory spaces $M (p, q)$ as the generators of a chain complex — was assembled in the 1980s out of Witten's 1982 physics heuristic and Floer's infinite-dimensional constructions, and given its first fully rigorous finite-dimensional account in Matthias Schwarz's Morse Homology (Progress in Mathematics 111, Birkhäuser, 1993), based on his 1992 ETH Zürich dissertation under Zehnder and Salamon. The Sard-Smale theorem that secures genericity is Smale's An infinite-dimensional version of Sard's theorem (American Journal of Mathematics 87, 1965, 861–866).

Bibliography Master

@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 de 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}
}

@article{Smale1965,
  author  = {Smale, Stephen},
  title   = {An infinite-dimensional version of {S}ard's theorem},
  journal = {American Journal of Mathematics},
  volume  = {87},
  number  = {4},
  pages   = {861--866},
  year    = {1965}
}

@book{Milnor1963,
  author    = {Milnor, John},
  title     = {Morse Theory},
  series    = {Annals of Mathematics Studies},
  volume    = {51},
  publisher = {Princeton University Press},
  year      = {1963},
  note      = {Based on lecture notes by M. Spivak and R. Wells}
}

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

@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}
}

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

Prerequisites

03.02.30
03.02.05
02.12.01

Used in

03.15.02

Tier anchors

beginner: Banyaga-Hurtubise Lectures on Morse Homology Ch. 3 (informal); Milnor Morse Theory §3 height-function pictures
intermediate: Schwarz Morse Homology Part I §1-2; Banyaga-Hurtubise Lectures on Morse Homology Ch. 4; Audin-Damian Morse Theory and Floer Homology §3
master: Schwarz Morse Homology Part I §2-3; Smale 1961 (On gradient dynamical systems); Audin-Damian §3; Banyaga-Hurtubise Ch. 6

References

03-modern-geometry/Bott-LecturesOnMorseTheory.pdf · Gradient flow of a Morse function; descending and ascending cells
Schwarz — Morse Homology (Progress in Mathematics 111, Birkhäuser 1993) · Part I §1-3 (the negative gradient flow, W^s/W^u, the Morse-Smale condition, trajectory spaces)
Smale — On gradient dynamical systems (1961) · Annals of Mathematics 74, pp. 199--206 (the Morse-Smale transversality condition)
Thom — Sur une partition en cellules associée à une fonction sur une variété (1949) · C.R. Acad. Sci. Paris 228, pp. 973--975 (the cell decomposition by descending manifolds)
Milnor — Morse Theory (Annals of Mathematics Studies 51, 1963) · Part I §3 (the gradient-flow deformation of sublevel sets)

Estimated time

beginner: 17m
intermediate: 42m
master: 78m