Morse Theory

Morse Theory studies the topology of a smooth manifold by analyzing the critical points of a generic real-valued function on it. One-line statement: the number and indices of critical points of a Morse function

f : M \to R

determine the homotopy type of

M

via attachment of cells. Key result:

M

has the homotopy type of a CW complex with one

k

-cell for each critical point of index

k

; this yields the Morse inequalities relating critical points to Betti numbers. Where it applies: differential topology, Riemannian geometry, gauge theory, calculus of variations (geodesics), symplectic geometry (Floer theory), data analysis (persistent homology). Who/when: Marston Morse, 1925–1934; refined by Bott, Smale, Milnor, Witten.

Morse theory provides a dictionary between the analysis of a smooth function $f$ on a manifold $M$ and the topology of $M$ . A Morse function is a smooth function whose critical points are all nondegenerate (the Hessian is invertible). The central insight is that as one sweeps through sublevel sets $M^{a} = f^{- 1} (- \infty, a]$ and increases $a$ , the topology changes only when passing a critical value, and the change is precisely the attachment of a cell whose dimension equals the Morse index of the critical point. This converts global topological questions into local computations at finitely many points. A surprising fact: a manifold that admits a Morse function with exactly two critical points must be homeomorphic to a sphere (Reeb's theorem) — the entire global shape is forced by minimal critical data.

The Core Insight

Topology is built from critical points. Where the gradient of $f$ vanishes, the level sets of $f$ change their topological type. By cataloguing how they change — quantified by the index of the Hessian — you reconstruct the manifold one cell at a time. The flow of $- \nabla f$ organizes the manifold into a CW structure, and the algebra of its critical points (the Morse complex) computes homology.

Formal Statement

Critical Points and Nondegeneracy

Let $M$ be a smooth $n$ -dimensional manifold and $f : M \to R$ a smooth function. A point $p \in M$ is a critical point if $d f_{p} = 0$ . In local coordinates $(x^{1}, \dots, x^{n})$ centered at $p$ , this means

\frac{\partial f}{\partial x ^{i}} (p) = 0, i = 1, \dots, n .

⟳the differential

d f_{p}

acts on tangent vectors by directional derivatives; it vanishes iff all partials vanish.

At a critical point the Hessian is the well-defined symmetric bilinear form on $T_{p} M$ :

H_{p} f (v, w) = v (w (\tilde{f})), coordinates: (H_{p} f)_{ij} = \frac{\partial ^{2} f}{\partial x ^{i} \partial x ^{j}} (p) .

⟳the ambiguity from choice of vector field extensions

\tilde{f}

vanishes precisely because

d f_{p} = 0

, making

H_{p} f

coordinate-independent.

A critical point $p$ is nondegenerate if $H_{p} f$ is a nondegenerate bilinear form ( $det (H_{p} f)_{ij} \neq = 0$ ). The index $λ (p)$ is the number of negative eigenvalues of $H_{p} f$ , i.e. the dimension of a maximal subspace on which $H_{p} f$ is negative definite. A function all of whose critical points are nondegenerate is a Morse function.

Morse Lemma

Let $p$ be a nondegenerate critical point of $f$ with index $λ$ . Then there exist local coordinates $(y^{1}, \dots, y^{n})$ centered at $p$ in which

f = f (p) - (y^{1})^{2} - \dots - (y^{λ})^{2} + (y^{λ + 1})^{2} + \dots + (y^{n})^{2} .

⟳iteratively complete the square using a smooth version of the Taylor remainder (Hadamard's lemma), simultaneously diagonalizing the Hessian by Sylvester's law of inertia.

In particular nondegenerate critical points are isolated, and the index is a coordinate-free invariant.

Fundamental Theorems of Morse Theory

Let $f : M \to R$ be a smooth function on a compact manifold, and $M^{a} = f^{- 1} (- \infty, a]$ .

(A) If

f^{- 1} [a, b]

is compact and contains no critical points, then

M^{a}

is diffeomorphic to

M^{b}

, and

M^{a}

is a deformation retract of

M^{b}

. (B) If

f^{- 1} [a, b]

contains exactly one critical point

p

, of index

λ

, with

a < f (p) < b

, then

M^{b}

has the homotopy type of

M^{a}

with a

λ

-cell attached:

M^{b} ≃ M^{a} \cup_{φ} e^{λ} .

⟳use the gradient flow of

- \nabla f

for (A) to push

M^{b}

down to

M^{a}

; for (B), the Morse Lemma normal form lets you explicitly attach the cell of descending directions.

(Corollary) If

M

is compact and

f

is Morse, then

M

has the homotopy type of a CW complex with one cell of dimension

λ

for each critical point of index

λ

Why It Holds / Derivation

Sketch of the Fundamental Theorems

Step 1 — Riemannian setup. Fix a Riemannian metric

g

M

, defining the gradient

\nabla f

g (\nabla f, \cdot) = df

. Consider the negative gradient flow

\dot{ϕ}_{t} = - \nabla f (ϕ_{t})

. Along a flow line,

\frac{d}{d t} f (ϕ_{t} (x)) = df (\dot{ϕ}_{t}) = - g (\nabla f, \nabla f) = - ∥\nabla f ∥^{2} \leq 0.

⟳chain rule plus the defining identity of the gradient; equality to zero holds only at critical points.

Step 2 — No critical values (Theorem A). On

f^{- 1} [a, b]

with no critical points,

∥\nabla f ∥ > 0

. Rescale the flow by

1/∥\nabla f ∥^{2}

f

decreases at unit rate. The time-

(b - a)

flow gives a diffeomorphism

M^{b} \to M^{a}

and a deformation retraction onto

M^{a}

. Step 3 — Passing one critical point (Theorem B). Use Morse coordinates from equation [3]. Let

ε

be small. The set

M^{c + ε}

deformation retracts onto

M^{c - ε}

together with the descending disk — the cell spanned by the negative (

y^{1}, \dots, y^{λ}

) directions — attached along its boundary sphere

S^{λ - 1}

via the gradient flow. Explicitly the model is

f = c - \sum_{1}^{λ} (y^{i})^{2} + \sum_{λ + 1}^{n} (y^{j})^{2}

, and one constructs

F

a modified function whose sublevel set is

M^{c - ε}

plus a "handle," then retracts. Step 4 — Assembling the CW complex. Order critical values

c_{1} < c_{2} < \dots < c_{k}

. Inductively each

M^{c_{i} + ε}

is homotopy equivalent to the previous sublevel set with cells attached. Compactness gives finitely many critical points (each isolated by the Morse Lemma). The result is a CW complex with the stated cells.

■

Morse Inequalities

Let $M$ be compact and $f$ Morse with $C_{λ}$ critical points of index $λ$ . Let $b_{λ} = dim_{F} H_{λ} (M; F)$ be the Betti numbers over a field $F$ . Then the weak Morse inequalities:

C_{λ} \geq b_{λ} for all λ, λ \sum (- 1)^{λ} C_{λ} = λ \sum (- 1)^{λ} b_{λ} = χ (M) .

⟳cellular homology of the Morse CW complex has chain groups of rank

C_{λ}

; ranks of chain groups dominate homology ranks, and Euler characteristic is a chain-level invariant.

The strong Morse inequalities sharpen this to, for every $k$ :

λ = 0 \sum k (- 1)^{k - λ} C_{λ} \geq λ = 0 \sum k (- 1)^{k - λ} b_{λ} .

⟳encode via the Morse polynomial

M (t) = \sum C_{λ} t^{λ}

and Poincaré polynomial

P (t) = \sum b_{λ} t^{λ}

; then

M (t) - P (t) = (1 + t) Q (t)

with

Q

having nonnegative coefficients.

Edward Witten | b. 1951

Although Marston Morse founded the subject, Witten's 1982 paper "Supersymmetry and Morse Theory" reinvented it. Witten deformed the de Rham differential by $d_{t} = e^{- t f} d e^{t f}$ and showed that as $t \to \infty$ the harmonic forms (and hence cohomology) localize at critical points of $f$ . This gave an analytic proof of the Morse inequalities, interpreted the Morse complex as instanton-counting tunnelling amplitudes between critical points in supersymmetric quantum mechanics, and seeded Floer theory. His physical reasoning — that perturbative ground states sit at minima while instantons (gradient flow lines) connect them — became the template for infinite-dimensional Morse theory in gauge theory and symplectic topology.

Building Intuition

Imagine a mountainous landscape $M$ flooded by rising water, where the height is $f$ . The sublevel set $M^{a}$ is the submerged region at water level $a$ . As the water rises:

Crossing a minimum (index 0): a new isolated puddle appears — you attach a $0$ -cell (a point).
Crossing a saddle (index 1): two bodies of water merge, or a handle forms — you attach a $1$ -cell (an arc connecting components).
Crossing a maximum (index 2 on a surface): a remaining lake fills a basin and disappears as an island gets submerged — you attach a $2$ -cell (a disk capping off a hole).

The index counts the number of independent downhill directions at a critical point. A minimum has none (all uphill, index 0), a maximum on an $n$ -manifold has all $n$ . The genius is that the global shape of the terrain is encoded entirely by how these merging/capping events happen, in what order, attached where.

The height function on a vertically-standing torus, showing four critical points and the cells they attach.

<svg viewBox="0 0 460 420" xmlns="http://www.w3.org/2000/svg">
<rect width="460" height="420" fill="#0a0e17"/>
<text x="230" y="28" fill="#e2e8f0" font-size="16" text-anchor="middle">Height function f on a standing torus</text>
<!-- axis -->
<line x1="40" y1="380" x2="40" y2="60" stroke="#94a3b8" stroke-width="1.5"/>
<polygon points="40,52 36,64 44,64" fill="#94a3b8"/>
<text x="22" y="220" fill="#e2e8f0" font-size="13" transform="rotate(-90 22 220)">f (height)</text>
<!-- torus outline -->
<ellipse cx="250" cy="220" rx="120" ry="150" fill="none" stroke="#94a3b8" stroke-width="2"/>
<ellipse cx="250" cy="220" rx="55" ry="80" fill="none" stroke="#94a3b8" stroke-width="2"/>
<!-- critical-point key curve highlight: vertical of critical heights -->
<path d="M250 70 L250 150 M250 290 L250 370" stroke="#d4a853" stroke-width="2.5" fill="none"/>
<!-- critical points -->
<circle cx="250" cy="70" r="5" fill="#d4a853"/>
<text x="270" y="74" fill="#e2e8f0" font-size="13">max, index 2</text>
<circle cx="250" cy="150" r="5" fill="#d4a853"/>
<text x="270" y="154" fill="#e2e8f0" font-size="13">saddle, index 1</text>
<circle cx="250" cy="290" r="5" fill="#d4a853"/>
<text x="270" y="294" fill="#e2e8f0" font-size="13">saddle, index 1</text>
<circle cx="250" cy="370" r="5" fill="#d4a853"/>
<text x="270" y="374" fill="#e2e8f0" font-size="13">min, index 0</text>
<!-- water level -->
<line x1="100" y1="320" x2="400" y2="320" stroke="#475569" stroke-width="1" stroke-dasharray="4 3"/>
<text x="105" y="335" fill="#94a3b8" font-size="11">sublevel set M^a</text>
<!-- cell counts -->
<text x="60" y="400" fill="#e2e8f0" font-size="12">C0=1, C1=2, C2=1  ⇒  χ = 1−2+1 = 0</text>
</svg>

Worked Examples

Morse Function on the 2-Torus

Given: the standing torus

T^{2} \subset R^{3}

with

f

= height (the

z

-coordinate). Find its critical points, indices, and verify the Morse inequalities over

Q

. Steps:

Critical points occur where the tangent plane is horizontal: the bottom minimum, the inner-bottom saddle, the inner-top saddle, the top maximum.
Compute indices via the Hessian:

- bottom min:

H = diag (+, +)

, index

λ = 0

. - lower saddle:

H = diag (+, -)

, index

λ = 1

. - upper saddle:

H = diag (+, -)

, index

λ = 1

. - top max:

H = diag (-, -)

, index

λ = 2

Critical counts: $C_{0} = 1, C_{1} = 2, C_{2} = 1$ .
Known Betti numbers of $T^{2}$ : $b_{0} = 1, b_{1} = 2, b_{2} = 1$ .
Weak inequalities: $C_{λ} \geq b_{λ}$ holds with equality everywhere — this is a perfect Morse function.
Euler characteristic: $\sum (- 1)^{λ} C_{λ} = 1 - 2 + 1 = 0 = χ (T^{2})$ . ✓

Answer:

C_{0} = 1, C_{1} = 2, C_{2} = 1; χ (T^{2}) = 0; Morse equalities are sharp.

The Standard Height on the n-Sphere

Given:

S^{n} = {x \in R^{n + 1} : ∣ x ∣ = 1}

with

f (x) = x_{n + 1}

(height). Determine the CW structure and homology. Steps:

Critical points: north pole $N = (0, \dots, 0, 1)$ and south pole $S = (0, \dots, 0, - 1)$ . Only two, both nondegenerate.
At $S$ (the minimum) all directions go up: index $λ = 0$ .
At $N$ (the maximum) all directions go down: index $λ = n$ .
Counts: $C_{0} = 1$ , $C_{n} = 1$ , all others zero. By the CW corollary $S^{n} ≃ e^{0} \cup e^{n}$ — a point with an $n$ -cell attached, the standard CW structure of the sphere.
Homology: chain complex has $Z$ in degrees $0$ and $n$ only, with zero differentials, giving $H_{0} = H_{n} = Z$ , all else $0$ . ✓
Reeb's theorem check: a compact manifold admitting a Morse function with exactly two critical points is homeomorphic to $S^{n}$ .

Answer:

C_{0} = C_{n} = 1; H_{k} (S^{n}) = Z for k \in {0, n}, 0 otherwise.

Common Misconceptions

Morse functions are not unique, but they are generic. Any smooth function can be perturbed (in the

C^{\infty}

topology) to a Morse function — Morse functions form an open dense set. Do not confuse a specific Morse function (which can have unnecessarily many critical points, e.g. a wiggly height function) with the minimal one (a perfect Morse function realizing the Betti numbers). Not every manifold admits a perfect Morse function:

RP^{3}

over

Z

has 2-torsion forcing extra critical points beyond the

Q

-Betti numbers — the strong Morse inequalities with field coefficients matter. The index is the number of negative eigenvalues, not positive. A common sign error flips minima and maxima. A function with a degenerate critical point fails the theory. Consider

f : R \to R

f (x) = x^{3}

. Then

f^{'} (0) = 0

0

is critical, but

f^{''} (0) = 0

— the Hessian is degenerate. The Morse Lemma fails:

f

has no quadratic normal form at

0

, and

0

is not even a local extremum. Such "monkey saddle"-type degeneracies (

f (x, y) = x^{3} - 3 x y^{2}

in 2D) are exactly what Morse theory excludes, and they can be removed by an arbitrarily small perturbation (e.g.

x^{3} + ε x

has either zero or two nondegenerate critical points). This illustrates why nondegeneracy is essential and why genericity arguments are needed.

Connections

Morse theory sits at the crossroads of analysis, topology, and geometry:

Homology Theory and CW Complexes: the Morse complex $(C_{*}, \partial)$ , where $C_{λ} = Z^{C_{λ}}$ and $\partial$ counts gradient flow lines between critical points of adjacent index, computes $H_{*} (M)$ directly — a finite-dimensional, geometrically transparent chain complex.
Riemannian Geometry: the gradient flow requires a metric, and Morse theory of the energy functional on the loop space $Ω M$ recovers geodesics as critical points, proving the existence of closed geodesics and underpinning Bott's periodicity theorem.
Symplectic Geometry and Floer Theory: infinite-dimensional Morse theory of the symplectic action functional yields Floer homology, resolving the Arnold conjecture on fixed points of Hamiltonian symplectomorphisms.
Handle Decompositions and Cobordism Theory: each index- $λ$ critical point corresponds to attaching a $λ$ -handle; Smale's handle-cancellation techniques powered the proof of the high-dimensional Poincaré conjecture.
de Rham Cohomology: Witten's deformation links the analytic spectrum of a deformed Laplacian to the Morse complex.

Real-World Applications

High-dimensional Poincaré conjecture (Smale, 1961): the $h$ -cobordism theorem, proved using Morse-theoretic handle cancellation, established that homotopy $n$ -spheres ( $n \geq 5$ ) are homeomorphic to $S^{n}$ .
Existence of closed geodesics: Morse theory on the loop space (Lyusternik–Fet) guarantees every compact Riemannian manifold has at least one closed geodesic; the energy functional's critical points are geodesics.
Bott periodicity: Bott computed the homotopy of classical Lie groups via Morse theory of energy on path spaces, foundational for topological K-theory.
Persistent homology / Topological Data Analysis: discrete Morse theory (Forman) on filtered simplicial complexes underlies persistence diagrams used in shape recognition, neuroscience connectomics, and materials science.
Computer graphics & mesh processing: Reeb graphs and Morse–Smale complexes segment 3D surfaces, extract ridges/valleys, and enable shape matching and feature extraction.
Scientific visualization: Morse–Smale complexes identify critical features (maxima, saddles, minima) in scalar fields from combustion simulations, cosmology (cosmic web structure), and medical imaging.
Gauge theory & 4-manifold topology: Floer homology of flat connections (instanton Floer homology) distinguishes smooth structures and feeds into Donaldson and Seiberg–Witten theory.

Historical Development

1925–1934 — Marston Morse develops the "calculus of variations in the large," proving the Morse inequalities and applying them to geodesics; his treatise The Calculus of Variations in the Large (1934) founds the field.
1929 — Morse establishes the relation between critical points of the energy functional on path spaces and geodesics, the infinite-dimensional template.
1949 — Lyusternik & Fet prove existence of a closed geodesic on any compact Riemannian manifold using Morse-theoretic methods on the free loop space.
1956–1959 — Raoul Bott proves the periodicity theorem for the stable homotopy of classical groups via Morse theory of energy functionals; introduces Morse–Bott theory for nondegenerate critical manifolds.
1960–1962 — Stephen Smale uses Morse theory and handle decompositions to prove the $h$ -cobordism theorem and the generalized Poincaré conjecture in dimensions $\geq 5$ .
1963 — John Milnor publishes Morse Theory, the canonical exposition, streamlining proofs of the fundamental theorems and the $h$ -cobordism setup.
1982 — Edward Witten gives an analytic proof of the Morse inequalities via supersymmetric deformation of the de Rham complex, linking it to physics and instantons.
1988–1989 — Andreas Floer constructs Floer homology for Hamiltonian systems and 3-manifolds, infinite-dimensional Morse theory resolving cases of the Arnold conjecture.
1998 — Robin Forman develops discrete Morse theory on CW/simplicial complexes, making the theory algorithmically usable for data and computation.

Subtleties & Limits

Morse–Smale transversality and the Morse complex. To define the boundary operator

\partial

counting flow lines, one needs the descending and ascending manifolds of critical points to intersect transversally (the Morse–Smale condition). This holds for generic metrics but requires careful proof; the moduli spaces of flow lines must be compactified (broken trajectories), and orientations must be coherently assigned to fix signs in

\partial

. Proving

\partial^{2} = 0

amounts to a gluing/compactness argument over 1-dimensional moduli of flow lines. Open and active directions:

Min-max and Lyusternik–Schnirelmann category: for degenerate settings one needs cup-length / category bounds; sharp results on the minimal number of critical points (the Arnold conjecture for non-degenerate vs. degenerate Hamiltonians) remain subtle.
Infinite-dimensional Morse theory (Floer-type) requires the Palais–Smale condition and Fredholm analysis; constructing well-defined theories in gauge-theoretic settings (e.g., reducibles, bubbling) is delicate and partly open.
Discrete vs. smooth correspondence: optimal (minimal critical-cell) discrete Morse functions are NP-hard to compute in general; bridging algorithmic efficiency and topological optimality is ongoing.

Morse Theory

The Core Insight

Formal Statement

Critical Points and Nondegeneracy

Morse Lemma

Fundamental Theorems of Morse Theory

Why It Holds / Derivation

Sketch of the Fundamental Theorems

Morse Inequalities

Edward Witten | b. 1951

Building Intuition

The height function on a vertically-standing torus, showing four critical points and the cells they attach.

Worked Examples

Morse Function on the 2-Torus

The Standard Height on the n-Sphere

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also

Before this

Where this leads