03.02.40 · differential-geometry / manifolds

The path space as a CW complex: the fundamental theorem of Morse theory

shipped3 tiersLean: none

Anchor (Master): Milnor Morse Theory Part III §§16--17 and Appendix; Bott 1956 Comm. Pure Appl. Math. 9; Bott 1959 Ann. Math. 70 (the periodicity theorem); Klingenberg Lectures on Closed Geodesics Ch. 2

Intuition Beginner

Fix two cities on a globe and think about every possible route between them — not just the shortest one, but all the wiggly paths, the long detours, the ones that loop over a pole. This whole collection of routes is a space in its own right. A single point of this new space is one entire journey from start to finish. Moving around in it means deforming one route into a nearby route.

To each route attach a number that measures how stretched it is: short, taut routes get a small number, long sprawling ones get a big number. The straightest-possible routes — the geodesics — are the special routes where this number stops changing under small wiggles. They are the resting points of the landscape of all routes.

Morse theory studies a landscape by its peaks, passes, and pits. Here the landscape is the space of all routes, and its special resting points are the geodesics. The remarkable payoff is that this enormous, infinite space of routes can be rebuilt out of a small number of simple building blocks — one block for each geodesic. A geodesic that is genuinely shortest contributes the simplest block; a geodesic that wanders past a focusing point contributes a fatter block.

So a question about every route at once turns into a question about a handful of straightest routes. That trade is the heart of this unit, and it is the engine behind one of the deepest patterns in topology.

Visual Beginner

Alt text: The left panel shows a sphere with two marked points and several curving paths between them; the shortest meridian geodesic is bold, and a couple of longer geodesics that wind further around are drawn lighter. A bracket gathers all these paths into one box labelled "space of all routes." An arrow points to the right panel, where the same routes are redrawn as a tidy tower of building blocks: a small block at the bottom for the shortest geodesic, then taller blocks stacked above for each longer geodesic. The picture shows that the unmanageable space of every route is assembled from one simple block per straightest route, with block height growing as the geodesic winds more.

Worked example Beginner

Take the round sphere and pick the north pole as the start and the south pole as the finish. What are the straightest routes between them?

Every line of longitude is one. There are infinitely many longitudes, but they all have the same length and the same shape, so as a family they form one circle's worth of shortest geodesics. After that come the geodesics that run from pole to pole, overshoot, and wrap around once more before stopping — these are longer, and they have passed a focusing point along the way. Then there are ones that wrap around twice, longer still, and so on.

The building-block recipe says: the shortest family gives the low blocks, and each extra wrap-around adds a taller block, because each extra wrap-around means the route has slid past one more focusing point. The height of a block is exactly the count of focusing points the geodesic has passed.

What this tells us. Even on the simplest curved surface, the space of all routes between two poles is layered: a bottom layer of genuine shortest paths, then ever-taller layers for the routes that wind further. Counting focusing points — which the previous units taught us to do — assigns each straightest route to its correct layer. The infinite jumble of routes is organised into clean shelves.

Check your understanding Beginner

Formal definition Intermediate+

Throughout, $(M, g)$ is a complete Riemannian manifold, $p, q \in M$ , and $Ω = Ω (M; p, q)$ is the path space of piecewise-smooth curves $ω : [0, 1] \to M$ with $ω (0) = p$ , $ω (1) = q$ , topologised by the metric $d (ω, ω^{'}) = max_{t} ρ (ω (t), ω^{'} (t)) + (\int_{0}^{1} ∣ \overset{ω}{˙} - \overset{ω}{˙}^{'} ∣^{2} d t)^{1/2}$ of 03.02.19. The energy is $E (ω) = \frac{1}{2} \int_{0}^{1} ∣ \overset{ω}{˙} (t) ∣^{2} d t$ , and for $c \geq 0$ the sublevel set is $$ \Omega^c = E^{-1}[0, c] = {, \omega \in \Omega : E(\omega) \le c ,}. $$ The critical points of $E$ are exactly the geodesics from $p$ to $q$ 03.02.19, and by Schwarz a geodesic of length $ℓ$ has energy $\frac{1}{2} ℓ^{2}$ , so $Ω^{c}$ contains exactly the geodesics of length $\leq 2 c$ .

Definition (broken-geodesic approximation). Fix a subdivision $0 = t_{0} < t_{1} < \dots < t_{k} = 1$ . Let $Ω (t_{0}, \dots, t_{k}) \subset Ω^{c}$ be the subspace of paths $ω$ that are a (minimal) geodesic on each $[t_{i - 1}, t_{i}]$ , with the subdivision chosen fine enough that each such restriction is the unique minimal geodesic between its endpoints. As established in 03.02.19, this is a finite-dimensional smooth manifold, coordinatised by the breakpoints $(ω (t_{1}), \dots, ω (t_{k - 1}))$ , and $E$ restricted to it is a smooth function whose critical points are precisely the smooth geodesics in $Ω^{c}$ , with the same index and nullity (in the sense of $E_{**}$ ) as on the full $Ω$ . Write $B^{c} = Ω (t_{0}, \dots, t_{k}) \cap Ω^{c}$ for the corresponding sublevel set of this finite-dimensional model.

Definition (non-conjugate endpoints). The pair $(p, q)$ is non-conjugate if $q$ is not conjugate to $p$ along any geodesic from $p$ to $q$ . By Sard's theorem applied to the exponential map, the set of $q$ conjugate to a fixed $p$ along some geodesic has measure zero, so non-conjugate pairs are generic. Under this hypothesis every geodesic from $p$ to $q$ is a nondegenerate critical point of $E$ : its nullity is $μ (1) = 0$ by the Morse Index Theorem 03.02.19, and its index $λ$ is finite and equals the number of interior conjugate points counted with multiplicity.

Definition (CW homotopy type). A topological space $X$ has the homotopy type of a CW complex $K$ if there is a homotopy equivalence $X ≃ K$ . The cells of $K$ are the building blocks of 03.12.10; a $λ$ -cell is attached along a map $S^{λ - 1} \to K^{(λ - 1)}$ of its boundary into the lower skeleton. The fundamental theorem below produces such a $K$ for $X = Ω (M; p, q)$ , with one $λ$ -cell per index- $λ$ geodesic.

Counterexamples to common slips

"One cell per geodesic, always." The clean one-cell-per-geodesic statement needs non-conjugate endpoints, so that each geodesic is a nondegenerate critical point of $E$ . If $q$ is conjugate to $p$ — e.g. $p, q$ antipodal on $S^{n}$ , where a whole sphere's worth of geodesics has the same length — the critical points form positive-dimensional manifolds and one instead attaches a cell-bundle's worth, handled by Morse-Bott theory rather than the bare statement here.
Sublevel set versus full space. $Ω^{c}$ is not itself a finite CW complex unless only finitely many geodesics have energy $\leq c$ ; that finiteness is what the broken-geodesic reduction supplies. The full $Ω$ is then the increasing union of the $Ω^{c}$ , and its homotopy type is recovered only after the monotone-union (direct-limit) passage of the Appendix.
Energy versus length. One must vary the energy $E$ , not the length $L$ , to get isolated critical points: $L$ is invariant under reparametrisation, so its critical set is never isolated. $E$ breaks the reparametrisation symmetry, and a geodesic (constant-speed) is the energy-critical representative of its length-critical orbit.

Key theorem with proof Intermediate+

Theorem (Fundamental Theorem of Morse Theory). Let $M$ be a complete Riemannian manifold and $p, q \in M$ a non-conjugate pair, with $Ω = Ω (M; p, q)$ . Then $Ω$ has the homotopy type of a countable CW complex containing one cell of dimension $λ$ for each geodesic from $p$ to $q$ of index $λ$ . Moreover each $Ω^{c}$ has the homotopy type of a finite CW complex with one $λ$ -cell per index- $λ$ geodesic of energy $\leq c$ , provided $c$ is not the energy of any geodesic.

Proof. Fix $c$ not a critical value of $E$ .

Step 1: pass to the finite-dimensional model. Choose a subdivision $0 = t_{0} < \dots < t_{k} = 1$ fine enough (Lebesgue number of an open cover by uniquely-geodesic balls, using $E \leq c$ to bound lengths) that $Ω (t_{0}, \dots, t_{k})$ contains every geodesic of energy $\leq c$ and approximates $Ω^{c}$ . Milnor's finite-dimensional approximation ^{[Milnor Part III §16]} shows the inclusion $B^{c} = Ω (t_{0}, \dots, t_{k}) \cap Ω^{c} ↪ Ω^{c}$ is a homotopy equivalence: the map sending a path to its broken-geodesic replacement (geodesic interpolation between the values at the $t_{i}$ ) is a deformation retraction of $Ω^{c}$ onto $B^{c}$ . So it suffices to give $B^{c}$ , a sublevel set of the smooth function $E ∣_{Ω (t_{0}, \dots, t_{k})}$ on a finite-dimensional manifold, its CW structure.

Step 2: $E$ is a Morse function on the model. On the finite-dimensional manifold $Ω (t_{0}, \dots, t_{k})$ the critical points of $E$ are the smooth geodesics, each nondegenerate because $(p, q)$ is non-conjugate: the Hessian of $E$ there is the index form $E_{**}$ restricted to the broken-geodesic tangent space, whose nullity is $μ (1) = 0$ 03.02.19. Its index equals the geodesic's index $λ = \sum_{0 < s < 1} μ (s)$ , the interior conjugate-point count, by the Morse Index Theorem 03.02.19. Thus $E ∣_{Ω (t_{0}, \dots, t_{k})}$ is a Morse function whose index- $λ$ critical points are exactly the index- $λ$ geodesics.

Step 3: assemble the cells by handle attachment. Apply the handle-attachment machinery of 03.02.31 to the Morse function $E$ on the finite-dimensional model. Crossing each critical value $E (γ) = \frac{1}{2} ℓ_{γ}^{2}$ of an index- $λ$ geodesic $γ$ , Milnor's Theorem 3.2 03.02.31 gives a homotopy equivalence $$ B^{c'} \simeq B^{c''} \cup_{\varphi_\gamma} e^\lambda $$ across the value, where $c^{''} < E (γ) < c^{'}$ are regular and $φ_{γ} : S^{λ - 1} \to B^{c^{''}}$ is the descending attaching sphere. Between critical values, Theorem 3.1 03.02.31 gives a deformation retraction, so the homotopy type is locally constant there. Iterating across the finitely many geodesics of energy $\leq c$ exhibits $B^{c}$ , and hence $Ω^{c}$ , as a finite CW complex with one $λ$ -cell per index- $λ$ geodesic of energy $\leq c$ .

Step 4: pass to the limit. Let $c \to \infty$ through a sequence $c_{1} < c_{2} < \dots$ of regular values with $c_{m} \to \infty$ (possible since geodesic energies are discrete and unbounded). Each inclusion $Ω^{c_{m}} ↪ Ω^{c_{m + 1}}$ is, by Steps 1--3, the inclusion of a subcomplex up to homotopy, adding the cells of the geodesics with energy in $(c_{m}, c_{m + 1}]$ . The Appendix lemma on the homotopy type of a monotone union (stated and proved below) identifies the homotopy type of the increasing union $Ω = ⋃_{m} Ω^{c_{m}}$ with the CW complex obtained as the direct limit of the $Ω^{c_{m}}$ . This is the countable CW complex with one $λ$ -cell per index- $λ$ geodesic, completing the proof. $■$

The theorem is the exact variational analogue of the CW-type corollary of 03.02.31: there a Morse function on a finite-dimensional closed manifold built it from handles; here the energy functional on the infinite-dimensional path space does the same, with the index theorem 03.02.19 supplying the index of each cell. The broken-geodesic model is the device that makes the finite-dimensional handle theory applicable, and the monotone-union lemma is the device that lifts the answer back to the infinite-dimensional space.

Bridge. This builds toward Bott periodicity: when $M$ is a symmetric space whose minimal geodesics from $p$ to $q$ form a manifold $Ω_{d}$ of minima and whose next geodesics all have index $\geq λ_{0}$ , the theorem says $Ω (M; p, q)$ is, up to dimension $λ_{0} - 1$ , homotopy equivalent to $Ω_{d}$ ; iterating this across $U (n) \supset \dots$ is exactly how Bott computed the stable homotopy of the classical groups, and this is the foundational reason periodicity has a geometric proof at all. The cell-per-geodesic dictionary appears again in 03.08.07, where the minimal-geodesic manifold on $SU (2 n)$ is a Grassmannian and the index gap drives the periodicity isomorphism. Putting these together, the variational Morse theory of 03.02.19 and the handle theory of 03.02.31 generalise from a single manifold to its entire loop space, and the central insight is that the index theorem turns "geodesic of index $λ$ " into " $λ$ -cell," so loop-space topology becomes a conjugate-point count — this is exactly the mechanism that the periodicity calculation exploits.

Exercises Intermediate+

Exercise (medium, short-answer). Explain why the broken-geodesic space $\Omega(t_0,\dots,t_k)$ is finite-dimensional while the full path space $\Omega$ is infinite-dimensional, and why the inclusion is nonetheless a homotopy equivalence on each sublevel set.

Hint

A broken geodesic is determined by its breakpoints; a general path is determined by its value at every $t$ .

Answer

Rubric. A broken geodesic in $Ω (t_{0}, \dots, t_{k})$ is determined by the finitely many breakpoint values $ω (t_{1}), \dots, ω (t_{k - 1}) \in M$ , giving dimension $(k - 1) dim M$ ; a general path needs a value at every parameter, an infinite-dimensional datum. Full credit notes that geodesic interpolation between breakpoints defines a deformation retraction of $Ω^{c}$ onto $Ω (t_{0}, \dots, t_{k}) \cap Ω^{c}$ (replacing each segment by its unique minimal geodesic strictly decreases or preserves energy and is canonical), so the inclusion is a homotopy equivalence on sublevel sets.

Exercise (medium, numeric). Let $p, q$ be non-antipodal points on the round $S^2$. The geodesics between them are the short great-circle arc and its complements wrapping around. The short arc has index $0$; the first wrap-around has crossed one conjugate point of multiplicity $1$. What is the index of the first wrap-around geodesic, and what cell does it give?

Hint

On $S^{2}$ the conjugate-point multiplicity is $n - 1 = 1$ . Index equals the conjugate-point count with multiplicity, from 03.02.19.

Answer

Index $1$ ; a $1$ -cell. The first wrap-around passes one interior conjugate point of multiplicity $1$ , so its index is $1$ and it contributes a $1$ -cell. Successive wrap-arounds add one to the index each time, giving cells of dimensions $0, 1, 2, \dots$ — recovering that $Ω (S^{2})$ has the rational homotopy type with cells in every dimension.

Exercise (hard, short-answer). State the monotone-union (Appendix) lemma and explain precisely where it is needed in the proof of the fundamental theorem.

Hint

Each $Ω^{c_{m}}$ is a finite CW complex; the full $Ω$ is their increasing union.

Answer

Rubric. Lemma: if $K_{1} \subseteq K_{2} \subseteq \dots$ is a sequence of spaces with each inclusion a homotopy equivalence onto a subcomplex (cofibration) of a CW structure, then the union $⋃_{m} K_{m}$ with the weak topology has the homotopy type of the direct-limit CW complex; a map into it is a homotopy equivalence iff its restriction to each $K_{m}$ is. It is needed at Step 4: Steps 1--3 only give each sublevel set $Ω^{c_{m}}$ as a finite CW complex, and the lemma is what upgrades the increasing union $Ω = ⋃_{m} Ω^{c_{m}}$ to a single (countable) CW complex with the assembled cells. Full credit identifies the lemma as the direct-limit/colimit passage and pins it to the infinite-dimensional Step 4.

Exercise (hard, short-answer). Sketch how the fundamental theorem yields a lower bound on the number of geodesics from $p$ to $q$ in terms of the Betti numbers of $\Omega(M;p,q)$.

Hint

This is the Morse-inequality argument of 03.02.31 transported to the path space.

Answer

Rubric. The CW model has $c_{λ}$ = number of index- $λ$ geodesics as its count of $λ$ -cells, and cellular homology 03.12.13 of that complex computes $H_{*} (Ω (M; p, q))$ . The weak Morse inequalities 03.02.31 give $c_{λ} \geq b_{λ} = dim H_{λ} (Ω (M; p, q))$ , so the number of index- $λ$ geodesics is at least the $λ$ -th Betti number of the path space. When $Ω (M; p, q)$ has nonzero homology in infinitely many degrees (as on a sphere), this forces infinitely many geodesics. Full credit names the cell-count = geodesic-count identity and applies the weak inequality.

Lean formalization Intermediate+

Mathlib has CWComplex, singular and cellular homology, the Levi-Civita connection, geodesics, and the exponential map, but no path-space variational layer and no Morse-theoretic bridge from a functional to a cell structure, so the fundamental theorem is not yet formalisable end-to-end. The statement-level shape one would target is below, written as Lean-compatible pseudo-Lean; it does not compile against current Mathlib (lean_status: none).

-- Statement target (NOT compiling against current Mathlib):
-- The path space Ω(M; p q) and the energy functional.
variable {M : Type*} [RiemannianManifold M] (p q : M)

def PathSpace : Type _ := { ω : C(unitInterval, M) // ω 0 = p ∧ ω 1 = q }

noncomputable def energy (ω : PathSpace p q) : ℝ :=
  (1/2) * ∫ t in Set.Icc 0 1, ‖tangent ω t‖ ^ 2

-- (p, q) non-conjugate: no geodesic has q conjugate to p.
def NonConjugatePair : Prop := ∀ γ : Geodesic p q, ¬ IsConjugate γ 1

-- Fundamental theorem (target): for a non-conjugate pair, the path space
-- is homotopy equivalent to a CW complex with one λ-cell per index-λ geodesic.
-- theorem fundamental_morse (h : NonConjugatePair p q) :
--   ∃ (K : CWComplex) (cells : Geodesic p q → K.Cell),
--     (∀ γ, (cells γ).dim = morseIndex γ) ∧
--     Nonempty (HomotopyEquiv (PathSpace p q) K)

the Mathlib gap analysis above enumerates the missing primitives: the topologised path space and energy functional, the finite-dimensional broken-geodesic model with its deformation-retraction theorem, the monotone-union (direct-limit) lemma for CW pairs, and the cell-per-geodesic identity itself.

Advanced results Master

The minimal-geodesic theorem and the index gap. The sharpest form of the fundamental theorem, the one Bott used, isolates a range of dimensions. Suppose $Ω_{d} \subset Ω (M; p, q)$ is the set of minimal geodesics and it is a smooth compact manifold, and suppose every non-minimal geodesic has index $\geq λ_{0}$ . Then the inclusion $Ω_{d} ↪ Ω (M; p, q)$ induces isomorphisms on homotopy groups $π_{i}$ for $i < λ_{0} - 1$ and a surjection at $i = λ_{0} - 1$ ^{[Milnor Part III §17]}. The reason is precisely the cell dictionary: the first non-minimal cell has dimension $\geq λ_{0}$ , so up to dimension $λ_{0} - 1$ the CW model is just $Ω_{d}$ . The number $λ_{0}$ is the index gap, and computing it is a conjugate-point count via the index theorem 03.02.19.

Bott periodicity from the loop space. On the special unitary group, the minimal geodesics from $I$ to $- I$ form a Grassmannian, and the index gap is large; the minimal-geodesic theorem then gives a homotopy equivalence, in a range growing with $n$ , between $Ω SU (2 n)$ and a Grassmannian, and iterating this across $U, O, Sp$ produces the eightfold and twofold periodicities $π_{i} (O) ≅ π_{i + 8} (O)$ , $π_{i} (U) ≅ π_{i + 2} (U)$ ^{[Bott 1959]}. This is the original 1959 proof of the periodicity theorem, and it is entirely Morse-theoretic: the loop space of a Lie group is studied through the energy functional, the geodesics are computed explicitly via the group exponential, and their indices via conjugate points. The whole of topological $K$ -theory 03.08.07 rests on this geometric foundation.

Closed geodesics and the free loop space. Replacing fixed endpoints $p, q$ by the condition $ω (0) = ω (1)$ (and quotienting by the circle action that rotates the parameter) turns $Ω$ into the free loop space $Λ M$ , whose critical points of $E$ are the closed geodesics. Bott's 1956 iteration formula ^{[Bott 1956]} computes how the index of an $m$ -fold iterate $γ^{m}$ grows with $m$ via the Sturm intersection number, and feeding this into the fundamental theorem for $Λ M$ is the route to existence results for infinitely many closed geodesics (Gromoll–Meyer, Vigué-Poirrier–Sullivan) when the loop-space homology is unbounded. The same cell-per-geodesic mechanism, applied to a different boundary condition, drives the entire closed-geodesic problem.

Palais–Smale and the Hilbert-manifold completion. Completing $Ω$ to the $H^{1}$ Hilbert manifold makes $E$ a $C^{\infty}$ function satisfying condition (C) of Palais and Smale, so the abstract infinite-dimensional Morse theory of Palais applies directly, bypassing the broken-geodesic model: the negative gradient flow of $E$ on the Hilbert manifold gives the handle attachments intrinsically. The broken-geodesic approach of Milnor and the Hilbert-manifold approach of Palais yield the same CW homotopy type; the former is more elementary, the latter generalises to other variational problems.

Synthesis. The fundamental theorem of Morse theory is the keystone that ties at least five threads into one structure. It is the path-space generalisation of the handle/CW theorem of 03.02.31: the energy functional plays the role of the Morse function, the broken-geodesic model the role of the manifold, and handle attachment proceeds identically — this is exactly the same machinery lifted from a finite-dimensional manifold to its loop space. It consumes the Morse Index Theorem of 03.02.19 as the computation of each cell's dimension, so the cell structure is read off conjugate points, and this is the foundational reason the loop space "knows" the curvature. It is dual to the homology computation of 03.12.13, in that the cell counts feed the cellular chain complex whose homology is that of $Ω$ , and the weak Morse inequalities then bound geodesic counts below by Betti numbers. It generalises the closed-manifold CW corollary to the infinite-dimensional setting via the monotone-union lemma, the central insight being that an increasing union of finite CW pairs is a CW complex. And putting these together, it is the bridge by which Bott periodicity 03.08.07 becomes a theorem of differential geometry: the stable homotopy of the classical groups is computed by counting conjugate points along geodesics in a symmetric space, which is the deepest single application of the entire Morse-theoretic apparatus assembled across 03.02.19, 03.02.31, and this unit.

Full proof set Master

Lemma (homotopy type of a monotone union — Milnor Appendix). Let $X$ be a space with the weak topology of an increasing sequence of subspaces $X_{1} \subseteq X_{2} \subseteq \dots$ , $X = ⋃_{m} X_{m}$ , such that each inclusion $X_{m} ↪ X_{m + 1}$ is a cofibration and a homotopy equivalence. Then each inclusion $X_{m} ↪ X$ is a homotopy equivalence, and if each $X_{m}$ has the homotopy type of a CW complex, so does $X$ .

Proof. Build the mapping telescope $T = ⨆_{m} X_{m} \times [m, m + 1] / \sim$ , gluing $(x, m + 1) \in X_{m} \times [m, m + 1]$ to $(x, m + 1) \in X_{m + 1} \times [m + 1, m + 2]$ via the inclusion. The projection $T \to X$ , $(x, s) \mapsto x$ , is a homotopy equivalence because the telescope deformation-retracts onto $X$ when the inclusions are cofibrations (collapse each cylinder onto its right end, compatibly). Within $T$ , the inclusion $X_{m} \times {m} ↪ T$ is a homotopy equivalence since each cylinder $X_{m} \times [m, m + 1]$ retracts to $X_{m} \times {m}$ and each inclusion $X_{m} ↪ X_{m + 1}$ is one by hypothesis. Composing, $X_{m} ↪ X$ is a homotopy equivalence. For the CW conclusion, each $X_{m}$ is homotopy equivalent to a CW complex $K_{m}$ , and the homotopy equivalences may be chosen compatibly with the inclusions (each inclusion is a cofibration, so the equivalence extends), realising $X$ as the colimit $lim K_{m}$ , which is a CW complex with the weak topology. $□$

Proposition (each non-critical sublevel set is a finite CW complex). If $c$ is not a critical value of $E$ and only finitely many geodesics from $p$ to $q$ have energy $\leq c$ , then $Ω^{c}$ has the homotopy type of a finite CW complex with one $λ$ -cell per index- $λ$ geodesic of energy $< c$ .

Proof. By Step 1 of the fundamental theorem, $Ω^{c}$ deformation-retracts onto $B^{c} = Ω (t_{0}, \dots, t_{k}) \cap Ω^{c}$ , a sublevel set of the smooth Morse function $E ∣_{Ω (t_{0}, \dots, t_{k})}$ on a finite-dimensional manifold (Step 2). By the closed-manifold-with-boundary version of the handle theorem 03.02.31, a sublevel set of a Morse function with finitely many critical points below level $c$ has the homotopy type of a finite CW complex with one cell of dimension equal to the index of each such critical point. The critical points of $E$ below $c$ are the geodesics of energy $< c$ , of index $λ$ each, by the Morse Index Theorem 03.02.19. $□$

Proposition (geodesic energies are discrete and unbounded; non-conjugate genericity). On a complete $M$ with $(p, q)$ non-conjugate, the set of energies of geodesics from $p$ to $q$ is a discrete subset of $[0, \infty)$ with no finite accumulation point, and is unbounded if $M$ is compact with $π_{1}$ infinite or $M$ is non-compact suitably.

Proof. Each geodesic of energy $\leq c$ is a nondegenerate critical point of $E$ (non-conjugacy), hence isolated in $Ω$ by the Morse lemma in the broken-geodesic model; on the compact set $B^{c}$ there are finitely many, so the critical values in $[0, c]$ are finite, giving discreteness with no accumulation below any $c$ . Unboundedness: if there were a largest energy $c_{0}$ , then $Ω = Ω^{c_{0}}$ would be a finite CW complex, contradicting that $Ω (M; p, q) ≃ Ω M$ (based loop space) has nonzero homology in arbitrarily high degree whenever $M$ has nontrivial higher homotopy or an infinite fundamental group. $□$

Proposition (cell count equals geodesic count; Morse inequality on $Ω$ ). In the CW model of $Ω (M; p, q)$ , the number of $λ$ -cells equals the number of index- $λ$ geodesics, and consequently the number of index- $λ$ geodesics is at least $b_{λ} = dim H_{λ} (Ω (M; p, q))$ .

Proof. The cell-count statement is the content of the fundamental theorem (one cell of dimension $λ$ per index- $λ$ geodesic). Cellular homology 03.12.13 of the CW model computes $H_{*} (Ω (M; p, q))$ , and the chain group $C_{λ}$ is free of rank $c_{λ}$ = (number of index- $λ$ geodesics). The weak Morse inequality of 03.02.31, $c_{λ} \geq b_{λ}$ , follows from rank-nullity of the cellular boundary maps exactly as in the finite-dimensional case. $□$

Connections Master

Jacobi fields, conjugate points, and the Morse Index Theorem 03.02.19. That unit supplies the two ingredients this one assembles: the broken-geodesic finite-dimensional model $Ω (t_{0}, \dots, t_{k})$ , and the computation of the index of a geodesic as its interior conjugate-point count. Here those are repackaged into cells — each index- $λ$ geodesic becomes a $λ$ -cell — so the index theorem is precisely the dimension formula for the cells of the path space.
Handle attachment, CW type, and the Morse inequalities 03.02.31. This unit is the path-space lift of that one. The handle-attachment theorems 3.1 and 3.2 are applied verbatim to the energy functional on the broken-geodesic model, and the Morse inequalities transport to bound the number of geodesics from $p$ to $q$ below by the Betti numbers of $Ω (M; p, q)$ . The finite-dimensional CW corollary there is the local engine of the infinite-dimensional theorem here.
CW complexes 03.12.10. The target object of the fundamental theorem is a CW complex in the precise sense of that unit: cells attached along sphere maps into lower skeleta. The monotone-union lemma needed for the infinite-dimensional passage is a statement about colimits of CW pairs, and the cellular-approximation theorem from that unit is what makes the iterated mapping cones an honest CW complex.
Bott periodicity 03.08.07. The deepest application: when $M$ is a symmetric space, the minimal-geodesic refinement of this theorem identifies the loop space, in a range of dimensions set by the index gap, with a manifold of minimal geodesics; iterating across the classical groups yields the periodicity isomorphisms underlying $K$ -theory. The cell-per-geodesic dictionary established here is the mechanism of that proof.
Cellular homology 03.12.13. The cells produced here feed the cellular chain complex, whose homology is the homology of the path space; this is the computational route from geodesic counts to the Betti numbers of $Ω (M; p, q)$ , and it is what makes the Morse inequalities on the loop space effective.

Historical & philosophical context Master

The idea that the calculus of variations should be studied "in the large" — globally, through the topology of the space of competing curves rather than locally through a single extremal — is Marston Morse's, articulated in his 1934 The Calculus of Variations in the Large (American Mathematical Society Colloquium Publications 18) ^{[Morse 1934]}. Morse already understood the geodesic problem as the study of critical points of a functional on a path space, but the clean statement that the path space is a CW complex assembled from one cell per geodesic awaited the CW formalism of J. H. C. Whitehead and the streamlined exposition of John Milnor.

Milnor's Morse Theory (Annals of Mathematics Studies 51, 1963), from notes by Spivak and Wells, gave the canonical treatment: Part III §§16--17 prove the fundamental theorem via the broken-geodesic approximation, and the Appendix isolates the monotone-union lemma that handles the infinite-dimensional passage ^{[Milnor Part III §§16--17]}. The motivating triumph, occupying the final sections, is Raoul Bott's periodicity theorem: Bott's 1959 The stable homotopy of the classical groups (Annals of Mathematics 70, 313--337) computed the stable homotopy of $O, U, Sp$ by realising their loop spaces, through this very theorem, as approximated by manifolds of minimal geodesics ^{[Bott 1959]}. Bott's earlier 1956 paper on the iteration of closed geodesics and the Sturm intersection theory ^{[Bott 1956]} supplied the index computations that make the periodicity calculation iterate. That a question in the stable homotopy of Lie groups — a purely topological question — should be answered by counting conjugate points along geodesics in a Riemannian symmetric space remains one of the most striking unifications in twentieth-century mathematics, and it is the philosophical centre of gravity of the whole subject: global topology read off the focusing of geodesics.

Bibliography Master

@book{Milnor1963Morse,
  author    = {Milnor, John W.},
  title     = {Morse Theory},
  series    = {Annals of Mathematics Studies},
  number    = {51},
  publisher = {Princeton University Press},
  year      = {1963},
  note      = {Based on lecture notes by M. Spivak and R. Wells. Part III, §§16--17 and Appendix}
}

@article{Bott1959Stable,
  author  = {Bott, Raoul},
  title   = {The stable homotopy of the classical groups},
  journal = {Annals of Mathematics},
  volume  = {70},
  pages   = {313--337},
  year    = {1959}
}

@article{Bott1956Iteration,
  author  = {Bott, Raoul},
  title   = {On the iteration of closed geodesics and the {S}turm intersection theory},
  journal = {Communications on Pure and Applied Mathematics},
  volume  = {9},
  pages   = {171--206},
  year    = {1956}
}

@book{Morse1934CVL,
  author    = {Morse, Marston},
  title     = {The Calculus of Variations in the Large},
  series    = {American Mathematical Society Colloquium Publications},
  volume    = {18},
  publisher = {American Mathematical Society},
  year      = {1934}
}

@book{doCarmo1992,
  author    = {do Carmo, Manfredo P.},
  title     = {Riemannian Geometry},
  publisher = {Birkh{\"a}user},
  year      = {1992}
}

@book{Klingenberg1978,
  author    = {Klingenberg, Wilhelm},
  title     = {Lectures on Closed Geodesics},
  series    = {Grundlehren der mathematischen Wissenschaften},
  number    = {230},
  publisher = {Springer},
  year      = {1978}
}

@book{Palais1963Morse,
  author    = {Palais, Richard S.},
  title     = {Morse theory on {H}ilbert manifolds},
  journal   = {Topology},
  volume    = {2},
  pages     = {299--340},
  year      = {1963}
}

Prerequisites

03.02.19
03.02.31
03.12.10

Tier anchors

beginner: Milnor Morse Theory §16 (the broken-geodesic picture); Needham Visual Differential Geometry (geodesics on the sphere) informal
intermediate: Milnor Morse Theory Part III §§16--17; do Carmo Riemannian Geometry Ch. 12; Lee Introduction to Riemannian Manifolds Ch. 11
master: Milnor Morse Theory Part III §§16--17 and Appendix; Bott 1956 Comm. Pure Appl. Math. 9; Bott 1959 Ann. Math. 70 (the periodicity theorem); Klingenberg Lectures on Closed Geodesics Ch. 2

References

03-modern-geometry/Bott-LecturesOnMorseTheory.pdf · The path space, the energy functional, the CW approximation, and periodicity
Milnor — Morse Theory (Annals of Mathematics Studies 51, 1963) · Part III §§16--17 (the fundamental theorem) and the Appendix (homotopy type of a monotone union)
Bott — The stable homotopy of the classical groups (1959) · Annals of Mathematics 70, pp. 313--337
Bott — On the iteration of closed geodesics and the Sturm intersection theory (1956) · Comm. Pure Appl. Math. 9, pp. 171--206
do Carmo — Riemannian Geometry (1992) · Ch. 12 (the fundamental theorem of Morse theory for the energy functional)

Estimated time

beginner: 18m
intermediate: 45m
master: 82m