03.15.03 · modern-geometry / morse-homology

Compactness: broken trajectories

shipped3 tiersLean: none

Anchor (Master): Schwarz *Morse Homology* Part I Ch. 4; Floer 1988; Salamon 1990

Intuition Beginner

You already met the downhill streams of a height function: each one starts sitting still at a high resting point and ends sitting still at a lower one. Now imagine you have a whole sequence of such streams between the same two resting points, and you watch what they settle down to as you go further along the sequence. Sometimes they settle onto another honest stream. But sometimes something stranger happens: the limiting path slows down so much near a resting point in the middle that it seems to stop there for an arbitrarily long time, then continue downward later.

In the limit, that single stream splits into two. The path runs from the top resting point down to a middle resting point, pauses there forever, and a second path carries on from that middle point down to the bottom. Two genuine streams, joined end to end at a middle stop. This joined object is called a broken trajectory.

The point of this unit is that nothing worse than breaking can happen. A sequence of downhill streams either settles onto one honest stream or onto such a chain of streams linked at resting points. Streams cannot run off to nowhere, develop kinks, or speed up without bound. The reason is a budget: every stream spends exactly the height drop between its two endpoints, and that drop is shared out among the pieces.

Visual Beginner

Alt text: An upright torus carrying its height function with a top, a middle saddle, and a bottom. A family of downhill streams from top to bottom is shown bending ever closer to the saddle and pausing there longer and longer; the limit is two separate streams joined at the saddle, a broken path. The picture conveys that a sequence of trajectories can break at a middle resting point in the limit.

Worked example Beginner

Take the round sphere with its height function: north pole $N$ on top, south pole $S$ at the bottom, and no resting point in between. Every downhill stream runs along a meridian from $N$ straight to $S$ . Pick a sequence of these streams whose meridians swing around toward, say, the meridian through London. As you go along the sequence, the streams simply approach the London meridian, an honest stream from $N$ to $S$ . Nothing breaks, because there is no middle resting point to break at.

Now tilt a doughnut on its edge instead. It has a top $t$ , a bottom $b$ , and a saddle $s$ partway down. Watch a sequence of streams from $t$ to $b$ whose paths drift toward the saddle. Each later stream passes nearer to $s$ and dawdles there longer before sliding on to $b$ . In the limit the dawdling becomes infinite: the path reaches $s$ and stops.

What this tells us: whether a sequence of streams stays whole or breaks depends on what resting points sit between the endpoints in height. With nothing in between (the sphere) the limit is whole. With a middle resting point on the way down (the doughnut) the limit can break there. The total height drop $t$ to $b$ equals the drop $t$ to $s$ plus the drop $s$ to $b$ — the budget is simply split between the two pieces.

Check your understanding Beginner

Question 1 (easy, multiple-choice). A broken trajectory from a top resting point $p$ to a bottom resting point $q$ is made of:

A single downhill path that loops back on itself.
Several downhill paths joined end to end at middle resting points, each strictly lower than the last.
A path that goes uphill for part of its length.
Two paths starting at $p$ that never meet.

Hint

A broken path pauses forever at a middle resting point, then continues downward from there.

Answer

B. A broken trajectory is a chain of honest downhill streams, joined at resting points, with the height strictly dropping at each junction. Feedback if correct: right — each piece is a genuine flow line, and they are linked at intermediate critical points. Feedback if wrong: downhill streams never go uphill or loop; breaking means the single stream splits into a chain of lower-and-lower pieces.

Formal definition Intermediate+

Fix a closed Riemannian manifold $(M^{n}, g)$ and a Morse-Smale pair $(f, g)$ , with $\nabla f$ the gradient and $μ (p)$ the Morse index of a critical point $p$ , exactly as set up in 03.15.01 and 03.15.02. For critical points $p \neq = q$ write $M (p, q)$ for the space of parametrized negative-gradient trajectories from $p$ to $q$ and $M (p, q) = M (p, q) / R$ for its quotient by time translation.

The quantity that controls every limit is the energy of a trajectory. For $γ \in M (p, q)$ ,

E (γ) = \int_{- \infty}^{+ \infty} \overset{γ}{˙} (t)_{g}^{2} d t = \int_{- \infty}^{+ \infty} \nabla f (γ (t))_{g}^{2} d t = f (p) - f (q),

where the last equality uses $\frac{d}{d t} f (γ) = - ∣\nabla f (γ) ∣_{g}^{2}$ and the asymptotic limits $γ (\mp \infty) = p, q$ . The energy of a trajectory depends only on its endpoints, and equals the height drop. It is finite because $M$ is compact, and it is additive: along any chain of critical points the drops sum.

A broken trajectory from $p$ to $q$ is a finite ordered tuple

v = (v_{1}, v_{2}, \dots, v_{k}), v_{i} \in M (z_{i - 1}, z_{i}),

where $z_{0} = p$ , $z_{k} = q$ , and $z_{0}, z_{1}, \dots, z_{k}$ are pairwise-distinct critical points with strictly decreasing critical values $f (z_{0}) > f (z_{1}) > \dots > f (z_{k})$ . The intermediate points $z_{1}, \dots, z_{k - 1}$ are the breaks; the integer $k \geq 1$ is the number of pieces. A single honest unparametrized trajectory is the case $k = 1$ . The Morse-Smale dimension count $dim M (z_{i - 1}, z_{i}) = μ (z_{i - 1}) - μ (z_{i}) - 1 \geq 0$ forces $μ (z_{0}) > μ (z_{1}) > \dots > μ (z_{k})$ , so a broken trajectory from $p$ to $q$ has at most $μ (p) - μ (q)$ pieces.

A sequence $[γ_{m}] \in M (p, q)$ converges geometrically to a broken trajectory $v = (v_{1}, \dots, v_{k})$ if there are representatives $γ_{m}$ and, for each piece $i$ , a sequence of time shifts $τ_{m}^{i} \in R$ such that the reparametrized curves $t \mapsto γ_{m} (t + τ_{m}^{i})$ converge to a representative of $v_{i}$ in $C_{loc}^{\infty} (R, M)$ as $m \to \infty$ , with the shifts ordered, $τ_{m}^{1} ≪ τ_{m}^{2} ≪ \dots ≪ τ_{m}^{k}$ , and the gaps $τ_{m}^{i + 1} - τ_{m}^{i} \to + \infty$ . Geometric convergence is the precise meaning of "the stream dawdles ever longer at each break."

The broken-trajectory compactification of $M (p, q)$ is the set

\overline{M} (p, q) = k \geq 1 ⨆ p = z_{0}, z_{k} = q f (z_{0}) > \dots > f (z_{k}) ⨆ M (z_{0}, z_{1}) \times \dots \times M (z_{k - 1}, z_{k}),

topologized so that geometric convergence is convergence in $\overline{M} (p, q)$ . The compactness theorem below asserts that this space is compact, and that $M (p, q)$ is its open, dense, top stratum.

Counterexamples to common slips

Breaking can only occur at critical points strictly between $p$ and $q$ in value, not merely in index. Two index- $1$ saddles at the same height never appear as consecutive breaks of one trajectory, because a piece $v_{i} \in M (z_{i - 1}, z_{i})$ requires $f (z_{i - 1}) > f (z_{i})$ . Equal-value saddles are exactly the Morse-Smale-violating degeneracy excluded by hypothesis.
Geometric convergence is not $C^{0}$ or uniform convergence on all of $R$ . The curves $γ_{m}$ have no single time-shift making them converge uniformly; one needs a separate shift $τ_{m}^{i}$ for each piece, with the gaps diverging. A common error is to seek one limit curve where the limit is a $k$ -tuple.
Energy bounds the number of breaks, but the bound $k \leq μ (p) - μ (q)$ uses Morse-Smale. Without transversality there can be flow lines $z_{i - 1} \to z_{i}$ with $μ (z_{i - 1}) \leq μ (z_{i})$ , and the index need not drop at each break; the clean dimension stratification fails.
The energy is the unparametrized invariant $f (p) - f (q)$ , not the arc length. Two trajectories with the same endpoints have equal energy but generally different lengths; it is the energy, fixed by the endpoints, that supplies the a-priori bound for Arzelà-Ascoli.

Key theorem with proof Intermediate+

The compactness theorem is the central analytic statement of Schwarz Part I Chapter 4 ^{[Schwarz Part I Ch. 4]}. It says the unparametrized moduli is precompact and its missing limits are broken trajectories.

Theorem (compactness by broken trajectories). Let $(f, g)$ be a Morse-Smale pair on a closed Riemannian manifold $(M^{n}, g)$ and let $p \neq = q$ be critical points. Every sequence $[γ_{m}] \in M (p, q)$ has a subsequence converging geometrically to a broken trajectory $(v_{1}, \dots, v_{k})$ from $p$ to $q$ with $1 \leq k \leq μ (p) - μ (q)$ . Consequently $\overline{M} (p, q)$ is compact, and $M (p, q)$ is precompact with the broken trajectories as its only added limit points.

Proof. Choose representatives $γ_{m} : R \to M$ of $[γ_{m}]$ . Every trajectory obeys the uniform first-derivative bound $∣ \overset{γ}{˙}_{m} (t) ∣_{g} = ∣\nabla f (γ_{m} (t)) ∣_{g} \leq max_{M} ∣\nabla f ∣_{g} =: C_{0}$ , finite by compactness of $M$ . Differentiating the flow equation gives $\overset{γ}{¨}_{m} = - Hess f (γ_{m}) \overset{γ}{˙}_{m}$ , so $∣ \overset{γ}{¨}_{m} ∣_{g} \leq C_{1} C_{0}$ with $C_{1} = max_{M} ∥ Hess f ∥_{g}$ ; all higher derivatives are bounded the same way from the flow equation and compactness. The family ${γ_{m}}$ is therefore equicontinuous with all derivatives uniformly bounded on $R$ .

Normalize each representative by a level-crossing condition: fix a regular value $c_{1}$ with $f (q) < c_{1} < f (p)$ and shift each $γ_{m}$ so that $f (γ_{m} (0)) = c_{1}$ . By Arzelà-Ascoli applied on each compact time interval and a diagonal argument, a subsequence (still indexed by $m$ ) converges in $C_{loc}^{\infty} (R, M)$ to a smooth curve $v_{1} : R \to M$ solving the flow equation $\overset{v}{˙}_{1} = - \nabla f (v_{1})$ . The limit is non-constant since $f (v_{1} (0)) = c_{1}$ is a regular value, so $\nabla f (v_{1} (0)) \neq = 0$ .

The limit $v_{1}$ has finite energy: $E (v_{1}) = \int ∣ \overset{v}{˙}_{1} ∣^{2} \leq lim inf_{m} E (γ_{m}) = f (p) - f (q) < \infty$ by Fatou. A finite-energy gradient trajectory on a closed manifold converges at each end to a critical point (the $ω$ - and $α$ -limit sets are connected, flow-invariant, and contained in ${\nabla f = 0}$ , hence single critical points since these are isolated). So $v_{1} \in M (z_{0}, z_{1})$ for critical points $z_{0}, z_{1}$ with $f (z_{0}) \geq f (p)$ and $f (z_{1}) \leq c_{1} < f (p)$ . Because $f$ decreases along $γ_{m}$ and $f (γ_{m} (0)) = c_{1}$ , one gets $f (z_{0}) \leq f (p)$ as well, hence $z_{0} = p$ — the backward limit is pinned to $p$ . Set $z_{1} = v_{1} (+ \infty)$ .

If $z_{1} = q$ the sequence converges to the single piece $v_{1}$ and $k = 1$ . Otherwise $f (z_{1}) > f (q)$ , and energy is lost into the tail: $E (v_{1}) = f (p) - f (z_{1}) < f (p) - f (q)$ , so a definite amount $f (z_{1}) - f (q) > 0$ of energy escapes past the level $c_{1}$ toward $q$ . Choose a new regular value $c_{2}$ with $f (q) < c_{2} < f (z_{1})$ and re-shift the *same* subsequence by times $τ_{m}^{2} \to + \infty$ so that $f (γ_{m} (τ_{m}^{2})) = c_{2}$ ; that such shifts exist and tend to $+ \infty$ is exactly the statement that the trajectories spend unbounded time near $z_{1}$ (the field $\nabla f$ is small near the critical point, so crossing from level $c_{1}$ down to $c_{2}$ takes arbitrarily long). Arzelà-Ascoli again extracts a non-constant limit $v_{2} \in M (z_{1}, z_{2})$ with $f (z_{1}) > f (z_{2})$ . The strict drop $f (z_{1}) > f (z_{2})$ and the finiteness of the total energy $f (p) - f (q)$ bound the number of iterations: each new piece consumes energy at least the spectral-gap-controlled minimum positive value $min_{z \neq = z^{'}} (f (z) - f (z^{'}))$ over critical values, so the process terminates after $k \leq (f (p) - f (q)) / min Δ f$ steps with $z_{k} = q$ . Morse-Smale further forces $μ (z_{i - 1}) > μ (z_{i})$ at each break, giving the sharper $k \leq μ (p) - μ (q)$ . The resulting tuple $(v_{1}, \dots, v_{k})$ is the geometric limit. Compactness of $\overline{M} (p, q)$ follows: any sequence has a geometrically convergent subsequence, and the limit lies in the disjoint union defining the compactification. $□$

Bridge. The compactness theorem builds toward 03.15.04, where gluing supplies the converse — every broken trajectory $p \to z \to q$ with consecutive index drops one is the limit of a unique nearby family of honest trajectories — so that breaking and gluing together identify the boundary of a compactified one-dimensional moduli. This is exactly the mechanism that appears again in 03.15.06: when $μ (p) - μ (q) = 2$ the space $\overline{M} (p, q)$ is a compact $1$ -manifold with boundary, its boundary the once-broken trajectories, and the foundational reason that $\partial^{2} = 0$ is that the boundary of a compact $1$ -manifold has signed count zero. Putting these together, the energy identity $E (γ) = f (p) - f (q)$ proved here is the finite-dimensional shadow of the symplectic-action and Yang-Mills energy bounds that drive Gromov and Uhlenbeck compactness; the bridge is that in each theory an a-priori energy bound plus an elliptic estimate yields subconvergence away from finitely many concentration points, and only the nature of the lost energy — a break here, a bubble or instanton there — distinguishes the settings.

Exercises Intermediate+

Exercise 4 (medium, numeric). For a Morse-Smale pair, $p$ has index $2$ and $q$ has index $0$, so $\widehat{\mathcal{M}}(p,q)$ has dimension $1$. Through index-$1$ critical points $r_1, r_2$ there are flow lines, and $\overline{\widehat{\mathcal{M}}}(p,q)$ is a compact $1$-manifold with boundary. If the boundary consists of the broken trajectories $p \to r_1 \to q$ and $p \to r_2 \to q$, how many boundary points does this compact $1$-manifold have?

Hint

Each once-broken trajectory $p \to r_{i} \to q$ is one boundary point; count them.

Answer

$2$ . Each broken trajectory $p \to r_{i} \to q$ contributes one boundary point, so there are $2$ . A compact $1$ -manifold with boundary has an even number of boundary points, consistent with $\overline{M} (p, q)$ being a single closed interval whose two ends are these two broken trajectories.

Exercise 5 (medium, short-answer). Explain why a sequence of trajectories in $\widehat{\mathcal{M}}(p,q)$ that "dawdles" near an intermediate critical point $z$ cannot be made to converge by a single time-shift, and why a separate shift per piece is needed.

Hint

The time spent near $z$ diverges; pick the shift that centres a fixed regular level for each piece.

Answer

Rubric. Full credit: as the trajectories spend unbounded time near $z$ , the portion of $γ_{m}$ before $z$ and the portion after $z$ are separated by an interval of length $\to \infty$ on which $γ_{m}$ sits near the rest point. A single shift can centre at most one of these portions at a fixed regular level; the other recedes to $\pm \infty$ and disappears in the $C_{loc}^{\infty}$ limit, registering only as the constant $z$ . To recover both pieces one normalizes each separately by a level-crossing condition at a regular value chosen between the relevant critical values, producing shifts $τ_{m}^{1} ≪ τ_{m}^{2}$ with $τ_{m}^{2} - τ_{m}^{1} \to \infty$ . The limit is the pair $(v_{1}, v_{2})$ , not a single curve.

Exercise 6 (hard, short-answer). Use the energy identity and the existence of a smallest positive gap between critical values to bound the number of breaks of any broken trajectory from $p$ to $q$ in terms of $f(p) - f(q)$, independently of the index count.

Hint

Each piece $v_{i}$ carries energy $f (z_{i - 1}) - f (z_{i}) \geq δ$ where $δ = min (f (z) - f (z^{'}))$ over distinct critical values.

Answer

Rubric. Let $δ = min {f (z) - f (z^{'}) : z, z^{'} \in Crit (f), f (z) > f (z^{'})} > 0$ be the least positive gap between distinct critical values (finite-many criticals, so the min is attained and positive). Each piece $v_{i} \in M (z_{i - 1}, z_{i})$ has energy $f (z_{i - 1}) - f (z_{i}) \geq δ$ . The total energy is $\sum_{i} (f (z_{i - 1}) - f (z_{i})) = f (p) - f (q)$ by telescoping. Hence $k δ \leq f (p) - f (q)$ , giving $k \leq (f (p) - f (q)) / δ$ . This bound uses only the energy budget, not transversality; the Morse-Smale sharpening $k \leq μ (p) - μ (q)$ requires the index to drop at each break.

Exercise 7 (hard, short-answer). On the round sphere $S^2$, the only critical points are $N$ (index $2$) and $S$ (index $0$), with no intermediate critical point. Explain why $\widehat{\mathcal{M}}(N,S) \cong S^1$ is already compact, with no broken trajectories, and reconcile this with the compactness theorem.

Hint

A broken trajectory from $N$ to $S$ would need an intermediate critical point of index strictly between $2$ and $0$ .

Answer

Rubric. Full credit: a break of a trajectory from $N$ to $S$ would occur at a critical point $z$ with $f (N) > f (z) > f (S)$ and, under Morse-Smale, $2 = μ (N) > μ (z) > μ (S) = 0$ , so $μ (z) = 1$ . The sphere's height function has no index- $1$ critical point, so no break is possible and $\overline{M} (N, S) = M (N, S)$ . The unparametrized moduli is the circle of meridian directions, already compact. The compactness theorem is consistent: it guarantees a geometrically convergent subsequence, and here every such subsequence converges to an honest (unbroken) trajectory because the compactification adds nothing. Compactness of $M (N, S)$ itself reflects the absence of intermediate critical levels.

Lean formalization Intermediate+

Mathlib carries the analytic inputs — Arzelà-Ascoli, Grönwall, flows of smooth vector fields — but not the moduli-space scaffolding on which a broken-trajectory compactness statement is phrased, so this unit ships at lean_status: none. The pseudo-Lean below indicates the intended shape; it does not compile because trajectorySpace, energy, brokenTrajectory, and GeometricConverges are not defined in Mathlib.

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

-- Energy of a gradient trajectory equals the height drop.
theorem energy_eq_drop (g) (f) (p q : M) (γ : trajectorySpace g f p q) :
    energy g γ = f p - f q := by sorry

-- A broken trajectory: a chain of unparametrised pieces with strictly
-- decreasing critical values.
structure BrokenTrajectory (g) (f) (p q : M) where
  break_pts : List M                          -- z_1, …, z_{k-1}
  pieces    : ...                              -- v_i ∈ Mhat(z_{i-1}, z_i)
  decreasing : Chain (fun a b => f a > f b) p (break_pts ++ [q])

-- Compactness: every sequence in Mhat(p,q) has a subsequence converging
-- geometrically to a broken trajectory.
theorem compactness_broken (g) (f) (hMS : MorseSmale g f) (p q : M) :
    ∀ (γ : ℕ → unparamTrajectorySpace g f p q),
      ∃ (φ : ℕ → ℕ) (StrictMono φ) (v : BrokenTrajectory g f p q),
        GeometricConverges (γ ∘ φ) v := by sorry

The hard analysis (uniform derivative bounds from compactness, the diagonal Arzelà-Ascoli extraction, finite-energy convergence to critical points) reduces to Mathlib-available lemmas once the moduli spaces exist; the named gap in Mathlib gap analysis is precisely the moduli/energy scaffolding and the geometric-convergence topology.

Advanced results Master

The compactness theorem refines into a stratified description of the compactified moduli, and the energy identity is the first instance of a phenomenon that recurs across geometric analysis.

Stratification of the compactification. For a Morse-Smale pair, $\overline{M} (p, q)$ is a stratified space whose stratum of $k$ -fold broken trajectories is $$ \bigsqcup_{\substack{p = z_0,, z_k = q \ \mu(z_0) > \cdots > \mu(z_k)}} \widehat{\mathcal{M}}(z_0,z_1) \times \cdots \times \widehat{\mathcal{M}}(z_{k-1}, z_k), $$ of dimension $\sum_{i} (μ (z_{i - 1}) - μ (z_{i}) - 1) = (μ (p) - μ (q)) - k$ . The top stratum $k = 1$ has dimension $μ (p) - μ (q) - 1$ ; each break drops the dimension by one. When $μ (p) - μ (q) = 1$ the top stratum is $0$ -dimensional and compact, hence finite, and there are no breaks (a break would need a stratum of dimension $- 1$ ): this finiteness is what makes the Morse differential a well-defined finite sum. When $μ (p) - μ (q) = 2$ the top stratum is a $1$ -manifold and the codimension-one stratum $k = 2$ is the finite set of once-broken trajectories $p \to z \to q$ with $μ (z) = μ (p) - 1$ ; together with gluing this exhibits $\overline{M} (p, q)$ as a compact $1$ -manifold with boundary.

Geometric convergence as Gromov-Hausdorff degeneration. The mode of convergence — reparametrize, extract a $C_{loc}^{\infty}$ limit on each piece, lose energy into necks where the trajectory lingers at a rest point — is the finite-dimensional template for the compactness theorems of gauge theory and symplectic topology. In Uhlenbeck compactness for anti-self-dual connections on a cylinder 03.07.20, energy concentrates at finitely many points (bubbling) or escapes down infinite necks into lower-energy instantons; the neck degeneration is the exact analogue of the break here, with the cylinder's translation symmetry playing the role of the time-translation $R$ . The breaking phenomenon was first identified by Floer ^{[Floer 1988]} for the unregularized gradient flow of the symplectic action, and the systematic finite-dimensional account is Salamon's survey ^{[Salamon 1990]} and Schwarz's monograph ^{[Schwarz Part I Ch. 4]}.

Sharpness of the energy bound. The a-priori bound $E (γ) = f (p) - f (q)$ is sharp and uniform: it depends only on the endpoints, not on the trajectory or the metric within a Morse-Smale family. This uniformity is what upgrades pointwise subconvergence to convergence of the whole moduli. Were the energy unbounded along a sequence — as happens when the asymptotic ends degenerate (Morse-Bott) and a family of trajectories can carry growing action — the limit could fail to be a broken trajectory of the same endpoints, and the compactification would acquire extra strata. Nondegeneracy of the critical points, through the exponential-decay estimates of 03.15.02, is precisely what pins the energy and the asymptotic limits.

The limit pieces are genuine trajectories, not just connecting orbits. Finite energy alone gives a curve whose limit sets are connected invariant subsets of ${\nabla f = 0}$ ; isolatedness of nondegenerate critical points upgrades these to single points, so each piece $v_{i}$ lies in an honest $M (z_{i - 1}, z_{i})$ . This is the Morse-theoretic shadow of Conley's connecting-orbit theory ^{[Conley 1978]}, where the limit objects are connections between isolated invariant sets in the absence of any nondegeneracy hypothesis; the Morse-Smale assumption collapses Conley's index pairs to the clean broken-trajectory picture.

Synthesis. The energy identity $E (γ) = f (p) - f (q)$ is the foundational reason compactness holds: it is exactly the a-priori bound that feeds Arzelà-Ascoli, and it is dual to the index identity $ind D_{γ} F = μ (p) - μ (q)$ of 03.15.02 in the sense that energy governs the analytic compactness while the index governs the dimension. Putting these together, the broken-trajectory compactification stratifies $\overline{M} (p, q)$ by the number of breaks, with the codimension of each stratum equal to the number of breaks, and this is exactly the structure that makes $\overline{M} (p, q)$ a compact $1$ -manifold with boundary when $μ (p) - μ (q) = 2$ . The construction generalises: replacing energy by the symplectic action gives Floer's breaking, and replacing it by the Yang-Mills functional gives Uhlenbeck's bubbling; the central insight common to all three is that a finite energy bound plus an elliptic estimate forces subconvergence away from finitely many concentration phenomena, and the Morse setting is where that insight is visible without the analytic overhead. The compactness proved here, the gluing of 03.15.04 that inverts it, and the coherent orientations of 03.15.05 are the three inputs that 03.15.06 assembles into $\partial^{2} = 0$ .

Full proof set Master

The compactness theorem is proved in full in the Key theorem section. The supporting structural claims are recorded here.

Proposition (energy identity and additivity). For $γ \in M (p, q)$ one has $E (γ) = \int_{R} ∣ \overset{γ}{˙} ∣_{g}^{2} d t = f (p) - f (q)$ , and for a broken trajectory $(v_{1}, \dots, v_{k})$ from $p$ to $q$ through $z_{0} = p, \dots, z_{k} = q$ the energies add: $\sum_{i = 1}^{k} E (v_{i}) = f (p) - f (q)$ .

Proof. Along $γ$ , $\frac{d}{d t} f (γ) = df (\overset{γ}{˙}) = g (\nabla f, - \nabla f) = - ∣\nabla f (γ) ∣_{g}^{2} = - ∣ \overset{γ}{˙} ∣_{g}^{2}$ . Integrating over $R$ and using $lim_{t \to \mp \infty} f (γ (t)) = f (p), f (q)$ gives $\int_{R} ∣ \overset{γ}{˙} ∣^{2} d t = f (p) - f (q)$ ; the integral is finite because $∣ \overset{γ}{˙} ∣$ decays exponentially at both ends by the nondegeneracy estimates of 03.15.02. For the chain, each piece $v_{i} \in M (z_{i - 1}, z_{i})$ has $E (v_{i}) = f (z_{i - 1}) - f (z_{i})$ , and the sum telescopes: $\sum_{i = 1}^{k} (f (z_{i - 1}) - f (z_{i})) = f (z_{0}) - f (z_{k}) = f (p) - f (q)$ . $■$

Proposition (finite-energy limit is a trajectory). Let $v : R \to M$ solve $\overset{v}{˙} = - \nabla f (v)$ on a closed Morse manifold with $E (v) = \int_{R} ∣ \overset{v}{˙} ∣^{2} < \infty$ . Then the limits $v (\pm \infty)$ exist and are critical points.

Proof. Finite energy forces $lim inf_{t \to + \infty} ∣ \overset{v}{˙} (t) ∣ = 0$ , so there are times $t_{j} \to + \infty$ with $\nabla f (v (t_{j})) \to 0$ ; by compactness a subsequence of $v (t_{j})$ converges to some $x$ with $\nabla f (x) = 0$ , a critical point. The $ω$ -limit set $ω (v) = ⋂_{T} \overline{{v (t) : t \geq T}}$ is non-empty, compact, connected, and flow-invariant. Since $f$ is non-increasing along $v$ and bounded below, $f$ is constant on $ω (v)$ , so $ω (v) \subseteq {\nabla f = 0} \cap f^{- 1} (c)$ for a single value $c$ ; being connected and contained in the finite, hence discrete, critical set, $ω (v)$ is a single critical point. Thus $v (+ \infty)$ exists and is critical, and likewise $v (- \infty)$ by the same argument for the reversed flow. Exponential convergence then follows from the nondegeneracy of that limit critical point as in 03.15.02. $■$

Proposition (a $1$ -manifold-with-boundary for index difference two). Let $(f, g)$ be Morse-Smale and $μ (p) - μ (q) = 2$ . Then, granting the gluing theorem of 03.15.04, $\overline{M} (p, q)$ is a compact $1$ -manifold with boundary, and its boundary is the finite set of once-broken trajectories ${(v_{1}, v_{2}) : v_{1} \in M (p, z), v_{2} \in M (z, q), μ (z) = μ (p) - 1}$ .

Proof. The top stratum $M (p, q)$ is a $1$ -manifold of dimension $μ (p) - μ (q) - 1 = 1$ by transversality 03.15.02. By the compactness theorem its only added limit points are broken trajectories; a break at $z$ requires $μ (p) > μ (z) > μ (q)$ , so $μ (z) = μ (p) - 1$ and the chain has exactly two pieces, each in a $0$ -dimensional (hence finite) moduli $M (p, z)$ , $M (z, q)$ . Thus the codimension-one stratum is the finite set of once-broken trajectories, and there are no deeper strata. Gluing 03.15.04 provides, for each once-broken trajectory, a half-open arc of honest trajectories converging to it, giving $\overline{M} (p, q)$ a manifold-with-boundary chart at each broken point. Compactness from the compactness theorem and the absence of further strata make $\overline{M} (p, q)$ a compact $1$ -manifold with boundary, with boundary the broken trajectories. $■$

The energy and finite-energy-limit propositions are proved here in full; the $1$ -manifold-with-boundary statement is stated with its gluing input quoted, the gluing construction itself being the content of 03.15.04. A complete account is in Schwarz ^{[Schwarz Part I Ch. 4]}.

Connections Master

03.15.02 (trajectory spaces, the Fredholm setup, and transversality) is the direct upstream input: it makes $M (p, q)$ a smooth manifold of dimension $μ (p) - μ (q) - 1$ and supplies the exponential-decay estimates at the ends that pin the asymptotic limits and bound the energy. This unit asks how non-compact that manifold is, and answers that the missing limits are broken trajectories. The Fredholm index there and the energy here are the two halves of the moduli's structure — one governing dimension, the other governing compactness.

03.15.04 (gluing of trajectories) is the immediate successor and the converse of this unit: where compactness says a sequence of trajectories can degenerate to a broken one, gluing says every once-broken trajectory $p \to z \to q$ with consecutive index drops one arises as the limit of a unique one-parameter family of honest trajectories. Compactness and gluing together identify the broken trajectories with the boundary points of the compactified one-dimensional moduli, which is the geometric content underlying the Morse differential's algebra.

03.15.06 (the Morse complex and $\partial^{2} = 0$ ) is the apex this unit feeds: when $μ (p) - μ (q) = 2$ , the compact $1$ -manifold-with-boundary $\overline{M} (p, q)$ has boundary equal to the once-broken trajectories, and the vanishing of the signed boundary count of a compact $1$ -manifold is exactly $\partial^{2} = 0$ . Without the compactness proved here there is no compact $1$ -manifold to take the boundary of, and the identity would have no geometric meaning.

03.07.20 (Uhlenbeck compactness for ASD equations on cylinders) is the infinite-dimensional analogue: the same reparametrize-and-extract-a-local-limit mechanism appears there, with energy escaping into infinite necks (breaking) or concentrating at points (bubbling). The finite-dimensional broken-trajectory compactness of this unit is the cleanest case of that phenomenon, with no bubbling, and is the template Schwarz designed Morse homology to rehearse before Floer theory.

Historical & philosophical context Master

The degeneration of a sequence of gradient (or pseudo-gradient) flow lines into a chain of flow lines joined at rest points was isolated as the central compactness mechanism of the new infinite-dimensional Morse theory by Andreas Floer, in The unregularized gradient flow of the symplectic action (Communications on Pure and Applied Mathematics 41, 1988, 775–813) and the companion instanton and Lagrangian papers of the same year. Floer's analysis of how trajectories of the symplectic-action functional break, rather than vanish, supplied the compactness needed to make his homology well-defined; the finite-dimensional case, where the moduli are honest manifolds, is logically prior but was abstracted only afterward. Dietmar Salamon's survey Morse theory, the Conley index and Floer homology (Bulletin of the London Mathematical Society 22, 1990, 113–140) placed the broken-trajectory compactification in the lineage of Charles Conley's connecting-orbit theory (Isolated Invariant Sets and the Morse Index, CBMS Regional Conference Series 38, American Mathematical Society, 1978), in which limit objects of flows between isolated invariant sets are studied without any nondegeneracy assumption.

The first self-contained, fully rigorous finite-dimensional treatment — the compactness theorem stated as the precompactness of $M (p, q)$ with a broken-trajectory compactification, proved by energy bounds and Arzelà-Ascoli — is Matthias Schwarz's Morse Homology (Progress in Mathematics 111, Birkhäuser, 1993), from his 1992 ETH Zürich dissertation under Eduard Zehnder and Dietmar Salamon. Schwarz's Part I Chapter 4 develops the compactness exactly as the finite-dimensional rehearsal for the Gromov and Uhlenbeck compactness theorems of symplectic and gauge theory.

Bibliography Master

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

@article{Floer1988action,
  author  = {Floer, Andreas},
  title   = {The unregularized gradient flow of the symplectic action},
  journal = {Communications on Pure and Applied Mathematics},
  volume  = {41},
  number  = {6},
  pages   = {775--813},
  year    = {1988}
}

@article{Salamon1990,
  author  = {Salamon, Dietmar},
  title   = {Morse theory, the {C}onley index and {F}loer homology},
  journal = {Bulletin of the London Mathematical Society},
  volume  = {22},
  number  = {2},
  pages   = {113--140},
  year    = {1990}
}

@book{Conley1978,
  author    = {Conley, Charles},
  title     = {Isolated Invariant Sets and the {M}orse Index},
  series    = {CBMS Regional Conference Series in Mathematics},
  volume    = {38},
  publisher = {American Mathematical Society, Providence, RI},
  year      = {1978}
}

@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

03.15.01
03.15.02

Tier anchors

beginner: Banyaga-Hurtubise *Lectures on Morse Homology* Ch. 4 (broken-flow-line pictures); Audin-Damian *Morse Theory and Floer Homology* §3.2 informal limit picture
intermediate: Schwarz *Morse Homology* Part I Ch. 4; Audin-Damian *Morse Theory and Floer Homology* §3.2; Banyaga-Hurtubise Ch. 6
master: Schwarz *Morse Homology* Part I Ch. 4; Floer 1988; Salamon 1990

References

03-modern-geometry/Bott-LecturesOnMorseTheory.pdf · Gradient flows; convergence of trajectories (background for broken-flow-line compactness)
Schwarz — Morse Homology (Progress in Mathematics 111, Birkhäuser 1993) · Part I, Ch. 4 (the compactness theorem and the broken-trajectory compactification)
Floer — The unregularized gradient flow of the symplectic action · Communications on Pure and Applied Mathematics 41 (1988), pp. 775-813 (breaking of flow lines in the Floer setting)
Salamon — Morse theory, the Conley index and Floer homology · Bulletin of the London Mathematical Society 22 (1990), pp. 113-140 (broken trajectories and the compactness of the moduli)
Conley — Isolated Invariant Sets and the Morse Index · CBMS Regional Conference Series 38 (1978) (connecting orbits and limit behaviour)

Estimated time

beginner: 18m
intermediate: 46m
master: 82m