03.15.05 · modern-geometry / morse-homology

Coherent orientations and characteristic signs

shipped3 tiersLean: none

Anchor (Master): Schwarz *Morse Homology* Part II Ch. 3; Floer-Hofer 1993; Donaldson 1987

Intuition Beginner

You have met the downhill streams that run between resting points of a height function, and you have learned to count them. To build a richer bookkeeping you want to do more than count: you want to attach a sign, a plus or a minus, to each isolated stream. The reason is the same reason a ledger uses debits and credits. If you only ever add, two cancelling effects look like a doubling; once you allow signs, cancellation becomes visible, and a great deal of structure that was hidden in the raw count comes into view.

So the goal of this unit is to decide, for each isolated downhill stream from one resting point to a lower one, whether it should count as $+ 1$ or $- 1$ . The choice cannot be made one stream at a time by whim. It has to be made by a single consistent rule, applied everywhere at once, so that when two streams are joined end to end the sign of the joined object is determined by the signs of its parts. A rule with that joining property is called coherent.

Where does the sign live? Each stream carries, at every moment, a tiny linear picture of how nearby streams spread apart or squeeze together. That linear picture has two natural directions of "extra room": the directions a stream can be pushed and still solve the flow, and the directions it fails to reach. Orienting that room — picking which way is positive — is what fixes the sign. Coherence means these choices, made along every stream, agree wherever streams meet.

Visual Beginner

Alt text: An upright torus with its height function, a top, two saddles, and a bottom. Two streams from the top to a saddle each carry a small transverse arrow marking an orientation; one is labelled plus, the other minus. An inset shows two streams joined at a saddle with the rule that the joined sign is the product of the part signs. The picture conveys that signs are attached by orienting the linear spread-directions of each stream and that they multiply under joining.

Worked example Beginner

Take the upright doughnut with a top $t$ , two saddles $s_{1}$ and $s_{2}$ partway down, and a bottom $b$ . Look only at the streams from the top $t$ down to the saddle $s_{1}$ . Because the height drops by exactly one index step, these streams are isolated points — finitely many separate paths — so each one is a candidate for a sign.

Suppose there are two such streams, call them $u$ and $u^{'}$ . Fix once and for all a direction of "positive spread" at the top $t$ and another at the saddle $s_{1}$ . Sliding the positive direction at $t$ along the stream $u$ down to $s_{1}$ , compare it with the positive direction chosen at $s_{1}$ . If they match, set the sign of $u$ to $+ 1$ ; if they are opposite, set it to $- 1$ . Do the same along $u^{'}$ . Say the comparison gives $u \mapsto + 1$ and $u^{'} \mapsto - 1$ .

The signed count of streams from $t$ to $s_{1}$ is then $(+ 1) + (- 1) = 0$ .

What this tells us: the raw count of streams was $2$ , but the signed count is $0$ . The signs detected that the two streams approach $s_{1}$ from opposite sides and should cancel. Had we picked different positive directions at $t$ or at $s_{1}$ , the individual signs could flip, but the signed total $0$ stays the same up to an overall sign — the cancellation is real, not an artefact of the choices.

Check your understanding Beginner

Formal definition Intermediate+

Fix a closed Riemannian manifold $(M^{n}, g)$ and a Morse-Smale pair $(f, g)$ , with the trajectory spaces $M (p, q)$ , the quotient $M (p, q) = M (p, q) / R$ , and the linearized operator $D_{γ} F = \frac{d}{d t} + A (t)$ exactly as in 03.15.02, where $A (t) = Hess f (γ (t))$ in a parallel frame and $D_{γ} F$ acts on $W^{1, 2} (γ^{*} T M) \to L^{2} (γ^{*} T M)$ , Fredholm of index $μ (p) - μ (q)$ .

For any real Fredholm operator $D : E \to F$ between Banach spaces, the determinant line is the one-dimensional real vector space $$ \det(D) ;=; \Lambda^{\max}\ker D ;\otimes; \big(\Lambda^{\max}\operatorname{coker} D\big)^{*}, $$ where $Λ^{m a x} V = Λ^{d i m V} V$ for a finite-dimensional $V$ , with the convention $Λ^{0} V = R$ . As $D$ varies continuously through Fredholm operators, the lines $det (D)$ assemble into a real line bundle $det \to Fred (E, F)$ over the space of Fredholm operators; the dimensions of $ker$ and $coker$ may jump, but their difference is the index and the determinant line varies continuously through the jumps. An orientation of $D$ is an orientation of the line $det (D)$ , that is, a choice of one of its two rays.

Along a trajectory $γ \in M (p, q)$ the operator $D_{γ} F$ is surjective (Morse-Smale transversality), so $coker D_{γ} F = 0$ and $det (D_{γ} F) = Λ^{m a x} ker D_{γ} F$ . Here $ker D_{γ} F = T_{γ} M (p, q)$ , so an orientation of $D_{γ} F$ is the same as an orientation of the tangent space to the moduli at $γ$ . The $R$ -action by time translation contributes the canonical section $\partial_{t} γ \in ker D_{γ} F$ ; quotienting by it orients $M (p, q)$ .

A coherent orientation is an assignment, to every pair $(p, q)$ and every $γ \in M (p, q)$ , of an orientation $o_{γ}$ of $det (D_{γ} F)$ , varying continuously in $γ$ and compatible with the gluing (catenation) isomorphism $$ \det(D_{\gamma_1}F) \otimes \det(D_{\gamma_2}F) ;\xrightarrow{;\cong;}; \det(D_{\gamma_1 #\rho \gamma_2}F) $$ for $γ_{1} \in M (p, z)$ , $γ_{2} \in M (z, q)$ , and a large gluing parameter $ρ$ , in the sense that the orientation of the right-hand side induced from $\mathfrak{o}{\gamma_1} \otimes \mathfrak{o}{\gamma_2} $a g r ees w i t h$ \mathfrak{o}{\gamma_1 #_\rho \gamma_2}$.

When $μ (p) - μ (q) = 1$ the moduli $M (p, q)$ is a finite set of points, each an isolated flow line $u$ . A coherent orientation $o$ assigns to such a $u$ a characteristic sign $$ n(u) = n_{\mathfrak{o}}(u) \in {+1, -1}, $$ defined by comparing $o_{u}$ — an orientation of the zero-dimensional $M (p, q)$ at $u$ , hence a $\pm$ — with the standard orientation of a point. The signed count is the integer $$ n(p,q) = \sum_{u \in \widehat{\mathcal{M}}(p,q)} n(u) ;\in; \mathbb{Z}. $$

Counterexamples to common slips

The determinant line is not the line $Λ^{m a x} ker D$ alone. When $D$ is not surjective the cokernel contributes the dual factor $(Λ^{m a x} coker D)^{*}$ ; dropping it makes the line discontinuous at the loci where $dim ker$ jumps. Surjectivity, granted here by Morse-Smale, is what lets one work with $Λ^{m a x} ker$ alone.
A coherent orientation is not a global orientation of any single moduli space. It is a family of orientations across all pairs $(p, q)$ tied together by gluing; there is no requirement (and in general no possibility) that the disjoint union $⨆ M (p, q)$ be orientable as one space.
The characteristic sign $n (u)$ depends on the chosen coherent orientation, and the absolute signs $n (u)$ can all flip when the orientation convention is changed. What is convention-independent is the chain isomorphism class of the resulting complex; the signs are well-defined only relative to a fixed coherent system.
Coherence is a condition on catenation, not on concatenation of arbitrary paths. The gluing isomorphism is available only for trajectories sharing an endpoint critical point $z$ with the index dropping by one at each step; it is not a tensor structure on all of path space.

Key theorem with proof Intermediate+

The existence of a coherent orientation is the central structural statement of Schwarz Part II Chapter 3 ^{[Schwarz Part II Ch. 3]}. It is what upgrades the $Z /2$ count of flow lines to a genuine $Z$ -valued count.

Theorem (existence and ambiguity of coherent orientations). Let $(f, g)$ be a Morse-Smale pair on a closed Riemannian manifold $(M^{n}, g)$ . Then a coherent orientation $o$ exists. Any two coherent orientations differ by a function $ε : Crit (f) \to {\pm 1}$ in the following sense: there is a choice of sign $ε (x)$ at each critical point so that the two characteristic signs are related by $$ n'(u) = \varepsilon(p),\varepsilon(q),n(u), \qquad u \in \widehat{\mathcal{M}}(p,q),\ \mu(p)-\mu(q)=1 . $$ Consequently the signed counts transform as $n^{'} (p, q) = ε (p) ε (q) n (p, q)$ , and the two resulting boundary operators are conjugate by the diagonal sign-change $x \mapsto ε (x) x$ .

Proof. The construction is by choosing orientations at the critical points and propagating. To each critical point $x$ associate, once and for all, an orientation $o_{x}$ of the determinant line $det (D_{x})$ of the based operator $D_{x} = \frac{d}{d t} + A_{x}$ on the half-line, where $A_{x} = Hess f (x)$ is the constant Hessian; this operator has index $μ (x)$ relative to the chosen boundary condition, and its determinant line is finite-dimensional. There are two choices at each $x$ .

Given such basepoint orientations, orient $det (D_{γ} F)$ for every $γ \in M (p, q)$ as follows. The asymptotically-constant operator $D_{γ} F$ on $R$ converges at $- \infty$ to $A_{p}$ and at $+ \infty$ to $A_{q}$ . The catenation isomorphism for determinant lines — a continuous, associative isomorphism $det (D^{'}) \otimes det (D^{''}) ≅ det (D^{'} # D^{''})$ natural in homotopies of the operators, established for asymptotically-constant first-order operators by the standard linear-gluing argument ^{[Floer-Hofer 1993]} — gives a canonical isomorphism $$ \det(D_p) ;\cong; \det(D_\gamma F) \otimes \det(D_q), $$ obtained by gluing $det (D_{γ} F)$ to the based operator at $q$ and comparing with the based operator at $p$ (both ends contribute, and the index adds: $μ (p) = (μ (p) - μ (q)) + μ (q)$ ). Solving for $det (D_{γ} F)$ and inserting the chosen orientations $o_{p}$ , $o_{q}$ determines an orientation $o_{γ}$ . Because the catenation isomorphism is natural under continuous deformation, $o_{γ}$ varies continuously in $γ$ ; because catenation is associative, the orientation of a glued trajectory $γ_{1} #_{ρ} γ_{2}$ obtained from $o_{γ_{1}}$ and $o_{γ_{2}}$ agrees with the one assigned directly. So $o = {o_{γ}}$ is coherent. This proves existence.

For the ambiguity, let $o$ and $o^{'}$ be two coherent orientations. They arise from two systems of basepoint orientations ${o_{x}}$ and ${o_{x}^{'}}$ ; record their discrepancy by $ε (x) = + 1$ if $o_{x}^{'} = o_{x}$ and $- 1$ otherwise. From the displayed catenation isomorphism, $o_{γ}$ depends multiplicatively on the basepoint orientations at the two ends $p, q$ : replacing $o_{p}$ by $ε (p) o_{p}$ and $o_{q}$ by $ε (q) o_{q}$ multiplies $o_{γ}$ by $ε (p) ε (q)$ . For $μ (p) - μ (q) = 1$ this multiplies the characteristic sign, giving $n^{'} (u) = ε (p) ε (q) n (u)$ , hence $n^{'} (p, q) = ε (p) ε (q) n (p, q)$ . The induced boundary operators $\partial$ , $\partial^{'}$ then satisfy $\partial^{'} = T \partial T^{- 1}$ with $T (x) = ε (x) x$ , so they are conjugate and define isomorphic complexes. $□$

Bridge. This existence theorem builds toward 03.15.06, where the signed counts $n (p, q)$ are exactly the matrix entries of the boundary operator $\partial$ . The foundational reason the sign rule had to be coherent rather than arbitrary is that $\partial^{2} = 0$ counts the boundary points of the compactified one-dimensional moduli $\overline{M} (p, r)$ with $μ (p) - μ (r) = 2$ , and the two ends of each boundary arc must receive opposite signs; this is exactly what coherence with the gluing isomorphism guarantees. The catenation law $det (D^{'}) \otimes det (D^{''}) ≅ det (D^{'} # D^{''})$ used here is dual to the breaking of trajectories proved in 03.15.03: breaking degenerates a trajectory into a catenation, and the determinant line of the limit is the tensor product of the determinant lines of the pieces, so orientations chosen coherently upstairs descend compatibly to the broken stratum. The ambiguity statement — coherent orientations form a torsor under sign-changes at critical points — appears again in 03.07.22, where the same determinant-line construction orients instanton moduli; putting these together, the finite-dimensional sign bookkeeping of this unit is the literal template for the orientation theory of every Floer homology, and the bridge is the catenation isomorphism of determinant lines, which is identical in form in both settings.

Exercises Intermediate+

Exercise 3 (medium, symbolic). Under a change of coherent orientation recorded by $\varepsilon : \mathrm{Crit}(f) \to \{\pm 1\}$, the signed count transforms as $n'(p,q) = \varepsilon(p)\varepsilon(q)\,n(p,q)$. Show that the quantity $n(p,q)\,n(q,r)\,n(p,r)$ (when all three index drops are one) is unchanged if $\mu(p)-\mu(r)=1$ would be needed — instead consider the product $n(p,q)\,n(q,r)$ for a two-step chain $p \to q \to r$ and determine its transformation factor.

Hint

Multiply the two transformation factors and look for which signs appear twice.

Answer

Factor $ε (p) ε (r)$ . One has $n^{'} (p, q) n^{'} (q, r) = ε (p) ε (q) n (p, q) \cdot ε (q) ε (r) n (q, r) = ε (p) ε (q)^{2} ε (r) n (p, q) n (q, r)$ . Since $ε (q)^{2} = 1$ , the intermediate sign cancels and the product transforms by $ε (p) ε (r)$ — the same factor as a single hop $p \to r$ . This telescoping of the interior signs is the algebraic shadow of coherence and is what makes $\partial^{2}$ well-defined up to conjugation.

Exercise 4 (medium, numeric). For a Morse-Smale pair, $\mu(p)-\mu(q)=2$, and $\overline{\widehat{\mathcal{M}}}(p,q)$ is a compact $1$-manifold with boundary whose boundary is four once-broken trajectories. For the boundary count of a compact $1$-manifold to vanish in the signed sense, how must the four boundary points pair up by sign?

Hint

The signed boundary of a compact $1$ -manifold is zero; each arc contributes two endpoints of opposite sign.

Answer

Two $+$ and two $-$ , paired by arcs. A compact $1$ -manifold with boundary is a finite union of circles and closed arcs; only arcs have boundary, and each arc has two endpoints carrying opposite induced orientations, hence one $+$ and one $-$ . Four boundary points form two arcs, giving two $+$ and two $-$ , and the signed total is $0$ . This is precisely the cancellation that yields $\partial^{2} = 0$ .

Exercise 5 (medium, short-answer). Explain why, when the linearized operator $D_\gamma F$ is surjective, the determinant line $\det(D_\gamma F)$ is canonically $\Lambda^{\max} T_\gamma\mathcal{M}(p,q)$, and why orienting it is the same as orienting the moduli at $\gamma$.

Hint

Surjectivity makes the cokernel vanish; the kernel is the tangent space to the zero set.

Answer

Rubric. Full credit: surjectivity gives $coker D_{γ} F = 0$ , so the dual factor $(Λ^{m a x} coker)^{*} = Λ^{0} \cdot^{*} = R$ is canonically the identity line, and $det (D_{γ} F) = Λ^{m a x} ker D_{γ} F$ . By the implicit-function theorem in the Banach setting (transversality), $M (p, q)$ is a manifold near $γ$ with tangent space $ker D_{γ} F$ . Hence $det (D_{γ} F) = Λ^{m a x} T_{γ} M (p, q)$ , and a ray in this top exterior power is exactly an orientation of the tangent space, i.e. a local orientation of the moduli. Quotienting by the time-translation direction $\partial_{t} γ$ transfers this to an orientation of $M (p, q)$ .

Exercise 6 (hard, short-answer). State the gluing (catenation) isomorphism for determinant lines and explain precisely what it means for a system of orientations to be coherent with respect to it. Then explain why coherence is exactly the hypothesis that makes the two boundary points of an arc in $\overline{\widehat{\mathcal{M}}}(p,q)$ ($\mu(p)-\mu(q)=2$) carry opposite signs.

Hint

Each boundary point is a catenation $γ_{1} # γ_{2}$ ; the orientation of the arc near it is induced from $o_{γ_{1}} \otimes o_{γ_{2}}$ via the isomorphism.

Answer

Rubric. The catenation isomorphism is a continuous, associative family of isomorphisms $det (D_{γ_{1}} F) \otimes det (D_{γ_{2}} F) ≅ det (D_{γ_{1} #_{ρ} γ_{2}} F)$ for $γ_{1} \in M (p, z)$ , $γ_{2} \in M (z, q)$ , natural under homotopy of the operators. A system ${o_{γ}}$ is coherent if for every such pair the orientation $o_{γ_{1}} \otimes o_{γ_{2}}$ maps to $o_{γ_{1} #_{ρ} γ_{2}}$ under the isomorphism. Near a once-broken trajectory $(γ_{1}, γ_{2})$ on the boundary of an arc in $\overline{M} (p, q)$ , the glued trajectories $γ_{1} #_{ρ} γ_{2}$ for large $ρ$ sweep out a half-open collar; the arc's boundary orientation at this endpoint is the one induced from $o_{γ_{1}} \otimes o_{γ_{2}}$ . The two endpoints of one arc are joined through the interior, so the manifold orientation forces their induced boundary orientations to be opposite; coherence is what makes the sign at each endpoint computable from the determinant-line tensor product and therefore makes "opposite" an honest statement about the product signs $n (γ_{1}) n (γ_{2})$ . Full credit requires both the precise isomorphism and the collar/opposite-orientation argument.

Exercise 7 (hard, short-answer). The determinant line of a Fredholm operator extends continuously across the loci where $\dim\ker$ jumps, even though $\Lambda^{\max}\ker$ alone does not. Explain why, using a path of operators where a one-dimensional kernel and a one-dimensional cokernel are created together.

Hint

Index is constant; a kernel direction and a cokernel direction appear simultaneously, and the dual cokernel factor compensates the new kernel factor.

Answer

Rubric. Consider a family $D_{s}$ of index- $0$ operators, invertible for $s \neq = 0$ , with $dim ker D_{0} = dim coker D_{0} = 1$ . For $s \neq = 0$ , $det (D_{s}) = Λ^{0} ker \otimes (Λ^{0} coker)^{*} = R \otimes R = R$ , with a canonical generator $1 \otimes 1$ . At $s = 0$ a kernel line $ker D_{0} = ⟨ k ⟩$ and a cokernel line $coker D_{0} = ⟨ c ⟩$ appear, so $det (D_{0}) = ⟨ k ⟩ \otimes ⟨ c^{*} ⟩$ . The continuous local framing is supplied by stabilisation: enlarge the operator by a small isomorphism between the emerging kernel and cokernel directions; as $s \to 0$ the generator $1 \otimes 1$ limits to $k \otimes c^{*}$ up to the chosen stabilisation, giving a continuous nonvanishing section of $det$ across $s = 0$ . The dual placement of the cokernel factor is exactly what makes the new kernel factor cancel in the limit, so the line stays one-dimensional and oriented continuously even as $Λ^{m a x} ker$ jumps from $R$ to $⟨ k ⟩$ . Full credit: the simultaneous creation of $ker$ and $coker$ and the cancelling dual factor.

Lean formalization Intermediate+

Mathlib carries finite-dimensional exterior algebra, the determinant of a linear endomorphism, and the Orientation of a finite-dimensional real space, but no determinant line of a Fredholm operator and no gluing law for it, so this unit ships at lean_status: none. The pseudo-Lean below indicates the intended shape; it does not compile because Fredholm, detLine, catenation, and CoherentOrientation are not defined in Mathlib.

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

-- The determinant line of a real Fredholm operator.
noncomputable def detLine {E F : Type*} (D : E →L[ℝ] F) [Fredholm D] : Module ℝ :=
  (Λtop (LinearMap.ker D)) ⊗[ℝ] (Λtop (LinearMap.range D)ᗮ)ᵈ   -- coker via complement

-- The catenation isomorphism (the algebraic engine of coherence).
theorem det_catenation {E F} (D₁ D₂ : E →L[ℝ] F) [Fredholm D₁] [Fredholm D₂]
    (ρ : ℝ) :
    detLine D₁ ⊗[ℝ] detLine D₂ ≃ₗ[ℝ] detLine (catenate D₁ D₂ ρ) := by sorry

-- A coherent orientation: a section of detLine over the moduli, gluing-invariant.
structure CoherentOrientation (g) (f) where
  orient   : ∀ p q (γ : trajectorySpace g f p q), Orientation ℝ (detLine (linF γ)) PUnit
  coherent : ∀ p z q γ₁ γ₂ ρ,
    (det_catenation _ _ ρ) (orient p z γ₁ ⊗ orient z q γ₂) = orient p q (glue γ₁ γ₂ ρ)

-- The characteristic sign of an isolated flow line.
noncomputable def charSign (o : CoherentOrientation g f) (p q : M)
    (u : unparamTrajectorySpace g f p q) (h : μ f p - μ f q = 1) : ℤ := by sorry

The finite-dimensional linear algebra of $Λ^{m a x} ker$ , the dual cokernel factor, and the tensor isomorphisms reduces to Mathlib's Module.Det, Orientation, and TensorProduct API once a Fredholm-operator structure with a continuously-varying determinant line is in place; the named gap in Mathlib gap analysis is exactly that determinant-line bundle and its catenation isomorphism.

Advanced results Master

The coherent-orientation construction refines into a precise count of the choices, a clean description of how the boundary operator transforms, and a determinant-line account of the gluing sign that drives $\partial^{2} = 0$ .

The torsor of coherent orientations. Fix a Morse-Smale pair with $r$ critical points. A coherent orientation is determined by a choice of orientation of the based determinant line $det (D_{x})$ at each critical point $x$ , two choices each, so the set of coherent orientations carries a free transitive action of $(Z /2)^{r}$ , the group of functions $ε : Crit (f) \to {\pm 1}$ . The induced action on boundary operators factors through conjugation by the diagonal matrix $diag (ε (x))$ , so it acts as the identity on the isomorphism class of the complex and on its homology. The orientation data is a genuine choice, but a choice without homological consequence — the finite-dimensional model of the statement that Floer homology with $Z$ -coefficients is well-defined up to canonical isomorphism once a coherent orientation is fixed.

The determinant line of the based operator and spectral flow. The based operator $D_{x} = \frac{d}{d t} + A_{x}$ with constant $A_{x} = Hess f (x)$ has, with the standard positive/negative spectral splitting boundary condition, kernel dimension equal to the number of negative eigenvalues of $A_{x}$ , namely $μ (x)$ . So $det (D_{x}) ≅ Λ^{μ (x)} E_{x}^{-}$ , where $E_{x}^{-}$ is the negative eigenspace of the Hessian — the tangent space to the unstable manifold $W^{u} (x)$ . An orientation of $det (D_{x})$ is therefore an orientation of $W^{u} (x)$ , and the coherent-orientation data is, equivalently, a choice of orientation of every unstable manifold. This is the concrete differential-topological picture behind the abstract determinant-line formalism: the sign $n (u)$ compares the orientation of $W^{u} (p)$ , transported along $u$ and intersected with $W^{s} (q)$ , against the orientation of $W^{u} (q)$ .

The gluing sign and $\partial^{2} = 0$ . For $μ (p) - μ (r) = 2$ , each interior arc of $\overline{M} (p, r)$ runs between two once-broken trajectories $(γ_{1}, γ_{2})$ and $(γ_{1}^{'}, γ_{2}^{'})$ . The catenation isomorphism computes the induced boundary orientation at each end as $n (γ_{1}) n (γ_{2})$ times a universal gluing sign $σ$ that depends only on the indices $μ (p), μ (z), μ (r)$ — not on the trajectories. Because the two ends of one arc receive opposite manifold-boundary orientations, the contributions $n (γ_{1}) n (γ_{2})$ and $n (γ_{1}^{'}) n (γ_{2}^{'})$ enter $\partial^{2}$ with opposite signs and cancel. Summing over all arcs, the coefficient of $r$ in $\partial^{2} p$ is $\sum_{z} \sum_{u \in M (p, z)} \sum_{v \in M (z, r)} n (u) n (v)$ , which equals the signed boundary count of the compact $1$ -manifold $\overline{M} (p, r)$ , hence $0$ . The universal sign $σ$ is absorbed once and for all into the grading convention; Floer-Hofer ^{[Floer-Hofer 1993]} isolate it as the gluing sign of the determinant-line catenation.

Relation to the orientation of intersection of $W^{u}$ and $W^{s}$ . With unstable manifolds oriented, the moduli $M (p, q) = W^{u} (p) ⋔ W^{s} (q)$ inherits an orientation from the orientations of $W^{u} (p)$ , $W^{u} (q)$ (equivalently of $W^{s} (q)$ via a co-orientation), and the ambient $M$ if $M$ is oriented; when $M$ is not oriented the determinant-line route still works, which is why Schwarz's formalism is stated for determinant lines rather than for oriented intersections. The two viewpoints agree where both apply, and the determinant-line one is the genuinely general construction ^{[Schwarz Part II Ch. 3]}.

Synthesis. The determinant line $det (D_{γ} F)$ is the foundational object: orienting it is exactly orienting the moduli, and the catenation isomorphism $det (D^{'}) \otimes det (D^{''}) ≅ det (D^{'} # D^{''})$ is dual to the breaking of 03.15.03, so a coherent orientation is precisely a system of moduli orientations stable under degeneration. This is exactly the structure that forces the boundary points of each arc of $\overline{M} (p, r)$ to cancel in signed pairs, which is the central insight making $\partial^{2} = 0$ hold over $Z$ rather than only over $Z /2$ . Putting these together, the torsor of coherent orientations under $(Z /2)^{Crit (f)}$ acts by conjugation and so leaves the homology canonically defined, which generalises verbatim to Floer theory: replacing the based operator $\frac{d}{d t} + Hess f (x)$ by the asymptotic operator of the Cauchy-Riemann or anti-self-duality equation gives the coherent orientations of 03.07.22 and of 05.08.02, and the bridge is the determinant-line catenation law, which has the same form in every dimension. The orientations built here, the gluing of 03.15.04 that supplies the collars, and the compactness of 03.15.03 that supplies the arcs are the three inputs 03.15.06 assembles into the integral boundary operator.

Full proof set Master

The existence-and-ambiguity theorem is proved in full in the Key theorem section. The supporting structural propositions are recorded here.

Proposition (determinant line is one-dimensional and continuous). For a real Fredholm operator $D : E \to F$ , the line $det (D) = Λ^{m a x} ker D \otimes (Λ^{m a x} coker D)^{*}$ is one-dimensional, and the lines fit into a continuous real line bundle over $Fred (E, F)$ , locally framed across kernel jumps by stabilisation.

Proof. For finite-dimensional $ker D$ and $coker D$ , $Λ^{m a x}$ of each is one-dimensional, so the tensor product of a line with the dual of a line is one-dimensional. For continuity, fix $D_{0}$ and a finite-dimensional subspace $W \subset F$ with $W + range D_{0} = F$ . For $D$ near $D_{0}$ , $W + range D = F$ as well, and the stabilised operator $D = D \oplus incl : E \oplus W \to F$ is surjective with $ker D$ of constant rank $index D + dim W$ . There is a canonical isomorphism $det (D) ≅ Λ^{m a x} ker D \otimes (Λ^{m a x} W)^{*}$ obtained from the exact sequence $0 \to ker D \to ker D \to W \to coker D \to 0$ , and $Λ^{m a x} ker D$ is a continuous line bundle because $ker D$ has constant rank near $D_{0}$ . This frames $det$ locally, so the lines form a continuous line bundle. $■$

Proposition (based determinant line is the orientation of the unstable manifold). For a nondegenerate critical point $x$ , the based operator $D_{x} = \frac{d}{d t} + A_{x}$ with $A_{x} = Hess f (x)$ and the spectral boundary condition has $det (D_{x}) ≅ Λ^{μ (x)} E_{x}^{-} = Λ^{m a x} T_{x} W^{u} (x)$ .

Proof. The operator $\frac{d}{d t} + A_{x}$ on the half-line with the boundary condition selecting decay toward $+ \infty$ has kernel consisting of solutions $ξ (t)$ with $\dot{ξ} = - A_{x} ξ$ that decay as $t \to + \infty$ ; these are the components of $ξ (0)$ in the negative eigenspaces of $A_{x}$ , since $ξ (t) = e^{- t A_{x}} ξ (0)$ decays exactly when $ξ (0) \in E_{x}^{-}$ , the sum of eigenspaces with negative eigenvalue. The cokernel vanishes for this boundary condition by the standard asymptotic-operator analysis. Hence $ker D_{x} ≅ E_{x}^{-}$ and $det (D_{x}) = Λ^{m a x} E_{x}^{-}$ . Finally $dim E_{x}^{-} = μ (x)$ by definition of the Morse index, and $E_{x}^{-} = T_{x} W^{u} (x)$ because the unstable manifold is tangent to the negative eigenspace of the Hessian at a hyperbolic rest point. $■$

Proposition (well-definedness of $n (p, q)$ over $Z$ ). Fix a coherent orientation. For $μ (p) - μ (q) = 1$ the set $M (p, q)$ is finite, each point carries a well-defined sign $n (u) \in {\pm 1}$ , and $n (p, q) = \sum_{u} n (u) \in Z$ is independent of the trajectory enumeration and continuous in the Morse-Smale pair within a connected family.

Proof. Finiteness: $M (p, q)$ has dimension $μ (p) - μ (q) - 1 = 0$ by transversality 03.15.02, and it is compact because the compactification 03.15.03 adds only broken trajectories, which need an intermediate critical point of index strictly between $μ (p)$ and $μ (q)$ — impossible when the indices differ by one. A compact $0$ -manifold is finite. The sign $n (u)$ is the comparison of the coherent orientation $o_{u}$ of the zero-dimensional moduli at $u$ with the standard point orientation, hence a well-defined element of ${\pm 1}$ . The sum over the finite set is a well-defined integer, manifestly independent of the order of summation. For a smooth path of Morse-Smale pairs along which no flow line is born or dies (no index-difference-zero or wall-crossing event), the finite set $M (p, q)$ and each $o_{u}$ vary continuously, so each $n (u)$ is locally constant and $n (p, q)$ is constant. $■$

The three propositions are proved here in full; the universal gluing sign $σ$ entering $\partial^{2} = 0$ is computed from the catenation isomorphism, whose construction for asymptotically-constant operators is in Floer-Hofer ^{[Floer-Hofer 1993]} and Schwarz ^{[Schwarz Part II Ch. 3]}.

Connections Master

03.15.02 (trajectory spaces, the Fredholm setup, and transversality) supplies the linearized operator $D_{γ} F$ whose determinant line carries the orientation. Transversality there makes $D_{γ} F$ surjective, so the determinant line is the top exterior power of the tangent space to the moduli, and the Fredholm index $μ (p) - μ (q)$ is the dimension that determines when isolated signed points occur. Without the Fredholm framing there is no operator to orient.

03.15.03 (compactness, broken trajectories) is dual to this unit through the catenation isomorphism. Breaking degenerates a trajectory into a catenation $γ_{1} # γ_{2}$ , and the determinant line of the limit is the tensor product of the determinant lines of the pieces; coherence is exactly the demand that the chosen orientations respect this tensor decomposition, so that the signs of the broken stratum are computed from the signs of the pieces.

03.15.06 (the Morse complex and $\partial^{2} = 0$ ) is the apex this unit feeds: the signed counts $n (p, q)$ are the matrix entries of the boundary operator $\partial x = \sum_{μ (y) = μ (x) - 1} n (x, y) y$ , and coherence is precisely what makes the once-broken trajectories on the boundary of each compactified arc cancel in signed pairs, giving $\partial^{2} = 0$ over $Z$ rather than only over $Z /2$ . The whole purpose of orienting the determinant lines is to make this integral boundary operator well-defined.

03.07.22 (orientations on instanton trajectory moduli) is the infinite-dimensional analogue: the same determinant-line construction, with the asymptotic operator of the anti-self-duality equation in place of $\frac{d}{d t} + Hess f$ , orients the instanton moduli and produces the signs of the instanton Floer differential. Schwarz's finite-dimensional orientation theory was designed as the rehearsal for exactly this gauge-theoretic construction of Donaldson and Floer-Hofer.

05.08.02 (Floer homology) inherits this orientation theory verbatim: coherent orientations of the determinant lines of the Cauchy-Riemann operators along Floer trajectories give the $Z$ -coefficient symplectic Floer differential, and the torsor-under-sign-changes ambiguity is identical to the one proved here.

Historical & philosophical context Master

The use of an orientation of a determinant line to fix the sign of a count of solutions to an elliptic equation was introduced into gauge theory by Simon Donaldson, in The orientation of Yang-Mills moduli spaces and 4-manifold topology (Journal of Differential Geometry 26, 1987, 397–428), where the determinant line of the anti-self-duality deformation operator is shown to be orientable and an orientation is used to define the integer-valued Donaldson invariants. The systematic algebraic formalism — determinant lines of Fredholm operators, the catenation isomorphism, and the notion of a coherent orientation as a gluing-invariant system — was developed by Andreas Floer and Helmut Hofer in Coherent orientations for periodic orbit problems in symplectic geometry (Mathematische Zeitschrift 212, 1993, 13–38), in the infinite-dimensional Floer setting.

The first self-contained finite-dimensional account, where the trajectory moduli are honest finite-dimensional manifolds and the determinant-line orientation reduces to an orientation of the unstable manifolds, 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 II Chapter 3 carries out the coherent-orientation construction as the finite-dimensional template for the Floer-Hofer theory, and it is this template that makes the integral Morse complex available; the expository synthesis appears in Salamon's IAS/Park City Lectures on Floer homology ^{[Salamon 1999]}. Earlier expositions of the Morse complex over $Z /2$ avoided the orientation question entirely, which is why the integral refinement is credited to this determinant-line lineage.

Bibliography Master

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

@article{FloerHofer1993,
  author  = {Floer, Andreas and Hofer, Helmut},
  title   = {Coherent orientations for periodic orbit problems in symplectic geometry},
  journal = {Mathematische Zeitschrift},
  volume  = {212},
  number  = {1},
  pages   = {13--38},
  year    = {1993}
}

@article{Donaldson1987,
  author  = {Donaldson, Simon K.},
  title   = {The orientation of {Y}ang-{M}ills moduli spaces and 4-manifold topology},
  journal = {Journal of Differential Geometry},
  volume  = {26},
  number  = {3},
  pages   = {397--428},
  year    = {1987}
}

@incollection{Salamon1999,
  author    = {Salamon, Dietmar},
  title     = {Lectures on {F}loer homology},
  booktitle = {Symplectic Geometry and Topology (Park City, UT, 1997)},
  series    = {IAS/Park City Mathematics Series},
  volume    = {7},
  pages     = {143--229},
  publisher = {American Mathematical Society, Providence, RI},
  year      = {1999}
}

@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

03.15.02
03.15.03

Tier anchors

beginner: Audin-Damian *Morse Theory and Floer Homology* §3.4 (signing the flow lines); Banyaga-Hurtubise *Lectures on Morse Homology* Ch. 5 (orientations, informal)
intermediate: Schwarz *Morse Homology* Part II Ch. 3; Audin-Damian §3.4; Banyaga-Hurtubise Ch. 5
master: Schwarz *Morse Homology* Part II Ch. 3; Floer-Hofer 1993; Donaldson 1987

References

03-modern-geometry/Bott-LecturesOnMorseTheory.pdf · Gradient flows and orientation of stable/unstable manifolds (background)
Schwarz — Morse Homology (Progress in Mathematics 111, Birkhäuser 1993) · Part II, Ch. 3 (orientations of the trajectory spaces and the characteristic signs)
Floer-Hofer — Coherent orientations for periodic orbit problems in symplectic geometry · Mathematische Zeitschrift 212 (1993), pp. 13-38 (the coherent-orientation formalism via determinant lines)
Donaldson — The orientation of Yang-Mills moduli spaces and 4-manifold topology · Journal of Differential Geometry 26 (1987), pp. 397-428 (determinant-line orientations of gauge-theoretic moduli)
Salamon — Lectures on Floer homology · Park City / IAS Lecture Notes (1999), §3 (orientations and signs in the Floer/Morse complex)

Estimated time

beginner: 17m
intermediate: 44m
master: 80m