03.15.09 · modern-geometry / morse-homology

Morse cohomology, cup product, and the ring structure

shipped3 tiersLean: none

Anchor (Master): Schwarz *Morse Homology* Part III; Fukaya 1993; Betz-Cohen 1994; Cohen-Norbury 2009

Intuition Beginner

You already have a counting machine that turns a height function on a curvy shape into an algebraic gadget: resting points sorted by how many downhill directions they have, wired together by signed counts of the streams between them. Reading that wiring downhill gives homology — the holes of the shape. This unit asks what happens when you read the very same streams in the opposite direction, uphill, and what new operation appears once you do.

Reading uphill repackages the same data as cohomology. Nothing about the shape changes; you have only swapped which end of each stream you call the start. The surprise is that cohomology carries an extra gift homology does not: a way to multiply two classes together and get a third. That multiplication is the cup product, and it turns the whole collection of cohomology classes into a ring — a system where you can add and also multiply.

Where does a multiplication come from, geometrically? Picture three streams meeting at a single junction, like the letter Y. Two of them arrive from resting points of one landscape; the third leaves toward a resting point of another. Counting how many such Y-junctions you can form, with signs, is the rule that multiplies the first two classes to produce the third. The number of rigid Y-shaped meetings is the product's recipe.

Visual Beginner

A schematic of a $Y$ -shaped flow graph drawn on a shaded surface. Two incoming gradient streams arrive at a central junction point from two upper resting points labelled $x_{1}$ and $x_{2}$ ; one outgoing stream leaves the junction and descends to a lower resting point labelled $y$ . Small arrows show the downhill direction along each of the three edges, and the junction is marked as the single shared meeting point where all three streams agree.

The picture captures the whole idea of the product: a class for $x_{1}$ times a class for $x_{2}$ contributes to the class for $y$ whenever a rigid $Y$ -junction connects them. Counting these junctions, with signs, fills in the multiplication table of the cohomology ring.

Worked example Beginner

Take the round two-sphere with the height function whose resting points are the south pole $p$ (a bottom, zero downhill directions) and the north pole $q$ (a top, two downhill directions). Compute its cohomology and the one interesting product.

Step 1. List the resting points by downhill count. The south pole $p$ has index $0$ ; the north pole $q$ has index $2$ . There is no index- $1$ resting point. So the cohomology has one class in degree $0$ , none in degree $1$ , and one class in degree $2$ .

Step 2. Name the classes. Call the degree- $0$ class $1$ (it is the unit, the all-of-the-sphere class) and the degree- $2$ class $a$ (it records the single top). Reading the streams uphill confirms there is nothing linking these two across a single step, so both survive into cohomology.

Step 3. Multiply. The product $1 \times a$ is just $a$ , because multiplying by the unit changes nothing. The only product left to check is $a \times a$ . Its degree would be $2 + 2 = 4$ , but the sphere is two-dimensional and has no degree- $4$ class. So $a \times a = 0$ .

Step 4. Read off the ring. The cohomology ring of the two-sphere is generated by one element $a$ in degree $2$ with $a \times a = 0$ . Written compactly, it is the system ${0, 1, a, and their sums}$ with the single rule $a^{2} = 0$ .

What this tells us: the same streams that compute the holes of the sphere also, when read as $Y$ -junctions, fill in a multiplication table. For the sphere the table is short — one nonzero generator that squares to zero — but it already matches the cup-product ring of the sphere computed by every other method.

Check your understanding Beginner

Exercise (easy, multiple choice).

Morse cohomology is built from the Morse complex by doing what?

A. Counting entirely different streams
B. Reading the same gradient streams in the reversed (uphill) direction and dualising
C. Throwing away the resting points of odd index
D. Replacing the manifold with a different one

Hint

Cohomology uses the same flow data as homology; only the direction of reading and the duality change.

Answer

B. Morse cohomology dualises the Morse complex: the cochains are the dual of the chains, and the coboundary counts the same gradient streams read in the upward direction. The resting points and streams are unchanged.

Feedback (correct): Right — the geometry is identical; only the bookkeeping flips.

Feedback (wrong): The flow data stays the same; what changes is reading it uphill and taking the algebraic dual.

Formal definition Intermediate+

Fix a closed Riemannian $n$ -manifold $M$ and a Morse-Smale pair $(f, g)$ , with Morse chain complex $(C_{*} (f), \partial)$ from 03.15.06: $C_{k} (f) = ⨁_{μ (x) = k} Z ⟨ x ⟩$ free abelian on the index- $k$ critical points, and $\partial ⟨ x ⟩ = \sum_{μ (y) = k - 1} n (x, y) ⟨ y ⟩$ the coherent signed count of isolated negative-gradient trajectories.

The Morse cochain complex is the degreewise dual $$ C^k(f) = \operatorname{Hom}{\mathbb{Z}}(C_k(f), \mathbb{Z}), \qquad \delta = \partial^{\vee} : C^k(f) \to C^{k+1}(f), $$ with $⟨ δ ξ, x ⟩ = ⟨ ξ, \partial x ⟩$ . Since $\partial^{2} = 0$ one gets $δ^{2} = 0$ , and Morse cohomology is $H M^{k} (f, g) = ker δ / im δ$ in degree $k$ . Concretely $\delta\langle y\rangle^{\vee} = \sum{\mu(x) = \mu(y)+1} n(x,y),\langle x\rangle^{\vee} $: t h eco b o u n d a r y o f t h e d u a l g e n er a t or o f$ y $co u n t s t h e in d e x - r ai s in g t r aj ec t or i es * in t o w hi c h *$ y $f l o w s, t ha t i s, t h es am e t r aj ec t or i eso f$ \partial$ read upward.

Geometrically this upward reading is flow reversal. The negative-gradient flow of $- f$ exchanges stable and unstable manifolds, $W^{u} (x; f) = W^{s} (x; - f)$ and $W^{s} (x; f) = W^{u} (x; - f)$ , and the index of a critical point of $- f$ is $n - μ_{f} (x)$ . This identifies $C^{k} (f) ≅ C_{n - k} (- f)$ as complexes, the chain-level shape of Poincaré duality.

The cup product is defined by counting $Y$ -graphs. Choose three generic Morse-Smale pairs $(f_{1}, g_{1}), (f_{2}, g_{2}), (f_{3}, g_{3})$ on $M$ . A $Y$ -configuration from $(x_{1}, x_{2})$ to $y$ , where $μ_{f_{1}} (x_{1}) = k$ , $μ_{f_{2}} (x_{2}) = l$ , $μ_{f_{3}} (y) = k + l$ , is a point of the fibre product $$ \mathcal{Y}(x_1, x_2; y) = W^u(x_1; f_1) ,\cap, W^u(x_2; f_2) ,\cap, W^s(y; f_3), $$ three half-trajectories joined at a common point of $M$ . For generic data this is a smooth manifold of dimension $μ_{f_{1}} (x_{1}) + μ_{f_{2}} (x_{2}) - μ_{f_{3}} (y)$ . When that dimension is $0$ it is a finite set; coherent orientations 03.15.05 assign each point a sign, and the signed count is $n (x_{1}, x_{2}; y) \in Z$ . The product $$ \smile ,:, C^k(f_1) \otimes C^l(f_2) \to C^{k+l}(f_3), \qquad \langle x_1\rangle^{\vee} \smile \langle x_2\rangle^{\vee} = \sum_{\mu_{f_3}(y) = k+l} n(x_1, x_2; y),\langle y\rangle^{\vee} $$ is a cochain map for $δ$ , hence descends to $⌣: H M^{k} \otimes H M^{l} \to H M^{k + l}$ . The codimension count $n - (k + l) = (n - k) + (n - l) - n$ is exactly the transverse-intersection condition $W^{u} (x_{1}) \cap W^{u} (x_{2}) \cap W^{s} (y)$ being a point: three submanifolds of codimensions $k$ , $l$ , $n - (k + l)$ in $M^{n}$ meeting in dimension $0$ .

Counterexamples to common slips

The cup product needs three independent Morse functions, not one. Using a single $f$ for all three legs forces the unstable manifolds of $x_{1}$ and $x_{2}$ to belong to the same flow and generically fail to meet transversally; the fibre product is then not cut out cleanly, and the count is ill-defined. The remedy is to perturb to three generic pairs — continuation invariance 03.15.07 guarantees the answer is independent of the choices.
The coboundary $δ$ is not obtained by counting brand-new trajectories. It uses exactly the trajectories of $\partial$ , transposed. A reader who recomputes flow lines from scratch for $δ$ is double-counting; $δ = \partial^{\lor}$ is a transpose of an integer matrix.
Degree must be additive: $⟨ x_{1} ⟩^{\lor} ⌣ ⟨ x_{2} ⟩^{\lor}$ lands in degree $μ (x_{1}) + μ (x_{2})$ , never the maximum or the sum-minus-one. A product landing in the wrong degree signals a mis-set index condition in the fibre product.

Key theorem with proof Intermediate+

Theorem (the cup product is a well-defined associative graded-commutative product; Schwarz Part III, Fukaya 1993). For generic Morse-Smale data the $Y$ -graph count defines a product $⌣$ on $HM^(M) $t ha t i s in d e p e n d e n t o f t h ec h ose n d a t a u pt o t h eco n t in u a t i o ni so m or p hi s m s, g r a d e d - co mm u t a t i v e, a ssoc ia t i v e, an d u ni t a l w i t h u ni tt h eco h o m o l o g y c l a sso f t h e minim u m o f a se l f - in d e x in g M or se f u n c t i o n . U n d er t h e M or seH o m o l o g y T h eor e m co m p a r i so n$ HM^(M) \cong H^(M; \mathbb{Z}) $, t h e p r o d u c t$ \smile$ corresponds to the singular cup product, so the isomorphism is one of graded rings.*

Proof. Chain-level well-definedness. For generic $(f_{i}, g_{i})$ the fibre product $Y (x_{1}, x_{2}; y)$ is cut out as the transverse zero set of a Fredholm section over the $Y$ -shaped path space; its dimension is $μ (x_{1}) + μ (x_{2}) - μ (y)$ by the index formula, and in dimension $0$ it is compact, hence finite, because broken or split limits would lie in negative-dimensional moduli and are excluded. So $n (x_{1}, x_{2}; y)$ is a finite integer.

Cochain-map / Leibniz identity. Consider the one-dimensional fibre products, $μ (x_{1}) + μ (x_{2}) - μ (y) = 1$ . The compactified moduli is a compact oriented $1$ -manifold with boundary; gluing 03.15.04 identifies its ends with the two ways a $Y$ -graph can degenerate: an edge into $x_{1}$ or $x_{2}$ breaks (contributing $δ$ on a factor), or the outgoing edge to $y$ breaks (contributing $δ$ on the product). Signed boundary count zero yields $$ \delta(\xi_1 \smile \xi_2) = (\delta\xi_1) \smile \xi_2 + (-1)^{|\xi_1|}, \xi_1 \smile (\delta\xi_2), $$ the graded Leibniz rule, so $⌣$ is a cochain map and descends to cohomology.

Associativity. Compare the two bracketings $(ξ_{1} ⌣ ξ_{2}) ⌣ ξ_{3}$ and $ξ_{1} ⌣ (ξ_{2} ⌣ ξ_{3})$ by introducing the $H$ -graph (two trivalent vertices joined by an internal edge, four outer legs). The moduli of such graphs with a variable internal edge length $ℓ \in [0, \infty]$ is a one-parameter family interpolating the two bracketings: $ℓ \to 0$ collapses to one bracketing, $ℓ \to \infty$ stretches to the other. Its one-dimensional stratum is a compact oriented cobordism between the two composite operations, so they agree on cohomology.

Graded commutativity. Swapping the two incoming legs of a $Y$ -graph is an orientation-preserving or reversing automorphism of the fibre product according to the Koszul sign $(- 1)^{k l}$ ; comparing the swapped and unswapped counts on the cohomology level gives $ξ_{1} ⌣ ξ_{2} = (- 1)^{∣ ξ_{1} ∣∣ ξ_{2} ∣} ξ_{2} ⌣ ξ_{1}$ .

Unitality and the ring isomorphism. Take $f_{3}$ self-indexing with a unique minimum $m$ ; the class $[m]^{\lor} \in H M^{0}$ acts as a two-sided unit because a $Y$ -graph with one leg landing at the unique source $W^{s} (m) = M$ reduces to a single continuation trajectory. Finally, the Morse Homology Theorem provides a chain homotopy equivalence to the singular cochain complex intertwining $⌣$ with the Alexander-Whitney cup product, established by a $Y$ -graph-versus-front-face/back-face comparison; on cohomology this is a ring isomorphism $H M^{*} (M) ≅ H^{*} (M; Z)$ . $□$

Bridge. This product builds toward the entire multiplicative theory of manifold invariants, and the foundational reason it exists is exactly the codimension arithmetic that makes three submanifolds of codimensions $k$ , $l$ , $n - k - l$ meet in isolated points. The central insight is that counting $Y$ -graphs is dual to intersecting cycles: the Morse cup product is the flow-graph shadow of the intersection product, and this is exactly the geometry that the singular cup product packages combinatorially. Putting these together, flow reversal identifies $C^{*} (f)$ with $C_{n -*} (- f)$ , which is the chain-level Poincaré duality that turns the cup product into the intersection pairing; the construction generalises from $Y$ -graphs to arbitrary trees, where it appears again in the $A_{\infty}$ and Morse-field-theory structures. The bridge is the recognition that one trajectory-counting machine, fed graphs with more vertices, produces every higher product, and the simplest such graph — the $Y$ — already identifies $H M^{*}$ with the singular cohomology ring.

Exercises Intermediate+

Exercise 4 (medium, short-answer).

Explain why the cup product requires three independent Morse functions rather than one, in terms of transversality of the fibre product.

Hint

Two unstable manifolds of the same flow cannot be perturbed independently; transversality of $W^{u} (x_{1}) \cap W^{u} (x_{2}) \cap W^{s} (y)$ needs independent data.

Answer

The product is the signed count of points in $W^{u} (x_{1}; f_{1}) \cap W^{u} (x_{2}; f_{2}) \cap W^{s} (y; f_{3})$ . For this to be a transverse intersection (cut out cleanly with the expected dimension $μ (x_{1}) + μ (x_{2}) - μ (y)$ ) the three submanifolds must be independently perturbable. With one Morse function, $W^{u} (x_{1})$ and $W^{u} (x_{2})$ belong to the same gradient flow and their relative position is fixed; generically they fail to meet $W^{s} (y)$ transversally, and the count is ill-defined. Choosing three generic pairs restores transversality, and continuation invariance 03.15.07 shows the resulting product on cohomology is independent of the choices. Rubric: full credit for the transversality explanation plus the continuation-invariance remark.

Exercise 5 (medium, short-answer).

Derive the graded Leibniz rule $δ (ξ_{1} ⌣ ξ_{2}) = (δ ξ_{1}) ⌣ ξ_{2} + (- 1)^{∣ ξ_{1} ∣} ξ_{1} ⌣ (δ ξ_{2})$ from the boundary of the one-dimensional $Y$ -graph moduli.

Hint

A compact oriented $1$ -manifold has signed boundary zero; classify the ways a $Y$ -graph degenerates at the boundary.

Answer

Fix degrees with $μ (x_{1}) + μ (x_{2}) - μ (y) = 1$ , so the fibre product is a compact oriented $1$ -manifold with boundary. Its boundary points are the broken configurations: either an incoming edge breaks (an extra index-raising trajectory splits off the $x_{1}$ - or $x_{2}$ -leg, contributing $(δ ξ_{1}) ⌣ ξ_{2}$ or $ξ_{1} ⌣ (δ ξ_{2})$ with the Koszul sign $(- 1)^{∣ ξ_{1} ∣}$ from reordering), or the outgoing edge breaks (contributing $δ (ξ_{1} ⌣ ξ_{2})$ ). A compact oriented $1$ -manifold has total signed boundary zero, so the three terms sum to zero, which rearranges to the stated Leibniz rule. Rubric: full credit for identifying the three boundary types and the signed-boundary-zero argument.

Exercise 7 (hard, short-answer).

Prove that on a closed oriented $n$ -manifold the Morse cup-product pairing $H M^{k} (f) \otimes H M^{n - k} (f) \to H M^{n} (f) ≅ Z$ is non-degenerate over $Q$ , using flow reversal.

Hint

Combine $C^{*} (f) ≅ C_{n -*} (- f)$ with the Morse Homology Theorem and identify the pairing with the intersection form.

Answer

Flow reversal gives $C^{k} (f) ≅ C_{n - k} (- f)$ , so $H M^{k} (f) ≅ H M_{n - k} (- f) ≅ H M_{n - k} (M)$ after continuation invariance 03.15.07. The cup-product pairing $H M^{k} \otimes H M^{n - k} \to H M^{n} ≅ Z$ counts $Y$ -graphs landing at the top critical point; under the comparison isomorphism it equals the singular cup-product pairing $H^{k} (M; Q) \otimes H^{n - k} (M; Q) \to Q$ evaluated on the fundamental class, which is Poincaré-dual to the intersection form on $H_{n - k} \otimes H_{k}$ . Poincaré duality 03.12.16 makes that form non-degenerate, so the Morse pairing is non-degenerate over $Q$ . Rubric: full credit for the flow-reversal identification plus the Poincaré-duality non-degeneracy.

Exercise 8 (hard, short-answer).

Sketch why the associativity of $⌣$ follows from a one-parameter family of $H$ -graphs with variable internal edge length, and identify what the two endpoints $ℓ \to 0$ and $ℓ \to \infty$ contribute.

Hint

The internal edge length is a free parameter; collapsing or stretching it recovers the two bracketings of a triple product.

Answer

An $H$ -graph has two trivalent vertices joined by an internal edge of length $ℓ$ , with four outer legs (three inputs, one output). For fixed critical-point labels with total expected dimension $1$ , the moduli over $ℓ \in [0, \infty]$ is a compact oriented $1$ -manifold-with-boundary. At $ℓ \to 0$ the internal edge collapses to a four-valent vertex that resolves into one bracketing, say $(ξ_{1} ⌣ ξ_{2}) ⌣ ξ_{3}$ ; at $ℓ \to \infty$ the edge stretches and breaks, gluing 03.15.04 identifying the limit with the other bracketing $ξ_{1} ⌣ (ξ_{2} ⌣ ξ_{3})$ . The remaining boundary contributions involve a $δ$ on one factor and cancel in cohomology. Signed boundary count zero then equates the two bracketings, giving associativity. (The same edge-length family, kept track of at the chain level rather than passed to cohomology, is the first homotopy of the Morse $A_{\infty}$ structure.) Rubric: full credit for the cobordism argument plus the identification of the two endpoints.

Advanced results Master

Theorem (the dual complex and Morse cohomology; Schwarz Part III). For a Morse-Smale pair $(f, g)$ on a closed $M$ , the dual $C^(f) = \operatorname{Hom}{\mathbb{Z}}(C(f), \mathbb{Z}) $w i t h$ \delta = \partial^{\vee} $i s a coc hain co m pl e x w h oseco h o m o l o g y$ HM^(f,g) $i s, v ia co n t in u a t i o n, anin v a r ian t$ HM^(M) $o f$ M $a l o n e . T h e u ni v er s a l - coe f f i c i e n t se q u e n cer e l a t es i tt o$ HM_ $:$ 0 \to \operatorname{Ext}(HM_{k-1}, \mathbb{Z}) \to HM^k \to \operatorname{Hom}(HM_k, \mathbb{Z}) \to 0 $. * T h eco b o u n d a r y$ \delta\langle y\rangle^{\vee} = \sum_{\mu(x)=\mu(y)+1} n(x,y)\langle x\rangle^{\vee} $i s t h e t r an s p oseo f$ \partial$; no new analysis is needed beyond the differential of 03.15.06.

Theorem (flow reversal and Poincaré duality; Schwarz Part III, Milnor). Reversing the gradient flow exchanges stable and unstable manifolds, $W^{u} (x; f) = W^{s} (x; - f)$ , and sends the Morse index $μ_{f} (x)$ to $n - μ_{f} (x) = μ_{- f} (x)$ . The induced chain isomorphism $C_{k} (f) ≅ C^{n - k} (- f)$ intertwines $\partial_{f}$ with $δ_{- f}$ and descends to a Poincaré duality isomorphism $H M_{k} (f) ≅ H M^{n - k} (- f) ≅ H M^{n - k} (M)$ (continued in 03.15.10). This is the Morse-theoretic proof of Poincaré duality 03.12.16: a single change of sign of the function, no triangulation or cap product with a fundamental class required at the chain level.

Theorem (the ring isomorphism; Fukaya 1993, Betz-Cohen 1994). The $Y$ -graph cup product makes $HM^(M) $a u ni t a l, a ssoc ia t i v e, g r a d e d - co mm u t a t i v er in g, an d t h e M or seH o m o l o g y T h eor e m co m p a r i so ni s ani so m or p hi s m o f g r a d e d r in g s$ HM^(M) \cong H^(M; \mathbb{Z}) $c a r r y in g$ \smile $t o t h es in g u l a r c u pp r o d u c t . * T h es t r u c t u r eco n s t an t s a r e t h e f ib r e - p r o d u c t co u n t s$ n(x_1, x_2; y) = #\bigl(W^u(x_1;f_1) \cap W^u(x_2;f_2) \cap W^s(y;f_3)\bigr)$ in the rigid case.

Theorem (graph operations and Morse field theory; Cohen-Norbury 2009). Assigning to a finite oriented trivalent tree the fibre product of the corresponding stable/unstable manifolds, with one Morse function per edge, defines on $C^(M) $t h es t r u c t u r eo f a M or se f i e l d t h eor y : t h e g e n u s - z er o t r ees g i v e an$ A_\infty $(in f a c t$ C_\infty $) a l g e b r a w h ose f i r s tp r o d u c t i s$ \smile $an d w h ose hi g h er p r o d u c t s a r e t h e M a ssey - t y p eo p er a t i o n s$ m_k $. * T h e a ssoc ia t i v i t y h o m o t o p y o f E x er c i se 8 i s$ m_3 $; t h eo p er a t i o n ss a t i s f y t h e$ A_\infty $r e l a t i o n s$ \sum \pm, m_i(\mathbf{1} \otimes m_j \otimes \mathbf{1}) = 0$, which are signed-boundary-zero statements for the corresponding one-dimensional tree moduli.

Theorem (Witten's analytic prediction; Witten 1982). The deformed differential $d_{t} = e^{- t f} d e^{t f}$ on $\Omega^(M) $ha s, a s$ t \to \infty $, a s p ec t r u m w h ose l o w - l y in g e i g e n s p a cesco n ce n t r a t e a tt h ecr i t i c a l p o in t so f$ f $; t h e t u nn e l l in g am pl i t u d es b e tw ee n cr i t i c a l p o in t sr e p r o d u ce t h e ma t r i x e l e m e n t s$ n(x,y) $o f t h e M or seco b o u n d a r y, an d t h e w e d g e p r o d u c t o f f or m s d e g e n er a t eso n t o t h e$ Y $- g r a p h c u pp r o d u c t . * T h e d e R ham co m pl e x$ (\Omega^*(M), d)$ with its wedge product is the analytic limit of the Morse cochain ring, the precise sense in which Morse cohomology is de Rham cohomology with its ring structure.

Synthesis. The cup product is the foundational reason the Morse complex is more than an additive bookkeeping device, and the central insight is that one trajectory-counting machine, fed graphs of increasing complexity, generates the entire multiplicative hierarchy: the $Y$ -graph gives $⌣$ , the $H$ -graph gives the associativity homotopy, and arbitrary trees give the $A_{\infty}$ operations. This is exactly the structure that identifies $H M^{*} (M)$ with the singular cohomology ring and, through flow reversal $C^{*} (f) ≅ C_{n -*} (- f)$ , with its Poincaré-dual intersection theory; the Morse cup product is dual to the intersection product, and putting these together the same fibre-product geometry that computes products computes dualities. The construction generalises from finite-dimensional $M$ to the loop space and the gauge-orbit space, where it appears again as the pair-of-pants product in symplectic Floer cohomology and the instanton product in Donaldson theory, and where the tree operations become the Fukaya-category compositions. The bridge is the recognition that Witten's deformed de Rham complex is the analytic incarnation of this graph calculus: the wedge product of forms is the continuum limit of the $Y$ -graph count, and the Morse cohomology ring is the de Rham cohomology ring read through the critical points of a single function.

Full proof set Master

Proposition (the coboundary squares to zero and is the transpose of $\partial$ ). With $δ = \partial^{\lor}$ , one has $δ^{2} = 0$ , and in the dual basis $δ ⟨ y ⟩^{\lor} = \sum_{μ (x) = μ (y) + 1} n (x, y) ⟨ x ⟩^{\lor}$ .

Proof. By definition $⟨ δ ξ, c ⟩ = ⟨ ξ, \partial c ⟩$ for $ξ \in C^{k}$ , $c \in C_{k + 1}$ . Then $⟨ δ^{2} ξ, c ⟩ = ⟨ δ ξ, \partial c ⟩ = ⟨ ξ, \partial^{2} c ⟩ = 0$ for all $c$ because $\partial^{2} = 0$ on the Morse complex 03.15.06; since the pairing is perfect (the $⟨ x ⟩$ form a basis), $δ^{2} ξ = 0$ . For the matrix form, evaluate $⟨ δ ⟨ y ⟩^{\lor}, ⟨ x ⟩⟩ = ⟨⟨ y ⟩^{\lor}, \partial ⟨ x ⟩⟩ = ⟨⟨ y ⟩^{\lor}, \sum_{y^{'}} n (x, y^{'}) ⟨ y^{'} ⟩⟩ = n (x, y)$ whenever $μ (x) = μ (y) + 1$ , and $0$ otherwise. So $δ$ is the matrix transpose of $\partial$ , and its nonzero entries are exactly the signed trajectory counts read upward. $□$

Proposition (the unit class). Let $f_{3}$ be self-indexing with a unique minimum $m$ . Then $[m]^{\lor} \in H M^{0} (f_{3})$ is a cocycle and acts as a two-sided unit for $⌣$ .

Proof. First $δ [m]^{\lor} = 0$ : the coboundary $δ ⟨ m ⟩^{\lor}$ counts trajectories from an index- $1$ critical point down into $m$ , summed as $\sum_{μ (x) = 1} n (x, m) ⟨ x ⟩^{\lor}$ . For a connected $M$ the augmentation $C_{0} (f_{3}) \to Z$ sending every minimum to $1$ is a chain map, and $m$ being the unique minimum makes $⟨ m ⟩^{\lor}$ the pullback of the generator of $H^{0} (pt)$ ; its coboundary vanishes because $[m]^{\lor}$ represents the cohomology class dual to the path component. For unitality, consider $⟨ x ⟩^{\lor} ⌣ [m]^{\lor}$ . The relevant $Y$ -graph has one incoming leg landing in $W^{s} (m; f_{3})$ . Because $m$ is the unique minimum of a self-indexing function, $W^{s} (m; f_{3}) = M$ up to a set of lower dimension, so the constraint imposed by that leg is vacuous and the fibre product $W^{u} (x; f_{1}) \cap W^{u} (m; f_{2}) \cap W^{s} (y; f_{3})$ collapses to a single continuation trajectory from $x$ to $y$ . The signed count is therefore the continuation map, which on cohomology is the identity after passing to $H M^{*} (M)$ . Hence $⟨ x ⟩^{\lor} ⌣ [m]^{\lor} = ⟨ x ⟩^{\lor}$ in cohomology, and symmetrically on the left. $□$

Proposition (graded commutativity). On cohomology, $ξ_{1} ⌣ ξ_{2} = (- 1)^{∣ ξ_{1} ∣∣ ξ_{2} ∣} ξ_{2} ⌣ ξ_{1}$ .

Proof. The two products are computed from the fibre products $W^{u} (x_{1}; f_{1}) \cap W^{u} (x_{2}; f_{2}) \cap W^{s} (y; f_{3})$ and $W^{u} (x_{2}; f_{2}) \cap W^{u} (x_{1}; f_{1}) \cap W^{s} (y; f_{3})$ , which are the same subset of $M$ ; the only difference is the ordering of the two incoming factors, hence of the oriented bases of $T W^{u} (x_{1})$ and $T W^{u} (x_{2})$ used to coorient the intersection. Reordering a coorientation of an intersection of submanifolds of codimensions $k = μ (x_{1})$ and $l = μ (x_{2})$ introduces the Koszul sign $(- 1)^{k l}$ . Therefore each contributing point carries opposite signs in the two counts up to $(- 1)^{k l}$ , giving $n (x_{1}, x_{2}; y) = (- 1)^{k l} n (x_{2}, x_{1}; y)$ and the stated identity on cohomology. The choice of three independent generic functions guarantees both fibre products are simultaneously transverse, so the comparison is term-by-term. $□$

Proposition (associativity), proof. The two bracketings of a triple cup product agree on cohomology. Form the moduli $H (x_{1}, x_{2}, x_{3}; y)$ of $H$ -graphs: two trivalent vertices joined by an internal edge of length $ℓ \in (0, \infty)$ , the four outer legs flowing from $x_{1}, x_{2}, x_{3}$ (along $f_{1}, f_{2}, f_{3}$ ) and to $y$ (along $f_{4}$ ). For labels with $\sum μ (x_{i}) - μ (y) = 1$ this is an oriented $1$ -manifold; its Gromov-type compactification adds boundary at $ℓ \to 0$ and $ℓ \to \infty$ together with once-broken outer-edge configurations. The $ℓ \to 0$ end is the four-valent degeneration that, after the standard resolution, computes $(ξ_{1} ⌣ ξ_{2}) ⌣ ξ_{3}$ ; the $ℓ \to \infty$ end splits the internal edge and, by gluing 03.15.04, computes $ξ_{1} ⌣ (ξ_{2} ⌣ ξ_{3})$ ; the broken-outer-edge boundaries contribute $δ$ -exact terms. Signed boundary count zero on a compact oriented $1$ -manifold yields $(ξ_{1} ⌣ ξ_{2}) ⌣ ξ_{3} - ξ_{1} ⌣ (ξ_{2} ⌣ ξ_{3}) = δ (\dots) \pm (\dots) δ$ , an identity that vanishes on $H M^{*}$ . $□$

Theorem (ring isomorphism onto singular cohomology), stated with proof reference — see Fukaya 1993 §§4–6 and Betz-Cohen 1994 ^{[source pending]}. The comparison chain map $Φ : C^{*} (f) \to C_{sing}^{*} (M; Z)$ of the Morse Homology Theorem is upgraded to a homomorphism of $A_{\infty}$ -algebras whose linear term is a quasi-isomorphism and whose quadratic term intertwines the $Y$ -graph product with the Alexander-Whitney cup product; on cohomology the linear term is the ring isomorphism. The full proof requires the gradient-tree moduli and their compactness/gluing theory, established by Fukaya for the $A_{\infty}$ structure and by Betz-Cohen for the graph-moduli formalism; the finite-dimensional case is the model that Cohen-Norbury 2009 axiomatised as a positive boundary topological field theory.

Connections Master

The Morse complex and $\partial^{2} = 0$ 03.15.06. Morse cohomology is the linear dual of that complex: $C^{*} (f) = Hom (C_{*} (f), Z)$ and $δ = \partial^{\lor}$ , so $δ^{2} = 0$ is the transpose of $\partial^{2} = 0$ . Every structure here — the coboundary, the ring, the duality — is read off the differential built in 03.15.06; this unit supplies its multiplicative and contravariant counterpart.
Cap product 03.12.17. The singular cup product and ring structure that this unit reconstructs Morse-theoretically are exactly the operations defined on singular cohomology there. The ring isomorphism $H M^{*} (M) ≅ H^{*} (M; Z)$ identifies the $Y$ -graph product with the Alexander-Whitney cup product, so the two units describe one ring by two routes — combinatorial chains versus gradient flow.
Poincaré duality 03.12.16. Flow reversal $W^{u} (x; f) = W^{s} (x; - f)$ gives $C^{*} (f) ≅ C_{n -*} (- f)$ , the chain-level Poincaré duality. The Morse cup-product pairing $H M^{k} \otimes H M^{n - k} \to Z$ becomes the intersection form under this identification, which is the singular Poincaré-duality pairing; the Morse-theoretic proof (a single sign change of $f$ ) is the subject of 03.15.10.
Continuation maps and invariance of $H M_{*}$ 03.15.07. Well-definedness of the product up to the three-function choices rests on continuation invariance: different generic data give chain-homotopic complexes and continuation-conjugate products, so the ring $H M^{*} (M)$ is an invariant of $M$ . The same machinery licenses the comparison with singular cohomology and the independence of the $Y$ -graph count from the chosen Morse functions.

Historical & philosophical context Master

The dual reading of the Morse complex is as old as the complex itself. Edward Witten's 1982 Supersymmetry and Morse theory (J. Differential Geom. 17, 661–692) ^{[source pending]} introduced the deformed differential $d_{t} = e^{- t f} d e^{t f}$ on differential forms and argued that as $t \to \infty$ the instanton tunnelling amplitudes between critical points reproduce the Morse differential, with the wedge product of forms degenerating onto a product on the critical-point complex. Witten's account was physical rather than rigorous, but it identified the de Rham ring as the analytic source of a Morse-theoretic product and set the agenda for a flow-line construction.

The rigorous $Y$ -graph construction emerged in the early 1990s. Kenji Fukaya's 1993 Morse homotopy, $A_{\infty}$ -category, and Floer homologies (Proc. GARC Workshop, Seoul) ^{[source pending]} organised the products from gradient trees into an $A_{\infty}$ -category, with the cup product as the first composition and higher Massey products from larger trees. Independently, Mélanie Betz and Ralph Cohen's 1994 Graph moduli spaces and cohomology operations (Turkish J. Math. 18, 23–41) ^{[source pending]} gave the fibre-product-of-trajectory-spaces formalism — the count $# W^{u} (x_{1}) \cap W^{u} (x_{2}) \cap W^{s} (y)$ — and proved it realises the singular cup product. Matthias Schwarz's 1993 monograph Morse Homology (Birkhäuser) had meanwhile established the dual complex, the ring structure, and Poincaré duality via flow reversal within the self-contained finite-dimensional theory, as the template every Floer theory would later reproduce. Ralph Cohen and Paul Norbury's Morse field theory (Asian J. Math. 16 (2012), 661–712) ^{[source pending]} axiomatised the graph operations as a positive-boundary topological field theory, closing the circle from Witten's physical intuition to a rigorous functorial structure.

Bibliography Master

@article{Witten1982SUSYMorse,
  author  = {Witten, Edward},
  title   = {Supersymmetry and {M}orse theory},
  journal = {Journal of Differential Geometry},
  volume  = {17},
  number  = {4},
  year    = {1982},
  pages   = {661--692}
}

@incollection{Fukaya1993MorseHomotopy,
  author    = {Fukaya, Kenji},
  title     = {{M}orse homotopy, $A_\infty$-category, and {F}loer homologies},
  booktitle = {Proceedings of the GARC Workshop on Geometry and Topology (Seoul, 1993)},
  series    = {Lecture Notes Series},
  volume    = {18},
  publisher = {Seoul National University},
  year      = {1993},
  pages     = {1--102}
}

@article{BetzCohen1994GraphModuli,
  author  = {Betz, M{\'e}lanie and Cohen, Ralph L.},
  title   = {Graph moduli spaces and cohomology operations},
  journal = {Turkish Journal of Mathematics},
  volume  = {18},
  number  = {1},
  year    = {1994},
  pages   = {23--41}
}

@book{Schwarz1993MorseHomology,
  author    = {Schwarz, Matthias},
  title     = {Morse Homology},
  series    = {Progress in Mathematics},
  volume    = {111},
  publisher = {Birkh{\"a}user Verlag},
  address   = {Basel},
  year      = {1993}
}

@article{CohenNorbury2012MorseFieldTheory,
  author  = {Cohen, Ralph L. and Norbury, Paul},
  title   = {Morse field theory},
  journal = {Asian Journal of Mathematics},
  volume  = {16},
  number  = {4},
  year    = {2012},
  pages   = {661--712}
}

@book{AudinDamian2014,
  author    = {Audin, Mich{\`e}le and Damian, Mihai},
  title     = {Morse Theory and Floer Homology},
  series    = {Universitext},
  publisher = {Springer},
  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},
  year      = {2004}
}

@book{Milnor1963MorseTheory,
  author    = {Milnor, John},
  title     = {Morse Theory},
  series    = {Annals of Mathematics Studies},
  volume    = {51},
  publisher = {Princeton University Press},
  year      = {1963}
}

Prerequisites

03.15.06
03.15.07
03.12.17

Tier anchors

beginner: Audin-Damian *Morse Theory and Floer Homology* §4.9 (products); Hutchings *Lecture notes on Morse homology* §4
intermediate: Schwarz *Morse Homology* Part III; Banyaga-Hurtubise *Lectures on Morse Homology* Ch. 12; Audin-Damian §4.9
master: Schwarz *Morse Homology* Part III; Fukaya 1993; Betz-Cohen 1994; Cohen-Norbury 2009

References

03-modern-geometry/Bott-LecturesOnMorseTheory.pdf · Gradient flows and the relation between Morse and de Rham theory (background for the dual complex)
Schwarz — Morse Homology (Progress in Mathematics 111, Birkhäuser 1993) · Part III (the Morse cohomology, the cup product via fibre products of trajectory spaces, and the ring isomorphism)
Fukaya — Morse homotopy, A∞-category, and Floer homologies · Proceedings of the GARC Workshop on Geometry and Topology (Seoul, 1993), Lecture Notes Series 18, pp. 1-102 (products and the A∞ structure via gradient trees)
Betz-Cohen — Graph moduli spaces and cohomology operations · Turkish Journal of Mathematics 18 (1994), pp. 23-41 (the Y-graph / fibre-product construction of the Morse cup product)
Witten — Supersymmetry and Morse theory · Journal of Differential Geometry 17 (1982), pp. 661-692 (the instanton differential and the de Rham/Morse dictionary underlying the dual complex)
Cohen-Norbury — Morse field theory · Asian Journal of Mathematics 16 (2012), pp. 661-712 (graph operations on Morse complexes, associativity and the ring structure)

Estimated time

beginner: 17m
intermediate: 42m
master: 82m