Poincaré duality

Intuition [Beginner]

Poincaré duality is the rule that says, on a closed oriented manifold of dimension $n$, the shape information that lives in dimension $k$ is the same as the shape information that lives in dimension $n-k$. A two-dimensional surface has its zero-dimensional homology (path components) matched up with its two-dimensional homology (top-degree cycles), and its one-dimensional homology paired with itself. A three-dimensional manifold has its one-dimensional cycles paired with its two-dimensional surfaces. The dimensions add to $n$ in every case.

Why should this be believable? Imagine drawing a curve on a closed surface like a torus and a second curve transverse to it. The two curves cross at isolated points, and the count of crossings (with signs) is a number that does not change when you wiggle either curve. That number is the intersection number, and the rule that you can always pair a $k$-dimensional cycle with an $(n-k)$-dimensional cycle by counting transverse intersections is the geometric heart of duality. The cap product is the algebraic packaging of this counting.

The reason this matters: duality is the bridge between cohomology and homology on manifolds. Cohomology is built from cochains and looks contravariant; homology is built from chains and looks covariant. On a closed oriented manifold the two become indistinguishable up to a dimension flip. Every manifold invariant computed via integration, intersection, or characteristic classes ultimately rests on this identification.

Visual [Beginner]

A schematic showing a closed orientable surface (a two-torus) with two transverse loops drawn on it. One loop is a meridian, one is a longitude, and they cross at a single point. Arrows indicate that the meridian is a one-dimensional homology class, the longitude is its dual under Poincaré duality, and the crossing point is the intersection number $1$ that records the pairing.

The picture captures the essential geometry: on a closed oriented surface, every one-dimensional cycle has a dual one-dimensional cycle, and the duality is detected by the intersection number. The same picture in higher dimensions replaces curves with submanifolds of complementary dimension.

Worked example [Beginner]

Compute the Poincaré-duality matching on the two-sphere $S^2$ and on the two-torus $T^2$, and check that it gives the right answer in each degree.

Step 1. The two-sphere $S^2$ has dimension $n=2$. Its homology is $H_0=\mathbb{Z}$ (one path component), $H_1=0$ (no one-dimensional holes), and $H_2=\mathbb{Z}$ (one closed surface, the sphere itself). Its cohomology is identical: $H^0=\mathbb{Z}$, $H^1=0$, $H^2=\mathbb{Z}$.

Step 2. Poincaré duality predicts $H^k(S^2)\cong H_{2-k}(S^2)$ for every $k$. Checking degree by degree: $H^0=\mathbb{Z}$ matches $H_2=\mathbb{Z}$; $H^1=0$ matches $H_1=0$; $H^2=\mathbb{Z}$ matches $H_0=\mathbb{Z}$. Every degree agrees. The duality is at work even on the simplest closed surface.

Step 3. The two-torus $T^2$ has dimension $n=2$. Its homology is $H_0=\mathbb{Z}$, $H_1={\mathbb{Z}}^2$ (two independent one-dimensional loops, the meridian and the longitude), and $H_2=\mathbb{Z}$.

Step 4. Poincaré duality predicts $H^k(T^2)\cong H_{2-k}(T^2)$. Degree zero matches degree two: each is $\mathbb{Z}$. Degree one matches itself: $H^1={\mathbb{Z}}^2=H_1$. Degree two matches degree zero: each is $\mathbb{Z}$. The two generators of $H_1$ are the meridian and the longitude, and Poincaré duality identifies the meridian as the cohomology class dual to the longitude, with intersection number $1$ at their unique transverse crossing point.

Step 5. Check the intersection. Draw the meridian and the longitude on the torus so they cross at exactly one point. The Poincaré-duality pairing of the meridian (as a one-cycle) with the longitude (as a one-cocycle, after duality) is the count of crossings with signs, which is $1$. This is the intersection number, and it is exactly the number Poincaré duality computes.

What this tells us: Poincaré duality is not abstract bookkeeping. On every closed oriented manifold the duality matches each homology class with a cohomology class of complementary dimension, and the pairing is computed by transverse intersection. The torus is the smallest example where the matching is interesting in the middle dimension.

Check your understanding [Beginner]

Formal definition [Intermediate+]

Let $M$ be a topological manifold of dimension $n$, meaning a Hausdorff second-countable space in which every point has a neighbourhood homeomorphic to ${\mathbb{R}}^n$. The manifold is closed when it is compact and without boundary. A local orientation at $x\in M$ is a generator ${\mu }_x$ of $H_n(M,M\setminus \{x\};\mathbb{Z})\cong \mathbb{Z}$, and an orientation of $M$ is a choice of local orientation at every point that is locally consistent: every point has a neighbourhood $U\cong {\mathbb{R}}^n$ on which all the local orientations come from a single class in $H_n(M,M\setminus U)\cong \mathbb{Z}$.

For $M$ a closed connected $n$-manifold, the homology $H_n(M;\mathbb{Z})$ is $\mathbb{Z}$ if $M$ is orientable and $0$ otherwise. When $M$ is orientable, an orientation determines a generator $[M]\in H_n(M;\mathbb{Z})$ called the fundamental class.

The cap product is the bilinear pairing $$ \frown : H^k(M; R) \otimes H_n(M; R) \to H_{n-k}(M; R) $$ defined at the chain level by $\sigma ⌢\phi =\phi (\sigma {|}_{[v_0,\dots ,v_k]})\cdot \sigma {|}_{[v_k,\dots ,v_n]}$ on a singular $n$-simplex $\sigma$ paired with a $k$-cochain $\phi$, then descended to (co)homology. Capping with the fundamental class defines the Poincaré duality map $$ D : H^k(M; \mathbb{Z}) \to H_{n-k}(M; \mathbb{Z}), \qquad D(\alpha) = \alpha \frown [M]. $$

The $\mathbb{Z}/2$ fundamental class $[M{]}_2\in H_n(M;\mathbb{Z}/2)$ exists for every closed $n$-manifold without an orientation hypothesis: the local coefficient system $H_n(M,M\setminus \{x\};\mathbb{Z}/2)\cong \mathbb{Z}/2$ has only one nonzero element, so the consistency requirement is automatic.

For a manifold with boundary $(M,\partial M)$ that is compact, the relative fundamental class is a generator $[M,\partial M]\in H_n(M,\partial M;\mathbb{Z})$ when $M$ is orientable; it restricts to the fundamental class of $\partial M$ under the connecting map of the long exact sequence of the pair.

For a non-compact manifold, the cohomology that pairs with ordinary homology is cohomology with compact supports $$ H^k_c(M; R) = \mathrm{colim}_{K \subseteq M \text{ compact}} H^k(M, M \setminus K; R), $$ and the dual of compact-support cohomology under duality is Borel-Moore homology $H_k^{BM}(M;R)$, the homology built from possibly-infinite locally-finite chains.

Counterexamples to common slips

• The orientability hypothesis matters with $\mathbb{Z}$ coefficients. The real projective plane $\mathbb{R}{P}^2$ is a closed two-manifold but not orientable: $H_2(\mathbb{R}{P}^2;\mathbb{Z})=0$, and Poincaré duality with integer coefficients fails. With $\mathbb{Z}/2$ coefficients, $H_2(\mathbb{R}{P}^2;\mathbb{Z}/2)=\mathbb{Z}/2$ and the duality $H^k(\mathbb{R}{P}^2;\mathbb{Z}/2)\cong H_{2-k}(\mathbb{R}{P}^2;\mathbb{Z}/2)$ holds.
• The closedness hypothesis matters. The open disk ${\mathring{D}}^2$ is an orientable two-manifold without boundary, but not compact: $H_2({\mathring{D}}^2;\mathbb{Z})=0$, and the ordinary cohomology $H^k$ does not pair with the ordinary homology $H_{n-k}$. The compact-supports / Borel-Moore replacement is what restores duality in the non-compact setting.
• The dual cell decomposition Poincaré used in 1895 makes integral duality visible at the chain level only for triangulable manifolds with a chosen triangulation. The cap-product formulation works for any topological manifold, with no triangulation required, and is what Lefschetz, Eilenberg, and Steenrod established as the modern statement.

Key theorem with proof [Intermediate+]

Theorem (Poincaré duality; Hatcher Theorem 3.30). Let $M$ be a closed oriented $n$-manifold with fundamental class $[M]\in H_n(M;\mathbb{Z})$. The cap product with $[M]$ induces an isomorphism $$ D : H^k(M; \mathbb{Z}) \xrightarrow{\cong} H_{n-k}(M; \mathbb{Z}) $$ for every $k\in \{0,1,\dots ,n\}$. The same statement holds with coefficients in any commutative ring $R$ once $M$ is $R$-orientable; in particular it holds for every closed $n$-manifold with $R=\mathbb{Z}/2$.

Proof. The argument has three steps. First, reduce to the local statement on ${\mathbb{R}}^n$ by a Mayer-Vietoris / open-cover induction. Second, prove the local statement directly. Third, assemble the pieces using a five-lemma argument.

Step 1: open-cover induction. Suppose $U,V\subseteq M$ are open subsets for which the duality map $D_U:H_c^k(U;\mathbb{Z})\to H_{n-k}(U;\mathbb{Z})$ is an isomorphism in every $k$, and similarly for $V$ and $U\cap V$. Then the same holds for $U\cup V$. The argument uses the Mayer-Vietoris sequence for compactly supported cohomology $$ \cdots \to H^k_c(U \cap V) \to H^k_c(U) \oplus H^k_c(V) \to H^k_c(U \cup V) \to H^{k+1}_c(U \cap V) \to \cdots $$ and the standard Mayer-Vietoris sequence for homology, with cap products by the relevant fundamental classes giving a commutative ladder. The five lemma upgrades the assumed isomorphisms to an isomorphism on $U\cup V$.

The same induction extends to nested unions: if $\{U_i\}$ is a directed family of open subsets with $D_{U_i}$ an isomorphism for each $i$, then $D_{{\bigcup }_i{U}_i}$ is an isomorphism. This uses commutation of compactly supported cohomology and homology with directed colimits.

Step 2: local statement on ${\mathbb{R}}^n$. For $U={\mathbb{R}}^n$ the homology is $H_k({\mathbb{R}}^n;\mathbb{Z})=\mathbb{Z}$ for $k=0$ and zero for $k>0$. The compactly supported cohomology is $H_c^k({\mathbb{R}}^n;\mathbb{Z})=\mathbb{Z}$ for $k=n$ and zero for $k<n$, computed by excision applied to the pair $({\mathbb{R}}^n,{\mathbb{R}}^n\setminus 0)\simeq (D^n,S^{n-1})$. The duality map $D:H_c^k({\mathbb{R}}^n)\to H_{n-k}({\mathbb{R}}^n)$ vanishes for $k\ne n$ and identifies the generator of $H_c^n({\mathbb{R}}^n)=\mathbb{Z}$ with the generator of $H_0({\mathbb{R}}^n)=\mathbb{Z}$, which is the class of any point. Both sides are $\mathbb{Z}$ in degree $n$ and zero otherwise; the cap-product pairing is the canonical generator-to-generator isomorphism.

Step 3: assembly. Cover $M$ by finitely many open subsets $U_1,\dots ,U_N$ each homeomorphic to ${\mathbb{R}}^n$. Such a cover exists because $M$ is compact and locally Euclidean. Write $V_j=U_1\cup \cdots \cup U_j$. By Step 2 the duality holds for each $U_i$. By Step 1, induction on $j$ extends the duality to $V_j$ for every $j$, since $V_{j+1}=V_j\cup U_{j+1}$ and the intersection $V_j\cap U_{j+1}$ is itself an open subset of the Euclidean $U_{j+1}$, hence a finite union of open Euclidean charts to which the same induction applies. After $N$ steps, $V_N=M$ and the duality map $D:H^k(M;\mathbb{Z})\to H_{n-k}(M;\mathbb{Z})$ is an isomorphism. The compactness of $M$ identifies $H_c^k(M)=H^k(M)$, completing the argument.

For the $\mathbb{Z}/2$ version, the same argument applies with $\mathbb{Z}/2$ coefficients throughout. The orientation hypothesis was used only to identify $H_n(M,M\setminus K;\mathbb{Z})=\mathbb{Z}$ globally, and with $\mathbb{Z}/2$ coefficients the local-orientation choice is automatic. $\square$

Bridge. Poincaré duality builds toward every manifold-theoretic invariant in algebraic topology and in differential geometry. The foundational reason it holds is exactly the local computation on ${\mathbb{R}}^n$: at each point of an $n$-manifold, the local pairing of $H_c^n$ with $H_0$ identifies the orientation generator with the point class, and the open-cover induction propagates this local matching to the global statement. This is exactly the same local-to-global principle that appears again in 03.04.06 (de Rham cohomology), where the integration pairing ${\int }_M:H_{dR}^k(M)\times H_{dR}^{n-k}(M)\to \mathbb{R}$ on a closed oriented smooth manifold is the smooth-manifold incarnation of the same duality. The central insight is that Poincaré duality identifies cohomology of degree $k$ with homology of degree $n-k$ on every closed oriented manifold, and the map is exactly cap product with the fundamental class. The bridge is the recognition that this duality generalises to manifolds with boundary as Lefschetz duality, to non-compact manifolds via compactly supported cohomology and Borel-Moore homology, and to non-orientable manifolds via $\mathbb{Z}/2$ coefficients. Putting these together, one Poincaré-duality framework produces every classical pairing on manifolds — intersection numbers, signature forms, the integration pairing on de Rham forms — and identifies them with one another at the level of cohomology.

Exercises [Intermediate+]

Lean formalization [Intermediate+]

Mathlib has the chain-complex infrastructure in Algebra.Homology and the singular-set construction in AlgebraicTopology.SingularSet, but no named Poincaré duality theorem. The intended formalisation reads schematically:

[object Promise]

The proof gap is substantive. Mathlib needs the orientation sheaf as a local system $H_n(M,M\setminus \{x\};\mathbb{Z})$, the fundamental class as a global section, the Alexander-Whitney formula for cap product as a chain map, the Mayer-Vietoris ladder for compactly supported cohomology, and the open-cover / direct-limit assembly. Each piece is formalisable from existing infrastructure but has not been packaged. The relative version (Lefschetz duality), the $\mathbb{Z}/2$ version, and the compact-support / Borel-Moore version are further targets, as is the de Rham incarnation predicated on Mathlib first acquiring a named de Rham cohomology object for smooth manifolds.

Advanced results [Master]

Theorem (Lefschetz duality; Hatcher Theorem 3.43). Let $(M,\partial M)$ be a compact oriented $n$-manifold-with-boundary with relative fundamental class $[M,\partial M]\in H_n(M,\partial M;\mathbb{Z})$. Cap product with $[M,\partial M]$ induces isomorphisms $$ H^k(M, \partial M; \mathbb{Z}) \xrightarrow{\cong} H_{n-k}(M; \mathbb{Z}), \qquad H^k(M; \mathbb{Z}) \xrightarrow{\cong} H_{n-k}(M, \partial M; \mathbb{Z}) $$ for every $k$. Together with the long exact sequences of the pair $(M,\partial M)$, the two duality maps are dual to one another and recover Poincaré duality on $\partial M$ as the boundary case.

The proof proceeds by doubling: the closed oriented $n$-manifold $DM=M{\cup }_{\partial M}M$ obtained by gluing two copies of $M$ along their common boundary admits ordinary Poincaré duality, and Mayer-Vietoris on $DM=M\cup M$ decomposes the duality isomorphism into the two pieces above. Naturality of cap product propagates the matching to a natural isomorphism of pairs.

Theorem (compact-supports duality; Hatcher Theorem 3.35). Let $M$ be a (not necessarily compact) oriented $n$-manifold. Cap product with the fundamental class in Borel-Moore homology induces an isomorphism $$ D_c : H^k_c(M; \mathbb{Z}) \xrightarrow{\cong} H_{n-k}(M; \mathbb{Z}) $$ for every $k$, and dually $$ D^c : H^k(M; \mathbb{Z}) \xrightarrow{\cong} H_{n-k}^{BM}(M; \mathbb{Z}) $$ where $H^c$iscohomologywithcompactsupportsand$H^{BM}$ is Borel-Moore (locally-finite) homology.

For a closed manifold, $H_c^k=H^k$ and $H_{\ast }^{BM}=H_{\ast }$ since every chain is automatically locally finite, and the two compact-support statements collapse to ordinary Poincaré duality. For an open manifold, the compact-support and Borel-Moore framings are the homotopical replacements that preserve duality.

Theorem (de Rham incarnation). Let $M$ be a closed oriented smooth $n$-manifold. The integration pairing $$ \int_M : H^k_{dR}(M; \mathbb{R}) \otimes H^{n-k}{dR}(M; \mathbb{R}) \to \mathbb{R}, \qquad [\alpha] \otimes [\beta] \mapsto \int_M \alpha \wedge \beta $$ *is non-degenerate. Equivalently, it identifies $H^{n-k}{dR}(M; \mathbb{R})$withthedualvectorspace$\mathrm{Hom}{\mathbb{R}}(H^k{dR}(M; \mathbb{R}), \mathbb{R})$,andunderthedeRhamisomorphism$H^_{dR}(M; \mathbb{R}) \cong H^(M; \mathbb{R})$ this is exactly Poincaré duality with real coefficients.*

This is the smooth-manifold incarnation of Poincaré duality. Bott-Tu §5 develops the theorem from the de Rham side first, deducing the topological statement as a consequence.

Theorem (Hirzebruch signature theorem; closed oriented $4k$-manifolds). For a closed oriented $4k$-manifold $M$, the cup product induces a non-degenerate symmetric bilinear form $$ Q : H^{2k}(M; \mathbb{R}) \times H^{2k}(M; \mathbb{R}) \to \mathbb{R}, \qquad Q(\alpha, \beta) = \langle \alpha \cup \beta, [M] \rangle. $$ The signature $\sigma (M)$ is the signature (number of positive eigenvalues minus number of negative eigenvalues) of $Q$. The Hirzebruch signature theorem identifies $\sigma (M)$ with $⟨L(TM),[M]⟩$, the integral of the Hirzebruch $L$-class against the fundamental class.

Poincaré duality is the input that makes the cup-product form non-degenerate; without it, the form would be only a graded pairing into $H^{4k}(M;\mathbb{R})$ and would carry no signature. The signature theorem is one of the original applications of the index-theory machinery, a special case of Atiyah-Singer.

Theorem (vanishing of odd Euler characteristic). For a closed oriented $n$-manifold with $n$ odd, $\chi (M)=0$.

The argument: Poincaré duality identifies $H^k(M;\mathbb{Q})$ with $H_{n-k}(M;\mathbb{Q})=H^{n-k}(M;\mathbb{Q})$ (the second equality is the universal coefficient theorem over a field, with no torsion). So $b_k=b_{n-k}$ for every $k$, and $\chi (M)={\sum }_k(-1{)}^k{b}_k$ pairs each $b_k$ with $b_{n-k}$ in opposite signs. When $n$ is odd, the sum cancels in pairs and $\chi (M)=0$. The $\mathbb{Z}/2$ version of this argument extends the conclusion to closed (not necessarily orientable) manifolds of odd dimension.

Theorem (intersection pairing on a closed oriented $n$-manifold). Cap product followed by the augmentation $H_0(M;\mathbb{Z})\to \mathbb{Z}$ produces a bilinear form $$ H_k(M; \mathbb{Z}) \otimes H_{n-k}(M; \mathbb{Z}) \to \mathbb{Z}. $$ Geometrically, for transverse submanifolds $A^k,B^{n-k}\subseteq M$, the pairing computes the algebraic count of intersection points $A⋔B$ with signs determined by the orientations.

This is the geometric content of Poincaré duality. The cap-product pairing on cohomology is dual to the intersection pairing on homology under the duality isomorphism, and the two pairings carry the same information.

Theorem (top-degree cohomology and integration). For a closed oriented connected $n$-manifold $M$, $H^n(M;\mathbb{Z})=\mathbb{Z}$ generated by the orientation cohomology class ${\mu }_M=D^{-1}([\mathrm{pt}])$, the Poincaré dual of the point class. Evaluation against the fundamental class $⟨{\mu }_M,[M]⟩=1$, and on a closed oriented smooth manifold the corresponding de Rham class is represented by any volume form normalised so that ${\int }_M\omega =1$.

The orientation class ${\mu }_M$ is the tautological generator: it is dual to the $0$-cycle "a point" under the cap-product duality, and integration of forms against it computes integrals of top forms on the manifold. Every classical integration theorem on manifolds — Gauss, Stokes, the Gauss-Bonnet theorem — is the de Rham incarnation of cap-product evaluation against ${\mu }_M$.

Synthesis. Poincaré duality is the foundational reason every cohomological invariant on a closed oriented manifold has a homological dual of complementary dimension. The central insight is that cap product with the fundamental class identifies $H^k(M;\mathbb{Z})$ with $H_{n-k}(M;\mathbb{Z})$ on every closed oriented $n$-manifold, and this is exactly the algebraic packaging of the geometric fact that $k$-dimensional cycles pair with $(n-k)$-dimensional cycles by transverse intersection. Putting these together, the cap-product duality, the Lefschetz extension to manifolds with boundary, the compact-support extension to non-compact manifolds, and the $\mathbb{Z}/2$ extension to non-orientable manifolds form one duality framework that adapts to every flavour of manifold geometry. The bridge between the topological statement and the analytic statement is the de Rham theorem: cap product with $[M]$ on the topological side is integration of differential forms on the smooth side, and the two pairings agree under the de Rham isomorphism. This is exactly the bridge that appears again in 03.04.06 (de Rham cohomology), where the integration pairing ${\int }_M:H_{dR}^k(M)\times H_{dR}^{n-k}(M)\to \mathbb{R}$ is the de Rham incarnation of Poincaré duality.

The duality identifies several pairings that look distinct at first inspection. The cup product on cohomology, evaluated against the fundamental class, is dual to the intersection pairing on homology. The integration pairing on de Rham cohomology generalises both. The signature of a closed oriented $4k$-manifold, computed from the middle-dimensional cup-product form, is identified by the Hirzebruch signature theorem with an integral of characteristic classes against the fundamental class. The bridge is that all of these are different presentations of the cap-product pairing $H^k(M)\otimes H_n(M)\to H_{n-k}(M)$ specialised to the fundamental class. Poincaré duality is dual to itself in a precise sense: the duality map $D:H^k\to H_{n-k}$ has an inverse $D^{-1}:H_{n-k}\to H^k$ given by Poincaré duality in degree $n-k$, and the composite $D\circ D^{-1}$ on $H_{n-k}$ is the identity. The recursion stabilises after one round-trip. The duality also generalises to spectra and to generalised homology theories: a Poincaré-duality space in the sense of Wall is a homotopy type that satisfies the same cap-product isomorphism, and Atiyah duality identifies the Spanier-Whitehead dual of a closed manifold with the Thom spectrum of its stable normal bundle, a duality at the spectrum level whose underlying homological consequence is exactly Poincaré duality.

Full proof set [Master]

Theorem (Poincaré duality), proof. Given in the Intermediate-tier section: open-cover induction reduces the global statement to the local statement on ${\mathbb{R}}^n$, the local statement is a direct excision computation $H_c^n({\mathbb{R}}^n)=\mathbb{Z}=H_0({\mathbb{R}}^n)$, and the Mayer-Vietoris five-lemma propagates the local matching to the global one. The compactness of $M$ identifies $H_c^k(M)=H^k(M)$, completing the closed case. $\square$

Theorem (Lefschetz duality), proof. Form the closed double $DM=M{\cup }_{\partial M}M$. This is a closed oriented $n$-manifold with fundamental class $[DM]$ obtained by gluing $[M,\partial M]$ in the two copies. Poincaré duality on $DM$ gives $H^k(DM)\cong H_{n-k}(DM)$ in every degree. The Mayer-Vietoris sequence of $DM=M{\cup }_{\partial M}M$ has the form $$ \cdots \to H^k(DM) \to H^k(M) \oplus H^k(M) \to H^k(\partial M) \to H^{k+1}(DM) \to \cdots, $$ and by symmetry the involution $\tau$ swapping the two copies of $M$ acts on each term. Decomposing into the $+1$ and $-1$ eigenspaces of $\tau$: the $+1$-eigenspace of $H^k(DM)$ is $H^k(M)$, the $-1$-eigenspace is $H^k(M,\partial M)$. The same decomposition on $H_{n-k}(DM)$ gives $H_{n-k}(M)$ and $H_{n-k}(M,\partial M)$. Poincaré duality on $DM$ commutes with $\tau$, so it restricts to isomorphisms on each eigenspace, yielding the two Lefschetz statements. Naturality is immediate from naturality of the Mayer-Vietoris and cap-product constructions. $\square$

Theorem (compact-supports duality), proof sketch. The argument repeats the open-cover induction of the closed case but with $H_c^k$ replacing $H^k$ and Borel-Moore $H_k^{BM}$ replacing $H_k$. Mayer-Vietoris in compactly supported cohomology has a different variance (the connecting map raises degree, as for ordinary cohomology, but the sequence is for $U\cup V$ in terms of $U$, $V$, and $U\cap V$ with a sign flip), and the directed-colimit step over compact subsets is what makes the induction work for a non-compact $M$. The local statement on ${\mathbb{R}}^n$ is the same, since ${\mathbb{R}}^n$ has $H_c^n({\mathbb{R}}^n)=\mathbb{Z}$ already. The full argument is in Hatcher §3.3, Theorems 3.35–3.36. $\square$

Theorem (de Rham incarnation), proof. The de Rham theorem identifies the singular cohomology of a smooth manifold with the cohomology of the de Rham complex, and identifies cup product with wedge product. Under this identification, the topological cap-product pairing $H^k(M;\mathbb{R})\otimes H_n(M;\mathbb{R})\to H_{n-k}(M;\mathbb{R})$ followed by the augmentation $H_0(M;\mathbb{R})\to \mathbb{R}$ becomes the integration pairing on de Rham forms. Specifically, for closed forms $\alpha \in {\Omega }^k(M)$, $\beta \in {\Omega }^{n-k}(M)$, $$ \langle [\alpha] \cup [\beta], [M] \rangle = \int_M \alpha \wedge \beta. $$ By Poincaré duality the bilinear pairing $H^k(M;\mathbb{R})\otimes H^{n-k}(M;\mathbb{R})\to \mathbb{R}$ is non-degenerate, so the integration pairing on de Rham cohomology is also non-degenerate. The smooth-manifold incarnation transfers the topological duality to an analytic statement about integration of differential forms. $\square$

Theorem (Hirzebruch signature theorem), stated without proof — see Hirzebruch 1956 Topological Methods in Algebraic Geometry Ch. 8 ^[pending]. The full signature theorem $\sigma (M)=⟨L(TM),[M]⟩$ requires the theory of multiplicative sequences and the $L$-genus, plus the Atiyah-Singer index theorem in its differential-operator form. Poincaré duality enters as the input that turns the middle-dimensional cup product on a $4k$-manifold into a non-degenerate symmetric form whose signature is well-defined.

Theorem (vanishing odd-dimensional $\chi$), proof. Over $\mathbb{Q}$, the universal coefficient theorem gives $H^k(M;\mathbb{Q})\cong \mathrm{Hom}(H_k(M;\mathbb{Z}),\mathbb{Q})=H_k(M;\mathbb{Q})$ since $\mathbb{Q}$ is a field. So $b_k(M)=\dim H^k(M;\mathbb{Q})=\dim H_k(M;\mathbb{Q})$. By Poincaré duality, $H^k(M;\mathbb{Q})\cong H_{n-k}(M;\mathbb{Q})$, hence $b_k=b_{n-k}$. The Euler characteristic $$ \chi(M) = \sum_{k=0}^n (-1)^k b_k = \sum_{k=0}^n (-1)^k b_{n-k} = (-1)^n \sum_{j=0}^n (-1)^j b_j = (-1)^n \chi(M), $$ where the second-to-last equality uses the substitution $j=n-k$. So $\chi (M)=(-1{)}^n\chi (M)$, and when $n$ is odd this forces $\chi (M)=0$. $\square$

Theorem (intersection pairing), proof. Cap product gives $H^k(M;\mathbb{Z})\otimes H_n(M;\mathbb{Z})\to H_{n-k}(M;\mathbb{Z})$. Composing with the duality isomorphism $D^{-1}:H_k(M;\mathbb{Z})\to H^{n-k}(M;\mathbb{Z})$ in the other slot, and pairing with the fundamental class, gives a bilinear form $H_k(M;\mathbb{Z})\otimes H_{n-k}(M;\mathbb{Z})\to \mathbb{Z}$. For transverse submanifolds $A,B\subseteq M$ of complementary dimension, this form computes the signed count of intersection points, a fact verifiable on any chosen triangulation in which $A$ and $B$ are subcomplexes meeting transversally. The signed count is invariant under transverse perturbation, which is what makes the intersection pairing well-defined on homology. $\square$

Theorem (top-degree cohomology), proof. The fundamental class $[M]\in H_n(M;\mathbb{Z})=\mathbb{Z}$ generates the top-degree homology. Poincaré duality identifies $H^n(M;\mathbb{Z})=\mathbb{Z}$ with $H_0(M;\mathbb{Z})=\mathbb{Z}$ (since $M$ is connected) via the cap product $\alpha \mapsto \alpha ⌢[M]$. The orientation class ${\mu }_M$ is the preimage of the point class, characterised by $⟨{\mu }_M,[M]⟩=1$. On a smooth manifold with chosen Riemannian metric, the de Rham representative of ${\mu }_M$ is the volume form normalised by ${\int }_M\omega =1$. $\square$

Connections [Master]

• Singular homology 03.12.11. Poincaré duality is a statement about the singular homology and singular cohomology of a closed oriented manifold; the fundamental class $[M]\in H_n(M;\mathbb{Z})$ is a singular-homology class, and the cap product is a chain-level operation on singular chains and cochains. Without singular homology the duality has no setting; with it, the duality is the key structural property distinguishing manifolds from arbitrary spaces.

• Excision 03.12.14. The local computation on ${\mathbb{R}}^n$ that is the input to the open-cover induction proof of duality uses excision applied to the pair $({\mathbb{R}}^n,{\mathbb{R}}^n\setminus 0)$ to identify $H_c^n({\mathbb{R}}^n;\mathbb{Z})=\mathbb{Z}$. The Mayer-Vietoris ladder propagating the local statement to a global one is itself derived from excision applied to an open cover. Excision is the load-bearing technical input to Poincaré duality.

• Eilenberg-Steenrod axioms 03.12.15. The seven axioms that characterise ordinary homology theories include excision, exactness of the long exact sequence of a pair, and naturality, all of which appear in the Mayer-Vietoris ladder used in the proof of Poincaré duality. The duality theorem is a consequence of the axioms together with the manifold hypothesis: any homology theory satisfying the axioms gives a Poincaré-duality theorem on closed oriented manifolds, with the appropriate coefficient system.

• De Rham cohomology 03.04.06. The smooth-manifold incarnation of Poincaré duality is the integration pairing ${\int }_M:H_{dR}^k(M)\times H_{dR}^{n-k}(M)\to \mathbb{R}$ on a closed oriented smooth manifold. Under the de Rham theorem, the topological duality and the analytic pairing are identified, and every classical integration theorem on manifolds (Gauss, Stokes, Gauss-Bonnet) is a manifestation of cap-product evaluation against the fundamental class.

• Cellular homology 03.12.13. On a CW manifold with chosen cell structure, the dual cell decomposition (originally due to Poincaré) realises Poincaré duality at the chain level: the dual cell of an $i$-cell is an $(n-i)$-cell in the dual decomposition, and the cellular chain complex of one decomposition is dual to that of the other. The cap-product duality on (co)homology is the homotopy-invariant version of this combinatorial chain-level statement.

Historical & philosophical context [Master]

Poincaré stated the duality that bears his name in his 1895 memoir Analysis Situs (Journal de l'École Polytechnique 1895) and refined it across the five Compléments published between 1899 and 1904 ^[pending]. The original statement was at the level of dual cell decompositions: for a triangulated closed orientable $n$-manifold, the number of $k$-dimensional cells in one decomposition equals the number of $(n-k)$-dimensional cells in the dual decomposition, with combinatorial identifications between boundaries. The Betti numbers $b_k$ and $b_{n-k}$ were equal as a corollary. Poincaré's proof was combinatorial and required a triangulation; the role of orientation was identified explicitly, and the failure of the integral statement for non-orientable manifolds was noted with the projective plane as an example.

The modern cap-product framing was assembled in stages. Solomon Lefschetz's Topology (AMS Colloquium Publications 1930) ^[pending] introduced the intersection pairing on homology and the relative version of duality for manifolds with boundary that now carries his name. James Alexander introduced cap product in 1936 in connection with his duality theorem for compact subsets of spheres. Hassler Whitney refined the cup-and-cap-product machinery in the late 1930s. Norman Steenrod's 1947 reformulation, presented in the Eilenberg-Steenrod monograph Foundations of Algebraic Topology (Princeton 1952), packaged Poincaré duality as a consequence of the seven axioms together with the manifold hypothesis, with cap product as the universal pairing. The monograph established the modern presentation: $D:H^k(M;\mathbb{Z})\to H_{n-k}(M;\mathbb{Z})$, $D(\alpha )=\alpha ⌢[M]$.

The smooth-manifold incarnation has a parallel history. Georges de Rham's 1931 thesis Sur l'analysis situs des variétés à n dimensions (J. Math. Pures Appl. 10, 115-200) ^[pending] established the isomorphism between singular cohomology and the cohomology of the complex of differential forms on a smooth manifold, opening the route to the integration pairing ${\int }_M\alpha \wedge \beta$ as the smooth analogue of cup-and-evaluate. André Weil's 1952 sheaf-theoretic reformulation, and Friedrich Hirzebruch's 1956 Topological Methods in Algebraic Geometry ^[pending], absorbed Poincaré duality into the wider framework of characteristic classes and the signature theorem. Atiyah-Singer's 1963 index theorem identified Poincaré duality, the de Rham integration pairing, and the analytic pairing of an elliptic operator with its formal adjoint as different facets of one structure. Wall's 1967 Poincaré Complexes I (Ann. Math. 86, 213-245) abstracted the duality from the manifold setting, defining a Poincaré-duality space as a CW complex satisfying the cap-product isomorphism, which became foundational for surgery theory and the classification of manifolds.

Bibliography [Master]

[object Promise]