Poincaré duality



Integration theory



Poincaré duality over some space is an equivalence (if it exists) relating cohomology with homology on that space.

The canonical example is the Poincaré duality in ordinary cohomology H (X)H^\bullet(X)/ordinary homology H (X)H_\bullet(X) which exists over an orientable closed manifold XX of dimension nn: any choice of volume form ω\omega induces by the cap product an isomorphism

()ω:H (X)H n(X). (-)\cap \omega \;\colon\; H_\bullet(X) \stackrel{\simeq}{\to} H^{n-\bullet}(X) \,.

More generally Poincaré duality is about dual objects in a generalized cohomology theory. For instance if XX above is furthermore K-oriented, hence oriented in the K-theory cohomology theory (hence if it has spin^c structure) then there is an isomorphism between its K-homology and K-theory. For more on this see at Poincaré duality algebra.

Poincaré duality is the mechanism behind Umkehr maps/push-forward in generalized cohomology: given a map of spaces f:XYf \colon X \to Y which enjoy Poincaré duality with respect to some generalized cohomology theory RR, one can pass from the canonically given pullback morphism

f *:R (Y)R (X) f^\ast \colon R^\bullet(Y) \to R^\bullet(X)

to the dual morphism

f !:R (X)R (X)f *R (Y)R (Y). f_! \colon R^\bullet(X) \simeq R_\bullet(X) \stackrel{f_\ast}{\to} R_\bullet(Y) \simeq R^\bullet(Y) \,.

Here the duality is typically exhibited in two steps:

  1. an Atiyah duality identifies the dual of R (X)R^\bullet(X) with the RR-cohomology of a Thom space of XX;

  2. a Thom isomorphism identified the RR-cohomology of the Thom space back with that of XX.

More generally, spaces XX are not self-dual in this way, but may at least be dual to themselves but equipped with a twist. This yields the twisted Umkehr maps in twisted cohomology (see for instance at Freed-Witten-Kapustin anomaly cancellation).

For more on this see at twisted Umkehr map.

Definition and statement

We first state the

and then its

Traditional formulation

Around 1895 Henri Poincaré made an observation about Betti numbers of closed manifolds, which in the 1930s was then formulated by Eduard Čech and Hassler Whitney in the following modern form:


Let XX be a closed manifold of dimension nn that is orientable. Then the cap product with any choice of orientation – in the form of a fundamental class [X]H n(X)[X] \in H_n(X) – induces isomorphisms of the form

()[X]:H k(X)H nk(X) (-)\cap [X] \;\colon\; H^k(X) \stackrel{\simeq}{\to} H_{n-k}(X)

between the ordinary cohomology and the ordinary homology groups of XX.

Later this was turned around and more general topological spaces satisfying this condition were considered


A topological space for which there is dd \in \mathbb{N} and a class [X]H d(X)[X] \in H_d(X) such that the cap product induces isomorphisms

()[X]:H (X)H d(X) (-) \cap [X] \;\colon\; H^\bullet(X) \stackrel{\simeq}{\to} H_{d-\bullet}(X)

is called a Poincaré duality space.

Refinement to homotopy theory

Traditionally Poincaré duality is stated as a duality of chain homology groups. The passage from chain complexes to their homology groups, hence the passage from full homotopy theory to just some invariants, however forgets a lot of information. But it turns out that this can always be lifted:


Let XX be a Poincaré duality space of dimension nn, def. 1. Then there is a quasi-isomorphism

C (X)Σ dC (X) C^\bullet(X) \stackrel{\simeq}{\to} \Sigma^{-d} C_\bullet(X)

between the chain complex of ordinary cohomology of XX with that dd-fold de-suspension of the chain complex of ordinary homology of XX.

This is such that its image in chain homology is (up to sign) the traditional Poincaré duality isomorphism

H (X)H d(X) H^\bullet(X) \stackrel{\simeq}{\to} H_{d -\bullet}(X)

of theorem 1. When =r\bullet=r, the sign is 1 r(r+1)/2-1^{r(r+1)/2}.

This is (EM 11, theorem, 2.5.2).


Since C (X)[C (X,),𝕀](C (X)) C^\bullet(X) \simeq [C_\bullet(X,), \mathbb{I}] \simeq (C_\bullet(X))^\vee is the dual object to C (X)C_\bullet(X) in the (∞,1)-category of unbounded chain complexes, this says that for a Poincaré duality space XX the chain complex C (X)C_\bullet(X) is almost a self-dual objects, except for a degree-shift, hence except for a twist of (itself) degree 0.


Recognition of manifolds

From [ALGTOP-L], Oct 5, 2010.

  • Jim Stasheff:

    Any one have a reference for obstructions which detect whether a space whose cohomology has Poincare duality is actually a manifold? thanks

  • John Klein:

    Ranicki’s total surgery obstruction does it in the top case, in surgery dimensions.

    • The total surgery obstruction. Algebraic topology, Aarhus 1978 (Proc. Sympos., Univ. Aarhus, Aarhus, 1978), pp. 275–316, Lecture Notes in Math., 763, Springer, Berlin, 1979. (pdf)
  • Nathanien Rounds:

    I believe this question is answered (simply connected case, over Q) in Dennis Sullivan’s paper Infinitesmal Computations in Topology (Theorem 13.2). The answer, as I understand it, is that outside dimension 4k any graded commutative algebra over Q wtih first betti number 0 satisfying Poincare Duality can be realized as the cohomoloyg ring of a manifold. In dimension 4k there is an obstruction related to the signature which is given in that paper. There won’t in general be a unique manifold corresponding to the ring; for example one can choose different rational Pontragin classes and change the homemorphism type but not the cohomology.

    The general answer (not simply connected, over Z) is given by Ranicki’s total surgery obstruction, as John Klein has already pointed out. One way to interpret this obstruction geometrically is that the Poincare duality map is always “local” but it need not have a “local” inverse, and the lack of a local inverse is an obstruction to having a anifold structure. See for example McCrory’s paper “A Characterization of Homology Manifolds” and also my thesis, which if you’re interested is here:

See also (Ranicki 96) and see at Poincaré complex.

Relation to Thom isomorphism

The Poincaré dual of a submanifold can be identified with the Thom class on its normal bundle


Relation to push-forward in cohomology

Given Poincare duality and hence (twisted-)self-dual objects, the induced dagger-structure allows torevert morphisms in cohomology. These “Umkehr maps” describe fiber integration in cohomology?.


The six operations of Grothendieck and Grothendieck duality are designed to ensure a generalized and relative versions of Poincaré duality and related phenomena in the setups like sheaf and topos theory, algebraic geometry. See Grothendieck duality for references.

Another generalization, for singular spaces, is with help of stratifications and via intersection cohomology.

Variants and generalizations



This historical work of Henri Poincaré is reviewed around page 28. pf

  • Jean Dieudonné, A History of Algebraic and Differential Topology, 1900 - 1960

Summaries of the traditional modern statement of Poincaré duality are for instance in

Discussion of generalizations to “chain duality” (see also at Poincare complex) is at

Discussion in etale cohomology is in

With an eye towards generalization in spectral geometry (spectral triples):

  • Alain Connes, page 10 of Noncommutative geometry and reality, J. Math. Phys. 36 (11), 1995 (pdf)

For equivariant cohomology

In equivariant cohomology:

For finite groups

For compact Lie groups

In equivariant K-theory:

For Hochschild cohomology

Poincaré duality on Hochschild (co)homology

  • M. Van den Bergh, A relation between Hochschild homology and cohomology for Gorenstein rings . Proc. Amer. Math. Soc. 126 (1998), 1345–1348; (JSTOR)

    Correction: Proc. Amer. Math. Soc. 130 (2002), 2809–2810.

with more on that in

  • U. Krähmer, Poincaré duality in Hochschild cohomology (pdf)

That this Poincare duality takes the Connes coboundary operator? to the BV operator is shown in

For spaces in higher geometry

Poincaré duality on Deligne-Mumford stacks/orbifolds is discussed in

For disucssion in noncommutative topology/KK-theory see at Poincaré duality algebra.

On the level of chains

Poincaré duality is traditional considered on the level of cohomology groups. One may asks if it lifts to a duality on the underlying chain complexes.

Such “derived” Poincaré duality on the level of chain complexes is claimed in theorem 2.5.2 of

based on – or at least inspired by – the discussion in (section 10) of

The article

claims a chain duality (even of A A_\infty-comodule chains but then possibly only exhibited by a bimodule) for the singular chain complex of any Poincaré duality topological space.

A review is in

Revised on September 29, 2014 02:27:44 by Anonymous Coward (