Introduction to Stable homotopy theory -- S



S4D2 – Graduate Seminar on Topology

\;\;\;\;\;\;\;\;\;\;\; Complex oriented cohomology

\;\;\;\;\;\;\;\;\;\;\; Dr. Urs Schreiber



Abstract. The category of those generalized cohomology theories that are equipped with a universal “complex orientation” happens to unify within it the abstract structure theory of stable homotopy theory with the concrete richness of the differential topology of cobordism theory and of the arithmetic geometry of formal group laws, such as elliptic curves. In the seminar we work through classical results in algebraic topology, organized such as to give in the end a first glimpse of the modern picture of chromatic homotopy theory.


Accompanying notes.

Main page: Introduction to Stable homotopy theory.


Seminar) Complex oriented cohomology

Outline. We start with two classical topics of algebraic topology that first run independently in parallel:

The development of either of these happens to give rise to the concept of spectra and via this concept it turns out that both topics are intimately related. The unification of both is our third topic


Literature. (Kochman 96).

S.1) Generalized cohomology

Idea. The concept that makes algebraic topology be about methods of homological algebra applied to topology is that of generalized homology and generalized cohomology: these are covariant functors or contravariant functors, respectively,

SpacesAb Spaces \longrightarrow Ab^{\mathbb{Z}}

from (sufficiently nice) topological spaces to \mathbb{Z}-graded abelian groups, such that a few key properties of the homotopy types of topological spaces is preserved as one passes them from Ho(Top) to the much more tractable abelian category Ab.

Literature. (Aguilar-Gitler-Prieto 02, chapters 7,8 and 12, Kochman 96, 3.4, 4.2, Schwede 12, II.6)

Generalized cohomology functors

Idea. A generalized (Eilenberg-Steenrod) cohomology theory is such a contravariant functor which satisfies the key properties exhibited by ordinary cohomology (as computed for instance by singular cohomology), notably homotopy invariance and excision, except that its value on the point is not required to be concentrated in degree 0. Dually for generalized homology. There are two versions of the axioms, one for reduced cohomology, and they are equivalent if properly set up.

An important example of a generalised cohomology theory other than ordinary cohomology is topological K-theory. The other two examples of key relevance below are cobordism cohomology and stable cohomotopy.

Literature. (Switzer 75, section 7, Aguilar-Gitler-Prieto 02, section 12 and section 9, Kochman 96, 3.4).


Reduced cohomology

The traditional formulation of reduced generalized cohomology in terms of point-set topology is this:


A reduced cohomology theory is

  1. a functor

    E˜ :(Top CW */) opAb \tilde E^\bullet \;\colon\; (Top^{\ast/}_{CW})^{op} \longrightarrow Ab^{\mathbb{Z}}

    from the opposite of pointed topological spaces (CW-complexes) to \mathbb{Z}-graded abelian groups (“cohomology groups”), in components

    E˜:(XfY)(E˜ (Y)f *E˜ (X)), \tilde E \;\colon\; (X \stackrel{f}{\longrightarrow} Y) \mapsto (\tilde E^\bullet(Y) \stackrel{f^\ast}{\longrightarrow} \tilde E^\bullet(X)) \,,
  2. equipped with a natural isomorphism of degree +1, to be called the suspension isomorphism, of the form

    σ E:E˜ ()E˜ +1(Σ) \sigma_E \;\colon\; \tilde E^\bullet(-) \overset{\simeq}{\longrightarrow} \tilde E^{\bullet +1}(\Sigma -)

such that:

  1. (homotopy invariance) If f 1,f 2:XYf_1,f_2 \colon X \longrightarrow Y are two morphisms of pointed topological spaces such that there is a (base point preserving) homotopy f 1f 2f_1 \simeq f_2 between them, then the induced homomorphisms of abelian groups are equal

    f 1 *=f 2 *. f_1^\ast = f_2^\ast \,.
  2. (exactness) For i:AXi \colon A \hookrightarrow X an inclusion of pointed topological spaces, with j:XCone(i)j \colon X \longrightarrow Cone(i) the induced mapping cone (def.), then this gives an exact sequence of graded abelian groups

    E˜ (Cone(i))j *E˜ (X)i *E˜ (A). \tilde E^\bullet(Cone(i)) \overset{j^\ast}{\longrightarrow} \tilde E^\bullet(X) \overset{i^\ast}{\longrightarrow} \tilde E^\bullet(A) \,.

(e.g. AGP 02, def. 12.1.4)

This is equivalent (prop. 1 below) to the following more succinct homotopy-theoretic definition:


A reduced generalized cohomology theory is a functor

E˜ :Ho(Top */) opAb \tilde E^\bullet \;\colon\; Ho(Top^{\ast/})^{op} \longrightarrow Ab^{\mathbb{Z}}

from the opposite of the pointed classical homotopy category (def., def.), to \mathbb{Z}-graded abelian groups, and equipped with natural isomorphisms, to be called the suspension isomorphism of the form

σ:E˜ +1(Σ)E˜ () \sigma \;\colon\; \tilde E^{\bullet +1}(\Sigma -) \overset{\simeq}{\longrightarrow} \tilde E^\bullet(-)

such that:

As a consequence (prop. 1 below), we find yet another equivalent definition:


A reduced generalized cohomology theory is a functor

E˜ :(Top */) opAb \tilde E^\bullet \;\colon\; (Top^{\ast/})^{op} \longrightarrow Ab^{\mathbb{Z}}

from the opposite of the category of pointed topological spaces to \mathbb{Z}-graded abelian groups, such that

and equipped with natural isomorphism, to be called the suspension isomorphism of the form

σ:E˜ +1(Σ)E˜ () \sigma \;\colon\; \tilde E^{\bullet +1}(\Sigma -) \overset{\simeq}{\longrightarrow} \tilde E^\bullet(-)

such that


The three definitions

  • def. 1

  • def. 2

  • def. 3

are indeed equivalent.


Regarding the equivalence of def. 1 with def. 2:

By the existence of the classical model structure on topological spaces (thm.), the characterization of its homotopy category (cor.) and the existence of CW-approximations, the homotopy invariance axiom in def. 1 is equivalent to the functor passing to the classical pointed homotopy category. In view of this and since on CW-complexes the standard topological mapping cone construction is a model for the homotopy cofiber (prop.), this gives the equivalence of the two versions of the exactness axiom.

Regarding the equivalence of def. 2 with def. 3:

This is the universal property of the classical homotopy category (thm.) which identifies it with the localization (def.) of Top */Top^{\ast/} at the weak homotopy equivalences (thm.), together with the existence of CW approximations (rmk.): jointly this says that, up to natural isomorphism, there is a bijection between functors FF and F˜\tilde F in the following diagram (which is filled by a natural isomorphism itself):

Top op F Ab γ Top F˜ Ho(Top) op(Top CW)/ \array{ Top^{op} &\overset{F}{\longrightarrow}& Ab^{\mathbb{Z}} \\ {}^{\mathllap{\gamma_{Top}}}\downarrow & \nearrow_{\mathrlap{\tilde F}} \\ Ho(Top)^{op}\simeq (Top_{CW})/_\sim }

where FF sends weak homotopy equivalences to isomorphisms and where () (-)_\sim means identifying homotopic maps.

Prop. 1 naturally suggests (e.g. Lurie 10, section 1.4) that the concept of generalized cohomology be formulated in the generality of any abstract homotopy theory (model category), not necessarily that of (pointed) topological spaces:


Let 𝒞\mathcal{C} be a model category (def.) with 𝒞 */\mathcal{C}^{\ast/} its pointed model category (prop.).

A reduced additive generalized cohomology theory on 𝒞\mathcal{C} is

  1. a functor

    E˜ :Ho(𝒞 */) opAb \tilde E^\bullet \;\colon \; Ho(\mathcal{C}^{\ast/})^{op} \longrightarrow Ab^{\mathbb{Z}}
  2. a natural isomorphism (“suspension isomorphisms”) of degree +1

    σ:E˜ E˜ +1Σ \sigma \; \colon \; \tilde E^\bullet \longrightarrow \tilde E^{\bullet+1} \circ \Sigma

such that

Finally we need the following terminology:


Let E˜ \tilde E^\bullet be a reduced cohomology theory according to either of def. 1, def. 2, def. 3 or def. 4.

We say E˜ \tilde E^\bullet is additive if in addition

  • (wedge axiom) For {X i} iI\{X_i\}_{i \in I} any set of pointed CW-complexes, then the canonical morphism

    E˜ ( iIX i) iIE˜ (X i) \tilde E^\bullet(\vee_{i \in I} X_i) \longrightarrow \prod_{i \in I} \tilde E^\bullet(X_i)

    from the functor applied to their wedge sum (def.), to the product of its values on the wedge summands, is an isomorphism.

We say E˜ \tilde E^\bullet is ordinary if its value on the 0-sphere S 0S^0 is concentrated in degree 0:

  • (Dimension) E˜ 0(𝕊 0)0\tilde E^{\bullet\neq 0}(\mathbb{S}^0) \simeq 0.

If E˜ \tilde E^\bullet is not ordinary, one also says that it is generalized or extraordinary.

A homomorphism of reduced cohomology theories

η:E˜ F˜ \eta \;\colon\; \tilde E^\bullet \longrightarrow \tilde F^\bullet

is a natural transformation between the underlying functors which is compatible with the suspension isomorphisms in that all the following squares commute

E˜ (X) η X F˜ (X) σ E σ F E˜ +1(ΣX) η ΣX F˜ +1(ΣX). \array{ \tilde E^\bullet(X) &\overset{\eta_X}{\longrightarrow}& \tilde F^\bullet(X) \\ {}^{\mathllap{\sigma_E}}\downarrow && \downarrow^{\mathrlap{\sigma_F}} \\ \tilde E^{\bullet + 1}(\Sigma X) &\overset{\eta_{\Sigma X}}{\longrightarrow}& \tilde F^{\bullet + 1}(\Sigma X) } \,.

We now discuss some constructions and consequences implied by the concept of reduced cohomology theories:


Given a generalized cohomology theory (E ,δ)(E^\bullet,\delta) on some 𝒞\mathcal{C} as in def. 4, and given a homotopy cofiber sequence in 𝒞\mathcal{C} (prop.),

XfYgZcoker(g)ΣX, X \stackrel{f}{\longrightarrow} Y \stackrel{g}{\longrightarrow} Z \stackrel{coker(g)}{\longrightarrow} \Sigma X \,,

then the corresponding connecting homomorphism is the composite

:E (X)σE +1(ΣX)coker(g) *E +1(Z). \partial \;\colon\; E^\bullet(X) \stackrel{\sigma}{\longrightarrow} E^{\bullet+1}(\Sigma X) \stackrel{coker(g)^\ast}{\longrightarrow} E^{\bullet+1}(Z) \,.

The connecting homomorphisms of def. 6 are parts of long exact sequences

E (Z)E (Y)E (X)E +1(Z). \cdots \stackrel{\partial}{\longrightarrow} E^{\bullet}(Z) \longrightarrow E^\bullet(Y) \longrightarrow E^\bullet(X) \stackrel{\partial}{\longrightarrow} E^{\bullet+1}(Z) \to \cdots \,.

By the defining exactness of E E^\bullet, def. 4, and the way this appears in def. 6, using that σ\sigma is by definition an isomorphism.

Unreduced cohomology

Given a reduced generalized cohomology theory as in def. 1, we may “un-reduce” it and evaluate it on unpointed topological spaces XX simply by evaluating it on X +X_+ (def.). It is conventional to further generalize to relative cohomology and evaluate on unpointed subspace inclusions i:AXi \colon A \hookrightarrow X, taken as placeholders for their mapping cones Cone(i +)Cone(i_+) (prop.).

In the following a pair (X,U)(X,U) refers to a subspace inclusion of topological spaces UXU \hookrightarrow X. Whenever only one space is mentioned, the subspace is assumed to be the empty set (X,)(X, \emptyset). Write Top CW Top_{CW}^{\hookrightarrow} for the category of such pairs (the full subcategory of the arrow category of Top CWTop_{CW} on the inclusions). We identify Top CWTop CW Top_{CW} \hookrightarrow Top_{CW}^{\hookrightarrow} by X(X,)X \mapsto (X,\emptyset).


A cohomology theory (unreduced, relative) is

  1. a functor

    E :(Top CW ) opAb E^\bullet : (Top_{CW}^{\hookrightarrow})^{op} \to Ab^{\mathbb{Z}}

    to the category of \mathbb{Z}-graded abelian groups,

  2. a natural transformation of degree +1, to be called the connecting homomorphism, of the form

    δ (X,A):E (A,)E +1(X,A). \delta_{(X,A)} \;\colon\; E^\bullet(A, \emptyset) \to E^{\bullet + 1}(X, A) \,.

such that:

  1. (homotopy invariance) For f:(X 1,A 1)(X 2,A 2)f \colon (X_1,A_1) \to (X_2,A_2) a homotopy equivalence of pairs, then

    E (f):E (X 2,A 2)E (X 1,A 1) E^\bullet(f) \;\colon\; E^\bullet(X_2,A_2) \stackrel{\simeq}{\longrightarrow} E^\bullet(X_1,A_1)

    is an isomorphism;

  2. (exactness) For AXA \hookrightarrow X the induced sequence

    E n(X,A)E n(X)E n(A)δE n+1(X,A) \cdots \to E^n(X, A) \longrightarrow E^n(X) \longrightarrow E^n(A) \stackrel{\delta}{\longrightarrow} E^{n+1}(X, A) \to \cdots

    is a long exact sequence of abelian groups.

  3. (excision) For UAXU \hookrightarrow A \hookrightarrow X such that U¯Int(A)\overline{U} \subset Int(A), then the natural inclusion of the pair i:(XU,AU)(X,A)i \colon (X-U, A-U) \hookrightarrow (X, A) induces an isomorphism

    E (i):E n(X,A)E n(XU,AU) E^\bullet(i) \;\colon\; E^n(X, A) \overset{\simeq}{\longrightarrow} E^n(X-U, A-U)

We say E E^\bullet is additive if it takes coproducts to products:

  • (additivity) If (X,A)= i(X i,A i)(X, A) = \coprod_i (X_i, A_i) is a coproduct, then the canonical comparison morphism

    E n(X,A) iE n(X i,A i) E^n(X, A) \overset{\simeq}{\longrightarrow} \prod_i E^n(X_i, A_i)

    is an isomorphism from the value on (X,A)(X,A) to the product of values on the summands.

We say E E^\bullet is ordinary if its value on the point is concentrated in degree 0

  • (Dimension): E 0(*,)=0E^{\bullet \neq 0}(\ast,\emptyset) = 0.

A homomorphism of unreduced cohomology theories

η:E F \eta \;\colon\; E^\bullet \longrightarrow F^\bullet

is a natural transformation of the underlying functors that is compatible with the connecting homomorphisms, hence such that all these squares commute:

E (A,) η (A,) F (A,) δ E δ F E +1(X,A) η (X,A) F +1(X,A). \array{ E^\bullet(A,\emptyset) &\overset{\eta_{(A,\emptyset)}}{\longrightarrow}& F^\bullet(A,\emptyset) \\ {}^{\mathllap{\delta_E}}\downarrow && \downarrow^{\mathrlap{\delta_F}} \\ E^{\bullet +1}(X,A) &\overset{\eta_{(X,A)}}{\longrightarrow}& F^{\bullet +1}(X,A) } \,.

e.g. (AGP 02, def. 12.1.1).


The excision axiom in def. 7 is equivalent to the following statement:

For all A,BXA,B \hookrightarrow X with X=Int(A)Int(B)X = Int(A) \cup Int(B), then the inclusion

i:(A,AB)(X,B) i \colon (A, A \cap B) \longrightarrow (X,B)

induces an isomorphism,

i *:E (X,B)E (A,AB) i^\ast \;\colon\; E^\bullet(X, B) \overset{\simeq}{\longrightarrow} E^\bullet(A, A \cap B)

(e.g Switzer 75, 7.2)


In one direction, suppose that E E^\bullet satisfies the original excision axiom. Given A,BA,B with X=Int(A)Int(B)X = \Int(A) \cup Int(B), set UXAU \coloneqq X-A and observe that

U¯ =XA¯ =XInt(A) Int(B) \begin{aligned} \overline{U} & = \overline{X-A} \\ & = X- Int(A) \\ & \subset Int(B) \end{aligned}

and that

(XU,BU)=(A,AB). (X-U, B-U) = (A, A \cap B) \,.

Hence the excision axiom implies E (X,B)E (A,AB) E^\bullet(X, B) \overset{\simeq}{\longrightarrow} E^\bullet(A, A \cap B).

Conversely, suppose E E^\bullet satisfies the alternative condition. Given UAXU \hookrightarrow A \hookrightarrow X with U¯Int(A)\overline{U} \subset Int(A), observe that we have a cover

Int(XU)Int(A) =(XU¯)Int(A) (XInt(A))Int(A) =X \begin{aligned} Int(X-U) \cup Int(A) & = (X - \overline{U}) \cap \Int(A) \\ & \supset (X - Int(A)) \cap Int(A) \\ & = X \end{aligned}

and that

(XU,(XU)A)=(XU,AU). (X-U, (X-U) \cap A) = (X-U, A - U) \,.


E (XU,AU)E (XU,(XU)A)E (X,A). E^\bullet(X-U,A-U) \simeq E^\bullet(X-U, (X-U)\cap A) \simeq E^\bullet(X,A) \,.

The following lemma shows that the dependence in pairs of spaces in a generalized cohomology theory is really a stand-in for evaluation on homotopy cofibers of inclusions.


Let E E^\bullet be an cohomology theory, def. 7, and let AXA \hookrightarrow X. Then there is an isomorphism

E (X,A)E (XCone(A),*) E^\bullet(X,A) \stackrel{\simeq}{\longrightarrow} E^\bullet(X \cup Cone(A), \ast)

between the value of E E^\bullet on the pair (X,A)(X,A) and its value on the unreduced mapping cone of the inclusion (rmk.), relative to a basepoint.

If moreover AXA \hookrightarrow X is (the retract of) a relative cell complex inclusion, then also the morphism in cohomology induced from the quotient map p:(X,A)(X/A,*)p \;\colon\; (X,A)\longrightarrow (X/A, \ast) is an isomorphism:

E (p):E (X/A,*)E (X,A). E^\bullet(p) \;\colon\; E^\bullet(X/A,\ast) \longrightarrow E^\bullet(X,A) \,.

(e.g AGP 02, corollary 12.1.10)


Consider U(Cone(A)A×{0})Cone(A)U \coloneqq (Cone(A)-A \times \{0\}) \hookrightarrow Cone(A), the cone on AA minus the base AA. We have

(XCone(A)U,Cone(A)U)(X,A) ( X\cup Cone(A)-U, Cone(A)-U) \simeq (X,A)

and hence the first isomorphism in the statement is given by the excision axiom followed by homotopy invariance (along the contraction of the cone to the point).

Next consider the quotient of the mapping cone of the inclusion:

(XCone(A),Cone(A))(X/A,*). ( X\cup Cone(A), Cone(A) ) \longrightarrow (X/A,\ast) \,.

If AXA \hookrightarrow X is a cofibration, then this is a homotopy equivalence since Cone(A)Cone(A) is contractible and since by the dual factorization lemma (lem.) and by the invariance of homotopy fibers under weak equivalences (lem.), XCone(A)X/AX \cup Cone(A)\to X/A is a weak homotopy equivalence, hence, by the universal property of the classical homotopy category (thm.) a homotopy equivalence on CW-complexes.

Hence now we get a composite isomorphism

E (X/A,*)E (XCone(A),Cone(A))E (X,A). E^\bullet(X/A,\ast) \overset{\simeq}{\longrightarrow} E^\bullet( X\cup Cone(A), Cone(A) ) \overset{\simeq}{\longrightarrow} E^\bullet(X,A) \,.

As an important special case of : Let (X,x)(X,x) be a pointed CW-complex. For p:(Cone(X),X)(ΣX,{x})p\colon (Cone(X), X) \to (\Sigma X,\{x\}) the quotient map from the reduced cone on XX to the reduced suspension, then

E (p):E (Cone(X),X)E (ΣX,{x}) E^\bullet(p) \;\colon\; E^\bullet(Cone(X),X) \overset{\simeq}{\longrightarrow} E^\bullet(\Sigma X, \{x\})

is an isomorphism.


(exact sequence of a triple)

For E E^\bullet an unreduced generalized cohomology theory, def. 7, then every inclusion of two consecutive subspaces

ZYX Z \hookrightarrow Y \hookrightarrow X

induces a long exact sequence of cohomology groups of the form

E q1(Y,Z)δ¯E q(X,Y)E q(X,Z)E q(Y,Z) \cdots \to E^{q-1}(Y,Z) \stackrel{\bar \delta}{\longrightarrow} E^q(X,Y) \stackrel{}{\longrightarrow} E^q(X,Z) \stackrel{}{\longrightarrow} E^q(Y,Z) \to \cdots


δ¯:E q1(Y,Z)E q1(Y)δE q(X,Y). \bar \delta \;\colon \; E^{q-1}(Y,Z) \longrightarrow E^{q-1}(Y) \stackrel{\delta}{\longrightarrow} E^{q}(X,Y) \,.

Apply the braid lemma to the interlocking long exact sequences of the three pairs (X,Y)(X,Y), (X,Z)(X,Z), (Y,Z)(Y,Z):

(graphics from this Maths.SE comment, showing the dual situation for homology)

See here for details.


The exact sequence of a triple in prop. 3 is what gives rise to the Cartan-Eilenberg spectral sequence for EE-cohomology of a CW-complex XX.


For (X,x)(X,x) a pointed topological space and Cone(X)=(X(I +))/XCone(X) = (X \wedge (I_+))/X its reduced cone, the long exact sequence of the triple ({x},X,Cone(X))(\{x\}, X, Cone(X)), prop. 3,

0E q(Cone(X),{x})E q(X,{x})δ¯E q+1(Cone(X),X)E q+1(Cone(X),{x})0 0 \simeq E^q(Cone(X), \{x\}) \longrightarrow E^q(X,\{x\}) \overset{\bar \delta}{\longrightarrow} E^{q+1}(Cone(X),X) \longrightarrow E^{q+1}(Cone(X), \{x\}) \simeq 0

exhibits the connecting homomorphism δ¯\bar \delta here as an isomorphism

δ¯:E q(X,{x})E q+1(Cone(X),X). \bar \delta \;\colon\; E^q(X,\{x\}) \overset{\simeq}{\longrightarrow} E^{q+1}(Cone(X),X) \,.

This is the suspension isomorphism extracted from the unreduced cohomology theory, see def. 9 below.


(Mayer-Vietoris sequence)

Given E E^\bullet an unreduced cohomology theory, def. 7. Given a topological space covered by the interior of two spaces as X=Int(A)Int(B)X = Int(A) \cup Int(B), then for each CABC \subset A \cap B there is a long exact sequence of cohomology groups of the form

E n1(AB,C)δ¯E n(X,C)E n(A,C)E n(B,C)E n(AB,C). \cdots \to E^{n-1}(A \cap B , C) \overset{\bar \delta}{\longrightarrow} E^n(X,C) \longrightarrow E^n(A,C) \oplus E^n(B,C) \longrightarrow E^n(A \cap B, C) \to \cdots \,.

e.g. (Switzer 75, theorem 7.19, Aguilar-Gitler-Prieto 02, theorem 12.1.22)

Relation between unreduced and reduced cohomology

(unreduced to reduced cohomology)

Let E E^\bullet be an unreduced cohomology theory, def. 7. Define a reduced cohomology theory, def. 1 (E˜ ,σ)(\tilde E^\bullet, \sigma) as follows.

For x:*Xx \colon \ast \to X a pointed topological space, set

E˜ (X,x)E (X,{x}). \tilde E^\bullet(X,x) \coloneqq E^\bullet(X,\{x\}) \,.

This is clearly functorial. Take the suspension isomorphism to be the composite

σ:E˜ +1(ΣX)=E +1(ΣX,{x})E (p)E +1(Cone(X),X)δ¯ 1E (X,{x})=E˜ (X) \sigma \;\colon\; \tilde E^{\bullet+1}(\Sigma X) = E^{\bullet+1}(\Sigma X, \{x\}) \overset{E^\bullet(p)}{\longrightarrow} E^{\bullet+1}(Cone(X),X) \overset{\bar \delta^{-1}}{\longrightarrow} E^\bullet(X,\{x\}) = \tilde E^{\bullet}(X)

of the isomorphism E (p)E^\bullet(p) from example 1 and the inverse of the isomorphism δ¯\bar \delta from example 2.


The construction in def. 9 indeed gives a reduced cohomology theory.

(e.g Switzer 75, 7.34)


We need to check the exactness axiom given any AXA\hookrightarrow X. By lemma 1 we have an isomorphism

E˜ (XCone(A))=E (XCone(A),{*})E (X,A). \tilde E^\bullet(X \cup Cone(A)) = E^\bullet(X \cup Cone(A), \{\ast\}) \overset{\simeq}{\longrightarrow} E^\bullet(X,A) \,.

Unwinding the constructions shows that this makes the following diagram commute:

E˜ (XCone(A)) E (X,A) E˜ (X) = E (X,{x}) E˜ (A) = E (A,{a}), \array{ \tilde E^\bullet(X\cup Cone(A)) &\overset{\simeq}{\longrightarrow}& E^\bullet(X,A) \\ \downarrow && \downarrow \\ \tilde E^\bullet(X) &=& E^\bullet(X,\{x\}) \\ \downarrow && \downarrow \\ \tilde E^\bullet(A) &=& E^\bullet(A,\{a\}) } \,,

where the vertical sequence on the right is exact by prop. 3. Hence the left vertical sequence is exact.


(reduced to unreduced cohomology)

Let (E˜ ,σ)(\tilde E^\bullet, \sigma) be a reduced cohomology theory, def. 1. Define an unreduced cohomolog theory E E^\bullet, def. 7, by

E (X,A)E˜ (X +Cone(A +)) E^\bullet(X,A) \coloneqq \tilde E^\bullet( X_+ \cup Cone(A_+))

and let the connecting homomorphism be as in def. 6.


The construction in def. 10 indeed yields an unreduced cohomology theory.

e.g. (Switzer 75, 7.35)


Exactness holds by prop. 2. For excision, it is sufficient to consider the alternative formulation of lemma 8. For CW-inclusions, this follows immediately with lemma 1.


The constructions of def. 10 and def. 9 constitute a pair of functors between then categories of reduced cohomology theories, def. 1 and unreduced cohomology theories, def. 7 which exhbit an equivalence of categories.


(…careful with checking the respect for suspension iso and connecting homomorphism..)

To see that there are natural isomorphisms relating the two composites of these two functors to the identity:

One composite is

E (E˜ :(X,x)E (X,{x})) ((E) :(X,A)E (X +Cone(A +)),*), \begin{aligned} E^\bullet & \mapsto (\tilde E^\bullet \colon (X,x) \mapsto E^\bullet(X,\{x\})) \\ & \mapsto ((E')^\bullet \colon (X,A) \mapsto E^\bullet( X_+ \cup Cone(A_+) ), \ast) \end{aligned} \,,

where on the right we have, from the construction, the reduced mapping cone of the original inclusion AXA \hookrightarrow X with a base point adjoined. That however is isomorphic to the unreduced mapping cone of the original inclusion (prop.- P#UnreducedMappingConeAsReducedConeOfBasedPointAdjoined)). With this the natural isomorphism is given by lemma 1.

The other composite is

E˜ (E :(X,A)E˜ (X +Cone(A +))) ((E˜) :XE˜ (X +Cone(* +))) \begin{aligned} \tilde E^\bullet & \mapsto (E^\bullet \colon (X,A) \mapsto \tilde E^\bullet(X_+ \cup Cone(A_+))) \\ & \mapsto ((\tilde E')^\bullet \colon X \mapsto \tilde E^\bullet(X_+ \cup Cone(*_+))) \end{aligned}

where on the right we have the reduced mapping cone of the point inclusion with a point adoined. As before, this is isomorphic to the unreduced mapping cone of the point inclusion. That finally is clearly homotopy equivalent to XX, and so now the natural isomorphism follows with homotopy invariance.

Finally we record the following basic relation between reduced and unreduced cohomology:


Let E E^\bullet be an unreduced cohomology theory, and E˜ \tilde E^\bullet its reduced cohomology theory from def. 9. For (X,*)(X,\ast) a pointed topological space, then there is an identification

E (X)E˜ (X)E (*) E^\bullet(X) \simeq \tilde E^\bullet(X) \oplus E^\bullet(\ast)

of the unreduced cohomology of XX with the direct sum of the reduced cohomology of XX and the unreduced cohomology of the base point.


The pair *X\ast \hookrightarrow X induces the sequence

E 1(*)δE˜ (X)E (X)E (*)δE˜ +1(X) \cdots \to E^{\bullet-1}(\ast) \stackrel{\delta}{\longrightarrow} \tilde E^\bullet(X) \stackrel{}{\longrightarrow} E^\bullet(X) \stackrel{}{\longrightarrow} E^\bullet(\ast) \stackrel{\delta}{\longrightarrow} \tilde E^{\bullet+1}(X) \to \cdots

which by the exactness clause in def. 7 is exact.

Now since the composite *X*\ast \to X \to \ast is the identity, the morphism E (X)E (*)E^\bullet(X) \to E^\bullet(\ast) has a section and so is in particular an epimorphism. Therefore, by exactness, the connecting homomorphism vanishes, δ=0\delta = 0 and we have a short exact sequence

0E˜ (X)E (X)E (*)0 0 \to \tilde E^\bullet(X) \stackrel{}{\longrightarrow} E^\bullet(X) \stackrel{}{\longrightarrow} E^\bullet(\ast) \to 0

with the right map an epimorphism. Hence this is a split exact sequence and the statement follows.

Generalized homology functors

All of the above has a dual version with generalized cohomology replaced by generalized homology. For ease of reference, we record these dual definitions:


A reduced homology theory is a functor

E˜ :(Top CW */)Ab \tilde E_\bullet \;\colon\; (Top^{\ast/}_{CW}) \longrightarrow Ab^{\mathbb{Z}}

from the category of pointed topological spaces (CW-complexes) to \mathbb{Z}-graded abelian groups (“homology groups”), in components

E˜ :(XfY)(E˜ (X)f *E˜ (Y)), \tilde E _\bullet \;\colon\; (X \stackrel{f}{\longrightarrow} Y) \mapsto (\tilde E_\bullet(X) \stackrel{f_\ast}{\longrightarrow} \tilde E_\bullet(Y)) \,,

and equipped with a natural isomorphism of degree +1, to be called the suspension isomorphism, of the form

σ:E˜ ()E˜ +1(Σ) \sigma \;\colon\; \tilde E_\bullet(-) \overset{\simeq}{\longrightarrow} \tilde E_{\bullet +1}(\Sigma -)

such that:

  1. (homotopy invariance) If f 1,f 2:XYf_1,f_2 \colon X \longrightarrow Y are two morphisms of pointed topological spaces such that there is a (base point preserving) homotopy f 1f 2f_1 \simeq f_2 between them, then the induced homomorphisms of abelian groups are equal

    f 1*=f 2*. f_1_\ast = f_2_\ast \,.
  2. (exactness) For i:AXi \colon A \hookrightarrow X an inclusion of pointed topological spaces, with j:XCone(i)j \colon X \longrightarrow Cone(i) the induced mapping cone, then this gives an exact sequence of graded abelian groups

    E˜ (A)i *E˜ (X)j *E˜ (Cone(i)). \tilde E_\bullet(A) \overset{i_\ast}{\longrightarrow} \tilde E_\bullet(X) \overset{j_\ast}{\longrightarrow} \tilde E_\bullet(Cone(i)) \,.

We say E˜ \tilde E_\bullet is additive if in addition

  • (wedge axiom) For {X i} iI\{X_i\}_{i \in I} any set of pointed CW-complexes, then the canonical morphism

    iIE˜ (X i)E˜ ( iIX i) \oplus_{i \in I} \tilde E_\bullet(X_i) \longrightarrow \tilde E^\bullet(\vee_{i \in I} X_i)

    from the direct sum of the value on the summands to the value on the wedge sum (prop.- P#WedgeSumAsCoproduct)), is an isomorphism.

We say E˜ \tilde E_\bullet is ordinary if its value on the 0-sphere S 0S^0 is concentrated in degree 0:

  • (Dimension) E˜ 0(𝕊 0)0\tilde E_{\bullet\neq 0}(\mathbb{S}^0) \simeq 0.

A homomorphism of reduced cohomology theories

η:E˜ F˜ \eta \;\colon\; \tilde E_\bullet \longrightarrow \tilde F_\bullet

is a natural transformation between the underlying functors which is compatible with the suspension isomorphisms in that all the following squares commute

E˜ (X) η X F˜ (X) σ E σ F E˜ +1(ΣX) η ΣX F˜ +1(ΣX). \array{ \tilde E_\bullet(X) &\overset{\eta_X}{\longrightarrow}& \tilde F_\bullet(X) \\ {}^{\mathllap{\sigma_E}}\downarrow && \downarrow^{\mathrlap{\sigma_F}} \\ \tilde E_{\bullet + 1}(\Sigma X) &\overset{\eta_{\Sigma X}}{\longrightarrow}& \tilde F_{\bullet + 1}(\Sigma X) } \,.

A homology theory (unreduced, relative) is a functor

E :(Top CW )Ab E_\bullet : (Top_{CW}^{\hookrightarrow}) \longrightarrow Ab^{\mathbb{Z}}

to the category of \mathbb{Z}-graded abelian groups, as well as a natural transformation of degree +1, to be called the connecting homomorphism, of the form

δ (X,A):E +1(X,A)E (A,). \delta_{(X,A)} \;\colon\; E_{\bullet + 1}(X, A) \longrightarrow E^\bullet(A, \emptyset) \,.

such that:

  1. (homotopy invariance) For f:(X 1,A 1)(X 2,A 2)f \colon (X_1,A_1) \to (X_2,A_2) a homotopy equivalence of pairs, then

    E (f):E (X 1,A 1)E (X 2,A 2) E_\bullet(f) \;\colon\; E_\bullet(X_1,A_1) \stackrel{\simeq}{\longrightarrow} E_\bullet(X_2,A_2)

    is an isomorphism;

  2. (exactness) For AXA \hookrightarrow X the induced sequence

    E n+1(X,A)δE n(A)E n(X)E n(X,A) \cdots \to E_{n+1}(X, A) \stackrel{\delta}{\longrightarrow} E_n(A) \longrightarrow E_n(X) \longrightarrow E_n(X, A) \to \cdots

    is a long exact sequence of abelian groups.

  3. (excision) For UAXU \hookrightarrow A \hookrightarrow X such that U¯Int(A)\overline{U} \subset Int(A), then the natural inclusion of the pair i:(XU,AU)(X,A)i \colon (X-U, A-U) \hookrightarrow (X, A) induces an isomorphism

    E (i):E n(XU,AU)E n(X,A) E_\bullet(i) \;\colon\; E_n(X-U, A-U) \overset{\simeq}{\longrightarrow} E_n(X, A)

We say E E^\bullet is additive if it takes coproducts to direct sums:

  • (additivity) If (X,A)= i(X i,A i)(X, A) = \coprod_i (X_i, A_i) is a coproduct, then the canonical comparison morphism

    iE n(X i,A i)E n(X,A) \oplus_i E^n(X_i, A_i) \overset{\simeq}{\longrightarrow} E^n(X, A)

    is an isomorphismfrom the direct sum of the value on the summands, to the value on the total pair.

We say E E_\bullet is ordinary if its value on the point is concentrated in degree 0

  • (Dimension): E 0(*,)=0E_{\bullet \neq 0}(\ast,\emptyset) = 0.

A homomorphism of unreduced homology theories

η:E F \eta \;\colon\; E_\bullet \longrightarrow F_\bullet

is a natural transformation of the underlying functors that is compatible with the connecting homomorphisms, hence such that all these squares commute:

E +1(X,A) η (X,A) F +1(X,A) δ E δ F E (A,) η (A,) F (A,). \array{ E_{\bullet +1}(X,A) &\overset{\eta_{(X,A)}}{\longrightarrow}& F_{\bullet +1}(X,A) \\ {}^{\mathllap{\delta_E}}\downarrow && \downarrow^{\mathrlap{\delta_F}} \\ E_\bullet(A,\emptyset) &\overset{\eta_{(A,\emptyset)}}{\longrightarrow}& F^\bullet(A,\emptyset) } \,.
Multiplicative cohomology theories

The generalized cohomology theories considered above assign cohomology groups. It is familiar from ordinary cohomology with coefficients not just in a group but in a ring, that also the cohomology groups inherit compatible ring structure. The generalization of this phenomenon to generalized cohomology theories is captured by the concept of multiplicative cohomology theories:


Let E 1,E 2,E 3E_1, E_2, E_3 be three unreduced generalized cohomology theories (def.). A pairing of cohomology theories

μ:E 1E 2E 3 \mu \;\colon\; E_1 \Box E_2 \longrightarrow E_3

is a natural transformation (of functors on (Top CW ×Top CW ) op(Top_{CW}^{\hookrightarrow}\times Top_{CW}^{\hookrightarrow})^{op} ) of the form

μ n 1,n 2:E 1 n 1(X,A)E 2 n 2(Y,B)E 3 n 1+n 2(X×Y,A×YX×B) \mu_{n_1,n_2} \;\colon\; E_1^{n_1}(X,A) \otimes E_2^{n_2}(Y,B) \longrightarrow E_3^{n_1 + n_2}(X\times Y \;,\; A\times Y \cup X \times B)

such that this is compatible with the connecting homomorphisms δ i\delta_i of E iE_i, in that the following are commuting squares

E 1 n 1(A)E 2 n 2(Y,B) δ 1id 2 E 1 n 1+1(X,A)E 2 n 2(Y,B) μ n 1,n 2 μ n 1+1,n 2 E 3 n 1+n 2(A×YX×B,X×B)E 3 n 1+n 2(A×Y,A×B) δ 3 E 3 n 1+n 2+1(X×Y,A×B) \array{ E_1^{n_1}(A) \otimes E_2^{n_2}(Y,B) &\overset{\delta_1 \otimes id_2}{\longrightarrow}& E_1^{n_1+1}(X,A) \otimes E_2^{n_2}(Y,B) \\ {}^{\mathllap{\mu_{n_1,n_2}}}\downarrow && \downarrow^{\mathrlap{\mu_{n_1+1, n_2}}} \\ \underoverset {E_3^{n_1 + n_2}(A \times Y \cup X \times B , X \times B)} {E_3^{n_1 + n_2}(A \times Y, A \times B)} {\simeq} &\overset{\delta_3}{\longrightarrow}& E_3^{n_1 + n_2+ 1}(X \times Y, A \times B) }


E 1 n 1(X,A)E 2 n 2(B) (1) n 1id 1δ 2 E 1 n 1+1(X,A)E 2 n 2(Y,B) μ n 1,n 2 μ n 1,n 2+1 E 3 n 1+n 2(A×YX×B,A×Y)E 3 n 1+n 2(X×B,A×B) δ 3 E 3 n 1+n 2+1(X×Y,A×B), \array{ E_1^{n_1}(X,A) \otimes E_2^{n_2}(B) &\overset{(-1)^{n_1} id_1 \otimes \delta_2}{\longrightarrow}& E_1^{n_1+1}(X,A) \otimes E_2^{n_2}(Y,B) \\ {}^{\mathllap{\mu_{n_1,n_2}}}\downarrow && \downarrow^{\mathrlap{\mu_{n_1, n_2 + 1}}} \\ \underoverset {E_3^{n_1 + n_2}(A \times Y \cup X \times B , A \times Y)} {E_3^{n_1 + n_2}(X \times B, A \times B)} {\simeq} &\overset{\delta_3}{\longrightarrow}& E_3^{n_1 + n_2+ 1}(X \times Y, A \times B) } \,,

where the isomorphisms in the bottom left are the excision isomorphisms.


An (unreduced) multiplicative cohomology theory is an unreduced generalized cohomology theory theory EE (def. 7) equipped with

  1. (external multiplication) a pairing (def. 13) of the form μ:EEE\mu \;\colon\; E \Box E \longrightarrow E;

  2. (unit) an element 1E 0(*)1 \in E^0(\ast)

such that

  1. (associativity) μ(idμ)=μ(μid)\mu \circ (id \otimes \mu) = \mu \circ (\mu \otimes id);

  2. (unitality) μ(1x)=μ(x1)=x\mu(1\otimes x) = \mu(x \otimes 1) = x for all xE n(X,A)x \in E^n(X,A).

The mulitplicative cohomology theory is called commutative (often considered by default) if in addition

  • (graded commutativity)

    E n 1(X,A)E n 2(Y,B) (uv)(1) n 1n 2(vu) E n 2(Y,B)E X,A n 1 μ n 1,n 2 μ n 2,n 1 E n 1+n 2(X×Y,A×YX×B) (switch (X,A),(Y,B)) * E n 1+n 2(Y×X,B×XY×A). \array{ E^{n_1}(X,A) \otimes E^{n_2}(Y,B) &\overset{(u \otimes v) \mapsto (-1)^{n_1 n_2} (v \otimes u) }{\longrightarrow}& E^{n_2}(Y,B) \otimes E^{n_1}_{X,A} \\ {}^{\mathllap{\mu_{n_1,n_2}}}\downarrow && \downarrow^{\mathrlap{\mu_{n_2,n_1}}} \\ E^{n_1 + n_2}( X \times Y , A \times Y \cup X \times B) &\underset{(switch_{(X,A), (Y,B)})^\ast}{\longrightarrow}& E^{n_1 + n_2}( Y \times X , B \times X \cup Y \times A) } \,.

Given a multiplicative cohomology theory (E,μ,1)(E, \mu, 1), its cup product is the composite of the above external multiplication with pullback along the diagonal maps Δ (X,A):(X,A)(X×X,A×XX×A)\Delta_{(X,A)} \colon (X,A) \longrightarrow (X\times X, A \times X \cup X \times A);

()():E n 1(X,A)E n 2(X,A)μ n 1,n 2E n 1+n 2(X×X,A×XX×A)Δ (X,A) *E n 1+n 2(X,AB). (-) \cup (-) \;\colon\; E^{n_1}(X,A) \otimes E^{n_2}(X,A) \overset{\mu_{n_1,n_2}}{\longrightarrow} E^{n_1 + n_2}( X \times X, \; A \times X \cup X \times A) \overset{\Delta^\ast_{(X,A)}}{\longrightarrow} E^{n_1 + n_2}(X, \; A \cup B) \,.

e.g. (Tamaki-Kono 06, II.6)


Let (E,μ,1)(E,\mu,1) be a multiplicative cohomology theory, def. 14. Then

  1. For every space XX the cup product gives E (X)E^\bullet(X) the structure of a \mathbb{Z}-graded ring, which is graded-commutative if (E,μ,1)(E,\mu,1) is commutative.

  2. For every pair (X,A)(X,A) the external multiplication μ\mu gives E (X,A)E^\bullet(X,A) the structure of a left and right module over the graded ring E (*)E^\bullet(\ast).

  3. All pullback morphisms respect the left and right action of E (*)E^\bullet(\ast) and the connecting homomorphisms respect the right action and the left action up to multiplication by (1) n 1(-1)^{n_1}


Regarding the third point:

For pullback maps this is the naturality of the external product: let f:(X,A)(Y,B)f \colon (X,A) \longrightarrow (Y,B) be a morphism in Top CW Top_{CW}^{\hookrightarrow} then naturality says that the following square commutes:

E n 1(*)E n 2(Y,B) μ n 1,n 2 E n 1+n 2(Y,B) (id,f *) f * E n 1(*)E n 2(X,A) μ n 1,n 2 E n 1+n 2(Y,B). \array{ E^{n_1}(\ast) \otimes E^{n_2}(Y,B) &\overset{\mu_{n_1,n_2}}{\longrightarrow}& E^{n_1 + n_2}(Y, B) \\ {}^{\mathllap{(id,f^\ast)}}\downarrow && \downarrow^{\mathrlap{f^\ast}} \\ E^{n_1}(\ast) \otimes E^{n_2}(X,A) &\overset{\mu_{n_1,n_2}}{\longrightarrow}& E^{n_1 + n_2}(Y,B) } \,.

For connecting homomorphisms this is the (graded) commutativity of the squares in def. 14:

E n 1(*)E n 2(A) (1) n 1(id,δ) E n 1(*)E n 2+2(X) μ n 1,n 2 μ n 1,n 2 E n 1+n 2(A) δ E 3 n 1+n 2+1(X,B). \array{ E^{n_1}(\ast)\otimes E^{n_2}(A) &\overset{(-1)^{n_1} (id, \delta)}{\longrightarrow}& E^{n_1}(\ast) \otimes E^{n_2 + 2}(X) \\ {}^{\mathllap{\mu_{n_1,n_2}}}\downarrow && \downarrow^{\mathrlap{\mu_{n_1,n_2}}} \\ E^{n_1 + n_2}(A) &\overset{\delta}{\longrightarrow}& E_3^{n_1 + n_2+ 1}(X,B) } \,.

Brown representability theorem

Idea. Given any functor such as the generalized (co)homology functor above, an important question to ask is whether it is a representable functor. Due to the \mathbb{Z}-grading and the suspension isomorphisms, if a generalized (co)homology functor is representable at all, it must be represented by a \mathbb{Z}-indexed sequence of pointed topological spaces such that the reduced suspension of one is comparable to the next one in the list. This is a spectrum or more specifically: a sequential spectrum .

Whitehead observed that indeed every spectrum represents a generalized (co)homology theory. The Brown representability theorem states that, conversely, every generalized (co)homology theory is represented by a spectrum, subject to conditions of additivity.

As a first application, Eilenberg-MacLane spectra representing ordinary cohomology may be characterized via Brown representability.

Literature. (Switzer 75, section 9, Aguilar-Gitler-Prieto 02, section 12, Kochman 96, 3.4)

Traditional discussion

Write Top 1 */Top */Top_{{\geq 1}}^{\ast/} \hookrightarrow Top^{\ast/} for the full subcategory of connected pointed topological spaces. Write Set */Set^{\ast/} for the category of pointed sets.


A Brown functor is a functor

F:Ho(Top 1 */) opSet */ F\;\colon \; Ho(Top_{\geq 1}^{\ast/})^{op} \longrightarrow Set^{\ast/}

(from the opposite of the classical homotopy category (def., def.) of connected pointed topological spaces) such that

  1. (additivity) FF takes small coproducts (wedge sums) to products;

  2. (Mayer-Vietoris) If X=Int(A)Int(B)X = Int(A) \cup Int(B) then for all x AF(A)x_A \in F(A) and x BF(B)x_B \in F(B) such that (x A)| AB=(x B)| AB(x_A)|_{A \cap B} = (x_B)|_{A \cap B} then there exists x XF(X)x_X \in F(X) such that x A=(x X)| Ax_A = (x_X)|_A and x B=(x X)| Bx_B = (x_X)|_B.


For every additive reduced cohomology theory E˜ ():Ho(Top */) opSet */\tilde E^\bullet(-) \colon Ho(Top^{\ast/})^{op}\to Set^{\ast/} (def. 2) and for each degree nn \in \mathbb{N}, the restriction of E˜ n()\tilde E^n(-) to connected spaces is a Brown functor (def. 15).


Under the relation between reduced and unreduced cohomology above, this follows from the exactness of the Mayer-Vietoris sequence of prop. 4.


(Brown representability)

Every Brown functor FF (def. 15) is representable, hence there exists XTop 1 */X \in Top_{\geq 1}^{\ast/} and a natural isomorphism

[,X] *F() [-,X]_{\ast} \overset{\simeq}{\longrightarrow} F(-)

(where [,] *[-,-]_\ast denotes the hom-functor of Ho(Top 1 */)Ho(Top_{\geq 1}^{\ast/}) (exmpl.)).

(e.g. AGP 02, theorem 12.2.22)


A key subtlety in theorem 2 is the restriction to connected pointed topological spaces in def. 15. This comes about since the proof of the theorem requires that continuous functions f:XYf \colon X \longrightarrow Y that induce isomorphisms on pointed homotopy classes

[S n,X] *[S n,Y] * [S^n,X]_\ast \longrightarrow [S^n,Y]_\ast

for all nn are weak homotopy equivalences (For instance in AGP 02 this is used in the proof of theorem 12.2.19 there). But [S n,X] *=π n(X,x)[S^n,X]_\ast = \pi_n(X,x) gives the nnth homotopy group of XX only for the canonical basepoint, while for a weak homotopy equivalence in general one needs to consider the homotopy groups at all possible basepoints, at least one for each connected component. But so if one does assume that all spaces involved are connected, hence only have one connected component, then indeed weak homotopy equivalences are equivalently those maps XYX\to Y making all the [S n,X] *[S n,Y] *[S^n,X]_\ast \longrightarrow [S^n,Y]_\ast into isomorphisms.

See also example 5 below.

The representability result applied degreewise to an additive reduced cohomology theory will yield (prop. 10 below) the following concept.


An Omega-spectrum XX (def.) is

  1. a sequence {X n} n\{X_n\}_{n \in \mathbb{N}} of pointed topological spaces X nTop */X_n \in Top^{\ast/}

  2. weak homotopy equivalences

    σ˜ n:X nW clσ˜ nΩX n+1 \tilde \sigma_n \;\colon\; X_n \underoverset{\in W_{cl}}{\tilde \sigma_n}{\longrightarrow} \Omega X_{n+1}

    for each nn \in \mathbb{N}, form each space to the loop space of the following space.


Every additive reduced cohomology theory E˜ ():(Top CW *) opAb \tilde E^\bullet(-) \colon (Top_{CW}^\ast)^{op} \longrightarrow Ab^{\mathbb{Z}} according to def. 2, is represented by an Omega-spectrum EE (def. 16) in that in each degree nn \in \mathbb{N}

  1. E˜ n()\tilde E^n(-) is represented by some E nHo(Top */)E_n \in Ho(Top^{\ast/});

  2. the suspension isomorphism σ n\sigma_n of E˜ \tilde E^\bullet is represented by the structure map σ˜ n\tilde \sigma_n of the Omega-spectrum in that for all XTop */X \in Top^{\ast/} the following diagram commutes:

    E˜ n(X) σ n(X) E˜ n+1(ΣX) [X,E n] * [X,σ˜ n] * [X,ΩE n+1] * [ΣX,E n+1] *, \array{ \tilde E^{n}(X) &\overset{\sigma_n(X)}{\longrightarrow}& &\longrightarrow& \tilde E^{n+1}(\Sigma X) \\ {}^{\mathllap{\simeq}}\downarrow && && \downarrow^{\mathrlap{\simeq}} \\ [X,E_n]_\ast &\overset{[X,\tilde \sigma_n]_\ast}{\longrightarrow}& [X, \Omega E_{n+1}]_\ast &\simeq& [\Sigma X, E_{n+1}]_\ast } \,,

    where [,] *Hom Ho(Top 1 */)[-,-]_\ast \coloneqq Hom_{Ho(Top_{\geq 1}^{\ast/})} denotes the hom-sets in the classical pointed homotopy category (def.) and where in the bottom right we have the (ΣΩ)(\Sigma\dashv \Omega)-adjunction isomorphism (prop.).


If it were not for the connectedness clause in def. 15 (remark 2), then theorem 2 with prop. 9 would immediately give the existence of the {E n} n\{E_n\}_{n \in \mathbb{N}} and the remaining statement would follow immediately with the Yoneda lemma, which says in particular that morphisms between representable functors are in natural bijection with the morphisms of objects that represent them.

The argument with the connectivity condition in Brown representability taken into account is essentially the same, just with a little bit more care:

For XX a pointed topological space, write X (0)X^{(0)} for the connected component of its basepoint. Observe that the loop space of a pointed topological space only depends on this connected component:

ΩXΩ(X (0)). \Omega X \simeq \Omega (X^{(0)}) \,.

Now for nn \in \mathbb{N}, to show that E˜ n()\tilde E^n(-) is representable by some E nHo(Top */)E_n \in Ho(Top^{\ast/}), use first that the restriction of E˜ n+1\tilde E^{n+1} to connected spaces is represented by some E n+1 (0)E_{n+1}^{(0)}. Observe that the reduced suspension of any XTop */X \in Top^{\ast/} lands in Top 1 */Top_{\geq 1}^{\ast/}. Therefore the (ΣΩ)(\Sigma\dashv \Omega)-adjunction isomorphism (prop.) implies that E˜ n+1(Σ())\tilde E^{n+1}(\Sigma(-)) is represented on all of Top */Top^{\ast/} by ΩE n+1 (0)\Omega E_{n+1}^{(0)}:

E˜ n+1(ΣX)[ΣX,E n+1 (0)] *[X,ΩE n+1 (0)] *[X,ΩE n+1] *, \tilde E^{n+1}(\Sigma X) \simeq [\Sigma X, E_{n+1}^{(0)}]_\ast \simeq [X, \Omega E_{n+1}^{(0)}]_\ast \simeq [X, \Omega E_{n+1}]_\ast \,,

where E n+1E_{n+1} is any pointed topological space with the given connected component E n+1 (0)E_{n+1}^{(0)}.

Now the suspension isomorphism of E˜\tilde E says that E nHo(Top */)E_n \in Ho(Top^{\ast/}) representing E˜ n\tilde E^n exists and is given by ΩE n+1 (0)\Omega E_{n+1}^{(0)}:

E˜ n(X)E˜ n+1(Σ,X)[X,ΩE n+1] \tilde E^n(X) \simeq \tilde E^{n+1}(\Sigma, X)\simeq [X,\Omega E_{n+1}]

for any E n+1E_{n+1} with connected component E n+1 (0)E_{n+1}^{(0)}.

This completes the proof. Notice that running the same argument next for (n+1)(n+1) gives a representing space E n+1E_{n+1} such that its connected component of the base point is E n+1 (0)E_{n+1}^{(0)} found before. And so on.



Every Omega-spectrum EE, def. 16, represents an additive reduced cohomology theory def. 1 E˜ \tilde E^\bullet by

E˜ n(X)[X,E n] * \tilde E^n(X) \coloneqq [X,E_n]_\ast

with suspension isomorphism given by

σ n:E˜ n(X)=[X,E n] *[X,σ˜ n][X,ΩE n+1] *[ΣX,E n+1]=E˜ n+1(ΣX). \sigma_n \;\colon\; \tilde E^n(X) = [X,E_n]_\ast \overset{[X,\tilde \sigma_n]}{\longrightarrow} [X, \Omega E_{n+1}]_\ast \overset{\simeq}{\to} [\Sigma X, E_{n+1}] = \tilde E^{n+1}(\Sigma X) \,.

The additivity is immediate from the construction. The exactnes follows from the long exact sequences of homotopy cofiber sequences given by this prop..


If we consider the stable homotopy category Ho(Spectra)Ho(Spectra) of spectra (def.) and consider any topological space XX in terms of its suspension spectrum Σ XHo(Spectra)\Sigma^\infty X \in Ho(Spectra) (exmpl.), then the statement of prop. 11 is more succinctly summarized by saying that the graded reduced cohomology groups of a topological space XX represented by an Omega-spectrum EE are the hom-groups

E˜ (X)[Σ X,Σ E] \tilde E^\bullet(X) \;\simeq\; [\Sigma^\infty X, \Sigma^\bullet E]

in the stable homotopy category, into all the suspensions (thm.) of EE.

This means that more generally, for XHo(Spectra)X \in Ho(Spectra) any spectrum, it makes sense to consider

E˜ (X)[X,Σ E] \tilde E^\bullet(X) \;\coloneqq\; [X,\Sigma^\bullet E]

to be the graded reduced generalized EE-cohomology groups of the spectrum XX.

See also in part 1 this example.

Application to ordinary cohomology

Let AA be an abelian group. Consider singular cohomology H n(,A)H^n(-,A) with coefficients in AA. The corresponding reduced cohomology evaluated on n-spheres satisfies

H˜ n(S q,A){A ifq=n 0 otherwise \tilde H^n(S^q,A) \simeq \left\{ \array{ A & if \; q = n \\ 0 & otherwise } \right.

Hence singular cohomology is a generalized cohomology theory which is “ordinary cohomology” in the sense of def. 5.

Applying the Brown representability theorem as in prop. 10 hence produces an Omega-spectrum (def. 16) whose nnth component space is characterized as having homotopy groups concentrated in degree nn on AA. These are called Eilenberg-MacLane spaces K(A,n)K(A,n)

π q(K(A,n)){A ifq=n 0 otherwise. \pi_q(K(A,n)) \simeq \left\{ \array{ A & if \; q = n \\ 0 & otherwise } \right. \,.

Here for n>0n \gt 0 then K(A,n)K(A,n) is connected, therefore with an essentially unique basepoint, while K(A,0)K(A,0) is (homotopy equivalent to) the underlying set of the group AA.

Such spectra are called Eilenberg-MacLane spectra HAH A:

(HA) nK(A,n). (H A)_n \simeq K(A,n) \,.

As a consequence of example 3 one obtains the uniqueness result of Eilenberg-Steenrod:


Let E˜ 1\tilde E_1 and E˜ 2\tilde E_2 be ordinary (def. 5) generalized (Eilenberg-Steenrod) cohomology theories. If there is an isomorphism

E˜ 1(S 0)E˜ 2(S 0) \tilde E_1(S^0) \simeq \tilde E_2(S^0)

of cohomology groups of the 0-sphere, then there is an isomorphism of cohomology theories

E˜ 1E˜ 2. \tilde E_1 \overset{\simeq}{\longrightarrow} \tilde E_2 \,.

(e.g. Aguilar-Gitler-Prieto 02, theorem 12.3.6)

Homotopy-theoretic discussion

Using abstract homotopy theory in the guise of model category theory (see the lecture notes on classical homotopy theory), the traditional proof and further discussion of the Brown representability theorem above becomes more transparent (Lurie 10, section 1.4.1, for exposition see also Mathew 11).

This abstract homotopy-theoretic proof uses the general concept of homotopy colimits in model categories as well as the concept of derived hom-spaces (“∞-categories”). Even though in the accompanying Lecture notes on classical homotopy theory these concepts are only briefly indicated, the following is included for the interested reader.


Let 𝒞\mathcal{C} be a model category. A functor

F:Ho(𝒞) opSet F \;\colon\; Ho(\mathcal{C})^{op} \longrightarrow Set

(from the opposite of the homotopy category of 𝒞\mathcal{C} to Set)

is called a Brown functor if

  1. it sends small coproducts to products;

  2. it sends homotopy pushouts in 𝒞Ho(𝒞)\mathcal{C}\to Ho(\mathcal{C}) to weak pullbacks in Set (see remark 4).


A weak pullback is a diagram that satisfies the existence clause of a pullback, but not necessarily the uniqueness condition. Hence the second clause in def. 17 says that for a homotopy pushout square

Z X Y XZY \array{ Z &\longrightarrow& X \\ \downarrow &\swArrow& \downarrow \\ Y &\longrightarrow& X \underset{Z}{\sqcup}Y }

in 𝒞\mathcal{C}, then the induced universal morphism

F(XZY)epiF(X)×F(Z)F(Y) F\left(X \underset{Z}{\sqcup}Y\right) \stackrel{epi}{\longrightarrow} F(X) \underset{F(Z)}{\times} F(Y)

into the actual pullback is an epimorphism.


Say that a model category 𝒞\mathcal{C} is compactly generated by cogroup objects closed under suspensions if

  1. 𝒞\mathcal{C} is generated by a set

    {S i𝒞} iI \{S_i \in \mathcal{C}\}_{i \in I}

    of compact objects (i.e. every object of 𝒞\mathcal{C} is a homotopy colimit of the objects S iS_i.)

  2. each S iS_i admits the structure of a cogroup object in the homotopy category Ho(𝒞)Ho(\mathcal{C});

  3. the set {S i}\{S_i\} is closed under forming reduced suspensions.


(suspensions are H-cogroup objects)

Let 𝒞\mathcal{C} be a model category and 𝒞 */\mathcal{C}^{\ast/} its pointed model category (prop.) with zero object (rmk.). Write Σ:X0X0\Sigma \colon X \mapsto 0 \underset{X}{\coprod} 0 for the reduced suspension functor.

Then the fold map

ΣXΣX0X0X00XXX00X0ΣX \Sigma X \coprod \Sigma X \simeq 0 \underset{X}{\sqcup} 0 \underset{X}{\sqcup} 0 \longrightarrow 0 \underset{X}{\sqcup} X \underset{X}{\sqcup} 0 \simeq 0 \underset{X}{\sqcup} 0 \simeq \Sigma X

exhibits cogroup structure on the image of any suspension object ΣX\Sigma X in the homotopy category.

This is equivalently the group-structure of the first (fundamental) homotopy group of the values of functor co-represented by ΣX\Sigma X:

Ho(𝒞)(ΣX,):YHo(𝒞)(ΣX,Y)Ho(𝒞)(X,ΩY)π 1Ho(𝒞)(X,Y). Ho(\mathcal{C})(\Sigma X, -) \;\colon\; Y \mapsto Ho(\mathcal{C})(\Sigma X, Y) \simeq Ho(\mathcal{C})(X, \Omega Y) \simeq \pi_1 Ho(\mathcal{C})(X, Y) \,.

In bare pointed homotopy types 𝒞=Top Quillen */\mathcal{C} = Top^{\ast/}_{Quillen}, the (homotopy types of) n-spheres S nS^n are cogroup objects for n1n \geq 1, but not for n=0n = 0, by example 4. And of course they are compact objects.

So while {S n} n\{S^n\}_{n \in \mathbb{N}} generates all of the homotopy theory of Top */Top^{\ast/}, the latter is not an example of def. 18 due to the failure of S 0S^0 to have cogroup structure.

Removing that generator, the homotopy theory generated by {S n} nn1\{S^n\}_{{n \in \mathbb{N}} \atop {n \geq 1}} is Top 1 */Top^{\ast/}_{\geq 1}, that of connected pointed homotopy types. This is one way to see how the connectedness condition in the classical version of Brown representability theorem arises. See also remark 2 above.

See also (Lurie 10, example

In homotopy theories compactly generated by cogroup objects closed under forming suspensions, the following strenghtening of the Whitehead theorem holds.


In a homotopy theory compactly generated by cogroup objects {S i} iI\{S_i\}_{i \in I} closed under forming suspensions, according to def. 18, a morphism f:XYf\colon X \longrightarrow Y is an equivalence precisely if for each iIi \in I the induced function of maps in the homotopy category

Ho(𝒞)(S i,f):Ho(𝒞)(S i,X)Ho(𝒞)(S i,Y) Ho(\mathcal{C})(S_i,f) \;\colon\; Ho(\mathcal{C})(S_i,X) \longrightarrow Ho(\mathcal{C})(S_i,Y)

is an isomorphism (a bijection).

(Lurie 10, p. 114, Lemma star)


By the ∞-Yoneda lemma, the morphism ff is a weak equivalence precisely if for all objects A𝒞A \in \mathcal{C} the induced morphism of derived hom-spaces

𝒞(A,f):𝒞(A,X)𝒞(A,Y) \mathcal{C}(A,f) \;\colon\; \mathcal{C}(A,X) \longrightarrow \mathcal{C}(A,Y)

is an equivalence in Top QuillenTop_{Quillen}. By assumption of compact generation and since the hom-functor 𝒞(,)\mathcal{C}(-,-) sends homotopy colimits in the first argument to homotopy limits, this is the case precisely already if it is the case for A{S i} iIA \in \{S_i\}_{i \in I}.

Now the maps

𝒞(S i,f):𝒞(S i,X)𝒞(S i,Y) \mathcal{C}(S_i,f) \;\colon\; \mathcal{C}(S_i,X) \longrightarrow \mathcal{C}(S_i,Y)

are weak equivalences in Top QuillenTop_{Quillen} if they are weak homotopy equivalences, hence if they induce isomorphisms on all homotopy groups π n\pi_n for all basepoints.

It is this last condition of testing on all basepoints that the assumed cogroup structure on the S iS_i allows to do away with: this cogroup structure implies that 𝒞(S i,)\mathcal{C}(S_i,-) has the structure of an HH-group, and this implies (by group multiplication), that all connected components have the same homotopy groups, hence that all homotopy groups are independent of the choice of basepoint, up to isomorphism.

Therefore the above morphisms are equivalences precisely if they are so under applying π n\pi_n based on the connected component of the zero morphism

π n𝒞(S i,f):π n𝒞(S i,X)π n𝒞(S i,Y). \pi_n\mathcal{C}(S_i,f) \;\colon\; \pi_n \mathcal{C}(S_i,X) \longrightarrow \pi_n\mathcal{C}(S_i,Y) \,.

Now in this pointed situation we may use that

π n𝒞(,) π 0𝒞(,Ω n()) π 0𝒞(Σ n(),) Ho(𝒞)(Σ n(),) \begin{aligned} \pi_n \mathcal{C}(-,-) & \simeq \pi_0 \mathcal{C}(-,\Omega^n(-)) \\ & \simeq \pi_0\mathcal{C}(\Sigma^n(-),-) \\ & \simeq Ho(\mathcal{C})(\Sigma^n(-),-) \end{aligned}

to find that ff is an equivalence in 𝒞\mathcal{C} precisely if the induced morphisms

Ho(𝒞)(Σ nS i,f):Ho(𝒞)(Σ nS i,X)Ho(𝒞)(Σ nS i,Y) Ho(\mathcal{C})(\Sigma^n S_i, f) \;\colon\; Ho(\mathcal{C})(\Sigma^n S_i,X) \longrightarrow Ho(\mathcal{C})(\Sigma^n S_i,Y)

are isomorphisms for all iIi \in I and nn \in \mathbb{N}.

Finally by the assumption that each suspension Σ nS i\Sigma^n S_i of a generator is itself among the set of generators, the claim follows.


(Brown representability)

Let 𝒞\mathcal{C} be a model category compactly generated by cogroup objects closed under forming suspensions, according to def. 18. Then a functor

F:Ho(𝒞) opSet F \;\colon\; Ho(\mathcal{C})^{op} \longrightarrow Set

(from the opposite of the homotopy category of 𝒞\mathcal{C} to Set) is representable precisely if it is a Brown functor, def. 17.

(Lurie 10, theorem


Due to the version of the Whitehead theorem of prop. 13 we are essentially reduced to showing that Brown functors FF are representable on the S iS_i. To that end consider the following lemma. (In the following we notationally identify, via the Yoneda lemma, objects of 𝒞\mathcal{C}, hence of Ho(𝒞)Ho(\mathcal{C}), with the functors they represent.)

Lemma (\star): Given X𝒞X \in \mathcal{C} and ηF(X)\eta \in F(X), hence η:XF\eta \colon X \to F, then there exists a morphism f:XXf \colon X \to X' and an extension η:XF\eta' \colon X' \to F of η\eta which induces for each S iS_i a bijection η():PSh(Ho(𝒞))(S i,X)Ho(𝒞)(S i,F)F(S i)\eta'\circ (-) \colon PSh(Ho(\mathcal{C}))(S_i,X') \stackrel{\simeq}{\longrightarrow} Ho(\mathcal{C})(S_i,F) \simeq F(S_i).

To see this, first notice that we may directly find an extension η 0\eta_0 along a map XX oX\to X_o such as to make a surjection: simply take X 0X_0 to be the coproduct of all possible elements in the codomain and take

η 0:X(iI,γ:S iFS i)F \eta_0 \;\colon\; X \sqcup \left( \underset{{i \in I,} \atop {\gamma \colon S_i \stackrel{}{\to} F}}{\coprod} S_i \right) \longrightarrow F

to be the canonical map. (Using that FF, by assumption, turns coproducts into products, we may indeed treat the coproduct in 𝒞\mathcal{C} on the left as the coproduct of the corresponding functors.)

To turn the surjection thus constructed into a bijection, we now successively form quotients of X 0X_0. To that end proceed by induction and suppose that η n:X nF\eta_n \colon X_n \to F has been constructed. Then for iIi \in I let

K iker(Ho(𝒞)(S i,X n)η n()F(S i)) K_i \coloneqq ker \left( Ho(\mathcal{C})(S_i, X_n) \stackrel{\eta_n \circ (-)}{\longrightarrow} F(S_i) \right)

be the kernel of η n\eta_n evaluated on S iS_i. These K iK_i are the pieces that need to go away in order to make a bijection. Hence define X n+1X_{n+1} to be their joint homotopy cofiber

X n+1coker((iI,γK iS i)(γ) iIγK iX n). X_{n+1} \coloneqq coker\left( \left( \underset{{i \in I,} \atop {\gamma \in K_i}}{\sqcup} S_i \right) \overset{(\gamma)_{{i \in I} \atop {\gamma\in K_i}}}{\longrightarrow} X_n \right) \,.

Then by the assumption that FF takes this homotopy cokernel to a weak fiber (as in remark 4), there exists an extension η n+1\eta_{n+1} of η n\eta_n along X nX n+1X_n \to X_{n+1}:

Then by the assumption that FF takes this homotopy cokernel to a weak fiber (as in remark 4), there exists an extension η n+1\eta_{n+1} of η n\eta_n along X nX n+1X_n \to X_{n+1}:

(iIγK iS i) (γ) iIγK i X n η n F (po h) η n+1 * X n+1 F(X n+1) * η n+1 epi * η n ker((γ *) iIγK i) * η n (pb) F(X n) (γ *) iIγK i iIγK iF(S i). \array{ \left( \underset{{i \in I}\atop {\gamma \in K_i}}{\sqcup} S_i \right) &\overset{(\gamma)_{{i \in I}\atop \gamma \in K_i}}{\longrightarrow}& X_n &\overset{\eta_n}{\longrightarrow}& F \\ \downarrow &(po^{h})& \downarrow & \nearrow_{\mathrlap{\exists \eta_{n+1}}} \\ \ast &\longrightarrow& X_{n+1} } \;\;\;\;\;\;\;\;\;\;\; \Leftrightarrow \;\;\;\;\;\;\;\;\;\;\; \array{ && F(X_{n+1}) &\longrightarrow& \ast \\ &{}^{\mathllap{\exists \eta_{n+1}}}\nearrow& \downarrow^{\mathrlap{epi}} && \downarrow \\ \ast &\overset{\eta_n}{\longrightarrow}& ker\left((\gamma^\ast\right)_{{i \in I} \atop {\gamma \in K_i}}) &\longrightarrow& \ast \\ &{}_{\mathllap{\eta_n}}\searrow& \downarrow &(pb)& \downarrow \\ && F(X_n) &\underset{(\gamma^\ast)_{{i \in I} \atop {\gamma \in K_i}} }{\longrightarrow}& \underset{{i \in I}\atop {\gamma\in K_i}}{\prod}F(S_i) } \,.

It is now clear that we want to take

Xlim nX n X' \coloneqq \underset{\rightarrow}{\lim}_n X_n

and extend all the η n\eta_n to that colimit. Since we have no condition for evaluating FF on colimits other than pushouts, observe that this sequential colimit is equivalent to the following pushout:

nX n nX 2n nX 2n+1 X, \array{ \underset{n}{\sqcup} X_n &\longrightarrow& \underset{n}{\sqcup} X_{2n} \\ \downarrow && \downarrow \\ \underset{n}{\sqcup} X_{2n+1} &\longrightarrow& X' } \,,

where the components of the top and left map alternate between the identity on X nX_n and the above successor maps X nX n+1X_n \to X_{n+1}. Now the excision property of FF applies to this pushout, and we conclude the desired extension η:XF\eta' \colon X' \to F:

nX n nX 2n+1 X nX 2n (η 2n+1) n η (η 2n) n F F(X) η epi *(η n) n lim nF(X n) nF(X 2n+1) n(X 2n) nF(X n), \array{ && \underset{n}{\sqcup} X_n \\ & \swarrow && \searrow \\ \underset{n}{\sqcup} X_{2n+1} &\longrightarrow& X' &\longleftarrow& \underset{n}{\sqcup} X_{2n} \\ & {}_{\mathllap{(\eta_{2n+1})_{n}}}\searrow& \downarrow^{\mathrlap{\exists \eta}} & \swarrow_{\mathrlap{(\eta_{2n})_n}} \\ && F } \;\;\;\;\;\;\;\;\; \Leftrightarrow \;\;\;\;\;\;\;\;\; \array{ && F(X') \\ &{}^{\mathllap{\exists \eta}}\nearrow& \downarrow^{\mathrlap{epi}} \\ &\ast \overset{(\eta_n)_n}{\longrightarrow}& \underset{\longleftarrow}{\lim}_n F(X_n) \\ & \swarrow && \searrow \\ \underset{n}{\prod}F(X_{2n+1}) && && \underset{n}{\prod}(X_{2n}) \\ & \searrow && \swarrow \\ && \underset{n}{\prod}F(X_n) } \,,

It remains to confirm that this indeed gives the desired bijection. Surjectivity is clear. For injectivity use that all the S iS_i are, by assumption, compact, hence they may be taken inside the sequential colimit:

X n(γ) γ^ S i γ X=lim nX n. \array{ && X_{n(\gamma)} \\ &{}^{\mathllap{ \exists \hat \gamma}}\nearrow& \downarrow \\ S_i &\overset{\gamma}{\longrightarrow}& X' = \underset{\longrightarrow}{\lim}_n X_n } \,.

With this, injectivity follows because by construction we quotiented out the kernel at each stage. Because suppose that γ\gamma is taken to zero in F(S i)F(S_i), then by the definition of X n+1X_{n+1} above there is a factorization of γ\gamma through the point:

0: S i γ^ X n(γ) η n F * X n(γ)+1 X \array{ 0 \colon & S_i &\overset{\hat \gamma}{\longrightarrow}& X_{n(\gamma)} &\overset{\eta_n}{\longrightarrow}& F \\ & \downarrow && \downarrow & \\ & \ast &\longrightarrow& X_{n(\gamma)+1} \\ & && \downarrow \\ & && X' }

This concludes the proof of Lemma (\star).

Now apply the construction given by this lemma to the case X 00X_0 \coloneqq 0 and the unique η 0:0!F\eta_0 \colon 0 \stackrel{\exists !}{\to} F. Lemma ()(\star) then produces an object XX' which represents FF on all the S iS_i, and we want to show that this XX' actually represents FF generally, hence that for every Y𝒞Y \in \mathcal{C} the function

θη():Ho(𝒞)(Y,X)F(Y) \theta \coloneqq \eta'\circ (-) \;\colon\; Ho(\mathcal{C})(Y,X') \stackrel{}{\longrightarrow} F(Y)

is a bijection.

First, to see that θ\theta is surjective, we need to find a preimage of any ρ:YF\rho \colon Y \to F. Applying Lemma ()(\star) to (η,ρ):XYF(\eta',\rho)\colon X'\sqcup Y \longrightarrow F we get an extension κ\kappa of this through some XYZX' \sqcup Y \longrightarrow Z and the morphism on the right of the following commuting diagram:

Ho(𝒞)(,X) Ho(𝒞)(,Z) η() κ() F(). \array{ Ho(\mathcal{C})(-,X') && \longrightarrow && Ho(\mathcal{C})(-, Z) \\ & {}_{\mathllap{\eta'\circ(-)}}\searrow && \swarrow_{\mathrlap{\kappa \circ (-)}} \\ && F(-) } \,.

Moreover, Lemma ()(\star) gives that evaluated on all S iS_i, the two diagonal morphisms here become isomorphisms. But then prop. 13 implies that XZX' \longrightarrow Z is in fact an equivalence. Hence the component map YZZY \to Z \simeq Z is a lift of κ\kappa through θ\theta.

Second, to see that θ\theta is injective, suppose f,g:YXf,g \colon Y \to X' have the same image under θ\theta. Then consider their homotopy pushout

YY (f,g) X Y Z \array{ Y \sqcup Y &\stackrel{(f,g)}{\longrightarrow}& X' \\ \downarrow && \downarrow \\ Y &\longrightarrow& Z }

along the codiagonal of YY. Using that FF sends this to a weak pullback by assumption, we obtain an extension η¯\bar \eta of η\eta' along XZX' \to Z. Applying Lemma ()(\star) to this gives a further extension η¯:ZZ\bar \eta' \colon Z' \to Z which now makes the following diagram

Ho(𝒞)(,X) Ho(𝒞)(,Z) η() η¯() F() \array{ Ho(\mathcal{C})(-,X') && \longrightarrow && Ho(\mathcal{C})(-, Z) \\ & {}_{\mathllap{\eta'\circ(-)}}\searrow && \swarrow_{\mathrlap{\bar \eta' \circ (-)}} \\ && F(-) }

such that the diagonal maps become isomorphisms when evaluated on the S iS_i. As before, it follows via prop. 13 that the morphism h:XZh \colon X' \longrightarrow Z' is an equivalence.

Since by this construction hfh\circ f and hgh\circ g are homotopic

YY (f,g) X h Y Z Z \array{ Y \sqcup Y &\stackrel{(f,g)}{\longrightarrow}& X' \\ \downarrow && \downarrow & \searrow^{\mathrlap{\stackrel{h}{\simeq}}} \\ Y &\longrightarrow& Z &\longrightarrow& Z' }

it follows with hh being an equivalence that already ff and gg were homotopic, hence that they represented the same element.


Given a reduced additive cohomology functor H :Ho(𝒞) opAb H^\bullet \colon Ho(\mathcal{C})^{op}\to Ab^{\mathbb{Z}}, def. 4, its underlying Set-valued functors H n:Ho(𝒞) opAbSetH^n \colon Ho(\mathcal{C})^{op}\to Ab\to Set are Brown functors, def. 17.


The first condition on a Brown functor holds by definition of H H^\bullet. For the second condition, given a homotopy pushout square

X 1 f 1 Y 1 X 2 f 2 Y 2 \array{ X_1 &\stackrel{f_1}{\longrightarrow}& Y_1 \\ \downarrow^{} && \downarrow \\ X_2 &\stackrel{f_2}{\longrightarrow}& Y_2 }

in 𝒞\mathcal{C}, consider the induced morphism of the long exact sequences given by prop. 2

H (coker(f 2)) H (Y 2) f 2 * H (X 2) H +1(Σcoker(f 2)) H (coker(f 1)) H (Y 1) f 1 * H (X 1) H +1(Σcoker(f 1)) \array{ H^\bullet(coker(f_2)) &\longrightarrow& H^\bullet(Y_2) &\stackrel{f^\ast_2}{\longrightarrow}& H^\bullet(X_2) &\stackrel{}{\longrightarrow}& H^{\bullet+1}(\Sigma coker(f_2)) \\ {}^{\mathllap{\simeq}}\downarrow && \downarrow && \downarrow && \downarrow^{\mathrlap{\simeq}} \\ H^\bullet(coker(f_1)) &\longrightarrow& H^\bullet(Y_1) &\stackrel{f^\ast_1}{\longrightarrow}& H^\bullet(X_1) &\stackrel{}{\longrightarrow}& H^{\bullet+1}(\Sigma coker(f_1)) }

Here the outer vertical morphisms are isomorphisms, as shown, due to the pasting law (see also at fiberwise recognition of stable homotopy pushouts). This means that the four lemma applies to this diagram. Inspection shows that this implies the claim.


Let 𝒞\mathcal{C} be a model category which satisfies the conditions of theorem 3, and let (H ,δ)(H^\bullet, \delta) be a reduced additive generalized cohomology functor on 𝒞\mathcal{C}, def. 4. Then there exists a spectrum object EStab(𝒞)E \in Stab(\mathcal{C}) such that

  1. HH\bullet is degreewise represented by EE:

    H Ho(𝒞)(,E ), H^\bullet \simeq Ho(\mathcal{C})(-,E_\bullet) \,,
  2. the suspension isomorphism δ\delta is given by the structure morphisms σ˜ n:E nΩE n+1\tilde \sigma_n \colon E_n \to \Omega E_{n+1} of the spectrum, in that

    δ:H n()Ho(𝒞)(,E n)Ho(𝒞)(,σ˜ n)Ho(𝒞)(,ΩE n+1)Ho(𝒞)(Σ(),E n+1)H n+1(Σ()). \delta \colon H^n(-) \simeq Ho(\mathcal{C})(-,E_n) \stackrel{Ho(\mathcal{C})(-,\tilde\sigma_n) }{\longrightarrow} Ho(\mathcal{C})(-,\Omega E_{n+1}) \simeq Ho(\mathcal{C})(\Sigma (-), E_{n+1}) \simeq H^{n+1}(\Sigma(-)) \,.

Via prop. 14, theorem 3 gives the first clause. With this, the second clause follows by the Yoneda lemma.

Milnor exact sequence

Idea. One tool for computing generalized cohomology groups via “inverse limits” are Milnor exact sequences. For instance the generalized cohomology of the classifying space BU(1)B U(1) plays a key role in the complex oriented cohomology-theory discussed below, and via the equivalence BU(1)P B U(1) \simeq \mathbb{C}P^\infty to the homotopy type of the infinite complex projective space (def. 43), which is the direct limit of finite dimensional projective spaces P n\mathbb{C}P^n, this is an inverse limit of the generalized cohomology groups of the P n\mathbb{C}P^ns. But what really matters here is the derived functor of the limit-operation – the homotopy limit – and the Milnor exact sequence expresses how the naive limits receive corrections from higher “lim^1-terms”. In practice one mostly proceeds by verifying conditions under which these corrections happen to disappear, these are the Mittag-Leffler conditions.

We need this for instance for the computation of Conner-Floyd Chern classes below.

Literature. (Switzer 75, section 7 from def. 7.57 on, Kochman 96, section 4.2, Goerss-Jardine 99, section VI.2, )

Lim 1Lim^1

Given a tower A A_\bullet of abelian groups

A 3f 2A 2f 1A 1f 0A 0 \cdots \to A_3 \stackrel{f_2}{\to} A_2 \stackrel{f_1}{\to} A_1 \stackrel{f_0}{\to} A_0


:nA nnA n \partial \;\colon\; \underset{n}{\prod} A_n \longrightarrow \underset{n}{\prod} A_n

for the homomorphism given by

:(a n) n(a nf n(a n+1)) n. \partial \;\colon\; (a_n)_{n \in \mathbb{N}} \mapsto (a_n - f_n(a_{n+1}))_{n \in \mathbb{N}}.

The limit of a sequence as in def. 19 – hence the group lim nA n\underset{\longleftarrow}{\lim}_n A_n universally equipped with morphisms lim nA np nA n\underset{\longleftarrow}{\lim}_n A_n \overset{p_n}{\to} A_n such that all

lim nA n p n+1 p n A n+1 f n A n \array{ && \underset{\longleftarrow}{\lim}_n A_n \\ & {}^{\mathllap{p_{n+1}}}\swarrow && \searrow^{\mathrlap{p_n}} \\ A_{n+1} && \overset{f_n}{\longrightarrow} && A_n }

commute – is equivalently the kernel of the morphism \partial in def. 19.


Given a tower A A_\bullet of abelian groups

A 3f 2A 2f 1A 1f 0A 0 \cdots \to A_3 \stackrel{f_2}{\to} A_2 \stackrel{f_1}{\to} A_1 \stackrel{f_0}{\to} A_0

then lim 1A \underset{\longleftarrow}{\lim}^1 A_\bullet is the cokernel of the map \partial in def. 19, hence the group that makes a long exact sequence of the form

0lim nA nnA nnA nlim n 1A n0, 0 \to \underset{\longleftarrow}{\lim}_n A_n \longrightarrow \underset{n}{\prod} A_n \stackrel{\partial}{\longrightarrow} \underset{n}{\prod} A_n \longrightarrow \underset{\longleftarrow}{\lim}^1_n A_n \to 0 \,,

The functor lim 1:Ab (,)Ab\underset{\longleftarrow}{\lim}^1 \colon Ab^{(\mathbb{N}, \geq)} \longrightarrow Ab (def. 20) satisfies

  1. for every short exact sequence 0A B C 0Ab (,)0 \to A_\bullet \to B_\bullet \to C_\bullet \to 0 \;\;\; \in Ab^{(\mathbb{N}, \geq)} then the induced sequence

    0lim nA nlim nB nlim nC nlim n 1A nlim n 1B nlim n 1C n0 0 \to \underset{\longleftarrow}{\lim}_n A_n \to \underset{\longleftarrow}{\lim}_n B_n \to \underset{\longleftarrow}{\lim}_n C_n \to \underset{\longleftarrow}{\lim}_n^1 A_n \to \underset{\longleftarrow}{\lim}_n^1 B_n \to \underset{\longleftarrow}{\lim}_n^1 C_n \to 0

    is a long exact sequence of abelian groups;

  2. if A A_\bullet is a tower such that all maps are surjections, then lim n 1A n0\underset{\longleftarrow}{\lim}^1_n A_n \simeq 0.

(e.g. Switzer 75, prop. 7.63, Goerss-Jardine 96, section VI. lemma 2.11)


For the first property: Given A A_\bullet a tower of abelian groups, write

L (A )[0nA ndeg0nA ndeg10] L^\bullet(A_\bullet) \coloneqq \left[ 0 \to \underset{deg \, 0}{\underbrace{\underset{n}{\prod} A_n}} \overset{\partial}{\longrightarrow} \underset{deg\, 1}{\underbrace{\underset{n}{\prod} A_n}} \to 0 \right]

for the homomorphism from def. 19 regarded as the single non-trivial differential in a cochain complex of abelian groups. Then by remark 5 and def. 20 we have H 0(L(A ))limA H^0(L(A_\bullet)) \simeq \underset{\longleftarrow}{\lim} A_\bullet and H 1(L(A ))lim 1A H^1(L(A_\bullet)) \simeq \underset{\longleftarrow}{\lim}^1 A_\bullet.

With this, then for a short exact sequence of towers 0A B C 00 \to A_\bullet \to B_\bullet \to C_\bullet \to 0 the long exact sequence in question is the long exact sequence in homology of the corresponding short exact sequence of complexes

0L (A )L (B )L (C )0. 0 \to L^\bullet(A_\bullet) \longrightarrow L^\bullet(B_\bullet) \longrightarrow L^\bullet(C_\bullet) \to 0 \,.

For the second statement: If all the f kf_k are surjective, then inspection shows that the homomorphism \partial in def. 19 is surjective. Hence its cokernel vanishes.


The category Ab (,)Ab^{(\mathbb{N}, \geq)} of towers of abelian groups has enough injectives.


The functor () n:Ab (,)Ab(-)_n \colon Ab^{(\mathbb{N}, \geq)} \to Ab that picks the nn-th component of the tower has a right adjoint r nr_n, which sends an abelian group AA to the tower

r n[idAidA=(r n) n+1idA=(r n) nid0=(r n) n1000]. r_n \coloneqq \left[ \cdots \overset{id}{\to} A \overset{id}{\to} \underset{= (r_n)_{n+1}}{\underbrace{A}} \overset{id}{\to} \underset{= (r_n)_n}{\underbrace{A}} \overset{id}{\to} \underset{= (r_n)_{n-1}}{\underbrace{0}} \to 0 \to \cdots \to 0 \to 0 \right] \,.

Since () n(-)_n itself is evidently an exact functor, its right adjoint preserves injective objects (prop.).

So with A Ab (,)A_\bullet \in Ab^{(\mathbb{N}, \geq)}, let A nA˜ nA_n \hookrightarrow \tilde A_n be an injective resolution of the abelian group A nA_n, for each nn \in \mathbb{N}. Then

A (η n) nnr nA nnr nA˜ n A_\bullet \overset{(\eta_n)_{n \in \mathbb{N}}}{\longrightarrow} \underset{n \in \mathbb{R}}{\prod} r_n A_n \hookrightarrow \underset{n \in \mathbb{N}}{\prod} r_n \tilde A_n

is an injective resolution for A A_\bullet.


The functor lim 1:Ab (,)Ab\underset{\longleftarrow}{\lim}^1 \colon Ab^{(\mathbb{N}, \geq)} \longrightarrow Ab (def. 20) is the first right derived functor of the limit functor lim:Ab (,)Ab\underset{\longleftarrow}{\lim} \colon Ab^{(\mathbb{N},\geq)} \longrightarrow Ab.


By lemma 2 there are enough injectives in Ab (,)Ab^{(\mathbb{N}, \geq)}. So for A Ab (,)A_\bullet \in Ab^{(\mathbb{N}, \geq)} the given tower of abelian groups, let

0A j 0J 0j 1J 1j 2J 2 0 \to A_\bullet \overset{j^0}{\longrightarrow} J^0_\bullet \overset{j^1}{\longrightarrow} J^1_\bullet \overset{j^2}{\longrightarrow} J^2_\bullet \overset{}{\longrightarrow} \cdots

be an injective resolution. We need to show that

lim 1A ker(lim(j 2))/im(lim(j 1)). \underset{\longleftarrow}{\lim}^1 A_\bullet \simeq ker(\underset{\longleftarrow}{\lim}(j^2))/im(\underset{\longleftarrow}{\lim}(j^1)) \,.

Since limits preserve kernels, this is equivalently

lim 1A (lim(ker(j 2) ))/im(lim(j 1)) \underset{\longleftarrow}{\lim}^1 A_\bullet \simeq (\underset{\longleftarrow}{\lim}(ker(j^2)_\bullet))/im(\underset{\longleftarrow}{\lim}(j^1))

Now observe that each injective J qJ^q_\bullet is a tower of epimorphism. This follows by the defining right lifting property applied against the monomorphisms of towers of the following form

0 0 0 id id id id id incl id id id 0 0 id id id id \array{ \cdots &\to & 0 &\to& 0 &\longrightarrow& 0 &\longrightarrow& \mathbb{Z} &\overset{id}{\longrightarrow}& \cdots &\overset{id}{\longrightarrow}& \mathbb{Z} &\overset{id}{\longrightarrow}& \mathbb{Z} \\ \cdots && \downarrow^{\mathrlap{id}} && \downarrow^{\mathrlap{id}} && \downarrow^{\mathrlap{incl}} && \downarrow^{\mathrlap{id}} && && \downarrow^{\mathrlap{id}} && \downarrow^{\mathrlap{id}} \\ \cdots &\to& 0 &\to& 0 &\to & \mathbb{Z} &\underset{id}{\longrightarrow}& \mathbb{Z} &\underset{id}{\longrightarrow}& \cdots &\underset{id}{\longrightarrow}& \mathbb{Z} &\underset{id}{\longrightarrow}& \mathbb{Z} }

Therefore by the second item of prop. 15 the long exact sequence from the first item of prop. 15 applied to the short exact sequence

0A j 0J 0j 1ker(j 2) 0 0 \to A_\bullet \overset{j^0}{\longrightarrow} J^0_\bullet \overset{j^1}{\longrightarrow} ker(j^2)_\bullet \to 0


0limA limj 0limJ 0limj 1lim(ker(j 2) )lim 1A 0. 0 \to \underset{\longleftarrow}{\lim} A_\bullet \overset{\underset{\longleftarrow}{\lim} j^0}{\longrightarrow} \underset{\longleftarrow}{\lim} J^0_\bullet \overset{\underset{\longleftarrow}{\lim}j^1}{\longrightarrow} \underset{\longleftarrow}{\lim}(ker(j^2)_\bullet) \longrightarrow \underset{\longleftarrow}{\lim}^1 A_\bullet \longrightarrow 0 \,.

Exactness of this sequence gives the desired identification lim 1A (lim(ker(j 2) ))/im(lim(j 1)). \underset{\longleftarrow}{\lim}^1 A_\bullet \simeq (\underset{\longleftarrow}{\lim}(ker(j^2)_\bullet))/im(\underset{\longleftarrow}{\lim}(j^1)) \,.


The functor lim 1:Ab (,)Ab\underset{\longleftarrow}{\lim}^1 \colon Ab^{(\mathbb{N}, \geq)} \longrightarrow Ab (def. 20) is in fact the unique functor, up to natural isomorphism, satisfying the conditions in prop. 17.


The proof of prop. 16 only used the conditions from prop. 15, hence any functor satisfying these conditions is the first right derived functor of lim\underset{\longleftarrow}{\lim}, up to natural isomorphism.

The following is a kind of double dual version of the lim 1\lim^1 construction which is sometimes useful:


Given a cotower

A =(A 0f 0A 1f 1A 2) A_\bullet = (A_0 \overset{f_0}{\to} A _1 \overset{f_1}{\to} A_2 \to \cdots)

of abelian groups, then for every abelian group BAbB \in Ab there is a short exact sequence of the form

0lim n 1Hom(A n,B)Ext 1(lim nA n,B)lim nExt 1(A n,B)0, 0 \to \underset{\longleftarrow}{\lim}^1_n Hom(A_n, B) \longrightarrow Ext^1( \underset{\longrightarrow}{\lim}_n A_n, B ) \longrightarrow \underset{\longleftarrow}{\lim}_n Ext^1( A_n, B) \to 0 \,,

where Hom(,)Hom(-,-) denotes the hom-group, Ext 1(,)Ext^1(-,-) denotes the first Ext-group (and so Hom(,)=Ext 0(,)Hom(-,-) = Ext^0(-,-)).


Consider the homomorphism

˜:nA nnA n \tilde \partial \;\colon\; \underset{n}{\oplus} A_n \longrightarrow \underset{n}{\oplus} A_n

which sends a nA na_n \in A_n to a nf n(a n)a_n - f_n(a_n). Its cokernel is the colimit over the cotower, but its kernel is trivial (in contrast to the otherwise formally dual situation in remark 5). Hence (as opposed to the long exact sequence in def. 20) there is a short exact sequence of the form

0nA n˜nA nlim nA n0. 0 \to \underset{n}{\oplus} A_n \overset{\tilde \partial}{\longrightarrow} \underset{n}{\oplus} A_n \overset{}{\longrightarrow} \underset{\longrightarrow}{lim}_n A_n \to 0 \,.

Every short exact sequence gives rise to a long exact sequence of derived functors (prop.) which in the present case starts out as

0Hom(lim nA n,B)nHom(A n,B)nHom(A n,B)Ext 1(lim nA n,B)nExt 1(A n,B)nExt 1(A n,B) 0 \to Hom(\underset{\longrightarrow}{\lim}_n A_n,B) \longrightarrow \underset{n}{\prod} Hom( A_n, B ) \overset{\partial}{\longrightarrow} \underset{n}{\prod} Hom( A_n, B ) \longrightarrow Ext^1(\underset{\longrightarrow}{\lim}_n A_n,B) \longrightarrow \underset{n}{\prod} Ext^1( A_n, B ) \overset{\partial}{\longrightarrow} \underset{n}{\prod} Ext^1( A_n, B ) \longrightarrow \cdots

where we used that direct sum is the coproduct in abelian groups, so that homs out of it yield a product, and where the morphism \partial is the one from def. 19 corresponding to the tower

Hom(A ,B)=(Hom(A 2,B)Hom(A 1,B)Hom(A 0,B)). Hom(A_\bullet,B) = ( \cdots \to Hom(A_2,B) \to Hom(A_1,B) \to Hom(A_0,B) ) \,.

Hence truncating this long sequence by forming kernel and cokernel of \partial, respectively, it becomes the short exact sequence in question.

Mittag-Leffler condition

A tower A A_\bullet of abelian groups

A 3A 2A 1A 0 \cdots \to A_3 \to A_2 \to A_1 \to A_0

is said to satify the Mittag-Leffler condition if for all kk there exists iki \geq k such that for all jikj \geq i \geq k the image of the homomorphism A iA kA_i \to A_k equals that of A jA kA_j \to A_k

im(A iA k)im(A jA k). im(A_i \to A_k) \simeq im(A_j \to A_k) \,.

(e.g. Switzer 75, def. 7.74)


The Mittag-Leffler condition, def. 21, is satisfied in particular when all morphisms A i+1A iA_{i+1}\to A_i are epimorphisms (hence surjections of the underlying sets).


If a tower A A_\bullet satisfies the Mittag-Leffler condition, def. 21, then its lim 1\underset{\leftarrow}{\lim}^1 vanishes:

lim 1A =0. \underset{\longleftarrow}{\lim}^1 A_\bullet = 0 \,.

e.g. (Switzer 75, theorem 7.75, Kochmann 96, prop. 4.2.3, Weibel 94, prop. 3.5.7)

Proof idea

One needs to show that with the Mittag-Leffler condition, then the cokernel of \partial in def. 19 vanishes, hence that \partial is an epimorphism in this case, hence that every (a n) nnA n(a_n)_{n \in \mathbb{N}} \in \underset{n}{\prod} A_n has a preimage under \partial. So use the Mittag-Leffler condition to find pre-images of a na_n by induction over nn.

Mapping telescopes

Given a sequence

X =(X 0f 0X 1f 1X 2f 2) X_\bullet = \left( X_0 \overset{f_0}{\longrightarrow} X_1 \overset{f_1}{\longrightarrow} X_2 \overset{f_2}{\longrightarrow} \cdots \right)

of (pointed) topological spaces, then its mapping telescope is the result of forming the (reduced) mapping cylinder Cyl(f n)Cyl(f_n) for each nn and then attaching all these cylinders to each other in the canonical way



X =(X 0f 0X 1f 1X 2f 2) X_\bullet = \left( X_0 \overset{f_0}{\longrightarrow} X_1 \overset{f_1}{\longrightarrow} X_2 \overset{f_2}{\longrightarrow} \cdots \right)

a sequence in Top, its mapping telescope is the quotient topological space of the disjoint union of product topological spaces

Tel(X )(n(X n×[n,n+1]))/ Tel(X_\bullet) \coloneqq \left( \underset{n \in \mathbb{N}}{\sqcup} \left( X_n \times [n,n+1] \right) \right)/_\sim

where the equivalence relation quotiented out is

(x n,n)(f(x n),n+1) (x_n, n) \sim (f(x_n), n+1)

for all nn\in \mathbb{N} and x nX nx_n \in X_n.

Analogously for X X_\bullet a sequence of pointed topological spaces then use reduced cylinders (exmpl.) to set

Tel(X )(n(X n[n,n+1] +))/ . Tel(X_\bullet) \coloneqq \left( \underset{n \in \mathbb{N}}{\sqcup} \left( X_n \wedge [n,n+1]_+ \right) \right)/_\sim \,.

For X X_\bullet the sequence of stages of a (pointed) CW-complex X=lim nX nX = \underset{\longleftarrow}{\lim}_n X_n, then the canonical map

Tel(X )X Tel(X_\bullet) \longrightarrow X

from the mapping telescope, def. 22, is a weak homotopy equivalence.


Write in the following Tel(X)Tel(X) for Tel(X )Tel(X_\bullet) and write Tel(X n)Tel(X_n) for the mapping telescop of the substages of the finite stage X nX_n of XX. It is intuitively clear that each of the projections at finite stage

Tel(X n)X n Tel(X_n) \longrightarrow X_n

is a homotopy equivalence, hence (prop.) a weak homotopy equivalence. A concrete construction of a homotopy inverse is given for instance in (Switzer 75, proof of prop. 7.53).

Moreover, since spheres are compact, so that elements of homotopy groups π q(Tel(X))\pi_q(Tel(X)) are represented at some finite stage π q(Tel(X n))\pi_q(Tel(X_n)) it follows that

lim nπ q(Tel(X n))π q(Tel(X)) \underset{\longrightarrow}{\lim}_n \pi_q(Tel(X_n)) \overset{\simeq}{\longrightarrow} \pi_q(Tel(X))

are isomorphisms for all qq\in \mathbb{N} and all choices of basepoints (not shown).

Together these two facts imply that in the following commuting square, three morphisms are isomorphisms, as shown.

lim nπ q(Tel(X n)) π q(Tel(X)) lim nπ q(X n) π q(X). \array{ \underset{\longleftarrow}{\lim}_n \pi_q(Tel(X_n)) &\overset{\simeq}{\longrightarrow}& \pi_q(Tel(X)) \\ {}^{\mathllap{\simeq}}\downarrow && \downarrow \\ \underset{\longleftarrow}{\lim}_n \pi_q(X_n) &\underset{\simeq}{\longrightarrow}& \pi_q(X) } \,.

Therefore also the remaining morphism is an isomorphism (two-out-of-three). Since this holds for all qq and all basepoints, it is a weak homotopy equivalence.

Milnor exact sequences

(Milnor exact sequence for homotopy groups)


X 3p 2X 2p 1X 1p 0X 0 \cdots \to X_3 \overset{p_2}{\longrightarrow} X_2 \overset{p_1}{\longrightarrow} X_1 \overset{p_0}{\longrightarrow} X_0

be a tower of fibrations (Serre fibrations (def.)). Then for each qq \in \mathbb{N} there is a short exact sequence

0lim i 1π q+1(X i)π q(lim iX i)lim iπ q(X i)0, 0 \to \underset{\longleftarrow}{\lim}^1_i \pi_{q+1}(X_i) \longrightarrow \pi_q(\underset{\longleftarrow}{\lim}_i X_i) \longrightarrow \underset{\longleftarrow}{\lim}_i \pi_q(X_i) \to 0 \,,

for π \pi_\bullet the homotopy group-functor (exact as pointed sets for i=0i = 0, as groups for i1i \geq 1) which says that

  1. the failure of the limit over the homotopy groups of the stages of the tower to equal the homotopy groups of the limit of the tower is at most in the kernel of the canonical comparison map;

  2. that kernel is the lim 1\underset{\longleftarrow}{\lim}^1 (def. 20) of the homotopy groups of the stages.

An elementary but tedious proof is indicated in (Bousfield-Kan 72, chapter IX, theorem 3.1. The following is a neat model category-theoretic proof following (Goerss-Jardine 96, section VI. prop. 2.15), which however requires the concept of homotopy limit over towers.


With respect to the classical model structure on simplicial sets or the classical model structure on topological spaces, a tower of fibrations as stated is a fibrant object in the injective model structure on functors [(,),sSet] inj[(\mathbb{N},\geq), sSet]_{inj} ([(,),Top] inj[(\mathbb{N},\geq), Top]_{inj}) (prop). Hence the plain limit over this diagram represents the homotopy limit. By the discussion there, up to weak equivalence that homotopy limit is also the pullback in

holimX nPath(X n) (pb) nX n (id,p n) n nX n×X n, \array{ holim X_\bullet &\longrightarrow& \underset{n}{\prod} Path(X_n) \\ \downarrow &(pb)& \downarrow \\ \underset{n}{\prod} X_n &\underset{(id,p_n)_n}{\longrightarrow}& \underset{n}{\prod} X_ n \times X_n } \,,

where on the right we have the product over all the canonical fibrations out of the path space objects. Hence also the left vertical morphism is a fibration, and so by taking its fiber over a basepoint, the pasting law gives a homotopy fiber sequence

nΩX nholimX nX n. \underset{n}{\prod} \Omega X_n \longrightarrow holim X_\bullet \longrightarrow \underset{n}{\prod} X_n \,.

The long exact sequence of homotopy groups of this fiber sequence goes

nπ q+1(X n)nπ q+1(X n)π q(limX )nπ q(X n)nπ q(X n). \cdots \to \underset{n}{\prod} \pi_{q+1}(X_n) \longrightarrow \underset{n}{\prod} \pi_{q+1}(X_n) \longrightarrow \pi_q (\underset{\longleftarrow}{\lim} X_\bullet) \longrightarrow \underset{n}{\prod} \pi_q(X_n) \longrightarrow \underset{n}{\prod} \pi_q(X_n) \to \cdots \,.

Chopping that off by forming kernel and cokernel yields the claim for positive qq. For q=0q = 0 it follows by inspection.


(Milnor exact sequence for generalized cohomology)

Let XX be a pointed CW-complex, X=lim nX nX = \underset{\longrightarrow}{\lim}_n X_n and let E˜ \tilde E^\bullet an additive reduced cohomology theory, def. 1.

Then the canonical morphisms make a short exact sequence

0lim n 1E˜ 1(X n)E˜ (X)lim nE˜ (X n)0, 0 \to \underset{\longleftarrow}{\lim}^1_n \tilde E^{\bullet-1}(X_n) \longrightarrow \tilde E^{\bullet}(X) \longrightarrow \underset{\longleftarrow}{\lim}_n \tilde E^{\bullet}(X_n) \to 0 \,,

saying that

  1. the failure of the canonical comparison map E˜ (X)limE˜ (X n)\tilde E^\bullet(X) \to \underset{\longleftarrow}{\lim} \tilde E^\bullet(X_n) to the limit of the cohomology groups on the finite stages to be an isomorphism is at most in a non-vanishing kernel;

  2. this kernel is precisely the lim 1\lim^1 (def. 20) of the cohomology groups at the finite stages in one degree lower.

e.g. (Switzer 75, prop. 7.66, Kochmann 96, prop. 4.2.2)



X =(X 0i 0X 1i 1X 2i 1) X_\bullet = \left( X_0 \overset{i_0}{\hookrightarrow} X_1 \overset{i_1}{\hookrightarrow} X_2 \overset{i_1}{\hookrightarrow} \cdots \right)

the sequence of stages of the (pointed) CW-complex X=lim nX nX = \underset{\longleftarrow}{\lim}_n X_n, write

A X nX 2n×[2n,2n+1]; B X nX (2n+1)×[2n+1,2n+2]. \begin{aligned} A_X &\coloneqq \underset{n \in \mathbb{N}}{\sqcup} X_{2n} \times [2n,{2n}+1]; \\ B_X &\coloneqq \underset{n \in \mathbb{N}}{\sqcup} X_{(2n+1)} \times [2n+1,{2n}+2]. \end{aligned}

for the disjoint unions of the cylinders over all the stages in even and all those in odd degree, respectively.

These come with canonical inclusion maps into the mapping telescope Tel(X )Tel(X_\bullet) (def.), which we denote by

A X B X ι A x ι B x Tel(X ). \array{ A_X && && B_X \\ & {}_{\mathllap{\iota_{A_x}}}\searrow && \swarrow_{\mathrlap{\iota_{B_x}}} \\ && Tel(X_\bullet) } \,.

Observe that

  1. A XB XTel(X )A_X \cup B_X \simeq Tel(X_\bullet);

  2. A XB XnX nA_X \cap B_X \simeq \underset{n \in \mathbb{N}}{\sqcup} X_n;

and that there are homotopy equivalences

  1. A XnX 2n+1A_X \simeq \underset{n \in \mathbb{N}}{\sqcup} X_{2n+1}

  2. B XnX 2nB_X \simeq \underset{n \in \mathbb{N}}{\sqcup} X_{2n}

  3. Tel(X )XTel(X_\bullet) \simeq X.

The first two are obvious, the third is this proposition.

This implies that the Mayer-Vietoris sequence (prop.) for E˜ \tilde E^\bullet on the cover ABXA \sqcup B \to X is isomorphic to the bottom horizontal sequence in the following diagram:

E˜ 1(A X)E˜ 1(B X) E˜ 1(A XB X) E˜ (X) (ι A x) *(ι B x) * E˜ (A X)E˜ (B X) E˜ (A XB X) = (id,id) nE˜ 1(X n) nE˜ 1(X n) E˜ (X) (i n *) n nE˜ (X n) nE˜ (X n), \array{ \tilde E^{\bullet-1}(A_X)\oplus \tilde E^{\bullet-1}(B_X) &\longrightarrow& \tilde E^{\bullet-1}(A_X \cap B_X) &\longrightarrow& \tilde E^\bullet(X) &\overset{(\iota_{A_x})^\ast - (\iota_{B_x})^\ast}{\longrightarrow}& \tilde E^\bullet(A_X)\oplus \tilde E^\bullet(B_X) &\overset{}{\longrightarrow}& \tilde E^\bullet(A_X \cap B_X) \\ \downarrow^{\mathrlap{\simeq}} && \downarrow^{\mathrlap{\simeq}} && \downarrow^{\mathrlap{=}} && {}^{\mathllap{(id, -id)}}\downarrow^{\mathrlap{\simeq}} && \downarrow^{\mathrlap{\simeq}} \\ \underset{n}{\prod}\tilde E^{\bullet-1}(X_n) &\underset{\partial}{\longrightarrow}& \underset{n}{\prod}\tilde E^{\bullet-1}(X_n) &\longrightarrow& \tilde E^\bullet(X) &\overset{(i_n^\ast)_{n}}{\longrightarrow}& \underset{n}{\prod}\tilde E^\bullet(X_n) &\underset{\partial}{\longrightarrow}& \underset{n}{\prod}\tilde E^\bullet(X_n) } \,,

hence that the bottom sequence is also a long exact sequence.

To identify the morphism \partial, notice that it comes from pulling back EE-cohomology classes along the inclusions ABAA \cap B \to A and ABBA\cap B \to B. Comonentwise these are the inclusions of each X nX_n into the left and the right end of its cylinder inside the mapping telescope, respectively. By the construction of the mapping telescope, one of these ends is embedded via i n:X nX n+1i_n \colon X_n \hookrightarrow X_{n+1} into the cylinder over X n+1X_{n+1}. In conclusion, \partial acts by

:(a n) n(a ni n *(a n+1)). \partial \;\colon\; (a_n)_{n \in \mathbb{N}} \mapsto ( a_n - i_n^\ast(a_{n+1}) ) \,.

(The relative sign is the one in (ι A x) *(ι B x) *(\iota_{A_x})^\ast - (\iota_{B_x})^\ast originating in the definition of the Mayer-Vietoris sequence and properly propagated to the bottom sequence while ensuring that E˜ (X) nE˜ (X n)\tilde E^\bullet(X)\to \prod_n \tilde E^\bullet(X_n) is really (i n *) n(i_n^\ast)_n and not (1) n(i n *) n(-1)^n(i_n^\ast)_n, as needed for the statement to be proven.)

This is the morphism from def. 19 for the sequence

E˜ (X n+1)i n *E˜ (X n)i n *E˜ (X n1). \cdots \to \tilde E^\bullet(X_{n+1}) \overset{i_n^\ast}{\longrightarrow} \tilde E^\bullet(X_n) \overset{i_n^\ast}{\longrightarrow} \tilde E^{\bullet}(X_{n-1}) \to \cdots \,.

Hence truncating the above long exact sequence by forming kernel and cokernel of \partial, the result follows via remark 5 and definition 20.

In contrast:


Let XX be a pointed CW-complex, X=lim nX nX = \underset{\longleftarrow}{\lim}_n X_n.

For E˜ \tilde E_\bullet an additive reduced generalized homology theory, then

lim nE˜ (X n)E˜ (X) \underset{\longrightarrow}{\lim}_n \tilde E_\bullet(X_n) \overset{\simeq}{\longrightarrow} \tilde E_\bullet(X)

is an isomorphism.

(Switzer 75, prop. 7.53)

There is also a version for cohomology of spectra:

For X,EHo(Spectra)X, E \in Ho(Spectra) two spectra, then the EE-generalized cohomology of XX is the graded group of homs in the stable homotopy category (def., exmpl.)

E (X) [X,E] [Σ X,E]. \begin{aligned} E^\bullet(X) & \coloneqq [X,E]_{-\bullet} \\ & \coloneqq [\Sigma^\bullet X, E] \end{aligned} \,.

The stable homotopy category is, in particular, the homotopy category of the stable model structure on orthogonal spectra, in that its localization at the stable weak homotopy equivalences is of the form

γ:OrthSpec(Top cg) stableHo(Spectra). \gamma \;\colon\; OrthSpec(Top_{cg})_{stable} \longrightarrow Ho(Spectra) \,.

In the following when considering an orthogonal spectrum XOrthSpec(Top cg)X \in OrthSpec(Top_{cg}), we use, for brevity, the same symbol for its image under γ\gamma.


For X,EOrthSpec(Top cg)X, E \in OrthSpec(Top_{cg}) two orthogonal spectra (or two symmetric spectra such that XX is a semistable symmetric spectrum) then there is a short exact sequence of the form

0lim n 1E +n1(X n)E (X)lim nE +n(X n)0 0 \to \underset{\longleftarrow}{\lim}^1_n E^{\bullet + n -1}(X_{n}) \longrightarrow E^\bullet(X) \longrightarrow \underset{\longleftarrow}{\lim}_n E^{\bullet + n}(X_n) \to 0

where lim 1\underset{\longleftarrow}{\lim}^1 denotes the lim^1, and where this and the limit on the right are taken over the following structure morphisms

E +n+1(X n+1)E +1n+1(Σ n X)E +n+1(X nS 1)E +n(X n). E^{\bullet + n + 1}(X_{n+1}) \overset{E^{\bullet+1n+1}(\Sigma^X_n)}{\longrightarrow} E^{\bullet+n+1}(X_n \wedge S^1) \overset{\simeq}{\longrightarrow} E^{\bullet + n}(X_n) \,.

(Schwede 12, chapter II prop. 6.5 (ii)) (using that symmetric spectra underlying orthogonal spectra are semistable (Schwede 12, p. 40))


For X,EHo(Spectra)X,E \in Ho(Spectra) two spectra such that the tower nE n1(X n)n \mapsto E^{n -1}(X_{n}) satisfies the Mittag-Leffler condition (def. 21), then two morphisms of spectra XEX \longrightarrow E are homotopic already if all their morphisms of component spaces X nE nX_n \to E_n are.


By prop. 18 the assumption implies that the lim 1lim^1-term in prop. 22 vanishes, hence by exactness it follows that in this case there is an isomorphism

[X,E]E 0(X)lim n[X n,E n]. [X,E] \simeq E^0(X) \overset{\simeq}{\longrightarrow} \underset{\longleftarrow}{\lim}_n [X_n, E_n] \,.

Serre-Atiyah-Hirzebruch spectral sequence

Idea. Another important tool for computing generalized cohomology is to reduce it to the computation of ordinary cohomology with coefficients. Given a generalized cohomology theory EE, there is a spectral sequence known as the Atiyah-Hirzebruch spectral sequence (AHSS) which serves to compute EE-cohomology of FF-fiber bundles over a simplicial complex XX in terms of ordinary cohomology with coefficients in the generalized cohomology E (F)E^\bullet(F) of the fiber. For E=E = HA this is known as the Serre spectral sequence.

The Atiyah-Hirzebruch spectral sequence in turn is a consequence of the “Cartan-Eilenberg spectral sequence” which arises from the exact couple of relative cohomology groups of the skeleta of the CW-complex, and whose first page is the relative cohomology groups for codimension-1 skeleta.

We need the AHSS for instance for the computation of Conner-Floyd Chern classes below.

Literature. (Kochman 96, section 2.2 and 4.2)

See also the accompanying lecture notes on spectral sequences.

Converging spectral sequences

A cohomology spectral sequence {E r p,q,d r}\{E_r^{p,q}, d_r\} is

  1. a sequence {E r ,}\{E_r^{\bullet,\bullet}\} (for rr \in \mathbb{N}, r1r \geq 1) of bigraded abelian groups (the “pages”);

  2. a sequence of linear maps (the “differentials”)

    {d r:E r ,E r +r,r+1} \{d_r \;\colon\; E_r^{\bullet,\bullet} \longrightarrow E_r^{\bullet+r, \bullet-r+1}\}

such that

  • H r+1 ,H_{r+1}^{\bullet,\bullet} is the cochain cohomology of d rd_r, i.e. E r+1 ,=H(E r ,,d r)E_{r+1}^{\bullet, \bullet} = H(E_r^{\bullet,\bullet},d_r), for all rr \in \mathbb{N}, r1r \geq 1.

Given a \mathbb{Z}-graded abelian group_ C C^\bullet equipped with a decreasing filtration

C F sC F s+1C 0 C^\bullet \supset \cdots \supset F^s C^\bullet \supset F^{s+1} C^\bullet \supset \cdots \supset 0

such that

C =sF sC and0=sF sC C^\bullet = \underset{s}{\cup} F^s C^\bullet \;\;\;\; and \;\;\;\; 0 = \underset{s}{\cap} F^s C^\bullet

then the spectral sequence is said to converge to C C^\bullet, denoted,

E 2 ,C E_2^{\bullet,\bullet} \Rightarrow C^\bullet


  1. in each bidegree (s,t)(s,t) the sequence {E r s,t} r\{E_r^{s,t}\}_r eventually becomes constant on a group

    E s,tE 1 s,tE_\infty^{s,t} \coloneqq E_{\gg 1}^{s,t};

  2. E ,E_\infty^{\bullet,\bullet} is the associated graded of the filtered C C^\bullet in that

    E s,tF sC s+t/F s+1C s+tE_\infty^{s,t} \simeq F^s C^{s+t} / F^{s+1}C^{s+t}.

The converging spectral sequence is called a multiplicative spectral sequence if

  1. {E 2 ,}\{E_2^{\bullet,\bullet}\} is equipped with the structure of a bigraded algebra;

  2. F C F^\bullet C^\bullet is equipped with the structure of a filtered graded algebra (F pC kF qC lF p+qC k+lF^p C^k \cdot F^q C^l \subset F^{p+q} C^{k+l});

such that

  1. each d rd_{r} is a derivation with respect to the (induced) algebra structure on E r ,{E_r^{\bullet,\bullet}}, graded of degree 1 with respect to total degree;

  2. the multiplication on E ,E_\infty^{\bullet,\bullet} is compatible with that on C C^\bullet.


The point of spectral sequences is that by subdividing the data in any graded abelian group C C^\bullet into filtration stages, with each stage itself subdivided into bidegrees, such that each consecutive stage depends on the previous one in way tightly controled by the bidegrees, then this tends to give much control on the computation of C C^\bullet. For instance it often happens that one may argue that the differentials in some spectral sequence all vanish from some page on (one says that the spectral sequence collapses at that page) by pure degree reasons, without any further computation.


The archetypical example of (co-)homology spectral sequences as in def. 24 are induced from a filtering on a (co-)chain complex, converging to the (co-)chain homology of the chain complex by consecutively computing relative (co-)chain homologies, relative to decreasing (increasing) filtering degrees. For more on such spectral sequences of filtered complexes see at Interlude -- Spectral sequences the section For filtered complexes.

A useful way to generate spectral sequences is via exact couples:


An exact couple is three homomorphisms of abelian groups of the form

D g D f h E \array{ D && \stackrel{g}{\longrightarrow} && D \\ & {}_{\mathllap{f}}\nwarrow && \swarrow_{\mathrlap{h}} \\ && E }

such that the image of one is the kernel of the next.

im(h)=ker(f),im(f)=ker(g),im(g)=ker(f). im(h) = ker(f)\,,\;\;\; im(f) = ker(g)\,, \;\;\; im(g) = ker(f) \,.

Given an exact couple, then its derived exact couple is

im(g) g im(g) f hg 1 H(E,hf), \array{ im(g) && \stackrel{g}{\longrightarrow} && im(g) \\ & {}_{\mathllap{f}}\nwarrow && \swarrow_{\mathrlap{h \circ g^{-1}}} \\ && H(E, h \circ f) } \,,

where g 1g^{-1} denotes the operation of sending one equivalence class to the equivalenc class of any preimage under gg of any of its representatives.


(cohomological spectral sequence of an exact couple)

Given an exact couple, def. 24,

D 1 g 1 D 1 f 1 h 1 E 1 \array{ D_1 && \stackrel{g_1}{\longrightarrow} && D_1 \\ & {}_{\mathllap{f_1}}\nwarrow && \swarrow_{\mathrlap{h_1}} \\ && E_1 }

its derived exact couple

D 2 g 2 D 2 f 2 h 2 E 2 \array{ D_2 && \stackrel{g_2}{\longrightarrow} && D_2 \\ & {}_{\mathllap{f_2}}\nwarrow && \swarrow_{\mathrlap{h_2}} \\ && E_2 }

is itself an exact couple. Accordingly there is induced a sequence of exact couples

D r g r D r f r h r E r. \array{ D_r && \stackrel{g_r}{\longrightarrow} && D_r \\ & {}_{\mathllap{f_r}}\nwarrow && \swarrow_{\mathrlap{h_r}} \\ && E_r } \,.

If the abelian groups DD and EE are equipped with bigrading such that

deg(f)=(0,0),deg(g)=(1,1),deg(h)=(1,0) deg(f) = (0,0)\,,\;\;\;\; deg(g) = (-1,1)\,,\;\;\; deg(h) = (1,0)

then {E r ,,d r}\{E_r^{\bullet,\bullet}, d_r\} with

d r h rf r =hg r+1f \begin{aligned} d_r & \coloneqq h_r \circ f_r \\ & = h \circ g^{-r+1} \circ f \end{aligned}

is a cohomological spectral sequence, def. 24.

(As before in prop. 23, the notation g ng^{-n} with nn \in \mathbb{N} denotes the function given by choosing, on representatives, a preimage under g n=gggntimesg^n = \underset{n\;times}{\underbrace{g \circ \cdots \circ g \circ g}}, with the implicit claim that all possible choices represent the same equivalence class.)

If for every bidegree (s,t)(s,t) there exists R s,t1R_{s,t} \gg 1 such that for all rR s,tr \geq R_{s,t}

  1. g:D s+R,tRD s+R1,tR1g \colon D^{s+R,t-R} \stackrel {\simeq}{\longrightarrow} D^{s+R -1, t-R-1};

  2. g:D sR+1,t+R20D sR,t+R1g\colon D^{s-R+1, t+R-2} \stackrel{0}{\longrightarrow} D^{s-R,t+R-1}

then this spectral sequence converges to the inverse limit group

G lim(gD s,sgD s1,s+1g) G^\bullet \coloneqq \underset{}{\lim} \left( \cdots \stackrel{g}{\to} D^{s,\bullet-s} \stackrel{g}{\longrightarrow} D^{s-1, \bullet - s + 1} \stackrel{g}{\to} \cdots \right)

filtered by

F pG ker(G D p1,p+1). F^p G^\bullet \coloneqq ker(G^\bullet \to D^{p-1, \bullet - p+1}) \,.

(e.g. Kochmann 96, lemma 2.6.2)


We check the claimed form of the E E_\infty-page:

Since ker(h)=im(g)ker(h) = im(g) in the exact couple, the kernel

ker(d r1)ker(hg r+2f) ker(d_{r-1}) \coloneqq ker(h \circ g^{-r+2} \circ f)

consists of those elements xx such that g r+2(f(x))=g(y)g^{-r+2} (f(x)) = g(y), for some yy, hence

ker(d r1) s,tf 1(g r1(D s+r1,tr+1)). ker(d_{r-1})^{s,t} \simeq f^{-1}(g^{r-1}(D^{s+r-1,t-r+1})) \,.

By assumption there is for each (s,t)(s,t) an R s,tR_{s,t} such that for all rR s,tr \geq R_{s,t} then ker(d r1) s,tker(d_{r-1})^{s,t} is independent of rr.

Moreover, im(d r1)im(d_{r-1}) consists of the image under hh of those xD s1,tx \in D^{s-1,t} such that g r2(x)g^{r-2}(x) is in the image of ff, hence (since im(f)=ker(g)im(f) = ker(g) by exactness of the exact couple) such that g r2(x)g^{r-2}(x) is in the kernel of gg, hence such that xx is in the kernel of g r1g^{r-1}. If r>Rr \gt R then by assumption g r1| D s1,t=0g^{r-1}|_{D^{s-1,t}} = 0 and so then im(d r1)=im(h)im(d_{r-1}) = im(h).

(Beware this subtlety: while g R s,t| D s1,tg^{R_{s,t}}|_{D^{s-1,t}} vanishes by the convergence assumption, the expression g R s,t| D s+r1,tr+1g^{R_{s,t}}|_{D^{s+r-1,t-r+1}} need not vanish yet. Only the higher power g R s,t+R s+1,t+2+2| D s+r1,tr+1g^{R_{s,t}+ R_{s+1,t+2}+2}|_{D^{s+r-1,t-r+1}} is again guaranteed to vanish. )

It follows that

E p,np =ker(d R)/im(d R) f 1(im(g R1))/im(h) fim(g R1)im(f) im(g R1)ker(g) \begin{aligned} E_\infty^{p,n-p} & = ker(d_R)/im(d_R) \\ & \simeq f^{-1}(im(g^{R-1}))/im(h) \\ & \underoverset{\simeq}{f}{\longrightarrow} im(g^{R-1}) \cap im(f) \\ & \simeq im(g^{R-1}) \cap ker(g) \end{aligned}

where in last two steps we used once more the exactness of the exact couple.

(Notice that the above equation means in particular that the E E_\infty-page is a sub-group of the image of the E 1E_1-page under ff.)

The last group above is that of elements xG nx \in G^n which map to zero in D p1,np+1D^{p-1,n-p+1} and where two such are identified if they agree in D p,npD^{p,n-p}, hence indeed

E p,npF pG n/F p+1G n. E_\infty^{p,n-p} \simeq F^p G^n / F^{p+1} G^n \,.

Given a spectral sequence (def. 24), then even if it converges strongly, computing its infinity-page still just gives the associated graded of the filtered object that it converges to, not the filtered object itself. The latter is in each filter stage an extension of the previous stage by the corresponding stage of the infinity-page, but there are in general several possible extensions (the trivial extension or some twisted extensions). The problem of determining these extensions and hence the problem of actually determining the filtered object from a spectral sequence converging to it is often referred to as the extension problem.

More in detail, consider, for definiteness, a cohomology spectral sequence converging to some filtered F H F^\bullet H^\bullet

E p,qH . E^{p,q} \;\Rightarrow\; H^\bullet \,.

Then by definition of convergence there are isomorphisms

E p,F pH p+/F p+1H p+. E_\infty^{p,\bullet} \simeq F^p H^{p + \bullet} / F^{p+1} H^{p + \bullet} \,.

Equivalently this means that there are short exact sequences of the form

0F p+1H p+F pH p+E p,0. 0 \to F^{p+1}H^{p +\bullet} \hookrightarrow F^p H^{p +\bullet} \longrightarrow E_\infty^{p,\bullet} \to 0 \,.

for all pp. The extension problem then is to inductively deduce F pH F^p H^\bullet from knowledge of F p+1H F^{p+1}H^\bullet and E p,E_\infty^{p,\bullet}.

In good cases these short exact sequences happen to be split exact sequences, which means that the extension problem is solved by the direct sum

F pH p+F p+1H p+E p,. F^p H^{p+\bullet} \simeq F^{p+1} H^{p+\bullet} \oplus E_\infty^{p,\bullet} \,.

But in general this need not be the case.

One sufficient condition that these exact sequences split is that they consist of homomorphisms of RR-modules, for some ring RR, and that E p,E_\infty^{p,\bullet} are projective modules (for instance free modules) over RR. Because then the Ext-group Ext R 1(E p,,)Ext^1_R(E_\infty^{p,\bullet},-) vanishes, and hence all extensions are trivial, hence split.

So for instance for every spectral sequence in vector spaces the extension problem is trivial (since every vector space is a free module).


The following proposition requires, in general, to evaluate cohomology functors not just on CW-complexes, but on all topological spaces. Hence we invoke prop. 1 to regard a reduced cohomology theory as a contravariant functor on all pointed topological spaces, which sends weak homotopy equivalences to isomorphisms (def. 3).


(Serre-Cartan-Eilenberg-Whitehead-Atiyah-Hirzebruch spectral sequence)

Let A A^\bullet be a an additive unreduced generalized cohomology functor (def.). Let BB be a CW-complex and let XπBX \stackrel{\pi}{\to} B be a Serre fibration (def.), such that all its fibers are weakly contractible or such that BB is simply connected. In either case all fibers are identified with a typical fiber FF up to weak homotopy equivalence by connectedness (this example), and well defined up to unique iso in the homotopy category by simply connectedness:

F X Fib cl B. \array{ F &\longrightarrow& X \\ && \downarrow^{\mathrlap{\in Fib_{cl}}} \\ && B } \,.

If at least one of the following two conditions is met

  • BB is finite-dimensional as a CW-complex;

  • A (F)A^\bullet(F) is bounded below in degree and the sequences A p(X n+1)A p(X n)\cdots \to A^p(X_{n+1}) \to A^p(X_n) \to \cdots satisfy the Mittag-Leffler condition (def. 21) for all pp;

then there is a cohomology spectral sequence, def. 24, whose E 2E_2-page is the ordinary cohomology H (B,A (F))H^\bullet(B,A^\bullet(F)) of BB with coefficients in the AA-cohomology groups A (F)A^\bullet(F) of the fiber, and which converges to the AA-cohomology groups of the total space

E 2 p,q=H p(B,A q(F))A (X) E_2^{p,q} = H^p(B, A^q(F)) \; \Rightarrow \; A^\bullet(X)

with respect to the filtering given by

F pA (X)ker(A (X)A (X p1)), F^p A^\bullet(X) \coloneqq ker\left( A^\bullet(X) \to A^\bullet(X_{p-1}) \right) \,,

where X pπ 1(B p)X_{p} \coloneqq \pi^{-1}(B_{p}) is the fiber over the ppth stage of the CW-complex B=lim nB nB = \underset{\longleftarrow}{\lim}_n B_n.


The exactness axiom for AA gives an exact couple, def. 24, of the form

s,tA s+t(X s) s,tA s+t(X s) s,tA s+t(X s,X s1)(A s+t(X s) A s+t(X s1) δ A s+t(X s,X s1) A s+t+1(X s,X s1)), \array{ \underset{s,t}{\prod} A^{s+t}(X_{s}) && \stackrel{}{\longrightarrow} && \underset{s,t}{\prod} A^{s+t}(X_{s}) \\ & \nwarrow && \swarrow \\ && \underset{s,t}{\prod} A^{s+t}(X_{s}, X_{s-1}) } \;\;\;\;\;\;\; \left( \array{ A^{s+t}(X_s) & \longrightarrow & A^{s+t}(X_{s-1}) \\ \uparrow && \downarrow_{\mathrlap{\delta}} \\ A^{s+t}(X_s, X_{s-1}) && A^{s+t+1}(X_{s}, X_{s-1}) } \right) \,,

where we take X 1=XX_{\gg 1} = X and X <0=X_{\lt 0} = \emptyset.

In order to determine the E 2E_2-page, we analyze the E 1E_1-page: By definition

E 1 s,t=A s+t(X s,X s1) E_1^{s,t} = A^{s+t}(X_s, X_{s-1})

Let C(s)C(s) be the set of ss-dimensional cells of BB, and notice that for σC(s)\sigma \in C(s) then

(π 1(σ),π 1(σ))(D n,S n1)×F σ, (\pi^{-1}(\sigma), \pi^{-1}(\partial \sigma)) \simeq (D^n, S^{n-1}) \times F_\sigma \,,

where F σF_\sigma is weakly homotopy equivalent to FF (exmpl.).

This implies that

E 1 s,t A s+t(X s,X s1) A˜ s+t(X s/X s1) A˜ s+t(σC(n)S sF +) σC(s)A˜ s+t(S sF +) σC(s)A˜ t(F +) σC(s)A t(F) C cell s(B,A t(F)), \begin{aligned} E_1^{s,t} & \coloneqq A^{s+t}(X_s, X_{s-1}) \\ & \simeq \tilde A^{s+t}(X_s/X_{s-1}) \\ & \simeq \tilde A^{s+t}(\underset{\sigma \in C(n)}{\vee} S^s \wedge F_+) \\ & \simeq \underset{\sigma \in C(s)}{\prod} \tilde A^{s+t}(S^s \wedge F_+) \\ & \simeq \underset{\sigma \in C(s)}{\prod} \tilde A^t(F_+) \\ & \simeq \underset{\sigma \in C(s)}{\prod} A^t(F) \\ & \simeq C^s_{cell}(B,A^t(F)) \end{aligned} \,,

where we used the relation to reduced cohomology A˜\tilde A, prop. 10 together with lemma 1, then the wedge axiom and the suspension isomorphism of the latter.

The last group C cell s(B,A t(F))C^s_{cell}(B,A^t(F)) appearing in this sequence of isomorphisms is that of cellular cochains (def.) of degree ss on BB with coefficients in the group A t(F)A^t(F).

Since cellular cohomology of a CW-complex agrees with its singular cohomology (thm.), hence with its ordinary cohomology, to conclude that the E 2E_2-page is as claimed, it is now sufficient to show that the differential d 1d_1 coincides with the differential in the cellular cochain complex (def.).

We discuss this now for π=id\pi = id, hence X=BX = B and F=*F = \ast. The general case works the same, just with various factors of FF appearing in the following:

Consider the following diagram, which commutes due to the naturality of the connecting homomorphism δ\delta of A A^\bullet:

*: C cell s1(X,A t(*)) = iI s1A t(*) iI sA t(*) = C cell s(X,A t(*)) iI s1A˜ s+t1(S s1) iI sA˜ s+t(S s) d 1: A s+t1(X s1,X s2) A s+t1(X s1) δ A s+t(X s,X s1) A s+t1(S s1,) A s+t1(S s1) δ A s+t(D s,S s1). \array{ \partial^\ast \colon & C^{s-1}_{cell}(X,A^t(\ast)) & =& \underset{i \in I_{s-1}}{\prod} A^t(\ast) && \longrightarrow && \underset{i \in I_s}{\prod} A^t(\ast) & = & C_{cell}^{s}(X,A^t(\ast)) \\ && & {}^{\mathllap{\simeq}}\downarrow && && \downarrow^{\mathrlap{\simeq}} \\ && & \underset{i \in I_{s-1}}{\prod} \tilde A^{s+t-1}(S^{s-1}) && && \underset{i \in I_s}{\prod} \tilde A^{s+t}(S^{s}) \\ && & {}^{\mathllap{\simeq}}\downarrow && && \downarrow^{\mathrlap{\simeq}} \\ && d_1 \colon & A^{s+t-1}(X_{s-1}, X_{s-2}) &\overset{}{\longrightarrow}& A^{s+t-1}(X_{s-1}) &\overset{\delta}{\longrightarrow}& A^{s+t}(X_s, X_{s-1}) \\ && & \downarrow && \downarrow && \downarrow \\ && & A^{s+t-1}(S^{s-1}, \emptyset) &\overset{}{\longrightarrow}& A^{s+t-1}(S^{s-1}) &\overset{\delta}{\longrightarrow}& A^{s+t}(D^s , S^{s-1}) } \,.

Here the bottom vertical morphisms are those induced from any chosen cell inclusion (D s,S s1)(X s,X s1)(D^s , S^{s-1}) \hookrightarrow (X_s, X_{s-1}).

The differential d 1d_1 in the spectral sequence is the middle horizontal composite. From this the vertical isomorphisms give the top horizontal map. But the bottom horizontal map identifies this top horizontal morphism componentwise with the restriction to the boundary of cells. Hence the top horizontal morphism is indeed the coboundary operator *\partial^\ast for the cellular cohomology of XX with coefficients in A (*)A^\bullet(\ast) (def.). This cellular cohomology coincides with singular cohomology of the CW-complex XX (thm.), hence computes the ordinary cohomology of XX.

Now to see the convergence. If BB is finite dimensional then the convergence condition as stated in prop. 23 is met. Alternatively, if A (F)A^\bullet(F) is bounded below in degree, then by the above analysis the E 1E_1-page has a horizontal line below which it vanishes. Accordingly the same is then true for all higher pages, by each of them being the cohomology of the previous page. Since the differentials go right and down, eventually they pass beneath this vanishing line and become 0. This is again the condition needed in the proof of prop. 23 to obtain convergence.

By that proposition the convergence is to the inverse limit

lim(A (X s+1)A (X s)). \underset{\longleftarrow}{\lim} \left( \cdots \stackrel{}{\to} A^\bullet(X_{s+1}) \longrightarrow A^\bullet(X_{s}) \to \cdots \right) \,.

If XX is finite dimensional or more generally if the sequences that this limit is over satisfy the Mittag-Leffler condition (def. 21), then this limit is A (X)A^\bullet(X), by prop. 18.

Multiplicative structure

For E E^\bullet a multiplicative cohomology theory (def. 14), then the Atiyah-Hirzebruch spectral sequences (prop. 24) for E (X)E^\bullet(X) are multiplicative spectral sequences.

A decent proof is spelled out in (Kochman 96, prop. 4.2.9). Use the graded commutativity of smash products of spheres to get the sign in the graded derivation law for the differentials. See also the proof via Cartan-Eilenberg systems at multiplicative spectral sequence – Examples – AHSS for multiplicative cohomology.


Given a multiplicative cohomology theory (A,μ,1)(A,\mu,1) (def. 14), then for every Serre fibration XBX \to B (def.) all the differentials in the corresponding Atiyah-Hirzebruch spectral sequence of prop. 24

H (B,A (F))A (X) H^\bullet(B,A^\bullet(F)) \;\Rightarrow\; A^\bullet(X)

are linear over A (*)A^\bullet(\ast).


By the proof of prop. 24, the differentials are those induced by the exact couple

s,tA s+t(X s) s,tA s+t(X s) s,tA s+t(X s,X s1)(A s+t(X s) A s+t(X s1) δ A s+t(X s,X s1) A s+t+1(X s,X s1)). \array{ \underset{s,t}{\prod} A^{s+t}(X_{s}) && \stackrel{}{\longrightarrow} && \underset{s,t}{\prod} A^{s+t}(X_{s}) \\ & \nwarrow && \swarrow \\ && \underset{s,t}{\prod} A^{s+t}(X_{s}, X_{s-1}) } \;\;\;\;\;\;\; \left( \array{ A^{s+t}(X_s) & \longrightarrow & A^{s+t}(X_{s-1}) \\ \uparrow && \downarrow_{\mathrlap{\delta}} \\ A^{s+t}(X_s, X_{s-1}) && A^{s+t+1}(X_{s}, X_{s-1}) } \right) \,.

consisting of the pullback homomorphisms and the connecting homomorphisms of AA.

By prop. 23 its differentials on page rr are the composites of one pullback homomorphism, the preimage of (r1)(r-1) pullback homomorphisms, and one connecting homomorphism of AA. Hence the statement follows with prop. 8.


For EE a homotopy commutative ring spectrum (def.) and XX a finite CW-complex, then the Kronecker pairing

, X:E 1(X)E 2(X)π 2 1(E) \langle-,-\rangle_X \;\colon\; E^{\bullet_1}(X) \otimes E_{\bullet_2}(X) \longrightarrow \pi_{\bullet_2-\bullet_1}(E)

extends to a compatible pairing of Atiyah-Hirzebruch spectral sequences.

(Kochman 96, prop. 4.2.10)

S.2) Cobordism theory

Idea. As one passes from abelian groups to spectra, a miracle happens: even though the latter are just the proper embodiment of linear algebra in the context of homotopy theory (“higher algebra”) their inspection reveals that spectra natively know about deep phenomena of differential topology, index theory and in fact string theory (for instance via a close relation between genera and partition functions).

A strong manifestation of this phenomenon comes about in complex oriented cohomology theory/chromatic homotopy theory that we eventually come to below. It turns out to be higher algebra over the complex Thom spectrum MU.

Here we first concentrate on its real avatar, the Thom spectrum MO. The seminal result of Thom's theorem says that the stable homotopy groups of MO form the cobordism ring of cobordism-equivalence classes of manifolds. In the course of discussing this cobordism theory one encounters various phenomena whose complex version also governs the complex oriented cohomology theory that we are interested in below.

Literature. (Kochman 96, chapter I and sections II.2, II6). A quick efficient account is in (Malkiewich 11). See also (Aguilar-Gitler-Prieto 02, section 11).

Classifying spaces and GG-Structure

Idea. Every manifold XX of dimension nn carries a canonical vector bundle of rank nn: its tangent bundle. There is a universal vector bundle of rank nn, of which all others arise by pullback, up to isomorphism. The base space of this universal bundle is hence called the classifying space and denoted BGL(n)BO(n)B GL(n) \simeq B O(n) (for O(n)O(n) the orthogonal group). This may be realized as the homotopy type of a direct limit of Grassmannian manifolds. In particular the tangent bundle of a manifold XX is classified by a map XBO(n)X \longrightarrow B O(n), unique up to homotopy. For GG a subgroup of O(n)O(n), then a lift of this map through the canonical map BGBO(n)B G \longrightarrow B O(n) of classifying spaces is a G-structure on XX

BG X BO(n) \array{ && B G \\ &\nearrow& \downarrow \\ X &\longrightarrow& B O(n) }

for instance an orientation for the inclusion SO(n)O(n)SO(n) \hookrightarrow O(n) of the special orthogonal group, or an almost complex structure for the inclusion U(n)O(2n)U(n) \hookrightarrow O(2n) of the unitary group.

All this generalizes, for instance from tangent bundles to normal bundles with respect to any embedding. It also behaves well with respect to passing to the boundary of manifolds, hence to bordism-classes of manifolds. This is what appears in Thom's theorem below.

Literature. (Kochman 96, 1.3-1.4), for stable normal structures also (Stong 68, beginning of chapter II)

Coset spaces

For XX a smooth manifold and GG a compact Lie group equipped with a free smooth action on XX, then the quotient projection

XX/G X \longrightarrow X/G

is a GG-principal bundle (hence in particular a Serre fibration).

This is originally due to (Gleason 50). See e.g. (Cohen, theorem 1.3)


For GG a Lie group and HGH \subset G a compact subgroup, then the coset quotient projection

GG/H G \longrightarrow G/H

is an HH-principal bundle (hence in particular a Serre fibration).


For GG a compact Lie group and KHGK \subset H \subset G closed subgroups, then the projection map on coset spaces

p:G/KG/H p \;\colon\; G/K \longrightarrow G/H

is a locally trivial H/KH/K-fiber bundle (hence in particular a Serre fibration).


Observe that the projection map in question is equivalently

G× H(H/K)G/H, G \times_H (H/K) \longrightarrow G/H \,,

(where on the left we form the Cartesian product and then divide out the diagonal action by HH). This exhibits it as the H/KH/K-fiber bundle associated to the HH-principal bundle of corollary 3.

Orthogonal and Unitary groups

The orthogonal group O(n)O(n) is compact topological space, hence in particular a compact Lie group.


The unitary group U(n)U(n) is compact topological space, hence in particular a compact Lie group.


The n-spheres are coset spaces of orthogonal groups:

S nO(n+1)/O(n). S^n \simeq O(n+1)/O(n) \,.

The odd-dimensional spheres are also coset spaces of unitary groups:

S 2n+1U(n+1)/U(n) S^{2n+1} \simeq U(n+1)/U(n)

Regarding the first statement:

Fix a unit vector in n+1\mathbb{R}^{n+1}. Then its orbit under the defining O(n+1)O(n+1)-action on n+1\mathbb{R}^{n+1} is clearly the canonical embedding S n n+1S^n \hookrightarrow \mathbb{R}^{n+1}. But precisely the subgroup of O(n+1)O(n+1) that consists of rotations around the axis formed by that unit vector stabilizes it, and that subgroup is isomorphic to O(n)O(n), hence S nO(n+1)/O(n)S^n \simeq O(n+1)/O(n).

The second statement follows by the same kind of reasoning:

Clearly U(n+1)U(n+1) acts transitively on the unit sphere S 2n+1S^{2n+1} in n+1\mathbb{C}^{n+1}. It remains to see that its stabilizer subgroup of any point on this sphere is U(n)U(n). If we take the point with coordinates (1,0,0,,0)(1,0, 0, \cdots,0) and regard elements of U(n+1)U(n+1) as matrices, then the stabilizer subgroup consists of matrices of the block diagonal form

(1 0 0 A) \left( \array{ 1 & \vec 0 \\ \vec 0 & A } \right)

where AU(n)A \in U(n).


For n,kn,k \in \mathbb{N}, nkn \leq k, then the canonical inclusion of orthogonal groups

O(n)O(k) O(n) \hookrightarrow O(k)

is an (n-1)-equivalence, hence induces an isomorphism on homotopy groups in degrees <n1\lt n-1 and a surjection in degree n1n-1.


Consider the coset quotient projection

O(n)O(n+1)O(n+1)/O(n). O(n) \longrightarrow O(n+1) \longrightarrow O(n+1)/O(n) \,.

By prop. 30 and by corollary 3, the projection O(n+1)O(n+1)/O(n)O(n+1)\to O(n+1)/O(n) is a Serre fibration. Furthermore, example 8 identifies the coset with the n-sphere

S nO(n+1)/O(n). S^{n}\simeq O(n+1)/O(n) \,.

Therefore the long exact sequence of homotopy groups (exmpl.)of the fiber sequence O(n)O(n+1)S nO(n)\to O(n+1)\to S^n has the form

π +1(S n)π (O(n))π (O(n+1))π (S n) \cdots \to \pi_{\bullet+1}(S^n) \longrightarrow \pi_\bullet(O(n)) \longrightarrow \pi_\bullet(O(n+1)) \longrightarrow \pi_\bullet(S^n) \to \cdots

Since π <n(S n)=0\pi_{\lt n}(S^n) = 0, this implies that

π <n1(O(n))π <n1(O(n+1)) \pi_{\lt n-1}(O(n)) \overset{\simeq}{\longrightarrow} \pi_{\lt n-1}(O(n+1))

is an isomorphism and that

π n1(O(n))π n1(O(n+1)) \pi_{n-1}(O(n)) \overset{\simeq}{\longrightarrow} \pi_{n-1}(O(n+1))

is surjective. Hence now the statement follows by induction over knk-n.



For n,kn,k \in \mathbb{N}, nkn \leq k, then the canonical inclusion of unitary groups

U(n)U(k) U(n) \hookrightarrow U(k)

is a 2n-equivalence, hence induces an isomorphism on homotopy groups in degrees <2n\lt 2n and a surjection in degree 2n2n.


Consider the coset quotient projection

U(n)U(n+1)U(n+1)/U(n). U(n) \longrightarrow U(n+1) \longrightarrow U(n+1)/U(n) \,.

By prop. 31 and corollary 3, the projection U(n+1)U(n+1)/U(n)U(n+1)\to U(n+1)/U(n) is a Serre fibration. Furthermore, example 8 identifies the coset with the (2n+1)-sphere

S 2n+1U(n+1)/U(n). S^{2n+1}\simeq U(n+1)/U(n) \,.

Therefore the long exact sequence of homotopy groups (exmpl.)of the fiber sequence U(n)U(n+1)S 2n+1U(n)\to U(n+1) \to S^{2n+1} is of the form

π +1(S 2n+1)π (U(n))π (U(n+1))π (S 2n+1) \cdots \to \pi_{\bullet+1}(S^{2n+1}) \longrightarrow \pi_\bullet(U(n)) \longrightarrow \pi_\bullet(U(n+1)) \longrightarrow \pi_\bullet(S^{2n+1}) \to \cdots

Since π 2n(S 2n+1)=0\pi_{\leq 2n}(S^{2n+1}) = 0, this implies that

π <2n(U(n))π <2n(U(n+1)) \pi_{\lt 2n}(U(n)) \overset{\simeq}{\longrightarrow} \pi_{\lt 2n}(U(n+1))

is an isomorphism and that

π 2n(U(n))π 2n(U(n+1)) \pi_{2n}(U(n)) \overset{\simeq}{\longrightarrow} \pi_{2n}(U(n+1))

is surjective. Hence now the statement follows by induction over knk-n.

Stiefel manifolds and Grassmannians

Throughout we work in the category Top cgTop_{cg} of compactly generated topological spaces (def.). For these the Cartesian product X×()X \times (-) is a left adjoint (prop.) and hence preserves colimits.


For n,kn, k \in \mathbb{N} and nkn \leq k, then the nnth real Stiefel manifold of k\mathbb{R}^k is the coset topological space.

V n( k)O(k)/O(kn), V_n(\mathbb{R}^k) \coloneqq O(k)/O(k-n) \,,

where the action of O(kn)O(k-n) is via its canonical embedding O(kn)O(k)O(k-n)\hookrightarrow O(k).

Similarly the nnth complex Stiefel manifold of k\mathbb{C}^k is

V n( k)U(k)/U(kn), V_n(\mathbb{C}^k) \coloneqq U(k)/U(k-n) \,,

here the action of U(kn)U(k-n) is via its canonical embedding U(kn)U(k)U(k-n)\hookrightarrow U(k).


For n,kn, k \in \mathbb{N} and nkn \leq k, then the nnth real Grassmannian of k\mathbb{R}^k is the coset topological space.

Gr n( k)O(k)/(O(n)×O(kn)), Gr_n(\mathbb{R}^k) \coloneqq O(k)/(O(n) \times O(k-n)) \,,

where the action of the product group is via its canonical embedding O(n)×O(kn)O(n)O(n)\times O(k-n) \hookrightarrow O(n) into the orthogonal group.

Similarly the nnth complex Grassmannian of k\mathbb{C}^k is the coset topological space.

Gr n( k)U(k)/(U(n)×U(kn)), Gr_n(\mathbb{C}^k) \coloneqq U(k)/(U(n) \times U(k-n)) \,,

where the action of the product group is via its canonical embedding U(n)×U(kn)U(n)U(n)\times U(k-n) \hookrightarrow U(n) into the unitary group.


For all nkn \leq k \in \mathbb{N}, the canonical projection from the Stiefel manifold (def. 25) to the Grassmannian is a O(n)O(n)-principal bundle

O(n) V n( k) Gr n( k) \array{ O(n) &\hookrightarrow& V_n(\mathbb{R}^k) \\ && \downarrow \\ && Gr_n(\mathbb{R}^k) }

and the projection from the complex Stiefel manifold to the Grassmannian us a U(n)U(n)-principal bundle:

U(n) V n( k) Gr n( k). \array{ U(n) &\hookrightarrow& V_n(\mathbb{C}^k) \\ && \downarrow \\ && Gr_n(\mathbb{C}^k) } \,.

By prop 3 and prop 29.


The real Grassmannians Gr n( k)Gr_n(\mathbb{R}^k) and the complex Grassmannians Gr n( k)Gr_n(\mathbb{C}^k) of def. 26 admit the structure of CW-complexes. Moreover the canonical inclusions

Gr n( k)Gr n( k+1) Gr_n(\mathbb{R}^k) \hookrightarrow Gr_n(\mathbb{R}^{k+1})

are subcomplex incusion (hence relative cell complex inclusions).

Accordingly there is an induced CW-complex structure on the classifying space (def. 27).

BO(n)lim kGr n( k). B O(n) \simeq \underset{\longrightarrow}{\lim}_k Gr_n(\mathbb{R}^k) \,.

A proof is spelled out in (Hatcher, section 1.2 (pages 31-34)).


The Stiefel manifolds V n( k)V_n(\mathbb{R}^k) and V n( k)V_n(\mathbb{C}^k) from def. 25 admits the structure of a CW-complex.

e.g. (James 59, p. 3, James 76, p. 5 with p. 21, Blaszczyk 07)

(And I suppose with that cell structure the inclusions V n( k)V n( k+1)V_n(\mathbb{R}^k) \hookrightarrow V_n(\mathbb{R}^{k+1}) are subcomplex inclusions.)


The real Stiefel manifold V n( k)V_n(\mathbb{R}^k) (def. 25) is (k-n-1)-connected.


Consider the coset quotient projection

O(kn)O(k)O(k)/O(kn)=V n( k). O(k-n) \longrightarrow O(k) \longrightarrow O(k)/O(k-n) = V_n(\mathbb{R}^k) \,.

By prop. 30 and by corollary 3, the projection O(k)O(k)/O(kn)O(k)\to O(k)/O(k-n) is a Serre fibration. Therefore there is induced the long exact sequence of homotopy groups of this fiber sequence, and by prop. 32 it has the following form in degrees bounded by nn:

π kn1(O(kn))epiπ kn1(O(k))0π kn1(V n( k))0π 1<kn1(O(k))π 1<kn1(O(kn)). \cdots \to \pi_{\bullet \leq k-n-1}(O(k-n)) \overset{epi}{\longrightarrow} \pi_{\bullet \leq k-n-1}(O(k)) \overset{0}{\longrightarrow} \pi_{\bullet \leq k-n-1}(V_n(\mathbb{R}^k)) \overset{0}{\longrightarrow} \pi_{\bullet-1 \lt k-n-1}(O(k)) \overset{\simeq}{\longrightarrow} \pi_{\bullet-1 \lt k-n-1}(O(k-n)) \to \cdots \,.

This implies the claim. (Exactness of the sequence says that every element in π n1(V n( k))\pi_{\bullet \leq n-1}(V_n(\mathbb{R}^k)) is in the kernel of zero, hence in the image of 0, hence is 0 itself.)



The complex Stiefel manifold V n( k)V_n(\mathbb{C}^k) (def. 25) is 2(k-n)-connected.


Consider the coset quotient projection

U(kn)U(k)U(k)/U(kn)=V n( k). U(k-n) \longrightarrow U(k) \longrightarrow U(k)/U(k-n) = V_n(\mathbb{C}^k) \,.

By prop. 31 and by corollary 3 the projection U(k)U(k)/U(kn)U(k)\to U(k)/U(k-n) is a Serre fibration. Therefore there is induced the long exact sequence of homotopy groups of this fiber sequence, and by prop. 33 it has the following form in degrees bounded by nn:

π 2(kn)(U(kn))epiπ 2(kn)(U(k))0π 2(kn)(V n( k))0π 1<2(kn)(U(k))π 1<2(kn)(U(kn)). \cdots \to \pi_{\bullet \leq 2(k-n)}(U(k-n)) \overset{epi}{\longrightarrow} \pi_{\bullet \leq 2(k-n)}(U(k)) \overset{0}{\longrightarrow} \pi_{\bullet \leq 2(k-n)}(V_n(\mathbb{C}^k)) \overset{0}{\longrightarrow} \pi_{\bullet-1 \lt 2(k-n)}(U(k)) \overset{\simeq}{\longrightarrow} \pi_{\bullet-1 \lt 2(k-n)}(U(k-n)) \to \cdots \,.

This implies the claim.

Classifying spaces

By def. 26 there are canonical inclusions

Gr n( k)Gr n( k+1) Gr_n(\mathbb{R}^k) \hookrightarrow Gr_n(\mathbb{R}^{k+1})


Gr n( k)Gr n( k+1) Gr_n(\mathbb{C}^k) \hookrightarrow Gr_n(\mathbb{C}^{k+1})

for all kk \in \mathbb{N}. The colimit (in Top, see there, or rather in Top cgTop_{cg}, see this cor.) over these inclusions is denoted

BO(n)lim kGr n( k) B O(n) \coloneqq \underset{\longrightarrow}{\lim}_k Gr_n(\mathbb{R}^k)


BU(n)lim kGr n( k), B U(n) \coloneqq \underset{\longrightarrow}{\lim}_k Gr_n(\mathbb{C}^k) \,,


Moreover, by def. 25 there are canonical inclusions

V n( k)V n( k+1) V_n(\mathbb{R}^k) \hookrightarrow V_n(\mathbb{R}^{k+1})


V n( k)V n( k+1) V_n(\mathbb{C}^k) \hookrightarrow V_n(\mathbb{C}^{k+1})

that are compatible with the O(n)O(n)-action and with the U(n)U(n)-action, respectively. The colimit (in Top, see there, or rather in Top cgTop_{cg}, see this cor.) over these inclusions, regarded as equipped with the induced O(n)O(n)-action, is denoted

EO(n)lim kV n( k) E O(n) \coloneqq \underset{\longrightarrow}{\lim}_k V_n(\mathbb{R}^k)


EU(n)lim kV n( k), E U(n) \coloneqq \underset{\longrightarrow}{\lim}_k V_n(\mathbb{C}^k) \,,


The inclusions are in fact compatible with the bundle structure from prop. 34, so that there are induced projections

(EO(n) BO(n))lim k(V n( k) Gr n( k)) \left( \array{ E O(n) \\ \downarrow \\ B O(n) } \right) \;\; \simeq \;\; \underset{\longrightarrow}{\lim}_k \left( \array{ V_n(\mathbb{R}^k) \\ \downarrow \\ Gr_n(\mathbb{R}^k) } \right)


(EU(n) BU(n))lim k(V n( k) Gr n( k)), \left( \array{ E U(n) \\ \downarrow \\ B U(n) } \right) \;\; \simeq \;\; \underset{\longrightarrow}{\lim}_k \left( \array{ V_n(\mathbb{C}^k) \\ \downarrow \\ Gr_n(\mathbb{C}^k) } \right) \,,

respectively. These are the standard models for the universal principal bundles for OO and UU, respectively. The corresponding associated vector bundles

EO(n)×O(n) n E O(n) \underset{O(n)}{\times} \mathbb{R}^n


EU(n)×U(n) n E U(n) \underset{U(n)}{\times} \mathbb{C}^n

are the corresponding universal vector bundles.

Since the Cartesian product O(n)×()O(n)\times (-) in compactly generated topological spaces preserves colimits, it follows that the colimiting bundle is still an O(n)O(n)-principal bundle

(EO(n))/O(n) (lim kV n( k))/O(n) lim k(V n( k)/O(n)) lim kGr n( k) BO(n), \begin{aligned} (E O(n))/O(n) & \simeq (\underset{\longrightarrow}{\lim}_k V_{n}(\mathbb{R}^k))/O(n) \\ & \simeq \underset{\longrightarrow}{\lim}_k (V_n(\mathbb{R}^k)/O(n)) \\ & \simeq \underset{\longrightarrow}{\lim}_k Gr_n(\mathbb{R}^k) \\ & \simeq B O(n) \end{aligned} \,,

and anlogously for EU(n)E U(n).

As such this is the standard presentation for the O(n)O(n)-universal principal bundle and U(n)U(n)-universal principal bundle, respectively. Its base space BO(n)B O(n) is the corresponding classifying space.


There are canonical inclusions

Gr n( k)Gr n+1( k+1) Gr_n(\mathbb{R}^k) \hookrightarrow Gr_{n+1}(\mathbb{R}^{k+1})


Gr n( k)Gr n+1( k+1) Gr_n(\mathbb{C}^k) \hookrightarrow Gr_{n+1}(\mathbb{C}^{k+1})

given by adjoining one coordinate to the ambient space and to any subspace. Under the colimit of def. 27 these induce maps of classifying spaces

BO(n)BO(n+1) B O(n) \longrightarrow B O(n+1)


BU(n)BU(n+1). B U(n) \longrightarrow B U(n+1) \,.

There are canonical maps

Gr n 1( k 1)×Gr n 2( k 2)Gr n 1+n 2( k 1+k 2) Gr_{n_1}(\mathbb{R}^{k_1}) \times Gr_{n_2}(\mathbb{R}^{k_2}) \longrightarrow Gr_{n_1 + n_2}(\mathbb{R}^{k_1 + k_2})


Gr n 1( k 1)×Gr n 2( k 2)Gr n 1+n 2( k 1+k 2) Gr_{n_1}(\mathbb{C}^{k_1}) \times Gr_{n_2}(\mathbb{C}^{k_2}) \longrightarrow Gr_{n_1 + n_2}(\mathbb{C}^{k_1 + k_2})

given by sending ambient spaces and subspaces to their direct sum.

Under the colimit of def. 27 these induce maps of classifying spaces

BO(n 1)×BO(n 2)BO(n 1+n 2) B O(n_1) \times B O(n_2) \longrightarrow B O(n_1 + n_2)


BU(n 1)×BU(n 2)BU(n 1+n 2) B U(n_1) \times B U(n_2) \longrightarrow B U(n_1 + n_2)

The colimiting space EO(n)=lim kV n( k)E O(n) = \underset{\longrightarrow}{\lim}_k V_n(\mathbb{R}^k) from def. 27 is weakly contractible.

The colimiting space EU(n)=lim kV n( k)E U(n) = \underset{\longrightarrow}{\lim}_k V_n(\mathbb{C}^k) from def. 27 is weakly contractible.


By propositions 37, and 38, the Stiefel manifolds are more and more highly connected as kk increases. Since the inclusions are relative cell complex inclusions by prop. 36, the claim follows.


The homotopy groups of the classifying spaces BO(n)B O(n) and BU(n)B U(n) (def. 27) are those of the orthogonal group O(n)O(n) and of the unitary group U(n)U(n), respectively, shifted up in degree: there are isomorphisms

π +1(BO(n))π O(n) \pi_{\bullet+1}(B O(n)) \simeq \pi_\bullet O(n)


π +1(BU(n))π U(n) \pi_{\bullet+1}(B U(n)) \simeq \pi_\bullet U(n)

(for homotopy groups based at the canonical basepoint).


Consider the sequence

O(n)EO(n)BO(n) O(n) \longrightarrow E O(n) \longrightarrow B O(n)

from def. 27, with O(n)O(n) the fiber. Since (by prop. 29) the second map is a Serre fibration, this is a fiber sequence and so it induces a long exact sequence of homotopy groups of the form

π (O(n))π (EO(n))π (BO(n))π 1(O(n))π 1(EO(n)). \cdots \to \pi_\bullet(O(n)) \longrightarrow \pi_\bullet(E O(n)) \longrightarrow \pi_\bullet(B O(n)) \longrightarrow \pi_{\bullet-1}(O (n)) \longrightarrow \pi_{\bullet-1}(E O(n)) \to \cdots \,.

Since by cor. 39 π (EO(n))=0\pi_\bullet(E O(n))= 0, exactness of the sequence implies that

π (BO(n))π 1(O(n)) \pi_\bullet(B O(n)) \overset{\simeq}{\longrightarrow} \pi_{\bullet-1}(O (n))

is an isomorphism.

The same kind of argument applies to the complex case.


For nn \in \mathbb{N} there are homotopy fiber sequence (def.)

S nBO(n)BO(n+1) S^n \longrightarrow B O(n) \longrightarrow B O(n+1)


S 2n+1BU(n)BU(n+1) S^{2n+1} \longrightarrow B U(n) \longrightarrow B U(n+1)

exhibiting the n-sphere ((2n+1)(2n+1)-sphere) as the homotopy fiber of the canonical maps from def. 28.

This means (thm.), that there is a replacement of the canonical inclusion BO(n)BO(n+1)B O(n) \hookrightarrow B O(n+1) (induced via def. 27) by a Serre fibration

BO(n) BO(n+1) weakhomotopyequivalence Serrefib. B˜O(n) \array{ B O(n) &\hookrightarrow& B O(n+1) \\ {}^{\mathllap{{weak \, homotopy} \atop equivalence}}\downarrow & \nearrow_{\mathrlap{Serre \, fib.}} \\ \tilde B O(n) }

such that S nS^n is the ordinary fiber of BO(n)B˜O(n+1)B O(n)\to \tilde B O(n+1), and analogously for the complex case.


Take B˜O(n)(EO(n+1))/O(n)\tilde B O(n) \coloneqq (E O(n+1))/O(n).

To see that the canonical map BO(n)(EO(n+1))/O(n)B O(n)\longrightarrow (E O(n+1))/O(n) is a weak homotopy equivalence consider the commuting diagram

O(n) id O(n) EO(n) EO(n+1) BO(n) (EO(n+1))/O(n). \array{ O(n) &\overset{id}{\longrightarrow}& O(n) \\ \downarrow && \downarrow \\ E O(n) &\longrightarrow& E O(n+1) \\ \downarrow && \downarrow \\ B O(n) &\longrightarrow& (E O(n+1))/O(n) } \,.

By prop. 29 both bottom vertical maps are Serre fibrations and so both vertical sequences are fiber sequences. By prop. 40 part of the induced morphisms of long exact sequences of homotopy groups looks like this

π (BO(n)) π ((EO(n+1))/O(n)) π 1(O(n)) = π 1(O(n)), \array{ \pi_\bullet(B O(n)) &\overset{}{\longrightarrow}& \pi_\bullet( (E O(n+1))/O(n) ) \\ {}^{\mathllap{\simeq}}\downarrow && \downarrow^{\mathrlap{\simeq}} \\ \pi_{\bullet-1}(O(n)) &\overset{=}{\longrightarrow}& \pi_{\bullet-1}(O(n)) } \,,

where the vertical and the bottom morphism are isomorphisms. Hence also the to morphisms is an isomorphism.

That BO(n)B˜O(n+1)B O(n)\to \tilde B O(n+1) is indeed a Serre fibration follows again with prop. 29, which gives the fiber sequence

O(n+1)/O(n)(EO(n+1))/O(n)(EO(n+1))/O(n+1). O(n+1)/O(n) \longrightarrow (E O(n+1))/O(n) \longrightarrow (E O(n+1))/O(n+1) \,.

The claim then follows with the identification

O(n+1)/O(n)S n O(n+1)/O(n) \simeq S^n

of example 8.

The argument for the complex case is directly analogous, concluding instead with the identification

U(n+1)/U(n)S 2n+1 U(n+1)/U(n)\simeq S^{2n+1}

from example 8.

GG-Structure on the Stable normal bundle

Given a smooth manifold XX of dimension nn and equipped with an embedding

i:X k i \;\colon\; X \hookrightarrow \mathbb{R}^k

for some kk \in \mathbb{N}, then the classifying map of its normal bundle is the function

g i:XGr kn( k)BO(kn) g_i \;\colon\; X \to Gr_{k-n}(\mathbb{R}^k) \hookrightarrow B O(k-n)

which sends xXx \in X to the normal of the tangent space

N xX=(T xX) k N_x X = (T_x X)^{\perp} \hookrightarrow \mathbb{R}^k

regarded as a point in G kn( k)G_{k-n}(\mathbb{R}^k).

The normal bundle of ii itself is the subbundle of the tangent bundle

T k k× k T \mathbb{R}^k \simeq \mathbb{R}^k \times \mathbb{R}^k

consisting of those vectors which are orthogonal to the tangent vectors of XX:

N i{xX,vT i(x) k|vi *T xXT i(x) k}. N_i \coloneqq \left\{ x\in X, v \in T_{i(x)}\mathbb{R}^k \;\vert\; v \,\perp\, i_\ast T_x X \subset T_{i(x)}\mathbb{R}^k \right\} \,.

A (B,f)(B,f)-structure is

  1. for each nn\in \mathbb{N} a pointed CW-complex B nTop CW */B_n \in Top_{CW}^{\ast/}

  2. equipped with a pointed Serre fibration

    B n f n BO(n) \array{ B_n \\ \downarrow^{\mathrlap{f_n}} \\ B O(n) }

    to the classifying space BO(n)B O(n) (def.);

  3. for all n 1n 2n_1 \leq n_2 a pointed continuous function

    g n 1,n 2:B n 1B n 2g_{n_1, n_2} \;\colon\; B_{n_1} \longrightarrow B_{n_2}

    which is the identity for n 1=n 2n_1 = n_2;

such that for all n 1n 2n_1 \leq n_2 \in \mathbb{N} these squares commute

B n 1 g n 1,n 2 B n 2 f n 1 f n 2 BO(n 1) BO(n 2), \array{ B_{n_1} &\overset{g_{n_1,n_2}}{\longrightarrow}& B_{n_2} \\ {}^{\mathllap{f_{n_1}}}\downarrow && \downarrow^{\mathrlap{f_{n_2}}} \\ B O(n_1) &\longrightarrow& B O(n_2) } \,,

where the bottom map is the canonical one from def. 28.

The (B,f)(B,f)-structure is multiplicative if it is moreover equipped with a system of maps μ n 1,n 2:B n 1×B n 2B n 1+n 2\mu_{n_1,n_2} \colon B_{n_1}\times B_{n_2} \to B_{n_1 + n_2} which cover the canonical multiplication maps (def.)

B n 1×B n 2 μ n 1,n 2 B n 1+n 2 f n 1×f n 2 f n 1+n 2 BO(n 1)×BO(n 2) BO(n 1+n 2) \array{ B_{n_1} \times B_{n_2} &\overset{\mu_{n_1, n_2}}{\longrightarrow}& B_{n_1 + n_2} \\ {}^{\mathllap{f_{n_1} \times f_{n_2}}}\downarrow && \downarrow^{\mathrlap{f_{n_1 + n_2}}} \\ B O(n_1) \times B O(n_2) &\longrightarrow& B O(n_1 + n_2) }

and which satisfy the evident associativity and unitality, for B 0=*B_0 = \ast the unit, and, finally, which commute with the maps gg in that all n 1,n 2,n 3n_1,n_2, n_3 \in \mathbb{N} these squares commute:

B n 1×B n 2 id×g n 2,n 2+n 3 B n 1×B n 2+n 3 μ n 1,n 2 μ n 1,n 2+n 3 B n 1+n 2 g n 1+n 2,n 1+n 2+n 3 B n 1+n 2+n 3 \array{ B_{n_1} \times B_{n_2} &\overset{id \times g_{n_2,n_2+n_3}}{\longrightarrow}& B_{n_1} \times B_{n_2 + n_3} \\ {}^{\mathllap{\mu_{n_1, n_2}}}\downarrow && \downarrow^{\mathrlap{\mu_{n_1,n_2 + n_3}}} \\ B_{n_1 + n_2} &\underset{g_{n_1+n_2, n_1 + n_2 + n_3}}{\longrightarrow}& B_{n_1 + n_2 + n_3} }


B n 1×B n 2 g n 1,n 1+n 3×id B n 1+n 3×B n 2 μ n 1,n 2 μ n 1+n 3,n 2 B n 1+n 2 g n 1+n 2,n 1+n 2+n 3 B n 1+n 2+n 3. \array{ B_{n_1} \times B_{n_2} &\overset{g_{n_1,n_1+n_3} \times id}{\longrightarrow}& B_{n_1+n_3} \times B_{n_2 } \\ {}^{\mathllap{\mu_{n_1, n_2}}}\downarrow && \downarrow^{\mathrlap{\mu_{n_1 + n_3 , n_2}}} \\ B_{n_1 + n_2} &\underset{g_{n_1+n_2, n_1 + n_2 + n_3}}{\longrightarrow}& B_{n_1 + n_2 + n_3} } \,.

Similarly, an S 2S^2-(B,f)(B,f)-structure is a compatible system

f 2n:B 2nBO(2n) f_{2n} \colon B_{2n} \longrightarrow B O(2n)

indexed only on the even natural numbers.

Generally, an S kS^k-(B,f)(B,f)-structure for kk \in \mathbb{N}, k1k \geq 1 is a compatible system

f kn:B knBO(kn) f_{k n} \colon B_{k n} \longrightarrow B O(k n)

for all nn \in \mathbb{N}, hence for all knkk n \in k \mathbb{N}.


Examples of (B,f)(B,f)-structures (def. 31) include the following:

  1. B n=BO(n)B_n = B O(n) and f n=idf_n = id is orthogonal structure (or “no structure”);

  2. B n=EO(n)B_n = E O(n) and f nf_n the universal principal bundle-projection is framing-structure;

  3. B n=BSO(n)=EO(n)/SO(n)B_n = B SO(n) = E O(n)/SO(n) the classifying space of the special orthogonal group and f nf_n the canonical projection is orientation structure;

  4. B n=BSpin(n)=EO(n)/Spin(n)B_n = B Spin(n) = E O(n)/Spin(n) the classifying space of the spin group and f nf_n the canonical projection is spin structure.

Examples of S 2S^2-(B,f)(B,f)-structures (def. 31) include

  1. B 2n=BU(n)=EO(2n)/U(n)B_{2n} = B U(n) = E O(2n)/U(n) the classifying space of the unitary group, and f 2nf_{2n} the canonical projection is almost complex structure (or rather: almost Hermitian structure).

  2. B 2n=BSp(2n)=EO(2n)/Sp(2n)B_{2n} = B Sp(2n) = E O(2n)/Sp(2n) the classifying space of the symplectic group, and f 2nf_{2n} the canonical projection is almost symplectic structure.

Examples of S 4S^4-(B,f)(B,f)-structures (def. 31) include

  1. B 4n=BU (n)=EO(4n)/U (n)B_{4n} = B U_{\mathbb{H}}(n) = E O(4n)/U_{\mathbb{H}}(n) the classifying space of the quaternionic unitary group, and f 4nf_{4n} the canonical projection is almost quaternionic structure.

Given a smooth manifold XX of dimension nn, and given a (B,f)(B,f)-structure as in def. 31, then a (B,f)(B,f)-structure on the stable normal bundle of the manifold is an equivalence class of the following structure:

  1. an embedding i X:X ki_X \; \colon \; X \hookrightarrow \mathbb{R}^k for some kk \in \mathbb{N};

  2. a homotopy class of a lift g^\hat g of the classifying map gg of the normal bundle (def. 30)

    B kn g^ f kn X g BO(kn). \array{ && B_{k-n} \\ &{}^{\mathllap{\hat g}}\nearrow& \downarrow^{\mathrlap{f_{k-n}}} \\ X &\overset{g}{\longrightarrow}& B O(k-n) } \,.

The equivalence relation on such structures is to be that generated by the relation ((i X) 1,g^ 1)((i X) ,g^ 2)((i_{X})_1, \hat g_1) \sim ((i_{X})_,\hat g_2) if

  1. k 2k 1k_2 \geq k_1

  2. the second inclusion factors through the first as

    (i X) 2:X(i X) 1 k 1 k 2 (i_X)_2 \;\colon\; X \overset{(i_X)_1}{\hookrightarrow} \mathbb{R}^{k_1} \hookrightarrow \mathbb{R}^{k_2}
  3. the lift of the classifying map factors accordingly (as homotopy classes)

    g^ 2:Xg^ 1B k 1ng k 1n,k 2nB k 2n. \hat g_2 \;\colon\; X \overset{\hat g_1}{\longrightarrow} B_{k_1-n} \overset{g_{k_1-n, k_2-n}}{\longrightarrow} B_{k_2-n} \,.

Thom spectra

Idea. Given a vector bundle VV of rank nn over a compact topological space, then its one-point compactification is equivalently the result of forming the bundle D(V)VD(V) \hookrightarrow V of unit n-balls, and identifying with one single point all the boundary unit n-spheres S(V)VS(V)\hookrightarrow V. Generally, this construction Th(C)D(V)/S(V)Th(C) \coloneqq D(V)/S(V) is called the Thom space of VV.

Thom spaces occur notably as codomains for would-be left inverses of embeddings of manifolds XYX \hookrightarrow Y. The Pontrjagin-Thom collapse map YTh(NX)Y \to Th(N X) of such an embedding is a continuous function going the other way around, but landing not quite in XX but in the Thom space of the normal bundle of XX in YY. Composing this further with the classifying map of the normal bundle lands in the Thom space of the universal vector bundle over the classifying space BO(k)B O(k), denoted MO(k)M O(k). In particular in the case that Y=S nY = S^n is an n-sphere (and every manifold embeds into a large enough nn-sphere, see also at Whitney embedding theorem), the Pontryagin-Thom collapse map hence associates with every manifold an element of a homotopy group of a universal Thom space MO(k)M O(k).

This curious construction turns out to have excellent formal properties: as the dimension ranges, the universal Thom spaces arrange into a spectrum, called the Thom spectrum, and the homotopy groups defined by the Pontryagin-Thom collapse pass along to the stable homotopy groups of this spectrum.

Moreover, via Whitney sum of vector bundle the Thom spectrum naturally is a homotopy commutative ring spectrum (def.), and under the Pontryagin-Thom collapse the Cartesian product of manifolds is compatible with this ring structure.

Literature. (Kochman 96, 1.5, Schwede 12, chapter I, example 1.16)

Thom spaces

Let XX be a topological space and let VXV \to X be a vector bundle over XX of rank nn, which is associated to an O(n)-principal bundle. Equivalently this means that VXV \to X is the pullback of the universal vector bundle E nBO(n)E_n \to B O(n) (def. 27) over the classifying space. Since O(n)O(n) preserves the metric on n\mathbb{R}^n, by definition, such VV inherits the structure of a metric space-fiber bundle. With respect to this structure:

  1. the unit disk bundle D(V)XD(V) \to X is the subbundle of elements of norm 1\leq 1;

  2. the unit sphere bundle S(V)XS(V)\to X is the subbundle of elements of norm =1= 1;

    S(V)i VD(V)VS(V) \overset{i_V}{\hookrightarrow} D(V) \hookrightarrow V;

  3. the Thom space Th(V)Th(V) is the cofiber (formed in Top (prop.)) of i Vi_V

    Th(V)cofib(i V) Th(V) \coloneqq cofib(i_V)

    canonically regarded as a pointed topological space.

S(V) i V D(V) (po) * Th(V). \array{ S(V) &\overset{i_V}{\longrightarrow}& D(V) \\ \downarrow &(po)& \downarrow \\ \ast &\longrightarrow& Th(V) } \,.

If VXV \to X is a general real vector bundle, then there exists an isomorphism to an O(n)O(n)-associated bundle and the Thom space of VV is, up to based homeomorphism, that of this orthogonal bundle.


If the rank of VV is positive, then S(V)S(V) is non-empty and then the Thom space (def. 33) is the quotient topological space

Th(V)D(V)/S(V). Th(V) \simeq D(V)/S(V) \,.

However, in the degenerate case that the rank of VV vanishes, hence the case that V=X× 0XV = X\times \mathbb{R}^0 \simeq X, then D(V)VXD(V) \simeq V \simeq X, but S(V)=S(V) = \emptyset. Hence now the pushout defining the cofiber is

i V X (po) * Th(V)X *, \array{ \emptyset &\overset{i_V}{\longrightarrow}& X \\ \downarrow &(po)& \downarrow \\ \ast &\longrightarrow& Th(V) \simeq X_* } \,,

which exhibits Th(V)Th(V) as the coproduct of XX with the point, hence as XX with a basepoint freely adjoined.

Th(X× 0)=Th(X)X +. Th(X \times \mathbb{R}^0) = Th(X) \simeq X_+ \,.

Let VXV \to X be a vector bundle over a CW-complex XX. Then the Thom space Th(V)Th(V) (def. 33) is equivalently the homotopy cofiber (def.) of the inclusion S(V)D(V)S(V) \longrightarrow D(V) of the sphere bundle into the disk bundle.


The Thom space is defined as the ordinary cofiber of S(V)D(V)S(V)\to D(V). Under the given assumption, this inclusion is a relative cell complex inclusion, hence a cofibration in the classical model structure on topological spaces (thm.). Therefore in this case the ordinary cofiber represents the homotopy cofiber (def.).

The equivalence to the following alternative model for this homotopy cofiber is relevant when discussing Thom isomorphisms and orientation in generalized cohomology:


Let VXV \to X be a vector bundle over a CW-complex XX. Write VXV-X for the complement of its 0-section. Then the Thom space Th(V)Th(V) (def. 33) is homotopy equivalent to the mapping cone of the inclusion (VX)V(V-X) \hookrightarrow V (hence to the pair (V,VX)(V,V-X) in the language of generalized (Eilenberg-Steenrod) cohomology).


The mapping cone of any map out of a CW-complex represents the homotopy cofiber of that map (exmpl.). Moreover, transformation by (weak) homotopy equivalences between morphisms induces a (weak) homotopy equivalence on their homotopy fibers (prop.). But we have such a weak homotopy equivalence, given by contracting away the fibers of the vector bundle:

VX V W cl W cl S(V) D(V). \array{ V-X &\longrightarrow& V \\ {}^{\mathllap{\in W_{cl}}}\downarrow && \downarrow^{\mathrlap{\in W_{cl}}} \\ S(V) &\hookrightarrow& D(V) } \,.

Let V 1,V 2XV_1,V_2 \to X be two real vector bundles. Then the Thom space (def. 33) of the direct sum of vector bundles V 1V 2XV_1 \oplus V_2 \to X is expressed in terms of the Thom space of the pullbacks V 2| D(V 1)V_2|_{D(V_1)} and V 2| S(V 1)V_2|_{S(V_1)} of V 2V_2 to the disk/sphere bundle of V 1V_1 as

Th(V 1V 2)Th(V 2| D(V 1))/Th(V 2| S(V 1)). Th(V_1 \oplus V_2) \simeq Th(V_2|_{D(V_1)})/Th(V_2|_{S(V_1)}) \,.

Notice that

  1. D(V 1V 2)D(V 2| IntD(V 1))S(V 1)D(V_1 \oplus V_2) \simeq D(V_2|_{Int D(V_1)}) \cup S(V_1);

  2. S(V 1V 2)S(V 2| IntD(V 1))IntD(V 2| S(V 1))S(V_1 \oplus V_2) \simeq S(V_2|_{Int D(V_1)}) \cup Int D(V_2|_{S(V_1)}).

(Since a point at radius rr in V 1V 2V_1 \oplus V_2 is a point of radius r 1rr_1 \leq r in V 2V_2 and a point of radius r 2r 1 2\sqrt{r^2 - r_1^2} in V 1V_1.)


For VV a vector bundle then the Thom space (def. 33) of nV\mathbb{R}^n \oplus V, the direct sum of vector bundles with the trivial rank nn vector bundle, is homeomorphic to the smash product of the Thom space of VV with the nn-sphere (the nn-fold reduced suspension).

Th( nV)S nTh(V)=Σ nTh(V). Th(\mathbb{R}^n \oplus V) \simeq S^n \wedge Th(V) = \Sigma^n Th(V) \,.

Apply prop. 44 with V 1= nV_1 = \mathbb{R}^n and V 2=VV_2 = V. Since V 1V_1 is a trivial bundle, then

V 2| D(V 1)V 2×D n V_2|_{D(V_1)} \simeq V_2\times D^n

(as a bundle over X×D nX\times D^n) and similarly

V 2| S(V 1)V 2×S n. V_2|_{S(V_1)} \simeq V_2\times S^n \,.

By prop. 45 and remark 7 the Thom space (def. 33) of a trivial vector bundle of rank nn is the nn-fold suspension of the base space

Th(X× n) S nTh(X× 0) S n(X +). \begin{aligned} Th(X \times \mathbb{R}^n) & \simeq S^n \wedge Th(X\times \mathbb{R}^0) \\ & \simeq S^n \wedge (X_+) \end{aligned} \,.

Therefore a general Thom space may be thought of as a “twisted suspension”, with twist encoded by a vector bundle (or rather by its underlying spherical fibration). See at Thom spectrum – For infinity-module bundles for more on this.

Correspondingly the Thom isomorphism (prop. 54 below) for a given Thom space is a twisted version of the suspension isomorphism (above).


For V 1X 1V_1 \to X_1 and V 2X 2V_2 \to X_2 to vector bundles, let V 1V 2X 1×X 2V_1 \boxtimes V_2 \to X_1 \times X_2 be the direct sum of vector bundles of their pullbacks to X 1×X 2X_1 \times X_2. The corresponding Thom space (def. 33) is the smash product of the individual Thom spaces:

Th(V 1V 2)Th(V 1)Th(V 2). Th(V_1 \boxtimes V_2) \simeq Th(V_1) \wedge Th(V_2) \,.

Given a vector bundle VXV \to X of rank nn, then the reduced ordinary cohomology of its Thom space Th(V)Th(V) (def. 33) vanishes in degrees <n\lt n:

H˜ <n(Th(V))H <n(D(V),S(V))0. \tilde H^{\bullet \lt n}(Th(V)) \simeq H^{\bullet \lt n}(D(V), S(V)) \simeq 0 \,.

Consider the long exact sequence of relative cohomology (from above)

H 1(D(V))i *H 1(S(V))H (D(V),S(V))H (D(V))i *H (S(V)). \cdots \to H^{\bullet-1}(D(V)) \overset{i^\ast}{\longrightarrow} H^{\bullet-1}(S(V)) \longrightarrow H^\bullet(D(V), S(V)) \longrightarrow H^{\bullet}(D(V)) \overset{i^\ast}{\longrightarrow} H^{\bullet}(S(V)) \to \cdots \,.

Since the cohomology in degree kk only depends on the kk-skeleton, and since for k<nk \lt n the kk-skeleton of S(V)S(V) equals that of XX, and since D(V)D(V) is even homotopy equivalent to XX, the morhism i *i^\ast is an isomorphism in degrees lower than nn. Hence by exactness of the sequence it follows that H <n(D(V),S(V))=0H^{\bullet \lt n}(D(V),S(V)) = 0.

Universal Thom spectra MGM G

For each nn \in \mathbb{N} the pullback of the rank-(n+1)(n+1) universal vector bundle to the classifying space of rank nn vector bundles is the direct sum of vector bundles of the rank nn universal vector bundle with the trivial rank-1 bundle: there is a pullback diagram of topological spaces of the form

(EO(n)×O(n) n) EO(n+1)×O(n+1) n+1 (pb) BO(n) BO(n+1), \array{ \mathbb{R}\oplus (E O(n)\underset{O(n)}{\times} \mathbb{R}^n) &\longrightarrow& E O(n+1) \underset{O(n+1)}{\times} \mathbb{R}^{n+1} \\ \downarrow &(pb)& \downarrow \\ B O(n) &\longrightarrow& B O(n+1) } \,,

where the bottom morphism is the canonical one (def.).

(e.g. Kochmann 96, p. 25)


For each kk \in \mathbb{N}, knk \geq n there is such a pullback of the canonical vector bundles over Grassmannians

{V n k,vV n,v n+1} {V n+1 k+1,vV n+1} Gr n( k) Gr n+1( k+1) \array{ \left\{ {V_{n}\subset \mathbb{R}^k, } \atop {v \in V_n, v_{n+1} \in \mathbb{R}} \right\} &\longrightarrow& \left\{ {V_{n+1} \subset \mathbb{R}^{k+1}}, \atop v \in V_{n+1} \right\} \\ \downarrow && \downarrow \\ Gr_n(\mathbb{R}^k) &\longrightarrow& Gr_{n+1}(\mathbb{R}^{k+1}) }

where the bottom morphism is the canonical inclusion (def.).

Now we claim that taking the colimit in each of the four corners of this system of pullback diagrams yields again a pullback diagram, and this proves the claim.

To see this, remember that we work in the category Top cgTop_{cg} of compactly generated topological spaces (def.). By their nature, we may test the universal property of a would-be pullback space already by mapping compact topological spaces into it. Now observe that all the inclusion maps in the four corners of this system of diagrams are relative cell complex inclusions, by prop. 35. Together this implies (via this lemma) that we may test the universal property of the colimiting square at finite stages. And so this implies the claim by the above fact that at each finite stage there is a pullback diagram.


The universal real Thom spectrum MOM O is the spectrum, which is represented by the sequential prespectrum (def.) whose nnth component space is the Thom space (def. 33)

(MO) nTh(EO(n)×O(n) n) (M O)_n \coloneqq Th(E O(n)\underset{O(n)}{\times}\mathbb{R}^n)

of the rank-nn universal vector bundle, and whose structure maps are the image under the Thom space functor Th()Th(-) of the top morphisms in prop. 47, via the homeomorphisms of prop. 45:

σ n:Σ(MO) nTh((EO(n)×O(n) n))Th(EO(n+1)×O(n+1) n+1)=(MO) n+1. \sigma_n \;\colon\; \Sigma (M O)_n \simeq Th(\mathbb{R}\oplus (E O(n)\underset{O(n)}{\times} \mathbb{R}^n)) \stackrel{}{\longrightarrow} Th(E O(n+1) \underset{O(n+1)}{\times} \mathbb{R}^{n+1}) = (M O)_{n+1} \,.

More generally, there are universal Thom spectra associated with any other tangent structure (“[[(B,f)]-structure]]”), notably for the orthogonal group replaced by the special orthogonal groups SO(n)SO(n), or the spin groups Spin(n)Spin(n), or the string 2-group String(n)String(n), or the fivebrane 6-group Fivebrane(n)Fivebrane(n),…, or any level in the Whitehead tower of O(n)O(n). To any of these groups there corresponds a Thom spectrum (denoted, respectively, MSOM SO, MSpin, MStringM String, MFivebraneM Fivebrane, etc.), which is in turn related to oriented cobordism, spin cobordism, string cobordism, et cetera.:


Given a (B,f)-structure \mathcal{B} (def. 31), write V n V^\mathcal{B}_n for the pullback of the universal vector bundle (def. 27) to the corresponding space of the (B,f)(B,f)-structure and with

V VO(n)×O(n) n (pb) B n f n BO(n) \array{ V^{\mathcal{B}} &\overset{}{\longrightarrow}& V O(n) \underset{O(n)}{\times} \mathbb{R}^n \\ \downarrow &(pb)& \downarrow \\ B_n &\underset{f_n}{\longrightarrow}& B O(n) }

and we write e n 1,n 2e_{n_1,n_2} for the maps of total space of vector bundles over the g n 1,n 2g_{n_1,n_2}:

V n 1 e n 1,n 2 V n 2 (pb) B n 1 g n 1,n 2 B n 2. \array{ V^{\mathcal{B}}_{n_1} &\overset{e_{n_1,n_2}}{\longrightarrow}& V^{\mathcal{B}}_{n_2} \\ \downarrow &(pb)& \downarrow \\ B_{n_1} &\underset{g_{n_1,n_2}}{\longrightarrow}& B_{n_2} } \,.

Observe that the analog of prop. 47 still holds:


Given a (B,f)-structure \mathcal{B} (def. 31), then the pullback of its rank-(n+1)(n+1) vector bundle V n+1 V^{\mathcal{B}}_{n+1} (def. 35) along the map g n,n+1:B nB n+1g_{n,n+1} \colon B_n \to B_{n+1} is the direct sum of vector bundles of the rank-nn bundle V n V^{\mathcal{B}}_n with the trivial rank-1-bundle: there is a pullback square

V n e n,n+1 V n+1 (pb) B n g n,n+1 B n+1. \array{ \mathbb{R} \oplus V^{\mathcal{B}}_n &\overset{e_{n,n+1}}{\longrightarrow}& V^{\mathcal{B}}_{n+1} \\ \downarrow &(pb)& \downarrow \\ B_n &\underset{g_{n,n+1}}{\longrightarrow}& B_{n+1} } \,.

Unwinding the definitions, the pullback in question is

(g n,n+1) *V n+1 =(g n,n+1) *f n+1 *(EO(n+1)×O(n+1) n+1) (g n,n+1f n+1) *(EO(n+1)×O(n+1) n+1) (f ni n) *(EO(n+1)×O(n+1) n+1) f n *i n *(EO(n+1)×O(n+1) n+1) f n *((EO(n)×O(n) n)) V n, \begin{aligned} (g_{n,n+1})^\ast V^{\mathcal{B}}_{n+1} & = (g_{n,n+1})^\ast f_{n+1}^\ast (E O(n+1)\underset{O(n+1)}{\times} \mathbb{R}^{n+1}) \\ & \simeq (g_{n,n+1} \circ f_{n+1})^\ast (E O(n+1)\underset{O(n+1)}{\times} \mathbb{R}^{n+1}) \\ & \simeq ( f_n \circ i_n )^\ast (E O(n+1)\underset{O(n+1)}{\times} \mathbb{R}^{n+1}) \\ & \simeq f_n^\ast i_n^\ast (E O(n+1)\underset{O(n+1)}{\times} \mathbb{R}^{n+1}) \\ & \simeq f_n^\ast (\mathbb{R} \oplus (E O(n)\underset{O(n)}{\times} \mathbb{R}^{n})) \\ &\simeq \mathbb{R} \oplus V^{\mathcal{B}_n} \,, \end{aligned}

where the second but last step is due to prop. 47.


Given a (B,f)-structure \mathcal{B} (def. 31), its universal Thom spectrum MM \mathcal{B} is, as a sequential prespectrum, given by component spaces being the Thom spaces (def. 33) of the \mathcal{B}-associated vector bundles of def. 35

(M) nTh(V n ) (M \mathcal{B})_n \coloneqq Th(V^{\mathcal{B}}_n)

and with structure maps given via prop. 45 by the top maps in prop. 48:

σ n:Σ(M) n=ΣTh(V n )Th(V n )Th(e n,n+1)Th(V n+1 )=(M) n+1. \sigma_n \;\colon\; \Sigma (M \mathcal{B})_n = \Sigma Th(V^{\mathcal{E}}_n) \simeq Th(\mathbb{R}\oplus V^{\mathcal{E}}_n) \overset{Th(e_{n,n+1})}{\longrightarrow} Th(V^{\mathcal{B}}_{n+1}) = (M \mathcal{B})_{n+1} \,.

Similarly for an S k(B,f)S^k-(B,f)-structure indexed on every kkth natural number (such as almost complex structure, almost quaternionic structure, example 10), there is the corresponding Thom spectrum as a sequential S kS^k spectrum (def.).

If B n=BG nB_n = B G_n for some natural system of groups G nO(n)G_n \to O(n), then one usually writes MGM G for MM \mathcal{B}. For instance MSOM SO, MSpin, MU, MSp etc.

If the (B,f)(B,f)-structure is multiplicative (def. 31), then the Thom spectrum MM \mathcal{B} canonical becomes a ring spectrum (for more on this see Part 1-2 the section on orthogonal Thom spectra ): the multiplication maps B n 1×B n 2B n 1+n 2B_{n_1} \times B_{n_2}\to B_{n_1 + n_2} are covered by maps of vector bundles

V n 1 V n 2 V n 1+n 2 V^{\mathcal{B}}_{n_1} \boxtimes V^{\mathcal{B}}_{n_2} \longrightarrow V^{\mathcal{B}}_{n_1 + n_2}

and under forming Thom spaces this yields (via prop. 46) maps

(M) n 1(M) n 2(M) n 1+n 2 (M \mathcal{B})_{n_1} \wedge (M \mathcal{B})_{n_2} \longrightarrow (M \mathcal{B})_{n_1 + n_2}

which are associative by the associativity condition in a multiplicative (B,f)(B,f)-structure. The unit is

(M) 0=Th(V 0 )Th(*)S 0, (M \mathcal{B})_0 = Th(V^{\mathcal{B}}_0) \simeq Th(\ast) \simeq S^0 \,,

by remark 7.


The universal Thom spectrum (def. 36) for framing structure (exmpl.) is equivalently the sphere spectrum (def.)

M1𝕊. M 1 \simeq \mathbb{S} \,.

Because in this case B n*B_n \simeq \ast and so E n nE^{\mathcal{B}}_n \simeq \mathbb{R}^n, whence Th(E n )S nTh(E^{\mathcal{B}}_n) \simeq S^n.

Pontrjagin-Thom construction

For XX a smooth manifold and i:X ki \colon X \hookrightarrow \mathbb{R}^k an embedding, then a tubular neighbourhood of XX is a subset of the form

τ iX{x k|d(x,i(X))<ϵ} \tau_i X \coloneqq \left\{ x \in \mathbb{R}^k \;\vert\; d(x,i(X)) \lt \epsilon \right\}

for some ϵ\epsilon \in \mathbb{R}, ϵ>0\epsilon \gt 0, small enough such that the map

N iXτ iX N_i X \longrightarrow \tau_i X

from the normal bundle (def. 30) given by

(i(x),v)(i(x),ϵ(1e |v|)v) (i(x),v) \mapsto (i(x), \epsilon (1-e^{- {\vert v\vert}}) v )

is a diffeomorphism.


(tubular neighbourhood theorem)

For every embedding of smooth manifolds, there exists a tubular neighbourhood according to def. 37.


Given an embedding i:X ki \colon X \hookrightarrow \mathbb{R}^k with a tubuluar neighbourhood τ iXhookrigtharrow k\tau_i X \hookrigtharrow \mathbb{R}^k (def. 37) then by construction:

  1. the Thom space (def. 33) of the normal bundle (def. 30) is homeomorphic to the quotient topological space of the topological closure of the tubular neighbourhood by its boundary:

    Th(N i(X))τ i(X)¯/τ i(X)¯Th(N_i(X)) \simeq \overline{ \tau_i(X)}/\partial \overline{\tau_i(X)};

  2. there exists a continous function

    kτ i(X)¯/τ i(X)¯ \mathbb{R}^k \longrightarrow \overline{ \tau_i(X)}/\partial \overline{\tau_i(X)}

    which is the identity on τ i(X) k\tau_i(X)\subset \mathbb{R}^k and is constant on the basepoint of the quotient on all other points.


For XX a smooth manifold of dimension nn and for i:X ki \colon X \hookrightarrow \mathbb{R}^k an embedding, then the Pontrjagin-Thom collapse map is, for any choice of tubular neighbourhood τ i(X) k\tau_i(X)\subset \mathbb{R}^k (def. 37) the composite map of pointed topological spaces

S k( k) *τ i(X)¯/τ i(X)¯Th(N iX) S^k \overset{\simeq}{\to} (\mathbb{R}^k)^\ast \longrightarrow \overline{ \tau_i(X)}/\partial \overline{\tau_i(X)} \overset{\simeq}{\to} Th(N_i X)

where the first map identifies the k-sphere as the one-point compactification of k\mathbb{R}^k; and where the second and third maps are those of remark 9.

The Pontrjagin-Thom construction is the further composite

ξ i:S kTh(N iX)Th(e i)Th(EO(kn)×O(kn) kn)(MO) kn \xi_i \;\colon\; S^k \longrightarrow Th(N_i X) \overset{Th(e_i)}{\longrightarrow} Th( E O(k-n) \underset{O(k-n)}{\times} \mathbb{R}^{k-n} ) \simeq (M O)_{k-n}

with the image under the Thom space construction of the morphism of vector bundles

ν e i EO(kn)×O(kn) kn (pb) X g i BO(kn) \array{ \nu &\overset{e_i}{\longrightarrow}& E O(k-n)\underset{O(k-n)}{\times} \mathbb{R}^{k-n} \\ \downarrow &(pb)& \downarrow \\ X &\underset{g_i}{\longrightarrow}& B O(k-n) }

induced by the classifying map g ig_i of the normal bundle (def. 30).

This defines an element

[S n+(kn)ξ i(MO) kn]π nMO [S^{n+(k-n)} \overset{\xi_i}{\to} (M O)_{k-n}] \in \pi_{n} M O

in the nnth stable homotopy group (def.) of the Thom spectrum MOM O (def. 34).

More generally, for XX a smooth manifold with normal (B,f)-structure (X,i,g^ i)(X,i,\hat g_i) according to def. 32, then its Pontrjagin-Thom construction is the composite

ξ i:S kTh(N iX)Th(e^ i)Th(V kn )(M) kn \xi_i \;\colon\; S^k \longrightarrow Th(N_i X) \overset{Th(\hat e_i)}{\longrightarrow} Th( V^{\mathcal{B}}_{k-n} ) \simeq (M \mathcal{B})_{k-n}


ν e^ i V kn (pb) X g^ i BO(kn). \array{ \nu &\overset{\hat e_i}{\longrightarrow}& V^{\mathcal{B}}_{k-n} \\ \downarrow &(pb)& \downarrow \\ X &\underset{\hat g_i}{\longrightarrow}& B O(k-n) } \,.

The Pontrjagin-Thom construction (def. 38) respects the equivalence classes entering the definition of manifolds with stable normal \mathcal{B}-structure (def. 32) hence descends to a function (of sets)

ξ:{n-manifoldswithstablenormal-structure}π n(M). \xi \;\colon\; \left\{ {n\text{-}manifolds\;with\;stable} \atop {normal\;\mathcal{B}\text{-}structure} \right\} \longrightarrow \pi_n(M\mathcal{B}) \,.

It is clear that the homotopies of classifying maps of \mathcal{B}-structures that are devided out in def. 32 map to homotopies of representatives of stable homotopy groups. What needs to be shown is that the construction respects the enlargement of the embedding spaces.

Given a embedded manifold Xi k 1X \overset{i}{\hookrightarrow}\mathbb{R}^{k_1} with normal \mathcal{B}-structure

B k 1n g^ i f kn X g i BO(k 1n) \array{ && B_{k_1-n} \\ & {}^{\mathllap{\hat g_i}}\nearrow & \downarrow^{\mathrlap{f_{k-n}}} \\ X &\underset{g_i}{\longrightarrow}& B O(k_1-n) }


α:S n+(k 1n)Th(E k 1n) \alpha \;\colon\; S^{n+(k_1-n)} \overset{}{\longrightarrow} Th(E^{\mathcal{B}_{k_1-n}})

for its image under the Pontrjagin-Thom construction (def. 38). Now given k 2k_2 \in \mathbb{N}, consider the induced embedding Xi k 1 k 1+k 2X \overset{i}{\hookrightarrow} \mathbb{R}^{k_1}\hookrightarrow \mathbb{R}^{k_1 + k_2} with normal \mathcal{B}-structure given by the composite

B k 1n g k 1n,k 1+k 2n B k 1+k 2n g^ i f k 1n×f k 2 f k 1+k 2n X g i BO(k 1n) BO(k 1+k 2n). \array{ && B_{k_1-n} &\overset{g_{k_1-n, k_1+ k_2 -n}}{\longrightarrow}& B_{k_1 + k_2-n} \\ & {}^{\mathllap{\hat g_i}}\nearrow & \downarrow^{\mathrlap{f_{k_1 - n} \times f_{k_2}}} && \downarrow^{\mathrlap{f_{k_1 + k_2-n}}} \\ X &\underset{g_i}{\longrightarrow}& B O(k_1-n) &\longrightarrow& B O(k_1 + k_2-n) } \,.

By prop. 48 and using the pasting law for pullbacks, the classifying map g^ i\hat g'_i for the enlarged normal bundle sits in a diagram of the form

(ν i k 2) (e^ iid) (V k 1n k 2) e k 1n,k 1+k 2n V k 1+k 2n (pb) (pb) X g^ i B k 1n g k 1n,k 1+k 2n B k 1+k 2n. \array{ (\nu_i \oplus \mathbb{R}^{k_2}) &\overset{(\hat e_i \oplus id)}{\longrightarrow}& (V^{\mathcal{B}}_{k_1-n} \oplus \mathbb{R}^{k_2}) &\overset{e_{k_1-n,k_1+k_2-n}}{\longrightarrow}& V^{\mathcal{B}}_{k_1 + k_2 - n} \\ \downarrow &(pb)& \downarrow &(pb)& \downarrow \\ X &\underset{\hat g_i}{\longrightarrow}& B_{k_1-n} &\underset{g_{k_1-n, k_1 + k_2 - n}}{\longrightarrow}& B_{k_1 +k_2 - n} } \,.

Hence the Pontrjagin-Thom construction for the enlarged embedding space is (using prop. 45) the composite

α k 2:S n+(k 1+k 2n)Th( k 2)S n+(k 1n)Th( k 2)Th(ν i)Th(id)Th(e^ i)Th( k 2)Th(E k 1n ))Th(e k 1n,k 1+k 2n)Th(V k 1+k 2n ). \alpha_{k_2} \;\colon\; S^{n + (k_1+ k_2 - n)} \simeq Th(\mathbb{R}^{k_2}) \wedge S^{n + (k_1 - n)} \overset{}{\longrightarrow} Th(\mathbb{R}^{k_2}) \wedge Th(\nu_i) \overset{Th(id)\wedge Th(\hat e_i)}{\longrightarrow} Th(\mathbb{R}^{k_2}) \wedge Th(E^{\mathcal{B}}_{k_1-n})) \overset{Th(e_{k_1-n, k_1 + k_2 - n})}{\longrightarrow} Th(V^{\mathcal{B}}_{k_1 + k_2 - n}) \,.

The composite of the first two morphisms here is S k kαS^{k_k}\wedge \alpha, while last morphism Th(e^ k 1n,k 1+k 2n)Th(\hat e_{k_1-n,k_1+k_2-n}) is the structure map in the Thom spectrum (by def. 36):

α k 2:S k 2S n+(k 1n)S k 2αS k 2Th(E k 1+k 2n )σ k 1n,k 1+k 2n MTh(V k 1+k 2n ) \alpha_{k_2} \;\colon\; S^{k_2} \wedge S^{n + (k_1 - n)} \overset{S^{k_2} \wedge \alpha}{\longrightarrow} S^{k_2} \wedge Th(E^{\mathcal{B}}_{k_1 + k_2 - n}) \overset{\sigma^{M \mathcal{B}}_{k_1-n,k_1 + k_2 - n} }{\longrightarrow} Th(V^{\mathcal{B}}_{k_1+k_2 - n})

This manifestly identifies α k 2\alpha_{k_2} as being the image of α\alpha under the component map in the sequential colimit that defines the stable homotopy groups (def.). Therefore α\alpha and α k 2\alpha_{k_2}, for all k 2k_2 \in \mathbb{N}, represent the same element in π (M)\pi_{\bullet}(M \mathcal{B}).

Bordism and Thom’s theorem

Idea. By the Pontryagin-Thom collapse construction above, there is an assignment

nManifoldsπ n(MO) n Manifolds \longrightarrow \pi_n(M O)

which sends disjoint union and Cartesian product of manifolds to sum and product in the ring of stable homotopy groups of the Thom spectrum. One finds then that two manifolds map to the same element in the stable homotopy groups π (MO)\pi_\bullet(M O) of the universal Thom spectrum precisely if they are connected by a bordism. The bordism-classes Ω O\Omega_\bullet^O of manifolds form a commutative ring under disjoint union and Cartesian product, called the bordism ring, and Pontrjagin-Thom collapse produces a ring homomorphism

Ω Oπ (MO). \Omega_\bullet^O \longrightarrow \pi_\bullet(M O) \,.

Thom's theorem states that this homomorphism is an isomorphism.

More generally, for \mathcal{B} a multiplicative (B,f)-structure, def. 31, there is such an identification

Ω π (M) \Omega_\bullet^{\mathcal{B}} \simeq \pi_\bullet(M \mathcal{B})

between the ring of \mathcal{B}-cobordism classes of manifolds with \mathcal{B}-structure and the stable homotopy groups of the universal \mathcal{B}-Thom spectrum.

Literature. (Kochman 96, 1.5)


Throughout, let \mathcal{B} be a multiplicative (B,f)-structure (def. 31).


Write I[0,1]I \coloneqq [0,1] for the standard interval, regarded as a smooth manifold with boundary. For c +c \in \mathbb{R}_+ Consider its embedding

e:I 0 e \;\colon\; I \hookrightarrow \mathbb{R}\oplus \mathbb{R}_{\geq 0}

as the arc

e:tcos(πt)e 1+sin(πt)e 2, e \;\colon\; t \mapsto \cos(\pi t) \cdot e_1 + \sin(\pi t) \cdot e_2 \,,

where (e 1,e 2)(e_1, e_2) denotes the canonical linear basis of 2\mathbb{R}^2, and equipped with the structure of a manifold with normal framing structure (example 10) by equipping it with the canonical framing

fr:tcos(πt)e 1+sin(πt)e 2 fr \;\colon\; t \mapsto \cos(\pi t) \cdot e_1 + \sin(\pi t) \cdot e_2

of its normal bundle.

Let now \mathcal{B} be a (B,f)-structure (def. 31). Then for Xi kX \overset{i}{\hookrightarrow}\mathbb{R}^k any embedded manifold with \mathcal{B}-structure g^:XB kn\hat g \colon X \to B_{k-n} on its normal bundle (def. 32), define its negative or orientation reversal (X,i,g^)-(X,i,\hat g) of (X,i,g^)(X,i, \hat g) to be the restriction of the structured manifold

(X×I(i,e) k+2,g^×fr) (X \times I \overset{(i,e)}{\hookrightarrow} \mathbb{R}^{k+2}, \hat g \times fr)

to t=1t = 1.


Two closed manifolds of dimension nn equipped with normal \mathcal{B}-structure (X 1,i 1,g^ 1)(X_1, i_1, \hat g_1) and (X 2,i 2,g^ 2)(X_2,i_2,\hat g_2) (def.) are called bordant if there exists a manifold with boundary WW of dimension n+1n+1 equipped with \mathcal{B}-strcuture (W,i W,g^ W)(W,i_W, \hat g_W) if its boundary with \mathcal{B}-structure restricted to that boundary is the disjoint union of X 1X_1 with the negative of X 2X_2, according to def. 39

(W,i W,g^ W)(X 1,i 1,g^ 1)(X 2,i 2,g^ 2). \partial(W,i_W,\hat g_W) \simeq (X_1, i_1, \hat g_1) \sqcup -(X_2, i_2, \hat g_2) \,.

The relation of \mathcal{B}-bordism (def. 40) is an equivalence relation.

Write Ω \Omega^\mathcal{B}_{\bullet} for the \mathbb{N}-graded set of \mathcal{B}-bordism classes of \mathcal{B}-manifolds.


Under disjoint union of manifolds, then the set of \mathcal{B}-bordism equivalence classes of def. 51 becomes an \mathbb{Z}-graded abelian group

Ω Ab \Omega^{\mathcal{B}}_\bullet \in Ab^{\mathbb{Z}}

(that happens to be concentrated in non-negative degrees). This is called the \mathcal{B}-bordism group.

Moreover, if the (B,f)-structure \mathcal{B} is multiplicative (def. 31), then Cartesian product of manifolds followed by the multiplicative composition operation of \mathcal{B}-structures makes the \mathcal{B}-bordism ring into a commutative ring, called the \mathcal{B}-bordism ring.

Ω CRing . \Omega^{\mathcal{B}}_\bullet \in CRing^{\mathbb{Z}} \,.

e.g. (Kochmann 96, prop. 1.5.3)

Thom’s theorem

Recall that the Pontrjagin-Thom construction (def. 38) associates to an embbeded manifold (X,i,g^)(X,i,\hat g) with normal \mathcal{B}-structure (def. 32) an element in the stable homotopy group π dim(X)(M)\pi_{dim(X)}(M \mathcal{B}) of the universal \mathcal{B}-Thom spectrum in degree the dimension of that manifold.


For \mathcal{B} be a multiplicative (B,f)-structure (def. 31), the \mathcal{B}-Pontrjagin-Thom construction (def. 38) is compatible with all the relations involved to yield a graded ring homomorphism

ξ:Ω π (M) \xi \;\colon\; \Omega^{\mathcal{B}}_\bullet \longrightarrow \pi_\bullet(M \mathcal{B})

from the \mathcal{B}-bordism ring (def. 52) to the stable homotopy groups of the universal \mathcal{B}-Thom spectrum equipped with the ring structure induced from the canonical ring spectrum structure (def. 36).


By prop. 50 the underlying function of sets is well-defined before dividing out the bordism relation (def. 40). To descend this further to a function out of the set underlying the bordism ring, we need to see that the Pontrjagin-Thom construction respects the bordism relation. But the definition of bordism is just so as to exhibit under ξ\xi a left homotopy of representatives of homotopy groups.

Next we need to show that it is

  1. a group homomorphism;

  2. a ring homomorphism.

Regarding the first point:

The element 0 in the cobordism group is represented by the empty manifold. It is clear that the Pontrjagin-Thom construction takes this to the trivial stable homotopy now.

Given two nn-manifolds with \mathcal{B}-structure, we may consider an embedding of their disjoint union into some k\mathbb{R}^{k} such that the tubular neighbourhoods of the two direct summands do not intersect. There is then a map from two copies of the k-cube, glued at one face

k k1 k k \Box^k \underset{\Box^{k-1}}{\sqcup} \Box^k \longrightarrow \mathbb{R}^k

such that the first manifold with its tubular neighbourhood sits inside the image of the first cube, while the second manifold with its tubular neighbourhood sits indide the second cube. After applying the Pontryagin-Thom construction to this setup, each cube separately maps to the image under ξ\xi of the respective manifold, while the union of the two cubes manifestly maps to the sum of the resulting elements of homotopy groups, by the very definition of the group operation in the homotopy groups (def.). This shows that ξ\xi is a group homomorphism.

Regarding the second point:

The element 1 in the cobordism ring is represented by the manifold which is the point. Without restriction we may consoder this as embedded into 0\mathbb{R}^0, by the identity map. The corresponding normal bundle is of rank 0 and hence (by remark 7) its Thom space is S 0S^0, the 0-sphere. Also V 0 V^{\mathcal{B}}_0 is the rank-0 vector bundle over the point, and hence (M) 0S 0(M \mathcal{B})_0 \simeq S^0 (by def. 36) and so ξ(*):(S 0S 0)\xi(\ast) \colon (S^0 \overset{\simeq}{\to} S^0) indeed represents the unit element in π (M)\pi_\bullet(M\mathcal{B}).

Finally regarding respect for the ring product structure: for two manifolds with stable normal \mathcal{B}-structure, represented by embeddings into k i\mathbb{R}^{k_i}, then the normal bundle of the embedding of their Cartesian product is the direct sum of vector bundles of the separate normal bundles bulled back to the product manifold. In the notation of prop. 46 there is a diagram of the form

ν 1ν 2 e^ 1e^ 2 V n 1 V n 2 κ n 1,n 2 V n 1+n 2 (pb) (pb) X 1×X 2 g^ 1×g^ 2 B k 1n 1×B k 2n 2 μ k 1n 1,k 2n 2 B k 1+k 2n 1n 2. \array{ \nu_1 \boxtimes \nu_2 &\overset{\hat e_1 \boxtimes \hat e_2}{\longrightarrow}& V^{\mathcal{B}}_{n_1} \boxtimes V^{\mathcal{B}}_{n_2} &\overset{\kappa_{n_1,n_2}}{\longrightarrow}& V^{\mathcal{B}}_{n_1 + n_2} \\ \downarrow &(pb)& \downarrow &(pb)& \downarrow \\ X_1 \times X_2 &\underset{\hat g_1 \times \hat g_2}{\longrightarrow}& B_{k_1-n_1} \times B_{k_2-n_2} &\underset{\mu_{k_1-n_1,k_2-n_2}}{\longrightarrow}& B_{k_1 + k_2 - n_1 - n_2} } \,.

To the Pontrjagin-Thom construction of the product manifold is by definition the top composite in the diagram

S n 1+n 2+(k 1+k 2n 1n 2) Th(ν 1ν 2) Th(e^ 1e^ 2) Th(V k 1n 1 V k 2n 2 ) κ k 1n 1,k 2n 2 Th(V k 1+k 2n 1n 2 ) = S n 1+(k 1n 1)S n 2+(k 2n 2) Th(ν 1)Th(ν 2) Th(e^ 1)Th(e^ 2) Th(V 1 )Th(V 2 ) κ k 1n 1,k 2n 2 Th(V k 1+k 2n 1n 2 ), \array{ S^{n_1 +n_2 + (k_1 + k_2 - n_1 - n_2)} &\overset{}{\longrightarrow}& Th(\nu_1 \boxtimes \nu_2) &\overset{Th(\hat e_1 \boxtimes \hat e_2)}{\longrightarrow}& Th(V^{\mathcal{B}}_{k_1-n_1} \boxtimes V^{\mathcal{B}}_{k_2-n_2}) &\overset{\kappa_{k_1-n_1, k_2-n_2}}{\longrightarrow}& Th(V^{\mathcal{B}}_{k_1 + k_2 - n_1 - n_2}) \\ {}^{\mathllap{\simeq}}\downarrow && \downarrow^{\mathrlap{\simeq}} && \downarrow^{\mathrlap{\simeq}} && \downarrow^{\mathrlap{=}} \\ S^{n_1 + (k_1 - n_1)} \wedge S^{n_2 + (k_2 - n_2)} &\overset{}{\longrightarrow}& Th(\nu_1) \wedge Th(\nu_2) &\overset{Th(\hat e_1)\wedge Th(\hat e_2)}{\longrightarrow}& Th(V^{\mathcal{B}}_1) \wedge Th(V^{\mathcal{B}}_2) &\overset{\kappa_{k_1-n_1, k_2-n_2}}{\longrightarrow}& Th(V^{\mathcal{B}}_{k_1 + k_2 - n_1 - n_2}) } \,,

which hence is equivalently the bottom composite, which in turn manifestly represents the product of the separate PT constructions in π (M)\pi_\bullet(M\mathcal{B}).


The ring homomorphsim in lemma 5 is an isomorphism.

Due to (Thom 54, Pontrjagin 55). See for instance (Kochmann 96, theorem 1.5.10).

Proof idea

Observe that given the result α:S n+(kn)Th(V kn)\alpha \colon S^{n+(k-n)} \to Th(V_{k-n}) of the Pontrjagin-Thom construction map, the original manifold Xi kX \overset{i}{\hookrightarrow} \mathbb{R}^k may be recovered as this pullback:

X i S n+(kn) g i (pb) α BO(kn) Th(V kn BO). \array{ X &\overset{i}{\longrightarrow}& S^{n + (k-n)} \\ {}^{\mathllap{g_i}}\downarrow &(pb)& \downarrow^{\mathrlap{\alpha}} \\ B O(k-n) &\longrightarrow& Th(V^{B O}_{k-n}) } \,.

To see this more explicitly, break it up into pieces:

X X + S n+(kn) (pb) (pb) X X +Th(X) Th(0) Th(ν i) (pb) (pb) B kn (B kn) +Th(B kn) Th(0) Th(V kn ) (pb) (pb) BO(kn) (BO(kn)) +Th(BO(kn)) Th(V kn BO). \array{ X &\longrightarrow& X_+ &\hookrightarrow& S^{n + (k-n)} \\ \downarrow &(pb)& \downarrow &(pb)& \downarrow \\ X &\longrightarrow& X_+ \simeq Th(X) &\overset{Th(0)}{\longrightarrow}& Th(\nu_i) \\ \downarrow &(pb)& \downarrow &(pb)& \downarrow \\ B_{k-n} &\longrightarrow& (B_{k-n})_+ \simeq Th(B_{k-n}) &\underset{Th(0)}{\longrightarrow}& Th(V^{\mathcal{B}}_{k-n}) \\ \downarrow &(pb)& \downarrow &(pb)& \downarrow \\ B O(k-n) &\longrightarrow& (B O(k-n))_+ \simeq Th(B O(k-n)) &\longrightarrow& Th(V^{B O}_{k-n}) } \,.

Moreover, since the n-spheres are compact topological spaces, and since the classifying space BO(n)B O(n), and hence its universal Thom space, is a sequential colimit over relative cell complex inclusions, the right vertical map factors through some finite stage (by this lemma), the manifold XX is equivalently recovered as a pullback of the form

X S n+(kn) g i (pb) Gr kn( k) i Th(V kn( k)×O(kn) kn). \array{ X &\longrightarrow& S^{n + (k-n)} \\ {}^{\mathllap{g_i}}\downarrow &(pb)& \downarrow \\ Gr_{k-n}(\mathbb{R}^k) &\overset{i}{\longrightarrow}& Th( V_{k-n}(\mathbb{R}^k) \underset{O(k-n)}{\times} \mathbb{R}^{k-n}) } \,.

(Recall that V kn V^{\mathcal{B}}_{k-n} is our notation for the universal vector bundle with \mathcal{B}-structure, while V kn( k)V_{k-n}(\mathbb{R}^k) denotes a Stiefel manifold.)

The idea of the proof now is to use this property as the blueprint of the construction of an inverse ζ\zeta to ξ\xi: given an element in π n(M)\pi_{n}(M \mathcal{B}) represented by a map as on the right of the above diagram, try to define XX and the structure map g ig_i of its normal bundle as the pullback on the left.

The technical problem to be overcome is that for a general continuous function as on the right, the pullback has no reason to be a smooth manifold, and for two reasons:

  1. the map S n+(kn)Th(V kn)S^{n+(k-n)} \to Th(V_{k-n}) may not be smooth around the image of ii;

  2. even if it is smooth around the image of ii, it may not be transversal to ii, and the intersection of two non-transversal smooth functions is in general still not a smooth manifold.

The heart of the proof is in showing that for any α\alpha there are small homotopies relating it to an α\alpha' that is both smooth around the image of ii and transversal to ii.

The first condition is guaranteed by Sard's theorem, the second by Thom's transversality theorem.


Thom isomorphism

Idea. If a vector bundle EpXE \stackrel{p}{\longrightarrow} X of rank nn carries a cohomology class ωH n(Th(E),R)\omega \in H^n(Th(E),R) that looks fiberwise like a volume form – a Thom class – then the operation of pulling back from base space and then forming the cup product with this Thom class is an isomorphism on (reduced) cohomology

(()ω)p *:H (X,R)H˜ +n(Th(E),R). ( (-) \cup \omega) \circ p^\ast \;\colon\; H^\bullet(X,R) \stackrel{\simeq}{\longrightarrow} \tilde H^{\bullet+n}(Th(E),R) \,.

This is the Thom isomorphism. It follows from the Serre spectral sequence (or else from the Leray-Hirsch theorem). A closely related statement gives the Thom-Gysin sequence.

In the special case that the vector bundle is trivial of rank nn, then its Thom space coincides with the nn-fold suspension of the base space (example 11) and the Thom isomorphism coincides with the suspension isomorphism. In this sense the Thom isomorphism may be regarded as a twisted suspension isomorphism.

We need this below to compute (co)homology of universal Thom spectra MUM U in terms of that of the classifying spaces BUB U.

Composed with pullback along the Pontryagin-Thom collapse map, the Thom isomorphism produces maps in cohomology that covariantly follow the underlying maps of spaces. These “Umkehr maps” have the interpretation of fiber integration against the Thom class.

Literature. (Kochman 96, 2.6)

Thom-Gysin sequence

The Thom-Gysin sequence is a type of long exact sequence in cohomology induced by a spherical fibration and expressing the cohomology groups of the total space in terms of those of the base plus correction. The sequence may be obtained as a corollary of the Serre spectral sequence for the given fibration. It induces, and is induced by, the Thom isomorphism.


Let RR be a commutative ring and let

S n E π B \array{ S^n &\longrightarrow& E \\ && \downarrow^{\mathrlap{\pi}} \\ && B }

be a Serre fibration over a simply connected CW-complex with typical fiber (exmpl.) the n-sphere.

Then there exists an element cH n+1(E;R)c \in H^{n+1}(E; R) (in the ordinary cohomology of the total space with coefficients in RR, called the Euler class of π\pi) such that the cup product operation c()c \cup (-) sits in a long exact sequence of cohomology groups of the form

H k(B;R)π *H k(E;R)H kn(B;R)c()H k+1(B;R). \cdots \to H^k(B; R) \stackrel{\pi^\ast}{\longrightarrow} H^k(E; R) \stackrel{}{\longrightarrow} H^{k-n}(B;R) \stackrel{c \cup (-)}{\longrightarrow} H^{k+1}(B; R) \to \cdots \,.

(e.g. Switzer 75, section 15.30, Kochman 96, corollary 2.2.6)


Under the given assumptions there is the corresponding Serre spectral sequence

E 2 s,t=H s(B;H t(S n;R))H s+t(E;R). E_2^{s,t} \;=\; H^s(B; H^t(S^n;R)) \;\Rightarrow\; H^{s+t}(E; R) \,.

Since the ordinary cohomology of the n-sphere fiber is concentrated in just two degees

H t(S n;R)={R fort=0andt=n 0 otherwise H^t(S^n; R) = \left\{ \array{ R & for \; t= 0 \; and \; t = n \\ 0 & otherwise } \right.

the only possibly non-vanishing terms on the E 2E_2 page of this spectral sequence, and hence on all the further pages, are in bidegrees (,0)(\bullet,0) and (,n)(\bullet,n):

E 2 ,0H (B;R),andE 2 ,nH (B;R). E^{\bullet,0}_2 \simeq H^\bullet(B; R) \,, \;\;\;\; and \;\;\; E^{\bullet,n}_2 \simeq H^\bullet(B; R) \,.

As a consequence, since the differentials d rd_r on the rrth page of the Serre spectral sequence have bidegree (r+1,r)(r+1,-r), the only possibly non-vanishing differentials are those on the (n+1)(n+1)-page of the form

E n+1 ,n H (B;R) d n+1 E n+1 +n+1,0 H +n+1(B;R). \array{ E_{n+1}^{\bullet,n} & \simeq & H^\bullet(B;R) \\ {}^{\mathllap{d_{n+1}}}\downarrow \\ E_{n+1}^{\bullet+n+1,0} & \simeq & H^{\bullet+n+1}(B;R) } \,.

Now since the coefficients RR is a ring, the Serre spectral sequence is multiplicative under cup product and the differential is a derivation (of total degree 1) with respect to this product. (See at multiplicative spectral sequence – Examples – AHSS for multiplicative cohomology.)

To make use of this, write

ι1H 0(B;R)E n+1 0,n \iota \coloneqq 1 \in H^0(B;R) \stackrel{\simeq}{\longrightarrow} E_{n+1}^{0,n}

for the unit in the cohomology ring H (B;R)H^\bullet(B;R), but regarded as an element in bidegree (0,n)(0,n) on the (n+1)(n+1)-page of the spectral sequence. (In particular ι\iota does not denote the unit in bidegree (0,0)(0,0), and hence d n+1(ι)d_{n+1}(\iota) need not vanish; while by the derivation property, it does vanish on the actual unit 1H 0(B;R)E n+1 0,01 \in H^0(B;R) \simeq E_{n+1}^{0,0}.)


cd n+1(ι)E n+1 n+1,0H n+1(B;R) c \coloneqq d_{n+1}(\iota) \;\; \in E_{n+1}^{n+1,0} \stackrel{\simeq}{\longrightarrow} H^{n+1}(B; R)

for the image of this element under the differential. We will show that this is the Euler class in question.

To that end, notice that every element in E n+1 ,nE_{n+1}^{\bullet,n} is of the form ιb\iota \cdot b for bE n+1 ,0H (B;R)b\in E_{n+1}^{\bullet,0} \simeq H^\bullet(B;R).

(Because the multiplicative structure gives a group homomorphism ι():H (B;R)E n+1 0,0E n+1 0,nH (B;R)\iota \cdot(-) \colon H^\bullet(B;R) \simeq E_{n+1}^{0,0} \to E^{0,n}_{n+1} \simeq H^\bullet(B;R), which is an isomorphism because the product in the spectral sequence does come from the cup product in the cohomology ring, see for instance (Kochman 96, first equation in the proof of prop. 4.2.9), and since hence ι\iota does act like the unit that it is in H (B;R)H^\bullet(B;R)).

Now since d n+1d_{n+1} is a graded derivation and vanishes on E n+1 ,0E_{n+1}^{\bullet,0} (by the above degree reasoning), it follows that its action on any element is uniquely fixed to be given by the product with cc:

d n+1(ιb) =d n+1(ι)b+(1) nιd n+1(b)=0 =cb. \begin{aligned} d_{n+1}(\iota \cdot b) & = d_{n+1}(\iota) \cdot b + (-1)^{n}\, \iota \cdot \underset{= 0}{\underbrace{d_{n+1}(b)}} \\ & = c \cdot b \end{aligned} \,.

This shows that d n+1d_{n+1} is identified with the cup product operation in question:

E n+1 s,n H s(B;R) d n+1 c() E n+1 s+n+1,0 H s+n+1(B;R). \array{ E_{n+1}^{s,n} & \simeq & H^s(B;R) \\ {}^{\mathllap{d_{n+1}}}\downarrow && \downarrow^{\mathrlap{c \cup (-)}} \\ E_{n+1}^{s+n+1, 0} & \simeq & H^{s+n+1}(B;R) } \,.

In summary, the non-vanishing entries of the E E_\infty-page of the spectral sequence sit in exact sequences like so

0 E s,n ker(d n+1) E n+1 s,n H s(B;R) d n+1 c() E n+1 s+n+1,0 H s+n+1(B;R) coker(d n+1) E s+n+1,0 0. \array{ 0 \\ \downarrow \\ E_\infty^{s,n} \\ {}^{\mathllap{ker(d_{n+1})}}\downarrow \\ E_{n+1}^{s,n} & \simeq & H^s(B;R) \\ {}^{\mathllap{d_{n+1}}}\downarrow && \downarrow^{\mathrlap{c \cup (-)}} \\ E_{n+1}^{s+n+1, 0} & \simeq & H^{s+n+1}(B;R) \\ {}^{\mathllap{coker(d_{n+1})}}\downarrow \\ E_\infty^{s+n+1,0} \\ \downarrow \\ 0 } \,.

Finally observe (lemma 6) that due to the sparseness of the E E_\infty-page, there are also short exact sequences of the form

0E s,0H s(E;R)E sn,n0. 0 \to E_\infty^{s,0} \longrightarrow H^s(E; R) \longrightarrow E_\infty^{s-n,n} \to 0 \,.

Concatenating these with the above exact sequences yields the desired long exact sequence.


Consider a cohomology spectral sequence converging to some filtered graded abelian group F C F^\bullet C^\bullet such that

  1. F 0C =C F^0 C^\bullet = C^\bullet;

  2. F sC <s=0F^{s} C^{\lt s} = 0;

  3. E s,t=0E_\infty^{s,t} = 0 unless t=0t = 0 or t=nt = n,

for some nn \in \mathbb{N}, n1n \geq 1. Then there are short exact sequences of the form

0E s,0C sE sn,n0. 0 \to E_\infty^{s,0} \overset{}{\longrightarrow} C^s \longrightarrow E_\infty^{s-n,n} \to 0 \,.

(e.g. Switzer 75, p. 356)


By definition of convergence of a spectral sequence, the E s,tE_{\infty}^{s,t} sit in short exact sequences of the form

0F s+1C s+tiF sC s+tE s,t0. 0 \to F^{s+1}C^{s+t} \overset{i}{\longrightarrow} F^s C^{s+t} \longrightarrow E_\infty^{s,t} \to 0 \,.

So when E s,t=0E_\infty^{s,t} = 0 then the morphism ii above is an isomorphism.

We may use this to either shift away the filtering degree

  • if tnt \geq n then F sC s+t=F (s1)+1C (s1)+(t+1)i s1F 0C (s1)+(t+1)=F 0C s+tC s+tF^s C^{s+t} = F^{(s-1)+1}C^{(s-1)+(t+1)} \underoverset{\simeq}{i^{s-1}}{\longrightarrow} F^0 C^{(s-1)+(t+1)} = F^0 C^{s+t} \simeq C^{s+t};

or to shift away the offset of the filtering to the total degree:

  • if 0t1n10 \leq t-1 \leq n-1 then F s+1C s+t=F s+1C (s+1)+(t1)i (t1)F s+tC (s+1)+(t1)=F s+tC s+tF^{s+1}C^{s+t} = F^{s+1}C^{(s+1)+(t-1)} \underoverset{\simeq}{i^{-(t-1)}}{\longrightarrow} F^{s+t}C^{(s+1)+(t-1)} = F^{s+t}C^{s+t}

Moreover, by the assumption that if t<0t \lt 0 then F sC s+t=0F^{s}C^{s+t} = 0, we also get

F sC sE s,0. F^{s}C^{s} \simeq E_\infty^{s,0} \,.

In summary this yields the vertical isomorphisms

0 F s+1C s+n F sC s+n E s,n 0 i (n1) i s1 = 0 F s+nC s+nE s+n,0 C s+n E s,n 0 \array{ 0 &\to& F^{s+1}C^{s+n} &\longrightarrow& F^{s}C^{s+n} &\longrightarrow& E_\infty^{s,n} &\to& 0 \\ && {}^{\mathllap{i^{-(n-1)}}}\downarrow^{\mathrlap{\simeq}} && {}^{\mathllap{i^{s-1}}}\downarrow^{\mathrlap{\simeq}} && \downarrow^{\mathrlap{=}} \\ 0 &\to& F^{s+n}C^{s+n} \simeq E_\infty^{s+n,0} &\longrightarrow& C^{s+n} &\longrightarrow& E_\infty^{s,n} &\to& 0 }

and hence with the top sequence here being exact, so is the bottom sequence.

Thom isomorphism

Let VBV \to B be a topological vector bundle of rank n>0n \gt 0 over a simply connected CW-complex BB. Let RR be a commutative ring.

There exists an element cH n(Th(V);R)c \in H^n(Th(V);R) (in the ordinary cohomology, with coefficients in RR, of the Thom space of VV, called a Thom class) such that forming the cup product with cc induces an isomorphism

H (B;R)c()H˜ +n(Th(V);R) H^\bullet(B;R) \overset{c \cup (-)}{\longrightarrow} \tilde H^{\bullet + n}(Th(V);R)

of degree nn from the unreduced cohomology group of BB to the reduced cohomology of the Thom space of VV.


Choose an orthogonal structure on VV. Consider the fiberwise cofiber

ED(V)/ BS(V) E \coloneqq D(V)/_B S(V)

of the inclusion of the unit sphere bundle into the unit disk bundle of VV (def. 33).

S n1 D n S n S(V) D(V) E p B = B = B \array{ S^{n-1} &\hookrightarrow& D^n &\longrightarrow& S^n \\ \downarrow && \downarrow && \downarrow \\ S(V) &\hookrightarrow& D(V) &\longrightarrow& E \\ \downarrow && \downarrow && \downarrow^{\mathrlap{p}} \\ B &=& B &=& B }

Observe that this has the following properties

  1. EpBE \overset{p}{\to} B is an n-sphere fiber bundle, hence in particular a Serre fibration;

  2. the Thom space Th(V)E/BTh(V)\simeq E/B is the quotient of EE by the base space, because of the pasting law applied to the following pasting diagram of pushout squares

    S(V) D(V) (po) B D(V)/ BS(V) (po) * Th(V) \array{ S(V) &\longrightarrow& D(V) \\ \downarrow &(po)& \downarrow \\ B &\longrightarrow& D(V)/_B S(V) \\ \downarrow &(po)& \downarrow \\ \ast &\longrightarrow& Th(V) }
  3. hence the reduced cohomology of the Thom space is (def.) the relative cohomology of EE relative BB

    H˜ (Th(V);R)H (E,B;R). \tilde H^\bullet(Th(V);R) \simeq H^\bullet(E,B;R) \,.
  4. EpBE \overset{p}{\to} B has a global section BsEB \overset{s}{\to} E (given over any point bBb \in B by the class of any point in the fiber of S(V)BS(V) \to B over bb; or abstractly: induced via the above pushout by the commutation of the projections from D(V)D(V) and from S(V)S(V), respectively).

In the following we write H ()H (;R)H^\bullet(-)\coloneqq H^\bullet(-;R), for short.

By the first point, there is the Thom-Gysin sequence (prop. 53), an exact sequence running vertically in the following diagram

H (B) p * H˜ (Th(V)) H (E) s * H (B) H n(B). \array{ && H^\bullet(B) \\ && {}^{\mathllap{p^\ast}}\downarrow & \searrow^{\mathrlap{\simeq}} \\ \tilde H^\bullet(Th(V)) &\longrightarrow& H^\bullet(E) &\underset{s^\ast}{\longrightarrow}& H^\bullet(B) \\ && \downarrow \\ && H^{\bullet-n}(B) } \,.

By the second point above this is split, as shown by the diagonal isomorphism in the top right. By the third point above there is the horizontal exact sequence, as shown, which is the exact sequence in relative cohomology H (E,B)H (E)H (B)\cdots \to H^\bullet(E,B) \to H^\bullet(E) \to H^\bullet(B) \to \cdots induced from the section BEB \hookrightarrow E.

Hence using the splitting to decompose the term in the middle as a direct sum, and then using horizontal and vertical exactness at that term yields

H (B) (0,id) H˜ (Th(V)) (id,0) H˜ (Th(V))H (B) (0,id) H (B) (id,0) H n(B) \array{ && H^\bullet(B) \\ && {}^{\mathllap{(0,id)}}\downarrow & \searrow^{\mathrlap{\simeq}} \\ \tilde H^\bullet(Th(V)) &\overset{(id,0)}{\hookrightarrow}& \tilde H^\bullet(Th(V)) \oplus H^\bullet(B) &\underset{(0,id)}{\longrightarrow}& H^\bullet(B) \\ && \downarrow^{\mathrlap{(id,0)}} \\ && H^{\bullet-n}(B) }

and hence an isomorphism

H˜ (Th(V))H n(B). \tilde H^\bullet(Th(V)) \overset{\simeq}{\longrightarrow} H^{\bullet-n}(B) \,.

To see that this is the inverse of a morphism of the form c()c \cup (-), inspect the proof of the Gysin sequence. This shows that H n(B)H^{\bullet-n}(B) here is identified with elements that on the second page of the corresponding Serre spectral sequence are cup products

ιb \iota \cup b

with ι\iota fiberwise the canonical class 1H n(S n)1 \in H^n(S^n) and with bH (B)b \in H^\bullet(B) any element. Since H (;R)H^\bullet(-;R) is a multiplicative cohomology theory (because the coefficients form a ring RR), cup producs are preserved as one passes to the E E_\infty-page of the spectral sequence, and the morphism H (E)B (B)H^\bullet(E) \to B^\bullet(B) above, hence also the isomorphism H˜ (Th(V))H (B)\tilde H^\bullet(Th(V)) \to H^\bullet(B), factors through the E E_\infty-page (see towards the end of the proof of the Gysin sequence). Hence the image of ι\iota on the E E_\infty-page is the Thom class in question.

Orientation in generalized cohomology

Idea. From the way the Thom isomorphism via a Thom class works in ordinary cohomology (as above), one sees what the general concept of orientation in generalized cohomology and of fiber integration in generalized cohomology is to be.

Specifically we are interested in complex oriented cohomology theories EE, characterized by an orientation class on infinity complex projective space P \mathbb{C}P^\infty (def. 43), the classifying space for complex line bundles, which restricts to a generator on S 2P S^2 \hookrightarrow \mathbb{C}P^\infty.

(Another important application is given by taking E=E = KU to be topological K-theory. Then orientation is spin^c structure and fiber integration with coefficients in EE is fiber integration in K-theory. This is classical index theory.)

Literature. (Kochman 96, section 4.3, Adams 74, part III, section 10, Lurie 10, lecture 5)

  • Riccardo Pedrotti, Complex oriented cohomology – Orientation in generalized cohomology, 2016 (pdf)


Universal EE-orientation

Let EE be a multiplicative cohomology theory (def. 14) and let VXV \to X be a topological vector bundle of rank nn. Then an EE-orientation or EE-Thom class on VV is an element of degree nn

uE˜ n(Th(V)) u \in \tilde E^n(Th(V))

in the reduced EE-cohomology ring of the Thom space (def. 33) of VV