Chapter 0_III

§11. Complements on homological algebra

Translator's note. In the OCR'd French source, the content of §11 is housed inside the file named for §10 (03-c0-s10-complements-modules-plats.md, lines 266–1485). It is given its own translated file here.

11.1. Recall on spectral sequences

11.1.1.

In what follows we shall use a notion of spectral sequence more general than the one defined in (T, 2.4); keeping the notation of (T, 2.4), we shall call a spectral sequence in an abelian category a system consisting of the following data:

a) A family of objects of defined for , , and .

b) A family of morphisms such that .

We set , , so that

c) A family of isomorphisms .

One then defines, for , by induction on , the subobjects and as the inverse images, under the canonical morphism , of the subobjects of this quotient identified by with the subobjects and respectively. It is clear that one then has, up to isomorphism,

for , and, if we further set and , one has the inclusion relations

  0 = B_1(E_2^{p,q}) ⊂ B_2(E_2^{p,q}) ⊂ B_3(E_2^{p,q}) ⊂ …
       … ⊂ Z_3(E_2^{p,q}) ⊂ Z_2(E_2^{p,q}) ⊂ Z_1(E_2^{p,q}) = E_2^{p,q}.       (11.1.1.2)

The remaining data of are then:

d) Two subobjects and of such that one has and, for every ,

  B_k(E_2^{p,q}) ⊂ B_∞(E_2^{p,q})    and    Z_∞(E_2^{p,q}) ⊂ Z_k(E_2^{p,q}).

One sets

e) A family of objects of , each equipped with a decreasing filtration . As usual we denote by the graded object associated to the filtered object , the direct sum of the .

f) For every pair , an isomorphism .

The family , without the filtrations, is called the abutment of the spectral sequence .

Suppose either that the category admits infinite direct sums, or that for every and every , the pairs such that and are finite in number (it suffices that this hold for ). Then the are defined, and denoting by the morphism whose restriction to is for each pair with , one has ; in other words, is a complex in , with derivation operator of degree +1, and it follows from c) that is isomorphic to for every .

11.1.2.

A morphism from a spectral sequence to a spectral sequence consists of systems of morphisms , , the being compatible with the filtrations of and , the diagrams

being commutative; moreover, on passing to quotients, induces a morphism , and one must have ; finally, one must have and ; on passing to quotients, then gives a morphism , and the diagram

must be commutative.

The preceding definitions show, by induction on , that if the are isomorphisms, then so are the for ; if one further knows that and and that the are isomorphisms, then one can conclude that is an isomorphism.

11.1.3.

Recall that if is a (decreasing) filtration of an object , the filtration is said to be separated if , discrete if there exists such that , exhaustive (or co-separated) if , co-discrete if there exists such that .

We shall say that a spectral sequence is weakly convergent if one has , (in other words, the objects and are determined by data a) through c) of the spectral sequence ). We shall say that the spectral sequence is regular if it is weakly convergent and moreover:

1° For every pair , the decreasing sequence is stationary; the hypothesis that is weakly convergent then implies for sufficiently large (depending on and ).

2° For every , the filtration of is discrete and exhaustive.

One says that the spectral sequence is co-regular if it is weakly convergent and moreover:

3° For every pair , the increasing sequence is stationary, which entails , and consequently .

4° For every , the filtration of is co-discrete.

Finally, one says that is biregular if it is both regular and co-regular; in other words, if the following conditions hold:

a) For every pair , the sequences and are stationary, and one has and for sufficiently large (which entails ).

b) For every , the filtration is discrete and co-discrete (which one also expresses by saying that it is finite).

The spectral sequences defined in (T, 2.4) are thus the biregular spectral sequences.

11.1.4.

Suppose that in the category filtered inductive limits exist and that the functor is exact (which is equivalent to saying that axiom AB 5) of (T, 1.5) is verified (cf. T, 1.8)). The condition that the filtration of an object be exhaustive then also reads . If a spectral sequence is weakly convergent, one has ; if moreover is a morphism of into a weakly convergent spectral sequence of , one has by exactness of . Moreover:

Proposition (11.1.5).

Let be an abelian category in which filtered inductive limits are exact, let , be two regular spectral sequences in , and let be a morphism of spectral sequences. If the are isomorphisms, then so is .

Proof. We already know (11.1.2) that the are isomorphisms and that

the hypothesis that and are regular also implies , and since is an isomorphism, so is ; one therefore concludes that is also an isomorphism. But since the filtrations of the and the are discrete and exhaustive, this entails that the are also isomorphisms (Bourbaki, Alg. comm., chap. III, § 2, n° 8, th. 1).

11.1.6.

It follows from (11.1.1.2) and from definition (11.1.1.3) that if, in a spectral sequence , one has , then one has for and . One says that a spectral sequence is degenerate if there exists an integer and, for every integer , an integer such that for every . From the preceding remark one first deduces that one also has for (including ) and . Moreover, the definition of shows that one has ; if is weakly convergent, one therefore also has ; in other words, for every , for and . If moreover the filtration of is discrete and exhaustive, the sequence is regular and one has up to isomorphism.

11.1.7.

Suppose that in the category filtered inductive limits exist and are exact, and let be an inductive system (over a filtered index set) of spectral sequences in . Then the inductive limit of this inductive system exists in the additive category of spectral sequences of objects of : to see this it suffices to define , , , , , , and as the respective inductive limits of , , , , , , and ; the verification of the conditions of (11.1.1) follows from the exactness of the functor in .

Remark (11.1.8).

Suppose that the category is the category of -modules over a Noetherian ring (resp. over a ring ). Then the definitions of (11.1.1) show that if, for a given , the are -modules of finite type (resp. of finite length), then so are all the modules for , as well as the . If moreover the filtration of the abutment is discrete and co-discrete for every , one concludes that each of the is also an -module of finite type (resp. of finite length).

11.1.9.

We shall have to consider conditions ensuring that a spectral sequence is biregular "uniformly" in . We shall then use the following lemma:

Lemma (11.1.10).

Let be a family of objects of linked by data a), b), c) of (11.1.1). For a fixed integer , the following properties are equivalent:

a) There exists an integer such that for , or , the morphisms are all zero.

b) There exists an integer such that for or , one has for .

c) There exists an integer such that for or , one has for .

d) There exists an integer such that for , one has and for .

Proof. Indeed, by conditions a), b), c) of (11.1.1), to say that is equivalent to saying that , and to say that is also equivalent to saying that ; the lemma follows at once from this remark.

11.2. The spectral sequence of a filtered complex

11.2.1.

Given an abelian category , we shall use notation such as for complexes of objects of in which the derivation operator is of degree +1, and notation such as for complexes of objects of in which the derivation operator is of degree . To any complex whose derivation operator is of degree +1, one can associate a complex by setting , the derivation operator being the operator ; and conversely, which will allow us, according to circumstances, to consider one type of complex or the other and to translate every result for one type into a result for the other. We shall denote similarly by notation such as (resp. ) bicomplexes of objects of in which the two derivation operators are of degree +1 (resp. ); one again passes from one type to the other by changing the signs of the indices, and one has analogous notation and remarks for arbitrary multicomplexes. The notation or will also be used for graded objects of , of type , which are not necessarily complexes (or which one may consider as such with zero derivation operators); for instance, we shall write for the cohomology of a complex whose derivation operator is of degree +1, and for the homology of a complex whose derivation operator is of degree ; when one passes from to by the operation described above, one has .

Recall in this connection that for a complex (resp. ), we shall generally write ("object of cocycles") and ("object of coboundaries") (resp. ("object of cycles") and ("object of boundaries")), so that (resp. ).

If (resp. ) is a complex in , and a functor from to an abelian category , we shall denote by (resp. ) the complex (resp. ) in .

We do not return to the definition of -functors (T, 2.1), except to point out that we shall also say -functor in place of -functor when the morphism lowers the degree by one unit; the context should make this clear when there could be doubt.

Finally, we shall say that a graded object of is bounded below (resp. bounded above) if there exists such that for (resp. ).

11.2.2.

Let be a complex in whose derivation operator is of degree +1, and suppose it equipped with a filtration formed of graded subobjects of ,

in other words ; moreover, we assume that for every . We now briefly recall how one functorially defines a spectral sequence from (M, XV, 4 and G, I, 4.3). For , the canonical morphism defines a morphism in cohomology

One denotes by the image of this morphism. Similarly, from the exact sequence

one deduces, by the long exact sequence of cohomology, a morphism

and one denotes by the image of this morphism; one shows that and one sets ; we shall not specify the definitions of the or the .

We note here that all the and , for , fixed, are subobjects of the same object , which one denotes ; one sets , so that the preceding definitions for and also apply for ; one further sets . One also defines the and so that the conditions of (11.1.1) are satisfied for . One defines on the other hand the subobjects , image of the morphism

and , image of the morphism

obtained as above from a long exact sequence of cohomology. One takes for and the canonical images in of and .

Finally, one denotes by the image in of the morphism coming from the canonical injection ; by the long exact sequence of cohomology, it is also the kernel of the morphism . One thus defines a filtration on ; we shall not give the definition of the isomorphisms here either.

11.2.3.

The functorial character of is to be understood as follows: given two filtered complexes , of and a morphism of complexes compatible with the filtrations, one deduces from it in an obvious way the morphisms (for ) and , and one shows that these morphisms are compatible with the , and in the sense of (11.1.2), so that they indeed define a morphism of spectral sequences. Moreover, one shows that if and are morphisms of the preceding type, homotopic of order , then for and for every (M, XV, 3.1).

11.2.4.

Suppose that in filtered inductive limits are exact. Then, if the filtration of is exhaustive, so is the filtration for every , since by hypothesis and the hypothesis on implies that cohomology commutes with inductive limits. Moreover, for the same reason, one has . One says that the filtration of is regular if for every there exists an integer such that for . This holds in particular when the filtration of is discrete. When the filtration of is regular and exhaustive, and filtered inductive limits in are exact, one shows (M, XV, 4) that the spectral sequence is regular.

11.3. The spectral sequences of a bicomplex

11.3.1.

As regards the conventions on bicomplexes, we follow those of (T, 2.4) rather than those of (M), the two derivations , (of degree +1) of such a bicomplex being therefore assumed to commute. Suppose that one of the two following conditions is verified: 1° infinite direct sums exist in ; 2° for every , there are only finitely many pairs such that and . Then the bicomplex defines a (simple) complex with , the derivation operator (of degree +1) of this complex being given by for . Whenever in what follows we speak of the (simple) complex defined by a bicomplex , it will always be understood that one of the preceding conditions is satisfied. One adopts analogous conventions for multicomplexes.

We denote by (resp. ) the simple complex (resp. ), and by , , (resp. , , ) its th (resp. th) objects of cocycles, coboundaries, and cohomology respectively; the derivation is a morphism of complexes, which therefore gives an operator on the cocycles, coboundaries, and cohomology,

and it is clear that for these operators, , , and are complexes; we shall denote the complex by , and its th objects of cocycles, coboundaries, and cohomology by , , and . One defines similarly the complexes and their objects of cohomology . Recall on the other hand that denotes the th object of cohomology of the (simple) complex defined by .

11.3.2.

On the complex defined by a bicomplex , one may consider two canonical filtrations and given by

  F_I^p(K^{•,•}) = (⊕_{i+j=n, i ≥ p} K^{i,j})_{n ∈ ℤ},   F_{II}^p(K^{•,•}) = (⊕_{i+j=n, j ≥ p} K^{i,j})_{n ∈ ℤ}.   (11.3.2.1)

which are by definition graded subobjects of the (simple) complex defined by , and thus make this complex into a filtered complex; moreover, it is clear that these filtrations are exhaustive and separated.

To each of these filtrations there corresponds a spectral sequence (11.2.2); we shall denote by and the spectral sequences corresponding to and respectively, called the spectral sequences of the bicomplex , both having as abutment the cohomology . One shows moreover (M, XV, 6) that

  'E_2^{p,q}(K^{•,•}) = H_I^p(H_{II}^q(K^{•,•})),    ″E_2^{p,q}(K^{•,•}) = H_{II}^p(H_I^q(K^{•,•})).   (11.3.2.2)

Any morphism of bicomplexes is ipso facto compatible with the filtrations of the same type on and , and thus defines a morphism for each of the two spectral sequences; moreover, two homotopic morphisms define a homotopy of order of the corresponding (simple) filtered complexes, hence the same morphism for each of the two spectral sequences (M, XV, 6.1).

Proposition (11.3.3).

Let be a bicomplex in an abelian category .

(i) If there exist and such that for or (resp. or ), the two spectral sequences and are biregular.

(ii) If there exist and such that for or (resp. if there exist and such that for or ), the two spectral sequences and are biregular.

Suppose moreover that in filtered inductive limits exist and are exact. Then:

(iii) If there exists such that for (resp. if there exists such that for ), the sequence is regular.

(iv) If there exists such that for (resp. if there exists such that for ), the sequence is regular.

Proof. The proposition follows at once from the definitions (11.1.3) and from (11.2.4), together with the following observations concerning the filtration F_I (and the analogous observations one deduces for by exchanging the roles of the two indices in ):

1° If there exists such that for , the filtration is discrete.

2° If there exists such that for , the filtration is co-discrete. One deduces at once that the same holds for the corresponding filtration for every ; moreover, the definition of corresponding to the filtration (11.2.2) shows that for every pair , the sequence is stationary.

3° If there exists such that for , one has

whenever , hence for ; on the other hand, for .

4° If there exists such that for , one has

  F_I^{p-r+1}(K^{•,•}) ∩ (⊕_{i+j=n} K^{i,j}) = ⊕_{i+j=n} K^{i,j}

whenever , hence for ; on the other hand, for .

11.3.4.

Suppose that the bicomplex is such that for or . It is known that one can then define for every a canonical "edge homomorphism"

(M, XV, 6). We briefly recall that this is due, on the one hand, to the fact that one has in the spectral sequence for , and, on the other hand, to the fact that , so that the isomorphism gives a homomorphism ; the equality of all the then makes it possible to define canonical homomorphisms for , in particular a homomorphism , whence by composition the edge homomorphism ; moreover, one verifies at once that, to the class mod. of an element such that , the edge homomorphism so defined associates in the class of mod. , and then, to this last, the cohomology class of in . One thus sees finally that the edge homomorphism (11.3.4.1) comes, by passage to cohomology, from the canonical injection (where is considered as a simple complex). One similarly interprets the edge homomorphism

as coming from the canonical injection .

11.3.5.

Let now be a bicomplex in whose two derivation operators are of degree . We shall then write (resp. ) for the simple complex (resp. ), (resp. ) for its th object of homology, (resp. ) for the complex (resp. ), (resp. ) for its th object of homology; analogous notation for the objects of cycles and boundaries; finally, will denote (when it exists) the th object of homology of the simple complex (with derivation operator of degree ) defined by .

Let with be the bicomplex with derivation operators of degree +1 associated to ; by definition, the spectral sequences of are those of , which one writes and , where one changes the notation, however, by setting

  'E_{p,q}^r(K_{•,•}) = 'E_r^{-p,-q}(K^{•,•}),    ″E_{p,q}^r(K_{•,•}) = ″E_r^{-p,-q}(K^{•,•})

for . With this notation, one has

  'E_{p,q}^2(K_{•,•}) = H_p^I(H_q^{II}(K_{•,•})),    ″E_{p,q}^2(K_{•,•}) = H_p^{II}(H_q^I(K_{•,•})).

To avoid sign errors, it will generally be preferable, for the relations between these spectral sequences and their abutment, to return to the complex . Let us note nonetheless the criteria corresponding to (11.3.3):

11.3.6.

The spectral sequences and are biregular in the following cases: a) There exist and such that for or for (resp. for

or for ); b) There exist and such that for and ; c) There exist and such that for and .

The sequence is regular if there exists such that for , or if there exists such that for .

The sequence is regular if there exists such that for , or if there exists such that for .

11.4. Hypercohomology of a functor with respect to a complex

11.4.1.

Let be an abelian category; recall that one calls a right resolution (or cohomological resolution) of an object of a complex of objects of , whose derivation operator is of degree +1,

equipped with a morphism called the augmentation of the resolution (and which one can consider as a morphism of complexes

  0 → A → 0 → 0 → …
        ↓   ↓   ↓
  0 → L^0 → L^1 → L^2 → … )

such that the sequence

  0 → A →^ε L^0 → L^1 → …

is exact; similarly, a left resolution (or homological resolution) of is a complex of objects of whose derivation operator is of degree , equipped with an augmentation , such that the sequence

  0 ← A ←^ε L_0 ← L_1 ← …

is exact.

When a right resolution of an object is such that for , one says that this resolution is of length . One defines similarly a left resolution of length . A resolution that is of length for some integer is said to be finite.

A resolution of is called projective (resp. injective) if the objects of other than that compose it are projective (resp. injective). When is the category of modules (left modules, say) over a ring, one will say similarly that a resolution of is flat (resp. free) when the modules other than that compose it are flat (resp. free).

11.4.2.

Let be a complex of objects of whose derivation operator is of degree +1.

We call a right Cartan–Eilenberg resolution of the pair consisting of a bicomplex with derivation operators of degree +1, with for , and a morphism of simple complexes , such that the following conditions are satisfied:

(i) For each index , the sequences

  0 → K^i →^ε L^{i,0} → L^{i,1} → …
  0 → B^i(K^•) →^ε B_I^i(L^{•,0}) → B_I^i(L^{•,1}) → …
  0 → Z^i(K^•) →^ε Z_I^i(L^{•,0}) → Z_I^i(L^{•,1}) → …
  0 → H^i(K^•) →^ε H_I^i(L^{•,0}) → H_I^i(L^{•,1}) → …

are exact; in other words, , , and are respectively resolutions of , , and .

(ii) For each , the simple complex is split; in other words, the exact sequences

are split.

One proves (M, XVII, 1.2) that if every object of is a subobject of an injective object, every complex of admits an injective Cartan–Eilenberg resolution, that is, one formed of injective objects (condition (ii) above then entails that the , and are also injective objects). Moreover, for every morphism of complexes of , every Cartan–Eilenberg resolution of and every injective Cartan–Eilenberg resolution of , there exists a morphism of bicomplexes compatible with and the augmentations; and if and are two homotopic morphisms from to , the corresponding morphisms from to are homotopic (loc. cit.).

When is bounded below (resp. bounded above), one may take such that for (resp. ) if for (resp. ) (M, XVII, 1.3).

Suppose on the other hand that there exists an integer such that every object of admits an injective resolution of length ; then one may suppose that for (M, XVII, 1.4).

11.4.3.

Let now be a covariant additive functor from to an abelian category . Given a complex of and an injective Cartan–Eilenberg resolution of , suppose that the (simple) complex defined by the bicomplex exists (cf. 11.3.1); then the two spectral sequences and of this bicomplex are called the spectral sequences of hypercohomology of with respect to the complex ; by virtue of (11.4.2) and (11.3.2), they in fact depend only on and not on the chosen injective Cartan–Eilenberg resolution ; moreover, they depend functorially on . They have a common abutment , called the hypercohomology of with respect to , and denoted . One shows that the terms E_2 of the two preceding spectral sequences are given by

denoting as usual the th derived functor of for ; denotes the complex . Unless expressly stated otherwise, we shall henceforth assume that every object of is a subobject of an injective object of , so that injective Cartan–Eilenberg resolutions exist for every complex of . Since for , the criteria of (11.3.3) show that the two hypercohomology spectral sequences of with respect to exist and are biregular in each of the two following cases: 1° is bounded below; 2° every object of admits an injective resolution of length at most equal to an integer (independent of the object considered). Indeed, in the first case, one may suppose (11.4.2) that there exists such that for , and in the second that there exists such that for ; in each of the two cases, it is moreover clear that for given , there are only finitely many pairs such that and , which establishes our assertions.

When one supposes that in filtered inductive limits exist and are exact (which implies in particular the existence in of infinite direct sums), then the complex defined by the bicomplex exists, and criterion (11.3.3) shows that the spectral sequence is always regular.

11.4.4.

When is a complex all of whose terms are zero except for a single , is isomorphic to , as follows at once from the definitions on taking a Cartan–Eilenberg resolution such that for .

If and are two complexes of , , two homotopic morphisms from to , then the morphisms deduced from and are identical, and likewise for the morphisms of the cohomology spectral sequences.

Proposition (11.4.5).

Suppose that in filtered inductive limits exist and are exact. If for every and every , one has functorial isomorphisms

for .

Proof. The only nonzero terms E_2 of the first spectral sequence (11.4.3.1) are then ; in other words, this sequence is degenerate; since it is regular (11.4.4), the conclusion follows from (11.1.6).

11.4.6.

Consider now, for example, a covariant bifunctor from to , where , , are three abelian categories; one assumes, for simplicity, that is additive in each of its arguments, and moreover that every object of and every object of is a subobject of an injective object, and that filtered inductive limits exist in and are exact. One then defines the hypercohomology of with respect to two complexes , of and respectively, with derivation operators of degree +1, by taking for (resp. ) an injective Cartan–Eilenberg resolution (resp. ); then is a quadricomplex of , which one regards as a bicomplex of by taking as degrees of the integers and . The hypercohomology of with respect to and is by definition the cohomology of this bicomplex (in other words, that of the associated simple complex),

denoted ; it is the abutment of two spectral sequences whose terms E_2 are given by

  'E_2^{p,q} = H^p(R^qT(K^•, K'^•))                                            (11.4.6.1)
  ″E_2^{p,q} = ⊕_{q'+q″=q} R^pT(H^{q'}(K^•), H^{q″}(K'^•))   (cf. M, XVII, 2). (11.4.6.2)

Here is the bicomplex and the second member of (11.4.6.1) is its cohomology when one regards it as a simple complex.

Moreover, the first spectral sequence is always regular, and the two spectral sequences are biregular when there exists such that every object of and every object of admits an injective resolution of length , or when and are bounded below; in the latter case, one may moreover omit the hypothesis that inductive limits exist in and .

If , are two other complexes of and respectively, , two homotopic morphisms from to , , two homotopic morphisms from to , then the morphisms deduced from and on the one hand, from and on the other, are identical, and likewise for the morphisms of the hypercohomology spectral sequences.

One generalizes easily to any covariant additive multifunctor.

Proposition (11.4.7).

Suppose that for every injective object of (resp. of ), (resp. ) is an exact functor. Then, with the notation of (11.4.6), one has canonical isomorphisms

  R^•T(K^•, K'^•) ⥲ H^•(T(L^{•,•}, K'^•)) ⥲ H^•(T(K^•, L'^{•,•}))               (11.4.7.1)

where the last two terms are the cohomology of the simple complexes defined by the tricomplexes and respectively.

Proof. Let us define, for example, the first of these isomorphisms. The quadricomplex can be considered as a bicomplex, by taking as degrees of the numbers and . As for each , is a resolution of , one has, for this bicomplex, by virtue of the hypothesis on , for and ; the first spectral sequence of this bicomplex is therefore degenerate; since for , this sequence is moreover regular (11.3.3), and the conclusion follows from (11.1.6).

One has analogous results for a covariant multifunctor in any number of arguments: in computing the hypercohomology, it is not necessary to replace all the complexes by a Cartan–Eilenberg resolution, but only of them, provided that when one fixes any arguments giving them as values injective objects, the covariant functor in the remaining argument is exact.

11.5. Passage to the inductive limit in hypercohomology

Lemma (11.5.1).

Let be a complex of , and for every integer , let be the complex such that for , for . Let be a covariant additive functor

from to , commuting with inductive limits (one assumes that filtered inductive limits exist and are exact in and ). Then is isomorphic to the inductive limit as tends to .

Proof. The construction of an injective Cartan–Eilenberg resolution of is performed by choosing arbitrarily, for each , an injective resolution of and an injective resolution of ; that done, the method of construction shows that the injective resolution of and the derivation operators depend only on the resolutions , and (M, XVII, 1.2). Now, it is clear that one has and for . One has, on the other hand, for each , a canonical injection , being the identity for . The preceding remark shows that if is an injective Cartan–Eilenberg resolution of , one can for each define an injective Cartan–Eilenberg resolution of such that for and for . One can, on the other hand, define a morphism of bicomplexes corresponding to , and the method of definition of this morphism (loc. cit.) shows once again that one may construct it so that is the identity for and . One has thus defined an inductive system of bicomplexes of , of which is evidently the inductive limit as tends to ; by reason of the commutativity of direct sums and inductive limits, the simple complex associated to is also the inductive limit of the simple complex associated to . Since commutes by hypothesis with inductive limits, and the same holds for cohomology (by exactness of the functor ), one indeed has up to isomorphism.

Lemma (11.5.1) allows one to extend to arbitrary complexes , by passage to the inductive limit, properties valid for complexes bounded below. As a first example, we shall prove:

Proposition (11.5.2).

Under the hypotheses of (11.5.1) concerning , and , is a cohomological functor in the abelian category of complexes of .

Proof. We show that we can reduce to the case of complexes bounded below: if one has an exact sequence of complexes, one evidently deduces from it for each an exact sequence , whence by hypothesis an exact sequence

  … → R^nT(K_{(r)}'^•) → R^nT(K_{(r)}^•) → R^nT(K_{(r)}″^•) →^∂ R^{n+1}T(K_{(r)}'^•) → …

these exact sequences forming an inductive system; lemma (11.5.1) and the exactness of the functor show that one has an exact sequence

  … → R^nT(K'^•) → R^nT(K^•) → R^nT(K″^•) →^∂ R^{n+1}T(K'^•) → …

To deal with the case of complexes bounded below, we may confine ourselves to those for which for ; these evidently form an abelian category .

Lemma (11.5.2.1).

In , let be the set of complexes having the

following properties: 1° Every is an injective object of ; 2° For every , one has , and is a direct factor of . Then:

(i) Every is an injective object of .

(ii) Every object of is isomorphic to a subcomplex of a complex belonging to .

Proof. (i) Let be an object of , a subobject of , an object of , and suppose given a morphism , which we wish to extend to a morphism . We shall use the language of the category of modules for simplicity (cf. [27]).

We identify with ; we proceed by induction on , supposing therefore the defined for , compatible with the derivation operators and for and such moreover that: 1° ; 2° If one sets for every , then coincides with on . The morphism gives, by composition with the projections, two morphisms and . Since carries into and vanishes on , it defines a morphism , and since coincides with on , coincides with on . Since , a direct factor of , is injective, there is a morphism which coincides with on and with on . Consider on the other hand the morphism , which vanishes on ; since is injective, there is a morphism , which coincides with on and with 0 on . It suffices then to take to be able to continue the induction.

(ii) To embed in a complex belonging to , one takes for each an injective object of such that there exists an injection . Then set for , and for , with the obvious derivation operator. Set for every (with for ); it is immediate that is injective for every , and that the are compatible with the derivation operators.

Corollary (11.5.2.2).

Every object of admits a right resolution formed of objects of . If is such a resolution, then for any resolution of formed of objects of , there is a morphism of bicomplexes compatible with the augmentations, and any two such morphisms , are homotopic.

Proof. This is none other than (M, V, 1.1 a)) applied to the abelian category .

11.5.2.3.

These preliminaries laid, consider an injective Cartan–Eilenberg resolution of and a resolution of formed of objects of , and let us show that one has an isomorphism . Indeed, one deduces from (11.5.2.2) a morphism of bicomplexes , and consequently a morphism of the first spectral sequences of these bicomplexes. Since by virtue of (11.3.3) these spectral sequences are regular, it suffices (11.1.5) to see that the preceding morphism is an isomorphism for the terms E_2, or, equivalently, that is equal to ; since is a right resolution of , one is reduced to proving the

Lemma (11.5.2.4).

If is a right resolution of an object of such that for every and every , then one has .

Proof. This is a particular case of (T, 2.5.2).

11.5.2.5.

The proof of (11.5.2) now concludes at once, for is an injective resolution of in the abelian category ; in other words, is none other than the right-derived cohomological functor of in the category (T, 2.3).

Proposition (11.5.3).

Under the hypotheses of (11.5.1) concerning , and , let be a bicomplex such that for and such that, for every , is a resolution of ; suppose finally that for every pair and every . Then there exists a functorial isomorphism

Proof. Let be the bicomplex such that for , for ; it is immediate that is the inductive limit of as tends to ; by virtue of the hypothesis on and of (11.5.1), it suffices therefore to prove the proposition when is bounded below, for example for , and for . Let then be a right resolution of formed of objects of (11.5.2.2); there is a morphism of bicomplexes compatible with the augmentations, whence a morphism for the first spectral sequences; lemma (11.5.2.4) shows, as in (11.5.2.3), that this morphism is an isomorphism, whence the conclusion.

Remark (11.5.4).

The preceding arguments prove that the conclusions of (11.5.2) and (11.5.3) are valid in the category of complexes bounded below, without supposing that commutes with filtered inductive limits. Moreover, when one considers only the category of complexes such that for , the characterization of as the system of right-derived functors of in shows that this cohomological functor is universal (T, 2.3).

Another case in which (11.5.2) is valid without making any additional hypothesis on is the case where there exists an integer such that every object of admits an injective resolution of length . Indeed, in the proof of (11.5.1), all the injective resolutions of objects of that intervene may be taken of length , whence it follows at once that the terms of total degree of the bicomplex are equal to those of and finite in number, as soon as is sufficiently large; this entails that for every , as soon as is sufficiently large. With the notation of (11.5.2), one therefore also has for sufficiently large (depending on ), and likewise for and , whence the conclusion. In the same way, (11.5.3) is valid without additional condition on when satisfies the preceding hypothesis and one supposes that the resolutions are of length .

11.5.5.

The result of (11.5.2) generalizes to covariant multifunctors. Consider for example the situation of (11.4.6), where one supposes that in , and

filtered inductive limits exist and are exact, and that commutes with inductive limits. Then is a bifunctor cohomological in each of the complexes , ; to see this, one reduces as in (11.5.2) to the case where and are bounded below; taking then injective resolutions of and of the type described in (11.5.2.2), one is reduced to the general property proved in (M, V, 4.1).

11.5.6.

Similarly, the results of (11.4.7) and (11.5.3) generalize as follows. Suppose (under the hypotheses of (11.5.5)) that one has two bicomplexes , such that and for , that for every , is a resolution of and is a resolution of , and finally that for and for every system of indices . Then one has a functorial isomorphism in and

  R^•T(K^•, K'^•) ⥲ H^•(T(L^{•,•}, L'^{•,•})).                                  (11.5.6.1)

This is established as in (11.5.3) by reducing to the case where and are bounded below.

Suppose moreover that for every pair and for every pair , the functors and are exact in and respectively. Then one also has functorial isomorphisms

  R^•T(K^•, K'^•) ⥲ H^•(T(L^{•,•}, K'^•)) ⥲ H^•(T(K^•, L'^{•,•})).              (11.5.6.2)

The proof is similar to that of (11.4.7).

11.5.7.

One again notes that the results of (11.5.5) and (11.5.6) are valid without supposing that commutes with inductive limits, provided one restricts oneself to complexes , bounded below, or one supposes that every object of (resp. ) admits an injective resolution of bounded length, and that in (11.5.6) the bicomplexes and have their second degree bounded above.

11.6. Hyperhomology of a functor with respect to a complex

11.6.1.

Let be an abelian category, a complex of objects of whose derivation operator is of degree . A left Cartan–Eilenberg resolution of consists of a bicomplex with derivation operators of degree with for , and a morphism of simple complexes , such that the conditions obtained from those of (11.4.2) by "reversal of arrows" are satisfied. If every object of is a quotient of a projective object, every complex of admits a projective Cartan–Eilenberg resolution, that is, one formed of projective objects , with functorial properties similar to those of (11.4.2). Moreover, if is bounded below (resp. bounded above), one may suppose that for (resp. ) if for (resp. ). If every object of admits a projective resolution of length , one may suppose that for .

11.6.2.

Suppose that is a covariant additive functor from to an abelian category . The definition of the hyperhomology and of the spectral sequences of hyperhomology of with respect to a complex of (when they exist) is performed again from (11.4.3) by "reversal of arrows", the terms of the two spectral sequences thus obtained being

where denotes the th derived functor of for , and 0 for ; denotes the complex .

The properties of hyperhomology are not all deducible by simple "reversal of arrows" from those of hypercohomology (unless one makes additional hypotheses of type AB 5*) of T, 1.5 on the category ) because of the regularity conditions on the two preceding spectral sequences, to which one must this time apply the criteria of (11.3.4). These last show that when one supposes that in filtered inductive limits exist and are exact, then the complex defined by the bicomplex exists, and the second spectral sequence is this time regular. If one supposes either that is bounded below, or that there exists an integer such that every object of admits a projective resolution of length , then the two hyperhomology spectral sequences exist (without hypothesis on ) and are biregular.

Proposition (11.6.3).

Let , be two abelian categories, a covariant additive functor from to . Then:

(i) Hyperhomology is a homological functor in the abelian category of complexes of bounded below.

(ii) Let be a complex of bounded below. If for every and every , one has functorial isomorphisms

for .

(iii) Let be a complex of bounded below. Let be a bicomplex such that for and such that, for every , is a resolution of ; suppose finally that for every pair and every . Then there exists a functorial isomorphism

The proofs proceed as those of (11.5.2), (11.4.5) and (11.5.3) in the case of complexes bounded below. We leave the details of these arguments to the reader.

11.6.4.

One has entirely analogous results for covariant multifunctors additive in each of the arguments. For example, for a bifunctor , one has the two hyperhomology spectral sequences with terms given by

  'E_{p,q}^2 = H_p(L_qT(K_•, K_•'))                                            (11.6.4.1)
  ″E_{p,q}^2 = ⊕_{q'+q″=q} L_pT(H_{q'}(K_•), H_{q″}(K_•')).                    (11.6.4.2)

Here too, it is the second spectral sequence which is regular, the two sequences being biregular when one deals with complexes , bounded below, or when the objects of the abelian categories considered have projective resolutions of fixed length.

Moreover:

Proposition (11.6.5).

Let , , be three abelian categories, a covariant bifunctor biadditive from to .

(i) is a homological bifunctor in each of the complexes bounded below , (formed respectively of objects of and of objects of ).

(ii) Suppose and bounded below. Let , be two bicomplexes such that and for , such that for every , is a resolution of and is a resolution of , and finally that for and every system . Then one has a functorial isomorphism

  L_•T(K_•, K_•') ⥲ H_•(T(L_{•,•}, L_{•,•}')).                                 (11.6.5.1)

(iii) Suppose moreover that for every pair and every pair , the functors and are exact in and respectively. Then one has functorial isomorphisms

  L_•T(K_•, K_•') ⥲ H_•(T(L_{•,•}, K_•')) ⥲ H_•(T(K_•, L_{•,•}')).             (11.6.5.2)

The proofs are analogous to those of (11.5.5) and (11.5.6).

11.7. Hyperhomology of a functor with respect to a bicomplex

11.7.1.

Let be an abelian category in which every object is a quotient of a projective object. Consider a bicomplex formed of objects of , and whose two degrees are bounded below; one may always restrict to the case where for or , and this is what we shall do henceforth. One may consider as a (simple) complex formed of objects of the abelian category of complexes of positive degrees of objects of . It follows from lemma (11.5.2.1) (or rather from the "dual" lemma obtained by "reversal of arrows") and from (M, V, 2.2) that admits a projective Cartan–Eilenberg resolution in the category ; such a resolution is a tricomplex of , with all degrees , formed of projective objects, such that for every , , , , constitute projective resolutions of , , , respectively in the category ; in particular, for every pair , is a projective resolution of in .

Proposition (11.7.2).

Let be a covariant additive functor from to an abelian category . With the notation of (11.7.1), the homology of the simple complex associated to the tricomplex is canonically isomorphic to the hyperhomology of the

simple complex associated to (11.6.2), and is the common abutment of six biregular spectral sequences denoted (with , , , , , or ), whose terms are given by the formulas

(Recall that we use the notation to denote the complex of objects for every complex ; for example denotes the complex , where is the hyperhomology of index of the functor with respect to the simple complex .)

Proof. Denote by the simple complex associated to , so that , and set ; it is clear that for each , is a projective resolution of in ; therefore by (11.6.3) and (11.6.4), one has a functorial isomorphism ; as the simple complex associated to the bicomplex is also associated to the tricomplex , this proves the first assertion of the statement.

Moreover, is the abutment of the two hyperhomology spectral sequences (11.6.2) of relative to the simple complex , which are nothing other than the sequences and respectively.

Consider now as a bicomplex with ; is again the abutment of the two spectral sequences of the bicomplex . Now, for every , is a bicomplex satisfying the conditions of (11.6.3) relative to the simple complex ; hence , and the first spectral sequence of is nothing other than . On the other hand, for every , is a Cartan–Eilenberg resolution of the simple complex ; the calculation done in (M, XV, 2) shows that , hence ; in other words, the simple complex is nothing other than the simple complex associated to the bicomplex . Now, for every , is a projective resolution of the simple complex in the category ; applying (11.6.3), one sees that one has

and one thus obtains the sequence . Finally, the sequences and are obtained by interchanging the roles of the indices and in the definition of the tricomplex and applying to this new tricomplex the preceding arguments.

One says that is the hyperhomology of relative to the bicomplex .

Remarks (11.7.3).

(i) It follows from (11.6.3) that is a homological functor in the category of bicomplexes of bounded below in each of their degrees.

(ii) Let be a tricomplex of such that for each pair , is a resolution of and that for all triples and every . Then one has an isomorphism ; indeed, with the notation of the proof of (11.7.2), is a resolution of such that for and every pair , and it suffices to apply (11.6.3, (iii)).

(iii) One generalizes at once the results of (11.7.2) to covariant multifunctors; for example, let be a second abelian category in which every object is a quotient of a projective object, a bicomplex of whose two degrees are bounded below, and a covariant additive bifunctor from to an abelian category . If and are the simple complexes associated to and respectively, one defines the hyperhomology of with respect to the two bicomplexes , as the hyperhomology ; applying (11.6.4) and (11.6.5), one has, as in (11.7.2), six spectral sequences abutting to this hyperhomology, which we leave the reader to write out.

11.8. Complements on the cohomology of simplicial complexes

11.8.1.

Let be a finite set, the set of finite sequences of elements of ("simplices" of ); one sets ; recall that the chain complex is the free graded abelian group generated by the elements of , being of degree , with a differential defined by

  d(α_0, …, α_h) = ∑_{i=0}^h (−1)^i (α_0, …, α̂_i, …, α_h).

The subgroup of generated by the chains for which two of the are equal, and by the chains , where for every permutation , and is the signature of , is a subcomplex of whose elements are called degenerate chains; one sets , and one has a natural homomorphism of complexes . One defines on the other hand a homomorphism of complexes as follows: one totally orders ; to the class mod. of a simplex , one associates 0 if two of the are equal, and the sequence of the arranged in increasing order otherwise. It is clear that is the identity of .

11.8.2.

Let be a second finite set; if , are the differentials of and , the tensor product complex can be considered as the free abelian group generated by the elements of , with the differential if .

The natural homomorphisms , define a homomorphism , this last tensor product being isomorphic to . Likewise, by means of the homomorphisms and defined in (11.8.1) (by means of total orders on and ), one defines a homomorphism such that is the identity.

With this notation:

Proposition (11.8.3).

There exists a homotopy such that is a linear combination of pairs of simplices with , , and such that for one has

  f − 1 = h ∘ d + d ∘ h.                                                       (11.8.3.1)

Proof. It suffices to define on each pair of simplices, reasoning by induction on the sum of the degrees of and , since one can take when this sum is 0. Let ; by the induction hypothesis and the definition of , one has . One has

  dω = f(d(σ, τ)) − d(σ, τ) − d(h(d(σ, τ))) = h(d(d(σ, τ))) = 0

by virtue of (11.8.3.1) and the induction hypothesis. Now, one has for (G, I, 3.7.4), hence also for , by virtue of Künneth's formula (G, I, 5.5.2). Applying this remark on replacing by and by , one sees that there exists an element of such that ; taking , one verifies (11.8.3.1) for the pair and the induction can continue.

11.8.4.

The notation being that of (11.8.2), we shall set if and . A system of coefficients on consists of a family of abelian groups, where depends only on the sets and , and a family of homomorphisms for , forming an inductive system for this preorder relation. One then defines a cochain complex as the set of families where runs over , with for every pair . The differential is given as follows: if , one has , for every , and one takes

  dλ(σ, τ) = ∑_i ± λ_i(σ_i, τ_i),

where denotes the canonical image of in .

We shall say that a cochain is bi-alternating if whenever one of the two simplices , has two equal terms, and if and for arbitrary permutations , of the indices. It is clear that these cochains generate a subcomplex of .

Proposition (11.8.5).

The canonical injection defines an isomorphism for the cohomology of these two complexes.

Proof. Note that if and have the meaning defined in (11.8.2), the maps and are defined in and respectively, the first being none other than the canonical injection. Since is the identity, it suffices to show that is homotopic to the identity; now, by (11.8.3), is defined in and one can therefore transpose the identity (11.8.3.1), which yields the desired result.

11.8.6.

Proposition (11.8.5) reduces the computation of the cohomology of to that of the cohomology of . Recall on the other hand that this last is, by the Eilenberg–Zilber theorem (G, I, 3.10.2), canonically isomorphic to the cohomology of the cochain complex defined as follows: one forms the chain complex , consisting of the linear combinations of the such that and have the same degree; the differential of this complex is given by if and ; one then has two canonical homomorphisms of complexes

  f : P_•(A, B) → C_•(A) ⊗ C_•(B),    g : C_•(A) ⊗ C_•(B) → P_•(A, B),

and one shows (loc. cit.) that there are homotopies , such that

  f ∘ g − 1 = d ∘ h + h ∘ d    and    g ∘ f − 1 = d ∘ h' + h' ∘ d.

Moreover, one has and and the homotopies , may be taken such that and . This point arises from the fact that the definition of and can be made by induction on the sum of the degrees of and , and from the fact that the and are zero for (loc. cit.); one reasons then as in (11.8.3) and the conclusion follows.

One then defines as the set of families where runs over the pairs whose terms have the same degree, with , and since one has and ,

  dλ(σ, τ) = ∑_{j, k} (−1)^{j+k} λ_{j,k}(σ_j, τ_k)

is defined and gives the differential of the complex . With this, the maps , , and are all defined by virtue of the preceding remarks; and are therefore homotopic to the identity, whence the desired isomorphism between the cohomology of and that of .

Remark (11.8.7).

The same reasoning as in (11.8.3), but applied to and , shows that if and are defined as in (11.8.1), verifies once more a relation (11.8.3.1), with , whence one deduces as in (11.8.5) an isomorphism of the cohomology of onto that of , these two complexes being defined obviously. This is the result whose proof is sketched in (G, I, 3.8.1).

11.8.8.

Take up now the notation and hypotheses of (11.8.2), and consider a complex of systems of coefficients on : for each , the therefore form a complex of abelian groups , and one has the commutative diagrams

for . Then one verifies at once that is a bicomplex of abelian groups, and is a sub-bicomplex of it.

Proposition (11.8.9).

The canonical injection defines an isomorphism for the cohomology of these two bicomplexes.

Proof. Set and for simplicity, and note that since for , the second spectral sequences of these bicomplexes are regular (11.3.3); the homomorphism therefore gives a morphism of spectral sequences which, for the terms E_2, reduces to

But for every , it follows from (11.8.3) that the homomorphism is bijective; the conclusion therefore follows from (11.1.5).

11.8.10.

Likewise, with the notation of (11.8.6), one has canonical homomorphisms of bicomplexes (with obvious notation), and the same reasoning as in (11.8.9), based this time on (11.8.6), shows that this homomorphism again gives an isomorphism in cohomology.

11.9. A lemma on complexes of finite type

Proposition (11.9.1).

Let be an abelian category, and parts of the set of objects of , such that , and verifying the following conditions:

(i) For every object and every epimorphism in , there exists an object and a morphism such that uv is an epimorphism.

(ii) For every pair of objects , and every epimorphism , belongs to .

(iii) The product of two objects of belongs to .

Let be a complex in , such that for every , and such that there exists with for . Then there exists a complex in such that for every and for , and a morphism of complexes such that the corresponding morphism is an isomorphism.

Proof. Let us first prove the following consequence of property (i):

(i bis) Let be an epimorphism in , an object of , a morphism in ; then there exists an object , an epimorphism and a morphism such that the diagram

                u'
            D    →    A
            ↓         ↓                                                        (11.9.1.1)
           v'         v
            ↓         ↓
            C    →    B
                u

is commutative.

Indeed, consider the fibre product in and the canonical projections , , making the diagram

                          q
            C ×_B A    →    A
                ↓             ↓                                                (11.9.1.2)
                p             v
                ↓             ↓
                C    →     B
                       u

commutative.

It is known ([27], p. I-12) that the cokernel of is the quotient of by ; since is an epimorphism, and , hence is an epimorphism; it then suffices to apply (i) to the epimorphism : there is an object and a morphism such that qw is an epimorphism; one takes , .

That being said, to prove the proposition, we proceed by induction. Suppose, for some , we have constructed, for , the objects , the morphisms and the morphisms so that for , that and for ; in addition, we suppose verified the following conditions:

(I_i) One has for and for .

(II_i) For , the homomorphism deduced from the family is an isomorphism.

(III_i) The composite morphism (where the left arrow is the restriction of and the right arrow is the canonical morphism) is an epimorphism.

Note that, by (ii), , the kernel of the epimorphism , belongs to by virtue of hypothesis (I_i). One again deduces from (ii) that also belongs to , taking into account hypothesis (III_i). By virtue of (i bis), there exists a , an epimorphism and a morphism , such that the diagram

is commutative.

Since the canonical morphism is an epimorphism and by hypothesis, it follows from (i) that there exists an object and a morphism such that the composite is an epimorphism. If one takes equal to 0, the diagram

is commutative, the horizontal arrow at the bottom being 0. Then take , which belongs to by virtue of (iii), and , . Since , one has and, with the usual notation, , which verifies (I_{i+1}). The commutativity of the diagrams (11.9.1.3) and (11.9.1.4) shows that . By definition of , the morphism deduced from the system of the () is the morphism deduced from by passage to quotients, hence it is an isomorphism since is an epimorphism, whence (II_{i+1}). Finally, one has by definition; the choice of shows that the morphism is an epimorphism, its restriction to already being so, whence (III_{i+1}). The induction can therefore continue, and the proposition is proved.

Corollary (11.9.2).

Let be a Noetherian ring (not necessarily commutative), a complex of right -modules. Suppose that the are -modules of finite type and that there exists such that for . Then there exists a complex formed of right -modules free of finite rank, such that for , and a homomorphism of complexes, such that the homomorphism corresponding to is bijective.

Proof. One applies (11.9.1) taking for the category of right -modules, for (resp. ) the set of -modules of finite type (resp. the set of -modules free of finite rank); verification of conditions (i), (ii) and (iii) of (11.9.1) is immediate, taking into account the hypothesis that is Noetherian.

Remarks (11.9.3).

(i) Under the conditions of (11.9.2), suppose moreover that the are flat right -modules. Then, for every left -module , the homomorphism of complexes still defines an isomorphism of the homology, as we shall see in chap. III.

(ii) The conclusion of (11.9.2) is no longer necessarily exact when one does not suppose Noetherian; indeed, applying it to a complex reduced to 0 except for a single term, one would conclude that every left -module of finite type admits a resolution by free modules of finite type, which is not true in general (cf. Bourbaki, Alg. comm., chap. I, § 2, exerc. 6).

However, instead of supposing Noetherian, one may suppose only that the have an -presentation finite (cf. chap. IV).

11.10. Euler–Poincaré characteristic of a complex of modules of finite length

11.10.1.

Let be a ring (not necessarily commutative),

  M^• : 0 → M^0 → M^1 → … → M^n → 0                                            (11.10.1.1)

a complex of left -modules of finite length. One calls Euler–Poincaré characteristic of this complex the number

  χ(M^•) = ∑_{i=0}^n (−1)^i long(M^i).                                          (11.10.1.2)

Proposition (11.10.2).

For every finite complex of left -modules of finite length, one has ( being considered as a complex for the trivial derivation). In particular, if the sequence (11.10.1.1) is exact, one has .

Proof. Set, to abbreviate, , , ; the , , are of finite length. From the exact sequences

one derives the relations

  long(Z^i) = long(H^i) + long(B^i)
  long(M^i) = long(Z^i) + long(B^{i+1})

whence

  long(M^i) − long(H^i) = long(B^{i+1}) + long(B^i)

Multiply this relation by and sum the relations obtained for , noting that ; the desired equality follows.

Corollary (11.10.3).

Let be a spectral sequence in the category of modules over a ring . Suppose that the are -modules of finite length and that there are only finitely many pairs such that . Then the Euler–Poincaré characteristics of all the complexes (11.1.1) are all equal. If, in addition, the sequence is weakly convergent and one sets for every , one also has , being considered as a complex with trivial derivation.

Proof. Note first that if , one has for , so all the complexes are finite and formed of -modules of finite length; the relation for every finite therefore follows from (11.10.2) and the isomorphism between and (as complexes with trivial derivation). The hypothesis that the are of finite length entails that for every pair , the sequences and are stationary; the hypothesis that is weakly convergent and that except for finitely many pairs entails therefore that there exists an integer such that for every pair ; whence the assertion concerning .