Homological Algebra With The Example of D-Modules: Date: ?

HOMOLOGICAL ALGEBRA
WITH THE EXAMPLE OF D-MODULES
Contents
Part 0. Announcements 6
0. What does the Homological algebra do? 6
0.1. Some examples of applications 6
0.2. Topics 6
0.3. The texts 7
Part 1. Intro 8
1. Algebraic topology: Homology from triangulations 8
1.1. Linear simplices 8
1.2. Topological simplices 9
1.3. Triangulations 9
1.4. Simplicial complexes 10
1.5. Complex C∗ (X, T ; k) 10
1.6. Examples 11
2. Duality for modules over rings 13
2.1. Rings 13
2.2. Duality and biduality of k-modules 13
2.3. What is the dual of the abelian group Zn ? 14
2.4. Resolutions 16
2.5. Derived category of k-modules 18
2.6. Derived versions of constructions 20
2.7. Duality for k = C[x1 , ..., xn ] 21
2.8. Homology and derived constructions 22
3. Sheaves 23
Date: ?
1
2
3.1. Definition 23
3.2. Global sections functor Γ : Sheaves(X) → Sets 24
3.3. Projective line P1 over C 24
3.4. Čech cohomology of a sheaf A with respect to a cover U 25
3.5. Cohomology of vector bundles 26
3.6. Geometric representation theory 27
3.7. Cohomology of the constant sheaf is dual to homology 27
4. D-modules 28
4.1. Intro 28
4.2. D-modules and differential equations 29
4.3. Higher solutions 30
4.4. Riemann-Hilbert correspondence: differential equations are the same as solutions 31
4.5. Differential equations (or D-modules) with Regular Singularities 31
4.6. Functoriality of D-modules 33
4.7. D-modules on a smooth algebraic variety X 34
4.8. Local systems 35
4.9. Constructible sheaves 36
4.10. Functoriality of sheaves 37
Part 2. Categories 38
5. Categories 38
5.1. Categories 38
5.2. Objects 39
5.3. Limits 41
5.4. Categories and sets 45
5.5. Functors 45
5.6. Natural transformations of functors (“morphisms of functors”) 47
5.7. Adjoint functors 48
5.8. Higher categories 50
5.9. Construction (description) of objects via representable functors 50
5.10. Completion of a category A to Ab 52
6. Abelian categories 54
3
6.1. Additive categories 54

6.2. (Co)kernels and (co)images 54
6.3. Abelian categories 56
6.4. Abelian categories and categories of modules 56
7. Exactness of functors and the derived functors 57
7.1. Exactness of functors 57
7.2. Left exact functors 58
7.3. Right exact functors 60
7.4. Projectives and the existence of projective resolutions 60
7.5. Injectives and the existence of injective resolutions 61
7.6. Exactness and the derived functors 63
8. Abelian category of sheaves of abelian groups 65
8.1. Categories of sheaves 65
8.2. Sheafification of presheaves 66
8.3. Inverse and direct images of sheaves 68
8.4. Stalks 70
8.5. Abelian category structure 72
Part 3. Derived categories of abelian categories 75

9. Homotopy category of complexes 75
9.1. Category C(A) of complexes in A 75
9.2. Mapping cones 77
9.3. The homotopy category K(A) of complexes in A 80
9.4. The triangulated structure of K(A) 81
9.5. Long exact sequence of cohomologies 84
9.6. Exact (distinguished) triangles and short exact sequences of complexes 84
9.7. Extension of additive functors to homotopy categories 85
9.8. Projective resolutions and homotopy 86
9.9. Derived functors LF : A → K − (B) and RG : A → K + (B) 87
10. Bicomplexes and the extension of resolutions and derived functors to complexes 87
10.1. Filtered and graded objects 88
10.2. Bicomplexes 88
4
10.3. Partial cohomologies 90

10.4. Resolutions of complexes 91
10.5. Spectral sequences 95
11. Derived categories of abelian categories 96
11.1. Derived category D(A) of an abelian category A 96
11.2. Truncations 98
11.3. Inclusion A֒→ D(A) 98
11.4. Homotopy description of the derived category 99
12. Derived functors 99
12.1. Derived functors Rp F : A → A 99
12.2. Derived functors RF : D + (A) → D + (B) 100
12.3. Usefulness of the derived category 102
Part 4. Homeworks 103

Homework 1 103
1.1. Homology of a torus 103
Homework 2 103
2.1. Homology of spheres 103
2.2. Euler characteristic 103
2.3. Duality for k-modules 103
2.4. Biduality for k-modules 103
2.5 Dualizing maps 104
Homework 3 104
3.1. The dual of the structure sheaf of a point in a plane 104
Homework 4 104
4.0. The sheaf of solutions is constant on an interval 104
4.1. Tensor product of A-modules. 104
4.2. The universal property of the tensor product ⊗A 105
4.3. Cancellation, Quotient as tensoring, Additivity, Free Modules 105
4.4. Tensoring of bimodules. 105
4.5. Tensoring over a commutative ring. 105
4.6. Tensoring over a field. 106
5
4.7. Tensoring of finite abelian groups over Z 106

Homework 5 106
5.0. Inverse image and the direct image of D-modules 106
5.1. Multiple tensor products. 106
5.2. Tensor algebra of an A-module. 107
5.3. Right exactness of tensor products. 108
5.4. Exterior algebra of an A-module. 108
5.5. Universal property of the exterior algebra 108
5.6. Basic properties of exterior algebras 109
5.7. Symmetric algebra of an A-module. 109
5.8. Universal property of the symmetric algebra 109
5.9. Basic properties of symmetric algebras. 109
Homework 6 110
6.0. Koszul complex 110
6.1. An example of inductive limit 110
6.2. An example of projective limit 110
6.3. Passage to a final subset in an inductive limit 110
6.4. Fibered products of sets 111
6.5. Use of coordinates in Linear Algebra 111
6.6. Construction of projective limits of sets 111
6.7. Construction of inductive limits of sets 111
Homework 7 111
7.1. Functoriality of modules under the change of rings 111
7.2. Representable functors 112
7.3. Direct image of sheaves 112
7.4. Examples of sheaves 112
7.5. Examples of maps of sheaves 112
7.6. Kernels, images and quotients of maps of sheaves 112
7.7. Stalks 113
So Okada has suggested a number of corrections.

6
Part 0. Announcements
0. What does the Homological algebra do?
Homological algebra is a general tool useful in various areas of mathematics. One tries
to apply it to constructions that morally should contain more information then meets the
eye. The homological algebra, if it applies, produces “derived” versions of the construction
(“the higher cohomology”), which contain the “hidden” information.
The goal is to understand the usefulness of homological ideas in applications and to use
this process as an excuse to visit various interesting topics in mathematics.
0.1. Some examples of applications. Algebraic analysis of linear differential

equations is based on the observation that any map between two spaces allows you to
move a system of linear equations on one of the spaces to the other. These operations
become most useful after passing to their derived versions.
Algebraic topology. It can loosely be described as a “systematic way of counting holes
in manifolds”. While we can agree that a circle has a 1-dimensional hole (in the sense of
“a hole that can be made by a one dimensional object”) and a sphere has a 2-dimensional
hole, algebraic topology finds that the surface of a pretzel has one 2-dimensional hole and
four 1-dimensional holes. These “holes” or “cycles” turn out to be essential in problems
in geometry and analysis.
Dual of a module over a ring. The dual V ∗ of a real vector space V is the space of
linear maps from V to real numbers. If one tries to do the same for a module M over
a ring k (say the ring of integers), it does not work as well since M ∗ can often be zero.
However in the “derived” world the construction works as well as for vector spaces.
Cohomology of sheaves. It deals with an omnipresent problem of relating local and
global information on a manifold.
Subtle spaces. In order to organize interesting objects such as all curves or all vector
bundles on a given curve, into a mathematically meaningful space, one requires an exten-
sion of a notion of a space. In contemporary physics the basic objects of string theory –
the D-branes – are expected to be highly sophisticated constructs of homological algebra.
0.2. Topics.
• Algebraic topology.
• Duality of abelian groups.
• Derived functors, Ext and Tor.
• Solutions of linear differential equations with singularities.
• Sheaves and cohomology of sheaves.
• Derived categories.
7
Possible advanced topics: (1) Differential graded algebras, (2) n-categories. (3) Ex-
tended notions of a space: stacks and dg-schemes. (4) Homotopical algebra.
0.3. The texts.

(1) Weibel, Charles A., An introduction to homological algebra, Cambridge Univer-
sity Press, Cambridge [England] ; New York : Cambridge studies in advanced
mathematics 38; ISBN/ISSN 0521435005.
(2) Gelfand, S. I., Manin Y. I., Methods of homological algebra, Springer, Berlin ; New
York : ISBN/ISSN 3540547460 (Berlin) 0387547460 (New York).
These are very different books. Manin and Gelfand Jr. are top mathematicians and their
book is full of exciting material from various areas, and it points towards hot develop-
ments. For that reason (and a laconic al style of Russian mathematics) it is also more
difficult. (Moreover it is said that the English translation has a huge number of confusing
typos.)
Weibel’s book deals with a more restricted subject, so it is less exciting but seems fairly
pleasant to read once one knows what one wants from homological algebra.
In any case I will hand out copies of chapters from both books, and other material sufficient
for the course.
8
Part 1. Intro
This part is an announcement for easily believable ideas we will revisit in more detail.
The idea is for me to pose some problems and to solve these problems with a little help.
1. Algebraic topology: Homology from triangulations
The goal is to measure properties of shapes i.e., of topological spaces. In a simple example
of a circle our measurements give the following observations: (0) it is connected, and (1)
it has a hole. We will find a systematic approach (homology of a topological space) to
finding such properties. This involves the following steps:
• In order to extract information about a topological space X we make a choice

of additional data – a triangulation T , i.e., we break X into oriented simplices
(points, intervals, triangles,...).
• Some information on (X, T ) is encoded into an algebraic construct, a complex
C• (X, T ; k) of vector spaces over a field k (such as Q, R, C): vector spaces Ci count
simplices of dimension i in the triangulation and the boundary maps between them
encode the way the simplices are attached to form X.
• Distill from this the interesting information: the homology H• (X, T ; k) of the
complex C• (X, T ; k) is called the “homology of the topological space X with co-
efficients in k (calculated using triangulation T )”. This is a smaller object then
C• (X, T ; k) – the passage to homology of the complex selects the interesting in-
formation. encoded in smaller vector spaces Hi .
• The end result, i.e., the homology H• (X, T ; k), does not depend on the auxiliary
choice of the triangulation T so we denote it H• (X; k).
1.1. Linear simplices. Calculation of homology of topological spaces is based on a geo-

metric idea of a simplex: 0-simplex is a point, 1-simplex is a closed interval, 2-simplex is
a triangle, 3-simplex is a pyramid, etc.
1.1.1. Standard simplices. The standard n-simplex is σn ⊆ Rn+1 , the convex closure σn =
Rn+1 . So, σn is in the first “quadrant” xi ≥ 0,
conv{e0 , ..., en } of the standard basis ofP
and there it is given by the hyperplane xi = 1.
More generally, we say that a linear i-simplex in Rn is the convex closure conv(v0 , ..., vi )
of a set of i + 1 vectors which lie in an i-dimensional affine subspace but do not lie in any
(i − 1)-dimensional affine subspace.
P
For instance e0 , ..., en lie in the n-dimensional hyperplane xi = 1 but not in any
(n − 1)-dimensional affine subspace.
9
1.1.2. Facets. We call V = {v0 , ..., vi } the vertices of the simplex conv(V), so an i-simplex
has (i + 1) vertices. The facets of the simplex conv(V) are the simplices associated to
subsets of the set of vertices – any subset W ⊆V defines a facet of conv(V) which is the
simplex conv(W ). The facets are closed under intersections: conv(W ′) ∩ conv(W ′′ ) =
conv(W ′ ∩ W ′′ ).
The facets of codimension 1 are called faces.
The facet of σn corresponding to a subset I⊆{0, ..., n} is σn (I) = conv{ei , i ∈ I} Observe
that the ordering of I = {i0 < · · ·ik } gives a canonical identification σn (I) ∼
= σk .
1.1.3. Orientations of simplices. We say that an ordering of the set of vertices V of a

simplex conv(V) gives an orientation of the simplex; and that two orderings give the same
orientation if they differ by an even permutation of vertices.
Standard simplices σn (and their facets) have standard orientation given by the ordering
e0 < · · · < en , we write this ordering from the right to the left as (en , ..., e0 ).
1.1.4. Barycentric coordinates

P on the simplex conv(v
P 0 , ..., vi ). Any point x in the simplex
can be written as x = xi vi with xi ≥ 0 and xi = 1. Barycentric coordinates xi are
unique.
P P
Proof. If v0 = 0 then v1 , ..., vi have to be independent. So x = i0 xp vp = i1 xp vp hence
P
xp , p > 0 are determined by x, and then so is x0 = 1 − i1 xp . In general, we can always
translate a simplex into another one with v0 = 0.
In particular one can recover the vertices from the simplex conv(v0 , ..., vi ) as the the points
with all but one coordinate zero.
The interior conv ◦ (V) of a linear simplex conv(V) consists of the points with all xi > 0.
1.2. Topological simplices. A topological i-simplex Y is a topological space Y with a

∼
=
homeomorphism φ : σi −
→Y . We say that the orientation of σn gives an orientation of Y .
We say that the facet of Y corresponding to a subset I⊆{0, ..., n} is φ(σn (I)), it has a
canonical structure of a topological simplex.
The faces of Y are the facets Y i corresponding to throwing out the ith vertex, i.e., the
image of conv(e0 , ..., ei−1 , ei+1 , ..., en ).
1.3. Triangulations. The idea of triangulation is to present a given topological space as

a combination of simple spaces – the simplices. Then we will extract the information on
X from the way the simplices are patched together.
There are some choices to be made in our approach. The notion of a simplicial complex
is a notion of a triangulation with certain properties SC1-3 that make it very easy to
describe how simplices fit together to form the space X - everything is stated in terms of
10
the set of vertices. However, the price for the properties SC1-3 is that in practice one
needs a large number of simplices.
A more loose notion of a CW-complex allows using few simplices, but makes the descrip-
tion of how they fit together more difficult. It is stated in terms i-cells in X, i.e., maps
φ
→ X such that the restriction to the interior σn◦ −
σi − →φ(σn◦ ) are homeomorphisms.
1.4. Simplicial complexes. A triangulation of a topological space X is a covering X =

{Xi , i ∈ I} of X by topological simplices Xi , such that (for topological simplices viewed
as subsets of X)
• (SC1) any facet of any simplex Xi is one of the simplices in X,

• (SC2) if Xi ⊆Xj then Xi is a facet of Xj ,
• (SC3) the intersection Xi ∩ Xj is ∅ or a simplex in X (so it is one of spaces Xi , Xj
or one of the facets of Xi or Xj ).
Now a simplex is determined by its vertices. Suppose that two simplices Xi and Xj have
the same set V of vertices. Then Xi ∩ Xj is 6= ∅ so it is a simplex Y . Now Y lies in both
Xi and Xj , hence it is a facet of Xi and of Xj . However a facet that contains all vertices
has to be the simplex itself.
This means that the way the simplices are attached will be completely described in terms
of the combinatorics of the set of vertices T 0 . For that reason one can think of a simplicial
complex as a combinatorial structure: a set V (set of vertices of T ), endowed by a family
F of subsets of V (these subsets correspond to simplices in T – for each simplex Y the set
of its vertices is a subset of V, and the mutual position of two simplices in X is recorded
in the intersection of the sets of their vertices).
1.4.1. Triangulations of S 1 . They are obtained by choosing an orientation of S 1 and n

distinct points A1 , ..., An that go in the direction of the orientation. The triangulation
is given by 0-simplices T 0 = {A1 , ..., An } and and 1-simplices T 1 = {A1 A0 , ..., A1 An } (I
denote by BA the segment from A to B).
If n = 1 this is not a simplicial complex since A1 A1 is not really a 1-simplex by our
definition - it is a circle hence not homeomorphic to σ1 . However, it is a 1-cell and
T = {A1 , A1 A1 } is an efficient CW-complex.
n = 2 still does not give a simplicial complex since A2 A1 ∩ A1 A2 is two points , hence not
a 0-simplex. For n ≥ 3 we do get a simplicial complex.
1.5. Complex C∗ (X, T ; k). Calculation of homology of X includes a choice of “coeffi-

cients group k on X”. k can be any abelian group but we usually choose it to be a
commutative ring. The most interesting case is k = Z but for now we will be happy with
k a field such as Q, R, C.
11
The space of i-chains for (X, T ) is the free k-module

def
Ci = Ci (X, T ; k) = ⊕α∈T i kα
with the basis given by the set of i-simplices T i in the triangulation T .
Pi k operator ∂i : Ci −
The boundary → Ci−1 sends an i-simplex Y to the sum of its faces,
∂i Y = 0 Y . Remember that any facet of a simplex Y ∈ T is some simplex in T , hence
a face Y k is some Z ∈ T – but thisPis only an equality of sets, so we are forgetting the
orientations. So to interpret ∂i Y = i0 Y k as en element of Ci−1 we replace Y i with Z if
the orientations of Y k and Z are the same and we replace Y i with −Z otherwise.
1.5.1. Notion of a complex. A complex of cochains is a sequence of k-modules and maps

∂ −2 ∂ −1 ∂0
· · · −−→ C −1 −−→ C 0 · · · −→ C 1 −
→· · ·,
such that the compositions of coboundary operators ∂ i are zero: ∂ i+1 ∂ i = 0, i ∈ Z.
We often omit the index on the coboundary operator, so we can write the preceding
requirement as ∂◦∂ = 0.
From a complex of cochains we get three sequences of k-modules
def
• i-cocycles Z i = Ker(∂ i )⊆C i ,
def
• i-coboundaries B i = Im(∂ i−1 ) = Im(∂ i−1 )⊆C i ,
def
• i-cohomologies H i = Z i /B i ,
Here we used B i ⊆Z i which follows from ∂∂ = 0.
A complex of chains is the same thing but with maps going down
∂−1 ∂
0 0∂
· · · ←−− C−1 ←− C0 · · · ←− C1 −
→· · ·.
The difference is only the terminology and the notation. In this case we lower the indices
def
and we talk of i-cycles Zi ⊆Ci , i-boundaries Bi ⊆Ci , and i-homologies Hi = Zi /Bi . When
we just say “complex” we usually mean “complex of cochains”.
1.5.2. Lemma. C∗ (X, T ; k) is a complex (of chains).

Proof. One first considers X = σn with the obvious triangulation Tn by all facets of σn .
One has T n = {σn } hence Cn = k·σn . Now one checks that deln−1 (∂n σn ) = 0.
The case of any (X, T ) follows. Any simplex Y ∈ T n comes with a homomorphism
φ
→ Y ⊆X, and this produces maps φi : C• (σn , Tn ; k) −
σn − →C• (σn , Tn ; k) such that ∂iX ◦fi =
σn
fi+1 ◦∂i . Therefore ∂∂σn = 0 implies ∂∂Y = 0.
1.6. Examples.
1.6.1. S n for n = 1 or 2.
12
1.6.2. Torus T 2 . We can view T 2 as a quotient of a rectangle, this makes the drawing of
triangles easier. There is a simple CW-triangulation where one divides the rectangle by
a diagonal into two triangles. It gives a fast calculation of homology.
One can get a simplicial complex, for instance by dividing the rectangle into nine rectan-
gles and each of these into two triangles. Then H0 and H2 are easy and the dimension of
H1 can be computed from the invariance of Euler characteristic under taking homology
(Homework 2.2).
Of a particular interest is a basis of H1 – one can see that it corresponds to two “main”
circles on the torus. Classically such basis controls the indeterminacy of elliptic integrals.
In the modern algebraic geometry one says that such basis produces the so called periods,
the basic invariants of elliptic curves.
13
2. Duality for modules over rings
It is easy to extend the construction of a dual vector space to duality for modules over
any ring. However this “naive” notion of duality is not very useful since it does not have
the standard properties of the duality for vector spaces. The correct notion of duality
involves going from modules to a larger world of complexes where we find the hidden part
of the dual of a module. This involves:
(1) thinking of abelian groups as complexes in degree 0,
(2) “Identifying” some complexes, in particular a module should be identified with its
resolution.
These steps mean that we change twice the realms (categories) in which calculate
(1) (2)
m(k) −→ C ∗ (m(k)) −→ D(m(k)),
from the category of k-modules m(k) to the category of complexes of k-modules C ∗ (m(k))
and to the derived category D(m(k)) of k-modules. Step (1) allows one to think of any
k-module in terms of particularly nice modules – the free modules. Step (2) leads to the
setting D(m(k)) which makes it precise what I mean by “identifying some complexes”.
2.1. Rings. Some of the classes of interesting rings k

• fields such as Q, R, C or the finite fields Fq with q elements,
• Z (related to number theory, i.e., everything),
• polynomial functions O(An ) = C[x1 , ..., xn ] (related to geometry),
• Differential operators DAn (related to linear differential equations).
By a module we mean a left module unless specified differently. Let m(k) = ml (k) be the
category of left k-modules and mr (k) the right k-modules.
2.2. Duality and biduality of k-modules. The dual of a left k-module M is the space
D(M) = M ∗ of linear functionals
def def
M ∗ = Homk (M, k) = {f : M −
→k; f (c·m) = c·f (m) & f (m′ +m′′ ) = f (m′ )+f (m′′ ), c ∈ k, m, m′ , m′′ ∈ M
For a right module N one asks that f (n·c) = f (n)·n. This construction is a functor, i.e., it
is defined not only on k-modules but also on maps of k-modules; the dual of f : M1 − →M2
is the adjoint map D(f ) = f ∗ : M2∗ −
→M1∗ , f ∗ (ν) m = hν, f mi, m ∈ M1 , ν ∈ M2∗ .
D D
2.2.1. Lemma. ml (k) −
→ mr (k) and mr (k) −
→ ml (k).
→ (M ∗ )∗ is well defined by
2.2.2. Lemma. For M ∈ ml (k), the canonical map ιM : M −
ιM (m) (λ) = hλ, mi, m ∈ M, λ ∈ M ∗ .
14
2.2.3. Lemma. If k is a field and M ∈ mf d (k) (i.e., M is a finite dimensional vector space
over k), then ιM is an isomorphism.
We would like to have the same situation for each k-module M. At least it works for
M = k (but not always, see 2.3):
2.2.4. Lemma. For a module M = k ∈ ml (k):

def
(a) The map that assigns to a ∈ k the operator of right multiplication Ra x = x·a, gives
R
an isomorphism of right k-modules k −
→ k∗ .
(b) ιk is an isomorphism.
∼
=
→(M ∗ )∗
When this is pushed a little further, we get a nice class of modules for which M −
:
2.2.5. Lemma. ιM is an isomorphism for any finitely generated free k-module.

Proof. It follows from the lemma 2.2.4b, and from
2.2.6. Sublemma. For two k- modules P, Q; (a) (P ⊕Q)∗ ∼

= P ∗⊕Q∗ , (b) map ιP ⊕Q is an
isomorphism iff both ιP and ιQ are isomorphisms.
2.3. What is the dual of the abelian group Zn ? We will notice that the duality
operation M7→M ∗ is not very good for arbitrary modules M of a ring k. To improve it
we will be forced to
• (i) pass to finitely generated free k-modules (here duality behaves as for finite
dimensional vector spaces), and
• (ii) from modules to complexes of modules (this is what it takes to be able to
describe any finitely generated module in terms of free finitely generated modules).
We will be guided through this by an example with k the ring of integers and M = Zn a
torsion module. For k = Z, category of Z-modules is just the category of abelian groups:
def
m(Z) = Ab. So we have the notion of a dual of an abelian group M ∗ = HomAb (M, Z).
def
For M = Zn = Z/nZ one has M ∗ = 0, so duality loses all information.
2.3.1. The passage from Zn to its resolution P • . We know that biduality works for the
abelian group M = Z (by 2.2.4), and Zn is clearly intimately related to Z. The quotient
q
map Z − → Zn relates Zn to Z, however it does not tell the whole story – the difference
between Zn and Z is in the kernel Ker(q) = nZ. However, the inclusion nZ⊆Z captures
the definition of Zn as Z/nZ, and since the abelian group nZ is isomorphic to Z by
Z ∋ x7→nx ∈ nZ, we will replace nZ by Z in this map. Then it becomes the multiplication
n
map Z − → Z.
15
n
We will think of Zn as encoded in the map Z −→ Z. For a more complicated k-module
such encoding will be more complicated, the proper setting will turn out to require to
n n
think of Z −
→ Z as a complex P • = (· · · − →Z −
→0 − →Z − →0 − → · · ·), with Z′ s in degrees
−1 and 0.
So we have passed from Zn ti a complex P •. Now we need to know how to dualize it.
2.3.2. Duality operation on complexes. The dual of a complex of k-modules C • = (· · · −

→
−1
C − →C −0
→C − 1
→ · · ·) is the complex DC obtained by applying D to modules and
•
maps. Since D is contravariant (i.e., it changes directions), we will also have to change
def −n−1
the indexing. So, (DC • )n = D(C −n ) and dnDC • is the adjoint of dC • .
In order to calculate DP • we will need
2.3.3. Sublemma. (a) k-linear maps between kr and ks can be described in terms of matri-
ces. If we denote for A ∈ Mr,s (k) by RA the right multiplication operator kr ∋ x7→xA ∈ ks
R
on row-vectors, then Mrs − → Homk (kr , ks ) is an isomorphism.
(b) The adjoint of RA is the left multiplication LAtr with the transpose of A (on column-
vectors).
2.3.4. Biduality is true on F reef g (k)-complexes. Let F reef g (k) be the category of all free
finitely generated k-modules.
2.3.5. Lemma. The biduality map ιC • is an isomorphism for any complex C • in

C ∗ (F reef g ).
Proof. Observe that (DDC • )n = DD(C n ), and define the map ιC • : C • − → DDC • as the
→ DDC n . If all C n are in F reef g (k) then all maps ιC n are
collection of maps ιC n : C n −
isomorphisms and hence so is ιC • .
def
2.3.6. The derived dual LD(Zn ) = D(P • ). We can identify DZ with Z and then the
n n n
adjoint of Z − → Z is again Z − → Z (see 2.3.3). So, DP • is the complex · · · − →Z −
→0 − →
Z− →0 − → · · ·, but this time Z’s are in degrees 0 and 1 ! So DP • ∼ = P • [−1] where one
denotes by C • [n] the shift of the complex C • by n places to the left.
Finally, it is natural to identify any N with a complex which has N in degree 0 and all
other terms zero (hence all maps are zero). So, since we have also identified Zn with P •,
def
we should identify the smart dual LD(Zn ) = DP • ∼ = P • [−1] with Zn [−1]. So,
LD(Zn ) = Zn [−1], the shift of Zn by one to the right ,
i.e., the complex which has Zn in degree 1 and all other terms zero.
16
2.3.7. Conclusion. The smart dual of Zn is not a module but a complex in degrees ≥ 0.
The fact that H 0 [LD(Zn )] = 0 corresponds to the fact that the naive definition of the
dual gives D(Zn ) = 0. So, the naive definition does not see the hidden part of the dual
which is H 1 [LD(Zn )] = Zn .
Observe that since the computation of the derived dual is in the setting of complexes
of free finitely generated modules, the biduality works (by 2.3.5), so the canonical map
LιZ
Zn −−→
n
(LD)(LD)(Zn ) is an isomorphism.
In 2.4 and 2.5 we will explain the ideas in the calculation of LD(Zn ) so that we can see
that they apply in many situations.
2.4. Resolutions. Here we repeat for any module M what we have been able to do for
Zn . The precise, formal, solution of our wish to describe a module M in terms of maps
between some (hopefully nicer) modules P n . is the notion of a resolution of M.
2.4.1. Exact complexes and short exact sequence. A complex of k-modules C • = (· · · − →

−1 0 1
C − →C − →C − →· · ·) is said to be exact if all of its cohomologies vanish, i.e., inclusion
B n ⊆Z n is equality. (Exact complexes are also called exact sequences.)
2.4.2. Examples. Suppose that in the complex C • all terms C n , n 6= 0 are zero. Then C •
is exact iff C 0 = 0.
If C n = 0 for n 6= −1, 0, then C • is exact iff d−1 : C −1 −
→C 0 is an isomorphism.
If C n = 0 for n 6= −1, 0, 1, then C • is exact iff d−1 : C −1 −
→ C 0 is injective, d0 : C 0 −
→ C1
0 0 −1
is surjective, and in C one has Ker(d ) = Im(d ). So all exact complexes with three
terms (and such complexes are also called short exact sequences), are of the following
form: module C 0 has a submodule C −1 and the quotient C 1 = C 0 /C −1 .
2.4.3. Resolutions. A left resolution of a module M is an exact complex

q
→P −2 −
···− →P −1 −
→P 0 −
→M −
→0 −
→· · ·.
n q
For example · · · − →Z −
→0 − →Z −
→Zn −
→0 −
→· · · (the numbers beneath are the positions
−1 0 1
in the complex) is a resolution of Zn .
→ P −2 −
We will also say that the complex · · · − → P −1 −
→P0 −
→0 − → · · · together with a
map q is a resolution of M. Another way to think of a resolution is as a morphism of
complexes where we extend q to a map complexes
· · · −−−→ P −2 −−−→ P −1 −−−→ P 0 −−−→ 0 −−−→ · · ·
    
    
y y y qy y
· · · −−−→ 0 −−−→ 0 −−−→ M −−−→ 0 −−−→ · · ·
Here,
17
2.4.4. Morphisms of complexes. A map of complexes f : A• −→B • is a system of maps f n

of the corresponding terms in complexes, which “preserves” the differential in the sense
that in the diagram
· · · −−−→ A−2 −−−→ A−1 −−−→ A0 −−−→ A1 −−−→ ···
     
     
y f −2 y f −1 y f 0y f 1y y
· · · −−−→ B −2 −−−→ B −1 −−−→ B 0 −−−→ B 1 −−−→ · · ·
all squares commute in the sense that two possible ways of following arrows give the same
n−1 n−1
result: f n ◦dA = dB ◦f n−1 , for all n; i.e., f ◦d = d◦f .
Now we have a category of complexes of k-modules C • [m(k)]: objects are complexes and
morphisms are maps of complexes.
2.4.5. Free resolutions. We consider subcategories of k-modules F reel (k)⊆ml (k)⊆ml f g (k)
consisting of free modules and of finitely generated modules. The intersection is
F reef g l (k). Now we can say that a free resolution of M is a resolution P • such that all
P i are free k-modules, etc.
2.4.6. Lemma. Any module M has a free resolution.

Proof. (1) There is a free module F and a surjective map F ։M (“a free cover of M”).
For this we choose any set G⊆M of generators of M (for instance G = M), and let F be
the free k-module with the basis G.
q
(2) Let P 0 −
→ M be the map F − → M from (1). If q has no kernel, we are done – we
choose P k = 0, k < 0. Otherwise we use again (1) to choose a free cover P −1 −
→ Ker(q),
−1 −1 0
then ∂ is the composition P − →Ker(q)⊆ P . Etc.
2.4.7. Lemma. If the ring k is noetherian any finitely generated module M has a resolution
by free finitely generated modules.
Proof. (1) If k-module M is finitely generated a free finitely generated module F and a
surjective map F ։M. This is as before, except that we can now choose a set G⊆M of
generators of M, to be a finite set.
To repeat the step (2), we need the new modules to be covered, such as
Ker(q)⊆ P 0 , Ker(∂ −1 )⊆ P −1 , ... are finitely generated. That’s what the noe-
therian assumption means: we say that a ring k is noetherian if any submodule of a
finitely generated module is finitely generated.
2.4.8. Projective modules. We say that a k-module P is projective if it is a summand of a

free k-module. So, free modules are projective and we get a larger class Proj(k)⊇F ree(k).
However it has the same good properties of free modules (so projective resolutions are as
good as free resolutions!).
18
For instance: if P is a summand of a finitely generated free k-module then the biduality
map ιP is again an isomorphism by sublemma 2.2.6.
2.4.9. Finite resolutions. The homological dimension of k is the infimum of all n such
that any k-module has a projective resolution of length ≤ n (i.e., P i = 0, i < −n).
A field has dimension 0, Z has dimension one and C[x1 , ..., xn ] has dimension n. However
the ring O(Y ) = C⊕xyC[x, y]⊆ C[x, y] of functions on the crossing Y = {xy = 0}⊆A2
has infinite dimension (because Y has a singularity).
2.5. Derived category of k-modules. By now, we can compute the dual of any k-
module M – we can find a free resolution P • following the proof of lemma 2.4.6, and then
def
LD(M) = D(P • ). The basic idea is therefore to replace M by its resolution. In what
sense are these the same things?
2.5.1. Identification of module with its resolution. I will explain it in two steps:
(i) already, the use of resolutions suggests that we have to pass from modules to the larger
world of complexes of modules. The most natural way is to associate to a module M the
complex M # which is M in degree zero and 0 in other degrees. This gives a functor
→ C ∗ (m(k)), M7→ M # .
m(k) −
(However, we will usually denote M # just by M again.)
(ii) Therefore the replacement of M by P • is seen in the realm of complexes as the
replacement of a complex M # by P • . The relation between them is given by the map
of complexes M # − → P • from 2.4.3. This map can not be an isomorphisms since the
complexes are really different, however it does become an isomorphism once we pass
to cohomology: the only non-zero cohomology group of P • is H 0 (P • ) ∼
= M, the same
# # •
is true for M and moreover the morphism of complexes M − → P induces idM on
→H 0 (P • ).
H 0 (M # ) −
2.5.2. Derived category via inverting quasi-isomorphisms. We say that a map of complexes
f : A• −→ B • is a quasi-isomorphism (“qis”) if the induced maps of cohomology groups
→H n (B • ), n ∈ Z, are all isomorphisms.
H (f ) : H n (A• ) −
n
So, M # and P • are “somewhat the same” in the sense that they are quasi-isomorphic.
However, we would like them to be truly the same thing, i.e., we would like to proclaim
all quasi-isomorphisms in C • (m(k)) to be isomorphisms! This is achieved by passing from
C • (m(k)) to the so called derived category of k-modules D(m(k)).
2.5.3. Localization of categories. Localization of a ring A with respect to a subset S⊆A

is the ring AS obtained by inverting all elements of S. More precisely, localization of a
i
ring A with respect to S⊆A is a pair of a ring AS and a map of rings A − → AS such that
def
i(S)⊆ (AS )∗ = the set of invertible elements of AS . There may be many such pairs, and
so we have to be still more precise, it is the universal such pair (i.e., the best such pair),
19
k
→ B) such that k(S)⊆B ∗ , there is a unique map of
in the sense that for each pair (B, A −
ι
rings AS −→B such that k = ι◦i.
Theorem. At least if A is commutative the localization of S⊆A exists (and can be de-
scribed).
The localization of a category A with respect to a class of morphisms S⊆Mor(A) is the
i
(universal!) functor, i.e., morphism of categories, A −→ AS such that the images of all
morphisms in S are isomorphisms in AS (i.e., have inverses in AS ). Again, localization
exists and can be described under some conditions.
2.5.4. Maps in the localized category. To make this less abstract I will sketch how one
goes about constructing AS . However we will return to this more precisely.
Observe that a map in A, say, α ∈ HomA (a, b) gives a map in AS , the map is i(α) ∈
HomAS [i(a), i(b)]. Moreover a wrong direction map σ ∈ HomA (b, a) again gives a map
from i(a) to i(b) in AS , the map is i(σ)−1 ∈ HomAS [i(a), i(b)]. Since these are the only
two kinds of maps that are asking to be in AS , it is natural that all maps in AS should
be generated from these two kinds of maps by using composition of maps.
This leads to the following idea: we will have Ob(AS ) = Ob(A) and the morphisms from
a to b in AS will come from diagrams in A
a−
→x ←
−p −
→· · ·q −
→y ←
−b,
where all backwards maps are in S. The last step is to decide when two such diagrams
give the same map in AS !
2.5.5. Derived category of modules and complexes of free modules. According to its def-
inition D(m(k)) looks like a somewhat abstract construction. Fortunately it turns out
that there is a simple description of D(m(k)) in terms of the category of complexes over
free modules C • (F ree(k)).
2.5.6. Do we really want the derived category? The historical origin of the idea is as we
have introduced it: it is a good setting for doing calculations with complexes. However,
the derived category D(A) of a category A (say A = m(k) as above), may be more “real”
than the simple category A we started with. One indication is that there are pairs of
very different categories A and B such that their derived categories D(A) and D(B) are
canonically equivalent. For instance A and B could be the categories of graded modules
•
for the symmetric algebra S(V ) and the exterior algebra ∧V ∗ for dual vector spaces V and
V ∗ . This turns out to be important, but there are more exciting examples: the relation
between linear differential equations and their solutions, mirror symmetry.
20
2.5.7. Bounded categories of complexes. We say that a complex C • is bounded from above
if C n = 0, n >> 0. The categories of such complexes is denoted C − (A) and D − (A)
(meaning that the complexes are allowed to stretch in the negative direction) Similarly
one has C + (A) and D + (A). We say that a complex C • is bounded (or finite) if C n = 0
for all but a finitely many n ∈ Z, this gives C b (A) and D b (A).
2.6. Derived versions of constructions.
2.6.1. Improving an object M. Let A = m(k) and M ∈ A. We improve M by replacing

it with a complex P • of free (or say, projective) modules. This can be described as
Projective α q
A
α
−−−→ resolutions M −−−→ (· · ·→P −1→P 0 →M→0→· · ·)
   
  
βy

, δy
βy δy
γ
C • (A) −−−→ C • (Proj(A)) (· · ·→0→M
γ →0→· · ·) −−−→ (· · ·→P −1 →P 0 →0→· · ·).
0
Combination of α and δ is intuitively clear: this is the description of M in terms of maps

between projective modules. The other route is a more formal formulation of the same
idea. Step β is the passage from modules to a larger world of complexes, and step γ is
the replacement of a complex by a quasi-isomorphic complex.
2.6.2. Any (additive) functor D : A − → B extends to complexes. Let A = m(k) and

B = m(k′ ) be categories of modules over two rings, and let D be a way to construct
from a module for k a module for k′ , i.e., a functor D : A −→ B. It extends to a functor
from A-complexes to B-complexes D • : C • (A) − → C • (B), that assigns to each A-complex
d−1 d0 d1
A• = (· · · −
→A−1 −−→ A0 −→ A1 −→ · · ·) a B-complex
D(d−1 ) D(d0 ) D(d1 )
D • (A• ) = (· · · −
→D(A−1 ) −−−−→ D(A0 ) −−−→ D(A1 ) −−−→ · · ·).
(As we know, if D is contravariant – for instance if D is the duality Dk – the formula for
D • A• would involve switching n and −n).
The main point is that this really is a complex: since D is a functor it preserves com-
positions of morphisms, hence D(dn )◦D(dn−1) = D(dn ◦dn−1 ) = D(0) = 0. Asking that
D is additive i.e., D(A′ ⊕A′′ ) = D(A′ )⊕D(A′′ ), A′ , A′′ ∈ A, is needed for the last step
D(0) = 0.
2.6.3. Left derived version LD of D. It really means that we do not apply D directly to
M but to its improved version P • :
∈ α
M −−−→ A −−−→ Projective
resolutions
 

βy

δy
γ D• ∋ def
C • (A) −−−→ C • (Proj(A)) −−−→ C • (B) −−−→ D • (P • ) = LD(M).
21
2.6.4. Left and right derived functors. In order to say that LD is really an improvement
of D, we need to know that H 0 [LD(M)] = D(M), then LD(M) contains the information
on D(M) and also a “hidden part” given by higher cohomologies H i [LD(M)], i > 0.
This is going to be true precisely if D has a property called right exactness (duality D is
right exact!). There are important functors which are not right exact but have a “dual”
property of left exactness, they will require a “dual” strategy: a right resolution of M:
··· − →I 0 −
→M − →I 1 −
→· · ·
by injective modules. We’ll be back to that.
2.7. Duality for k = C[x1 , ..., xn ]. The commutative ring k = C[x1 , ..., xn ] is the algebra
def
of functions on the n-dimensional affine space An = Cn . Natural examples of k-modules
have geometric meaning.
2.7.1. Affine algebraic varieties. We say that an affine algebraic variety is a subset Y of
some An which is given by polynomial conditions: Y = {z = (z1 , ..., zn ) ∈ Cn ; f1 (z) =
· · · = fc (z) = 0}. The set IY of functions that vanish on Y is an ideal in k (i.e., a
k-submodule of the k-module k). We define the ring O(Y ) of polynomial functions on Y
as the all restrictions f |Y of polynomials f ∈ k to Y . So O(Y ) = k/IY is also a module
for k = O(An ).
2.7.2. Duality.
P We will consider the k-module
P O(Y ) where Y is the origin in A . Then
n
IY = xi ·k and therefore O(Y ) = k/ xi ·k is isomorphic to C as a ring (C-valued

functions on a point!). However it is more interesting as a k-module.
x
2.7.3. n = 1. Here C[x] and IY = xC[x], so we have a resolution · · · − →0 −→ C[x] − →
q
C[x] −→ O(Y ) −→0 − → · · · and the computation of the dual of O(Y ) is the same as in the
case of Zn . One finds that D[O(Y )] ∼= O(Y )[−1].
2.7.4. n = 2. Then O(A2 ) = C[x, y] and O(Y ) = C[x, y]/hx, yi = C[x, y]/(xC[x, y] +
q
yC[x, y]) = k/(xk + yk). The kernel of the covering P 0 = k −
→ O(Y ) is xk + yk. We
α
can cover it turn with P 0 = k⊕k −
→ xk + yk, α(f, g) = xα + yβ. This covering still
contains surplus: Ker(α) = {(−yh, xh); h ∈ k}. However, this is a free module so the
β
next covering P −2 = k −
→ Ker(α)⊆P −1 , β(h)(−yh, xh). This gives a resolution
β α q
···−
→0 − → C[x, y]⊕C[x, y] −
→C[x, y] − → C[x, y] −
→O(Y ) −
→0 −
→· · ·.
As a complex this resolution is
β=(−y,x) α=(x,y)
P • = [· · · − →C[x, y] −−−−−→ C[x, y]⊕C[x, y] −−−−→ C[x, y] −
→0 − →0 −
→· · ·].
−2 −1 0
22
Therefore, computing the adjoints by lemma 2.3.3 gives

α∗ =(x,y) β ∗ =(−y,x)
LD[O(Y )] = DP • = [· · · − →C[x, y] −−−−−→ C[x, y]⊕C[x, y] −−−−−−→ C[x, y] −
→0 − →0 −
→· · ·].
0 1 2
The cohomology of this complex is easy to compute (nothing new!), it gives
Lemma. LD[O(Y )] ∼
= O(Y )[−2].
2.7.5. The general n. The resolutions above for n = 1, 2 are example of the Koszul com-
plex which we will meet later.
2.7.6. Geometric nature of integers. For the ring C[x1 , ..., xn ] = O(An ) of functions on
the n-dimensional affine space, the derived dual of the C[x1 , ..., xn ]-module O(origin) is
the same module O(origin) shifted to the degree n. So the shift is clearly the codimension
of the origin in An , and it is equal to n because dim(An ) = n.
In view of the similarity of computations for O(A1 ) = C[x] and for Z, we may expect that
Z is also a ring of functions on some geometric object, and that its dimension is one. So Z
should correspond to some geometric object which we will denote Spec(Z), and Spec(Z)
is some sort of a curve.
2.8. Homology and derived constructions.
2.8.1. Resolutions and triangulations. The idea is the same – explain complicated objects
in terms of combining simple ones. This can be done in several ways so in the end one has
to check that whatever we produced is independent of choices. (This we leave for later.)
23
3. Sheaves
Sheaves are a machinery which addresses an essential problem – the relation between local
and global information – so they appear throughout mathematics.
3.1. Definition.
3.1.1. Smooth functions on R. Let X be R or any smooth manifold. The notion of smooth
functions on X gives the following data:
• for each open U⊆X an algebra C ∞ (U) (the smooth functions on U),
ρU
• for each inclusion of open subsets V ⊆U⊆X a map of algebras C ∞ (U) −→
V
C ∞ (V )
(the restriction map);
and these data have the following properties
(1) (transitivity of restriction) ρUV ◦ρUV = ρUW for W ⊆V ⊆U,
(2) (gluing) if the functions fi ∈ C ∞ (Ui ) on open subsets Ui ⊆X, i ∈ I, are compatible
def
in the sense that fi = fj on the intersections Uij = Ui ∩ Uj , then they glue into
a unique smooth function f on U = ∪i∈I Ui .
The context of dealing with objects which can be restricted and glued compatible pieces
is formalized in the notion of
3.1.2. Sheaves on a topological space. A sheaf of sets S on a topological space (X, T )

consists of the following data:
• for each open U⊆X a set S(U),
V ρU
• for each inclusion of open subsets V ⊆U⊆X a map S(U) −→ S(V ) (called the
restriction map);
and these data are required to satisfy
(1) (transitivity of restriction) ρUV ◦ρUV = ρUW for W ⊆V ⊆U,
(2) (gluing) Let U = (Ui )i∈I be an open cover of an open U⊆X. For a family of
elements fi ∈ S(Ui ), i ∈ I, compatible in the sense that ρUUiij fi = ρUUiij fj in S(Uij )
for i, j ∈ I; there is a unique f ∈ S(U) such that on the intersections ρUUiij f = fi
in S(Ui ), i ∈ I.
(3) S(U) = ∅.
We can equally define sheaves of abelian groups, rings, modules, etc – only the last, and
least interesting requirement has to be modified, say in abelian groups we would ask that
S(U) is the trivial group {0}.
24
3.1.3. Examples. (1) Structure sheaves. On a topological space X one has a sheaf of
∞
continuous functions CX . If X is a smooth manifold there is a sheaf CX of smooth
functions, etc., holomorphic functions HX on a complex manifold, “polynomial” functions
OX on an algebraic variety. In each case the topology on X and the sheaf contain all
information on the structure of X.
(2) The constant sheaf SX on X associated to a set S: SX (U) is the set of locally constant
functions from U to X.
(3) Constant functions do not form a sheaf, neither do the functions with compact support.
A given class C of objects forms a sheaf if it is defined by local conditions. For instance,
being a (i) function with values in S, (ii) non-vanishing (i.e., invertible) function, (iii)
solution of a given system (∗) of differential equations; are all local conditions: they can
be checked in a neighborhood of each point.
3.2. Global sections functor Γ : Sheaves(X) → Sets. Elements of S(U) are called the
sections of a sheaf S on U⊆X (this terminology is from classical geometry). By Γ(X, S)
we denote the set S(X) of global sections.
The construction S7→Γ(S) means that we are looking at global objects of a given class of
objects defined by local conditions. We will see that it has a hidden part, the cohomology
S7→H • (X, S) of sheaves on X.
Γ recovers connected components i.e., H0 . On a smooth manifold X, Γ(C ∞ ) = C ∞ (X) is
huge and that will be all: H >0 will turn out to be zero (nothing hidden). The holomorphic
setting is more subtle in this sense.
3.2.1. Solutions of differential equations. Solutions of a system (∗) of differential equation

on X form a sheaf Sol(∗) . If X is an interval in R and (∗) is one equation of the form
y (n) + a1 (t)y (n−1) + · · · + a0 (t) = 0 with ai ∈ C ∞ (X) then any point c ∈ X gives evaluation
Ec
isomorphism of vector spaces Sol(∗) (X) −→ Cn by Ec (y) = (y(c), ..., y (n−1)(c)) (solutions
correspond to initial conditions!).
Lemma. Sol(∗) is a constant sheaf on X.

On the other hand let (∗) be the equation zy ′ = λy considered as equation in holomorphic
∼
=
functions on X = C∗ . On any disc c ∈ D⊆X, evaluation still gives Sol(∗) (D)− →C, the
solutions are multiples of functions z λ defined using a branch of logarithm on D). However,
Γ(X, Sol(∗) ) = 0 if λ 6∈ Z (any global solution would change by a factor of e2πiλ as we move
once around the origin). So locally there is the expected amount but nothing globally.
3.3. Projective line P1 over C. P1 = C ∪ ∞ can be covered by U1 = U = C and

U2 = V = P1 − {0}. We think of X = P1 as a complex manifold by identifying U and V
with C using coordinates u, v such that on U ∩ V one has uv = 1.
25
3.3.1. Lemma. Γ(P1 , OP1 ) = C.

Proof. (1) Proof using a P
P cover. A holomorphic function
P f on X restricts
P to f |U =
n≥0 βn v . On U ∩ V = C ,
n n ∗ n −n
n≥0 αn u and to f |V = n≥0 αn u = n≥0 βn u , and
therefore αn = βn = 0 for n 6= 0.
(2) Proof using maximum modulus principle. The restriction of a holomorphic function
f on X to U = C is a bounded holomorphic function (since X is compact), hence a
constant.
3.4. Čech cohomology of a sheaf A with respect to a cover U. Let U = (Ui )i∈I be
def
an open cover of X. We will use finite intersections Ui0 ,...,ip = Ui0 ∩ · · · ∩ Uip .
3.4.1. Calculations of global sections using a cover. Motivated by the calculation of global
sections in 3.3.1, to a sheaf A on X we associate
def Q
• Set C 0 (U, A) = i∈I A(Ui ) whose elements are systems f = (fi )i∈I with one
section fi ∈ A(Ui ) for each open set Ui ,
def Q
• C 1 (U, A) = (i,j)∈I 2 A(Uij ) whose elements are systems g = (gij )I 2 of sections
gij on all intersections Uij .
Now, if A is a sheaf of abelian groups we can reformulate the calculations of global sections
of A in terms of “linear algebra data”:
∼
= d
→ Ker[C 0 (U, A) −
Γ(A)− → C 1 (U, A)]
for the map d which sends f = (fi )I ∈ C 0 to df ∈ C 1 with
U
(df )ij = ρUjij fj − ρUUiij fi .
More informally, (df )ij = fj |Uij − fi |Uij .
3.4.2. Čech complex Č • (U, A). Emboldened, we try more of the same and define the
abelian groups
def
Y
Č n (U, A) = A(Ui0 ,...,in )
(i0 ,...,in)∈I n
dn
of systems of sections on multiple intersections, and relate them by the maps Č n (U, A) −→
Č n+1 (U, A) which sends f = (fi0 ,...,in )I n ∈ C n to dn f ∈ Č n+1 with
n+1
X
n
(d f )i0 ,...,in+1 = (−1)s fi0 ,...,is−1,is+1 ,...,in+1 .
s=0
3.4.3. Lemma. (Č • (U, A), d•) is a complex, i.e., d.◦dn−1 = 0.

26
3.4.4. Čech cohomology Ȟ • (X, U; A). It is defined as the cohomology of the Čech complex
Č • (U, A). We have already observed that
3.4.5. Lemma. Ȟ 0 (X, U; A) = Γ(A).
3.4.6. The “small Čech complex”. If the set I has a complete ordering, we can choose
def Q def Q
in Č n (U, A) = n
(i0 ,...,in)∈I n A(Ui0 ,...,in ) a subgroup C (U, A) = i0 <···<in A(Ui0 ,...,in ).
This is what we will usually use in computations since it is smaller but it also computes
the Čech cohomology:
3.4.7. Lemma. (a) C • (U, A)⊆Č • (U, A) is a subcomplex (i.e., it is invariant under the
differential). (b) Map of complexes C • (U, A)⊆Č • (U, A) is a quasi-isomorphism.
Proof. Intuitively plausible since the data in Č which is not in C • is a duplication of data
in C • .
Lemma. Ȟ • (P1 , OP1 ) = Ȟ 0 (P1 , OP1 ) = C.

Proof. Since the cover we use U = {U1 , U2 } has two elements, C n = 0 for n > 1. We know
Ȟ 0 = Γ(OP1 ) = C, so it remains
P+∞ to understand Ȟ 1 = C 1 /dC 0 = O(U ∩V )/[O(U)+O(V
P+∞ ),
i.e., all Laurent series φ = −∞ γn u that converge on C , modulo the series 0 λn un
n ∗
P+∞
and βn u−n , that converge on C and on P1 − 0. However, if a Laurent series
P+∞ 0 P
φ = −∞ γn un converges on C∗ ,then Laurent series φ+ = +∞ 0 γn un converges on C,
P −1
and φ− = −∞ γn un converge on C∗ ∪ ∞.
3.5. Cohomology of vector bundles.
3.5.1. Moebius strip. It projects to the central curve S 1 and the fibers are real lines.
3.5.2. Vector bundle over space X. One can extend various notions to relative setting
over some “base” X. A reasonable notion of a “vector space over a set X” is a collection
V = (Vx )x∈X of vector spaces, one for each point of X. Then the total space V = ⊔x∈X Vx
maps to X and the fibers are vector spaces. If X is a topological space, we want the
family of Vx to be continuous in x. This leads to the notion of a vector bundle over a
topological space, and similarly over a manifold.
3.5.3. Examples. On each manifold X there are the tangent and cotangent vector bundles
T X, T ∗ X. In terms of local coordinates xi at a, the fibers are Ta X = ⊕ R ∂x∂ i and
Ta X = ⊕ Rdxi .
Moebius strip is a line bundle over S 1 . One can describe it in terms of gluing of trivial
vector bundles on an open cover.
27
This gives a general construction of vector bundles V by gluing trivial vector bundles
Vi = Ui ×kn on an open cover U = (Ui )i∈I , and the transition functions φij : Uij −
→GLn (k).
(Here k = R or C.)
V = [⊔i∈I Ui ×kn ]/ ∼ . where Vi ∋ (ui , z) ∼ (uj , w) ∈ Vj if ui = uj and z = φij (uj )·w.
def
3.5.4. Sheaf V associate to a vector bundle V . Over an open U⊆X, V(U) = “sections of
the vector bundle V over U”. If V is obtained by gluing trivial vector bundles Vi = Ui ×Cn
by transition functions φij , then V(U) consists of all systems of fi ∈ H(Ui ∩ U, Cn ) such
that on all intersections Uij ∩ U one has fi = φij fj .
3.5.5. Line bundles Ln on P1 . On P1 let Ln be the vector bundle obtained by gluing trivial
vector bundles U×C, V ×C over U ∩ V by identifying (u, ξ) ∈ U×C and (v, ζ) ∈ V ×C if
uv = 1 and ζ = un ·ξ. So for U1 = U and Us = V one has φ12 (u) = un , U ∈ U ∩ V ⊆U.
Let Ln be the sheaf of holomorphic sections of Ln .
def
3.5.6. Lemma. (a) Γ(P1 , Ln ) ∼
= Cn [x, y] = homogeneous polynomials of degree n. So, it
is zero if n < 0 and for n ≥ 0 the dimension is n + 1.
(b) H 1 (P1 , U; Ln ) ∼
= ?.
3.5.7. Remark. For distinct points P1 , ..., Pn on P1 , and integers Di , define the sheaf
P def
L = O( Di Pi ) by L(U)
P= “all ∼ holomorphic functions f on U − {P1 , ..., Pn }, such that
ordPi f ≥ −Di . Then O( Di P i ) = L P D i .
3.6. Geometric representation theory. SL2 acts on C2 and therefore on (i) O(C2 ) =
C[x, y], and (ii) each Cn [x, y]; on (iii) P1 = lines in C2 , and less obviously on (iv) each Ln ,
hence also on (v) each H • (P1 , Ln ). In fact,
3.6.1. Lemma. Cn [x, y] = Γ(P1 , Ln ), n = 0, 1, 2, .. is the list of

(1) all irreducible finite dimensional holomorphic representations of SL2 .
(2) all irreducible finite dimensional representations of SU(2) on complex vector
spaces.
3.6.2. Borel-Weil-Bott theorem. For each semisimple (reductive) complex group G there
is a space B (the flag variety of G) such that all irreducible finite dimensional holomorphic
representations of G are obtained as global sections of all line bundles on B.
3.7. Cohomology of the constant sheaf is dual to homology.
Lemma. For X = S 1 , Ȟ ∗ (X, U; kX ) is dual to H∗ (X, k).

28
4. D-modules
A “D-module” on a space X means a module for the ring DX of differential operators (or
a sheaf of rings DX ).
4.1. Intro.
4.1.1. (i) The origin of the theory is the idea of Sato to study a function f on X, by
algebraic study of a DX -module Mf associated to f . (ii) Similarly, a system (∗) of linear
differential equations on X, has an algebraic reformulation – a module M∗ for the ring of
differential operators.
These two ideas may be viewed as parallel to the classical ideas: (i) algebraic number
theory studies numbers via the algebraic equations they satisfy, (ii) once we observe that
the choice of coordinates can be misleading, we pass from the study of systems of linear
equations to linear algebra on vector spaces.
4.1.2. We will use D-modules as an illustration of usefulness of homological algebra:
(1) Riemann-Hilbert correspondence. The notion of a solution of a system (∗)

of homogeneous linear differential equations, can be formulated in terms of an
associated D-module M∗ . This algebraic setting offers an extended meaning of
solutions: the solutions of differential equations with singularities form a complex
(rather then a single vector space). The higher cohomologies of this complex can
be called “higher solutions” of (∗).
The strength of this is illustrated by the observation that on the level of com-
plexes (i.e., when one uses the language of derived categories), the process of pass-
ing from differential equations to solutions gives an identification (the Riemann-
Hilbert correspondence), of two realms:
(a) differential equations with regular singularities,
(b) constructible sheaves (a topological notion).
(2) Functoriality. As a tool for differential equations, we want the ability
(a) to integrate a given system (∗) in certain directions,
(b) to restrict (∗) to submanifolds,
(c) to glue the information from various restrictions into information on the the
original system (∗).
This amounts, abstractly speaking, to the following “functoriality” of differential
equations (or D-modules):
• when two spaces are related by a map π : X − → Y we want to be able move
differential equations from one to another.
This is well understood in the algebro-geometric setting of D-modules, and again,
it is most efficient when performed on the level of derived categories.
29
In the exposition bellow we will intertwine (1) and (2), and use the RH-correspondence as
an additional motivation for functoriality. Moreover, we will need to notice that sheaves
have the same kind of functoriality as D-modules.
4.2. D-modules and differential equations.
4.2.1. The algebra DAn . Let X be the affine space An over C. Let DX the ring of linear
differential operators with polynomial coefficients, hence DX = ⊕I,J CxI ∂ J and
4.2.2. Lemma. DX is the C-algebra generated by xi ’s,∂i ’s and the relations:
(i) xi ’s commute, (ii) ∂i ’s commute, (iii) [∂i , xj ] = δij .
4.2.3. Natural modules for DX . These are various function spaces:

def
• (i) for U open in Rn : C ∞ (U), D(U) = the distributions on U, hyper-functions on
U, etc;
• (ii) for V open in Cn : H(V ) = holomorphic functions on V ,
• (iii) for W Zariski-open in Cn (i.e., the complement of some affine algebraic sub-
def
variety): O(W ) = polynomial functions on W .
Observe that these examples of spaces of functions actually form sheaves of DX -modules:
restriction of functions is a map of DX -modules.
4.2.4. A system of linear differential equations with polynomial coefficients. Such system
q
X
(∗) dij yi = 0, 1 ≤ j ≤ p;
i=1
of p equations in q unknowns, is given by a matrix d = (dij ) ∈ Mpq (DX ). One can consider
solutions in any D-module F :
q
X
q
Sol∗ (F ) = {f ∈ F , dij fi = 0, 1 ≤ j ≤ p}.
i=1
Moreover, if F is a sheaf, then we can define a sheaf of solutions Sol∗ (F ) – its sections
on U⊆X are the solutions on U:
def
(Sol∗ (F ))(U) = Sol∗ (F (U)).
30
4.2.5. The D-module associated to a system (∗). This passage (∗)7→ M∗ , from linear differ-
ential equations to D-modules, is the same as the passage from systems of linear equations
to linear algebra (you loose the coordinates). Having q unknowns yi , translates into a free
DX -module F = (DX )q = ⊕q1 DX ·Ei , where we denote its canonical DX -basis by Ei .
def
Imposing the equations (∗) translates into passing to the quotient M∗ = F/R by the
def P p Pq
relations submodule R = 1 DX ·( i=1 dij Ei ) ⊆ F .
P
Denote by Fi be image of Ei in the quotient M∗ . These elements generate
P M ∗ = DX ·Fi .
Observe that Fi ’s form a “formal” solution of (∗), i.e., one has qi=1 dij Fi = 0 in M∗ .
This is actually the “universal” solution:
4.2.6. Lemma. For any D-module F :

Sol∗ (F ) ∼
= HomDX (M, F ).
4.2.7. Example. For (∗) given by y ′ = 0, i.e., ∂y = 0, one has F = DX and R = DX ·∂⊆F ,
∼
=
hence M = DX /DX ·∂ ← −C[x]. Now,
Sol∗ (C ∞ (R)) = constant f unctions on R = HomDX (M, C ∞ (R)).
4.3. Higher solutions. Singularities of differential equations will affect solutions in two
ways:
• (i) there may be local solutions but no global solutions (due to “monodromy”, i.e.,
a non-consistent behavior of solutions around a singularity),
• (ii) there may be no ordinary solutions but there may be higher (“hidden”) solu-
tions.
def
For (i), we saw that the equation zy ′ = λy has a singularity ( = bad behavior), at 0 (for
λ 6∈ Z): on a disc D that does not contain 0, solutions form a 1-dimensional vector space,
but if D ∋ 0 then there are no solutions. The simplest example of (ii) is
4.3.1. δ-function. We consider the simplest singular differential equations zy = 0. Its

solutions in distributions are the multiples of the δ-function. If we view the problem in
the holomorphic functions H(A1 ), then there are no solutions though morally there should
be (the δ-function). To find these “hidden” solution we introduce the
4.3.2. The derived notion of solutions. We will derive the notion of solution using the
algebraic formulation Sol∗ (F ) ∼
= HomDX (M∗ , F ). Our derived notion of solutions is
def
LSol∗ (F ) = HomDX (P • , F )
for a free resolution P • of M∗ .
31
∼
= R
z
4.3.3. δ-function. Here M∗ = DX /z·DX ← −C[∂] has a resolution · · · −→0 − → DX −→
q Rz
DX − → M∗ − →0 − → · · ·, so P • = (· · · −
→0 −
→ DX −→ DX −→0 − → · · ·) is a free resolution
−1 0
and therefore,
def z
LSol∗ (F ) = Hom(P •, F ) = (· · · − →HomDX (DX , F ) −
→0 − → HomDX (DX , F ) −
→0 −
→· · ·)
0 1
∼ z
= (· · · −
→0 −
→F −
→F −
→0 −
→· · ·).
0 1
0 z
→ F ) and H 1 [LSol∗ (F )] = F /zF .
Therefore, H [LSol∗ (F )] = Ker(F −
So, if F = H(A1 ), then H 0 = Ker(z) is zero, and the the evaluation at zero map H 1 =
ev0
H(A1 )/z·H(A1 ) −→ C is an isomorphism. So, there are no solutions, but there is a higher
solution. The whole complex LSol∗ (H(A1 )) is quasi-isomorphic to H(A1 )/zH(A1 )[−1] ∼ =
C[−1], via the quasi-isomorphism
L
(· · · −−−→ 0 −−−→ H(A1 ) −−−z→ H(A1 ) −−−→ 0 −−−→ · · ·)
0 1
     
      .
y y y ev0 y y y
(· · · −−−→ 0 −−−→ 0 −−−→ C −−−→ 0 −−−→ · · ·)
We can say that the δ-function appears in the holomorphic picture in degree 1 (the
codimension of the point!). In contrast, in the distribution picture it appears in degree
z
zero: for F = D(R) one has H 0 [LSol∗ (F )] = Ker(F − → F ) = Cδ0 and H 1 [LSol∗ (F )] =
F /zF = 0.
4.4. Riemann-Hilbert correspondence: differential equations are the same as

solutions. Let X be a smooth complex algebraic variety. On the category of DX -modules
def
we will consider the functor Sol(M) = HomDX (M, HX ) of solutions in the sheaf of holo-
morphic functions HX , and the corresponding derived construction of complexes of solu-
tion sheaves
def
LSol(M) = Sol(P • ) = HomDX (P •, HX ),
which is defined by replacing M with a projective resolution P • .
4.4.1. Theorem. The derived construction of solutions in the sheaf of holomorphic func-
tions LSol gives a (contravariant) equivalence of derived categories
LSol
D b [mRS (DX )] −−→ D b [Sheavesc (X)].
This requires some explanation, or maybe a lot.
4.5. Differential equations (or D-modules) with Regular Singularities. Notation

“RS” will mean “with regular singularities”, and we now explain which D-modules are
said to have regular singularities.
32
4.5.1. Regular singular differential equations on curves. Let X be open in C and consider
a differential equation (∗) on X
Y (n) + A1 (z)Y (n−1) + · · · + An−1 (z)Y ′ + A0 (z)Y = 0;

with Y ∈ H(X, Cn ) = H(X)n (a column vector) and the (matrix valued) coefficient
functions Ai ∈ H(X, Mn ) = Mn (H(X)).
Consider a point a ∈ C − X such that X contains some punctured disc D ∗ = D − {a}
around a. Choose a local coordinate z on D centered at a. The coefficients of the equation
need not be holomorphic at a, i.e., “(∗) may have a singularity at a”. We say that (∗)
has a regular singularity at a if the equation can be rewritten in terms of the coordinate z
and ∂ = ∂/∂z near a, so that it is of the type dy = 0 where d can be written in the form
d = (z∂)n + B1 (z)(z∂)n−1 + · · · + Bn−1 (z)(z∂) + Bn (z)
with rational (matrix valued) functions Bi .
Here, the appearance of z in front of ∂ causes a singularity (remember the example
d = z∂ − λ). Regular means that we do not allow z 2 in front of ∂!
4.5.2. Example. Solve (z n ∂ − 1)y = 0 for n = 0, 1, 2, 3; and compare how complicated are
the solutions.
4.5.3. Frobenius method. There is a satisfactory method for solving differential equations
with RS. It usually appears as the first example of usefulness of linear algebra (the only
simpler example I can think of is the Frobenius method for difference equations).
4.5.4. D-modules on curves. Via the relation between D-modules and differential equa-
tions, we will say that a D-module on X has RS if it comes from a differential equation
with RS.
4.5.5. Algebraic varieties. For a general X, a DX -module is said to have RS if its restric-
tion to any algebraic curve in X has RS.
4.5.6. Remarks. (1) Differential equations with regular singularities are omnipresent in
mathematics (Gauss-Manin connections and the Picard-Fuchs equations in algebraic ge-
ometry, differential equations that define hypergeometric functions and all classes of spe-
cial functions,...). However, the equations of mathematical physics are usually irregular,
and the irregular ones are less understood already in dimension 2.
(2) The general definition of RS uses the idea of restricting a D-module. This is a part of
the functoriality property of D-modules:
33
4.6. Functoriality of D-modules. This means the correspondences of (complexes of)

π
D-modules on spaces X and Y , that come from a map X − → Y . Here we describe the
functoriality explicitly when X and Y are affine spaces. Next, in 4.7 we list the additional
ingredients needed when X and Y are arbitrary smooth algebraic varieties.
π
4.6.1. Affine spaces. Let X = An − → Y = Am . Then there are two direct image (push-
π! ,π∗
forward) constructions m(DX ) −−→ m(DY ), and two inverse image (pull-back) construc-
π ! ,π ∗
tions m(DX ) ←−−− m(DY ). Among the properties are
(1) (σ◦τ )? = σ? ◦τ? and (σ◦τ )? = τ ? ◦σ ? for ? ∈ {!, ∗}.
(2) These functors are best used in the larger world of complexes.
The graph of π gives a factorization of π into a composition of an embedding and a
projection
i:x7→(x,π(x)) q=pr2
X −−−−−−−→ (X×Y ) −−−→ Y,
so by (1) it suffices to explain π? and π ? when π is an embedding or a projection.
4.6.2. Embeddings. Let X = An and Z = Ap have coordinates xi and zj , so that on

Y = X×Z = An+p one has xi , zj ’s. Let i : X֒→Y, a7→(a, 0). (If i(a) = (a, π(a)) we can
change the coordinates to xi ’s and zj = yj − πj (x1 , ..., xn ), to be in the above situation.)
Now
∂ J
Lemma. (a) Define both i∗ (M) and i! (M) as ⊕J ( ∂z ) M ∼
= C[ ∂z
∂
]⊗C M. These are
! i =i∗
functors m(DX ) −−→ m(DY ).
def P io
(b) io (N) = N/ yj N defines a functor m(DX ) ←
− m(DY ).
4.6.3. Projections. Let X = An and Z = Am have coordinates xi and zj , so that on

Y = Z×X one has xi , zj ’s. Let q : Y ։X, (a, b)7→b be the second projection.
def P q∗
Lemma. (a) q∗ (N) = = N/ ∂j N defines a functor m(DX ) ←
− m(DY ).
q o
def
(b) q o M = ⊕J z J M ∼
= C[z1 , ..., zp ]⊗C M, defines a functor m(DX ) −
→ m(DY ).
4.6.4. What does the functoriality mean for functions or differential equations? Let us
look for justification of the above definitions on grounds of algebra or of our experience
with functions/differential equations.
• Growth from a smaller to a larger space:
– i∗ . The formula comes from the idea that we can already act on M by xi ,
∂xi , and also by any function f on Y (we just need to first restrict it to X),
so we have to add the action by ∂zj ’s, and this we do by “formally” adding
the derivatives normal to X in Y .
34
– q o . Similarly, we can already act on M by xi , ∂xi , and also it is natural that

∂zj ’s should act by 0 on M since the q-pull-back means “extending M to Y
in a way constant along the fibers”. So we only need to “formally” add the
functions zj on the fibers of q.
• Reduction from a larger space to a smaller one:
– q∗ . The natural ways of producing from a function φ on Y a function
Rq! φ on X boil down to taking average over the fibers: (q! φ)(x1 , ..., xn ) =
φ(x1R, ..., xn , z1 , ..., zp ) dz1 · · ·dzp . Let us look only at q : R − → pt, so that
q! φ = R φ(z) dz, in order that it makes sense we need to be in the world
of (something like) the compactly supported functions Cc∞ (R). Then the in-
tegral q! kills derivatives: q! (∂φ) = φ(b) − φ(a) = 0 for a sufficiently P large
interval (a, b). In honor of this fact we have defined q∗ N as N/ ∂zj N, i..e.,
the quotient obtained by killing all fiberwise derivatives in N.
– io . The restriction io N of N to X should kill Pall zj ’s since the restriction of
these functions to X is 0, this motivates N/ ∂zj N.
4.6.5. Derived functoriality. It turns out that the “growth” functors above are exact (see
7.1), so we are only interested in deriving the “reduction” functors. We use left derived
functors, i.e., we replace module N by its projective (usually free) resolution P • . So,
def def
Lq∗ (N) = q∗ P • and Lio (N) = io P •.
4.7. D-modules on a smooth algebraic variety X. When working with arbitrary

smooth algebraic varieties one needs (i) to replace the ring of differential operators with
the sheaf of differential operators, (ii) use derived categories of D-modules.
On an algebraic variety X there may not be enough global differential operators (for
instance a curve of genus > 1), so one is forced to look at local ones, i.e., at the sheaf DX
of differential operators on X. (Remember how few global holomorphic functions there
are on X = P1 !) A D-module on X will now mean a DX -module.
4.7.1. Sheafifying algebra. We define the notions of (i) a sheaf of rings A on X, (ii) a
module M for a sheaf of rings A, (iii) the category m(A) of A-modules.
4.7.2. Sheaf of rings DX . Let X be an algebraic variety, i.e., locally X can be identified
with an affine algebraic variety (i.e., locally X is the set of solutions of several polynomial
equations in some Cn ), and assume that X is smooth (i.e., locally looks like an open part
of some Cm ). Then we know what “differential operators with polynomial coefficients”
def
means in local coordinates. So, for any U Zariski-open in X we can define DX (U) = DU .
π
4.7.3. Functoriality of D-modules. Let X −
→ Y be a map of smooth algebraic varieties.
When working with general algebraic varieties one is forced to use derived categories! So
now the statement is:
35
Theorem. There are two direct image (push-forward) constructions

π ,π∗
D b [m(DX )] −−→
!
D b [m(DY )],
and two inverse image (pull-back) constructions
π ! ,π ∗
D b [m(DX )] ←−−− D b [m(DY )].
4.8. Local systems. Now we turn to the structure of solutions of differential equations.
The solutions of (“nice”) differential equations are complexes of constructible sheaves. A
sheaf is said to be constructible if it can be obtained by gluing together “local systems”.
We will see that local systems have an algebraic description as representations of groups.
4.8.1. Definition. A sheaf L on a topological space X is said to be a local system if it is

locally constant, i.e., X can be covered by open sets U such that the restriction L|U is
isomorphic to some constant sheaf on U.
4.8.2. Example. The basic examples of local systems are the sheaves Sλ of solutions of
(z∂ − λ)y = 0 on X = C∗ . Locally (say on each disk in X), they are constant sheaves
– but not globally. For that observe that following a section around the origin gives the
µ-multiple of the section, where the number µ = e2πiλ is called the monodromy of the
local system Sλ (on the positively oriented loop around the origin).
We will now restate this formally in the general case. In general monodromy will not be
just a number µ ∈ C∗ , but a representation of the so called fundamental group of X.
4.8.3. Monodromy representation attached to a local system. Let L be a local system (of
vector spaces) on X and fix a point b ∈ X. Since L is locally isomorphic to a constant
sheaf, there is a neighborhood D of a such that there is an isomorphism of L|D with a
constant sheaf VD on D for some vector space V . We assume that X is locally connected,
so that one can choose D connected.
Let γ be an oriented loop in X starting and ending at b. We cover γ with some connected
open sets Di , 0 ≤ i ≤ n, so that
(1) D0 = D = Dn
(2) Di ∩ Di−1 is non-empty and connected,
(3) sequence D0 , ..., Dn follows the direction of γ.
This gives an operator (monodromy operator along the sequence of Di ’s) µγ,D0 ,...,Dn on
L(D) defined by: Any s ∈ L(D) extends in a unique way to a sequence si ∈ L(Di ) so that
s0 = s and si = si−1 on Di ∩ Di−1 . Now, the action of the monodromy is the evolution of
the section along the loop γ (or along the sequence D0 , ..., Dn ):
def
µγ,D0 ,...,Dn s = sn .
Now one observes that
36
(1) µγ,D0 ,...,Dn = µγ , i.e., it depends only on µ and not on the choice of Di ’s.
(2) Actually, if γ1 and γ2 are close enough (so that both lie in the same Di ’s), then
µγ1 = µγ2 .
(3) As a consequence we note that if γ1 can be continuously deformed to γ2 then
µγ1 = µγ2 (we can pass from one to another through a sequence of close enough
curves). Therefore,
(4) µγ only depends on the homotopy class [γ] of the loop γ (we say that γ1 and γ2 are
homotopic if γ1 can be continuously deformed to γ2 , this is an equivalence relation
on loops through b).
(5) The set π1 (X, b) of all homotopy classes of loops through b is a group: one can
compose and reverse loops up to a deformation.
(6) µ : π1 (X, b) − → GL(L(D)) is a representation of the fundamental group on the
vector space L(D)), i.e.,
• µ[γ] is a linear operator,
• µb = idL(D) for the constant loop b, and
• µ[γ2 ]·[γ1 ] = µ[γ2 ] ◦µ[γ1 ] .
One can check that L can be reconstructed from the representation µ, actually
4.8.4. Lemma. Monodromy gives a bijection of isomorphism classes of local systems on

X and isomorphism classes of representations of the fundamental group.
A more fancy version is (the additional information is that monodromy preserves all
relations between objects):
4.8.5. Lemma. Monodromy gives an equivalence of categories of LocSyst(X) of local

systems on X and Rep(Π1 (X, b)) of representations of the fundamental group.
4.9. Constructible sheaves. “Sheavesc (X)” means “constructible sheaves” on X. A

sheaf A on X is constructible if there is a stratification of X, i.e., a decomposition X =
⊔i∈I Xi into smooth algebraic subvarieties, such that the restriction of A to each stratum
is locally isomorphic to a constant sheaf (i.e., it is a “local system”).
A typical example is the sheaf of solutions of zy ′ = λy on X = C.
4.9.1. Remarks. (1) For 1 = b ∈ XC∗ one has π1 (∗X, b) ∼ = Z since any loop can be
deformed to a loop which runs on the unit circle n times around the origin for some
n ∈ Z. A representation of π1 (∗X, b) ∼
= Z on a vector space V is the same as an invertible
linear operator on V and if dim(V ) = 1 it is the same as a non-zero number.
(2) The definition of constructible sheaves uses the notion of a restriction of a sheaf to a
subvariety. This is a particular case of
37
π
4.10. Functoriality of sheaves. Let X − → Y be a map of topological spaces. Then
π! ,π∗
there are two direct image (push-forward) constructions Sheaves(X) −−→ Sheaves(Y ),
π ! ,π ∗
and two inverse image (pull-back) constructions Sheaves(X) ←−−− Sheaves(Y ). Among
the properties are
(1) (σ◦τ )? = σ? ◦τ? and (σ◦τ )? = τ ? ◦σ ? for ? ∈ {!, ∗}.
(2) These functors are best used in the larger world of complexes.
As usual one becomes familiar with these by studying the particular cases of an embedding
or a projection.
def π
Lemma. (a) π∗ (M) (V ) = M(π −1 V ) defines a functor Sheaves(X) −→
∗
Sheaves(Y ).
def π −1
(b) π −1 (N ) (U) = lim N (V ) defines a functor preSheaves(X) −−→ preSheaves(Y ).
→
V ⊇π(U )
4.10.1. Remarks. (1) Actually, the second construction can be corrected (improved) to
π −1
give a functor Sheaves(X) −−→ Sheaves(Y ). More about it later. It will turn out that
the operation π −1 has a more complicated definition but a much simpler behavior then
π∗ (π −1 is exact while π∗ is only left exact).
i def
(2) For S ⊆X by the restriction of a sheaf A on X to S we mean the sheaf A|S = i−1 A.
38
Part 2. Categories
Our first goal is to restrict a sheaf to a point, this involves the idea of a limit in a category.
Section 5 covers basic categorical tricks, one of which is the appropriate notion of limits.
Next section 6 deals with our main goal. We want to extend the ideas that work in
categories of modules over a ring to categories of sheaves of abelian groups (or sheaves of
modules), and maybe some other categories. This is achieved by a more abstract setting
of abelian categories which gives a good (but not at all most general) framework for
homological algebra.
Finally, in section 8 we check that the category of sheaves of abelian groups on a given
topological space is an abelian category and that homological algebra can be used in this
category (it has enough injectives).
4.10.2. Why categories? The notion of a category is misleadingly elementary. It formal-

izes the idea that we study certain kind of objects (i.e., endowed with some specified
structures) and that it makes sense to go from one such object to another via some-
thing (a “morphism”) that preserves the relevant structures. Since this is indeed what we
usually do, the language of categories is convenient.
However, soon one finds that familiar notions and constructions (such as (i) empty set,
(ii) union of sets, (iii) product of sets, (iv) abelian group, ...) categorify, i.e., have
analogues (and often more then one) in general categories (respectively: (i) initial object,
final object, zero object; (ii) sum of objects or more generally a direct (inductive) limit of
objects; (iii) product of objects or more generally the inverse (projective) limit of objects;
(iv) additive category, abelian category; ...). This enriched language of categories was
recognized as fundamental for describing various complicated phenomena, and the study
of special kinds of categories mushroomed to the level of the study of functions with
various properties in analysis.
5. Categories
5.1. Categories. A category C consists of
(1) a class Ob(C) whose elements are called objects of C,

(2) for any a, b ∈ Ob(C) a set HomC (a, b) whose elements are called morphisms
(“maps”) from a to b,
(3) for any a, b, c ∈ Ob(C) a function HomC (b, c)×HomC (a, b) −
→ HomC (a, c), called
composition,
(4) for any a ∈ Ob(C) an element 1a ∈ HomC (a, a),
such that the composition is associative and 1a is a neutral element for composition.
39
5.1.1. Examples. (1) Categories of sets with additional structures: Sets, Ab, m(k) for a
def
ring k (denoted also Vect(k) if k is a field), Groups, Rings, T op, OrdSets = category
of ordered sets, ... (2) Another kind of examples: any partially ordered set (I, ≤) defines
a category with Ob = I and Hom(a, b) = point (call this point (a, b)) if a ≤ b and
∅ otherwise. (3) Sheaves of sets on X, Sheaves of abelian groups on X, Constructible
sheaves on X, ...
A category C is said to be small if Ob(C) is a set. Instead of a ∈ Ob(C) we will just say
a ∈ C.
5.2. Objects.
5.2.1. Some special objects and maps. We say that i ∈ C is an initial object if for any
a ∈ C set HomC (i, a) has one element. t ∈ C is a terminal object if for any a ∈ C set
HomC (a, t) has one element. We say that z ∈ C is a zero object if it is both initial and
terminal.
A map φ ∈ HomC (a, b) is said to be an isomorphism if it is invertible, i.e., if there is a
ψ ∈ HomC (b, a) such that ...
Lemma. Initial object in C (if it exists) is unique up to a canonical isomorphism, i.e., for
any two initial objects i, j in C there is a canonical isomorphism. (The same for terminal
and zero objects.)
5.2.2. The useful notion of “equality” of objects. This introduces an essential subtlety:
while in a set two elements may be equal or different, in a category the corresponding
notion for two objects has versions: (i) the same, (ii) isomorphic, (iii) isomorphic by a
canonical (given) isomorphism. It turns out that (i) is too restrictive, (ii) is too lax and
(iii) is the most useful. So we will often say that a = b and mean that we have in mind a
specific isomorphism φ : a − →b.
5.2.3. Sums and products.
5.2.4. Products. A product of objects a and b in C is a triple (Π, p, q) where Π ∈ C is

an object while p ∈ HomC (Π, a), q ∈ HomC (Π, b) are maps such that for any x ∈ C the
function
HomC (x, Π) ∋ φ7→ (p◦φ, q◦φ) ∈ HomC (x, a)×HomC (x, b)
is a bijection.
40
Remarks. (0) If C = Sets then the product of sets Π = a×b together with projections p, q
satisfies this property.
(1) While we expect that a product of a and b should be a specific object built from a and
b, this is not what the categorical definition above says. For given a, b there may be many
triples (Π, p, q) satisfying the product property – however it is easy to see that any two
such (Πi , pi , qi ), i = 1, 2; are related by a canonical isomorphism φ : Π1 −
→Φ2 provided by
the the defining property of the product. This is another example of 5.2.2.
(2) We often abuse the language and say that “Π is product of a and b”, what we mean is
that we remember the additional data p, q but are too lazy to mention these. Moreover,
we may even denote Π by a×b suggesting (in general incorrectly) that the product can
be constructed naturally from a and b – what we mean by notation a×b is some object
Π supplied with maps p, q with properties as above (again an example of the idea 5.2.2).
(3) The product of a and b is an example of a standard construction of “an object defined
by a universal property” (in this case the property that a map into a product is the same
as a pair of maps into a and b), or “an object (co)representing a functor” (in this case
def
Π corepresents a contravariant functor C ∋ x7→ F (x) = HomC (x, a)×HomC (x, b) ∈ Sets,
in the sense that the functor F is identified with the functor HomC (−, Π) that one gets
from Π). As we see, “an object defined by a universal property”: (i) is not really one
object but a system of various objects related by (compatible) isomorphisms, (ii) is not
really just one object but an object supplied with some additional data consisting of some
morphisms (such as p, q above).
(4) A product of a and b in C need not exist!
5.2.5. Sums. A sum of objects a and b in C is a triple (Σ, i, j) where Σ ∈ C is an object

while i ∈ HomC (a, Σ), j ∈ HomC (b, Σ) are maps such that for any x ∈ C the function
HomC (Σ, x) ∋ φ7→ (φ◦i, φ◦j) ∈ HomC (a, x)×HomC (b, x)
is a bijection.
Example. In Sets the sums exist and the sum of a and b is the disjoint union a ⊔ b.
5.2.6. Sums and products of families of objects. This is the same as for two objects. A
product of ai ∈ C, i ∈ I is a pair (P, (pi )i∈I )) where pi : P −
→ai gives a bijection
HomC (x, P ) ∋ φ7→ (pi ◦φ)i∈I ∈ Πi∈I HomC (x, ai ).
Q
We use the notation i∈I ai . A sum of ai ∈ C, i ∈ I is a pair (S, (ji )i∈I )) where ji : ai −
→S
gives a bijection
HomC (S, x) ∋ φ7→ (φ◦ji )i∈I ∈ Πi∈I HomC (ai , x).
The notation is ⊔i∈I ai or ⊕i∈I ai .
41
Q
Lemma. For a ring k category m(k) has sums and products. The product i∈I Mi is
(as a set) just the product of sets, so it consists of all families m = (mi )i∈I with mi ∈
P def
Mi , i ∈ I (such family is often written as a possibly
Q infinite sum i∈I mi = (mi )i∈I ).
The sum ⊕i∈I ai happens to be the submodule of i∈I Mi consisting of all finite families
m = (mi )i∈I , i.e.families such that mi = 0 for all but finitely many i ∈ I.
In this way we get familiar with categorical constructions: by checking what they mean
in familiar categories.
5.3. Limits. Consider a system of increasing subsets A0 ⊆ A1 ⊆· · · of a set A, we will

say that its limit lim Ai is the subset ∪i≥0 Ai of A. Similarly, the limit of a decreasing
→
def
sequence of subsets B0 ⊇ B1 ⊆· · · of A, will be the subset lim Bi = ∩i≥0 Bi of A. Now
←
we give precise meaning to this:
5.3.1. Inductive limits. An inductive system of objects of C over a partially ordered set
(I, ≤), consists of objects ai ∈ C, i ∈ I; and maps φji : ai −
→ aj for all i ≤ j in I; such
that φii = 1ai , i ∈ I and φkj ◦φji = φki when i ≤ j ≤ k.
Its limit is a pair (a, (ρi )i∈I ) of a ∈ C and maps ρi : ai −
→a such that
• ρj ◦φji = ρi for i ≤ j, and moreover

• (a, (ρi )i∈I ) is universal with respect to this property in the sense that for any
(a′ , (ρ′i )i∈I ) that satisfies ρ′j ◦φji = ρ′i for i ≤ j, there is a unique map ρ : a −
→ a′
such that ρ′i = ρ◦ρi , i ∈ I.
Informally, we write lim ai = a.

→ I,≤
5.3.2. Example. A system of increasing subsets A0 ⊆ A1 ⊆· · · of A really form an inductive

system in the category Sets (and over N with the standard order) and lim Ai = ∪i≥0 Ai .
→
5.3.3. Projective limits. A projective system of objects of C over a partially ordered set
(I, ≤), consists of objects ai ∈ C, i ∈ I; and maps φij : aj −
→ ai for all i ≤ j in I; such
that φii = 1ai , i ∈ I and φij ◦φjk = φik when i ≤ j ≤ k.
Its limit is a pair (a, (σi )i∈I ) of a ∈ C and maps σi : a −
→ai such that
• φij ◦σj = σi for i ≤ j, and

• (a, (σi )i∈I ) is universal in the sense that for any (a′ , (σi′ )i∈I ) that satisfies φij ◦σj′ =
σi′ for i ≤ j, there is a unique map σ : a′ − →a such that σi′ = σi ◦σ, i ∈ I.
Informally, lim ai = a.
← I,≤
42
Example. A decreasing sequence of subsets B0 ⊇ B1 ⊆· · · of A forms a projective system

and lim Bi = ∩i≥0 Bi of A.
←
In these basic examples we have: lim = growth) and lim = decline, but it can also go the
→ ←
opposite way:
5.3.4. Lemma. (a) Let (I, ≤) be {1, 2, 3, ...} with the order i ≤ j if i divides j. In Ab let
Ai = Q/Z for all i ∈ I, and let φji be the multiplication by j/i when i divides j. This is
an inductive system and lim Ai =?.
→
(b) Let (I, ≤) be N = {0, 1, ...} with the standard order. In Rings let kn = C[x]/xn+1 and
for i ≤ j let φij be the obvious quotient map. This is a projective system and lim ki =?.
←
5.3.5. Limits are functorial. Let IS(I,≤) (C) be the category of inductive systems in the
category C and over a partially ordered set (I, ≤). Objects are the inductive systems
(Ai )i∈I , (φji )i≤j and a map µ from (A′i )i∈I , (φ′ji )i≤j to (A′′i )i∈I , (φ′′ji )i≤j is a system of maps
µi : A′i − → A′′i , i ∈ I, which intertwine the structure maps of the two inductive systems
i.e., for i ≤ j the diagram
µi
A′i −−−→ A′′i
 

φ′ji y φ′′

ji y
µi
A′j −−−→ A′′j
commutes.
lim µi
→
Lemma. If limits of both systems exist then a map µ of systems defines a map lim A′i −−−→
→
lim A′′i .
→
Proof. By definition of lim.

→
5.3.6. Limits in sets, abelian groups, modules and such. In the category Sets one has
inductive and projective limits (i.e., each inductive or projective system has a limit):
5.3.7. Lemma. Let (I, ≤) be a partially ordered set such that for any i, j ∈ I there is
some k ∈ I such that i ≤ k ≥ j.
(a) (Construction of projective limits of sets.) Let (Ai )i∈I and maps (φij : Aj −→Ai )i≤j be
a projective system of sets. Then
Y
lim Ai is the subset of Ai consisting of all families a = (ai )i∈I such that φij aj = ai .]
←
i∈I
b) (Construction of inductive limits of sets.) Let the family of sets (Ai )i∈I and maps
(φji : Ai −
→Aj )i≤j be an inductive system of sets.
43
def
(1) Show that the relation ∼ defined on the disjoint union ⊔i∈I Ai = ∪i∈I Ai ×{i} by
• (a, i) ∼ (b, j) (for a ∈ Ai , b ∈ Aj ), if there is some k ≥ i, j such that “a = b
in Ak ”, i.e., if φki a = φkj b ,
is an equivalence relation.
(2) Show that lim Ai is the quotient [⊔i∈I Ai ]/ ∼ of the disjoint union by the above
→
equivalence relation.
5.3.8. Remarks. (1) The above lemma gives simple descriptions of limits, lim Ai is a
Q ←
subset of the product Ai , and lim Ai is a quotient of a sum (=disjoint union). Even
→
better, lim Ai can be described by
→
• for i ∈ I, any a ∈ Ai defines an element a of lim Ai ,

→
• all elements of lim Ai arise in this way, and
→
• for a ∈ Ai and b ∈ Aj one has a = b iff for some k ∈ I with i ≤ k ≥ j one has
a = b in Ak .
(This is just a simple retelling of the lemma, here a is the image of (a, i) in the quotient
[⊔i∈I Ai ]/ ∼.)
(2) The same existence and description of limits works in many categories such as abelian
groups, k-modules, Groups, etc (these are all categories where objects are sets with some
additional structure). For instance for an inductive system of abelian groups Ai over
(I, ≤), the inductive limit lim Ai always exists, it can be described as in (1), but one has
→
to also explain what is the group structure (addition) on the set [⊔i∈I Ai ]/ ∼. This is
clear, if a ∈ Ai and b ∈ Aj then a + b = φki a + φkj b for any k with i ≤ k ≥ j.
(c) Actually the lemma generalizes to limits in an arbitrary category (see lemma 5.3.12
bellow), but then we have to explicitly ask for the existence of some standard construc-
tions. For instance the category of finite dimensional vector spaces (or even simpler, the
category of finite sets) clearly does not have infinite sums or products, hence can not have
limits.
5.3.9. Stalks of a sheaf. We want to restrict a sheaf F on a topological space X to a point

def
a ∈ X. The restriction F |a is a sheaf on a point, so it just one set Fa = (F |a)({a})
called the stalk of F at a. What should Fa be? It has to be related to all F (U) where
U⊆X is is open and contains a, and F (U) should be closer to Fa when U is a smaller
neighborhood. A formal way to say this is that
• (i) the set Na of neighborhoods of a in X is partially ordered by U ≤ V if V ⊆U,
(ii) the values of F on neighborhoods (F (U))U ∈Na form an inductive system, (iii)
def
we define the stalk by Fa = lim F (U).
→
U ∈Na
44
Lemma. (a) The stalk of a constant sheaf of sets SRn at any point is canonically identified
with the set S.
(b) The stalk of a the sheaf HC of holomorphic functions at the origin is canonically
identified with the ring of convergent power series. (“Convergent” means that the series
converges on some disc around the origin.)
5.3.10. Fibered products as limits. First, notice that sums and products are special cases
of limits:
Lemma. Let ai ∈ C, i ∈ I. If we supply I with the discrete partialQorder (i.e., i ≤ j iff

i = j), then lim ai is the same as ⊕i∈I ai , and lim ai is the same as i∈I ai .
→ ←
p q
Next, given a1 −
→b ← −a2 in C, we say that the fibered product of a1 and a2 over the base
b (denoted a1 ×b a2 ) is the projective limit of the projective system ai , i ∈ I where
• (i) I = {1, 2, k} and ak = b,
• (ii) the only nontrivial inequalities in I are 1 ≥ k ≤ 2.
p q
Lemma. (a) In Sets fibered products exist and the fibered product of A −
→S ←
−B is the
set A×B = {...}.
S
⊆ ⊇
(b) If A −
→S←
− B are inclusions of subsets of S the fibered product A×B = A ∩ B is
S
just the intersection A ∩ B.
5.3.11. Existence and construction of limits. The following lemma gives a criterion for
existence of limits in a category C, and a way to describe them in terms of simpler
constructions. The particular case when C is the category of sets is the lemma 5.3.6 – it is
less abstract since we assume some familiar properties of sets. The general proof is “the
same”.
5.3.12. Lemma. (a) If a category C has

(1) products of families of objects and
σ
(2) any pair of maps µ, ν ∈ HomC (a, b) has an equalizer (i.e., a map e −
→ a universal
among all maps φ into a such that µ◦φ = ν◦φ),
then C has projective limits (and they can be described in terms of products and equal-
izers).
(b) Dually, if a category C has
(1) sums of families of objects and
σ
(2) any pair of maps µ, ν ∈ HomC (a, b) has a coequalizer (i.e., a map b −
→ c universal
among all maps φ from b such that φ◦µ = φ◦ν),
45
then C has inductive limits (and they can be described in terms of sums and coequalizers).
5.4. Categories and sets. Some of the relations:

• All sets form a category Sets which can be viewed as the basic example of a
category.
• Each set S defines a (small) category S with Ob(S) = S and for a, b ∈ S
HomS (a, b) is {1a } if a = b and it is empty otherwise. In the opposite (and
more stupid) direction, each small category C gives a set Ob(C) (we just forget the
morphisms).
• The structure of a category can be viewed as a more advanced version of the
structure of a set.
5.4.1. Question. If all sets form a structure more complicated then a set – a category
Sets, what do all categories form? (All categories form a more complicated structure, a
2-category. Moreover all n-categories form an (n + 1)-category ...)
5.4.2. Operations on categories. The last remark suggests that what we can do with sets,
we should be able to do with categories (though it may get more complicated).
For instance the product of sets lifts to a notion of a product of cate-
gories A and B, The category A×B has Ob(A×B) = Ob(A)×Ob(B) and
def
HomA×B [(a′ , b′ ), (a′′ , b′′ )] = HomA [a′ , a′′ ]×HomB [b′ , b′′ ].
However, since we are dealing with a finer structure there are operations on categories
that do not have analogues in sets. Say the dual (opposite) category of A is the category
Ao with Ob(Ao) = Ob(A), but
def
HomAo (a, b) = HomA (b, a).
This is the formal meaning of the observation that reversing the arrows gives a “duality
operation” for constructions in category theory. For instance, a projective system in A
is the same as an inductive system in Ao , a sum in A is the same as a product in Ao,
etc. This is useful: for any statement we prove for projective systems there is a “dual”
statement for inductive systems which is automatically true.
5.5. Functors. The analogue on the level of categories of a function between two sets is
a functor between two categories.
A functor F from a category A to a category B consists of
• for each object a ∈ A an object F (a) ∈ B,
• for each map α ∈ HomA (a′ , a′′ ) in A a map F (α) ∈ HomB (F a′ , F a′′ )
such that F preserves compositions and units, i.e., F (β◦α) = F (β)◦F (α) and F (1a ) =
1F a .
46
φ
5.5.1. Examples. (1) A functor means some construction, say a map of rings k −
→ h gives
• a pull-back functor φ∗ : m(h) −
→ m(k) where φ∗ N = N as an abelian group, but
now it is considered as module for k via φ.
def
• a push-forward functor φ∗ : m(k) − → m(h) where φ∗ M = h⊗k M. This is called
“change of coefficients”.
To see that these are functors, we need to define them also on maps. So, a map β : N ′ − →
N ′′ in m(h) gives a map φ∗ (β) : φ∗ (N ′ ) −
→ φ∗ (N ′′ ) in m(k) which as a function between
sets is really just β : N ′ −
→ N ′′ . On the other hand, α : M ′ −→ M ′′ in m(k) gives φ∗ (α) :
′
φ∗ (M ) −→φ∗ (M ) in m(h), this is just the map 1h ⊗α : h⊗k M ′ −
′′
→h⊗k M ′′ , c⊗x7→ c⊗α(x).
Here we see a general feature: functors often come in pairs (“adjoint pairs of functors”)
and usually one of them is stupid and the other one an interesting construction.
F
(2) For any category A there is the identity functor 1A : A −
→ A. Two functors A −
→B
G G◦F
and B −→ C can be composed to a functor A −−→ C.
(3) An object a ∈ A defines two functors, HomA (a, −) : A −
→ Sets, and HomA (−, a) :
Ao −→Sets. Moreover, HomA (−, −) is a functor from Ao ×A to sets!
(4) For a ring k, tensoring is a functor −⊗k − : mr (k)×ml (k) −
→Ab.
5.5.2. Contravariant functors. We say that a contravariant functor F from A to B is given

by assigning to any a ∈ A some F (a) ∈ B, and for each map α ∈ HomA (a′ , a′′ ) in A a
map F (α) ∈ HomB (F a′′ , F a′ ) – notice that we have changed the direction of the map so
now we have to require F (β◦α) = F (α)◦F (β) (and F (1a ) = 1F a ).
This is just a way of talking, not a new notion since a contravariant functor F from A to
B is the same as a functor F from A to Bo (or a functor F from Ao to B).
5.5.3. Categorification of “subset”. The categorical analogue of the notion of a subset of

a set is the notion of a (full) subcategory of a category.
A subcategory C ′ of a category C is given by a subclass Ob(C ′ )⊆Ob(C) and for any a, b ∈
Ob(C ′ ) a subset HomC ′ (a, b)⊆HomC (a, b) such that HomC ′ (a, a) ∋ 1a , a ∈ C ′ , and the sets
HomC ′ (a, b), a, b ∈ C ′ are closed under the composition in C.
A full subcategory C ′ of a category C is a subcategory C ′ such that HomC ′ (a, b)⊆HomC (a, b)
for a, b ∈ C ′ .
5.5.4. Examples. (1) F ree(k)⊆m(k). (2) Category C defines subcategory C ∗ where objects
are the same and morphisms are the isomorphisms from C.
5.5.5. Some categorifications of “injection”, “surjection” and “bijection”. The following

properties of a functor F : A֒→B are categorical analogues of the injectivity property of
a function:
47
• F is faithful (or “an embedding of categories”) if all maps HomA (a′ , a′′ ) −→
′ ′′ ′ ′′
HomB (F a , F a ), a , a ∈ A are injective.
• F is called fully faithful (or a full embedding) if all maps HomA (a′ , a′′ ) −
→
′ ′′
HomB (F a , F a ) are bijections.
Analogue of surjectivity:
• F is said to be essentially surjective, if it is surjective on isomorphism classes of
objects, i.e., any b ∈ B is isomorphic to F a for some a ∈ A.
Analogue of a bijection:
• F is said to be an equivalence of categories if it is essentially surjective and fully
faithful.
Here we took a point of view that a bijection is a function which is both injective and
surjective.
5.5.6. Examples. For φ : k − → h consider φ∗ : m(h) −→ m(k). For instance if φ is Z֒→k or

R֒→
C then φ∗ is the forgetful functor m(k) −→ Ab or VectC − → VectR . is faithful but not full.
φ∗ is always faithful but in our examples it is neither full nor essentially surjective.
Example. Let k be a field and Vk the category such that Ob(Vk ) = N = {0, 1, ...} and
Hom(n, m) = Mmn (matrices with m rows and n columns. It is equivalent to the category
ι
Vectfk g of finite dimensional vector spaces by the functor Vk −
→Vectfk g , here ι(n) = kn and
for a matrix α ∈ Mmn , ια : km − →kn is the multiplication by α.
Notice that Vk and Vectfk g seem very different (say only the first one is small), their
content is the same (the linear algebra), but one is more convenient for computation and
the other for thinking. Historically, equivalence ι is roughly the observation that one can
do linear algebra without always choosing coordinates.
5.6. Natural transformations of functors (“morphisms of functors”). A natural

transformation η of a functor F : A − → B into a functor G : A −
→ B consists of maps
′ ′′
ηa ∈ HomB (F a, Ga), a ∈ A such that for any map α : a −→a in A the following diagram
commutes
F (α)
F (a′ ) −−−→ F (a′′ )
 

ηa′ y

ηa′′ y , i.e., ηa′′ ◦F (α) = G(α)◦ηa′ .
G(α)
G(a′ ) −−−→ G(a′′ )
So, η relates values of functors on objects in a way compatible with the values of functors
on maps. In practice, any “natural” choice of maps ηa will have the compatibility property.
48
5.6.1. Example. For the functors φ∗ M = h⊗k M and φ∗ N = N from 5.5.1(1), there are
canonical morphisms of functors
α
α : φ∗ ◦φ∗ −
→1m(h) , φ∗ ◦φ∗ (N) = h⊗k N −→
N
N = 1m(h) (N)
is the action of h on N and
βM
→φ∗ ◦φ∗ ,
β : 1m(k) − φ∗ ◦φ∗ (M) = h⊗k M ←− M = 1m(M) (M)
is the map m7→1h ⊗m.
For any functor F : A −
→ B one has 1F : F −
→ F with (1F )a = 1F a : F a −
→ F a. For three
functors F, G, H from A to B one can compose morphisms µ : F − → G and ν : G −→H to
ν◦µ : F −
→H
5.7. Adjoint functors. This is often the most useful categorical idea.
F G
5.7.1. Useful definition. An adjoint pair of functors is a pair of functors (A −
→ B, B −
→ A)
together with “natural identifications”
∼
=
ζa,b : HomB (F a, b)−
→HomA (a, Gb), a ∈ A, b ∈ B.
Here,
(1) “natural” means behaving naturally in a and b, and by this we mean that ζ is a
natural transform of functors ζ : HomB (F −, −) − → HomA (−, G−) from Ao ×B to
sets.
(2) “identification” means that each function ζa,b is a bijection.
We say that F is the left adjoint of G and that that G is the left adjoint of F (in the
identity of homomorphisms F appears on the left in Hom and G on the right).
5.7.2. Lemma. Functors (φ∗ , φ∗ ) from 5.5.1(1) form an adjoint pair, i.e., there is a canon-
ical identification
ηM,N
Homm(h) (φ∗ M, N) −−−→ Homm(k) (M, φ∗ N), M ∈ m(k), N ∈ m(h).
σ τ ηM,N
If h⊗k M − → N, then Homm(h) (h⊗k M, N) −−−→ Homm(k) (M, N) by
→ N, M −
η(σ)(m) = σ(1⊗m) and η −1 (τ )(c⊗m) = cτ (m), m ∈ M, c ∈ h.
5.7.3. Remark. As in this example, often an adjoint pair appears in the following way:
there is an obvious functor A (so obvious that we usually do not pay it any attention), but
it has an adjoint B which is an interesting construction. The point is that this “interesting
construction” B is intimately tied to the original “stupid” functor A, hence the properties
of B can be deduced from the properties of the original simpler construction A. In fact
B is produced from A in an explicit way as the following lemma shows.
49
α β
5.7.4. What is the relation between morphisms of functors φ∗ ◦φ∗ − → 1m(h) and φ∗ ◦φ∗ ←−
η
→ Homm(k) (−, φ∗ −)
1m(k) , from 5.6.1, and the isomorphism of functors Homm(h) (φ∗ −, −) −
from 5.7.2 ? They are really the same thing, i.e., two equivalent ways to describe adjoint-
ness.
5.7.5. Lemma. (Existence of the right adjoint.)

(a) If F has a right adjoint F then for each b ∈ B the functor
HomB (F −, b) : A −
→Sets, a7→ HomB (F a, b)
is representable (see 5.9).
(b) Suppose that for each b ∈ B the functor HomB (F −, b) : A −
→ Sets is representable.
For each b ∈ B choose a representing object Gb ∈ A, then G is a functor from B to A
and it s the right adjoint of F .
(c) The right adjoint of F , if it exists, is unique up to a canonical isomorphism.
Of course the symmetric claims hold for left adjoints.
5.7.6. Left adjoints of forgetful functors. We say that a functor F is forgetful if it consists
in dropping part of the structure of an object. Bellow we will denote its left adjoint by
G. Standard construction (that add to the structure of an object), are often adjoints of
forgetful functors
→ Sets then G sends set S to the the free k-module k[S] = ⊕s∈S k·s
(1) If F : m(k) −
with a basis S.
(2) Let k be a commutative ring. For F : k − ComAlg − → Set from commutative
k-algebras to sets, G sends a set S to the polynomial ring k[xs , s ∈ S] where
variables are given by all elements of S.
(3) If F : k − ComAlg − → m(k), then for a k-module M, G(M) is the symmetric
algebra S(M). (To get exterior algebras in the same way one needs the notion of
super algebras.)
(4) For the functor F : k − Alg − → m(k) from k-algebras to k-modules, G(M) is the
tensor algebra S(M).
(5) Forgetful functor F : T opSets has a left adjoint D that sends a set S to S with
the discrete topology, and also the right adjoint C such that C(S) is S with the
topology such that only S and φ are open.
5.7.7. Functors between categories of modules. (1) For φ : k − → h, one has adjoint triple
def
(φ∗ , φ , φ⋆ ) with φ⋆ (M) = Homk (h, M), i.e., (φ∗ , φ ) and (φ , φ⋆ ) are adjoint pairs. So φ∗
∗ ∗ ∗
has both a left and a right adjoint and they are very different.
def
(2) (k, h)-bimodule X gives X∗ : m(h) −
→ m(k), with X∗ (N) = X⊗h N, what is its right
adjoint?
50
5.7.8. More categorifications of “injection”, “surjection”, “bijection”.
5.8. Higher categories. Notice that

def
(1) The class Cat of all categories has a category like structure with Ob(Cat) = cat-
def def
egories, and (for any two categories A and B) HomCat (A, B) = F unct(A, B) =
functors from A to B. However, the class HomCat (A, B) need not be a set unless
the categories A and B are small.
(2) Moreover, for any two categories A and B, HomCat (A, B) = F unct(A, B) is ac-
tually a category with Ob(F unct(A, B)) = functors from A to B, and for F, G ∈
def
F unct(A, B), the morphisms HomF unct(A,B) (F, G) = natural transforms from F
to G.
The two structures (1) and (2) (together with natural compatibility conditions between
them) make Cat into what is called a 2-category. We leave this notion vague as it is not
central to what we do now. (We will see that complexes and topological spaces are also
2-categories.)
Geometrically, a category A defines a 1-dimensional simplicial complex |A| (the nerve
of A) where vertices=Ob(A) and (directed) edges between vertices a and b are given by
HomC (a, b). A 2-category B defines a 2-dimensional topological object |B| (the nerve of
B). If say, B = Cat then vertices Ob(Cat) = categories, edges between vertices A and B
F,G
corresponds to all functors from A to B, and for two functors A −−→ B (i.e., two edges
η
from the point A to the point B), morphisms F − → G correspond to (directed) 2-cells in
|Cat| whose boundary is the union of edges corresponding to F and G. So one dimensional
topology controls the level of our thinking when we use categories, 2-dimensional when we
use 2-categories and in this way one can continue to define more complicated frameworks
for thinking of mathematics, modeled on more complicated topology: the nerve of an
n-category is an n-dimensional topological object for n = 0, 1, 2, ..., ∞ (here “1-category”
means just “category” and a 0-category is a set).
5.9. Construction (description) of objects via representable functors. Yoneda

lemma bellow says that passing from an object a ∈ A to the corresponding functor
HomA (−, a) does not loose any information – a can be recovered from the functor
HomA (−, a). This has the following applications:
(1) One can describe an object a by describing the corresponding functor HomA (−, a).
This turns out to be the most natural description of a.
(2) One can start with a functor F : Ao − →Sets and ask whether it comes from some
objects of a. (Then we say that a represents F and that F is representable).
(3) Functors F : Ao − → Sets behave somewhat alike the objects of A, and we can
think of their totality as a natural enlargement of A (like one completes Q to R).
51
b To a category A one can associate a category Ab def

5.9.1. Category A. = F unct(Ao , Sets)
of contravariant functors from A to sets. Recall that each object a ∈ A defines a functor
ιa = HomA (−, a) ∈ A. b The following statement essentially says that one can recover a
form the functor HomA (−, a), i.e., that this functor contains all information about a.
5.9.2. Theorem. (Yoneda lemma)

(a) Construction ι is a functor and it is a full embedding ι : A − b
→ A.
(b) Functor HomA (−, a) determines a up to a unique isomorphism.
5.9.3. Representable functors. We will say that a functor F : Ao − →Sets is representable

if there is some a ∈ A and an isomorphism of functors η : HomA (−, a) − → F . Then we
say that a represents F . This is the basic categorical trick used to describe an object a up
to a canonical isomorphism: instead of describing a we describe a functor F isomorphic
to HomA (−, a).
For instance a product of a and b is an object that represents the functor
A ∋ x7→ Hom(x, a)×HomA (x, b) ∈ Sets.
More generally, a projective system ai , i ∈ I, in A defines a functor lim ai ∈ Ab by
←
def
(lim ai )(x) = lim HomA (x, ai ).
← ←
Here, the limit on the right is in sets and we know that the category Sets has limits!
(there is no limit on the LHS – this is just the notation for a new functor). From definition
one finds that lim ai is representable iff lim ai exists (and then the limit represents the
← ←
functor).
Dually, a functor G : A −
→ Sets is said to be corepresentable if there is some a ∈ A and
an isomorphism of functors η : HomA (a, −) −→G. Then we say that a corepresents F . A
sum of a and b is an object that corepresents the functor
A ∋ x7→ Hom(a, x)×HomA (b, x) ∈ Sets.
An inductive system ai , i ∈ I, in A defines a functor lim ai : A −
→Sets by
→
def
(lim ai )(x) = lim HomA (ai , x)
→ ←
which is corepresentable by lim ai (iff it exists).

→
So, we say that we represent contravariant functors from A to Sets, and that we we
corepresent covariant functors from A to Sets. One can also omit “co” and talk about
representing both times (there is no confusion since the nature of functors distinguishes
the two situations).
52
5.10. Completion of a category A to A. b This is a categorical analogue of one of

the basic tricks in analysis: since among functions one can not find beauties like the δ-
functions, we extend the notion of of functions by adding distributions. Remember that
the distributions on an open U⊆Rn are the (nice) linear functionals on the vector space
of of (nice) functions: D(U, C)⊆ Cc∞ (U, C)∗ = HomC [Cc∞ (U), C].
The translation of this procedure in the categorical setting is the observation that the
category Ab may contain many beauties that should morally be in A. One example will
b In particular we will see inductive systems
be a way of treating inductive systems in A.
of infinitesimal geometric objects that underlie the differential calculus.
5.10.1. Limits. For instance, a limit of an inductive system in A need not exist in A but
it always exists in a larger category A.b An inductive system ai , i ∈ I, in A defines an
object in A if the limit lim ai exists, however it always defines a functor lim ai ∈ Ab as
→ →
above. This allows us to think of the inductive system ai , i ∈ I, in A, as if it were an
object lim ai in A – this object would be characterized by the property that the maps
→
into it are systems of compatible maps into ai ’s:
HomA (x, lim ai ) = lim HomA (x, ai ), x ∈ A,
→ →
and by Yoneda lemma it would really characterize the object. For this reason an inductive
b
system in A is called an ind-object of A (while it really gives an object of A).
Similarly one calls projective systems pro-objects of A. However, the completion Ab is a
good setting only for inductive systems – projective ones are related to covariant functors
from A to sets as we will see in the following example.
5.10.2. Affine schemes. The geometry we use here is the algebraic geometry. Its geometric
objects are called schemes and they are obtained by gluing today schemes of a somewhat
special type, which are called affine schemes (like manifolds are all obtained by gluing
open pieces of Rn ’s). An affine schemes S over C is determined by its algebra of functions
O(S) which is a C-algebra. Moreover, any commutative C-algebra A is the algebra of
functions on some scheme – the scheme is called the spectrum of A and denoted Spec(A).
So, affine algebras are really the same as commutative C-algebras, except that a map
φ φ∗
of affine schemes X − → Y defines a map of functions O(Y ) −→ O(X) in the opposite
direction (the pull-back φ∗ (f ) = f ◦φ). The statement “information contained in two
kinds of objects is the same but the directions reverse when one passes from geometry to
algebra” is categorically stated as categories SchemesC and (ComAlgC )o are equivalent.
We will simplify this kind of thinking and define the category of affine schemes over C as
the the opposite of the category of commutative C-algebras. The part that I skipped is
how one develops a geometric point of view on affine schemes defined in this way. Basically
one thinks geometrically and translates geometric ideas in algebra, and proves geometric
53
theorems in algebra. Once sufficiently many geometric statements are verified in this way
one can build up on these and do everything in geometry.
5.10.3. Schemes over C (or over any commutative ring k). Let us repeat the above an-
nounced steps. The category Af f SchemesC of affine C-schemes is defined as ComAlgCo .
We denote the canonical contravariant functor ComAlgCo − → Af f SchemesC , by Spec,
and in the opposite direction one has the operation of taking functions on a scheme
O
Af f SchemesC −→ ComAlgCo . For an algebra A in we think of Spec(A) ∈ SchemesC as a
geometric object such that A is the algebra of functions on it: O(Spec(A)) = A.
Now we consider some projective systems of algebras and the corresponding inductive
systems of affine schemes.
5.10.4. Some projective systems of algebras. Formal power series C[[x]] is lim C[x]/xn in
←
ComAlgC , however the projective system of commutative C-algebras C[x]/xn , n = 1, 2, ...
is itself something like a commutative C-algebra. It is a pro-object in ComAlgC and it
defines a functor lim C[x]/xn : ComAlgC − →Sets by
←
def
(A) = lim Hom(C[x]/xn , A) = nilpotent elements of A.
←
Equivalently, one can think of the pro-object lim C[x]/xn as a topological C-algebra
←
(C[[[x]], T ) with the topology T which one can describe by: a basis of neighborhoods of
0 is given by all ideals xn C[[x]], n > 0. The reason is that any topological algebra A also
gives an a functor A : ComAlgC − →Sets via
def
ComAlgC ∋ A7→ A(A) = HomT op.Algebras (A, A)
where A is considered as a topological algebra with discrete topology. Now we have a
way of comparing a pro-system and a topological algebra by comparing the corresponding
functors, and we find that the functors are the same
lim C[x]/xn = {nilpotents in −} = (C[[[x]], T ).
←
5.10.5. Some injective systems of geometric objects. The nth infinitesimal neighborhood
INAn1 (0) of the point 0 in the line A1 is the scheme defined by the algebra C[x]/xn+1
def
INAn1 (0) = Spec(C[x]/xn+1 ) hence O(INAn1 (0)) = C[x]/xn+1 .
Now, pro-algebra lim C[x]/xn in ComAlgC , is thought of as the algebra of functions on
←
the ind-object in schemes lim INAn1 (0) which we call the formal neighborhood F NA1 (0)
→
of 0 in A1 :
def def
F NA1 (0) = lim INAn1 (0) and O(F NA1 (0)) = lim O(INAn1 (0)) = lim C[x]/xn+1 .
→ ← ←
54
It turns out that the finite infinitesimal neighborhoods and formal neighborhoods are
precisely what the classics were thinking about when they based calculus on infinitesimally
small quantities – the only problem that obscured the rigorous basis of the differential
calculus on infinitesimals was the lack of the language of schemes.
6. Abelian categories
An abelian category is a category A which has the formal properties of the category Ab,
i.e., we can do in A all computations that one can do in Ab.
6.1. Additive categories. Category A is additive if

• (A0) For any a, b ∈ A, HomA (a, b) has a structure of abelian group such that then
compositions are bilinear.
• (A1) A has a zero object,
• (A2) A has sums of two objects,
• (A3) A has products of two objects,
6.1.1. Lemma. (a) Under the conditions (A0),(A1) one has (A2) ⇔ (A3).
(b) In an additive category a⊕b is canonically the same as a×b,
→B is additive if the maps HomA (a′ , a′′ ) −
For additive categories A, B a functor F : A − →
′ ′′
HomB (F a , F a ) are always morphisms of abelian groups.
def
6.1.2. Examples. (1) m(k), (2) F ree(k), (3) F iltVectk = filtered vector spaces over k.
6.2. (Co)kernels and (co)images. In module categories a map has kernel, cokernel
and image. To incorporate these notions into our project of defining abelian categories
we will find their abstract formulations.
6.2.1. Kernels: Intuition. Our intuition is based on the category of type m(k). For a map
α
of k-modules M − →N
• the kernel Ker(α) is a subobject of M,
• the restriction of α to it is zero,
• and this is the largest subobject with this property
6.2.2. Categorical formulation. Based on this, our general definition (in an additive cat-
α
egory A), of “k is a kernel of the map a −
→ b”, is
σ
• we have a map k − → M from k to M,
• if we follow this map by α the composition is zero,
55
σ
• map k − → M is universal among all such maps, in the sense that
τ
· all maps into a, x −
→ a, which are killed by α,
σ
· factor uniquely through k (i.e., through k −
→ a).
So, all maps from x to a which are killed by α are obtained from σ (by composing
it with some map x − →k). This is the “universality” property of the kernel.
6.2.3. Reformulation in terms of representability of a functor. A compact way to restate

the above definition is:
α
• The kernel of a −
→ b is any object that represents the functor
def
A ∋ x7→ α HomA (x, a) = {γ ∈ HomA (x, a); α◦γ = 0 }.
One should check that this is the same as the original definition.
We denote the kernel by Ker(α), but as usual, remember that
• this is not one specific object – it is only determined up to a canonical isomorphism,
• it is not only an object but a pair of an object and a map into a
.
α
6.2.4. Cokernels. In m(k) the cokernel of M − → N is N/α(M). So N maps into it,
composition with α kills it, and the cokernel is universal among all such objects. When
stated in categorical terms we see that we are interested in the functor
def
x7→HomA (b, x)α = {τ ∈ HomA (b, x); τ ◦α = 0 },
and the formal definition is symmetric to the definition of a kernel:
• The cokernel of f is any object that represents the functor A ∋ x7→ HomA (b, x)α .
So this object Coker(α) is supplied with a map b −
→ Coker(α) which is universal among
maps from b that kill α.
6.2.5. Images and coimages. In order to define the image of α we need to use kernels
and cokernels. In m(k), Im(α) is a subobject of N which is the kernel of N − → α(M).
We will see that the categorical translation obviously has a symmetrical version which
we call coimage. Back in m(k) the coimage is M/Ker(α), hence there is a canonical map
Coim(α) = M/Ker(α) − → Im(α), and it is an isomorphism. This observation will be the
final ingredient in the definition of abelian categories. Now we define
def
• Assume that α has cokernel b −→ Coker(α), the image of α is Im(α) = Ker[b −
→
Coker(α)] (if it exists).
def
• Assume that α has kernel Ker(α) −
→a, the coimage of α is Coim(α) = Coker[Ker(α) −
→
a]. (if it exists).
56
6.2.6. Lemma. If α has image and coimage, there is a canonical map Coim(α) −
→Im(α),
and it appears in a canonical factorization of α into a composition
a−
→Coim(α) −
→Im(α) −
→b.
6.2.7. Examples. (1) In m(k) the categorical notions of a (co)kernel and image have the
usual meaning, and coimages coincide with images.
(2) In F ree(k) kernels and cokernels need not exist.
def
(3) In F V = F iltVectk for φ ∈ HomF V (M∗ , N∗ ) (i.e., φ : M −
→ N such that
φ(Mk )⊆Nk , k ∈ Z), one has
• KerF V (φ) = KerVect(φ) with the induced filtration KerF V (φ)n = KerVect(φ) ∩ Mn ,
• CokerF V (φ) = N/φ(M) with the induced filtration CokerF V (φ)n =
image of Nn in N/φ(M) = [Nn + φ(M)]/φ(M) ∼ = Nn /φ(M) ∩ Nn .
• CoimF V (φ) = M/Ker(φ) with the induced filtration CoimF V (φ)n =
image of Mn in M/Ker(φ) = Mn + Ker(φ)/Ker(φ) ∼ = = Mn /Mn ∩ Ker(φ),
• ImF V (φ) = ImVect(φ)⊆N, with the induced filtration ImF V (φ)n = ImVect(φ) ∩
Nn .
Observe that the canonical map CoimF V (φ) − → ImF V (φ) is an isomorphism of vector
spaces M/Ker(φ) − → ImVect(φ), however the two spaces have filtrations induced from
filtrations on M and N respectively, and these need not coincide.
For instance one may have M and N be two filtrations on the same space V , if Mk ⊆Nk
then φ = 1V is a map of filtered spaces M − →N and Ker = 0Coker so that CoimF V (φ) =
M and ImF V (φ) = N and the map CoimF V (φ) − →ImF V (φ) is the same as φ, but φ is an
isomorphism iff the filtrations coincide: Mk = Nk .
6.3. Abelian categories. Category A is abelian if

• (A0-3) It is additive,
• It has kernels and cokernels (hence in particular it has images and coimages!),
• The canonical maps Coim(φ) − →Im(φ) are isomorphisms
6.3.1. Examples. Some of the following are abelian categories: (1) m(k) including Ab =
m(Z). (2) mf g (k) if k is noetherian. (3) F ree(k)⊆ Proj(k)⊆ m(k). (4) C • (A). (5)
Filtered vector spaces.
6.4. Abelian categories and categories of modules.
6.4.1. Exact sequences in abelian categories. Once we have the notion of kernel and cok-
ernel (hence also of image), we can carry over from module categories m(k) to general
abelian categories our homological train of thought. For instance we say that
57
• a map i : a − → b makes a into a subobject of b if Ker(i) = 0 (we denote it a֒→b

or even informally by a⊆b, one also says that i is a monomorphism or informally
that it is an inclusion),
• a map q : b − → c makes c into a quotient of b if Coker(q) = 0 (we denote it b։c
and say that q is an epimorphism or informally that q is surjective),
i def
• the quotient of b by a subobject a −
→b is b/a = Coker(i),
dn
• a complex in A is a sequence of maps · · ·An −→ An+1 −
→· · · such that dn+1 ◦dn = 0,
its cocycles, coboundaries and cohomologies are defined by B n = Im(dn ) is a
subobject of Z n = Ker(dn ) and H n = Z n /B n ;
µ ν
• sequence of maps a − →b−→ c is exact (at b) if ν◦µ = 0 and the canonical map
Im(µ) − →Ker(ν) is an isomorphism.
Now with all these definitions we are in a familiar world, i.e., they work as we expect.
α β
For instance, sequence 0 −→ a′ −→a− → a′′ −→ 0 is exact iff a′ is a subobject of a and a′′ is
′
the quotient of a by a , and if this is true then
Ker(α) = 0, Ker(β) = a′ , Coker(α) = a′′ , Coker(β) = 0, Im(α) = a′ , Im(β) = a′′ .
The difference between general abelian categories and module categories is that while in
a module category m(k) our arguments often use the fact that k-modules are after all
abelian groups and sets (so we can think in terms of their elements), the reasoning valid
in any abelian category has to be done more formally (via composing maps and factoring
maps through intermediate objects). However, this is mostly appearances – if we try to
use set theoretic arguments we will not go wrong:
6.4.2. Theorem. [Mitchell] Any abelian category is equivalent to a full subcategory of

some category of modules m(k).
7. Exactness of functors and the derived functors
7.1. Exactness of functors. As we have observed, any functor F : A − → B between

abelian categories extends to a functor between the corresponding categories of complexes
C • (F ) : C • (A) −
→ C • (B). We would like next to extend it to a functor between derived
categories D(F ) : D(A) − → D(B) – this is what one calls the derived version of F . We
may denote it again by F , or by D(F ), or use some other notation which reflects on the
way we produce the extension.
This extension is often obtained using exactness properties of F , i.e., it will depend on how
much does F preserves exact sequences. Say, if F is exact then D(F ) is obvious: it is just
the functor F applied to complexes. If F is right-exact D(F ) is the “left derived functor
LF ” obtained by replacing objects with projective resolutions. If F is left-exact, D(F )
is the “right derived functor RF ” obtained by replacing objects with injective resolutions
(see 7.5).
58
a α
7.1.1. Exactness of a sequence of maps. A sequence of maps in an abelian category Ma −→
αb−1
Ma+1 − →··· −→ Mb−1 −−→ Mb is said to be exact at Mi (for some i with a < i < b) if
Im(αi−1 ) = Ker(αi ). The sequence is said to be exact if it is exact at all Mi , a < i < b.
(The sequence may possibly be infinite in one or both directions.)
7.1.2. Exact functors. Let F : A −

→B be an additive functor between abelian categories.
(One can think of the case where A = m(k) and B = m(h) since the general case works
the same.)
We will say that F is exact if it preserves short exact sequences, i.e., for any SES 0 −
→
α β F (α)
A′ − → A′′ −
→A − →F (A′ ) −−→
→0 in A, its F -image in B is exact, i.e., the sequence F (0) −
F (β)
F (A) −−→ F (A′′ ) −
→F (0) is a SES in B.
Example. The pull-back functors φ∗ from 5.5.1 are exact since they do not change the
structure of abelian groups and the exactness for modules only involves the level of abelian
groups.
In practice few interesting functors are exact so we have to relax the notion of exactness:
7.2. Left exact functors. We say, that F is left exact if for any SES its F -image F (0) −
→
F (α) F (β)
F (A′ ) −−→ F (A) −−→ F (A′′ ) −
→F (0) is exact except possibly in the A′′ -term, i.e., F (β)
need not be surjective.
7.2.1. Lemma. The property of left exactness is the same as asking that F preserves
α β α β
→C ′ −
exactness of sequences of the form 0 − →C −→ C ′′ , i.e., if 0 −
→A′ − → A′′ is exact
→A −
F (α) F (β)
→F (A′ ) −−→ F (A) −−→ F (A′′ ) is exact in B.
in A, its F -image F (0) −
Proof. The property in the lemma seems a priori stronger because it produces the same
conclusion in a larger number of cases. However it follows from the definition of left
exactness by a diagram chase which uses left exactness in two places.
7.2.2. Example: Invariants are left exact. For a group G let Repk (G) be the category of
all representations of G over a field k. A representations of G is a pair (V, π) of a vector
π
space V and a map of groups G − → GL(V ). We often denote π(g)v by gv, and we omit V
or π from the notation for representations. Representations of G are the same as modules
def
for the algebra k[G] = ⊕g∈G kg (multiplication is obvious). So short exact sequences,
etc., make sense in Repk (G).
def def
→ Vectk , by I(V, π) = V G = {v ∈
Lemma. The functor of invariants, I : Repk (G) −
V, gv = v, g ∈ G} is left exact.
Proof is easy. It is more interesting to see how exactness fails at the right end.
59
Counterexample. For G = Z, a representation is the same as a vector space V with an

invertible linear operator A (= π(1)). Therefore, I(V, A) is the 1-eigenspace V1 of A. Short
exact sequences in Repk (Z) are all isomorphic to the ones of the form 0 − → (V ′ , A|V ′ ) −
→
′′
(V, A) −→ (V , A) − → 0, i.e., one has a vector space V with an invertible linear operator
A, an A-invariant subspace V ′ (we restrict A to it), and the quotient space V ′′ = V /V ′
(we factor A to it). So the exactness on the right means that any w ∈ V /V ′ such that
Aw = w comes from some v in V such that Av = v.
If V = Ce1 ⊕Ce2 with A = ( 10 11 ) and V ′ = Ce1 then 0 − → I(V ′ , A|V ′ ) −
→ I(V, A) −→
id 0
′′
I(V , A) −
→0, is just 0 − → Ce1 −
→Ce1 − →V − ′′
→0, and the exactness fails on the right.
The principle “Invariants are left exact”. It applies to many other situations. Also, since
I(V ) = HomG (k, V ),
it is a special case of the next lemma. The meaning of the last equality is:
• k denotes the trivial one dimensional representation of G on the vector space k.
• Moreover, the equality notation I(V ) = HomG (k, V ) is only a remainder of
η
a more precise statement: there is a canonical isomorphism of functors I − →
HomRepk (G) (k, −) from Repk (G) to Vectk .
• The map ηV sends a G-fixed vector w ∈ I(V ) to a linear map ηV (w) : k − →V,
given by multiplying w with scalars: k ∋ c7→ c·v ∈ V . One easily checks that ηV
is an isomorphism of vector spaces.
7.2.3. Lemma. Let A be an abelian category, for any a ∈ A, the functor

HomA (a, −) : A −
→Ab,
is left exact!
α β
Proof. For an exact sequence 0 −→ b′ − → b′′ −
→ b − → 0 we consider the corresponding
′ α∗ β ∗ ′′
sequence HomA (a, b ) −→ HomA (a, b) −→ HomA (a, b ).
µ def
→ b′ and 0 = α∗ (µ) = α◦µ, then µ factors through the kernel
(1) α∗ is injective. if a −
Ker(α) (by the definition of the kernel). However, Ker(α) = 0 (by definition of a short
exact sequence), hence µ = 0.
ν
(2) Ker(β ∗ ) = Im(α∗ ). First, β∗ ◦α∗ = (β◦α)∗ = 0∗ = 0, hence Im(α∗ )⊆Ker(β ∗ ). If a −
→b
and 0 = β∗ (ν), i.e., 0 = β◦ν, then ν factors through the kernel Ker(β). But Ker(β) = a′
and the factorization now means that ν is in Im(α∗ ).
α
7.2.4. Counterexample. Let A = Ab and apply Hom(a, −) for a = Z/2Z to 0 −
→ 2Z −
→
β
Z−→ Z/2Z − → 0. Then idZ/2Z does not lift to a map from Z/2Z to Z. So β∗ need not be
surjective.
60
7.3. Right exact functors. F is right exact if it satisfies one of two equivalent properties
F (α) F (β)
→ F (A′ ) −−→ F (A) −−→ F (A′′ ) −
(1) F -image F (0) − → F (0) of a SES is exact except
′
possibly in the A -term, i.e., F (α) may fail to be injective.
α β
(2) F preserves exactness of sequences of the form C ′ − → C ′′ −
→C − →0.
Version (1) is easier to check and (2) is easier to apply.
7.3.1. Lemma. Tensoring is right exact in each argument, i.e., for any left k-module M
the functor M⊗k − : mr (k) −
→ Ab is right exact, and so is −⊗k N : m(k) − → Ab for any
right k-module N.
7.3.2. Contravariant case. Let us state the definition of right exactness also in the case
that F is a contravariant from A to B. The choice of terminology is such that one requires
that the functor F : A −→ Bo is exact, this boils down to asking (again) that exactness is
preserved except possibly at A′ (the left end of the original sequence. So we need one of
the following equivalent properties
α β F (β)
→ A′ −
(1) For a SES 0 − → A′′ −
→ A − → F (A′′ ) −−→
→ 0 in A, its F -image F (0) −
F (α)
F (A) −−→ F (A′ ) −
→F (0) is exact except possibly at F (A′ ), i.e., F (α) may fail to
be surjective.
α β
(2) F preserves exactness of sequences of the form C ′ −
→C − → C ′′ − →0.
The most important example is
7.3.3. Lemma. For any M ∈ A, the contravariant functor GM = HomA (−, M) is right
exact.
7.4. Projectives and the existence of projective resolutions. Let A be an abelian

category.
7.4.1. Projectives. We say that p ∈ A is a projective object if the functor HomA (p, −) :
A− → Ab is exact. Since HomA (p, −) is known to be always left exact, what we need is
α β
that for any short exact sequence 0 − →a − →b− →c− → 0 map Hom(p, b) − → Hom(p, c) is
surjective. In other words, if c is a quotient of b then any map from p to the quotient
γ
p−→ c lifts to a map to b, i.e., there is a map p −
→ b such that γ = β◦γ̃ for the quotient
γ̃
β
map b −
→ c.
7.4.2. Lemma. ⊕i∈I pi is projective iff all summands pi are projective.

This definition of projectivity generalizes our earlier definition in module categories since
61
7.4.3. Lemma. For a k-module P , functor Homm(k) (P, −) : m(k) −

→Ab is exact iff P is a
summand of a free module.
We say that abelian category A has enough projectives if any object is a quotient of a
projective object.
corr Module categories have enough projectives.
The importance of “enough projectives” comes from
7.4.4. Lemma. For an abelian category A the following is equivalent

(1) Any object of A has a projective resolution (i.e., a left resolution consisting of
projective objects).
(2) A has enough projectives.
7.5. Injectives and the existence of injective resolutions. Dually, we say that i ∈ A
→Abo is exact.
is an injective object if the functor HomA (−, i) : A −
Again, since HomA (−, i) is always right exact, we need for any short exact sequence
α β α∗
0− →a − →b − →c − → 0 that the map Hom(b, p) −→ Hom(a, p), α∗ (φ) = φ◦α; be surjective.
γ
This means that if a is a subobject of b then any map a − → i from a subobject a to i
γ̃
extends to a map from b to i, i.e., there is a map b −
→ i such that γ = γ̃◦α. So, an object
i is injective if each map from a subobject a′ ֒→a to i, extends to the whole object a.
7.5.1. Example. Z is projective in Ab but it is not injective in Ab: Z⊆ n1 Z and the map
1Z : Z −
→Z does not extend to n1 Z −→Z.
Q
7.5.2. Lemma. Product i∈I Ji is injective iff all factors Ji are injective.
7.5.3. Lemma. A Z-module I is injective iff I is divisible, i.e., for any a ∈ I and n ∈
{1, 2, 3, ...} there is some ã ∈ I such that a = n·ã. (i.e., multiplication n : I −
→ I with
n ∈ {1, 2, 3, ...} is surjective.)
The proof will use the Zorn lemma which is an essential part of any strict definition of
set theory:
• Let (I, ≤) be a (non-empty) partially ordered set such that any chain J in I (i.e.,
any totally ordered subset) is dominated by some element of I (i.e., there is some
i ∈ I such that i ≥ j, j ∈ J). Then I has a maximal element.
α
Proof. For any a ∈ I and n > 0 we can consider n1 Z⊇Z −→ I with α(1) = a. If I is
e : nZ −
injective then α extends to α 1
→I and a = ne 1
α( n ).
β
Conversely, assume that I is divisible and let A⊇B −→ I. Consider the set E of all
pairs (C, γ) with B⊆C⊆A and γ : C − → I an extension of β. It is partially ordered
62
with (C, γ) ≤ (C ′ , γ ′ ) if C⊆C ′ and γ ′ extends γ. From Zorn lemma and the following
observations it follows that E has an element (C, γ) with C = A:
(1) For any totally ordered subset E ′ ⊆E there is an element (C, γ) ∈ E which domi-
nates all elements of E ′ (this is clear: take C = ∪(C ′ ,γ ′ )∈E ′ C ′ and γ is then obvious).
(2) If (C, γ) ∈ E and C 6= A then (C, γ) is not maximal:
• choose a ∈ A which is not in C and let C e = C + Z·a and C ∩ Z·a = Z·na
e
with n ≥ 0. If n = 0 then C = C⊕Z·a and one can extend γ to C by zero on
Z·a. If n > 0 then γ(na) ∈ I is n-divisible, i.e., γ(na) = nx for some x ∈ I.
Then one can extend γ to C e by eγ (a) = x (first define a map on C⊕Z·a, and
then descend it to the quotient C). e
7.5.4. We say that abelian category A has enough injectives if any object is a subobject
of an injective object.
7.5.5. Lemma. For an abelian category A the following is equivalent

(1) Any object of A has an injective resolution (i.e., a right resolution consisting of
injective objects).
(2) A has enough injectives.
c = Hom(M, Q/Z). Then the canonical

7.5.6. Lemma. For any abelian group M denote M
ρ c
map M − →Mc is injective.
(b) Category of abelian groups has enough injectives.
Proof. (a) For m ∈ M, χ ∈ M c, ρ(m) (χ) def
= χ(m). So, ρ(m) = 0 means that m is killed
c
by each χM (“each character of M”). If m 6= 0 then Z·m is isomorphic to Z or to one of
χ0
Z/nZ, in each case we can find a Z·m −→ Q/Z which is 6= 0 on the generator m. Since
Q/Z is injective we can extend χ0 to M.
Q c
c Q/Z·x = (Q/Z) , its
M
(b) To M we associate a huge injective abelian group IM = x∈M
elements are Mc-families c = (cχ ) c of elements of Q/Z (we denote such family also as
Pχ∈M
a (possibly infinite) formal sum χ∈M c cχ ·χ). By part (a), canonical map io is injective
ι
X
M− →IM , ι(m) = (χ(m))χ∈M c = χ(m)·χ, m ∈ M.
c
χ∈M
7.5.7. Lemma. Let F : A − → B be an exact functor between abelian categories. Its right
adjoint preserves injectivity and its left adjoint preserves projectivity.
Proof. For the right adjoint G and an injective b ∈ B, functor HomA [−, Gb] ∼
= Hom[F −, b]
is exact.
63
φ
corr For a map of rings k − → h functor φ∗ : m(k) −
→ m(h), φ∗ (M) = h⊗k M preserves
projectivity and φ⋆ : m(k) −
→m(h), φ⋆ (M) = Homk (h, M) preserves injectivity.
Proof. These are the two adjoints of the forgetful functor φ∗ .
7.5.8. Theorem. Module categories m(k) have enough injectives.

Proof. The problem will be reduced to the case k = Z via the canonical map of rings
φ
Z − → k. Any k-module M gives a Z-module φ∗ M, and by lemma 7.5.6 there is an
ι
embedding M − → IM into an injective abelian group. Moreover, by corollary 7.5.7 φ⋆ IM
is an injective k-module. So it suffices to have an embedding M֒→φ⋆ IM .
ζ
The adjoint pair (φ∗ , φ⋆) gives a map of k-modules M −
→ φ⋆ φ∗ M = HomZ (k, M), by
ζ(m) c = cm, m ∈ M, c ∈ k. It remains to check that both maps in the composition
ζ φ⋆ (ι)
M− → φ⋆ (M) −−→ φ⋆ (IM ) are injective. For ζ it is obvious since ζ(m) 1k = m, and for φ⋆
we recall that HomZ (k, −) is left exact, hence takes injective maps to injective maps.
7.5.9. Examples. (1) An injective resolution of the Z-module Z: 0 −

→Z −
→Q −
→Q/Z −
→0.
(2) Injective resolutions are often big, hence more difficult to use in specific calculations
then say, the free resolutions. We will need them mostly for the functor Γ(X, −) of global
sections of sheaves, and the functors HomA (a, −).
7.6. Exactness and the derived functors. This is a preliminary motivation for the
precise construction of derived functors in the next chapters.
7.6.1. Left derived functor RF of a right exact functor F . We observe that if F is right
exact then the correct way to extend it to a functor on the derived level is the construction
def
LF (M) = F (P •), i.e., replacement of the object by a projective resolution. “Correct”
means here that LF is really more then F – it contains the information of F in its zeroth
def
cohomology, i.e., L0 F ∼
= F for Li F (M) = H i [LF (M)]. Letter L reminds us that we use
a left resolution.
7.6.2. Lemma. If the functor F : A −

→ B is right exact, there is a canonical isomorphism
of functors H 0 (LF ) ∼
= F.
d−1 q
→P −2 −
Proof. Let · · · − →P −1 −−→ P 0 −
→M −
→0 be a projective resolution of M.
d−1
→P −2 −
(A)The case when F is covariant. Then LF (M) = F [· · · − →P −1 −−→ P 0 −
→0 −
→
· · ·] equals
F (d−1 )
→F (P −2) −
[· · · − →F (P −1 ) −−−−→ F (P 0 ) −
→0 −
→· · ·],
so H 0 [LF (M)] = F (P 0 )/F (d−1)F (P −1 ).
64
d−1 q
If we apply F to the exact sequence P −1 −−→ P 0 −
→M −
→ 0, the right exactness gives
F (d−1 ) F (q)
an exact sequence F (P −1) −−−−→ F (P 0 ) −−→ F (M) −
→ 0. Therefore, F (q) factors to a
0 −1 −1
canonical map F (P )/F (d )F (P ) − → F (M) which is an isomorphism.
→ P −2 −
(B)The case when F is contravariant. This is similar, LF (M) = F (· · · − →
d−1
P −1 −−→ P 0 −
→0 −
→· · ·) equals
d−1
··· − →F (P −0) −−→ F (P −1 ) −
→0 − →F (P −2 ) −
→0 −
→· · ·,
d−1
and we get H 0 [LF (M)] = Ker[F (P −0) −−→ F (P −1 )]. However, applying F to the exact
d−1 F (d−1 )
sequence P −1 −−→ P 0 −
→M −
→ 0 gives an exact sequence 0 − → F (P 0) −−−−→
→ F (M) −
d−1
F (P −1). So the canonical map F (M) −
→ Ker[F (P −0 ) −−→ F (P −1)] is an isomorphism.
7.6.3. Remark. As we see the argument is categorical and would not simplify if we only
considered module categories.
→ m(DA0 ), io M = M/xM. It is right

7.6.4. Example. Recall the functor io : m(DA1 ) −
exact by 7.3.1 since M/xM = M⊗k[x] k[x]/xk[x], and therefore H 0[Lio (M)] ∼
∼ = io (M) by
the preceding lemma.
7.6.5. Right derived functor RF of a left exact functor F . Obviously, we want to define
for any left exact functor F : A −→ B a right derived functor RF by replacing an object
by its injective resolution. Then, as above H 0 (RF ) ∼
= F.
65
8. Abelian category of sheaves of abelian groups
For a topological space X we will denote by Sh(X) = Sheaves(X, Ab) the category of
sheaves of abelian groups on X. Since a sheaf of abelian groups is something like an abelian
group smeared over X we hope to Sh(X) is again an abelian category. When attempting
to construct cokernels, the first idea does not quite work – it produces something like a
sheaf but without the gluing property. This forces us to
• (i) generalize the notion of sheaves to a weaker notion of a presheaf,

• (ii) find a canonical procedure that improves a presheaf to a sheaf.
(We will also see that a another example that requires the same strategy is the pull-back
operation on sheaves.)
Now it is easy to check that we indeed have an abelian category. What allows us to
compute in this abelian category is the lucky break that one can understand kernels,
cokernels, images and exact sequences just by looking at the stalks of sheaves.
8.1. Categories of sheaves. A presheaf of sets S on a topological space (X, T ) consists

of the following data:
• for each open U⊆X a set S(U),

V ρU
• for each inclusion of open subsets V ⊆U⊆X a map S(U) −→ S(V ) (called the
restriction map);
and these data are required to satisfy
• (Sh0)(Transitivity of restriction) ρUV ◦ρUV = ρUW for W ⊆V ⊆U
A sheaf of sets on a topological space (X, T ) is a presheaf S which also satisfies
• (Sh1) (Gluing) Let U = (Ui )i∈I be an open cover of an open U⊆X (We denote
Uij = Ui ∩Uj etc.). We ask that any family of compatible sections fi ∈ S(Ui ), i ∈ I,
glues uniquely. This means that if sections fi agree on intersections in the sense
that ρUUiij fi = ρUUiij fj in S(Uij ) for any i, j ∈ I; then there is a unique f ∈ S(U)
such that ρUUi f = fi in S(Ui ), i ∈ I.
• S(∅) is a point.
8.1.1. Remarks. (1) Presheaves of sets on X form a category preSheaves(X, Sets) when
Hom(A, B) consists of all systems φ = (φU )U ⊆X open of maps φU : A(U) −
→ B(U) which
66
are compatible with restrictions, i.e., for V ⊆U

U φ
A(U) −−−→ B(U)
 
 U .
ρU
V y ρVy
V φ
A(V ) −−−→ B(V )
(One reads the diagram above as : “the diagram ... commutes”.) The sheaves form a full
subcategory preSheaves(X, Sets) of Sheaves(X, Sets).
(2) We can equally define categories of sheaves of abelian groups, rings, modules, etc.
For a sheaf of abelian groups we ask that all A(U) are abelian groups, all restriction
morphisms are maps of abelian groups, and we modify the least interesting requirement
(Sh2): S(φ) is the trivial group {0}. In general, for a category A one can define categories
preSheaves(X, A) and Sheaves(X, A) similarly (the value on ∅ should be the final object
of A).
8.2. Sheafification of presheaves. We will use the wish to pull-back sheaves as a mo-
tivation for a procedure that improves presheaves to sheaves.
π
8.2.1. Functoriality of sheaves. Recall that for any map of topological spaces X −
→Y
π −1
one wants a pull-back functor Sheaves(Y ) −−→ Sheaves(X). As we have seen in the
definition of a stalk of a sheaf (pull ,back to a point), the natural formula is
def
π −1 (N ) (U) = lim N (V ),
→
V ⊇π(U )
where limit is over open V ⊆Y that contain π(U), and we say that V ′ ≤ V ′′ if V ′′ better
approximates π(U), i.e., if V ′′ ⊆V ′ .
π −1
8.2.2. Lemma. This gives a functor of presheaves preSheaves(X) −−→ preSheaves(Y ).
Proof. For U ′ ⊆U open, π −1 N (U ′ ) = lim N (V ) and π −1 N (U) = lim N (V )
→ V ⊇π(U ′ ) → V ⊇π(U )
are limits of inductive systems of N (V )’s, and the second system is a subsystem of the
first one, this gives a canonical map π −1 N (U) −
→π −1 N (U ′ ).
8.2.3. Remark. Even if N is a sheaf, π −1 (N ) need not be sheaf.

For that let Y = pt and let N = SY be the constant sheaf(of sets on Y given by a set
∅ if U = ∅,
S. So, SY (∅) = ∅ and SY (Y ) = S. Then π −1 (SY ) (U) = . We can say:
S U 6= ∅
π −1 (SY ) (U) = constant functions from U to S. However, we have noticed that constant
functions do not give a sheaf, so we need to correct the procedure π −1 to get sheaves from
67
sheaves. For that remember that for the presheaf of constant functions there is a related
sheaf SX of locally constant functions.
Our problem is that the presheaf of constant functions is defined by a global condition
(constancy) and we need to change it to a local condition (local constancy) to make it
into a sheaf. So we need the procedure of
e
8.2.4. Sheafification. This is a way to improve any presheaf of sets S into a sheaf of sets S.
We will imitate the way we passed from constant functions to locally constant functions.
More precisely, we will obtained the sections of the sheaf S̃ associated to the presheaf S
in two steps:
(1) we glue systems of local sections si which are compatible in the weak sense that
they are locally the same, and
(2) we identify two results of such gluing if the local sections in the two families are
locally the same.
e
Formally these two steps are performed by replacing S(U) with the set S(U), defined as
the set of all equivalence classes of systems (Ui , si )i∈I where
b
(1) Let S(U) be the class of all systems (Ui , si )i∈I such that
• (Ui )i∈I is an open cover of U and si is a section of S on Ui ,
• sections si are weakly compatible in the sense that they are locally the same,
i.e., for any i′ , i′′ ∈ I sections si′ and si′′ are the same near any point x ∈ Ui′ i′′ .
(Precisely, this means that there is neighborhood W such that si′ |W = si′′ |W .)
(2) We say that two systems (Ui , si )i∈I and (Vj , tj )j∈J are ≡, iff for any i ∈ I, j ∈ J
sections si and tj are weakly equivalent (i.e., for each x ∈ Ui ∩ Vj , there is an open
set W with x ∈ W ⊆Ui ∩ Vj such that “si = tj on W ” in the sense of restrictions
being the same).
8.2.5. Remark. The relation ≡ on S(U) b really says that (Ui , si )i∈I ≡ (Vj , tj )j∈J iff the
b
disjoint union (Ui , si )i∈I ⊔ (Vj , tj )j∈J is again in S(U).
8.2.6. Lemma. (a) ≡ is an equivalence relation.

q
e
(b) S(U) is a presheaf and there is a canonical map of presheaves S − e
→ S.
(c) Se is a sheaf.
Proof. (a) is obvious.
(b) The restriction of a system (Ui , si )i∈I to V ⊆U is the system (Ui ∩ V, si |Ui ∩ V )i∈I . The
weak compatibility of restrictions si |U ∩ V follows from the weak compatibility of sections
si . Finally, restriction is compatible with ≡, i.e., if (Ui′ , s′i )i∈I and (Uj′′ , s′′j )j∈J are ≡, then
so are (Ui′ ∩ V, s′i |Ui′ ∩ V )i∈I and (Uj′′ ∩ V, s′′j |Uj′′ ∩ V )j∈J .
68
The map S(U) − e

→ S(U) is given by interpreting a section s ∈ S(U) as a (small) system:
open cover of (Ui )i∈{0} is given by U0 = U and s0 = s.
(c’) Compatible systems of sections of the presheaf Se glue. Let V j , j ∈ J, be an open
cover of an open V ⊆X, and for each j ∈ J let σ j = [(Uij , sji )i∈ Ij ] be a section of Se on
Vj . So, σ j is an equivalence class of the system (Uij , sji )i∈ Ij consisting of an open cover
Uij , i ∈ Ij , of Vj and weakly compatible sections sij ∈ S(Uji ).
Now, if for any j, k ∈ J sections σ j = [(Upj , sjp )p∈ Ij ] and σ k = [(Uqk , skq )q∈ Ik ] of Se on V j
and V k , agree on the intersection V jk . This means that for any j, k σ j |V jk = σ k |V jk , i.e.,
(Upj ∩ V jk , sjp |Upj ∩ V jk )p∈ Ij ≡ (Uqk ∩ V jk , skq |Uqk ∩ V jk )q∈ Ik .
This in turn means that for j, k ∈ J and any p ∈ Ij , q ∈ Ik , sections sjp and skq are weakly
compatible. Since all sections sjp , j ∈ J, p ∈ Ij are weakly compatible, the disjoint union
b ). Its equivalence class σ is a section
of all systems (Uij , sji )i∈ Ij , j ∈ J is a system in S(V
of Se on V , and clearly σ|V j = σ j .
(c”) Compatible systems of sections of the presheaf Se glue uniquely. If τ ∈ S(V e ) is the
class of a system (Ui , si )i∈I and τ |V j = σ j then σ’s are compatible with all sjp ’s, hence
(Ui , si )i∈I ≡ ⊔j∈J (Uij , sji )i∈ Ij , hence τ = σ.
8.2.7. Sheafification as a left adjoint of the forgetful functor. As usual, we have not in-
vented something new: it was already there, hidden in the more obvious forgetful functor
8.2.8. Lemma. Sheafification functor preSheaves ∋ S7→Se ∈ Sheaves, is the left adjoint
of the inclusion Sheaves⊆preSheaves, i.e, for any presheaf S and any sheaf F there is a
natural identification
∼
e F )−
HomSheaves (S,
=
→ HompreSheaves (S, F ).
α ιS α
Explicitly, the bijection is given by (ιS )∗ α = α◦ιS , i.e., (Se −
→ F )7→ (S −→ Se −
→ F ).
8.3. Inverse and direct images of sheaves.
8.3.1. Pull back of sheaves (finally!) Now we can define for any map of topological spaces
π
X− → Y a pull-back functor
π −1
π^
def
Sheaves(Y ) −−→ Sheaves(X), π −1 N = −1 N .
ρ
→ X. Then ρ−1 S is the
8.3.2. Examples. (a) A point a ∈ X can be viewed as a map {a} −
stalk Sa .
→pt, for any set S one has SX = a−1 S.
(b) Let a : X −
69
8.3.3. Direct image of sheaves. Besides the pull-back of sheaves which we defined in 8.3.1,
there is also a much simpler procedure of the push-forward of sheaves:
π
8.3.4. Lemma. (Direct image of sheaves.) Let X −
→ Y be a map of topological spaces.
For a sheaf M on X, formula
def
π∗ (M) (V ) = M(π −1 V ),
∗ π
defines a sheaf π∗ M on Y , and this gives a functor Sheaves(X) −→ Sheaves(Y ).
8.3.5. Adjunction between the direct and inverse image operations. The two basic opera-
tions on sheaves are related by adjunction:
Lemma. For sheaves A on X and B on Y one has a natural identification
Hom(π −1 B, A) ∼
= Hom(B, π∗ A).
Proof. We want to compare β ∈ Hom(B, π∗ A) with α in
HomSh(X) (π −1 B, A) = HomSh(X) (π]

−1 B, A) ∼
= HompreSh(X) (π −1 B, A).
α is a system of maps
α
lim B(V ) = π −1 B(U) −→
U
A(U), for U open in X,
→ V ⊇π(U )
and β is a system of maps

β
V
B(V ) −→ A(π −1 V ), for V open in Y .
Clearly, any β gives some α since

lim βV
A(π −1 V ) −
→
lim B(V ) −−−−→ lim →A(U),
→ V ⊇π(U ) → V ⊇π(U )
the second map comes from the restrictions A(π −1V ) −

→ A(U) defined since V ⊇π(U)
implies π −1 V ⊇U .
For the opposite direction, any α gives for each V open in Y , a map lim B(W ) =
→ W ⊇π(π −1 V )
α −1
π −1 B(π −1 V ) −−π−−→
V
A(π −1 V ). Since B(V ) is one of the terms in the inductive system we
have a canonical map B(V ) − → lim B(W ), and the composition with the first
→ W ⊇π(π −1 V )
απ−1 V
−1
map B(V ) −
→ lim B(W ) −−−−→ A(π V ), is the wanted map βV .
→ W ⊇π(π −1 V )
70
π τ
8.3.6. Lemma. (a) If X − →Y − → Z then
τ∗ (π∗ A) ∼
= (τ ◦π)∗ A and τ∗ (π∗ A) ∼
= (τ ◦π)∗ A.
(b) (1X )∗ A ∼
= A∼
= (1X )−1 A.
Proof. The statements involving direct image are very simple and the claims for inverse
image follow by adjunction.
corr (Pull-back preserves the stalks) For a ∈ X one has (π −1 N )a ∼
= Nπ(a) .
This shows that the pull-back operation which was difficult to define is actually very
simple in its effect on sheaves.
8.4. Stalks. Part (a) of the following lemma is the recollection of the description of
inductive limits of abelian groups from the remark 5.3.8.
8.4.1. Lemma. (Inductive limits of abelian groups.) (a) For an inductive system of
abelian groups (or sets) Ai over (I, ≤), inductive limit lim Ai can be described by
→
• for i ∈ I any a ∈ Ai defines an element a of lim Ai ,

→
• all elements of lim Ai arise in this way, and
→
• for a ∈ Ai and b ∈ Aj one has a = b iff for some k ∈ I with i ≤ k ≥ j one has
a = b in Ak .
(b) For a subset K⊆I one has a canonical map lim Ai −
→ lim Ai .
→ i∈K → i∈I
Proof. In general (b) is clear from the definition of lim, and for abelian groups also from
→
(a).
8.4.2. Stalks of (pre)sheaves. Remember that the stalk of a presheaf Ab at a point x is

def
Ax = lim A(U), the limit over (diminishing) neighborhoods u of x. This means that
→
• any s ∈ A(U) with U ∋ x defines an element sx of the stalk,

• all elements of Ax arise in this way, and
• For s′ ∈ A(U ′ ) and s′′ ∈ A(U ′′ ) one has s′x = s′′x iff for some neighborhood W of x
in U ′ ∩ U ′′ one has s′ = s′′ on W .
→ Se is an isomorphism on stalks.
8.4.3. Lemma. For a presheaf S, the canonical map S −
i
Proof. We consider a point a ∈ X as a map pt = {a} − → X, so that Ax = i−1 A. For a
sheaf B on the point
e B) ∼
HomSh(pt) (i−1 S, = HomSh(X) (S,e i∗ B) ∼
= HompreSh(X)(S, i∗ B)
∼
= HompreSh(pt) (i−1 S, B) = HomSh(pt) (i−1 S, B).
71
8.4.4. Germs of sections and stalks of maps. For any neighborhood U of a point x we
def
have a canonical map S(U) −→lim S(V ) = Sx (see lemma 8.4.1), and we denote the
→ V ∋x
image of a section s ∈ Γ(U, S) in the stalk Sx by sx , and we call it the germ of the section
at x. The germs of two sections are the same at x if the sections are the same on some
(possibly very small) neighborhood of x (this is again by the lemma 8.4.1).
A map of sheaves φ : A→B defines for each x ∈ M a map of stalks Ax →Bx which we
call φx . It comes from a map of inductive systems given by φ, i.e., from the system of
maps φU : A(U) − → B(U), U ∋ x (see 5.3.5) ; and on germs it is given by φx (ax ) =
[φU (a)]x , a ∈ A(U).
For instance, let A = HC be the sheaf of holomorphic functions on C. Remember that
the stalk at a ∈ C can be identified with all convergent power series in z − a. Then the
germ of a holomorphic function f ∈ HC (U) at a can be thought of as the power series
Φ
expansion of f at a. An example of a map of sheaves HC − → HC is the multiplication by
an entire function φ ∈ HC (C), its stalk at a is the multiplication of the the power series
at a by the power series expansion of φ at a.
8.4.5. The following lemma from homework shows how much the study of sheaves reduces
to the study of their stalks.
Lemma. (a) Maps of sheaves φ, ψ : A→B are the same iff the maps on stalks are the
same, i.e., φx = ψx for each x ∈ M.
(b) Map of sheaves φ : A→B is an isomorphism iff φx is an isomorphism for each x ∈ M.
8.4.6. Sheafifications via the etale space of a presheaf. We will construct the sheafification
•
of a presheaf S (once again) in an “elegant” way, using the etale space S of the presheaf.
It is based on the following example of sheaves
p
Example. Let Y −
→ X be a continuous map. For any open U⊆X the elements of
def
Σ(U) = {s : U −
→Y, s is continuous and p◦s = 1u }
are called the (continuous) sections of p over U. Σ is a sheaf of sets.
•
To apply this construction we need a space S that maps to X:
•
• Let S be the union of all stalks Sm , m ∈ X.
•
• Let p : S→X be the map such that the fiber at m is the stalk at m.
• For any pair (U, s) with U open in X and s ∈ S(U), define a section s̃ of p over U
by
def •
s̃(x) = sx ∈ Sx ⊂ S, x ∈ U.
72
Lemma. (a) If for two sections si ∈ S(Ui ), i = 1, 2; of S, the corresponding sections s̃1
and s̃2 of p agree at a point then they agree on some neighborhood of of this point (i.e.,
def
if s̃1 (x) = s̃2 (x) for some x ∈ U12 = U1 ∩ U2 , then there is a neighborhood W of x such
that s̃1 = s̃2 on W ).
•
(b) All the sets s̃(U) (for U⊆X open and s ∈ S(U)), form a basis of a topology on S.
•
Map p : S→M is continuous.
(c) Let S̃(U) denote the set of continuous sections of p over U. Then S̃ is a sheaf and
there is a canonical map of presheaves ι : S→S̃.
Remark. Moreover, p is “etale” meaning “locally an isomorphism”, i.e., for each point σ ∈
• •
S there are neighborhoods σ ∈ W ⊆S and p(σ)⊆U⊆X such that p|W is a homeomorphism
∼
=
W− →U.
Lemma. The new Se and the old Se (from 8.2.4) are the same sheaves (and the same holds
for the canonical maps ι : S − e
→ S).
b ≡
Proof. Sections of p over U⊆X are the same as the equivalence classes of systems S/
defined in 8.2.4.
α
8.5. Abelian category structure. Let us fix a map of sheaves A − → B since the non-
trivial part is the construction of (co)kernels. Consider the example where the space is
the circle X = {z ∈ C, |z| = 1} and A = B is the sheaf CX ∞
of smooth functions on
∂
X, and the map α is the differentiation ∂ = ∂θ with respect to the angle θ. For U⊆X
∞ ∞
open, Ker(∂U ) : CX (U) −
→CX (U) consists of locally constant functions and the cokernel
∞ ∞
CX (U)/∂U CX (U) is
• zero if U 6= X (then any smooth function on U is the derivative of its indefinite
integral defined by using the exponential chart z = eiθ which identifies U with an
open subset of R), R
• one dimensional if U = X – for g ∈ C ∞ (X) one has X ∂g = 0 so say constant
functions on X are not derivatives (and for functions with integral zero the first
argument applies).
So by taking kernels at each level we got a sheaf but by taking cokernels we got a presheaf
which is not a sheaf (local sections are zero but there are global non-zero sections, so the
object is not controlled by its local properties).
8.5.1. Subsheaves. For (pre)sheaves S and S ′ we say that S ′ is a sub(pre)sheaf of S if

ρ′
S ′ (U)⊆S(U) and the restriction maps for S ′ , S ′ (U) −
→ S ′ (V ) are restrictions of the
ρ
restriction maps for S, S(U) −
→ S(V ).
73
α
8.5.2. Lemma. (Kernels.) Any map A −
→ B has a kernel and Ker(α)(U) =
φ(U )
Ker[A(U) −−→ B(U)] is a subsheaf of A.
def φ(U ) µ
Proof. First, K(U) = Ker[A(U) −−→ B(U)] is a sheaf, and then a map C −
→ A is killed
by α iff it factors through the subsheaf K of A.
α def
Lemma. (Cokernels.) Any map A − → B defines a presheaf C(U) = B(U)/αU (A(U)), the
associated sheaf C is the cokernel of α.
Proof. For a sheaf S one has
= HompreSheaves (C, S) ∼
HomSheaves (B, S)α ∼ = HomSheaves (C, S).
φ
The second identification is the adjunction. For the first one, a map B −
→ S is killed by
α, i.e., 0 = φ◦α, if for each U one has 0 = (φ◦α)U A(U) = φU (αU A(U)); but then it gives
φ
a map C − → S, with φU : C(U) = B(U)/αU A(U) − → S(U) the factorization of φU . The
opposite direction is really obvious, any ψ : C −
→ S can be composed with the canonical
map B −→ C (i.e., B(U) − → B(U)/αU A(U)) to give map B − → S which is clearly killed by
α.
α
8.5.3. Lemma. (Images.) Consider a map A −
→ B.
def
(a) It defines a presheaf I(U) = αU (A(U))⊆ B(U) which is a subpresheaf of B. The
associated sheaf I is the image of α.
def
(b) It defines a presheaf c(U) = A(U)/Ker(αU ), the associated sheaf is the coimage of
α.
(c) The canonical map Coim(α) −
→Im(α) is isomorphism.
def
Proof. (a) Im(α) = Ker[B − → Coker(α)] is a subsheaf of B and b ∈ B(U) is a section of
Im(α) iff it becomes zero in Coker(α). But a section b + αU A(U) of C on U is zero in B
iff it is locally zero in C, i.e., there is a cover Ui of U such that b|Ui ∈ αUi A(Ui ). But this
is the same as saying that b is locally in the subpresheaf I of B, i.e., the same as asking
that b is in the corresponding presheaf I of B.
def
(b) The coimage of α is by definition Coim(α) = Coker[Ker(α) −
→ A], i.e., the sheaf
associated to the presheaf U7→ A(U)/Ker(α)(U) = c(U).
(c) The map of sheaves Coim(α) − → Im(α) is associated to the canonical map
of presheaves c −
→ I, however already the map of presheaves is an isomorphism:
def
c(U) = A(U)/Ker(α)(U) ∼
= αU = A(U) = I(U).
8.5.4. Stalks of kernels, cokernels and images; exact sequences of sheaves.

74
α
8.5.5. Lemma. For a map of sheaves A −
→ B and x ∈ X
α
• (a) Ker(A −
→ B)x = Ker(αx : Ax −→Bx ),
α
• (b) Coker(A −
→ B)x = Coker(αx : Ax −
→Bx ),
α
• (c) Im(A −→ B)x = Im(αx : Ax −
→Bx ).
def
Proof. (a) Let x ∈ U and a ∈ A(U). The germ ax is killed by αx if 0 = αx (ax ) = (αU (a))x ,
i.e., iff αU (a) = 0 on some neighborhood U ′ of x in U. But this is the same as saying
that 0 = αU (a)|U ′ = αU ′ (a|U ′ ), i.e., asking that some restriction of a to a smaller
neighborhood of x is a section of the subsheaf Ker(α). And this in turn, is the same as
saying that the germ ax lies in the stalk of Ker(α).
q qx
(b) Map B − → Coker(α) is killed by composing with α, so the map of stalks Bx −→
Coker(α)x is killed by composing with αx .
To see that qx is surjective consider some element of the stalk Coker(α)x . It comes from
a section of a presheaf U7→B(U)/αU A(U), so it is of the form [b + αU (A(U))]x for some
section b ∈ B(U) on some neighborhood U of x. Therefore it is the image αx (bx ) of an
element bx of Bx .
To see that qx is injective, observe that a stalk bx ∈ Bx (of some section bB(U)), is killed
by qx iff its image αx (bx ) = [b + αU (A(U))]x is zero in Coker(α), i.e., iff there is a smaller
neighborhood U ′ ⊆U such that the restriction [b + αU (A(U))]|U ′ = b|U ′ + αU ′ (A(U ′ )) is
zero, i.e., b|U ′ is in αU ′ A(U ′ ). But the existence of such U ′ is the same as saying that bx
is in the image of αx .
(c) follows from (a) and (b) by following how images are defined in terms of kernels and
cokernels.
corr A sequence of sheaves is exact iff at each point the corresponding sequence of stalks
of sheaves is exact.
75
Part 3. Derived categories of abelian categories
9. Homotopy category of complexes
On the way to identifying any two quasi-isomorphic complexes, i.e., to inverting all quasi-
isomorphisms, in the first step we will invert a special kind of isomorphisms – the homo-
topy equivalences.
This is achieved by passing from the category of complexes C(A) to the so called “homo-
topy category of complexes” K(A). Category K(A) is no more an abelian category but it
has a similar if less familiar structure of a “triangulated category”. The abelian structure
of the category C(A) provides the notion of short exact sequences. This is essential since
one can think of putting B into a short exact sequences 0 − →A − →B − →C − → 0, as
describing the complex B in terms of simpler complexes A and C; and indeed it turns out
that the cohomology groups of B are certain combinations of cohomology groups of A and
B. Since K(A) is not abelian we are forced to find an analogue of short exact sequences
which works in K(A), this is the notion of “distinguished triangles”, also called “exact
triangles”. The properties of exact triangles in K(A) formalize into the concept of a “tri-
angulated category” which turns out to be the best standard framework for homological
algebra.
In particular, the passage to (A) will solve the remaining foundational problem in the
definition of derived functors:
• identify any two projective resolutions of one object
(so the derived functors will be well defined since we remove the dependence on the choice
of a projective resolution).
9.1. Category C(A) of complexes in A. We observe some of the properties of the

category C(A).
9.1.1. Structures.
• Shift functors. For any integer n define a shift functor [n] : C(A) −
→ C(A) by
p+n
(−1)n dA
p def n+p p p+1 p+n
(A[n]) = A , and the differential (A[n]) − → A[n]) given as A −−−−−−→
Ap+1+n .
• Functors Hi : C(A) − →A.
• Special class of morphisms related to cohomology functors: the quasi-
isomorphisms.
α β γ
• Triangles. These are diagrams of the form A − →B − →C − → A[1].
• Subcategories C b (A) etc. If ? is one of the symbols b, −, + we define a full subcat-
egory C ? (A) of C ? (A), consisting respectively of bounded complexes (i.e. An = 0
76
for |n| >> 0), complexes bounded from bellow: An = 0 for n << 0 (hence al-
lowed to stretch in the + direction), complexes bounded from above (so they may
stretch in the − direction). Moreover, for a subset Z⊆Z we can define C Z (A)
as a full subcategory consisting of all complexes A with An = 0 for n 6∈ Z. In
def def
particular one has C ≤0 (A) = C {···,−2,−1,0} and C ≥0 (A) = C {0,1,2,···} , and C {0} (A)
is equivalent to A.
9.1.2. Properties. The next two lemmas give basic properties of the above structures on
the category C(A).
9.1.3. Lemma. C(A) is an abelian category and a sequence of complexes is exact iff it is
exact on each level!
α αn
Proof. For a map of complexes A − → B we can define K n = Ker(An −→ B n ) and
C n = An /αn (B n ). This gives complexes since dA induces a differential dK on K and
dB a differential dC on C. Moreover, it is easy to check that in category C(A) one has
K = Ker(α) and C = Coker(α). Now one finds that Im(α)n = Im(αn ) = αn (An ) and
Coim(α)n = Coim(αn ) = An /Ker(αn ), so the canonical map Coim − → Im is given by
∼
n n = n n
isomorphisms A /Ker(α )− → α (A ). Exactness claim follows.
9.1.4. Lemma. A short exact sequence of complexes 0 −

→A −
→B −
→C −
→ 0 gives a long
exact sequence of cohomologies.
∂ n−1 Hn (α) Hn (β) ∂n Hn+1 (α) Hn+1 (β)
· · · −−→ Hn (A) −−−→ Hn (B) −−−→ Hn (C) −→ Hn+1 (A) −−−−→ Hn+1 (B) −−−−→ · · ·
Proof. We need to construct for a class γ ∈ Hn (C) a class ∂γ ∈ Hn+1 . So if γ = [c] is the
class of a cocycle c, we need
(1) From a cocycle c ∈ Z n (C) a cocycle a ∈ Z n+1 .

(2) Independence of [a] on the choice of c or any other auxiliary choices.
(3) The sequence of cohomology groups is exact.
Recall that a sequence of complexes. 0 −

→A −→B −→C −→ 0 is a short exact sequence if
→An −
for each integer n the sequence 0 − →B n −
→C n −
→0 is exact.
The following calculation is in the set-theoretic language appropriate for module categories
but can be rephrased in the language of abelian categories (and also the result for module
categories implies the result for abelian categories since any abelian category is equivalent
to a full subcategory of a module category).
(1) Since β n is surjective, c = β n b for b ∈ B n , Now db = αn+1 a for some a ∈ An+1 since
β n+1 (db) = dβ n+1b = dc = 0. Moreover, a is a cocycle since αn+2 (dn+1 a) = dn+1 (αn+1 a) =
dn+1 (dn b) = 0.
77
(2) So we want to associate to γ = [c] the class α = [a] ∈ Hn+1 (A). For that [a] should
be independent of the choices of c, B and a. So let [c] = [c′ ] and c′ = β n b′ with b′ ∈ B n ,
and db′ = αn+1 a′ for some a′ ∈ An+1 .
Since [c] = [c′ ] one has c′ = c+dz for some z ∈ C n−1 . Choose y ∈ B n−1 so that z = β n−1y,
then
β n b′ = c′ = c + dz = β n b + d(β n−1y) = β n b + β n dy = β n (b + dy).
The exactness at B now shows that b′ = b + dy + αn x for some x ∈ An . So,
αn+1 a′ = db′ = db + dαn x = αn+1 a + αn (dx) = αn+1 (a + dx).
Exactness at A implies that actually a′ = a + dx.
(3) I omit the easier part: the compositions of any two maps are zero.
Exactness at Hn (B). Let b ∈ Z n (B), then Hn (β)[b] = [β n b] is zero iff β n b = dz for some
z ∈ C n−1 . Let us lift this z to some y ∈ B n−1 , i.e., z = β n−1y. Then β n (b−dy) = dz−dz =
0, hence b − dy = αn a for some a ∈ An . Now a is a cocycle since αn (da) = d(b − dy) = 0,
and [b] = [b − dy] = Hn (α)[a].
Exactness at Hn (A). Let a ∈ Z n (A) be such that Hn (α)[a] = [αn a] is zero, i.e., αn a = db
for some b ∈ B n−1 . Then c = βb is a cocycle since dc = β n (db) = β n αn a = 0; and by the
definition of the connecting morphisms (in (1)), [a] = δ n−1 [c].
Exactness at Hn (C). Let c ∈ Z n (C) be such that ∂ n [c] = 0. Remember that this means
that c = βb and db = αa with [a] = 0, i.e., a = dx with x ∈ An−1 . But then db =
α(dx) = d(αx), so b − αx is a cocycle, and then c = β(b) = β(b − α(x)) implies that
[c] = Hn (β)[b − α(x)].
α
9.2. Mapping cones. The idea is that the cone of a map of complexes A − → B is a
complex C α which measures how far α is from being an isomorphism. However, the main
role of the mapping cone here is that it partially reformulates the short exact sequences
of complexes. It will turn out that the following data are the same:
(1) a mapping cone and its associated triangle

(2) a short exact sequence of complexes which splits in each degree.
This is the content of the next three lemmas. These results, together with the preceding
lemmas 9.1.4 and 9.1.3, all have analogues for the homotopy category of complex K(A)
which we meet in 9.3. These analogues (theorem 9.4.1), will be more complicated and
will give rise to a notion of a triangulated category.
α
9.2.1. Lemma. (a) A map A −
→ B defines a complex C called the cone of α by
• C n = B n ⊕An+1 ,
78
• dnC : B n ⊕An+1 −
→B n+1 ⊕An+2 combines the differentials in A and B and the map
α by:
def
d(bn ⊕an+1 ) = (dnB bn + αn+1 an+1 )⊕ − dn+1
A a
n+1
(b) The cone appears in a canonical triangle

• ••
α α α • ••n def
A− →C −
→B − → A[1], with αn (bn ) = bn ⊕0 and α (bn ⊕an+1 ) = an+1 .
Proof. (a) One has dC = (dB + α)⊕ − dA , hence

(dC )2 = dC ◦[(dB + α)⊕ − dA ] = (d2B + dB ◦α + α◦ − dA )⊕ (−dA )2 = 0,
in more details
dn+1 dn (bn ⊕an+1 ) = dn+1 [(dnB bn + αn+1 an+1 )⊕ − dn+1
A a
n+1
]
= (dn+1 n n
B dB b + d
n+1 n+1 n+1
α a + αn+1 (−dn+1
A a
n+1
)) ⊕ − dn+2 n+1 n+1
A dA a = 0.
• •• •
(b) The claim is just that α and α are maps of complexes. Clearly, (dC ◦α)(bn ) =
•
dC (bn ⊕0) = dB bb ⊕0 = α(dB bn ) and
•• ••
(dnA[1] ◦ α)(bn ⊕an+1 ) = (dnA[1] an+1 ) = −dn+1
A a
n+1
) = α(dB bn + αan+1 )⊕ − dn+1
A a
n+1
)
••
= ( α◦dC )(bn ⊕an+1 ).
9.2.2. Cone triangles and short exact sequences of complexes. We will see that the trian-
gles that arise from cones are reformulations of some short exact sequences of complexes.
α
Lemma. (a) For any map A −
→ B of complexes, the corresponding distinguished triangle
• ••
α α α
→ Cα −
→B −
A− → A[1] contains an exact sequence of complexes
• ••
α α
→ Cα −
→B −
0− → A[1] −
→0.
Moreover, this exact sequence of complexes splits canonically in each degree, i..e, short
• ••
α α
→ Bn −
exact sequences 0 − → (C α )n −
→ (A[1])n − → 0 in A have canonical splittings, i.e.,
identifications
∼
=
→ B n ⊕(A[1])n .
(C α )n −
(b) Actually, one can recover the whole triangle from the short exact sequence and its
splittings.
• ••
Proof. In (a), exactness is clear since α is the inclusion of the first factor and α is the
projection to the second factor. The splitting statement is the claim that the image of
B n ֒→Cαn has a complement (which is then automatically isomorphic to (A[1])n ), but the
image is B n ⊕0 and the complement is just 0⊕An+1 .
79
∼
= σn τn
(b) Identification (C α )n −
→ B n ⊕(A[1])n involves two maps (A[1])n ֒→ (C α )n ։ B n (ac-
αn+1
tually σ n determines τ n and vice versa). Now we can recover the map An+1 −−−→ B n+1
dn
as a part of the differential (C α )n −→
C
(C α )n+1 . Precisely, αn+1 is the composition
σn dn τ n+1
(A[1])n ֒→ (C α )n −→
C
(C α )n+1 ։ B n+1 .
(dB +α)⊕−dA
For this one just recalls that dnC is B n ⊕An+1 −−−−−−−−→ B n+1 ⊕An+2 , i.e.,
def
dnC (bn ⊕an+1 ) = (dnB bn + αn+1an+1 )⊕ − dn+1
A a
n+1
.
9.2.3. From a short exact sequences to a cone triangle. Conversely, we will see that any
short exact sequence that splits on each level, produces a cone triangle. However, the two
procedures of passing between cone triangles and short exact sequence that splits on each
level, turn out not to be inverse to each other. This gets resolved in 9.6.1 by adopting a
“correct” categorical setting – the homotopy category of complexes.
Lemma. A short exact sequence of complexes

φ ψ
0−
→P −
→Q −
→R−
→0
and a splitting Qn ∼
= P n ⊕Rn for each n ∈ Z, define canonically a distinguished triangle
• ••
α φ=α ψ= α
R[−1] −
→ P −−→ Q −−−→ (R[−1])[1].
(b) Moreover, this triangle is isomorphic in C(A) to a cone triangle.
σ n τn
(c) Explicitly, if the splitting is given by maps Rn ֒→ Qn ∼
= ։ P n , then the map
αn+1 dn
Q
Rn = (R[−1])n+1 −−−→ P n+1 is a component of the differential Qn −→ Qn+1 , i.e., it is
the composition
n dn τ n+1
n σ ∼
= Qn −→ Qn+1 ⊕P n+1 ∼
n+1 n n Q
α = (R ֒→ P ⊕R = P n+1⊕Rn+1 ։ P n+1).
Proof. (a) We just need to know that χ = α[1] : R − → P [1] is a map of complexes. Since
χn = τ n+1 ◦dnQ ◦σ n (r n ) one can decompose the action of dQ on the image of σ n into the P
and R components, by
dQ σ n (r n )) = τ n+1 dQ σ n (r n )] + σ n+1 dR (r n ), r n ∈ Rn .
Therefore, the differential on Q is given by (pn ∈ P n , r n ∈ Rn )
dQ (φn (pn ) + σ n (r n )) = dQ (φn (pn )) + dQ σ n (r n )) = φn+1 (dP pn )) + dQ σ n (r n ))
= [φn+1 (dP pn ))+τ n+1dQ σ n (r n )] + σ n+1 dR (r n ) = [φn+1 (dP pn ))+χn (r n )] + σ n+1 dR (r n ).
Now,
0 = d2Q (φn (pn ) + σ n (r n )) = [φn+2 (d2P pn )) + dQ χn (r n ) + χn+1 σ n+1 dR (r n )] + σ n+2 d2R (r n )
shows that χ◦dr = −dP ◦χ = dP [1] ◦χ.
80
(b) is clear: Q is isomorphic to the cone of α since Qn = τ n (P n )⊕σ n (Rn ) ∼

= P n ⊕Rn =
P n ⊕(R[−1])n+1 , and via these identification the differential in Q is precisely the cone
differential
dQ (pn ⊕r n ) = (dP pn + χn r n ) ⊕ dR r n = (dP pn + αn r n ) ⊕ − dR[−1] r n .
9.3. The homotopy category K(A) of complexes in A. We say that two maps of
α,β
complexes A −−→ B are homotopic (we denote this α mod β), if there is a sequence h of
maps hn : An −→B n−1 , such that
β − α = dh + hd, i.e., β n − α n = dB
n−1 n
h + hn+1 dnA .
We say that h is a homotopy from α to β.

α
A map of complexes A − → B is said to be a homotopical equivalence if there is a map β
in the opposite direction such that β◦α ≡ 1A and α◦β ≡ 1B .
9.3.1. Lemma. (a) Homotopic maps are the same on cohomology.

(b) Homotopical equivalences are quasi-isomorphisms.
(c) A complex A is homotopy equivalent to the zero object iff 1A = hd + dh. Then the
complex A is acyclic, i.e., H∗ (A) = 0.
(d) α ≡ β implies µ◦α ≡ µ◦β and α◦ν ≡ β◦ν.
Proof. (a) Denote for a ∈ Z n (A) by [a] the corresponding cohomology class in Hn (A).
n−1 n n−1 n
Then H(b)[a] − H(α)[a] = [dB h (a) + hn+1 dnA (a)] = [dB (h a)] = 0.
(b) If β◦α ≡ 1A and α◦β ≡ 1B then H(β)◦H(α) = H(β◦α) = H(1A ) = 1H(A) etc.
α β
(c) The map 0 −→ A with α = 0 is a homotopy equivalence if there is a map A −
→ 0 (then
necessarily β = 0) such that β◦α ≡ 10 and α◦β ≡ 1A . Since the LHS is always zero, the
only condition is that 0 ≡ 1A .
ν α µ
(d) If β − α = dB ◦h + h◦dA then for X −
→A −
→B −
→ Y one has µ◦β − µ◦α = dC ◦(µ◦h) +
(µ◦h)◦dA etc.
9.3.2. Homotopy category K(A). The objects are again just the complexes but the maps
are the homotopy classes [φ] of maps of complexes φ
def
HomK(A) (A, B) = HomC(A) (A, B)/ ≡ .
def
Now identity on A in K(A) is [1A ] and the composition is defined by [β]◦[α] = [β◦α],
this makes sense by the part (d) of the lemma.
81
9.3.3. Remarks. (1) Observe that for a homotopy equivalence α : A −

→B the correspond-
ing map in K(A), [α] : A −
→B is an isomorphism. So we have accomplished a part of our
long term goal – we have inverted some quasi-isomorphisms: the homotopy equivalences!
(2) More precisely, we know what are isomorphisms in K(A). The homotopy class [α] of
a map of complexes α, is an isomorphism in K(A) iff α is a homotopy equivalence!
9.4. The triangulated structure of K(A). K(A) is not an abelian category but there
are features that allow us to make similar computations:
• It is an additive category.
• It has shift functors [n].
• It has a class E of “distinguished triangles” (or “exact triangles”), defined as all
triangles isomorphic (in K(A) to a cone of a map of complexes.
• It has cohomology functors Hi : K(A) − →A.
9.4.1. Theorem. Distinguished triangles have the following properties

• (T0) The class of distinguished triangles is closed under isomorphisms.
• (T1) Any map α appears as the first map in some distinguished triangle.
1A 0
• (T2) For any object A, triangle A −→ A− →0 − →A[1] is distinguished.
α β γ
• (T3) (Rotation) If A −
→B −
→C −
→ A[1] is distinguished, so is
β γ −α[1]
B−
→C −
→ A[1] −−−→ B[1].
• (T4) Any diagram with distinguished rows
α β γ
A −−−→ B −−−→ C −−−→ A[1]
 

µy

νy
α′ β′ γ′
A′ −−−→ B ′ −−−→ C ′ −−−→ A′ [1]
can be completed to a morphism of triangles
α β γ
A −−−→ B −−−→ C −−−→ A[1]
   

µy

νy

ηy

µ[1]y
α′ β′ γ′
A′ −−−→ B ′ −−−→ C ′ −−−→ A′ [1].
α β γ=β◦α
• (T5) (Octahedral axiom) If maps A −
→B −
→ C and the composition A −−−−→ C,
appear in distinguished triangles
α α′ α′′
(1) A −
→B − → C1 −→ A[1],
β β′ β ′′
(2) B −
→C −
→ A1 −→ B[1],
γ γ′ γ ′′
(3) A −
→C −
→ B1 −→ C[1];
82
then there is a distinguished triangle

φ ψ χ
C1 −
→ B1 −
→ A1 −
→ C1 [1]
that fits into the commutative diagram
α α′ α′′
A −−−→ B −−−→ C1 −−−→ A[1]
   

=y

βy

φy

=y
γ γ′ γ ′′
A −−−→ C −−−→ B1 −−−→ A[1]
   

βy

=y

ψy

β[1]y
β β′ β ′′
B −−−→ C −−−→ C1 −−−→ B[1]
 

χy

=y
α′ [1]
C1 [1] ←−−− B1 [1]
9.4.2. Remarks. (1) Octahedral axiom (T5) is the most complicated and the least used
part.
(2) In (T4), the map γ is not unique nor is there a canonical choice. This is a source of
some subtleties in using triangulated categories.
Proof. (T0) is a part of the definition of E.
(T1) Any map in K(A), α ∈ HomK(A) (A, B) is a homotopy class [α] of some map of
α β γ
complexes α ∈ HomC(A) (A, B). The cone of α gives a triangle A −
→B −
→C −
→ A[1] in
[α] [β] [γ]
C(A) such that its image in K(A) is a distinguished triangle A −→ B −→ C −→ A[1] in
K(A) which starts with [α].
(T2) means that the cone C 1A = C of the identity map on A is isomorphic in K(A) to the
zero complex, i.e., that the cone C 1A is homotopically equivalent to zero. The homotopy
hn : C n −
→C n−1 is simply identity on the common summand and zero on its complement
def
An ⊕An+1 −
→An−1 ⊕An , hn (an ⊕an+1 ) = 0⊕an .
For that we calculate
n−1 n
(dC h + hn+1 dnC )(an ⊕an+1 ) = dC
n−1
(0⊕an ) + hn+1 [(dA an + 1A an+1 )⊕ − dn+1
A a
n+1
)]
= (1A an ⊕ − dA an ) + 0⊕(dA an + 1A an+1 ) = an ⊕an+1 .
(T4) The obvious strategy is to lift the first diagram in (T4) to the level of C(A). First,
•
0 0 α α
one can replace the rows with isomorphic ones which are cone triangles A −→ B −→
•• •′ ••′
α0 ′ α′0 ′ α0 α0
C α0 −→ A[1] and A −→ B −→ C α′0 −→ A′ [1] for maps α0 , α0′ in C(A) (these maps
83
are some representatives of homotopy classes α, α′ ). So, the diagram takes form (for any
representatives µ0 , ν0 of µ, ν)
• ••
[α0 ] [α0 ] [ α 0]
A −−−→ B −−−→ C α0 −−−→ A[1]
 

[µ0 ]y

[ν0 ]y
•′ ••′
[α′0 ] [α0 ] [ α 0]
′ ′
A −−−→ B −−−→ C α′0 −−−→ A′ [1]
It would be nice to lift the diagram completely into C(A) in the sense that we look
for representatives µ0 , ν0 of µ, ν so that the diagram in C(A) still commutes, then a
representative η0 of η would simply come from the functoriality (“naturality”) of the cone
construction. However we will make piece with the homotopic nature of the diagram and
incorporate the homotopy corrections. Homotopical commutativity [ν0 ]◦[α0 ] = [α0′ ]◦[µ0 ]
means that one has maps hn : B n − →(C α′0 )n−1 such that
ν0 ◦α0 − α0′ ◦µ0 = dh + hd.
η0
Now, we construct a map C α0 −→ C α′0 such that the diagram
0α β0 γ0
A −−−→ B −−−→ C α0 −−−→ [A1]
  

ν0 y

η0 y

µ0 y
α′ β′ γ′
A′ −−−
0
→ B ′ −−−
0 0
→ C α′0 −−−→ A′ [1]
commutes in C(A), by
→(B ′ )n ⊕(A′ )n+1 ,
η0 : B n ⊕An+1 − bn ⊕an+1 7→(ν0 bn + hn+1 an+1 )⊕µ0 an+1 .
α β γ
(T3) says that if one applies rotation to any cone triangle in C(A), A −
→B −
→ Cα −
→ A[1],
the resulting triangle
β γ −α[1]
B−→ Cα − → A[1] −−−→ B[1]
is isomorphic in K(A) to the cone triangle
β φ µ
B−
→ Cα −
→ Cβ −
→ B[1].
ζ
This requires a homotopy equivalence A −
→ C β such that the following diagram commutes
in K(A):
β γ −α[1]
B −−−→ C α −−−→ A[1] −−−→ B[1]
   

=y

=y

ζy

=y
β φ µ
B −−−→ C α −−−→ C β −−−→ B[1].
We define the map
ζn
An+1 = (A[1])n −→ (C β )n = (B n ⊕An+1 )⊕B n+1 , an+1 7→ 0⊕an+1 ⊕ − α(an+1 );
84
def
and in the opposite direction ξ(bn ⊕an+1 ⊕bn+1 ) = an+1 . It suffices to check that
(1) ζ and ξ are maps of complexes,

(2) ξ◦ζ = 1A[1] ,
(3) ζ◦ξ ≡ 1C β ,
(4) ζ◦γ = φ
(5) µ◦ζ = −α[1]
In (3), the homotopy h such that 1C β − ζ◦ξ = dh + hd, is the map hn : (C β )n −

→(C β )n−1
given by B n ⊕An+1 ⊕B n+1 ∋ bn ⊕an+1 ⊕bn+1 7→ 0⊕0⊕bn ∈ B n−1 ⊕An ⊕B n .
(T5) is a description of a certain “complicated” relation between cones of maps and
compositions of maps.
9.4.3. Triangulated categories. These are additive categories with a functor [1] (called
shift) and a class of distinguished triangles E, that satisfy the conditions (T 0 − T 5).
So, K(A) is our first triangulated category.
9.5. Long exact sequence of cohomologies.
9.5.1. Lemma. Any distinguished triangle X −

→Y −
→Z −
→ X[1] gives a long exact se-
quence of cohomologies
→Hi (X) −
··· − →Hi (Y ) −
→Hi (Z) −
→Hi+1 (X) −
→· · ·.
Proof. Up to isomorphism in K(A), we can replace the triangle X −

→Y −
→Z −
→X[1] with
• ••
α α α
→ Cα −
→Y −
the homotopy image of a cone triangle X − → X[1] in C(A). We have observed
• ••
α α α
→ Cα −
→Y −
that this triangle contains a short exact sequence of complexes 0 − → X[1] −
→0,
and the short exact sequence of complexes does indeed provide a long exact sequence of
cohomologies.
The proof was based on the relation of
9.6. Exact (distinguished) triangles and short exact sequences of complexes.

α
By definition distinguished triangles in K(A) come from maps A −
→ B in C(A). We will
just restate it as:
• distinguished triangles in K(A) come from short exact sequences in C(A) that
split on each level.
85
9.6.1. Lemma. (a) Any short exact sequence of complexes in C(A)
0−
→P −
→Q −
→R −
→0
which splits on each level, defines a distinguished triangle in K(A) of the form
−χ
P −
→Q −
→R −→ P [1].
(b) Any distinguished triangle in K(A) is isomorphic to one that comes from a short
exact sequence of complexes that splits on each level.
(c) Explicitly, in (a) the map χ in C(A) comes from a choice of splittings Qn ∼
= P n ⊕Rn .
χn dn
Q
The map Rn −→ P n+1 is the component of the differential Qn −→ Qn+1 , i.e.,
dn
χn = (Rn ֒→P n ⊕Rn ∼
= Qn −→ Qn+1 P n+1 ∼
Q
= P n+1⊕Rn+1 ։P n+1).
Proof. We know that the short exact sequence associated to a cone of a map splits
canonically on each level (lem 9.2.2). In the opposite direction, by lemma 9.2.3 any short
exact sequence of complexes
φ ψ
0−
→P −
→Q −
→R−
→0
with a splitting Qn ∼
= P n ⊕Rn on each level, defines a canonical cone triangle in C(A)
• ••
α φ=α ψ= α
R[−1] −
→ P −−→ Q −−−→ (R[−1])[1],
hence an exact triangle in K(A).

Since we are in K(A) we can rotate this triangle backwards using the property (T3) to
get an exact triangle P −→R − →Q − → P [1]. Now the two procedures of going between
short exact sequences in C(A) and exact triangles in K(A) are “inverse to each other”.
The formulas in (c) come from lemma 9.2.3c.
9.7. Extension of additive functors to homotopy categories. The following lemma

is quite obvious:
9.7.1. Lemma. (a) There is a canonical functor C(A) −

→K(A) which sends each complex
A to itself and each map of complexes φ to its homotopy class [φ].
F
(b) Any additive functor between abelian categories A −
→ B extends to a functor
C(F )
C(A) −−−→ C(B), here [C(F )A]n = F (An ) and the differential dnC(F ) is F (dn ).
86
C(F ) K(F )
(c) Moreover, C(A) −−−→ C(B), factors to a functor K(A) −−−→ K(B), i.e., there is a
unique functor K(F ) such that
C(F )
C(A) −−−→ C(B)
 
 
y y
K(F )
K(A) −−−→ K(B).
It is the same as C(F ) on objects and the action of K(F ) on homotopy classes of maps
of complexes comes from the action of C(F ) on maps of complexes.
9.8. Projective resolutions and homotopy.
9.8.1. Lemma. If we have two complexes

→P −n −
···− →P 0 −
→· · · − →a −
→0 −
→· · ·
→B −n −
···− →B 0 −
→· · · − →b −
→0 −
→· · ·
such that all P k are projective and the second complex is exact, then any map α : a −
→b
φ
lifts to a map P −→ B, i.e.,
· · · −−−→ P −n −−−→ · · · −−−→ P 0 −−−→ a −−−→ 0 −−−→ · · ·
   
   
φ−n y φ0 y αy =y
· · · −−−→ B −n −−−→ · · · −−−→ B 0 −−−→ b −−−→ 0 −−−→ · · ·.

(b) Any two such lifts are homotopic.
α◦ε
Proof. (a) Since εB : B 0 −
→ B is surjective and P 0 is projective, the map P 0 −−−→
A
B
phi0
factors through εB , i.e., it lifts to P 0 −−→ B 0 .
Actually, φ0 ◦dA −1 goes to Ker(εB ) since
εB (φ0 dA −1 ) = (α◦εA )dA = α◦0 = 0.
Exactness of the second complex, shows that dB −1 gives a surjective map dB −1 : B −1 − →
Ker(εB ). So, since P −1 is projective φ0 ◦dA −1 factors through dB −1 , giving a map φ−1 :
P −1 −
→B −1 , such that φ0 ◦dA −1 = dB −1 ◦φ−1 .
In this way we construct all φn inductively.
(b) If we have another solution ψ then
εB φ0 = α◦εA = εB ψ 0
gives εB (φ0 − ψ 0 ) = 0, hence
(φ0 − ψ 0 )P 0 ⊆Ker(εB ) = d−1 B (B −1 ).
87
→Im(dB −1 ) lifts to h0 : P 0 −
Now, since P 0 is projective, map φ0 − ψ 0 : P 0 − →B −1 , i.e.,
φ0 − ψ 0 = dB −1 ◦h0 .
One continuous similarly
dB −1 ◦(φ−1 − ψ −1 ) = (φ0 − ψ 0 )◦dA −1 = dB −1 ◦h0 ,
hence
Im[(φ−1 − ψ −1 ] − dB −1 ◦h0 ⊆ Ker(DB −1 ) = Im(d−2
B ).
So,
φ−1 − ψ −1 = h−1 ◦d−2
A
for some map h−1 : P −1 −
→B −2 . Etc.
corr (a) If P and Q are projective resolutions of objects a and b in A, then any map a −
→b
lifts uniquely to a map P −→Q.
(b) Any two projective resolutions of the same object of A are canonically isomorphic in
K(A).
9.9. Derived functors LF : A → K − (B) and RG : A → K + (B).
9.9.1. Lemma. If A has enough projectives:

(1) There is a canonical projective resolution functor P : A − →K − (A).
(2) For any additive right exact functor F : A − →B its left derived functor LF : A −
→
K ≤0 (B) is well defined by replacing objects with their projective resolutions
def
LF (A) = F (P(A))
and its zero cohomology is just the original functor F :
H0 [(LF )(A) ∼ = F (A).
Proof. (1) is clear from the corollary and then (2) follows.
10. Bicomplexes and the extension of resolutions and derived functors to

complexes
For a right exact functor F : A − → B we have defined its left derived functor LF : A − →
−
K (B) (if A has enough projectives) by replacing objects with their projective resolutions.
However, it is necessary to extend this construction to LF : K − (A) − → K − (B). For one
thing, for calculational reasons we need the property L(F ◦G)A = LF (LG(A)), but this
means that we have to apply LF to a complex.
We will need the notion of bicomplexes (roughly “complexes that stretch in a plane rather
then on a line”), as a tool for finding projective resolutions of complexes.
88
10.1. Filtered and graded objects. A graded object A of A is a sequence of objects

An ∈ A, n ∈ Z. We think of it as a sum A = ⊕Z An (and this is precise if the sum exists
in A).
An increasing (resp. decreasing) filtration on an object a ∈ A is an increasing (resp.
decreasing) sequence of subobjects an ֒→a, n ∈ Z. When we talk of a filtration F we
denote an by Fn a.
def
A filtration defines a graded object Gr(a) with Gr n (a) = an /an−1 (resp.
def
Gr n (a) = an /an+1 ). Also, a graded object A = ⊕ An has a canonical increas-
ing filtration An = ⊕i≤n Ai (if the sums exist), and a canonical decreasing filtration
An = ⊕i≥n Ai . (However, these grading and filtering are far from being inverse to each
other.)
We will be interesting in decreasing filtrations F of complexes (A = ⊕ Ai , d). This is a
sequence of subcomplexes Fn A = ⊕ Fn (Ai ), i.e.,
(1) · · ·F−1 (Ai )⊆ F0 (Ai )⊆ F1 (Ai )⊇· · ·⊆Ai is a filtration of Ai and

(2) d(Fn Ai )⊆ Fn (Ai+1 ).
A filtration F of a complex (A = ⊕ Ai , d) gives a filtration F of its cohomology groups

by
def
Fn [Hi (A)] = Im[Hi (Fn A) −
→Hi (A)].
10.2. Bicomplexes.
10.2.1. Bicomplexes. A bicomplex is a bigraded object B = ⊕p.q∈Z B p,q with differentials

d′ d′′
→ B p+1,q and B p,q −→ B p,q+1 , such that d = d′ + d′′ is also a differential. So we ask
B p,q −
that i.e.,
0 = d2 = (d′ + d′′ )2 = (d′ )2 + d′ d′′ + d′′ d′ + (d′′ )2 = d′ d′′ + d′′ d′ .

89
We draw a bicomplex as a two dimensional object:

d′ .. d′ .. d′ .. d′ .. d′ .. d′
· · · −−−→ . −−−→ . −−−→ . −−−→ . −−−→ . −−−→ · · ·
x x x x x x x
      
d 
′′ d′′  d′′  d′′  d′′  d′′  d 
′′
d′ d′ d′ d′ d′ d′
· · · −−−→ B −1,2 −−−→ B 0,2 −−−→ B 1,2 −−−→ B 2,2 −−−→ B 3,2 −−−→ · · ·
x x x x x x x
      
d 
′′ d 
′′ d 
′′ d 
′′ d 
′′ d 
′′ d 
′′
d′ d′ d′ d′ d′ d′
· · · −−−→ B −1,1 −−−→ B 0,1 −−−→ B 1,1 −−−→ B 2,1 −−−→ B 3,1 −−−→ · · ·
x x x x x x x
      
d′′  d′′  d′′  d′′  d′′  d′′  d′′ 
d′ d′ d′ d′ d′ d′
· · · −−−→ B −1,0 −−−→ B 0,0 −−−→ B 1,0 −−−→ B 2,0 −−−→ B 3,0 −−−→ · · ·
x x x x x x x
      
d′′  d′′  d′′  d′′  d′′  d′′  d′′ 
d′ d′ d′ d′ d′ d′
· · · −−−→ B −1,−1 −−−→ B 0,−1 −−−→ B 1,−1 −−−→ B 2,−1 −−−→ B 3,−1 −−−→ · · ·
x x x x x x x
      
d′′  d′′  d′′  d′′  d′′  d′′  d′′ 
d′ .. d′ . d′ . d′ . d′ . d′
· · · −−−→ . −−−→ .. −−−→ .. −−−→ .. −−−→ .. −−−→ · · · .
So, B pq has horizontal position p and height q, and d′ is a horizontal differential while d′′
is a vertical differential.
10.2.2. Remarks. (1) Anti-commutativity relation d′ d′′ + d′′ d′ = 0 can be interpreted as

commutativity in the correct framework: the super-mathematics.
(2) If d′ and d′′ would happen to commute, we would have to correct one of these (say
def
replace d′′ by (dˇ′′ )p,q = (−1)p (d′′ )p,q ).
10.2.3. The total complex T ot(B) of a bicomplex and the cohomology of a bicomplex. The
def
total complex of a bicomplex is the complex (T ot(B), d) with T ot(B)n = ⊕p+q=n B p,q .
The cohomology of B is by definition the cohomology of the complex T ot(B).
10.2.4. Decreasing filtrations ′ F and ′′ F on a bicomplex and on the total complex. The
fact that the complex T ot(B) has come from a bicomplex will be used to produce two
decreasing filtrations on the complex T ot(B). Actually, any complex A has a stupid
decreasing filtration F where the subcomplex Fn A is obtained by erasing all terms Ak
with k < n:
def
Fn A = (· · · −
→0 −
→0 − →An − →An+1 − →An+2 −→· · ·).
In turn, any bicomplex B has two decreasing filtrations ′ F and ′′ F . The sub-bicomplex
′
Fi B of a bicomplex B is obtained by erasing the part of B which is on the left from the
90
ith column, and symmetrically, ′′ Fj B is obtained by erasing beneath the j th row. Say, the
subbicomplex ′ Fi B is given by
d′ . d′ . d′ .. d′ .. d′ .. d′
· · · −−−→ .. −−−→ .. −−−→ . −−−→ . −−−→ . −−−→ · · ·
x x x x x x x
      
d 
′′ d 
′′ d 
′′ d 
′′ d 
′′ d 
′′ d 
′′
d′ d′ d′ d′ d′ d′
· · · −−−→ 0 −−−→ B i,2 −−−→ B i+1,2 −−−→ B i+2,2 −−−→ B i+3,2 −−−→ · · ·
x x x x x x x
      
d 
′′ d 
′′ d 
′′ d 
′′ d 
′′ d 
′′ d 
′′
d′ d′ d′ d′ d′ d′
· · · −−−→ 0 −−−→ B i,1 −−−→ B i+1,1 −−−→ B i+2,1 −−−→ B i+3,1 −−−→ · · ·
x x x x x x x
      
d 
′′ d 
′′ d 
′′ d 
′′ d 
′′ d 
′′ d 
′′
d′ d′ d′ d′ d′ d′
· · · −−−→ 0 −−−→ B i,0 −−−→ B i+1,0 −−−→ B i+2,0 −−−→ B i+3,0 −−−→ · · ·
x x x x x x x
      
d 
′′ d 
′′ d 
′′ d 
′′ d 
′′ d 
′′ d 
′′
d′ d′ d′ d′ d′ d′
· · · −−−→ 0 −−−→ B i,−1 −−−→ B i+1,−1 −−−→ B i+2,−1 −−−→ B i+3,−1 −−−→ · · ·
x x x x x x x
      
d′′  d′′  d′′  d′′  d′′  d′′  d′′ 
d′ . d′ . d′ .. d′ .. d′ .. d′
· · · −−−→ .. −−−→ .. −−−→ . −−−→ . −−−→ . −−−→ · · ·.
This then induces filtrations on the total complex, say
def
[′ Fi T ot(B)]n = T ot(′ Fi B)n = ⊕p+q=n, p≥i B p,q ⊆ T ot(B)n ⊇ ⊕p+q=n, q≥j B p,q = ′′ Fj [T ot(B)n ].
Finally, ‘F and ′′ F induce filtrations on the cohomology
′ def
Fi Hn (T ot B) = Im[Hn (T ot ′ Fi B) −
→Hn (T ot ′ Fi B)],
so the cohomology groups are extensions of pieces
′ def
′
Fi Hn (T ot B)
Gri [Hn (T ot B)] = ′F n .
i+1 H (T ot B)
These pieces can be calculated by the method of spectral sequences (see 10.5).
10.3. Partial cohomologies. By taking the “horizontal” cohomology we obtain a bi-

graded object ′ H(B) with
d′
′ p.q def p •,q Ker(B p,q −
→ B p+1.q )
H(B) = H (B ) = d′
.
Im(B p−1,q −
→ B p.q )
The vertical differential d′′ on B factors to a differential on ′ H(B) which we denote again
by d′′ :
′ d′′
H(B)p,q −→ ′ H(B)p,q+1 .
91
Next, we take the “vertical” cohomology of ′ H(B) (i.e., with respect to the new d′′ ), and
get a bigraded object ′′ H(′ H(B)) with
d′′
′′ ′ p,q def q ′ p,• Ker[′ H(B)p,q −→ ′ H(B)p.q+1 ]
( H(B)) = H ( H(B) ) = d′′
.
Im[′ H(B)p,q−1 −→ ′ H(B)p.q ]
One defines ′′ H(B) and ′ H(′′ H(B)) by switching the roles of the first and second coordi-
nates.
10.3.1. Remark. Constructions ′ H(B) and ′′ H(B) are upper bounds on the cohomology of
the bicomplex, and ′ H(′′ H(B)) and ′′ H(′ H(B)) are even better upper bounds. The precise
relation is given via the notion of spectral sequences (see 10.5).
10.4. Resolutions of complexes. An injective resolution of a complex A ∈ C ∗ A) is

→ I such that all I n are injective objects of the abelian category
a quasi-isomorphism A −
A. The next two theorems will state that injective resolutions of complexes exist and and
are can be chosen compatible with short exact sequences of complexes.
10.4.1. Theorem. If A has enough injectives any A ∈ C + (A) has an injective resolution.
More precisely,
(a) There is a bicomplex
d′ . d′ . d′ . d′ . d′ . d′
· · · −−−→ .. −−−→ .. −−−→ .. −−−→ .. −−−→ .. −−−→ · · ·
x x x x x x x
      
d′′  d′′  d′′  d′′  d′′  d′′  d′′ 
d′ d′ d′ d′ d′ d′
· · · −−−→ I −1,2 −−−→ I 0,2 −−−→ I 1,2 −−−→ I 2,2 −−−→ I 3,2 −−−→ · · ·
x x x x x x x
      
d′′  d′′  d′′  d′′  d′′  d′′  d′′ 
d′ d′ d′ d′ d′ d′
· · · −−−→ I −1,1 −−−→ I 0,1 −−−→ I 1,1 −−−→ I 2,1 −−−→ I 3,1 −−−→ · · ·
x x x x x x x
      
d 
′′ d 
′′ d 
′′ d 
′′ d 
′′ d 
′′ d 
′′
d′ d′ d′ d′ d′ d′
· · · −−−→ I −1,0 −−−→ I 0,0 −−−→ I 1,0 −−−→ I 2,0 −−−→ I 3,0 −−−→ · · ·
x x x x x x x
      
d 
′′ d 
′′ d 
′′ d 
′′ d 
′′ d 
′′ d 
′′
d′ d′ d′ d′ d′ d′
· · · −−−→ A−1 −−−→ A0 −−−→ A1 −−−→ A2 −−−→ A3 −−−→ · · ·
x x x x x x x
      
d 
′′ d 
′′ d 
′′ d 
′′ d 
′′ d 
′′ d 
′′
d′ d′ d′ d′ d′ d′
· · · −−−→ 0 −−−→ 0 −−−→ 0 −−−→ 0 −−−→ 0 −−−→ · · ·
such that the columns are injective resolutions of terms in the complex A.
(b) For any such bicomplex the canonical map A −
→T ot(I) is an injective resolution of A.
92
10.4.2. Theorem. Let P and R be projective resolutions of objects A and C that appear
in an exact sequence 0A − →B − →C −
→ 0. Then Q = P ⊕R appears in a short exact
sequence of projective resolutions
We start with the baby case of the theorem 10.4.2.
10.4.3. Lemma. Assume that A has enough injectives. A short exact sequence in A can
always be lifted to a short exact sequence of injective resolutions. More precisely,
(a) Let I and K be injective resolutions of objects A and C that appear in an exact
sequence 0 − →A − →B − →C − → 0. Then all J n = I n ⊕K n appear in a short exact sequence
of injective resolutions
.. . . . .
. −−−→ .. −−−→ .. −−−→ .. −−−→ ..
x x x x x
 2 2 2  
 dI dJ dK 
α2 β2
0 −−−→ I 2 −−−→ J 2 −−−→ K 2 −−−→ 0
x x x x x
 1 1 1  
 dI dJ dK 
α1 β1
0 −−−→ I 1 −−−→ J 1 −−−→ K 1 −−−→ 0
x x x x x
    
 dI 
0 dJ 
0 dK 
0

α0 β0
0 −−−→ I 0 −−−→ J 0 −−−→ K 0 −−−→ 0
x x x x x
    
 ι A ι B ι C 
α β
0 −−−→ A −−−→ B −−−→ C −−−→ 0
x x x x x
    
    
0 −−−→ 0 −−−→ 0 −−−→ 0 −−−→ 0
with α the inclusion of the first summand and β n the projection to the second summand.
n
(b) Any short exact sequence of injective resolutions

0 −−−→ I −−−→ J −−−→ K −−−→ 0
x x x x x
    
 ιA  ιB  ιC  
α β
0 −−−→ A −−−→ B −−−→ C −−−→ 0
necessarily splits on each level, i.e., J n ∼
= I n ⊕K n . (However, complex J is not a sum of
I and K.)
Proof. (a) We need to define ιB so that the middle column is a resolution and the diagram
commutes. Define ιB : B − →J 0 = I 0 ⊕K 0 by
def
ιB (b) = e
ιA (b)⊕ιC (βb)
93
where e → I 0 is any extension of ιA : A −

ιA : B − → I 0 (it exists since I 0 is injective). This
choice ensures that the two squares that contain ιB commute. Moreover, ιB is injective
since the kernel of the second component ιC ◦β is Ker(β) = A and on A⊆B ιB is ιA .
To continue in this way, we denote B e = Coker(ιB ) = (I 0 ⊕K 0 )/ιB (B) and notice that the
second projection gives a surjection from B e def
e to C = K 0 /ιC (C). Its kernel is the inverse of
ιC (C)⊆K 0 under the second projection, taken modulo ιB (B), i.e., (I 0 ⊕ιC (C))/ιB (B) ∼ =
0 def e
I /ιA (A) = A. So we have a commutative diagram
α1 β1
0 −−−→ I 1 −−−→ I 1 ⊕K 1 −−−→ K 1 −−−→ 0
x x x x
   
 dI 
0 dK 
0

1 β1
0 −−−→ A e −−α−→ e
B −−−→ C e −−−→ 0
x x x x x
    
 q
′ q q ′′  
α0 β0
0 −−−→ I 0 −−−→ I 0 ⊕K 0 −−−→ K 0 −−−→ 0
x x x x x
    
 ι A ι B ι C 
α β
0 −−−→ A −−−→ B −−−→ C −−−→ 0
in which d1I and d1K are factorizations of d1I and d1K through the the canonical quotient
maps q ′ and q ′′ . Maps d1I and d1K are embeddings, and we need to supply an embedding
1
Be −d→
J
I 1 ⊕K 1 which would give two more commuting squares, then d1J is defined as a
composition of d1J and the quotient map q. However, this is precisely the problem we
solved in the first step.
φn
(b) Since I n is injective, one can extend the identity map on I n to J n −→ I n , and gives a
splitting i.e., a complement Ker(φn ) to I n in J n .
10.4.4. Existence of injective resolutions of complexes: proof of the theorem 10.4.2. .

(0) About the proof. We choose injective resolutions of coboundaries and cohomologies
→ J n = J •,n , Hn (A) −
of A: B n (A) − → K n = K •,n . Then the nth column of I is I ν , n =
n n n+1
J ⊕K ⊕J and the horizontal differentials are the compositions
I p,q = J p,q ⊕K p,q ⊕J p,q+1 ։J p,q+1⊆ J p,q+1⊕K p,q+1⊕J p,q+2 = I p,q+1.
The vertical differential make the nth column into a complex I •,n such that
• J •,n ⊆ I •,n is a subcomplex, and

• K •,n ⊆ I •,n /J •,n is a subcomplex.
94
The fist task is to choose suitable injective resolutions of everything in site. We start by
choosing injective resolutions of coboundaries and cohomologies
B n (A) −
→Bn = Bn,• , Hn (A) −
→Hn = Hn,• .
Now, cocycles are an extension of cohomologies and coboundaries, i.e., there is an exact
sequence, and then An is an extension of Z n (A) and B n+1 , i.e., there are exact sequences
→B n (A) −
0− →Hn (A) −
→Z n (A) − →0 and →Z n (A) −
0− →An −
→B n+1 (A) −
→0.
By the preceding lemma 10.4.3 we can combine H and B to get injective resolutions of
these short exact sequences
0 −−−→ Bn −−−→ Zn −−−→ Hn −−−→ 0
x x x x x
    
 ι ι ι 
0 −−−→ B n (A) −−−→ Z n (A) −−−→ Hn (A) −−−→ 0
and
0 −−−→ Zn −−−→ An −−−→ Bn+1 −−−→ 0
x x x x x
    
 ι ι ι 
0 −−−→ Z n (A) −−−→ An −−−→ B n+1 (A) −−−→ 0
Since An is an injective resolution of An , we can use them to build a bicomplex as in
the the part (a) of the theorem, with the vertical differentials d′′ the differentials in An ’s.
d′ n
Now we need the horizontal differentials An −→ An+1 ), these are the compositions
d′ n def ⊆ ⊆
(An −→ An+1 ) = [An −
→Bn+1 −
→ Z n+1 −
→ An+1 ].
Since these are morphisms of complexes vertical and horizontal differentials commute,
however this is easily corrected to d′ d′′ + d′′ d′ = 0 by 10.2.2(2).
A ι
(b) Let I be a bicomplex from the part (a). A canonical map A −→ T ot(I) comes from
n ιAn n,0 n
A −−→ I ⊆ T ort(I) . Moreover, ιA is a quasi isomorphism since all maps are quasi
isomorphisms. This is easy to see directly and there is an elegant tool for such problems
– the concept of spectral sequences (see 10.5).
10.4.5. Existence of injective resolutions of exact sequences of complexes: proof of the

theorem 10.4.2. This is a combination of ideas in proofs of the lemma 10.4.3 and the
theorem 10.4.1.
corr If A has enough injectives any short exact sequence in C + (A) defines a distinguished
triangle in K(A).
Proof. By the theorem any short exact sequence in C + (A) is isomorphic in K(A) to a
short exact sequence of complexes with injective objects. Since such sequence splits on
each level it defines a distinguished triangle in K(A).
95
10.5. Spectral sequences. In general, spectral sequences are associated to filtered com-
plexes but we will be happy with the special case of spectral sequences associated to
bicomplexes. The idea of a spectral sequence is to relate the cohomology of a complex
with the cohomology of a simplified complex which may be more accessible.
10.5.1. Abstract definition. A spectral sequence is a sequence of complexes (Er , dr ), r ≥ 0

such that
(1) Er is bigraded, Er = ⊕ Erp,q and the differential dr has type (r, 1 − r), i.e.,
dr : Erp,q −
→Erp+r,q+1−r .
(2) Er+1 = H(Er , dr ), i.e.,
p,q d d
Er+1 = Ker(Erp,q −→
r
Erp+r,q+1−r )/Im(Erp−r,q+r−1 −→
r
Erp,q ).
10.5.2. The limit of a spectral sequence. The limit can be defined for any spectral sequence
but the most interesting case is when the spectral sequence stabilizes. We will say that
the (p, q)-term stabilizes in the r th term if for s ≥ r the differentials ds from the (p, q)-term
and into the (p, q)-term are zero. Then clearly
Er(p,q) ∼
= Er+1 ∼ = Er+2 ∼
(p,q) (p,q)
= · · ·.
Then we say that
(p,q) def
E∞ = Er(p,q) .
We will say that the spectral sequence stabilizes (degenerates) in the r th term if ds =
0, s ≥ r. Then E∞ is by definition Er .
10.5.3. Theorem. To any bicomplex B one associates two (“symmetric”) spectral se-
quences ′ E and ′′ E. The first one satisfies
(1) ′
E0p,q = B p,q
′ p,q
(2) E1 = ′′ Hp,q (B)
′ p,q
(3) E2 = ′ Hp [′′ H•,q (B)
(4) ′ p,q
E∞ = ′ Grp [Hp+q (T ot B)]
10.5.4. Remark. The basic consequence is that the piece ′ Grp [Hn (T ot B)] of Hn (T ot B) is
i,j
a subquotient of ′ Hn−p [′′ H•,q (B)], hence in particular, of ′′ Hp,n−p(B) (since Er+1 is always
i,j n
a subquotient of Er ). This gives upper bounds on the dimension of H (T ot B).
10.5.5. Degeneration criteria. One general result that guarantees the stabilization of each
term in some Er (with r depending on p and q):
Theorem. If for each n the sum (T ot B)n = ⊕i+j=n B i,j is finite, then each term stabilizes.
This applies for instance if B lies in first quadrant.
96
Remarks. (a) The simplest way to prove that a spectral sequence degenerates at Er is if
one can see that for s ≥ r and any p, q one of objects Erp,q or Erp+s,q+1−s is zero.
(2) Since ds is of type (s, 1 − s) it changes the total parity p + q. So one way to satisfy
the requirement in (1) is to know that all p + q with Erpq 6= 0 are of the same parity.
(3) We are fond of bicomplexes such that the first spectral sequence degenerates at E2 ,
then we recover the constituents of Hn (T ot B) from partial cohomology
Gr• [Hn (T ot B)] ∼
= ⊕p+q=n ′ Hp [′′ H•,n−p (B).
11. Derived categories of abelian categories
Historically the notion of triangulated categories has been discovered independently in

• Algebra: derived category of an abelian category A is a convenient setting for
doing homological algebra – i.e., the calculus of complexes in A (and much more).
• Topology: the stable homotopy theory deals with the category whose objects are
topological spaces (rather then complexes!), but the total structure is the same
as for a derived category D(A) (shift is giving by the operation of suspension of
topological spaces, exact triangles come from the topological construction of the
mapping cone Cf corresponding to a continuous map f : X − → Y of topological
spaces ...)
Here we will only be concerned with the (more popular) appearance of derived categories
in algebra.
11.1. Derived category D(A) of an abelian category A. The objects of D(A) are
again the complexes in A, however HomD(A) (A, B) is an equivalence class of diagrams in
K(A):
] D(A) (A, B) be the class of all diagrams in K(A) Of the form
(1) Let Hom
φ s
A−
→X ←
−B, with s a quasi-isomorphism.
φi i s
(2) Two diagrams A − → Xi ← − B, i = 1, 2; are equivalent iff they are quasi-isomorphic
ui
to a third diagram, in the sense that there are quasi-isomorphisms Xi − → X, i =
1, 2; such that
φ1 s
A −−−→ X1 ←−1−− B
  

=y

u1 y

=y
u1 ◦φ1 = u2 ◦φ2 φ s
, i.e., A −−−→ X ←−−− B
u1 ◦s1 = u2 ◦s2 x x x

=

u2 

=
φ2 s
A −−−→ X2 ←−2−− B.
97
11.1.1. Symmetry. It follows from the next lemma that one can equivalently represent
s φ
morphisms by diagrams A ← −C − → B with s a quasi-isomorphism.
s
11.1.2. Lemma. If s is a quasi-isomorphism a diagram A ←
−B in K(A) can be canonically
completed to a commutative diagram
ψ
Y −−−→ B
 

uy

sy
φ
A −−−→ X
with u a quasi-isomorphism. (Conversely, one can also complete u, ψ to s, φ.)
φ s
11.1.3. Remarks. (0) We can denote the map represented by the diagram A − →X ← −B by
−1
[s, φ], its intuitive meaning is that it the morphism s ◦φ once s is inverted. Identifications
of diagrams should correspond to equalities s1 −1 ◦φ1 = s2 −1 ◦φ2 , to compare it with the
requirement (2) rewrite it as first as (u1 ◦s1 )−1 ◦(u1 ◦φ1 ) = (u2 ◦s2 )−1 ◦(u2 ◦φ2 ), and then as
(u1 ◦φ1 )◦(u2 ◦φ2 )−1 = (u1 ◦s1 )◦(u2 ◦s2 )−1 ; then (2) actually says that both sides are equal
to the identity 1X .
(1) The composition of (equivalence classes of) diagrams is based on lemma 11.1.2.
φ s ψ u
Putting together two diagrams [s, φ] and [u, ψ] gives A −
→X ← −B − → Y ← − C. Now
s ψ ψ′ s′
lemma 11.1.2 allows us to replace the inner part X ←
−B − → Y with s′
→ Y by X ←− Z −
φ ψ′ s′ u
a quasi-isomorphism. This gives a diagram A −
→ X ←− Z −
→Y ←
− C and we define
def
= [u, ψ]◦[s, φ]◦[s′ ◦u, ψ ′◦φ].
(2) The above procedure inverts quasi-isomorphisms in K(A). We can actually invert
quasi-isomorphisms directly in C(A), however lemma 11.1.2 does not hold in C(A), so
one is forced to take a more complicated definition of maps as coming from long diagrams
φ0 0 s φ11 s 2 φ2 s φn−1 n sn−1 φn s
A −→ X0 ←
− B0 −→ X0 ←
− B1 −→ X2 ←
− · · · −−−→ Xn−1 ←−− Bn−1 −→ Xn ←− Bn = B
which are composed in the obvious way.
11.1.4. Theorem. D(A) is a triangulated category if we define the exact triangles as im-
ages of exact triangles in K(A).
One can also define triangulated subcategories D ? (A) for ? ∈ {+, b, −}, these are full
subcategories of all complexes A such that H • (A) ∈ C ? (A) (with zero differential). For
Z⊆Z one can also define a full subcategory D Z (A) by again requiring a condition on
cohomology: that H • (A) ∈ C Z (A).
98
11.1.5. The origin of exact triangles in D(A). They can be associated to either of the
following:
(1) a map of complexes,
(2) a short exact sequence of complexes that splits on each level,
(3) if A has enough injectives, any short exact sequence of complexes in C + (A), and
if A has enough projectives, any short exact sequence of complexes in C − (A).
11.1.6. Cohomology functors Hi : D(A) − → A. Hi is clearly defined on objects, on mor-

def
phisms it is well defined by Hi ([s, φ]) = Hi (s)−1 ◦Hi (φ).
11.2. Truncations. These are functors

τ≤n τ≤n
D ≤n (A) ←−− D(A) −−→ D ≥n ,
they come with canonical maps τ≤n A −
→A −
→τ≥n A defined by
· · · −−−→ An−2 −−−→ An−1 −−−→ Z n (A) −−−→ 0 −−−→ 0 −−−→ · · · (τ≤n (A))
        
        
y y y y y y yy y yy
· · · −−−→ An−2 −−−→ An−1 −−−→ An −−−→ An+1 −−−→ An+2 −−−→ · · · (τ≤n (A))
x x x x x x xx x xx
        
        
· · · −−−→ 0 −−−→ 0 −−−→ An /B n (A) −−−→ An+1 −−−→ An+2 −−−→ · · · (τ≥n (A))
11.2.1. Lemma. (a) τ≤n is the left adjoint to the inclusion D ≤n (A)⊆D(A), and τ≥n is the
right adjoint to the inclusion D ≥n (A)⊆D(A).
n o n o
i
Hi (A) i≥n
(b) Hi (τ≤n A) = H 0(A) i≤n
i>n
, and H i
(τ≥n A) = 0 i<n
.
11.3. Inclusion A֒→ D(A).
Lemma. (a) By interpreting each A ∈ A as a complex concentrated in degree zero, one

identifies A with a full subcategory of C(A), K(A) or D(A).
(b) The inclusion of A into the full subcategory D 0 (A) (all complexes A with H i (A) = 0
for i 6= 0) is an equivalence of categories. (The difference is that D 0 (A) is closed in D(A)
under isomorphisms and A is not).
Proof. In (a) we need to see that for A, B in A the canonical map HomA (A, B) −
→
HomX (A, B is an isomorphism:
• when X = C(A) since any homotopy between two complexes concentrated in
degree zero is clearly 0 (recall that hn : AN −
→B n−1 ),
• when X = D(A) one shows that
99
φ s
(1) any diagram in K(A) of the form A −
→X ←
−B with s a quasi-isomorphism, is
H0 (φ) H0 (s) H0 (s)−1 H0 (φ)
equivalent to a diagram in A A −−−→ H0 (X) ←−−− B, hence to A −−−−−−−−→
1 H0 (s)−1 H0 (φ)
B ←B− B, hence it comes from a map A −−−−−−−−→ B in A,
αi 1
(2) Two diagrams of the form A −→ B ←B− B, are equivalent iff α1 = α2 .
Remark. Part (b) of the lemma describes A inside D(A) (up to equivalence) by only using
the functors H i on D(A).
11.4. Homotopy description of the derived category. The following provides a down
to earth description of the derived category and gives us a way to calculate in the derived
category.
11.4.1. Theorem. Let IA be the full subcategory of A consisting of all injective objects.
(a) IA is an abelian subcategory.
(b) If A has enough injectives the canonical functors
σ τ
K + (IA ) −
→ D + (IA ) −
→ D + (A)
are equivalences of categories.
Proof. (A sketch.) One first observes that
• Quasi-isomorphism between complexes in C + (I) are always homotopical equiva-
lences.
This tells us that quasi-isomorphisms in K + (I) are actually isomorphisms in K + (I). Since
quasi-isomorphisms in K + (A) are already invertible the passage to D + (I) obviously gives
an equivalence K + (I) − →D + (I).
We know that any complex A in C + (A) is quasi-isomorphic to its injective resolution I,
and also any map A′ − → A′′ is quasi-isomorphic to a map of injective resolutions I ′ −
→ I ′′ .
This is the surjectivity of τ on objects and morphisms.
The remaining observation is that
• For I, J ∈ D + (I) map HomD(I) (I, J) −
→HomD(A) (I, J) is injective.
12. Derived functors
Let F : A −
→B be an additive functor between abelian categories.
12.1. Derived functors Rp F : A → A. Suppose that A has enough injectives and set
Rp F (A) = Hp (F I) for any injective resolution I of A.
We call these the (right) derived functors of F .
100
12.1.1. Theorem. (a) Functors Rp are well defined.

(b) R0 F ∼
= F.
α β
→A′ −
(c) Any short exact sequence 0 − → A′′ −
→A− → 0 in A defines a long exact sequence
of derived functors
R0 F (α) RF n (β) ∂0 ∂ n−1 Rn F (α) RF n (β) ∂
→R0 F (A′ ) −−−−→ R0 F (A) −−−−→ R0 F (A′′ ) −→ · · · −−→ Rn F (A′ ) −−−−→ Rn F (A) −−−−→ Rn F (A′′ ) −
0−
Proof. (a) and (b) follow from 9.9. (c) follows from the lemma 10.4.3. First we can choose
α β α β
an injective resolution 0 − →I ′ −→I − → I ′′ −
→0 of the short exact sequence 0 − →A′ − →A − →
α β
A′′ −→ 0, and then we apply F to it. The sequence of complexes 0 − → F (I ′ ) −
→ F (I) −
→
F (I ′′ ) −
→ 0 is exact since the short exact sequence of resolutions splits level-wise (lemma
10.4.3b), and F is additive.
12.1.2. Remark. , even if F is not left exact the above construction produces a left exact
functor RF .
12.2. Derived functors RF : D + (A) → D + (B). We start with the definition of its right
F
derived functor RF as the universal one among all extensions of A − → B to D + (A) − →
D + (B). Then we see that the “replacement by injective resolution” construction satisfies
the universality property.
12.2.1. Notion of the derived functor of F . A functor between two triangulated categories
is said to be a morphism of triangulated categories (a triangulated or ∂-functor) if it
preserves all structure of these categories:
(1) it is additive,
(2) it preserves shifts
(3) it preserves exact triangles
The right derived functor of an additive functor F : A −

→ B is the universal one among
+ +
all extensions of F to D (A) −→D (B). Precisely, it consists of the following data
• A triangulated functor RF : D + (A) −

→D + (B),
ξ
• a morphism of functors iB ◦F −
→RF ◦iA ;
and these data should satisfy the universality property:

e morphism ξe factors uniquely through ξ, i.e., there
g , ξ),
• for any other such pair (RF
µ ξe
is a unique morphism of functors RF − → RFg such that ξe = µ◦ξ, i.e., iB ◦F − →
ξ µ◦1A
e ◦iA is the composition iB ◦F −
RF →RF ◦iA −−−→ e ◦iA .
RF
101
12.2.2. Remark. (0) For A ∈ A, iA A is A viewed as a complex, and ξA relates objects

ξA
that should be the same if RF extends F : RF (iA A) −→ iB (F A). So, intuitively ξ is the
F
A −−−→ B
 

“commutativity constraint” for the diagram, iA y  , it takes care of the
iB y
RF
D + (A) −−−→ D + (B)
fact that the functors iB ◦F and RF ◦iA are not literally the same but only canonically
isomorphic (though the universality property does not require ξ to be an isomorphism, in
practice it will be an isomorphism).
(1) The problem is when does the universal extension exist? The simplest case is when
F is exact, then then one can define RF simply as F acting on complexes. In general the
most useful criterion is
12.2.3. Theorem. Suppose that A has enough injectives. Then for any additive functor
F
A− →B :
(a) RF : D + (A) −
→D(B) exists.
(b) For any complex A ∈ D + (A), there is a canonical isomorphism (RF )A ∼
= F (I) for
any injective resolution I of A.
(c) In particular, for A ∈ A the cohomologies of (RF )A are the derived functors (Ri F )A
introduced above.
σ τ
Proof. Recall from 11.4.1 that one has equivalences K + (IA ) −
→ D + (IA ) −
→ D + (A),
def F
and from recall that an additive functor FI = (IA ⊆A −
→ B) has a canonical extension
K(FI ) : K(A) −→K(B). So we can define RF as a composition
def (τ ◦σ)−1 K(FI )
RF = [D + (A) −−−−→ K + (IA ) −−−→ K + (B) −
→D + (B)].
Our construction satisfies the description of RF in (b). We have (τ σ)−1 I = I and a quasi-
def
isomorphism A − → I. Therefore, RF (A) = [K(FI )◦(τ ◦σ)−1 ] A = K(FI ) I = F (I).
(c) follows from (b).
It remains to check that our RF is the universal extension.
RF preserves shifts by its definition. Any exact triangle in D ( A) is isomorphic to a
α β γ α β γ
triangle A −
→B −
→C −
→ A[1] that comes from an exact sequence 0 −
→A−
→B −
→C −
→0
α β
→ A′ −
in C(A) that splits on each level. Moreover, we can replace the exact sequence 0 − →
γ ′′
A− →A − → 0 with an isomorphic (in D(A)) short exact sequence of injective resolutions
a b F (a)
→ I′ −
0 − → I ′′ −
→I − → F (I ′) −−→
→ 0. Since it also splits on each level its F -image 0 −
F (b)
F (I) −−→ F (I ′′ ) −
→0 is an exact sequence in C(B). Therefore it defines an exact triangle
F (a) F (b) γ
e
F (I ′ ) −−→ F (I) −−→ F (I ′′ ) −
→ F (I ′ )[1] in D(A). To see that this is the triangle we
102
observe that by definition F (I ′ ) = RF (A′ ) F (I) = RF (A) F (I ′′ ) = RF (A′′ ), and also

F (a) = RF (α) and F (b) = RF (β). It remains to see that γe = RF (γ). For this recall
that γ and e
γ are defined using splittings of the first and second row of
α β γ
0 −−−→ A′ −−−→ A −−−→ A′′ −−−→ 0
  
  
ι′ y ιy ι′′ y
a b
0 −−−→ I ′ −−−→ I −−−→ I ′′ −−−→ 0.
So, we need to be able to choose the splitting of the second row compatible with the one
in the first row. However, this is clearly possible by the construction of the second row.
12.3. Usefulness of the derived category. We list a few more reasons to rejoice in
derived categories.
12.3.1. Some historical reasons for the introduction of derived categories. (1) For a left
exact functor F we have derived functors Ri F and and then concept of derived category
allows us to glue them into one functor RF . Does this make a difference?
It is easy to check from definitions that for two left exact functors F and G one has
R(G◦F ) ∼ = RG◦RF . If we instead work with Ri F , Rj G and Rn (G◦F ) the above for-
mula degenerates to a weak and complicated relation in terms of spectral sequences. So,
seemingly more complicated construction RF is more natural and has better properties
then a bunch of functors Ri F .
(2) If G : B −→ C is right exact and B has enough projectives then one similarly has
i
→C and they glue to LG : D − (B) −
functors L G : B − →D − (C).
F G
However, if one has a combination of two functors A −→B− → C and F is left exact while
b LG◦RF b
G is right exact, we have a functor D (A) −−−−→ D (C) which is often very useful but
has no obvious analogue as a family of functors from A to C since the composition G◦F
need not be neither left nor right exact.
(3) Some of essential objects and tools exist only on the level of derived categories: the
dualizing sheaves in topology and algebraic geometry, the perverse sheaves.
12.3.2. Some unexpected gains. It turns out that in practice many deep relations between
abelian categories A and B become understandable only on the level of derived categories,
for instance D(A) and D(B) are sometimes equivalent though the abelian categories A
and B are very different.
103
Part 4. Homeworks
Homework 1
1.1. Homology of a torus. Calculate the dimensions of the homology groups of the
torus T 2 using a triangulation into a simplicial complex.
Homework policy.
(1) The grade will be based on homeworks.

(2) It is a good idea to discuss the problems that resist reasonable effort with others,
but the actual writing should be done in isolation.
(3) It is not necessary to turn out perfect homeworks, rather it is expected that your
work shows that that you are acquiring the material of the course.
(4) The amount of detail in your expositions (i) should be based on what is most
useful to you, i.e., you need to make it perfectly clear to yourself, (ii) but it should
be also written so that another person can follow the line of thought.
(5) If you are quite sure that you understand certain problem (or a part of), to the
extent that writing the solution would be a waste of time, just say so and formulate
what is happening in a few words. I reserve the right to ask about the missing
ideas. Homeworks are due next week on Thursday.
Homework 2
2.1. Homology of spheres. S n is the boundary of σn , hence has a triangulation by

all facets of σn except σn itself.
• (a) Use this to compute the dimension of homology groups H∗ (S 3 , k) for a field k.
• (b) Compute H∗ (S n , k) by an intuitive argument, using a covering by two cells.
2.2. Euler characteristic. The Euler characteristic of a finite sequence of finite di-
def P
mensional vector spaces A• is e(A• ) = i i
i (−1) dim(A ). For a finite complex of vector
spaces C • show that e(C • ) = e(H • (C • )).
2.3. Duality for k-modules. Prove lemma 2.2.1, i.e. show that for M ∈ ml (k), M ∗ is
really in mr (k) and for N ∈ mr (k), N ∗ is in ml (k).
2.4. Biduality for k-modules. Prove lemma 2.2.2, i.e., show that
• (1) for m ∈ M ι(m) is really in (M ∗ )∗ , and

• (2) ιM is a map of k-modules.
104
2.5 Dualizing maps. For a module M = k ∈ ml (k) show that

def
(a) The map that assigns to a ∈ k the operator of right multiplication Ra x = x·a, gives
R
an isomorphism of right k-modules k −
→ k∗ .
(b) ιk is an isomorphism.
Homework 3
3.1. The dual of the structure sheaf of a point in a plane. For the point Y in a
plane A2 show that
LD[O(Y )] ∼
= O(Y )[−2]
(i.e., calculate the cohomology of DP for the resolution P • from 2.7.2).
•
Homework 4
SHEAVES
4.0. The sheaf of solutions is constant on an interval. Let X be an interval in R

and ∗ be the differential equation
y (n) + a1 (t)y (n−1) + · · · + a0 (t) = 0
with ai ∈ C ∞ (X). Find some identification of the sheaf Sol∗ of solutions of ∗, with a
constant sheaf VX for some vector space V ∼
= Cn .
MULTI-LINEAR ALGEBRA: Tensor product of modules over a ring
If needed see S.Lang’s Algebra or N.Bourbaki’s Algebra.
Let A be a ring with a unit 1.
4.1. Tensor product of A-modules. For a left A-module U and a right A-module V ,
we define a free abelian group F = FU,V , with a basis U×V : F = ⊕u∈U, v∈V Z·(u, v). The
def
tensor product of U and V is the abelian group U⊗A V defined as a quotient U⊗A V = F/S
of F by the subgroup R generated by the elements of one of the following forms (here
u, ui ∈ U, v, vi ∈ V, a ∈ A):
(1) (u1 +u2 , v)−(u1, v)−(u2, v), (2) (u, v1 +v2 )−(u, v1 )−(u, v2 ), (3) (u·a, v)−(u, a·v), .
The image of (u, v) ∈ F in U⊗A V is denoted u⊗v. Let π : U×V →U⊗A V be the compo-
sition of maps U×V ֒→F →U⊗A V , so that π(u, v) = u⊗v.
(a) Show that
(a1 ) (u1 + u2 )⊗v = u1 ⊗v + u2 ⊗v,
(a2 ) u⊗(v1 + v2 ) = u⊗v1 + u⊗v2 ,
(a3 ) (u·a)⊗v) = u⊗(a·v).
105
Pn
(b) Show that each element of U⊗A V is a finite sum of the form i=1 ui ⊗vi , for some
ui ∈ U, vi ∈ V .
4.2. The universal property of the tensor product ⊗A . We say that a map
φ : U×V →H with values in an abelian group H, is A-balanced if it satisfies the conditions
φ(u1 +u2 , v) = φ(u1, v)+φ(u2, v), φ(u, v1 +v2 ) = φ(u, v1)+φ(u, v2) , φ(u·a, v) = φ(u, a·v),
for u, ui ∈ U, v, vi ∈ V, a ∈ A.
Show that the balanced maps φ : U×V →H are in a one-to-one correspondence with the
def
morphisms of abelian groups ψ : U⊗A V →H, by ψ7→φ = ψ◦π.
[The above notion of ”balanced maps” is a version of the notion of bilinear maps
which makes sense even for non-commutative rings A. In one direction of the above
one-to-one correspondence above says that the tensor product reformulates balanced
maps in terms of linear maps. In the opposite direction, the universal property tells
us how to constructs maps from a tensor product U⊗A V – one needs to construct a
balanced map from the product U×V . (This direction is used all the time!)]
4.3. Cancellation, Quotient as tensoring, Additivity, Free Modules. .

(a) Cancellation. Show that the map
α
X X
U⊗A A −
→ A, α( ui ⊗ai ) = ui ·ai ,
is (i) well defined, (ii) an isomorphism of abelian groups.
(b) Quotient by relations as tensoring. For each left ideal I in A, the map
α
X X
U⊗A A/I −→ U/U·I, α[ ui ⊗(ai + I)] = ( ui ·ai ) + U·I ,
is (i) well defined, (ii) an isomorphism of abelian groups.
(c) Additivity of tensoring. Let V = ⊕i∈I Vi be a direct sum of left A-modules, then
U⊗A V = ⊕i∈I U⊗A Vi .
(d) Free Modules. If V is a free left A-module with a basis vi , i ∈ I, then U⊗A V ∼
= UI,
i.e., it is a sum of I copies of U.
4.4. Tensoring of bimodules. If U is a bimodule for a pair of rings (R, A) and V is

a bimodule for a pair of rings (A, S), show that U⊗A V is a bimodule for (R, S).
4.5. Tensoring over a commutative ring. (a) If A is a commutative ring then the
left and right modules coincide, say a right A-module U becomes a left A-module with
def
the action defined by a·u = u·a.
(b) In that case the above general construction allows us to tensor two left modules U, V :
def
U⊗A V = ⊕u∈U, v∈V Z·(u, v)/R where R is the subgroup generated by the elements of
106
the form (1), (2) and (3′ ) (a·u, v) − (u, a·v). One has again properties (a1 ), (a2 ) and
(a′ 3 ) (a·u)⊗v) = u⊗(a·v), and the analogous universal property.
(c) Moreover, for a commutative ring A, U⊗A V is again an A-module with the action
def
a· (u⊗v) = (a· u)⊗v = u⊗(a·v).
4.6. Tensoring over a field. If A is a field then A-modules are vector spaces over A.
Let ui , i ∈ I, vj , j ∈ J, be bases of U and V , show that ui ⊗vj , i ∈ I, j ∈ J; is a basis of
U⊗A V and dim(U⊗A V ) = dim(U)· dim(V ).
4.7. Tensoring of finite abelian groups over Z. Show that Zn ⊗Zm ∼

= Zk for some
Z
k and calculate k.
Homework 5
D-modules
i
5.0. Inverse image and the direct image of D-modules. Let Y = {0}⊆ X =
q def
A1 −
→ Z = pt. Calculate the dimensions of cohomology groups (Ls io )M = H i [Lio (M)]
def
and (Ls q∗ )M = H i [Lq∗ (M)] for
(1) M = DX ,
(2) M = O(X),
def
(3) M = δ for δ = DX /DX z,
(4) M = M∗ for ∗ given by zy ′ = λy (the answer will depend somewhat on λ).
(Recall that i0 M = M/zM and q∗ M = M/∂M.)
MULTI-LINEAR ALGEBRA: Multiple tensor products and Tensor algebras

of modules
5.1. Multiple tensor products. Let Mi be an (Ai−1 , Ai )-bimodule for i = 1, ..., n.

The tensor product M1 ⊗ M2 ⊗ · · · ⊗ Mn is defined as a quotient of a free abelian group
A1 A2 An−1
F with a basis M1 ×· · ·×Mn , by the subgroup A generated by the elements of one of the
following forms (mi ∈ Mi , ai ∈ Ai ):
(1) (m1 , ..., m′i + m′′i , ..., mn ) − (m1 , ..., m′i , ..., mn ) − (m1 , ..., m′′i , ..., mn ),
(2) (m1 , ..., mi−1 ·ai , ..., mn ) − (m1 , ..., ai ·mi , ..., mn ).
107
The image of (m1 , ..., mn ) ∈ F in U⊗A V is denoted m1 ⊗· · ·⊗mn . Let π :

M1 ×· · ·×Mn →M1 ⊗ M2 ⊗ · · · ⊗ Mn be the composition M1 ×· · ·×Mn ֒→F →M1 ⊗ M2 ⊗ · · · ⊗ Mn ,
A1 A2 An−1 A1 A2 An−1
so that π(m1 , ..., mn ) = m1 ⊗· · ·⊗mn .
(a) Show that
(1) m1 ⊗· · ·⊗m′i + m′′i ⊗· · ·⊗mn = m1 ⊗· · ·⊗m′i ⊗· · ·⊗mn + m1 ⊗· · ·⊗m′′i ⊗· · ·⊗mn ,
(2) m1 ⊗· · ·⊗mi−1 ·ai ⊗mi ⊗· · ·⊗mn = m1 ⊗· · ·⊗mi−1 ⊗ai ·mi ⊗· · ·⊗mn .

P
(b) Each element of M1 ⊗ M2 ⊗ · · · ⊗ Mn is a finite sum of the form pk=1 mk,1 ⊗· · ·⊗mk,n .
A1 A2 An−1
(c) Formulate and prove the universal property of tensor products.

(e) Show that M1 ⊗ M2 ⊗ · · · ⊗ Mn is an (A0 , An )-bimodule.
A1 A2 An−1
(f) This definition is associative in the sense that there are canonical isomorphisms
(M1 ⊗ · · · ⊗ Mp1 ) ⊗ (Mp1 +1 ⊗ · · · ⊗ Mp1 +p2 ) ⊗ · · ·⊗(Mp1 +···+pk−1 +1 ⊗· · ·⊗Mp1 +···+pk )
A1 Ap1 −1 A p1 Ap1 +1 Ap1 +p2 −1 Ap1 +p2
∼
= M1 ⊗ M2 ⊗ · · · ⊗ Mp1 +···+pk .
A1 A2 Ap1 +···+pk −1
(e) For n = 2 this notion of a tensor product agrees with the one introduced previously.
5.2. Tensor algebra of an A-module. Let M be a module for a commutative ring

A with a unit. We will denote the n-tuple tensor product M⊗A · · ·⊗A M by TAn (M) =
n def
⊗A M = M ⊗n . For n = 0 this is - by definition - A itself (so it does not depend on M).
For n = 1 this is the module M.
def P n
(a) Show that T (M) = n≥0 T (M) has a unique structure of an associative A-algebra,
such that for all p, q ≥ 0 and mi , nj ∈ M,
(m1 ⊗· · ·⊗mp )·(n1 ⊗· · ·⊗nq ) = m1 ⊗· · ·⊗mp ⊗n1 ⊗· · ·⊗nq .
For this algebra structure structure m1 ⊗· · ·⊗mp is the product m1 · · ·mp of mi ∈ M ⊗1 =
M.
(b) Universal property of the tensor algebra. For each A-algebra B restriction
Homassoc. A−alg. with 1 (T M, B) ∋ φ7→φ|M ∈ HomA−modules (M, B),
is a bijection.
[We say that T (M) is that A-algebra defined by the A-module M, or that T (M) is universal
among A-algebras B endowed with a map of A-modules M→B.]
(c) Algebra A(M, R) defined by an A-module M and the relations R. What
we mean by algebraic relations between elements of an A-module M are intuitively the
108
conditions of type
k
X
(∗) ai ·mi,1 · · ·mi,ni = 0
i=1
for some ai ∈ A, mi,j ∈ M. The precise meaning of that is that the expression on the left
P
hand side defines an element r = ki=1 ai ·mi,1 ⊗· · ·⊗mi,ni of the tensor algebra. So any
def
set of such relations defines (i) a subset R⊆T (M), (ii) an A-algebra A(M, R) = T (M)/ <
R > where < R > denotes the 2sided ideal in T (M) generated by R, with (iii) a canonical
def
map of A-modules ι = [M⊆T (M)։A(M, R)].
Show that for each A-algebra B, Homassoc. A−alg. with 1 [A(M, R), B] is naturally identified
P
with the set of all β ∈ HomA−modules (M, B), such that for all r = ki=1 ai ·mi,1 ⊗· · ·⊗mi,ni
in R, the following relation between (images of) elements of M holds in B:
Pk
i=1 ai ·β(mi,1 )· · ·β(mi,ni ) = 0.
[Therefore, an algebraic relation of type (∗) between elements of M acquires meaning in
any algebra B supplied with a map of A-modules M→B. Algebra A(M, R) is universal
among all such A-algebras B that satisfy relations from R.]
5.3. Right exactness of tensor products. (a) Let A be a ring, L a right A-module and
0→M ′ →M→M ′′ →0 a short exact sequence of left A-modules. Show that there is a short
exact sequence of abelian groups L⊗A M ′ →L⊗A M→L⊗A M ′′ →0. Find an example when
the sequence 0→L⊗A M ′ →L⊗A M→L⊗A M ′′ →0 is not exact, i.e., L⊗A M ′ 6 ⊆L⊗A M.
(b) Let A be a ring, L a right A-module and M a left A-module. Show that any algebra
morphism φ : B→A, gives a surjective map L⊗M→L⊗A M, with the kernel generated by
B
elements of the form x·a⊗y − x⊗a·y, x ∈ L, y ∈ M, a ∈ A.
MULTI-LINEAR ALGEBRA: Exterior and Symmetric algebras of a module
5.4. Exterior algebra of an A-module. Let M be a module for a commutative ring

• •
A with a unit. The exterior algebra ∧M = ∧M is the associative A-algebra with 1
A
generated by M and by anti-commutativity relations R = {x⊗y + y⊗x, x, y ∈ M}. The
•
multiplication operation in ∧M is denoted ∧, so that the image of m1 ⊗· · ·⊗mn ∈ T (M)
• n
in ∧M is denoted m1 ∧· · ·∧mn ∈ ∧M.
•
(a) ∧M is a graded algebra. Show that (a0 ) the ideal I =< R > in T (M) is
def
homogeneous, i.e., I = ⊕n≥0 I for I n = I ∩ T n M. Show that the quotient algebra
n
• • n n def •
∧M = T (M)/I satisfies (a1 ) ∧M ∼= ⊕n≥0 ∧M for ∧M = T n (M)/I n ; and (a2 ) ∧M is a
n m n+m
graded algebra, i.e., ∧M ·∧M⊆ ∧ M, n, m ≥ 0.
5.5. Universal property of the exterior algebra. Show that for each
•
A-algebra B, Homassoc. A−alg. with 1 [∧M, B] can be identified with a set of all
109
φ : HomA−moduli (M, B), such that the φ-images of elements of M anti-commute in B,

i.e., φ(y)φ(x) = −φ(x)φ(y), x, y ∈ M.
0
5.6. Basic properties of exterior algebras. (a) Low degrees. ∧M = T 0 (M) = A
1
and ∧M = T 1 (M) = M.
(b) Bilinear forms extend to exterior algebras. For any A-modules L and M, and
any A-bilinear map < , >: L×M→A (i.e., linear in each variable); there is a unique A-
n n
bilinear map < , >: ∧L×∧M→A, such that < l1 ∧· · ·hn , m1 ∧· · ·∧mn >= det(< li , mj >).
k
(c) Free modules If M is a free A-module with a basis e1 , ..., ed , then ∧M is a free
A-module with a basis eJ = ej1 ∧· · ·∧ejk , indexed by all subsets J = {j1 < · · · < jn }⊆I
with k elements.
•
(d) Dimension. dim(∧ Cn ) = 2n .
5.7. Symmetric algebra of an A-module. Let M be a module for a commutative

ring A with a unit. The symmetric algebra S(M) = SA (M) of M is the algebra generated
by M and the relations R = {x⊗y − y⊗x, x, y ∈ M}.
(a) S(M) is a graded algebra. Show that
• (a0 ) the ideal I =< R > in T (M) is homogeneous and S(M) = T (M)/I satisfies
def
• (a1 ) S(M) ∼ = ⊕n≥0 S n (M) for S n (M) = T n (M)/I n ; and
• (a2 ) S(M) is a graded algebra.
5.8. Universal property of the symmetric algebra. (a) Show that (i) S(A) is
commutative, (ii) for any commutative A-algebra B, Homassoc. A−alg. with 1 [S(M), B] can
be identified with HomA−modules (M, B).
5.9. Basic properties of symmetric algebras. (a) T 0 (M) = S 0 (M) = A and

T 1 (M) = S 1 (M) = M.
a free A-module with a basis ei , i ∈ I, then S n (M) is a free A-module with a
(b) If M is Q
basis eJ = i∈I eJi i , indexed by all maps J : I→N with the integral n.
(c) The algebra of polynomials C[x1 , ..., xn ] is isomorphic to the symmetric algebra
S(Cx1 ⊕· · ·⊕Cxn ).
The next topic (Koszul resolution) will use both symmetric and exterior algebras.
110
Homework 6
6.0. Koszul complex. Koszul complex of the vector space V is the complex
d−k−1 k d−k d−3 −2 d−2 k d−1 0
· · · −−−→ ∧ V ⊗ S(V ) −−→ · · · −−→ ∧ V ⊗ S(V ) −−→ ∧ V ⊗ S(V ) −−→ ∧ V ⊗ S(V ) −
→0 −
→· · ·
If dim(V ) = n it is finite
n d−n d−3 −2 d−2 d−1
···−
→0 −
→ ∧ V ⊗ S(V ) −−→ · · · −−→ ∧ V ⊗ S(V ) −−→ V ⊗ S(V ) −−→ V ⊗ S(V ) −
→0 −
→· · ·.
(a) Show that the maps d−k are well defined by the formula (here vi ∈ V, f ∈ S(V ) andb
means that we are omitting this term)
k
X
−k
d (v1 ∧· · ·∧vk ⊗ f ) = (−1)k−i v1 ∧· · ·∧b
vi ∧· · ·∧vk ⊗ vi f.
1
(b) Show that this is a complex.

(c) Consider the Koszul complex for V = C and V = C2 , where have you used it before?
6.1. An example of inductive limit. Let (I, ≤) be {1, 2, 3, ...} with the order i ≤ j
if i divides j. In Ab let Ai = Q/Z for all i ∈ I, and let φji be the multiplication by j/i
when i divides j. This is an inductive system and lim Ai =?.
→
6.2. An example of projective limit. Let (I, ≤) be N = {0, 1, ...} with the standard
order. In Rings let kn = C[x]/xn+1 and for i ≤ j let φij be the obvious quotient map.
This is a projective system and lim ki =?.
←
6.3. Passage to a final subset in an inductive limit. A subset K of a partially

ordered set is called final if for each i ∈ I there is some k ∈ K with i ≤ k. Let
(ci , i ∈ I; φji , i ≤ j) be an inductive system in a category C. Show that if
• K⊆I is final and

• for any i′ , i′′ ∈ I there is some i ∈ I that dominates both,
the canonical map

ι
lim ck −
→ lim ci
→ →
k∈K i∈I
is an isomorphism. (More precisely, if one of these objects exists so does the other and
there is a canonical isomorphism.)
111
⊆ ⊇
6.4. Fibered products of sets. (a) If in the diagram of sets A − →S ←
− B, S is a
point, then the fibered product A×B is just the usual product A×B.
S
⊆ ⊇
(b) If A −
→S ←
− B are inclusions of subsets of S the fibered product A×B is just the
S
intersection A ∩ B.
p q
(c) In Sets fibered products exist and the fibered product of A −
→S ←
− B is the set
A×B =?.
S
6.5. Use of coordinates in Linear Algebra. Let k be a field and Vk the category
such that Ob(Vk ) = N = {0, 1, ...} and Hom(n, m) = Mmn (matrices with m rows and n
ι
columns. Let ι be the functor Vk −→Vectfk g into finite dimensional vector spaces, given by
ι(n) = kn and for a matrix α ∈ Mmn , ια : km − →kn is the multiplication by α.
(a) Show that ι is an equivalence of categories.
(b) Let us choose for each finite dimensional vector space V a basis eV1 , ..., eVdim(V ) . Use
this to construct a functor G which is both left and right adjoint to ι. (It plays the role
of the inverse of ι.)
6.6. Construction of projective limits of sets. Let (I, ≤) be a partially ordered set
such that for any i, j ∈ I there is some k ∈ I such that i ≤ k ≥ j. Let (Ai )i∈I and maps
(φij : Aj −
→Ai )i≤j be a projective system of sets. Then
Y
lim Ai is the subset of Ai consisting of all families a = (ai )i∈I such that φij aj = ai .]
←
i∈I
6.7. Construction of inductive limits of sets. Let (I, ≤) be a partially ordered set
such that for any i, j ∈ I there is some k ∈ I such that i ≤ k ≥ j. Let (Ai )i∈I and maps
(φji : Ai −
→Aj )i≤j be an inductive system of sets.
def
(a) Show that the relation ∼ defined on the disjoint union ⊔i∈I Ai = ∪i∈I Ai ×{i} by
• (a, i) ∼ (b, j) (for a ∈ Ai , b ∈ Aj ), if there is some k ≥ i, j such that “a = b in
Ak ”, i.e., if φki a = φkj a ,
is an equivalence relation.
(b) Show that lim Ai is the quotient [⊔i∈I Ai ]/ ∼ of the disjoint union by the above
→
equivalence relation.
Homework 7
7.1. Functoriality of modules under the change of rings. For a map of rings
φ:k− →h show that the direct image of modules φ∗ M = h⊗k M has a right adjoint.
112
7.2. Representable functors. Find functors which are (co)represented by an initial

(resp. final) object.
π
7.3. Direct image of sheaves. Let X −
→ Y be a map of topological spaces.
(a) Show that for a sheaf M on X, formula
def
π∗ (M) (V ) = M(π −1 V ),
defines a sheaf π∗ M on Y .
∗ π
(b) Show that this gives a functor Sheaves(X) −→ Sheaves(Y ).
7.4. Examples of sheaves. (a) Let φ : Y →X be a continuous map of topological

spaces. Denote S(U) the set of all sections over U i.e. all continuous maps s : U→Y such
def
that s(x) ∈ Yx = φ−1 x, x ∈ U. Show that S is a sheaf of sets.
(b) Which of the following classes of functions on R form a presheaf, a sheaf: (0) all func-
tions, (1) continuous functions, (2) smooth functions, (3) even functions, (4) L2 functions,
(5) compactly supported continuous functions, (6) constant functions. Why?
(c) Let p : V →M be a holomorphic vector bundle and denote the set of holomorphic
sections over an open set U ⊂ M by V(U). Show that V is a sheaf of complex vector
spaces.
(d) Let M be a complex manifold and OM (U) the holomorphic functions on U. Show
that OM is a sheaf of algebras on M.
7.5. Examples of maps of sheaves. Consider the following sheaves of vector spaces
on M: constant sheaf CM and the zero sheaf 0 = {0}M (if M is any topological space),
∞
sheaf CM of smooth functions on M (if M is open in Rn ), sheaf of holomorphic functions
HM (if M is open in C).
(1) If M is open in C define canonical maps of sheaves of vector spaces:
a b ∞ c
0M →CM →CM →OM .
(2) Which of the following are maps of sheaves of vector spaces (and why?):
∞
(a) (Differentiation) CM (U) ∋ f 7→ f ′ ∈ CM∞
(U),
∞ 2 ∞
(b) (Squaring) CM (U) ∋ f 7→ f ∈ CM (U),
(c) (multiplication by a function) OM (U) ∋ f 7→ zf ∈ OM (U),
∞ ∞
(d) (translation by 1) CM (U) ∋ f (x) 7→ f (x + 1) ∈ CM (U).
α
7.6. Kernels, images and quotients of maps of sheaves. Let A −
→ B be one of
the following maps of sheaves of vector spaces:
• (a) M = R and α = ∂x
∂ ∞
, i.e., αU : CM ∞
(U)→CM (U) is given by f 7→ f ′ ;
• (b) M = C and α = z, i.e., αU : OM (U)→OM (U) by f 7→ zf.
113
In both cases
α
U
(1) Show that K(U) = Ker[A(U) −→ B(U)] is a sheaf and a subsheaf of A, and it is
the kernel of α,
αU
(2) Show that I(U) = Im[A(U) −→ B(U)] is a sub-presheaf of B.
(3) Show that C(U) = B(U)/I(U) is a presheaf.
(4) Describe the sheaf K.
(5) Calculate the sheaf C associated to the presheaf C, and show that this is the
cokernel of α.
(6) Calculate the subsheaf I of B associated to I, and show that this is the image of
α.
(Describe the sections I(U), Q(U) and the stalks Ix , Qx , as well as you can, for instance
find the dimensions of these vector spaces.)
7.7. Stalks. (a) Show that any map of sheaves φ : A→B defines for each x ∈ M a map
of stalks Ax →Bx .
(b) Show that maps of sheaves φ, ψ : A→B are the same iff the maps on stalks are the
same, i.e., φx = ψx for each x ∈ M.
(c) Map of sheaves φ : A→B is an isomorphism iff φx is an isomorphism for each x ∈ M.

Homological Algebra With The Example of D-Modules: Date: ?

Uploaded by

Copyright:

Available Formats

Homological Algebra With The Example of D-Modules: Date: ?

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Homological Algebra With The Example of D-Modules: Date: ?

Uploaded by

Copyright:

Available Formats

HOMOLOGICAL ALGEBRA

WITH THE EXAMPLE OF D-MODULES

6.1. Additive categories 54

Part 3. Derived categories of abelian categories 75

10.3. Partial cohomologies 90

Part 4. Homeworks 103

4.7. Tensoring of finite abelian groups over Z 106

So Okada has suggested a number of corrections.

0. What does the Homological algebra do?

0.1. Some examples of applications. Algebraic analysis of linear differential

0.3. The texts.

1. Algebraic topology: Homology from triangulations

• In order to extract information about a topological space X we make a choice

1.1. Linear simplices. Calculation of homology of topological spaces is based on a geo-

1.1.3. Orientations of simplices. We say that an ordering of the set of vertices V of a

1.1.4. Barycentric coordinates

1.2. Topological simplices. A topological i-simplex Y is a topological space Y with a

1.3. Triangulations. The idea of triangulation is to present a given topological space as

1.4. Simplicial complexes. A triangulation of a topological space X is a covering X =

• (SC1) any facet of any simplex Xi is one of the simplices in X,

1.4.1. Triangulations of S 1 . They are obtained by choosing an orientation of S 1 and n

1.5. Complex C∗ (X, T ; k). Calculation of homology of X includes a choice of “coeffi-

The space of i-chains for (X, T ) is the free k-module

1.5.1. Notion of a complex. A complex of cochains is a sequence of k-modules and maps

1.5.2. Lemma. C∗ (X, T ; k) is a complex (of chains).

2. Duality for modules over rings

2.1. Rings. Some of the classes of interesting rings k

2.2.4. Lemma. For a module M = k ∈ ml (k):

2.2.5. Lemma. ιM is an isomorphism for any finitely generated free k-module.

2.2.6. Sublemma. For two k- modules P, Q; (a) (P ⊕Q)∗ ∼

2.3.2. Duality operation on complexes. The dual of a complex of k-modules C • = (· · · −

2.3.5. Lemma. The biduality map ιC • is an isomorphism for any complex C • in

2.4.1. Exact complexes and short exact sequence. A complex of k-modules C • = (· · · − →

2.4.3. Resolutions. A left resolution of a module M is an exact complex

2.4.4. Morphisms of complexes. A map of complexes f : A• −→B • is a system of maps f n

2.4.6. Lemma. Any module M has a free resolution.

2.4.8. Projective modules. We say that a k-module P is projective if it is a summand of a

2.5.3. Localization of categories. Localization of a ring A with respect to a subset S⊆A

2.6. Derived versions of constructions.

2.6.1. Improving an object M. Let A = m(k) and M ∈ A. We improve M by replacing

Combination of α and δ is intuitively clear: this is the description of M in terms of maps

2.6.2. Any (additive) functor D : A − → B extends to complexes. Let A = m(k) and

IY = xi ·k and therefore O(Y ) = k/ xi ·k is isomorphic to C as a ring (C-valued

Therefore, computing the adjoints by lemma 2.3.3 gives

The cohomology of this complex is easy to compute (nothing new!), it gives

2.8. Homology and derived constructions.

3.1.2. Sheaves on a topological space. A sheaf of sets S on a topological space (X, T )

3.2.1. Solutions of differential equations. Solutions of a system (∗) of differential equation

Lemma. Sol(∗) is a constant sheaf on X.

3.3. Projective line P1 over C. P1 = C ∪ ∞ can be covered by U1 = U = C and

3.3.1. Lemma. Γ(P1 , OP1 ) = C.

3.4.3. Lemma. (Č • (U, A), d•) is a complex, i.e., d.◦dn−1 = 0.

3.4.5. Lemma. Ȟ 0 (X, U; A) = Γ(A).

Lemma. Ȟ • (P1 , OP1 ) = Ȟ 0 (P1 , OP1 ) = C.

3.5. Cohomology of vector bundles.

3.6.1. Lemma. Cn [x, y] = Γ(P1 , Ln ), n = 0, 1, 2, .. is the list of

3.7. Cohomology of the constant sheaf is dual to homology.

Lemma. For X = S 1 , Ȟ ∗ (X, U; kX ) is dual to H∗ (X, k).

4.1.2. We will use D-modules as an illustration of usefulness of homological algebra: