Claus Michael Ringel, Jan Schröer - Representation Theory of Algebras I - Modules (Preliminary Version Feb 10, 2009) (2009) PDF

February 10, 2009
REPRESENTATION THEORY OF ALGEBRAS I: MODULES
CLAUS MICHAEL RINGEL AND JAN SCHRÖER
Abstract. This is the first part of a planned book “Introduction to Representa-

tion Theory of Algebras”. These notes are based on a 3 Semester Lecture Course
on Representation Theory held by the first author at the University of Bielefeld
from Autumn 1993 and on an (ongoing) 4 Semester Lecture Course by the second
author at the University of Bonn from Autumn 2006.
Preliminary version
Contents
1. Introduction 10
1.1. About this course 10
1.2. Notation and conventions 11
1.3. Acknowledgements 11
Part 1. Modules I: J-Modules 12

2. Basic terminology 12
2.1. J-modules 12
2.2. Isomorphisms of J-modules 12
2.3. Submodules 12
2.4. Factor modules 14
2.5. The lattice of submodules 15
2.6. Examples 16
2.7. Decompositons and direct sums of modules 18

2.8. Products of modules 19
2.9. Examples: Nilpotent endomorphisms 19
1
2 CLAUS MICHAEL RINGEL AND JAN SCHRÖER
2.10. Exercises 22
3. Homomorphisms between modules 23
3.1. Homomorphisms 23
3.2. Definition of a ring 24
3.3. Definition of an algebra 25
3.4. Homomorphism Theorems 25
3.5. Homomorphisms between direct sums 27

3.6. Idempotents and direct decompositions 29
3.7. Split monomorphisms and split epimorphisms 30
3.8. Short exact sequences and Hom-functors 30
3.9. Exercises 35
4. Digression: Categories 37
4.1. Categories 37
4.2. Functors 38
4.3. Equivalences of categories 39

4.4. Module categories 39
4.5. Hom-functors 39
4.6. Exercises 40
5. Examples of infinite dimensional 1-modules 40

5.1. The module N(∞) 40
5.2. Polynomial rings 41
d
5.3. The module (K[T ], dT ) 42
5.4. The module (K[T ], T ·) 43

5.5. The module (K(T ), T ·) 46
5.6. Exercises 46
6. Semisimple modules and their endomorphism rings 48

6.1. Semisimple modules 48
REPRESENTATION THEORY OF ALGEBRAS I: MODULES 3
6.2. Endomorphism rings of semisimple modules 52

6.3. Exercises 54
7. Socle and radical of a module 54

7.1. Socle of a module 54
7.2. Radical of a module 55
7.3. Large and small submodules 57
7.4. Exercises 61
Part 2. Modules of finite length 63
8. Filtrations of modules 63
8.1. Schreier’s Theorem 63
8.2. The Jordan-Hölder Theorem 66
8.3. Exercises 67
9. Digression: Local rings 67

9.1. Local rings 68
9.2. Exercises 71
10. Modules of finite length 72
10.1. Some length formulas for modules of finite length 72

10.2. The Fitting Lemma 75
10.3. The Harada-Sai Lemma 76
References 78
10.4. Exercises 78
11. Direct summands of finite direct sums 79
11.1. The Exchange Theorem 79
11.2. Consequences of the Exchange Theorem 80
References 82
11.3. Examples 82
11.4. Exercises 83
Part 3. Modules II: A-Modules 85

12. Modules over algebras 85
12.1. Representations of an algebra 85

12.2. Modules over an algebra 86
12.3. Modules and representations 86
12.4. A-modules and |A|-modules 87
12.5. Free modules 88

12.6. The opposite algebra 90
12.7. Right A-modules 90
12.8. Examples 91
12.9. Direct decompositions of the regular representation 91

12.10. Modules defined by idempotents 93
12.11. Modules over factor algebras 96
12.12. Modules over products of algebras 96
12.13. Bimodules 97
12.14. Modules over tensor products of algebras 99
12.15. Exercises 99
13. Semisimple algebras 99
13.1. Semisimple algebras and their modules 99

13.2. Examples: Group algebras 101
13.3. Remarks 102
13.4. Exercises 103
14. The Jacobson radical of an algebra 104

14.1. The radical of an algebra 104
14.2. Exercises 107
15. Quivers and path algebras 107
15.1. Quivers and path algebras 107

15.2. Examples 108

15.3. Idempotents in path algebras 109
15.4. Representations of quivers 110

15.5. Examples 111
15.6. Representations of quivers and modules over path algebras 112
15.7. Exercises 113
16. Digression: Classification problems in Linear Algebra 115

16.1. Classification of endomorphisms 115
16.2. Classification of homomorphisms 115
16.3. The Kronecker problem 116
16.4. The n-subspace problem 116

16.5. Exercises 118
Part 4. Projective modules 119

17. Projective modules 119
17.1. Definition and basic properties 119
17.2. The radical of a projective module 121
17.3. Cyclic projective modules 123

17.4. Projective covers 123
18. Injective modules 127
18.1. Definition and basic properties 127
18.2. Injective envelopes 129

19. Digression: The stable module category 132
20. Projective modules over finite-dimensional algebras 134
21. Projective modules over basic algebras 136
22. Direct summands of infinite direct sums 139

22.1. The General Exchange Theorem 139
22.2. The Krull-Remak-Schmidt-Azumaya Theorem 141
22.3. The Crawley-Jønsson-Warfield Theorem 143

22.4. Kaplansky’s Theorem 143
23. Projective modules over semiperfect algebras 146

24. Digression: Projective modules in other areas of mathematics 148
Part 5. Homological Algebra I: Resolutions and extension groups 149

25. Pushout and pullback 149
25.1. Pushout 149
25.2. Pullback 150
25.3. Properties of pushout and pullback 151

25.4. Induced exact sequences 153
25.5. Examples 156
25.6. Schanuel’s Lemma 156
25.7. Short exact sequences with projective middle term 157

25.8. Exercises 158
26. Homological Algebra 159
26.1. The Snake Lemma 159
26.2. Complexes 163

26.3. From complexes to modules 165
26.4. Homology of complexes 165
26.5. Homotopy of morphisms of complexes 167
26.6. The long exact homology sequence 168

27. Projective resolutions and extension groups 172
27.1. Projective resolutions 172
27.2. Ext 175
27.3. Induced maps between extension groups 176

27.4. Some properties of extension groups 176
27.5. Long exact Ext-sequences 177
27.6. Short exact sequences and the first extension group 178
27.7. The vector space structure on the first extension group 181
27.8. Injective resolutions 183

28. Digression: Homological dimensions 183
28.1. Projective, injective and global dimension 183
28.2. Hereditary algebras 184
28.3. Selfinjective algebras 184

28.4. Finitistic dimension 184
28.5. Representation dimension 184
28.6. Dominant dimension 185
28.7. Auslander algebras 185

28.8. Gorenstein algebras 185
29. Tensor products, adjunction formulas and Tor-functors 185
29.1. Tensor products of modules 185
29.2. Adjoint functors 188

29.3. Tor 189
Part 6. Homological Algebra II: Auslander-Reiten Theory 191

30. Irreducible homomorphisms and Auslander-Reiten sequences 191
30.1. Irreducible homomorphisms 191
30.2. Auslander-Reiten sequences and Auslander-Reiten quivers 191
30.3. Properties of irreducible homomorphisms 192

30.4. Dual statements 196
30.5. Examples: Irreducible maps in N f.d. 197
30.6. Exercises 199
31. Auslander-Reiten Theory 199

31.1. The transpose of a module 199
31.2. The Auslander-Reiten formula 200
31.3. The Nakayama functor 202

31.4. Proof of the Auslander-Reiten formula 203
31.5. Existence of Auslander-Reiten sequences 205

31.6. Properties of τ , Tr and ν 208
31.7. Properties of Auslander-Reiten sequences 210
31.8. Digression: The Brauer-Thrall Conjectures 214
31.9. The bimodule of irreducible morphisms 217

31.10. Translation quivers and mesh categories 221
31.11. Examples of Auslander-Reiten quivers 225
31.12. Knitting preprojective components 227
31.13. More examples of Auslander-Reiten quivers 233

32. Grothendieck group and Ringel form 242
32.1. Grothendieck group 242
32.2. The Ringel form 244
33. Reachable and directing modules 245

33.1. Reachable modules 245
33.2. Computations in the mesh category 250
33.3. Directing modules 253
33.4. The quiver of an algebra 256

33.5. Exercises 257
34. Cartan and Coxeter matrix 257
34.1. Coxeter matrix 257
34.2. Cartan matrix 262

34.3. Exercises 266
35. Representation theory of quivers 266
35.1. Bilinear and quadratic forms 267
35.2. Gabriel’s Theorem 271

36. Cartan matrices and (sub)additive functions 271
Part 7. Extras 274

37. Classes of modules 274
38. Classes of algebras 275

39. Dimensions 277
References 278
Index 279
***********************************************************************
1. Introduction
1.1. About this course. This is the first part of notes for a lecture course “In-
troduction to Representation Theory”. As a prerequisite only a good knowledge of
Linear Algebra is required. We will focus on the representation theory of quivers
and finite-dimensional algebras.
The intersection between the content of this course and a classical Algebra course
just consists of some elementary ring theory. We usually work over a fixed field K.
Field extensions and Galois theory do not play a role.
This part contains an introduction to general module theory. We prove the classical
theorems of Jordan-Hölder and Krull-Remak-Schmidt, and we develop the represen-
tation theory of semisimple algebras. (But let us stress that in this course, semisim-
ple representations carry the label “boring and not very interesting”.) We also start
to investigate short exact sequences of modules, pushouts, pullbacks and properties
of Auslander Reiten sequences. Some first results on the representation theory of
path algebras (or equivalently, the representation theory of quivers) are presented
towards the end of this first part. We study the Jacobson radical of an algebra,
decompositions of the regular representation of an algebra, and also describe the
structure of semisimple algebras (which is again regarded as boring). Furthermore,
we develop the theory of projective modules.
As you will notice, this first part of the script concentrates on modules and algebras.
But what we almost do not study yet are modules over algebras. (An exception are
semisimple modules and projective modules. Projective modules will be important
later on when we begin to study homological properties of algebras and modules.)
Here are some topics we will discuss in this series of lecture courses:
• Representation theory of quivers and finite-dimensional algebras

• Homological algebra
• Auslander-Reiten theory
• Knitting of preprojective components
• Tilting theory
• Derived and triangulated categories
• Covering theory
• Categorifications of cluster algebras
• Preprojective algebras
• Ringel-Hall algebras, (dual)(semi) canonical bases of quantized enveloping
algebras
• Quiver representations and root systems of Kac-Moody Lie algebras
• Homological conjectures
• Tame and wild algebras
• Functorial filtrations and applications to the representation theory of clans
and biserial algebras
• Gabriel-Roiter measure
• Degenerations of modules
• Decomposition theory for irreducible components of varieties of modules
1.2. Notation and conventions. Throughout let K be a (commutative) field. Set

K ∗ = K \{0}. Sometimes we will make additional assumptions on K. (For example,
we often assume that K is algebraically closed.)
Typical examples of fields are Q (the field of rational numbers), R (the real numbers),
C (the complex numbers), the finite fields Fp = Z/pZ where p is a prime number.
The field C is algebraically closed.
Let N = {0, 1, 2, 3, . . .} be the natural numbers (including 0).
All vector spaces will be K-vector spaces, and all linear maps are assumed to be
K-linear.
If I is a set, we denote its cardinality by |I|. If I ′ is subset of I we write I ′ ⊆ I. If

additionally I ′ 6= I we also write I ′ ⊂ I.
For a set M let Abb(M, M) be the set of maps M → M. By 1M (or idM ) we
denote the map defined by 1M (m) = m for all m ∈ M. Given maps f : L → M and
g : M → N, we denote the composition by gf : L → N. Sometimes we also write
g ◦ f instead of gf .
1.3. Acknowledgements. We thank Tim Eickmann, Alexander Ivanov, Julian

Kühlshammer, Nicola Pace and Jeff Serbus for typo hunting.
***********************************************************************
Part 1. Modules I: J-Modules
2. Basic terminology
2.1. J-modules. Our aim is to study modules over algebras. Before defining what
this means, we introduce a very straightforward notion of a module which does not
involve an algebra:
Let J be a set (finite or infinite). This set is our “index set”, and in fact only the
cardinality of J is of interest to us. If J is finite, then we often take J = {1, . . . , n}.
We also fix a field K.
A J-module is given by (V, φj )j∈J where V is a vector space and for each j ∈ J we
have a linear map φj : V → V .
Often we just say “module” instead of J-module, and we might say “Let V be a
module.” without explicitly mentioning the attached linear maps φj .
For a natural number m ≥ 0 an m-module is by definition a J-module where
J = {1, . . . , m}.
2.2. Isomorphisms of J-modules. Two J-modules (V, φj )j and (W, ψj )j are iso-
morphic if there exists a vector space isomorphism f : V → W such that
f φj = ψj f
for all j ∈ J.
f
V / W
φj ψj
f
V / W
The dimension of a J-module (V, φj )j is just the dimension of the vector space V .
Matrix version: If V and W are finite-dimensional, choose a basis v1 , . . . , vn of V and
a basis w1 , . . . , wn of W . Assume that the isomorphism f : V → W is represented by
a matrix F (with respect to the chosen bases), and let Φj and Ψj be a corresponding
matrices of φj and ψj , respectively. Then F Φj = Ψj F for all j, i.e. F −1 Ψj F = Φj
for all j.
If two modules V and W are isomorphic we write V ∼
= W.
2.3. Submodules. Let (V, φj )j be a module. A subspace U of V is a submodule

of V if φj (u) ∈ U for all u ∈ U and all j ∈ J. Note that the subspaces 0 and V are
always submodules of V . A submodule U of V is a proper submodule if U ⊂ V ,
i.e. U 6= V .
Example: Let
0 1
φ= .
0 0
Then the 1-module (K 2 , φ) has exactly three submodules, two of them are proper
submodules.
Matrix version: If V is finite-dimensional, choose a basis v1 , . . . , vn of V such that
v1 , . . . , vs is a basis of U. Let φj,U : U → U be the linear map defined by φj,U (u) =
φj (u) for all u ∈ U. Observe that (U, φj,U )j is again a J-module. Then the matrix
Φj of φj (with respect to this basis) is of the form

Aj Bj
Φj = .
0 Cj
In this case Aj is the matrix of φj,U with respect to the basis v1 , . . . , vs .
Let V be a vector space and X a subset of V , then hXi denotes the subspace of
V generated by X. This is the smallest subspace of V containing X. Similarly, for
elements x1 , . . . , xn in V let hx1 , . . . , xn i be the subspace generated by the xi .
P
Let I be a set, and for each i ∈ I let Ui be S a subspace of V . Then the sum i∈I Ui
is defined as the subspace hXi where X = i∈I Ui .
Let V = (V, φj )j be a module, and let X be a subset of V . The intersection U(X)
of all submodules U of V with X ⊆ U is the submodule generated by X. We
call X a generating set of U(X). If U(X) = V , then we say that V is generated
by X.
Lemma 2.1. Let X be a subset of a module V . Define a sequence of subspaces Ui of
V as follows: Let U0 be the subspace of V which is generated by X. If Ui is defined,
let X
Ui+1 = φj (Ui ).
j∈J
Then X
U(X) = Ui .
i≥0
Proof. Set
X
UX = Ui .
i≥0
One can easily check that UX is a submodule of V , and of course UX contains X.

Thus U(X) ⊆ UX . Vice versa, one can show by induction that every submodule
U with X ⊆ U contains all subspaces Ui , thus U also contains UX . Therefore
UX ⊆ U(X).
Let now c be a cardinal number. We say that a module V is c-generated, if V can

be generated by a set X with cardinality at most c. A module which is generated
by a finite set is called finitely generated.
By ℵ0 we denote the smallest infinite cardinal number. We call V countably

generated if V can be generated by a countable set. In other words, V is countably
generated if and only if V is ℵ0 -generated.
If V can be generated by just one element, then V is a cyclic module.
A generating set X of a module V is called a minimal generating set if there
exists no proper subset X ′ of X which generates V . If Y is a finite generating set
of V , then there exists a subset X ⊆ Y , which is a minimal generating set of V .
Warning: Not every module has a minimal generating set. For example, let V be
a vector space with basis {ei | i ∈ N1 }, and let φ : V → V be the endomorphism
defined by φ(ei ) = ei−1 for all i ≥ 2 and φ(e1 ) = 0. Then every generating set of the
module N(∞) = (V, φ) is infinite.
Lemma 2.2. If V is a finitely generated module, then every generating set of V
contains a finite generating set.
Proof. Let X = {x1 , . . . , xn } be a finite generating set of V , and let Y be an arbitrary

generating set of V . As before we have
X
V = U(Y ) = Ui .
i≥0
P
We have xj = i≥0 uij for some uij ∈ Ui and all but finitely many of the uij are
P
zero. Thus there exists some N ≥ 0 such that xj = N i=0 uij for all 1 ≤ j ≤ n. Each
element in Ui is a finite linear combination of elements of the form φji · · · φj1 (y) for
some j1 , . . . , ji ∈ J and y ∈ Y . This yields the result.
Warning: Finite minimal generating sets of a module V do not always have the
same cardinality: Let V = M2 (K) be the vector space of 2 × 2-matrices, and take
the module given by V together with all linear maps A : V → V , A ∈ M2 (K). Then
{( 10 01 )} and {( 10 00 ) , ( 00 10 )} are minimal generating sets of V .
Lemma 2.3. A module V is finitelyPgenerated if and only if for each family Ui ,
i ∈ I of submodules
P of V with V = i∈I Ui there exists a finite subset L ⊆ I such
that V = i∈L Ui .
Proof.
P Let x1 , . . . , xn be a generating set of V , and letPUi be submodules with V =
U . Then each element xl lies in a finite sum i∈I(l) Ui . This implies V =
Pni∈I Pi
l=1 i∈I(l) Ui .
Vice versa, let X be an arbitrary generatingPset of V . For x ∈ X let Ux be the cyclic

submodule generated
P by x. We get V = x∈X Ux . If there exists a finite subset
Y ⊆ X with V = x∈Y Ux , then Y is a generating set of V .
2.4. Factor modules. Let U be a submodule of V . Recall that

V /U = {v = v + U | v ∈ V }
and v + U = v ′ + U if and only if v − v ′ ∈ U. Define φj : V /U → V /U by φj (v + U) =

φj (v) + U. This is well defined since U is a submodule.
Then (V /U, φj )j is a J-module, the factor module corresponding to U.

Matrix version: In the situation of Section 2.3, we have that vs+1 + U, . . . , vn + U is
a basis of V /U and the matrix of φj with respect to this basis is Cj .
2.5. The lattice of submodules. A partially ordered set (or poset) is given
by (S, ≤) where S is a set and ≤ is a relation on S, i.e. ≤ is transitive (s1 ≤ s2 ≤ s3
implies s1 ≤ s3 ), reflexive (s1 ≤ s1 ) and anti-symmetric (s1 ≤ s2 and s2 ≤ s1
implies s1 = s2 ).
One can try to visualize a partially ordered set (S, ≤) using its Hasse diagram:
This is an oriented graph with vertices the elements of S, and one draws an arrow
s → t if s < t and if s ≤ m ≤ t implies s = m or m = t. Ususally one tries to draw
the diagram with arrows pointing upwards and then one forgets the orientation of
the arrows and just uses unoriented edges.
For example, the following Hasse diagram describes the partially ordered set (S, ≤)
with three elements s1 , s2 , t with si < t for i = 1, 2, and s1 and s2 are not comparable
in (S, ≤).
t
@
@
s1 s2
For a subset T ⊆ S an upper bound for T is some s ∈ S such that t ≤ s for all
t ∈ T . A supremum s0 of T is a smallest upper bound, i.e. s0 is an upper bound
and if s is an upper bound then s0 ≤ s.
Similarly, define a lower bound and an infimum of T .

The poset (S, ≤) is a lattice if for any two elements s, t ∈ S there is a supremum
and an infimum of T = {s, t}. In this case write s + t (or s ∪ t) for the supremum
and s ∩ t for the infimum.
One calls (S, ≤) a complete lattice if there is a supremum and infimum for every
subset of S.
Example: The natural numbers N together with the usual ordering form a lattice,
but this lattice is not complete. For example, the subset N itself does not have a
supremum in N.
A lattice (S, ≤) is called modular if

s1 + (t ∩ s2 ) = (s1 + t) ∩ s2
for all elements s1 , s2 , t ∈ S with s1 ≤ s2 .
This is not a lattice:
r
@
@@r
r
A
A
A r
A @
Ar @@r
This is a complete lattice, but it is not modular:

r
@
@@r
r
A
A
A r
A
Ar
The following lemma is straightforward:

Lemma 2.4. Sums and intersections of submodules are again submodules.
Lemma 2.5. Let (V, φj )j be a module. Then the set of all submodules of V is a
complete lattice where U1 ≤ U2 if U1 ⊆ U2 .
Proof.
P Straightforward: TheTsupremum of a set {Ui | i ∈ I} of submodules is
i∈I Ui , and the infimum is i∈I Ui .
Lemma 2.6 (Dedekind). Let U1 , U2 , W be submodules of a module V such that
U1 ⊆ U2 . Then we have
U1 + (W ∩ U2 ) = (U1 + W ) ∩ U2 .
Proof. It is sufficient to proof the statement for subspaces of vector spaces. The
inclusion ⊆ is obvious. For the other inclusion let u ∈ U1 , w ∈ W and assume
u + w ∈ U2 . Then w = (u + w) − u belongs to U2 and thus also to W ∩ U2 . Thus
u + w ∈ U1 + (W ∩ U2 ).
Thus the lattice of submodules of a module is modular.
2.6. Examples. (a): Let K be a field, and let V = (K 2 , φ, ψ) be a 2-module where

c1 0 0 1
φ= and ψ =
0 c2 1 0
and c1 6= c2 . By e1 and e2 we denote the canonical basis vectors of K 2 . The module
V is simple, i.e. V does not have any non-zero proper submodule. The 1-module
(K 2 , φ) has exactly two non-zero proper submodules. Let

0 1
θ= .
0 0
Then (K 2 , φ, θ) has exactly one non-zero proper submodule, namely U = he1 i. We

have U ∼= (K, c1 , 0), and V /U ∼ = (K, c2 , 0). In particular, U and V /U are not
isomorphic.
(b): Let
   
c1 0 0 0 0 0
φ =  0 c2 0  and ψ = 0 0 0
0 0 c3 1 1 0
with pairwise different ci . Then the lattice of submodules of (K 3 , φ, ψ) looks like
this:
r
@
@@r
r
@
@
@r
The non-zero proper submodules are he3 i, he1 , e3 i and he2 , e3 i.

(c): Let
   
c1 0 0 1
 0 c2   
φ=  and ψ = 1 0 
 c1 0   0 1
0 c2 1 0
with c1 6= c2 . If K = F3 , then the lattice of submodules of (K 4 , φ, ψ) looks like this:
r
@
AA

r r A@
r@ r
@A
@
@ AA
r
The non-zero proper submodules are he1 , e2 i, he3 , e4 i, he1 + e3 , e2 + e4 i and he1 +
2e3 , e2 + 2e4 i.
(d):
Let
   
c1 0 0 1
 0 c2   
φ=  and ψ = 1 0 
 c3 0   0 1
0 c4 1 0
with pairwise different ci . Then the lattice of submodules of (K 4 , φ, ψ) looks like
this:
r
@
@@r
r
@
@
@r
The non-zero proper submodules are he1 , e2 i and he3 , e4 i.
2.7. Decompositons and direct sums of modules. Let (V, φj )j be a module,

and let U1 and U2 be submodules of V . If U1 ∩ U2 = 0 and U1 + U2 = V , then this
is called a direct decomposition of V , and we say (V, φj )j is the direct sum of
the submodules U1 and U2 . In this case we write V = U1 ⊕ U2 .
A submodule U of V is a direct summand of V if there exists a submodule U ′
such that U ⊕ U ′ = V . In this case we say that U ′ is a direct complement of U
in V .
Matrix version: Assume that V is finite-dimensional. Choose a basis v1 , . . . , vs of U1
and a basis vs+1 , . . . , vn of U2 . Then the matrix Φj of φj with respect to the basis
v1 , . . . , vn of V is of the form
Aj 0
Φj =
0 Bj
where Aj and Bj are the matrices of φj,U1 and φj,U2 , respectively.
Vice versa, let (V, φj )j and (W, ψj )j be modules. Define

(V, φj )j ⊕ (W, ψj )j = (V ⊕ W, φj ⊕ ψj )j
where
V ⊕ W = V × W = {(v, w) | v ∈ V, w ∈ W }
and (φj ⊕ ψj )(v, w) = (φj (v), ψj (w)).
In this case V ⊕ W is the direct sum of the submodules V ⊕ 0 and 0 ⊕ W .

On the other hand, if (V, φj )j is the direct sum of two submodules U1 and U2 , then
we get an isomorphism
U1 ⊕ U2 → V
defined by (u1 , u2 ) 7→ u1 + u2 .
A module (V, φj )j is indecomposable if the following hold:
• V 6= 0,
• Let U1 and U2 be submodules of V with U1 ∩ U2 = 0 and U1 + U2 = V , then
U1 = 0 or U2 = 0.
If (V, φj )j is not indecomposable, then it is called decomposable.

More generally, we can construct direct sums of more than two modules, and we
can look at direct decompositions of a module into a direct sum of more than two
(i)
modules. This is defined in the obvious way. For modules (Vi , φj )j , 1 ≤ i ≤ t we
write
t
M
(1) (t) (i)
(V1 , φj )j ⊕ · · · ⊕ (Vt , φj )j = (Vi , φj )j .
i=1
2.8. Products of modules. Let I be a set, and for each i ∈ I let Vi be a vector
space. The product of the vector spaces Vi is by definition the set of all sequences
(vi )i∈I with vi ∈ Vi . We denote the product by
Y
Vi .
i∈I
With componentwise addition and scalar multiplication, this is again a vector space.
The Vi are called the factors of the product. For linear maps fi : Vi → Wi with i ∈ I
we define their product
Y Y Y
fi : Vi → Wi
i∈I i∈I i∈I
Q L Q
by i∈I fi ((vQ
i )i ) = (fi (v
Li ))i . Obviously, i∈I Vi is a subspace of i∈I Vi . If I is a
finite set, then i∈I Vi = i∈I Vi .
(i)
Now for each i ∈ I let Vi = (Vi , φj )j be a J-module. Then the product of the
modules Vi is defined as
!
Y Y (i)
Y Y (i)
(V, φj )j = Vi = (Vi , φj )j = Vi , φj .
i∈I i∈I i∈I i∈I j
Thus V is the product of the vector spaces Vi , and φj is the product of the linear
(i)
maps φj .
2.9. Examples: Nilpotent endomorphisms. Sometimes one does not study all
J-modules, but one assumes that the linear maps associated to the elements in J
satisfy certain relations. For example, if J just contains one element, we could
study all J-modules (V, f ) such that f n = 0 for some fixed n. Or, if J contains two
elements, then we can study all modules (V, f, g) such that f g = gf .
Assume |J| = 1. Thus a J-module is just (V, φ) with V a vector space and φ : V → V
a linear map. We additionally assume that φ is nilpotent, i.e. φm = 0 for some m
and that V is finite-dimensional. We denote this class of modules by N f.d..
We know from LA that there exists a basis v1 , . . . , vn of V such that the correspond-
ing matrix Φ of φ is of the form
 J(λ ) 
1
J(λ2 )
Φ= .. 
.
J(λt )
where J(λi ), 1 ≤ i ≤ t is a λi × λi -matrix of the form

0 1 
0 1
J(λi ) =  .. .. 
. .
0 1
0
for some partition λ = (λ1 , . . . , λt ) of n.
A partition of some n ∈ N is a sequence λ = (λ1 , . . . , λt ) of integers with λ1 ≥
λ2 ≥ · · · ≥ λt ≥ 1 and λ1 + · · · + λt = n.
Example: The partitions of 4 are (4), (3, 1), (2, 2), (2, 1, 1) and (1, 1, 1, 1).
One can visualize partitons with the help of Young diagrams: For example the
Young diagram of the partition (4, 2, 2, 1, 1) is the following:
Let e1 , . . . , em be the standard basis of K m where

   
1 0
0  ... 
e1 =  
 ...  , . . . , em =  .
 0
0 1
To each partition λ = (λ1 , . . . , λt ) of n we associate a module

t
M
N(λ) = N(λi ) = (K n , φλ)
i=1
where for m ∈ N we have
N(m) = (K m , φm )
with φm the endomorphism defined by φm (ej ) = ej−1 for 2 ≤ j ≤ m and φm (e1 ) = 0.
In other words, the matrix of φm with respect to the basis e1 , . . . , em is J(m).
We can visualize N(λ) with the help of Young diagrams. For example, for λ =
(4, 2, 2, 1, 1) we get the following diagram:
e14
e13
e12 e22 e32
e11 e21 e31 e41 e51
Here the vectors

{eij | 1 ≤ i ≤ 5, 1 ≤ j ≤ λi }
denote a basis of
K 10 = K 4 ⊕ K 2 ⊕ K 2 ⊕ K 1 ⊕ K 1 .
Let φλ : K 10 → K 10 be the linear map defined by φλ (eij ) = eij−1 for 2 ≤ j ≤ λi and
φλ (ei1 ) = 0. Thus N(λ) = (K 10 , φλ ).
So φλ operates on the basis vectors displayed in the boxes of the Young diagram
by mapping them to the vector in the box below if there is a box below, and by
mapping them to 0 if there is no box below.
The matrix of φλ with respect to the basis
e11 , e12 , e13 , e14 , e21 , e22 , e31 , e32 , e41 , e51
is 0 1 
 J(4)  0 1
 0 1 
J(2) 0
 J(2) =
 0 1
0

.
J(1)  0 1 
J(1) 0
0
0
Similarly, for an arbitrary partition λ = (λ1 , . . . , λt ) of n we will work with a basis

{eij | 1 ≤ i ≤ t, 1 ≤ j ≤ λi } of K n , and we define a linear map φλ : K n → K n by
φλ (eij ) = eij−1 for 2 ≤ j ≤ λi and φλ (ei1 ) = 0. For simplicity, define ei0 = 0 for all
i.
Theorem 2.7. For every module (V, φ) with V an n-dimensional vector space and
φ a nilpotent linear map V → V there exists a unique partition λ of n such that
(V, φ) ∼
= N(λ).
Proof. Linear Algebra (Jordan Normal Form).
Now let λ = (λ1 , . . . , λt ) be a again a partition of n, and let x ∈ N(λ) = (K n , φ).

Thus
X
x= cij eij
i,j
for some cij ∈ K. We want to compute the submodule U(x) ⊆ N(λ) generated by
x:
We get
!
X X X
φ(x) = φ cij eij = cij φ(eij ) = cij eij−1 .
i,j i,j i,j:j≥2
Similarly, we can easily write down φ2 (x), φ3 (x), etc. Now let r be maximal such
that cir 6= 0 for some i. This implies φr−1 (x) 6= 0 but φr (x) = 0. It follows that the
vectors x, φ(x), . . . , φr−1 (x) generate U(x) as a vector space, and we see that U(x)
is isomorphic to N(r).
For example, the submodule U(eij ) of N(λ) is isomorphic to N(j) and the corre-
sponding factor module N(λ)/U(eij ) is isomorphic to
M
N(λi − j) ⊕ N(λa ).
a6=i
Let us look at a bit closer at the example λ = (3, 1):
e13
e12
e11 e21
We get
U(e21 ) ∼
= N(1), N(3, 1)/U(e21 ) ∼
= N(3),
U(e11 ) ∼
= N(1), N(3, 1)/U(e11 ) ∼
= N(2, 1),
U(e12 ) ∼
= N(2), N(3, 1)/U(e12 ) ∼
= N(1, 1),
U(e13 ) ∼
= N(3), N(3, 1)/U(e13 ) ∼
= N(1),
U(e12 + e21 ) ∼
= N(2), N(3, 1)/U(e12 + e21 ) ∼
= N(2).
Let us check the last of these isomorphisms: Let x = e12 + e21 ∈ N(3, 1) = (K 4 , φ).
We get φ(x) = e11 and φ2 (x) = 0. It follows that U(x) is isomorphic to N(2).
Now as a vector space, N(3, 1)/U(x) is generated by the residue classes e13 and
e12 . We have φ(e13 ) = e12 and φ(e12 ) = e11 . In particular, φ(e12 ) ∈ U(x). Thus
N(3, 1)/U(x) ∼= N(2).
2.10. Exercises. 1: Let W and Ui , i ∈ I be a set of submodules of a module

(V, φj )j such that for all k, l ∈ I we have Uk ⊆ Ul or Uk ⊇ Ul . Show that
X [
Ui = Ui
i∈I i∈I
and !
[ [
(W ∩ Ui ) = W ∩ Ui .
i∈I i∈I
2: Let K be a field and let V = (K 4 , φ, ψ) be a module such that

 
λ1
 λ2 
φ=


λ3
λ4
with pairwise different λi ∈ K. How can the lattice of submodules of V look like?
3: Which of the following lattices can be the lattice of submodules of a 4-dimensional
module of the form (V, φ, ψ)? In each case you can work with a field K of your choice.
Of course it is better if you find examples which are independent of the field, if this
is possible.
(see the pictures distributed during the lecture)

4: Classify all submodules U of V = N(2, 1), N(3, 1), N(2, 2) and determine in each
case the isomorphism class of U and of the factor module V /U.
For K = F2 and K = F3 draw the corresponding Hasse diagrams.
Let K = Fp with p a prime number, and let λ and µ be partitions. How many
= N(λ) and V /U ∼
submodules U of V with U ∼ = N(µ) are there?
5: Let U be a maximal submodule of a module V , and let W be an arbitrary
submodule of V . Show that either W ⊆ U or U + W = V .
6: Find two 2 × 2-matrices A and B with coefficients in K such that (K 2 , A, B) has

exactly 4 submodules.
7: Show: If V is a 2-dimensional module with at least 5 submodules, then every
subspace of V is a submodule.
8: Let V be a 2-dimensional module with at most 4 submodules. Show that V is

cyclic.
————————————————————————————-
3. Homomorphisms between modules
3.1. Homomorphisms. Let (V, φj )j and (W, ψj )j be two modules. A linear map
f : V → W is a homomorphism (or module homomorphism) if
f φj = ψj f
for all j ∈ J.
f
V / W
φj ψj
f
V / W
We write f : (V, φj )j → (W, ψj )j or just f : V → W . An injective homomorphism

is also called a monomorphism, and a surjective homomorphism is an epimor-
phism. A homomorphism which is injective and surjective is an isomorphism,
compare Section 2.2.
If f : (V, φj )j → (W, ψj )j is an isomorphism, then the inverse f −1 : W → V is also a
homomorphism, thus also an isomorphism: We have
f −1 ψj = f −1 ψj f f −1 = f −1 f φj f −1 = φj f −1 .
For modules (U, µj )j , (V, φj )j , (W, ψj )j and homomorphisms f : U → V and g : V →

W the composition gf : U → W is again a homomorphism.
Here is a trivial example of a homomorphism: Let (V, φj )j be a module, and let U be

a submodule of V . Then the map ι : U → V defined by ι(u) = u is a homomorphism,
which is called the (canonical) inclusion.
Similarly, the map π : V → V /U defined by π(v) = v + U is a homomorphism, which
is called the (canonical) projection.
If f : (V, φj )j → (W, ψj )j is a homomorphism, then define
Ker(f ) = {v ∈ V | f (v) = 0},
the kernel of f , and
Im(f ) = {f (v) | v ∈ V },
the image of f . Furthermore, Cok(f ) = W/ Im(f ) is the cokernel of f .
One can easily check that Ker(f ) is a submodule of V : For v ∈ Ker(f ) and j ∈ J
we have f φj (v) = ψj f (v) = ψj (0) = 0.
Similarly, Im(f ) is a submodule of W : For v ∈ V and j ∈ J we have ψj f (v) =
f φj (v), thus ψj f (v) is in Im(f ).
For a homomorphism f : V → W let f1 : V → Im(f ) defined by f1 (v) = f (v) (the
only difference between f and f1 is that we changed the target module of f from
W to Im(f )), and let f2 : Im(f ) → W be the canonical inclusion. Then f1 is an
epimorphism and f2 a monomorphism, and we get f = f2 f1 . In other words, every
homomorphism is the composition of an epimorphism followed by a monomorphism.
Let V and W be J-modules. For homomorphisms f, g : V → W define

f + g: V → W
by (f + g)(v) = f (v) + g(v). This is again a homomorphism. Similarly, for c ∈ K
we can define
cf : V → W
by (cf )(v) = cf (v), which is also a homomorphism. Thus the set of homomorphisms
V → W forms a subspace of the vector space HomK (V, W ) of linear maps from
V to W . This subspace is denoted by HomJ (V, W ) and sometimes we just write
Hom(V, W ).
A homomorphism V → V is also called an endomorphism. The set HomJ (V, V )
of endomorphisms is denoted by EndJ (V ) or just End(V ). This is a K-algebra with
multiplication given by the composition of endomorphims. One often calls End(V )
the endomorphism algebra (or the endomorphism ring) of V .
3.2. Definition of a ring. A ring is a set R together with two maps + : R×R → R,
(a, b) 7→ a + b (the addition) and · : R × R → R, (a, b) 7→ ab (the multiplication)
such that the following hold:
• Associativity of addition: (a + b) + c = a + (b + c) for all a, b, c ∈ R,

• Commutativity of addition: a + b = b + a for all a, b ∈ R,
• Existence of a 0-element: There exists exactly one element 0 ∈ R with a +
0 = a for all a ∈ R,
• Existence of an additive inverse: For each a ∈ R there exists exactly one
element −a ∈ R such that a + (−a) = 0,
• Associativity of multiplication: (ab)c = a(bc) for all a, b, c ∈ R,
• Existence of a 1-element: There exists exactly one element 1 ∈ R with 1a =
a1 = a for all a ∈ R,
• Distributivity: (a + b)c = ac + bc and a(b + c) = ab + ac for all a, b, c ∈ R.
A ring R is commutative if ab = ba for all a, b ∈ R.
3.3. Definition of an algebra. A K-algebra is a K-vector space A together with

a map · : A×A → A, (a, b) 7→ ab (the multiplication) such that the following hold:
• Associativity of multiplication: (ab)c = a(bc) for all a, b, c ∈ A;

• Existence of a 1-element: There exists an element 1 which satisfies 1a =
a1 = a for all a ∈ A;
• Distributivity: a(b + c) = ab + ac and (a + b)c = ac + ac for all a, b, c ∈ A;
• Compatibility of multiplication and scalar multiplication: λ(ab) = (λa)b =
a(λb) for all λ ∈ K and a, b ∈ A.
The element 1 is uniquely determined and we often also denoted it by 1A .

In other words, a K-algebra is a ring A, which is also a K-vector space such that
additionally λ(ab) = (λa)b = a(λb) for all λ ∈ K and a, b ∈ A.
In contrast to the definition of a field, the definitions of a ring and an algebra do not
require that the element 0 is different from the element 1. Thus there is a ring and
an algebra which contains just one element, namely 0 = 1. If 0 = 1, then R = {0}.
3.4. Homomorphism Theorems.

Theorem 3.1 (Homomorphism Theorem). If V and W are J-modules, and if
f : V → W is a homomorphism, then f induces an isomorphism
f : V / Ker(f ) → Im(f )
defined by f (v + Ker(f )) = f (v).
Proof. One easily shows that f is well defined, and that it is a homomorphism.
Obviously f is injective and surjective, and thus an isomorphism.
Remark: The above result is very easy to prove. Nevertheless we call it a Theorem,
because of its importance.
We derive some consequences from Theorem 3.1:

Corollary 3.2 (First Isomorphism Theorem). If U1 ⊆ U2 are submodules of a
module V , then
V /U2 ∼
= (V /U1 )/(U2 /U1 ).
Proof. Note that U2 /U1 is a submodule of V /U1 . Thus we can build the factor
module (V /U1 )/(U2 /U1 ). Let
V → V /U1 → (V /U1 )/(U2 /U1 )
be the composition of the two canonical projections. This homomorphism is obvi-
ously surjective and its kernel is U2 . Now we use Theorem 3.1.
Corollary 3.3 (Second Isomorphism Theorem). If U1 and U2 are submodules of a
module V , then
U1 /(U1 ∩ U2 ) ∼
= (U1 + U2 )/U2 .
Proof. Let
U1 → U1 + U2 → (U1 + U2 )/U2
be the composition of the inclusion U1 → U1 + U2 and the projection U1 + U2 →
(U1 + U2 )/U2 . This homomorphism is surjective (If u1 ∈ U1 and u2 ∈ U2 , then
u1 +u2 +U2 = u1 +U2 is the image of u1 .) and its kernel is U1 ∩U2 (An element u1 ∈ U1
is mapped to 0 if and only if u1 + U2 = U2 , thus if and only if u1 ∈ U1 ∩ U2 .).
U1 + UH2
vv HH
vv HH
v HH
vv HH
vv
U1 HH U2
HH vv
HH vvv
HH vv
H vv
U1 ∩ U2
In particular, if U1 ⊂ U2 and W are submodules of a module V , then the above

results yield the isomorphisms
(U2 ∩ W )/(U1 ∩ W ) ∼
= (U1 + U2 ∩ W )/U1 ,
U2 /(U1 + U2 ∩ W ) ∼
= (U2 + W )/(U1 + W ).
The module (U1 + U2 ∩ W )/U1 is a submodule of U2 /U1 , and U2 /(U1 + W ∩ U2 ) is
the corresponding factor module.
U2
U1 + U2 ∩OW U2 + WL
rr OOO o LLL
rrr OOO ooooo LLL
rrr OOO oo LLL
rrr
O ooo L
U1 LLL U2 ∩ W U1 + WO r
U2
LLL ooo OOO
OOO rrr
LLL
ooooo OOO rrrr
LL ooo O rrr
U1 ∩ W U1 + U2 ∩ W
3.5. Homomorphisms between direct sums. Let

n
M
V = Vj
j=1
be a direct sum of modules. By

ιV,j : Vj → V
we denote the canonical inclusion and by
πV,j : V → Vj
P
the canonical projection. (Each v ∈ V is of the form v = nj=1 vj where the vj ∈ Vj
are uniquely determined. Then πV,j (v) = vj .) These maps are all homomorphisms.
They satisfy
πV,j ◦ ιV,j = 1Vj ,
πV,i ◦ ιV,j = 0 if i 6= j,
n
X
ιV,j ◦ πV,j = 1V .
j=1
Now let V and W be modules, which are a finite direct sum of certain submodules,
say
M n Mm
V = Vj and W = Wi .
j=1 i=1
If f : V → W is a homomorphism, define
fij = πW,i ◦ f ◦ ιV,j : Vj → Wi
We can write f : V → W in matrix form
 
f11 · · · f1n
f=  .. .. 
. .
fm1 · · · fmn
and we can use the usual matrix calculus: Let us write elements v ∈ V and w ∈ W
as columns
   
v1 w1
v= . .
.  and w = ... 

vn wm
with vj ∈ Vj and wi ∈ Wi . If f (v) = w we claim that
     Pn   
f11 · · · f1n v1 j=1 f1j (vj ) w1
.
 .. . . 
..   ..  =  .  .
..  =  ..  .
Pn
fm1 · · · fmn vn j=1 fmj (vj ) wm
Namely, if v ∈ V we get for 1 ≤ i ≤ m
n
X n
X
fij (vj ) = (πW,i ◦ f ◦ ιV,j ) (vj )
j=1 j=1
n
!!
X
= πW,i ◦ f ◦ ιV,j ◦ πV,j (v)
j=1
= (πW,i ◦ f )(v) = wi .
The first term is the matrix product of the ith row of the matrix of f with the
column vector v, the last term is the ith entry in the column vector w.
Vice versa, if fij : Vj → Wi with 1 ≤ j ≤ n and 1 ≤ i ≤ m are homomorphisms,

then we obtain with X
ιW,i ◦ fij ◦ πV,j
i,j
a homomorphism f : V → W , and of course we can write f as a matrix

 
f11 · · · f1n
f =  ... ..  .
.
fm1 · · · fmn
The composition of such morphisms given by matrices can be realized via matrix
multiplication.
If A is a matrix, we denote its transpose by t A. In particular, we can write the

column vector v we looked at above as v = t [v1 , . . . , vn ].
Now f 7→ (fij )ij defines an isomorphism of vector spaces
n m
! n M m
M M M
HomJ Vj , Wi → HomJ (Vj , Wi ).
j=1 i=1 j=1 i=1
In particular, for every module X we obtain isomorphisms of vector spaces

m
! m
M M
HomJ X, Wi → HomJ (X, Wi )
i=1 i=1
and !
n
M n
M
HomJ Vj , X → HomJ (Vj , X).
j=1 j=1
3.6. Idempotents and direct decompositions. An element r in a ring R is an

idempotent if r 2 = r. We will see that idempotents in endomorphism rings of
modules play an important role.
Let V = U1 ⊕ U2 be a direct decomposition of a module V . Thus U1 and U2 are
submodules of V such that U1 ∩ U2 = 0 and U1 + U2 = V . Let ιi : Ui → V and
πi : V → Ui be the corresponding inclusions and projections. We can write these
homomorphisms in matrix form
ι1 = t [1 0], ι2 = t [0 1], π1 = [1 0], π2 = [0 1].
Define e1 = ι1 π1 and e2 = ι2 π2 . Then both e1 and e2 are idempotents in the
endomorphism ring End(V ) of V . (For example, e21 = ι1 π1 ι1 π1 = ι1 1U1 π1 = e1 .) Set
e(U1 , U2 ) = e1 .
Proposition 3.4. Let V be a J-module. If we associate to an idempotent e ∈
End(V ) the pair (Im(e), Ker(e)), then we obtain a bijection between the set of all
idempotents in EndJ (V ) and the set of pairs (U1 , U2 ) of submodules of V such that
V = U1 ⊕ U2 .
Proof. Above we associated to a direct decompositon V = U1 ⊕ U2 the idempotent

e1 = ι1 π1 ∈ End(V ). This idempotent is uniquely determined by the following
two properties: For all u1 ∈ U1 we have e1 (u1 ) = u1 , and for all u2 ∈ U2 we have
e1 (u2 ) = 0. From e1 we can easily obtain the above direct decomposition: We have
U1 = Im(e1 ) and U2 = Ker(e1 ).
Vice versa, let e ∈ End(V ) be an idempotent. Define U1 = Im(e) and U2 = Ker(e).
Of course U1 and U2 are submodules of V . We also get U1 ∩ U2 = 0: If x ∈ U1 ∩ U2 ,
then x ∈ U1 = Im(f ), thus x = e(y) for some y, and x ∈ U2 = Ker(e), thus e(x) = 0.
Since e2 = e we obtain x = e(y) = e2 (y) = e(x) = 0.
Finally, we show that U1 + U2 = V : If v ∈ V , then v = e(v) + (v − e(v)) and
e(v) ∈ Im(e) = U1 . Furthermore, e(v − e(v)) = e(v) − e2 (v) = 0 shows that
v − e(v) ∈ Ker(e) = U2 .
Thus our idempotent e yields a direct decomposition V = U1 ⊕ U2 . Since e(u1 ) = u1

for all u1 ∈ U1 and e(u2 ) = 0 for all u2 ∈ U2 , we see that e is the idempotent
corresponding to the direct decomposition V = U1 ⊕ U2 .
The endomorphism ring End(V ) of a module V contains of course always the idem-
potents 0 and 1. Here 0 corresponds to the direct decomposition V = 0 ⊕ V , and 1
corresponds to V = V ⊕ 0.
If e is an idempotent in a ring, then 1 − e is also an idempotent. (Namely (1 − e)2 =
1 − e − e + e2 = 1 − e.)
If the idempotent e ∈ End(V ) corresponds to the pair (U1 , U2 ) with V = U1 ⊕ U2 ,

then 1 − e corresponds to (U2 , U1 ). (One easily checks that Im(1 − e) = Ker(e) and
Ker(1 − e) = Im(e).)
Corollary 3.5. For a module V the following are equivalent:
• V is indecomposable;
• V =
6 0, and 0 and 1 are the only idempotents in End(V ).
Later we will study in more detail the relationship between idempotents in endo-
morphism rings and direct decompositions.
3.7. Split monomorphisms and split epimorphisms. Let V and W be mod-

ules. An injective homomorphism f : V → W is called split monomorphism if
Im(f ) is a direct summand of W . A surjective homomorphism f : V → W is a split
epimorphism if Ker(f ) is a direct summand of V .
Lemma 3.6. Let f : V → W be a homomorphism. Then the following hold:
(i) f is a split monomorphism if and only if there exists a homomorphism

g : W → V such that gf = 1V ;
(ii) f is a split epimorphism if and only if there exists a homomorphism h : W →
V such that f h = 1W .
Proof. Assume first that f is a split monomorphism. Thus W = Im(f ) ⊕ C for some
submodule C of W . Let ι : Im(f ) → W be the inclusion homomorphism, and let
π : W → Im(f ) be the projection with kernel C. Let f0 : V → Im(f ) be defined by
f0 (v) = f (v) for all v ∈ V . Thus f = ιf0 . Of course, f0 is an isomorphism. Define
g = f0−1 π : W → V . Then we get
gf = (f0−1 π)(ιf0 ) = f0−1 (πι)f0 = f0−1 f0 = 1V .
Vice versa, assume there is a homomorphism g : W → V such that gf = 1V . Set

e = f g. This is an endomorphism of W , and we have
e2 = (f g)(f g) = f (gf )g = f 1V g = e,
thus e is an idempotent. In particular, the image of e is a direct summand of W .
But it is easy to see that Im(e) = Im(f ): Since e = f g we have Im(e) ⊆ Im(f ), and
f = f 1V = f gf = ef yields the other inclusion Im(f ) ⊆ Im(e). Thus Im(f ) is a
direct summand of W .
This proves part (i) of the statement. We leave part(ii) as an exercise.
3.8. Short exact sequences and Hom-functors. Let V, W, X, Y be modules,

and let f : V → W and h : X → Y be homomorphisms. For g ∈ HomJ (W, X) we
define a map
HomJ (f, h) : HomJ (W, X) → HomJ (V, Y ), g 7→ hgf.
It is easy to check that HomJ (f, h) is a linear map of vector spaces: For g, g1, g2 ∈
HomJ (W, X) and c ∈ K we have
h(g1 + g2 )f = hg1 f + hg2 f and h(cg)f = c(hgf ).
If V = W and f = 1V , then instead of HomJ (1V , h) we mostly write
HomJ (V, h) : HomJ (V, X) → HomJ (V, Y ),
thus by definition HomJ (V, h)(g) = hg for g ∈ HomJ (V, X). If X = Y and h = 1X ,
then instead of HomJ (f, 1X ) we write
HomJ (f, X) : HomJ (W, X) → HomJ (V, X),
thus HomJ (f, X)(g) = gf for g ∈ HomJ (W, X).
Let U, V, W be modules, and let f : U → V and g : V → W be homomorphisms. If
Im(f ) = Ker(g), then (f, g) is called an exact sequence. Mostly we denote such
an exact sequence in the form
f g
U−→V − → W.
We also say, the sequence is exact at V . Given such a sequence with U = 0,
exactness implies that g is injective. (For U = 0 we have Im(f ) = 0 = Ker(g), thus
g is injective.) Similarly, if W = 0, exactness yields that f is surjective. (For W = 0
we have Ker(g) = V , but Im(f ) = V means that f is surjective.)
Given modules Vi with 0 ≤ i ≤ t and homomorphisms fi : Vi−1 → Vi with 1 ≤ i ≤ t,
then the sequence (f1 , . . . , ft ) is an exact sequence if
Im(fi−1 ) = Ker(fi )
for all 2 ≤ i ≤ t. Also here we often write
f1 f2 ft
V0 −
→ V1 −
→ ··· −
→ Vt .
Typical examples of exact sequences can be obtained as follows: Let V and W

be modules and let g : V → W be a homomorphism. Let ι : Ker(g) → V be the
inclusion, and let π : W → Cok(g) be the projection. Then the sequence
ι g π
0 → Ker(g) −
→V −
→W −
→ Cok(g) → 0
is exact. (Recall that Cok(g) = W/ Im(g).)
Vice versa, if we have an exact sequence of the form
f g
0→U −
→V −
→W
then f is injective and Im(f ) = Ker(g). Similarly, if
g h
U−
→V −
→W →0
is an exact sequence, then h is surjective and Im(g) = Ker(h).
f g
Lemma 3.7. Let 0 → U − → V −→ W be an exact sequence of J-modules. Then
gf = 0, and for every homomorphism b : X → V with gb = 0 there exists a unique
homomorphism b′ : X → U with b = f b′ .
X
b′
~
~ b
~
~~ f g
0 / U / V / W
Proof. Of course we have gf = 0. Let now b : X → V be a homomorphism with

gb = 0. This implies that Im(b) ⊆ Ker(g). Set U ′ = Ker(g), and let ι : U ′ → V
be the inclusion. Thus b = ιb0 for some homomorphism b0 : X → U ′ . There is an
isomorphism f0 : U → U ′ with f = ιf0 . If we define b′ = f0−1 b0 , then we obtain
f b′ = (ιf0 )(f0−1b0 ) = ιb0 = b.
We still have to show the uniqueness of b′ : Let b′′ : X → U be a homomorphism
with f b′′ = b. Then the injectivity of f implies b′ = b′′ .
There is the following reformulation of Lemma 3.7:

f g
Lemma 3.8. Let 0 → U − →V − → W be an exact sequence of J-modules. Then for
every J-module X, the sequence
HomJ (X,f ) HomJ (X,g)
0 → HomJ (X, U) −−−−−−→ HomJ (X, V ) −−−−−−→ HomJ (X, W )
is exact. (“HomJ (X, −) is a left exact functor.”)
Proof. We have HomJ (X, g) ◦ HomJ (X, f ) = 0: For any homomorphism a : X → U

we get
(HomJ (X, g) ◦ HomJ (X, f ))(a) = gf a = 0.
This implies Im(HomJ (X, f )) ⊆ Ker(HomJ (X, g)).
Vice versa, let b ∈ Ker(HomJ (X, g)). Thus b : X → V is a homomorphism with gb =

0. We know that there exists some b′ : X → U with f b′ = b. Thus HomJ (X, f )(b′ ) =
f b′ = b. This shows that b ∈ Im(HomJ (X, f )). The uniqueness of b′ means that
HomJ (X, f ) is injective.
Here are the corresponding dual statements of the above lemmas:

f g
Lemma 3.9. Let U − → V −→ W → 0 be an exact sequence of J-modules. Then
gf = 0, and for every homomorphism c : V → Y with cf = 0 there exists a unique
homomorphism c′ : W → Y with c = c′ g.
f g
U / V /W / 0
}
c′ }
c }
~}
Y
Proof. Exercise.
And here is the corresponding reformulation of Lemma 3.8:

f g
Lemma 3.10. Let U − →V − → W → 0 be an exact sequence of J-modules. Then for
every J-module X, the sequence
HomJ (g,Y ) HomJ (f,Y )
0 → HomJ (W, Y ) −−−−−−→ HomJ (V, Y ) −−−−−−→ HomJ (U, Y )
is exact. (“HomJ (−, Y ) is a left exact contravariant functor.”)
Proof. Exercise.
An exact sequence of the form

f g
0→U −
→V −
→W →0
is called a short exact sequence. This sequence starts in U and ends in W .
Its middle term is V and its end terms are U and W . For such a short exact
sequence we often write (f, g) instead of (0, f, g, 0).
Two short exact sequences
f g
0 → U −→ V −→ W → 0
and
f′ g′
0 → U −→ V ′ −→ W → 0
are equivalent if there exists a homomorphism h : V → V ′ such that the following
diagram is commutative:
f g
0 / U / V / W / 0
h
f′ g′
0 / U / V′ / W / 0
Remark: The expression commutative diagram means the following: Given are
certain modules and between them certain homomorphisms. One assumes that for
any pair of paths which start at the same module and also end at the same module,
the compositions of the corresponding homomorphisms coincide. It is enough to
check that for the smallest subdiagrams. For example, in the diagram appearing in
the next lemma, commutativity means that bf = f ′ a and cg = g ′b. (And therefore
also cgf = g ′ f ′ a.) In the above diagram, commutativity just means hf = f ′ and
g = g ′ h. We used the homomorphisms 1U and 1W to obtain a nicer looking diagram.
Arranging such diagrams in square form has the advantage that we can speak about
rows and columns of a diagram. A frequent extra assumption is that certain columns
or rows are exact. In this lecture course, we will see many more commutative
diagrams.
Lemma 3.11. Let
f g
0 / U / V / W / 0
a b c
f′ g′
0 / U′ / V′ / W′ / 0
be a commutative diagram with exact rows. If a and c are isomorphisms, then b is

also an isomorphism.
Proof. First, we show that b is injective: If b(v) = 0 for some v ∈ V , then cg(v) =
g ′b(v) = 0. This implies g(v) = 0 since c is an isomorphism. Thus v belongs to
Ker(g) = Im(f ). So v = f (u) for some u ∈ U. We get f ′ a(u) = bf (u) = b(v) = 0.
Now f ′ a is injective, which implies u = 0 and therefore v = f (u) = 0.
Second, we prove that b is surjective: Let v ′ ∈ V ′ . Then c−1 g ′ (v ′ ) ∈ W . Since g
is surjective, there is some v ∈ V with g(v) = c−1 g ′ (v ′ ). Thus cg(v) = g ′ (v ′). This
implies
g ′ (v ′ − b(v)) = g ′(v ′ ) − g ′ b(v) = g ′ (v ′ ) − cg(v) = 0.
So v ′ − b(v) belongs to Ker(g ′ ) = Im(f ′ ). Therefore there exists some u′ ∈ U ′ with
f ′ (u′) = v ′ − b(v). Let u = a−1 (u′ ). Because f ′ (u′ ) = f ′ a(u) = bf (u), we get
v ′ = f ′ (u′ ) + b(v) = b(f (u) + v). Thus v ′ is in the image of b. So we proved that b
is an isomorphism.
The method used in the proof of the above lemma is called “Diagram chasing”.
Lemma 3.11 shows that equivalence of short exact sequences is indeed an equivalence
relation on the set of all short exact sequences starting in a fixed module U and
ending in a fixed module W :
Given two short exact exact sequences (f, g) and (f ′ , g ′) like in the assumption of
Lemma 3.11. If there exists a homomorphism h : V → V ′ such that hf = f ′ and
g = g ′ h, then h−1 satisfies h−1 f ′ = f and g ′ = gh−1 . This proves the symmetry of
the relation.
If there is another short exact sequence (f ′′ , g ′′ ) with f ′′ : U → V ′′ and g ′′ : V ′′ →
W and a homomorphism h′ : V ′ → V ′′ such that h′ f ′ = f ′′ and g ′ = g ′′ h′ , then
h′ h : V → V ′′ is a homomorphism with h′ hf = f ′′ and g = g ′′ h′ h. This shows our
relation is transitive.
Finally, (f, g) is equivalent to itself, just take h = 1V . Thus the relation is reflexive.
A short exact sequence

f g
0→U −
→V −
→W →0
is a split exact sequence (or splits) if Im(f ) is a direct summand of V . In other
words, the sequence splits if f is a split monomorphism, or (equivalently) if g is a
split epimorphism. (Remember that Im(f ) = Ker(g).)
Lemma 3.12. A short exact sequence 0 → U → V → W → 0 splits if and only if
it is equivalent to the short exact sequence
1 ι 2 π
0→U −
→ U ⊕ W −→ W → 0,
where ι1 is the inclusion of U into U ⊕ W , and π2 is the projection from U ⊕ W
onto W with kernel U.
Proof. Let (f, g) with f : U → V and g : V → W be a short exact sequence. If it

splits, then f is a split monomorphism. Thus there exists some f ′ : V → U with
f ′ f = 1U . So
f g
0 / U / V / W / 0
h i
f′
g
ι1 π2
0 U / U ⊕W /W / /0
is a commutative diagram: If we write ι1 = t [1, 0] and π2 = [0, 1], then we see that
t ′
[f , g]f = t [1, 0] = ι1 and g = [0, 1] ◦ t [f ′ , g] = π2 ◦ t [f ′ , g]. Thus (f, g) is equivalent
to (ι1 , π2 ).
Vice versa, assume that (f, g) and (ι1 , π2 ) are equivalent. Thus there exists some
h : V → U ⊕W such that hf = ι1 and g = π2 h. Let π1 be the projection from U ⊕W
onto U with kernel W . Then π1 hf = π1 ι1 = 1U . Thus f is a split monomorphism.

3.9. Exercises. 1: Prove part (ii) of the above lemma.

2: Let K be a field of characteristic 0. For integers i, j ∈ Z with i ≤ j let M(i, j)
be the 2-module (K j−i+1 , Φ, Ψ) where
i  0 1 
i+1 0 1
Φ= ..  and Ψ =  .. ..  .
. . .
j−1 0 1
j 0
Compute Hom(M(i, j), M(k, l)) for all integers i ≤ j and k ≤ l.
3: Let
0 1 2
V = (K , ).
0 0

a b
Show: End(V ) is the set of matrices of the form with a, b ∈ K.
0 a
Compute the idempotents in End(V ).

Compute all direct sum decompositions V = V1 ⊕ V2 , with V1 and V2 submodules of
V.
4: Let    
0 0 0 0 0 0
V = (K 3 , 1 0 0 , 0 0 0).
0 0 0 1 0 0
Show: End(V ) is the set of matrices of the form

 
a 0 0
b a 0
c 0 a
with a, b, c ∈ K.
Use this to show that V is indecomposable.
Show that V is not simple.

5: Let V and W be J-modules. We know that V × W is again a J-module.
Let f : V → W be a module homomorphism, and let
Γf = {(v, f (v)) | v ∈ V }
be the graph of f .
Show: The map f →7 Γf defines a bijection between HomJ (V, W ) and the set of
submodules U ⊆ V × W with U ⊕ (0 × W ) = V × W .
6: Let  
0 1 0
V = (K 3 , 0 0 0).
0 0 0
Compute End(V ) (as a set of 3 × 3-matrices).

Determine all idempotents e in End(V ).
Determine all direct sum decompositions V = V1 ⊕V2 (with drawings in case K = R).
Describe the map e 7→ (Im(e), Ker(e)) (where e runs through the set of idempotents
in End(V )).
7: Let
f1 f2
0 / V1 / V / V2 / 0
a
g1 g2
0 / W1 / W / W2 / 0
be a diagram of J-modules with exact rows.
Show: There exists a homomorphism a1 : V1 → W1 with af1 = g1 a1 if and only if
there exists a homomorphism a2 : V2 → W2 with g2 a = a2 f2 .
8: Let
f1 f2 f3 f4
V1 / V2 / V3 / V4 / V5
a1 a2 a3 a4 a5
g1 g2 g3 g4
W1 / W2 / W3 / W4 / W5
be a commutative diagramm of J-modules with exact rows.
Show: If a1 is an epimorphism, and if a2 and a4 are monomorphisms, then a3 is a
monomorphism.
If a5 is a monomorphism, and if a2 and a4 are epimorphisms, then a3 is an epimor-

phism.
If a1 , a2 , a4 , a5 are isomorphisms, then a3 is an isomorphism.

9: Let
f g
0→U − →V −→W →0
be a short exact sequence of J-modules.
Show: The exact sequence (f, g) splits if and only if for all J-modules X the sequence
HomJ (X,f ) HomJ (X,g)
0 → HomJ (X, U) −−−−−−→ HomJ (X, V ) −−−−−−→ HomJ (X, W ) → 0
is exact. (By the results we obtained so far, it is enough to show that HomJ (X, g)
is surjective for all X.)
10: If the sequence
fi gi
0 → Ui −
→ Vi −
→ Ui+1 → 0
is exact for all i ∈ Z, then the sequence
fi gi−1 fi+1 gi
· · · → Vi−1 −−−→ Vi −−−→ Vi+1 → · · ·
is exact.
11: Construct an example of a short exact sequence

0 → U → U′ ⊕ W → W → 0
such that U 6∼
= U ′.
————————————————————————————-
4. Digression: Categories
This section gives a quick introduction to the concept of categories.
4.1. Categories. A category C consists of objects and morphisms, the objects

form a class, and for any objects X and Y there is a set C(X, Y ), the set of mor-
phisms from X to Y . Is f such a morphism, we write f : X → Y . For all objects
X, Y, Z in C there is a composition map
C(Y, Z) × C(X, Y ) → C(X, Z), (g, f ) 7→ gf,
which satisfies the following properties:
• For any object X there is a morphism 1X : X → X such that f 1X = f and

1X g = g for all morphisms f : X → Y and g : Z → X.
• The composition of morphisms is associative: For f : X → Y , g : Y → Z
and h : Z → A we assume (hg)f = h(gf ).
For morphisms f : X → Y and g : Y → Z we call gf : X → Z the composition of

f and g.
A morphism f : X → Y is an isomorphism if there exists a morphism g : Y → X

such that gf = 1X and f g = 1Y .
When necessary, we write Ob(C) for the class of objects in C. However for an object
X, we often just say “X lies in C” or write “X ∈ C”.
Remark: Note that we speak of a “class” of objects, and not of sets of objects, since
we want to avoid set theoretic difficulties: For example the J-modules do not form
a set, otherwise we would run into contradictions. (Like: “The set of all sets.”)
If C ′ and C are categories with Ob(C ′ ) ⊆ Ob(C) and C ′ (X, Y ) ⊆ C(X, Y ) for all
objects X, Y ∈ C ′ such that the compositions of morphisms in C ′ coincide with the
compositions in C, then C ′ is called a subcategory of C. In case C ′ (X, Y ) = C(X, Y )
for all X, Y ∈ C ′ , one calls C ′ a full subcategory of C.
We only look at K-linear categories: We assume additionally that the morphism
sets C(X, Y ) are K-vector spaces, and that the composition maps
C(Y, Z) × C(X, Y ) → C(X, Z)
are K-bilinear. In K-linear categories we often write Hom(X, Y ) instead of C(X, Y ).
By Mod(K) we denote the category of K-vector spaces. Let mod(K) be the category
of finite-dimensional K-vector spaces.
4.2. Functors. Let C and D be categories. A covariant functor F : C → D

associates to each object X ∈ C an object F (X) ∈ D, and to each morphism
f : X → Y in C a morphism F (f ) : F (X) → F (Y ) in D such that the following
hold:
• F (1X ) = 1F (X) for all objects X ∈ C;

• F (gf ) = F (g)F (f ) for all morphisms f, g in C such that their composition
gf is defined.
By a functor we always mean a covariant functor. A trivial example is the following:

If C ′ is a subcategory of C, then the inclusion is a functor.
Similarly, a contravariant functor F : C → D associates to any object X ∈ C an
object F (X) ∈ D, and to each morphism f : X → Y in C a morphism F (f ) : F (Y ) →
F (X) such that the following hold:
• F (1X ) = 1F (X) for all objects X ∈ C;

• F (gf ) = F (f )F (g) for all morphisms f, g in C such that their composition
gf is defined.
Thus if we deal with contravariant functors, the order of the composition of mor-
phisms is reversed.
If C and D are K-linear categories, then a covariant (resp. contravariant) functor

F : C → D is K-linear, if the map C(X, Y ) → D(F (X), F (Y )) (resp. C(X, Y ) →
D(F (Y ), F (X))) defined by f 7→ F (f ) is K-linear for all objects X, Y ∈ C.
In Section 3.8 we will see examples of functors.
4.3. Equivalences of categories. Let F : C → D be a functor. Then F is called

full, if for all objects X, Y ∈ C the map C(X, Y ) → D(F (X), F (Y )), f 7→ F (f ) is
surjective, and F is faithful if these maps are all injective. If every object X ′ ∈ D
is isomorphic to an object F (X) for some X ∈ C, then F is dense.
A functor which is full, faithful and dense is called an equivalence (of categories).
If F : C → D is an equivalence, then there exists an equivalence G : D → C such that
for all objects C ∈ C the objects C and GF (C) are isomorphic, and for all objects
D ∈ D the objects D and F G(D) are isomorphic.
If F : C → D is an equivalence of categories such that Ob(C) → Ob(D), X → F (X)
is bijective, then F is called an isomorphism (of categories). If F is such an
isomorphism, then there exists a functor G : D → C such that C = GF (C) for all
objects C ∈ C and D = F G(D) for all objects D ∈ D. Then G is obviously also an
isomorphism. Isomorphisms of categories are very rare. In most constructions which
yield equivalences F of categories, it is difficult to decide if F sends two isomorphic
objects X 6= Y to the same object.
4.4. Module categories. Given a class M of J-modules, which is closed under

isomorphisms and under finite direct sums. Then M (together with the homomor-
phisms between the modules in M) is called a module category.
(Thus we assume the following: If V ∈ M and if V ∼ = V ′ , then V ′ ∈ M. Also, if
V1 , . . . , Vt are modules in M, then V1 ⊕ · · · ⊕ Vt ∈ M.)
If we say that f : X → Y is a homomorphism in M, then this means that both

modules X and Y lie in M (and that f is a homomorphism).
The module category of all J-modules is denoted by M(J). Thus Mod(K) = M(∅).
For J = {1, . . . , n} set M(n) := M(J).
4.5. Hom-functors. Typical examples of functors are Hom-functors: Let M be

a module category, which consists of J-modules. Each J-module V ∈ M yields a
functor
HomJ (V, −) : M → mod(K)
which associate to any module X ∈ M the vector space HomJ (V, X) and to any
morphism h : X → Y in M the morphism HomJ (V, h) : HomJ (V, X) → HomJ (V, Y )
in mod(K).
Similarly, every object X ∈ M yields a contravariant functor
HomJ (−, X) : M → mod(K).
4.6. Exercises. 1: For c ∈ K let Nc be the module category of 1-modules (V, φ)

with (φ − c1V )m = 0 for some m. Show that all module categories Nc are isomorphic
(as categories) to N := N0 .
————————————————————————————-
5. Examples of infinite dimensional 1-modules
5.1. The module N(∞). Let V be a K-vector space with basis {ei | i ≥ 1}. Define
a K-linear endomorphism φ : V → V by φ(e1 ) = 0 and φ(ei ) = ei−1 for all i ≥ 2.
We want to study the 1-module

N(∞) := (V, φ).
We clearly have a chain of submodules
N(0) ⊂ N(1) ⊂ · · · ⊂ N(i) ⊂ N(i + 1) ⊂ · · ·
of N(∞) where N(0) = 0, and N(i) is the submodule with basis e1 , . . . , ei where
i ≥ 1. Clearly, [
N(∞) = N(i).
i∈N0
The following is clear and will be used in the proof of the next lemma: Every
submodule of a J-module is a sum of cyclic modules.
Lemma 5.1. The following hold:
(i) The N(i) are the only proper submodules of N(∞);

(ii) The N(i) are cyclic, but N(∞) is not cyclic;
(iii) N(∞) is indecomposable.
Proof. First we determine the cyclic submodules: Let x ∈ V . Thus there exists
some n such that x ∈ N(n) and
Xn
x= ai ei .
i=1
If x = 0, the submodule U(x) generated by x is just N(0) = 0. Otherwise, U(x) is

equal to N(i) where i the the maximal index 1 ≤ j ≤ n such that aj 6= 0. Note that
the module N(∞) itself is therefore not cyclic.
Now let U be any submodule of V . It follows that U is a sum of cyclic modules,
thus X
U= N(i)
i∈I
for some I ⊆ N0 . If I is finite, we get U = N(max{i ∈ I}), otherwise we have
U = N(∞). In particular, this implies that N(∞) is indecomposable.
A J-module V is uniform if for any non-zero submodules U1 and U2 one has U1 ∩

U2 6= 0. It follows from the above considerations that N(∞) is a uniform module.
5.2. Polynomial rings. This section is devoted to study some interesting and im-
portant examples of modules arising from the polynomial ring K[T ] in one variable
T.
As always, K is a field. Recall that the characteristic char(K) is by definition the

minimum n such that the n-fold sum 1 + 1 + · · · + 1 of the identity element of K is
zero, if such a minimum exists, and char(K) = 0 otherwise. One easily checks that
char(K) is either 0 or a prime number.
The elements in K[T ] are of the form

m
X
f= ai T i
i=0
with ai ∈ K for all i and m ≥ 0. We set T 0 = 1. One calls f monic if an = 1 where

n is the maximal 1 ≤ i ≤ m such that ai 6= 0. If f 6= 0, then the degree of f is the
maximum of all i such that ai 6= 0. Otherwise the degree of f is −∞.
By P we denote the set of monic, irreducible polynomials in K[T ]. For example, if

K = C we have P = {T − c | c ∈ C}.
Exercise: Determine P in case K = R. (Hint: All irreducible polynomials over R
have degree at most 2.)
Note that {1, T 1, T 2 , . . .} is a basis of the K-vector space K[T ].
Let
T · : K[T ] → K[T ]
be the K-linear map which maps a polynomial f to T f . In particular, T i is mapped
to T i+1 .
Another important K-linear map is
d
: K[T ] → K[T ]
dT
P
which maps a polynomial m i
i=0 ai T to its derivative
X m
d
(f ) = ai iT i−1 .
dT i=1
Of course, in the above expression, i stands for the i-fold sum 1 + 1 + · · · + 1 of

the identity 1 of K. Thus, if char(K) = p > 0, then i = 0 in K if and only if i is
d
divisible by p. In particular dT (T np ) = 0 for all n ≥ 0.
We know that every polynomial p can be written as a product
p = cp1 p2 · · · pt
where c is a constant (degree 0) polynomial, and the pi are monic irreducible poly-
nomials. The polynomials pi and c are uniquely determined up to reordering.
d
5.3. The module (K[T ], dT ). We want to study the 1-module
d
V := (K[T ], ).
dT
Let Vn be the submodule of polynomials of degree ≤ n in V . With respect to the

basis 1, T 1 , . . . , T n we get
0 1 
0 2
Vn ∼
= (K n+1 ,  . . . . ).
. .
0 n
0
Exercise: If char(K) = 0, then

0 1  0 1 
0 2 0 1
(K n+1 ,  . . . . ) ∼
= (K n+1 ,  . . . . ).
. . . .
0 n 0 1
0 0
Proposition 5.2. We have

(
d ∼ N(∞) if char(K) = 0,
(K[T ], )= L
dT i∈N0 N(p) if char(K) = p.
Proof. Define a K-linear map

d
f : (K[T ], ) → N(∞)
dT
by T i →
7 i! · ei+1 where i! := i(i − 1) · · · 1 for i ≥ 1. Set 0! = 1. We have

d
f (T ) = f (iT i−1 ) = if (T i−1 ) = i(i − 1)! · ei = i! · ei .
i
dT
On the other hand,
φ(f (T i )) = φ(i! · ei+1 ) = i! · ei .
This implies that the diagram
f
K[T ] / N(∞)
d
dT
φ
f

K[T ] / N(∞)
commutes, and therefore f is a homomorphism of 1-modules. If char(K) = 0, then
f is an isomorphism with inverse
1
f −1 : ei+1 7→ · T i
i!
where i ≥ 0.
Now assume char(K) = p > 0. We get i! = 0 if and only if i ≥ p.
The 1-module M
(W, φ) := N(p)
i∈N0
has as a basis {eij | i ∈ N0 , 1 ≤ j ≤ p} where
(
0 if j = 1,
φ(eij ) =
ei,j−1 otherwise.
Define a K-linear map
f : W → K[T ]
by
1
eij 7→ T ip+j−1.
(j − 1)!
Since j ≤ p we know that p does not divide (j − 1)!, thus (j − 1)! 6= 0 in K. One
easily checks that f defines a vector space isomorphism.
Exercise: Prove that
d
f (φ(eij )) = (f (eij ))
dT
and determine f −1 .
We get that f is an isomorphism of 1-modules.
5.4. The module (K[T ], T ·). Next, we want to study the 1-module
V := (K[T ], T ·).
Pn
Let a = i=0 ai T i be a polynomial in K[T ]. The submodule U(a) of V generated
by a is
(a) := U(a) = {ab | b ∈ K[T ]}.
We call (a) the principal ideal generated by a.
Proposition 5.3. All ideals in the ring K[T ] are principal ideals.
Proof. Look it up in any book on Algebra.
In other words: Each submodule of V is of the form (a) for some a ∈ K[T ].
Now it is easy to check that (a) = (b) if and only if a|b and b|a if and only if there
exists some c ∈ K ∗ with b = ca. (For polynomials p and q we write p|q if q = pf for
some f ∈ K[T ].)
It follows that for submodules (a) and (b) of V we have
(a) ∩ (b) = l.c.m.(a, b)
and
(a) + (b) = g.c.d.(a, b).
Here l.c.m.(a, b) denotes the lowest common multiple, and g.c.d.(a, b) is the greatest
common divisor.
Let R = K[T ] be the polynomial ring in one variable T , and let a1 , . . . , an be

elements in R.
Lemma 5.4 (Bézout). Let R = K[T ] be the polynomial ring in one variable T ,
and let a1 , . . . , an be elements in R. There exists a greatest common divisor d of
a1 , . . . , an , and there are elements ri in R such that
Xn
d= ri ai .
i=1
It follows that d is the greatest common divisor of elements a1 , . . . , an in K[T ] if and

only if the ideal (a1 , . . . , an ) generated by the ai is equal to the ideal (d) generated
by d.
The greatest common divisor of elements a1 , . . . , an in K[T ] is 1 if and only if there
exists elements r1 , . . . , rn in K[T ] such that
X n
1= ri ai .
i=1
Let P be the set of monic irreducible polynomials in K[T ]. Recall that every poly-
nomial p 6= 0 in K[T ] can be written as
p = cpe11 pe22 · · · pet t
where c ∈ K ∗ , ei ≥ 1 and the pi are pairwise different polynomials in P. Further-
more, c, the ei and the pi are uniquely determined (up to reordering).
If b|a then there is an epimorphism
K[T ]/(a) → K[T ]/(b)
defined by p + (a) 7→ p + (b).
Now let p be a non-zero polynomial with

p = cpe11 pe22 · · · pet t
as above.
Proposition 5.5 (Chinese Reminder Theorem). There is an isomorphism of 1-
modules
t
Y
K[T ]/(p) → K[T ]/(pei i ).
i=1
Proof. We have pei i |p and therefore there is an epimorphism (of 1-modules)

πi : K[T ]/(p) → K[T ]/(pei i ).
This induces a homomorphism
t
Y
π : K[T ]/(p) → K[T ]/(pei i )
i=1
defined by π(a) = (π1 (a), . . . , πt (a)). Clearly, a ∈ Ker(π) if and only if πi (a) = 0 for
all i if and only if pei i |a for all i if and only if p|a. This implies that π is injective.
For a polynomial a of degree n we have dim K[T ]/(a) = n, and the residue classes
of 1, T, . . . , T n−1 form a basis of K[T ]/(a).
In particular, dim K[T ]/(pei i ) = deg(pei i ) and
t
Y
dim K[T ]/(pei i ) = deg(p).
i=1
Thus for dimension reasons we get that π must be also surjective.
Exercises: Let p be an irreducible polynomial in K[T ].

Show: The module (K[T ]/(p), T ·) is a simple 1-module, and all simple 1-modules
(over a field K) are isomorphic to a module of this form.
Show: The submodules of the factor module K[T ]/(pe ) are
0 = (pe )/(pe ) ⊂ (pe−1 )/(pe ) ⊂ · · · ⊂ (p)/(pe ) ⊂ K[T ]/(pe ),
and we have
= (pi )/(pi+1 ) ∼
((pi )/(pe ))/((pi+1 )/(pe )) ∼ = K[T ]/(p).
Special case: The polynomial T is an irreducible polynomial in K[T ], and one easily
checks that the 1-modules (K[T ]/(T e ), T ·) and N(e) are isomorphic.
Notation: Let p ∈ P be a monic, irreducible polynomial in K[T ]. Set

n
N = (K[T ]/(pn ), T ·).
p

This is a cyclic and indecomposable 1-module. The modules N 1p are the only
simple 1-modules (up to isomorphism).
Exercise: If p = T − c for some c ∈ K, then we have
c 1 
c 1
n ∼
N = (K n , Φ :=  . . . . . . ).
p c 1
c
i n

The residue classes of the elements (T − c) , 0 ≤ i ≤ n − 1 form a basis of N p
.
We have
T · (T − c)i = (T − c)i+1 + c(T − c)i .
The module c 1 
c 1
(K n ,  . . . . )
. .
c 1
c
has as a basis the canonical basis vectors e1 , . . . , en . We have Φ(e1 ) = ce1 and
Φ(ei ) = cei + ei−1 if i ≥ 2. Then
f : (T − c)i 7→ en−i
for i ≥ 0 yields an isomorphism of 1-modules: One easily checks that
f (T · (T − c)i ) = Φ(f ((T − c)i ))
for all i ≥ 0.
Conclusion: If we can determine the set P of irreducible polynomials in K[T ], then
one has quite a good description of the submodules and also the factor modules of
(K[T ], T ·). But of course, describing P is very hard (or impossible) if the field K is
too complicated.
5.5. The module (K(T ), T ·). Let K(T ) be the ring of rational functions in one
variable T . The elements of K(T ) are of the form pq where p and q are polynomials
′
in K[T ] wit q 6= 0. Furthermore, pq = pq′ if and only if pq ′ = qp′ . Copying the
ususal rules for adding and multiplying fractions, K(T ) becomes a ring (it is even
a K-algebra). Clearly, all non-zero elements in K(T ) have an inverse, thus K(T ) is
also a field. It contains K[T ] as a subring, the embedding given by p 7→ 1p .
Set K[T ] = (K[T ], T ·) and K(T ) = (K(T ), T ·).

Obviously, K[T ] is a submodule of K(T ). But there are many other interesting
submodules:
For p ∈ P, set

−1 q
K[T, p ] = | q ∈ K[T ], n ∈ N0 ⊂ K(T ).
pn
For example, if p = T , we can think of the elements of K[T, T −1 ] as linear combi-
nations X
ai T i
i∈Z
1
with only finitely many of the ai being non-zero. Here we write T −m = Tm
for
m ≥ 1.
5.6. Exercises. 1: Show: The module K[T, T −1 ]/K[T ] is isomorphic to N(∞). Its
basis are the residue classes of T −1 , T −2, . . ..
2: Let K[T ] be the vector space of polynomials in one variable

P T with coefficients in
d
a field K, and let dT be the differentiation map, i.e if p = ni=0 ai T i is a polynomial,
then
Xn
d
(p) = ai iT i−1 .
dT i=1
d
Show that the 1-module (K[T ], dT ) is indecomposable if char(K) = 0.
d
Write (K[T ], dT ) as a direct sum of indecomposable modules if char(K) > 0.
3: Let T · be the map which sends a polynomial p to T p.

d
Show that the 2-module (K[T ], dT , T ·) is simple and that K ∼ d
= End(K[T ], dT , T ·) if
char(K) = 0.
d
Compute End(K[T ], dT , T ·) if char(K) > 0.
d
Show that (K[T ], dT , T ·) is not simple in case char(K) > 0.
For endomorphisms f and g of a vector space let [f, g] = f g − gf be its commutator.

d
Show that [T ·, dT ] = 1.
4: Let K be a field, and let P be the set of monic irreducible polynomials in K[T ].
For a ∈ K(T ) set
K[T ]a = {f a | f ∈ K[T ]} ⊂ K(T ).
For every p ∈ P let K[T, p−1 ] be the subalgebra of K(T ) generated by T and p−1 .
In other words,
q
K[T, p−1 ] = { | q ∈ K[T ], n ∈ N0 } ⊂ K(T ).
pn
a: Show: The modules K[T ]p−n /K[T ] and K[T ]/(pn ) are isomorphic. Use this to
determine the submodules of K[T ]p−n /K[T ].
b: If U is a proper submodule of K[T ]p−n /K[T ], then U = K[T ]p−n /K[T ] for some
n ∈ N0 .
c: We have
X
K(T ) = K[T, p−1 ].
p∈P
Let
ιp : K[T, p−1 ]/K[T ] → K(T )/K[T ]
be the inclusion.
Show: The homomorphism
M X
ι= ιp : (K[T, p−1 ]/K[T ]) → K(T )/K[T ]
p∈P p∈P
is an isomorphism.
d: Determine the submodules of K(T )/K[T ].
————————————————————————————-
6. Semisimple modules and their endomorphism rings
Some topics discussed in this section are also known as “Artin-Wedderburn Theory”.
Just open any book on Algebra.
6.1. Semisimple modules. A module V is simple (or irreducible) if V 6= 0 and

the only submodules are 0 and V .
A module V is semisimple if V is a direct sum of simple modules.
A proper submodule U of a module V is called a maximal submodule of V , if

there does not exist a submodule U ′ with U ⊂ U ′ ⊂ V . It follows that a submodule
U ⊆ V is maximal if and only if the factor module V /U is simple.
Theorem 6.1. For a module V the following are equivalent:
(i) V is semisimple;
(ii) V is a sum of simple modules;
(iii) Every submodule of V is a direct summand.
The proof of Theorem 6.1 uses the Axiom of Choice:

Axiom 6.2 (Axiom of Choice). Let f : I → L be a surjective map of sets. Then
there exists a map g : L → I such that f g = 1L .
Let I be a partially ordered set. A subset C of I is a chain in I if for all c, d ∈ C

we have c ≤ d or d ≤ c. An equivalent formulation of the Axiom of Choice is the
following:
Axiom 6.3 (Zorn’s Lemma). Let I be a non-empty partially ordered set. If for
every chain in I there exists a supremum, then I contains a maximal element.
This is not surprising: The implication (ii) =⇒ (i) yields the existence of a basis
of a vector space. (We just look at the special case J = ∅. Then J-modules are just
vector spaces. The simple J-modules are one-dimensional, and every vector space
is a sum of its one-dimensional subspaces, thus condition (ii) holds.)
Proof of Theorem 6.1. The implication (i) =⇒ (ii) is obvious. Let us show (ii) =⇒
(iii): Let V be a sum of simple submodules, and let U be a submodule of V . Let W
be the set of submodules W of V with U ∩ W = 0. Together with the inclusion ⊆,
the set W is a partially ordered set. Since 0 ∈ W, we know that W is non-empty.
If W ′ ⊆ W is a chain, then
[
W′ = W
W ∈W ′
belongs to W: If x ∈ U ∩ W ′, then x belongs to some W in W ′ , and therefore
x ∈ U ∩ W = 0.
Now Zorn’s Lemma 6.3 says that W contains a maximal element. So let W ∈ W be
maximal. We know that U ∩ W = 0. On the other hand, we show that U + W = V :
Since V is a sum of simple submodules, it is enough to show that each simple
submodule of V is containd in U + W . Let S be a simple submodule of V . If we
assume that S is not contained in U + W , then (U + W ) ∩ S is a proper submodule
of S. Since S is simple, we get (U + W ) ∩ S = 0, and therefore U ∩ (W + S) = 0:
If u = w + s with u ∈ U, w ∈ W and s ∈ S, then u − w = s ∈ (U + W ) ∩ S = 0.
Thus s = 0 and u = w ∈ U ∩ W = 0.
This implies that W + S belongs to W. The maximality of W in W yields that
W = W +S and therefore we get S ⊆ W , which is a contradiction to our assumption
S 6⊆ U + W . Thus we see that U + W = V . So W is a direct complement of U in
V.
(iii) =⇒ (ii): Let S be the set of submodules of V , which are a sum of simple
submodules of V . We have 0 ∈ S. (We can think of 0 as the sum over an empty set
of simple submodules of V .)
Together with the inclusion ⊆, the set S forms a partially ordered set. Since 0
belongs to S, we know that S is non-empty.
If S ′ is a chain in S, then
[
U
U ∈S ′
belongs to S. Zorn’s Lemma tells us that S contains a maximal element. Let U be

such a maximal element.
We claim that U = V : Assume there exists some v ∈ V with v ∈ / U. Let W be the
set of submodules W of V with U ⊆ W and v ∈ / W . Again we interpret W together
with the inclusion ⊆ as a partially ordered set. Since U ∈ W, we know that W is
non-empty, and if W ′ is a chain in W, then
[
W
W ∈W ′
belongs to W. Zorn’s Lemma yields a maximal element in W, say W . Let W ′ be

the submodule generated by W and v. Since v ∈ / W , we get W ⊂ W ′ . On the
other hand, if X is a submodule with W ⊆ X ⊂ W ′ , then v cannot be in X, since
W ′ is generated by W and v. Thus X belongs to W, and the maximality of W
implies W = X. Thus we see that W is a maximal submodule of W ′ . Condition
(iii) implies that W has a direct complement C. Let C ′ = C ∩ W ′ . We have
W ∩ C ′ = W ∩ (C ∩ W ′ ) = 0, since W ∩ C = 0. Since the submodule lattice of a
module is modular (and since W ⊆ W ′ ), we get
W + C ′ = W + (C ∩ W ′ ) = (W + C) ∩ W ′ = V ∩ W ′ = W ′ .
This implies
W ′ /W = (W + C ′ )/W ∼
= C ′ /(W ∩ C ′ ) = C ′ .
Therefore C ′ is simple.
V A
{ AA
{{ AA
{{{ A AA
{{
W′C C
{ CC ~
{{ CC ~~
{{{ CC ~~~
{{ C ~~
WC C′
C CC {{
CC {{{
CC {
{{
0
Because U is a sum of simple modules, we get that U +C ′ is a sum of simple modules,
thus it belongs to S. Now U ⊂ U + C ′ yields a contradiction to the maximality of
U in S.
(ii) =⇒ (i): We show the following stronger statement:
LemmaP6.4. Let V be a module, and let U be the set of simple submodules
L U of V .
If V = U ∈U U, then there exists a subset U ′ ⊆ U such that V = U ∈U ′ U.
P
Proof. A subset U ′ of U is called independent, if the sum U ∈U ′ U is a direct sum.
Let T be the set of independent subsets of U, together with the inclusion of sets ⊆
this is a partially ordered set. Since the empty set belongs to T we know that T is
non-empty. If T ′ is a chain in T , then
[
U′
U ′ ∈T ′
is obviously in T . Thus by Zorn’s Lemma there exists a maximal element in T . Let

U ′ be such a maximal element. Set
X
W = U.
U ∈U ′
Since U ′ belongs to T , we know that this is a direct sum. We claim that W = V :

Otherwise there would exist a submodule U in U with U 6⊆ W , because V is the
sum of the submodules in U. Since U is simple, this would imply U ∩ W = 0. Thus
the set U ′ ∪ {U} is independent and belongs to T , a contradiction to the maximality
of U ′ in T .
This finishes the proof of Theorem 6.1.
Here is an important consequence of Theorem 6.1:

Corollary 6.5. Submodules and factor modules of semisimple modules are semisim-
ple.
Proof. Let V be a semisimple module. If W is a factor module of V , then W = V /U

for some submodule U of V . Now U has a direct complement C in V , and C is
isomorphic to W . Thus every factor module of V is isomorphic to a submodule of

V . Therefore it is enough to show that all submodules of V are semisimple.
V @
~~ @ @@
~~~ @@
~~ @
U@ C
@ @@ ~~
@@ ~~~
@@ ~
~~
0
Let U be submodule of V . We check condition (iii) for U: Every submodule U ′ of
U is also a submodule of V . Thus there exists a direct complement C of U ′ in V .
Then C ∩ U is a direct complement of U ′ in U.
V GG
ww GG
www GG
ww GG
ww G
C FF U
x AA
FF xx AA
FF x
FF xxx AA
A
F xx
C ∩ UG U′
GG }
GG }}
GG
GG }}}
}}
0
Of course U ′ ∩ (C ∩ U) = 0, and the modularity yields U ′ + (C ∩ U) = (U ′ + C) ∩ U =
V ∩ U = U.
Let V be a semisimple module. For every simple module S let VS be the sum of
all submodules U of V such that U ∼= S. The submodule VS depends only on the
isomorphism class [S] of S. Thus we obtain a family (VS )[S] of submodules of V
which are indexed by the isomorphism classes of simple modules. The submodules
VS are called the isotypical components of V .
Proposition 6.6. Let V be a semisimple module. Then the following hold:
L
• V = [S] VS ;
• If V ′ is a submodule of V , then VS′ = V ′ ∩ VS ;
• If W is another semisimple module and f : V → W is a homomorphism,
then f (VS ) ⊆ WS .
P submodule of V∼, and if W is

Proof. First, we show the following: If U is a simple
a set of simple submodules
P of V such that V = W ∈W W , then U = W
L for some
′
W ∈ W: Since V = W ∈W W , there is a subset W of W such that V = W ∈W ′ W .
For every W ∈ W ′ let πW : V → W be the corresponding projection. Let
M
ι: U → V = W
W ∈W ′
be the inclusion homomorphism. If U and W are not isomorphic, then πW ◦ ι = 0.

Since ι 6= 0 there must be some W ∈ W ′ with πW ◦ ι 6= 0. Thus U and W are
isomorphic.
P
Since V is semisimple, we have V = [S] VS . To show that this sum is direct, let us
look at a fixed isomorphism class [S]. Let T be the set of all isomorphism classes of
simple modules different from [S]. Define
X
U = VS ∩ VT .
[T ]∈T
Since U is a submodule of V , we know that U is semisimple. Thus U is generated

by simple modules. If U ′ is a simple submodule of U, then U ′ is isomorphic to S,
because U and therefore
P also U ′ are submodules of VS . On the other hand, since U ′
is a submodule of [T ]∈T VT , we get that U ′ is isomorphic to some T with [T ] ∈ T ,
a contradiction. Thus U contains no simple submodules, and therefore U = 0.
If V ′ is a submodule of V , then we know that V ′ is semisimple. Obviously, we have

VS′ ⊆ V ′ ∩VS . On the other hand, every simple submodule of V ′ ∩VS is isomorphic to
S and therefore contained in VS′ . Since V ′ ∩ VS is generated by simple submodules,
we get V ′ ∩ VS ⊆ VS′ .
Finally, let W be also a semisimple module, and let f : V → W be a homomorphism.

If U is a simple submodule of VS , then U ∼ = S. Now f (U) is either 0 or again
isomorphic to S. Thus f (U) ⊆ WS . Since VS is generated by its simple submodules,
we get f (VS ) ⊆ WS .
6.2. Endomorphism rings of semisimple modules. A skew field is a ring D

(with 1) such that every non-zero element in D has a multiplicative inverse.
Lemma 6.7 (Schur (Version 1)). Let S be a simple module. Then the endomorphism
ring End(S) is a skew field.
Proof. We know that End(S) is a ring. Let f : S → S be an endomorphism of S. It

follows that Im(f ) and Ker(f ) are submodules of S. Since S is simple we get either
Ker(f ) = 0 and Im(f ) = S, or we get Ker(f ) = S and Im(f ) = 0. In the first case,
f is an isomorphism, and in the second case f = 0. Thus every non-zero element in
End(S) is invertible.
Let us write down the following reformulation of Lemma 6.7:

Lemma 6.8 (Schur (Version 2)). Let S be a simple module. Then every endomor-
phism S → S is either 0 or an isomorphism.
Let V be a semisimple module, and as before let VS be its isotypical components.

We have M
V = VS ,
[S]
and every endomorphism f of V maps VS to itself. Let fS : VS → VS be the homo-

morphism obtained from f via restriction to VS , i.e. fS (v) = f (v) for all v ∈ VS .
Then f 7→ (fS )[S] defines an algebra isomorphism
Y
End(V ) → End(VS ).
[S]
Products of rings: Let I be an index set, and for each i ∈ I let Ri be a ring. By
Y
Ri
i∈I
we denote the product of the rings Ri . Its elements are the sequences (ri )i∈I
with ri ∈ Ri , and the addition and multiplication is defined componentwise, thus
(ri )i + (ri′ )i = (ri + ri′ )i and (ri )i · (ri′ )i = (ri ri′ )i .
The above isomorphism tells us, that to understand End(V ), we only have to un-
derstand the rings End(VS ). Thus assume V = VS . We have
M
V = S
i∈I
for some index set I. The structure of End(V ) only depends on the skew field
D = End(S) and the cardinality |I| of I.
If I is finite, then |I| = n and End(V ) is just the ring Mn (D) of n × n-matrices with
entries in D.
If I is infinite, we can interpret End(V ) as an “infinite matrix ring”: Let MI (D)

be the ring of column finite matrices: Let R be a ring. Then the elements of
MI (R) are double indexed families (rij )ij with i, j ∈ I and elements rij ∈ R such
that for every j only finitely many of the rij are non-zero. Now one can define the
multiplication of two such column finite matrices as
!
X
′ ′
(rij )ij · (rst )st = rij rjt .
j∈I it
The addition is defined componentwise. (This definition makes also sense if I is

finite, where we get the usual matrix ring with rows and columns indexed by the
elements in I and not by {1, . . . , n} as usual.)
Lemma 6.9. For every index set I and every finitely generated module W we have
!
M
End W ∼ = MI (End(W )).
i∈I
L L
Proof. Let ιj : W → i∈I W be the canonical inclusions, and let πj : i∈I W → W
be the canonical projections. We map
!
M
f ∈ End W
i∈I
to the double indexed family (πi ◦ f L

◦ ιj )ij . Since W is finitely generated, the
L image
of every homomorphism f : W → i∈I W is contained in a submodule i∈I ′ W
′
where I is a finite subset of I. This yields that the matrix (πi ◦ f ◦ ιj )ij is column
finite.
6.3. Exercises. 1: Let K be an algebraically closed field.

Classify the simple 1-modules (V, φ).
Classify the 2-dimensional simple 2-modules (V, φ, ψ).
For every n ≥ 1 construct an n-dimensional simple 2-module (V, φ, ψ).
2: Show that every simple 1-module is finite-dimensional.

Show: If K is algebraically closed, then every simple 1-module is 1-dimensional.
Show: If K = R, then every simple 1-module is 1- or 2-dimensional.
3: Let (V, φ1 , φ2 ) be a 2-module with V 6= 0 and [φ1 , φ2 ] = 1.
Show: If char(K) = 0, then V is infinite dimensional.

Hint: Assume V is finite-dimensional, and try to get a contradiction. You could
work with the trace (of endomorphisms of V ). Which endomorphisms does one have
to look at?

1 0
4: Let A = ∈ M(2, C). Find a matrix B ∈ M(2, C) such that (C2 , A, B) is
0 2
simple.

2 0
5: Let A = ∈ M(2, C). Show that there does not exist a matrix B ∈
0 2
M(2, C) such that (C2 , A, B) is simple.
6: Let V = (V, φj )j∈J be a finite-dimensional J-module such that all φj are diago-
nalizable.
Show: If [φi , φj ] = 0 for all i, j ∈ J and if V is simple, then V is 1-dimensional.
————————————————————————————-
7. Socle and radical of a module
7.1. Socle of a module. The socle of a module V is by definition the sum of

all simple submodules of V . We denote the socle of V by soc(V ). Thus soc(V ) is
semisimple and every semisimple submodule U of V is contained in soc(V ).
Let us list some basic properties of socles:
Lemma 7.1. We have V = soc(V ) if and only if V is semisimple.
Proof. Obvious.
Lemma 7.2. soc(soc(V )) = soc(V ).
Proof. Obvious.
Lemma 7.3. If f : V → W is a homomorphism, then f (soc(V )) ⊆ soc(W ).
Proof. The module f (soc(V )) is isomorphic to a factor module of soc(V ), thus it is

semisimple. As a semisimple submodule of W , we know that f (soc(V )) is contained
in soc(W ).
Lemma 7.4. If U is a submodule of V , then soc(U) = U ∩ soc(V ).
Proof. Since soc(U) is semisimple, it is a submodule of soc(V ), thus of U ∩ soc(V ).

On the other hand, U ∩ soc(V ) is semisimple, since it is a submodule of soc(V ).
Because U ∩soc(V ) is a semisimple submodule of U, we get U ∩soc(V ) ⊆ soc(U).
Lemma 7.5. If Vi with i ∈ I are modules, then
!
M M
soc Vi = soc(Vi ).
i∈I i∈I
L
Proof. Let V = i∈I Vi . Every submodule soc(Vi ) is semisimple, thus it is contained
in soc(V ). Vice versa, let U be a simple submodule of V , and let πi : V → Vi be
the canonical projections. ThenL πi (U) is either 0 or simple, thus it is contained
in soc(Vi ). This implies U ⊆ i∈I soc(V
L i ). The simple submodules of V generate
soc(V ), thus we also have soc(V ) ⊆ i∈I soc(Vi ).
7.2. Radical of a module. The socle of a module V is the largest semisimple

submodule. One can ask if every module has a largest semisimple factor module.
For |J| = 1 the example V = (K[T ], T ·) shows that this is not the case: For every
irreducible polynomial p in K[T ], the module K[T ]/(p) is simple with basis the
residue classes of 1, T, T 2 , . . . , T m−1 where m is the degree of the polynomial p.
Now assume that W = K[T ]/U is a largest semisimple factor module of V . This
would imply U ⊆ (p) for every irreducible polynomial p. Since
\
(p) = 0,
p∈P
we get U = 0 and therefore W = K[T ]. Here P denotes the set of all irreducible
polynomials in K[T ]. But V is not at all semisimple. Namely V is indecomposable
and not simple. In fact, V does not contain any simple submodules.
Recall: A submodule U of a module V is called a maximal submodule if U ⊂ V
and if U ⊆ U ′ ⊂ V implies U = U ′ .
By definition the radical of V is the intersection of all maximal submodules of V .

The radical of V is denoted by rad(V ).
Note that rad(V ) = V if V does not contain any maximal submodule. For example,
rad(N(∞)) = N(∞).
The factor module V / rad(V ) is called the top of V and is denoted by top(V ).
Lemma 7.6. Let V be a module. The radical of V is the intersection of all submod-
ules U of V such that V /U is semisimple.
Proof. Let r(V ) be the intersection of all submodules U of V such that V /U is

semisimple. Clearly, we get r(V ) ⊆ rad(V ). To get the other inclusion rad(V ) ⊆
r(V ), let U be a submodule of V with V /U L semisimple. We can write V /U as a
direct sum of simple modules Si , say V /U = i∈I Si . For every i ∈ I let Ui be the
kernel of the projection V → V /U → Si . This isTa maximal submodule of V , and
therefore we know that rad(V ) ⊆ Ui . Since U = i∈I Ui , we get rad(V ) ⊆ U which
implies rad(V ) ⊆ r(V ).
Note that in general the module V / rad(V ) does not have to be semisimple: If
V = (K[T ], T ·), then from the above discussion we get V / rad(V ) = V and V is not
semisimple. However, if V is a “module of finite length”, then the factor module
V / rad(V ) is semisimple. This will be discussed in Part 2, see in particular Lemma
10.9.
Let us list some basic properties of the radical of a module:
Lemma 7.7. We have rad(V ) = 0 if and only if 0 can be written as an intersection
of maximal submodules of V .
Lemma 7.8. If U is a submodule of V with U ⊆ rad(V ), then rad(V /U) =
rad(V )/U. In particular, rad(V / rad(V )) = 0.
Proof. Exercise.
Lemma 7.9. If f : V → W is a homomorphism , then f (rad(V )) ⊆ rad(W ).
Proof. We show that f (rad(V )) is contained in every maximal submodule of W : Let

U be a maximal submodule of W . If f (V ) ⊆ U, then we get of course f (rad(V )) ⊆ U.
Thus, assume f (V ) 6⊆ U. It is easy to see that U ∩ f (V ) = f (f −1 (U)).
t
W LL
tt LLL
ttt LLL
LLL
t
tt
tt
U II f (V )
II s
II
II sssss
II s
sss
f (f −1 (U))
Thus
W/U ∼ = f (V )/f (f −1(U)) ∼
= V /f −1 (U)
is simple, and therefore f −1 (U) is a maximal submodule of V and contains rad(V ).
So we proved that f (rad(V )) ⊆ f (f −1 (U)) for all maximal submodules U of W .
Since rad(V ) ⊆ f −1 (U) for all such U, we get
f (rad(V )) ⊆ f f −1 (rad(W )) ⊆ rad(W ).

Lemma 7.10. If U is a submodule of V , then (U + rad(V ))/U ⊆ rad(V /U).
Proof. Exercise.
In Lemma 7.10 there is normally no equality: Let V = (K[T ], T ·) and U = hT 2i =

(T 2 ). We have rad(V ) = 0, but rad(V /U) = (T )/(T 2) 6= 0.
Lemma 7.11. If Vi with i ∈ I are modules, then
!
M M
rad Vi = rad(Vi ).
i∈I i∈I
L
Proof. Let V = i∈I Vi , and let πi : V → Vi be the Lcanonical projections. We
have πi (rad(V )) ⊆ rad(Vi ), and therefore rad(V ) ⊆ i∈I rad(Vi ). Vice versa, let
U be a maximal submodule of V . Let Ui be the kernel of the composition Vi →
V → V /U of the obvious canonical homomorphisms. We get that either Ui is a
maximal submodule of Vi or Ui = Vi . In both cases we get rad(Vi ) ⊆ Ui . Thus
L
L i ) ⊆ U. Since rad(V ) is the intersection of all maximal submodules of V ,
i∈I rad(V
we get i∈I rad(Vi ) ⊆ rad(V ).
7.3. Large and small submodules. Let V be a module, and let U be a submodule
of V . The module U is called large in V if U ∩ U ′ 6= 0 for all non-zero submodules
U ′ of V . The module U is small in V if U + U ′ ⊂ V for all proper submodules U ′
of V .
Lemma 7.12. Let U1 and U2 be submodules of a module V . If U1 and U2 are large
in V , then U1 ∩ U2 is large in V . If U1 and U2 are small in V , then U1 + U2 is small
in V .
Proof. Let U1 and U2 be large submodules of V . If U is an arbitrary non-zero

submodule of V , then U2 ∩ U 6= 0, since U2 is large. But we also get U1 ∩ (U2 ∩ U) =
U1 ∩ U2 ∩ U 6= 0, since U1 is large. This implies (U1 ∩ U2 ) ∩ U 6= 0. Thus U1 ∩ U2 is
large as well.
If U1 and U2 are small submodule of V , and if U is an arbitrary submodule of V
with U1 + U2 + U = V , then U2 + U = V , since U1 is small. But this implies U = V ,
since U2 is small as well.
Lemma 7.13. For 1 ≤ i ≤ n let Ui be a submodule of a module Vi . Set U =
U1 ⊕ · · · ⊕ Un and V = V1 ⊕ · · · ⊕ Vn . Then the following hold:
• U is large in V if and only if Ui is large in Vi for all i;

• U is small in V if and only if Ui is small in Vi for all i.
Proof. Let U be large in V . For some j let Wj 6= 0 be a submodule of Vj . Now we

consider Wj as a submodule of V . Since U is large in V , we get that Wj ∩ U 6= 0.
But we have
Wj ∩ U = (Wj ∩ Vj ) ∩ U = Wj ∩ (Vj ∩ U) = Wj ∩ Uj .
This shows that Wj ∩ Uj 6= 0. So we get that Uj is large in Wj .
To show the converse, it is enough to consider the case n = 2. Let Ui be large in
Vi for i = 1, 2. Set V = V1 ⊕ V2 . We first show that U1 ⊕ V2 is large in V : Let
W 6= 0 be a submodule of V . If W ⊆ V2 , then 0 6= W ⊆ U1 ⊕ V2 . If W 6⊆ V2 , then
V2 ⊂ W + V2 and therefore V1 ∩ (W + V2 ) 6= 0. This is a submodule of V1 , thus
U1 ∩ V1 ∩ (W + V2 ) 6= 0 because U1 is large in V1 . Since U1 ∩ (W + V2 ) 6= 0, there
exists a non-zero element u1 ∈ U1 with u1 = w + v2 where w ∈ W and v2 ∈ V2 . This
implies w = u1 − v2 ∈ W ∩ (U1 ⊕ V2 ). Since 0 6= u1 ∈ V1 and v2 ∈ V2 we get w 6= 0.
Thus we have shown that U1 ⊕ V2 is large in V . In the same way one shows that
V1 ⊕ U2 is large in V . The intersection of these two modules is U1 ⊕ U2 . But the
intersection of two large modules is again large. Thus U1 ⊕ U2 is large in V .
Next, assume that U is small in V . For some j let Wj be a submodule of Vj with
Uj + Wj = Vj . Set
M
W := Wj ⊕ Vi .
i6=j
This is a submodule of V with U + W = V . Since U is small in V , we get W = V ,
and therefore Wj = Vj .
To show the converse, it is enough to consider the case n = 2. For i = 1, 2 let Ui be
small in Vi , and set V = V1 ⊕ V2 . We show that U1 = U1 ⊕ 0 is small in V : Let W
be a submodule of V with U1 + W = V . Since U1 ⊆ V1 we get
U1 + (W ∩ V1 ) = (U1 + W ) ∩ V1 = V ∩ V1 = V1 .
Now U1 is small in V1 , which implies W ∩ V1 = V1 . Therefore V1 ⊆ W . In particular,
U1 ⊆ W and W = U1 + W = V . In the same way one shows that U2 = 0 ⊕ U2 is
small in V . Since the sum of two small modules is again small, we conclude that
U1 ⊕ U2 is small in V .
Let V be a module, and let U be a submodule of V . A submodule U ′ of V is called

a maximal complement of U in V if the following hold:
• U ∩ U ′ = 0;
• If U ′′ is a submodule with U ′ ⊂ U ′′ , then U ∩ U ′′ 6= 0.
If U ′ is a maximal complement of U, then U + U ′ = U ⊕ U ′ .

Lemma 7.14. Let V be a module. Every submodule U of V has a maximal comple-
ment. If U ′ is a maximal complement of U, then U ⊕ U ′ is large in V .
Proof. We show the existence by using Zorn’s Lemma: Let U be a submodule of V ,

and let W be the set of all submodules W of V with U ∩ W = 0. Clearly, this set
is non-empty, and if Wi , i ∈ I form a chain in W, then also
[
Wi
i∈I
is in W. Thus W contains maximal elements. But if U ′ is maximal in W, then U ′

is a maximal complement of U.
If U ′ is a maximal complement of U in V , and if W is a submodule of V with

(U ⊕ U ′ ) ∩ W = 0, then U + U ′ + W = U ⊕ U ′ ⊕ W . Thus U ∩ (U ′ ⊕ W ) = 0. The
maximality of U ′ yields U ′ ⊕ W = U ′ . This implies W = 0. It follows that U ⊕ U ′
is large in V .
Recall that a module V is called uniform if U1 ∩ U2 6= 0 for all non-zero submodules

U1 and U2 of V . It is easy to show that V is uniform if and only if all non-zero
submodules of V are large. It follows that a module V is uniform and has a simple
socle if and only if V contains a large simple submodule.
Lemma 7.15. Let U 6= 0 be a cyclic submodule of a module V , and let W be a
submodule of V with U 6⊆ W . Then there exists a submodule W ′ of V with W ⊆ W ′
such that U 6⊆ W ′ and W ′ is maximal with these properties. Furthermore, for each
such W ′ , the module V /W ′ is uniform and has a simple socle, and we have
soc(V /W ′ ) = (U + W ′ )/W ′.
Proof. Assume U is generated by x. Let V be the set of all submodules V ′ of V with

W ⊆ V ′ and x ∈
/ V ′.
Since W belongs to V, S we known that V is non-empty. If Vi , i ∈SI is a chain of
submodules in V, then i∈I Vi also belongs to V. (For each y ∈ i∈I Vi we have
y ∈ Vi for some i.) Now Zorn’s Lemma yields a maximal element in V.
Let W ′ be maximal in V. Thus we have W ⊆ W ′ , x ∈ / W ′ and U 6⊆ W ′ . If now
W ′′ is a submodule of V with W ′ ⊂ W ′′ , then W ′′ does not belong to V. Therefore
x ∈ W ′′ and also U ⊆ W ′′ .
Since W ′ ⊂ U + W ′ , we know that (U + W ′ )/W ′ 6= 0. Every non-zero submodule of

V /W ′ is of the form W ′′ /W ′ for some submodule W ′′ of V with W ′ ⊂ W ′′ . It follows
that U ⊆ W ′′ and that (U + W ′ )/W ′ ⊆ W ′′ /W ′. This shows that (U + W ′ )/W ′ is
simple. We also get that (U + W ′ )/W ′ is large in V /W ′ . This implies soc(V /W ′) =
(U + W ′ )/W ′ .
Corollary 7.16. Let U 6= 0 be a cyclic submodule of a module V , and let W be a
submodule of V with U 6⊆ W . If U + W = V , then there exists a maximal submodule
W ′ of V with W ⊆ W ′ .
Proof. Let W ′ be a submodule of V with W ⊆ W ′ and U 6⊆ W ′ such that W ′ is

maximal with these properties. Assume U + W = V . This implies U + W ′ = V . By
Lemma 7.15 we know that

V /W ′ = (U + W ′ )/W ′ = soc(V /W ′)
is simple. Thus W ′ is a maximal submodule of V .
Corollary 7.17. For a finitely generated module V the following hold:
(i) rad(V ) is small in V ;

(ii) If V 6= 0, then rad(V ) ⊂ V ;
(iii) If V 6= 0, then V has maximal submodules.
Proof. Clearly, (i) implies (ii) and (iii). Let us prove (i): Assume V is a finitely
generated module, and let x1 , . . . , xn be a generating set of V . Furthermore, let W
be a proper submodule of V . We show that W + rad(V ) is a proper submodule: For
0 ≤ t ≤ n let Wt be the submodule of V which is generated by W and the elements
x1 , . . . , xt . Thus we obtain a chain of submodules
W = W0 ⊆ W1 ⊆ · · · ⊆ Wn = V.
Since W ⊂ V , there exists some t with Wt−1 ⊂ Wt = V . Let U be the (cyclic)
submodule generated by xt . We get
U + Wt−1 = Wt = V,
and U 6⊆ Wt−1 . By Corollary 7.16 this implies that there exists a maximal submodule
W ′ of V with Wt−1 ⊆ W ′ . Since W ′ is a maximal submodule of V , we get rad(V ) ⊆
W ′ . Thus
W + rad(V ) ⊆ W + W ′ = W ′ ⊂ V.
This shows that rad(V ) is small in V .
Note that a corresponding statement for the socle of a module is in general wrong:
For example, the 1-module V = (K[T ], T ·) is finitely generated, and we have
soc(V ) = 0. So the socle is not large in V in this case.
Corollary 7.18. Every proper submodule of a finitely generated module V is con-
tained in a maximal submodule of V .
Proof. This follows from the proof of Corollary 7.17.

Proposition 7.19. Let V be a module. The intersection of all large submodules of
V is equal to soc(V ).
Proof. Let U0 be the intersection of all large submodules of V . We want to show that
soc(V ) is contained in every large submodule of V . This implies then soc(V ) ⊆ U0 .
Let U be a large submodule of V . Assume soc(V ) is not contained in U. Then
U ∩ soc(V ) is a proper submodule of soc(V ). Since soc(V ) is generated by simple
submodules, there exists a simple submodule S of V which is not contained in U.
Now S is simple and therefore U ∩ S = 0. Since S 6= 0, this is a contradiction. This
implies soc(V ) ⊆ U0 .
Vice versa, we claim that U0 is semisimple: Let W be a submodule of U0 . We have to

show that W is a direct summand of U0 . Let W ′ be a maximal complement of W in
V . Since W ∩W ′ = 0, we get W ∩(W ′ ∩U0 ) = 0. It follows that W +(W ′ ∩U0 ) = U0 :
Since W + W ′ is large in V , we have U0 ⊆ W + W ′ . Thus
W + (W ′ ∩ U0 ) = (W + W ′ ) ∩ U0 = U0 .
Here we used modularity. Summarizing, we see that W ′ ∩ U0 is a direct complement
of W in U0 . Thus W is a direct summand of U0 . This shows that U0 is semisimple,
which implies U0 ⊆ soc(V ).
Proposition 7.20. Let V be a module. The sum of all small submodules of V is
equal to rad(V ). A cyclic submodule U of V is small in V if and only if U ⊆ rad(V ).
Proof. Let W be a maximal submodule of V . If U is a small submodule of V , we

get U ⊆ W . (Otherwise W ⊂ U + W = V by the maximality of W , and therefore
W = V since U is small in V .) Thus every small submodule of V is contained in
rad(V ). The same is true, if there are no maximal submodules in V , since in this
case we have rad(V ) = V .
Let U be a cyclic submodule contained in rad(V ). We want to show that U is small

in V . Let U ′ be a proper submodule of V . Assume that U + U ′ = V . Since U ′ is
a proper submodule, U cannot be a submodule of U ′ . Thus there exists a maximal
submodule W ′ with U ′ ⊆ W ′ . Since U + U ′ = V , we obtain U + W ′ = V . In
particular, U is not contained in W ′ . But U lies in the radical of V , and is therefore
a submodule of any maximal submodule of V , a contradiction. This proves that
U + U ′ ⊂ V , thus U is small in V .
Let U0 be the sum of all small submodule of V . We have shown already that
U0 ⊆ rad(V ). Vive versa, we show that rad(V ) ⊆ U0 : Let x ∈ rad(V ). The cyclic
submodule U(x) generated by x is small, thus it is contained in U0 . In particular,
x ∈ U0 . Thus we proved that U0 = rad(V ).
7.4. Exercises. 1: Show: If the submodules of a finite-dimensional module V form

a chain (i.e. if for all submodules U1 and U2 of V we have U1 ⊆ U2 or U2 ⊆ U1 ),
then U is cyclic.
d
2: Assume char(K) = 0. Show: The submodules of the 1-module (K[T ], dT ) form
d
a chain, but (K[T ], dT ) is not cyclic.
3: For λ ∈ K let J(λ, n) be the Jordan block of size n × n with eigenvalue λ. For
λ1 6= λ2 in K, show that the 1-module (K n , J(λ1 , n)) ⊕ (K m , J(λ2 , m)) is cyclic.
4: Classify the small submodules of (K[T ], T ·) and N(∞).
Are (K[T ], T ·) and N(∞) uniform modules?

5: Find an example of a module V and a submodule U of V such that U is large
and small in V .
6: When is 0 large (resp. small) in a module V ?
***********************************************************************
Part 2. Modules of finite length
8. Filtrations of modules
8.1. Schreier’s Theorem. Let V be a module and let U0 , . . . , Us be submodules

of V such that
0 = U0 ⊆ U1 ⊆ · · · ⊆ Us = V.
This is called a filtration of V with factors Ui /Ui−1 . The length of this filtration
is
|{1 ≤ i ≤ s | Ui /Ui−1 6= 0}|.
A filtration
0 = U0′ ⊆ U1′ ⊆ · · · ⊆ Ut′ = V
is a refinement of the filtration above if
{Ui | 0 ≤ i ≤ s} ⊆ {Uj′ | 0 ≤ j ≤ t}.
Two filtrations U0 ⊆ U1 ⊆ · · · ⊆ Us and V0 ⊆ V1 ⊆ · · · ⊆ Vt of V are called

isomorphic if s = t and there exists a bijection π : [0, s] → [0, t] (where for integers
i and j we write [i, j] = {k ∈ Z | i ≤ k ≤ j}) such that
Ui /Ui−1 ∼
= Vπ(i) /Vπ(i)−1
for 1 ≤ i ≤ s.
Theorem 8.1 (Schreier). Any two given filtrations of a module V have isomorphic
refinements.
Before we prove this theorem, we need the following lemma:

Lemma 8.2 (Butterfly Lemma). Let U1 ⊆ U2 and V1 ⊆ V2 be submodules of a
module V . Then we have
(U1 + V2 ∩ U2 )/(U1 + V1 ∩ U2 ) ∼
= (U2 ∩ V2 )/((U1 ∩ V2 ) + (U2 ∩ V1 ))
∼
= (V1 + U2 ∩ V2 )/(V1 + U1 ∩ V2 ).
The name “butterfly” comes from the picture
U1 + V2 ∩ UT2 V1 + U2 ∩ V2
TTTT jjj
TTT
TTTT jjjjjjj
TTT j
jjjj
U1 + V1 ∩ UT2 U2 ∩ V2 V1 + U1 ∩ V2
TTTT kkk
TTTT
TTTT kkkkkkk
k
TT kkkk
(U1 ∩ V2 ) + (U2 ∩ V1 )
which occurs as the part marked with ⋆ of the picture
◦A
} AA
}} AA
}}} AA
}} A
◦ ?? ◦@
~~ ?? @@
~~~ ?? @@
@@
~~ ??

U2 @ ◦ ?? V2
@@ ?? ~~
@@ ?? ~~
@@ ?? ~
~~
⋆ AA } ◦ AA } ⋆
AA }} AA }}
}A A}
}} AA }} AA
}} A }} A
⋆ ?? ⋆ ⋆@
~~ ?? @@
~~~ ?? @@
@@
~~ ??

U1 @ ⋆? V1
@@ ?? ~
@@ ??
?? ~~~
@@ ? ~
~~
◦A ◦
AA }}
AA }}
AA }}
A }}
◦
Note that V1 + U2 ∩ V2 = (V1 + U2 ) ∩ V2 = V1 + (U2 ∩ V2 ), since V1 ⊆ V2 . But we

have (V2 + U2 ) ∩ V1 = V1 and V2 + (U2 ∩ V1 ) = V2 . Thus the expression V2 + U2 ∩ V1
would not make any sense.
Proof of Lemma 8.2. Note that U1 ∩ V2 ⊆ U2 ∩ V2 . Recall that for submodules U

and U ′ of a module V we always have
U/(U ∩ U ′ ) ∼
= (U + U ′ )/U ′ .
U + UH′
ww HH
ww HH
ww HH
w HH
ww
U GG v
U′
GG v
GG vv
GG vvv
G vv
U ∩ U′
Since U1 ⊆ U2 and V1 ⊆ V2 we get

(U1 + V1 ∩ U2 ) + (U2 ∩ V2 ) = U1 + (V1 ∩ U2 ) + (U2 ∩ V2 )
= U1 + (U2 ∩ V2 )
= U1 + V2 ∩ U2
and
(U1 + V1 ∩ U2 ) ∩ (U2 ∩ V2 ) = (U1 + V1 ) ∩ U2 ∩ (U2 ∩ V2 )
= (U1 + V1 ∩ U2 ) ∩ V2
= ((V1 ∩ U2 ) + U1 ) ∩ V2
= (V1 ∩ U2 ) + (U1 ∩ V2 ).
The result follows.
Proof of Theorem 8.1. Assume we have two filtrations

0 = U0 ⊆ U1 ⊆ · · · ⊆ Us = V
and
0 = V0 ⊆ V1 ⊆ · · · ⊆ Vt = V
of a module V . For 1 ≤ i ≤ s and 0 ≤ j ≤ t define
Uij = Ui−1 + Vj ∩ Ui .
Thus we obtain
0 = U10 ⊆ U11 ⊆ · · · ⊆ U1t = U1
U1 = U20 ⊆ U21 ⊆ · · · ⊆ U2t = U2
.. ..
. .
Us−1 = Us0 ⊆ Us1 ⊆ · · · ⊆ Ust = Us = V.
Similarly, set
Vji = Vj−1 + Ui ∩ Vj .
This yields
0 = V10 ⊆ V11 ⊆ · · · ⊆ V1s = V1
V1 = V20 ⊆ V21 ⊆ · · · ⊆ V2s = V2
.. ..
. .
Vt−1 = Vt0 ⊆ Vt1 ⊆ · · · ⊆ Vts = Vt = V.
For 1 ≤ i ≤ s and 1 ≤ j ≤ t define
Fij = Uij /Ui,j−1 and Gji = Vji /Vj,i−1.
The filtration (Uij )ij is a refinement of the filtration (Ui )i and its factors are the
modules Fij . Similarly, the filtration (Vji)ji is a refinement of (Vj )j and has factors
Gji. Now the Butterfly Lemma 8.2 implies Fij ∼

= Gji , namely
Fij = Uij /Ui,j−1
= (Ui−1 + Vj ∩ Ui )/(Ui−1 + Vj−1 ∩ Ui )
∼
= (Vj−1 + Ui ∩ Vj )/(Vj−1 + Ui−1 ∩ Vj )
= Gji .
This finishes the proof.
A filtration
0 = U0 ⊆ U1 ⊆ · · · ⊆ Us = V.
of a module V with all factors Ui /Ui−1 (1 ≤ i ≤ s) being simple is called a com-
position series of V . In this case we call s (i.e. the number of simple factors) the
length of the composition series. We call the Ui /Ui−1 the composition factors of
V.
8.2. The Jordan-Hölder Theorem. As an important corollary of Theorem 8.1

we obtain the following:
Corollary 8.3 (Jordan-Hölder Theorem). Assume that a module V has a composi-
tion series of length l. Then the following hold:
• Any filtration of V has length at most l and can be refined to a composition

series;
• All composition series of V have length l.
Proof. Let
0 = U0 ⊂ U1 ⊂ · · · ⊂ Ul = V
be a composition series, and let
0 = V0 ⊆ V1 ⊆ · · · ⊆ Vt = V
be a filtration. By Schreier’s Theorem 8.1 there exist isomorphic refinements of these
filtrations. Let Fi = Ui /Ui−1 be the factors of the filtration (Ui )i . Thus Fi is simple.
If (Ui′ )i is a refinement of (Ui )i , then its factors are F1 , . . . , Fl together with some
0-modules. The corresponding refinement of (Vj )j has exactly l + 1 submodules.
Thus (Vj )j has at most l different non-zero factors. In particular, if (Vj )j is already
a composition series, then t = l.
If V has a composition series of length l, then we say V has length l, and we write
l(V ) = l. Otherwise, V has infinite length and we write l(V ) = ∞.
Assume l(V ) < ∞ and let S be a simple module. Then [V : S] is the number
of composition factors in a (and thus in all) composition series of V which are
isomorphic to S. One calls [V : S] the Jordan-Hölder multiplicity of S in V .
Let l(V ) < ∞. Then ([V : S])S∈S is called the dimension vector of V , where S
is a complete set of representatives of isomorphism classes of the simple modules.
Note that only finitely many entries of the dimension vector are non-zero. We get
X
[V : S] = l(V ).
S∈S
Example: If J = ∅, then a J-module is just given by a vector space V . In this

case l(V ) = dim V if V is finite-dimensional. It also follows that V is simple if and
only if dim V = 1. If V is infinite-dimensional, then dim V is a cardinality and we
usually write l(V ) = ∞.
For modules of finite length, the Jordan-Hölder multiplicities and the length are
important invariants.
8.3. Exercises. 1: Let V be a module of finite length, and let V1 , . . . , Vt be sub-

modules of V . Show: If !
X t
l Vi = l(V ),
i=1
Lt
then V = i=1 Vi .
2: Let V1 and V2 be modules, and let S be a factor of a filtration of V1 ⊕ V2 . Show:

If S is simple, then there exists a filtration of V1 or of V2 which contains a factor
isomorphic to S.
3: Construct indecomposable modules V1 and V2 with l(V1 ) = l(V2 ) = 2, and a
filtration of V1 ⊕ V2 containing a factor T of length 2 such that T is not isomorphic
to V1 or V2 .
4: Determine all composition series of the 2-module V = (K 5 , φ, ψ) where
   
c0 0 0 1 0 1
 c0   0 0 1 1
   
φ=  c1  and ψ = 
  0 

 c2   0 
c3 0
with pairwise different elements c0 , c1 , c2 , c3 in K.
————————————————————————————-
9. Digression: Local rings
We need some basic notations from ring theory. This might seem a bit boring but
will be of great use later on.
9.1. Local rings. Let R be a ring. Then r ∈ R is right-invertible if there exists

some r ′ ∈ R such that rr ′ = 1, and r is left-invertible if there exists some r ′′ ∈ R
such that r ′′ r = 1. We call r ′ a right inverse and r ′′ a left inverse of r. If r is
both right- and left-invertible, then r is invertible.
Example: Let V be a vector space with basis {ei | i ≥ 1}. Define a linear map
f : V → V by f (ei ) = ei+1 for all i, and a linear map g : V → V by g(e1 ) = 0 and
g(ei) = ei−1 for all i ≥ 2. Then we have gf = 1V , thus f is left-invertible and g is
right-invertible. Note also that f g 6= 1V , since for example f g(e1) = 0.
Lemma 9.1. If r ′ is a right inverse and r ′′ a left inverse of an element r, then
r ′ = r ′′ . In particular, there is only one right inverse and only one left inverse.
Proof. We have r ′ = 1r ′ = r ′′ rr ′ = r ′′ 1 = r ′′ .
Lemma 9.2. Assume that r is right-invertible. Then the following are equivalent:
• r is left-invertible;
• There exists only one right inverse of r.
Proof. Assume that r is right-invertible, but not left-invertible. Then rr ′ = 1 and

r ′ r 6= 1 for some r ′ . This implies
r(r ′ + r ′ r − 1) = rr ′ + rr ′r − r = 1.
But r ′ + r ′ r − 1 6= r ′ .
An element r in a ring R is nilpotent if r n = 0 for some n ≥ 1.

Lemma 9.3. Let r be a nilpotent element in a ring R, then 1 − r is invertible.
Proof. We have (1 − r)(1 + r + r 2 + r 3 + · · · ) = 1. (Note that this sum is finite, since

r is nilpotent.) One also easily checks that (1 + r + r 2 + r 3 + · · · )(1 − r) = 1. Thus
(1 − r) is right-invertible and left-invertible and therefore invertible.
A ring R is local if the following hold:
• 1 6= 0;
• If r ∈ R, then r or 1 − r is invertible.
(Recall that the only ring with 1 = 0 is the 0-ring, which contains just one element.
Note that we do not exclude that for some elements r ∈ R both r and 1 − r are
invertible.)
Local rings occur in many different contexts. For example, they are important in
Algebraic Geometry: One studies the local ring associated to a point x of a curve
(or more generally of a variety or a scheme) and hopes to get a “local description”,
i.e. a description of the curve in a small neighbourhood of the point x.
Examples: K[T ] is not local (T is not invertible, and 1 − T is also not invertible),
Z is not local, every field is a local ring.
Let U be an additive subgroup of a ring R. Then U is a right ideal of R if for all

u ∈ U and all r ∈ R we have ur ∈ U, and U is a left ideal if for all u ∈ U and all
r ∈ R we have ru ∈ U. One calls U an ideal if it is a right and a left ideal.
If I and J are ideals of a ring R, then the product
P IJ is the additive subgroup of
R generated by all (finite) sums of the form s is js where is ∈ I and js ∈ J. It
is easy to check that IJ is again an ideal. For n ≥ 0, define I 0 = R, I 1 = I and
I n+2 = I(I n+1 ) = (I n+1 )I.
A left ideal U is a maximal left ideal if it is maximal in the set of all proper left
ideals, i.e. if U ⊂ R and for every left ideal U ′ with U ⊆ U ′ ⊂ R we have U = U ′ .
Similarly, define a maximal right ideal.
Recall that an element e ∈ R is an idempotent if e2 = e.
Lemma 9.4. Let e ∈ R be an idempotent. If e is left-invertible or right-invertible,
then e = 1
Proof. If e is left-invertible, then re = 1 for some r ∈ R. Also e = 1e = (re)e =

re = 1. The other case is done similarly.
Lemma 9.5. Assume that R is a ring which has only 0 and 1 as idempotents. Then
all left-invertible and all right-invertible elements are invertible.
Proof. Let r be left-invertible, say r ′ r = 1. Then rr ′ is an idempotent, which by

our assumption is either 0 or 1. If rr ′ = 1, then r is right-invertible and therefore
invertible. If rr ′ = 0, then 1 = r ′ r = r ′ rr ′ r = 0, thus R = 0. The only element
0 = 1 in R = 0 is invertible. The other case is done similarly.
Proposition 9.6. The following properties of a ring R are equivalent:
(i) We have 0 6= 1, and if r ∈ R, then r or 1 − r is invertible (i.e. R is a local

ring);
(ii) There exist non-invertible elements in R, and the set of these elements is
closed under +;
(iii) The set of non-invertible elements in R form an ideal;
(iv) R contains a proper left ideal, which contains all proper left ideals;
(v) R contains a proper right ideal which contains all proper right ideals.
Remark: Property (iv) implies that R contains exactly one maximal left ideal. Using
the Axiom of Choice, the converse is also true.
Proof. We first show that under the assumptions (i), (ii) and (iv) the elements 0
and 1 are the only idempotents in R, and therefore every left-invertible element and
every right-invertible element will be invertible.
Let e ∈ R be an idempotent. Then 1 − e is also an idempotent. It is enough to show

that e or 1 − e are invertible: If e is invertible, then e = 1. If 1 − e is invertible, then
1 − e = 1 and therefore e = 0.
Under (i) we assume that either e or 1 − e are invertible, and we are done. Also
under (ii) we know that e or 1 −e is invertible: If e and 1 −e are both non-invertible,
then 1 = e + (1 − e) is non-invertible, a contradiction. Finally, assume that under
(iv) we have a proper left ideal I containing all proper left ideals, and assume that
e and 1 − e are both non-invertible. We claim that both elements and therefore
also their sum have to belong to I, a contradiction, since 1 cannot be in I. Why
does e ∈ I hold? Since e is non-invertible, we know that e is not left-invertible.
Therefore Re is a proper left ideal, which must be contained in I. Since 1 − e is also
non-invertible, we know that 1 − e ∈ I.
(i) =⇒ (ii): 0 6= 1 implies that 0 is not invertible. Assume r1 , r2 are not invertible.
Assume also that r1 + r2 is invertible. Thus x(r1 + r2 ) = 1 for some x ∈ R. We get
xr1 = 1 − xr2 . Now (i) implies that xr2 or 1 − xr2 is invertible. Without loss of
generality let xr1 be invertible. Thus there exists some y such that 1 = yxr1 . This
implies that r1 is left-invertible and therefore invertible, a contradiction.
(ii) =⇒ (i): The existence of non-invertible elements implies R 6= 0, and therefore
we have 0 6= 1. Let r ∈ R. If r and 1 − r are non-invertible, then by (ii) we get that
1 = r + (1 − r) is non-invertible, a contradiction.
(ii) =⇒ (iii): Let I be the set of non-invertible elements in R. Then by (ii) we
know that I is a subgroup of (R, +). Given x ∈ I and r ∈ R we have to show that
rx ∈ I and xr ∈ I. Assume rx is invertible. Then there is some y with yrx = 1,
thus x is left-invertible and therefore x is invertible, a contradiction. Thus rx ∈ I.
Similarly, we can show that xr ∈ I.
(iii) =⇒ (iv): Let I be the set of non-invertible elements in R. By (iii) we get
that I is an ideal and therefore a left ideal. Since 1 ∈
/ I we get I ⊂ R. Let U ⊂ R
be a proper left ideal. Claim: U ⊆ I. Let x ∈ U, and assume x ∈ / I. Thus x is
invertible. So there is some y ∈ R such that yx = 1. Then for r ∈ R we have
r = r1 = (ry)x ∈ U. Thus R ⊆ U which implies U = R, a contradiction. Similarly
we prove (iii) =⇒ (v).
(iv) =⇒ (i): Let I ′ be a proper left ideal of R that contains all proper left ideals.
We show that all elements in R \ I ′ are invertible: Let r ∈ / I ′ . Then Rr is a left
ideal of R which is not contained in I ′ , thus we get Rr = R. So there is some r ′ ∈ R
such that r ′ r = 1, in other words, r is left-invertible and therefore invertible. Now
let r ∈ R be arbitrary. We claim that r or 1 − r belong to R \ I ′ : If both elements
belong to I ′ , then so does 1 = r + (1 − r), a contradiction. Thus either r or 1 − r is
invertible. Similarly we prove (v) =⇒ (i).
If R is a local ring, then
I := {r ∈ R | r non-invertible }
is called the radical (or Jacobson radical) of R.

Corollary 9.7. The Jacobson radical I of a local ring R is the only maximal left
ideal and also the only maximal right ideal of R. It contains all proper left and all
proper right ideals of R.
Proof. Part (iii) of the above proposition tells us that I is indeed an ideal in R.
Assume now I ⊂ I ′ ⊆ R with I ′ a left (resp. right) ideal. Take r ∈ I ′ \ I. Then r is
invertible. Thus there exists some r ′ such that r ′ r = rr ′ = 1. This implies I ′ = R.
So we proved that I ′ is a maximal left and also a maximal right ideal. Furthermore,
the proof of (iii) =⇒ (iv) in the above proposition shows that I contains all proper
left ideals, and similarly one shows that I contains all proper right ideals of R.
If I is the Jacobson radical of a local ring R, then the radical factor ring R/I is
a ring without left ideals different from 0 and R/I. It is easy to check that R/I is
a skew field. (For r ∈ R/I with r 6= 0 and r ∈ R \ I there is some s ∈ R such that
sr = 1 = rs. In R/I we have s · r = sr = 1 = rs = r · s.)
Example: For c ∈ K set
R = {f /g | f, g ∈ K[T ], g(c) 6= 0},
m = {f /g ∈ R | f (c) = 0, g(c) 6= 0}.
Then m is an ideal in the ring R. In fact, m = (T − c)R: One inclusion is obtained
since
f (T − c)f
(T − c) = ,
g g
and the other inclusion follows since f (c) = 0 implies f = (T −c)h for some h ∈ K[T ]
and therefore
f h
= (T − c) .
g g
If r ∈ R \ m, then r = f /g with f (c) 6= 0 and g(c) 6= 0. Thus r −1 = g/f ∈ R is an

inverse of r.
If r = f /g ∈ m, then r is not invertible: For any f ′ /g ′ ∈ R, the product f /g · f ′/g ′ =
f f ′/gg ′ always lies in m, since (f f ′ )(c) = f (c)f ′ (c) = 0 and thus it cannot be the
identity and therefore r is not invertible.
Thus we proved that R \ m is exactly the set of invertible elements in R, and the set
m of non-invertible elements forms an ideal. So by the above theorem, R is a local
ring.
9.2. Exercises. 1: A module V is called local if it contains a maximal submodule

U, which contains all proper submodules of V . Show: If V is local, then V con-
tains exactly one maximal submodule. Construct an example which shows that the
converse is not true.
2: Show: Every module of finite length is a sum of local submodules.
3: Let V be a module of length n. Show: V is semisimple if and only if V cannot

be written as a sum of n − 1 local submodules.
4: Let R = K[X, Y ] be the polynomial ring in two (commuting) variables. Show:
• R is not local;
• 0 and 1 are the only idempotents in R;
• The Jacobson radical of R is 0.
————————————————————————————-
10. Modules of finite length
10.1. Some length formulas for modules of finite length.

Lemma 10.1. Let U be a submodule of a module V and let W = V /U be the
corresponding factor module. Then V has finite length if and only if U and W have
finite length. In this case, we get
l(V ) = l(U) + l(W )
and for every simple module S we have
[V : S] = [U : S] + [W : S].
Proof. Assume that V has length n. Thus every chain of submodules of V has length
at most n. In particular this is true for all chains of submodules of U. This implies
l(U) ≤ n. The same holds for chains of submodules which all contain U. Such
chains correspond under the projection homomorphism V → V /U to the chains of
submodules of V /U = W . Thus we get l(W ) ≤ n. So if V has finite length, then so
do U and W .
Vice versa, assume that U and W = V /U have finite length. Let
0 = U0 ⊂ U1 ⊂ · · · ⊂ Us = U and 0 = W0 ⊂ W1 ⊂ · · · ⊂ Wt = W
be composition series of U and W , respectively. We can write Wj in the form
Wj = Vj /U for some submodule U ⊆ Vj ⊆ V . We obtain a chain
0 = U0 ⊂ U1 ⊂ · · · ⊂ Us = U = V0 ⊂ V1 ⊂ · · · ⊂ Vt = V
of submodules of V such that
Vj /Vj−1 ∼
= Wj /Wj−1
for all 1 ≤ j ≤ t. This chain is a composition series of V , since the factors Ui /Ui−1
with 1 ≤ i ≤ s, and the factors Vj /Vj−1 with 1 ≤ j ≤ t are simple. If S is simple,
then the number of composition factors of V which are isomorphic to S is equal
to the number of indices i with Ui /Ui−1 ∼ = S plus the number of indices j with
Vj /Vj−1 ∼= S. In other words, [V : S] = [U : S] + [W : S].
Corollary 10.2. Let V be a module of finite length. If 0 = U0 ⊆ U1 ⊆ · · · ⊆ Ut = V

is a filtration of V , then
X t
l(V ) = l(Ui /Ui−1 ).
i=1
Corollary 10.3. Let U1 and U2 be modules of finite length. Then

l(U1 ⊕ U2 ) = l(U1 ) + l(U2 ).
Corollary 10.4. Let V be a module of finite length, and let U1 and U2 be submodules
of V . Then
l(U1 ) + l(U2 ) = l(U1 + U2 ) + l(U1 ∩ U2 ).
Proof. Set U ′ := U1 ∩ U2 and U ′′ := U1 + U2 .

U ′′ B
|| BB
|| BB
|| BB
|| B
U1 B U2
BB ||
BB ||
BB ||
B ||
U′
Then
U ′′ /U ′ ∼
= (U1 /U ′ ) ⊕ (U2 /U ′ ).
Thus
l(U1 ) = l(U ′ ) + l(U1 /U ′ ),
l(U2 ) = l(U ′ ) + l(U2 /U ′ ),
l(U ′′ ) = l(U ′ ) + l(U1 /U ′ ) + l(U2 /U ′ ).
This yields the result.
Corollary 10.5. Let V and W be modules and let f : V → W be a homomorphism.
If V has finite length, then
l(V ) = l(Ker(f )) + l(Im(f )).
If W has finite length, then
l(W ) = l(Im(f )) + l(Cok(f )).
= Im(f ) and W/ Im(f ) ∼

Proof. Use the isomorphisms V / Ker(f ) ∼ = Cok(f ).
Recall that for every homomorphism f : V → W there are short exact sequences
0 → Ker(f ) → V → Im(f ) → 0
and
0 → Im(f ) → W → Cok(f ) → 0.
Corollary 10.6. For every short exact sequence
0→U →V →W →0
of modules with l(V ) < ∞, we have l(V ) = l(U) + l(W ).
Corollary 10.7. Let V be a module of finite length, and let f : V → V be an

endomorphism of V . Then the following statements are equivalent:
(i) f is injective;
(ii) f is surjective;
(iii) f is an isomorphism;
(iv) l(Im(f )) = l(V ).
Lemma 10.8. If V is a module of finite length, then V is a finite direct sum of

indecomposable modules.
Proof. This is proved by induction on l(V ). The statement is trivial if V is inde-

composable. Otherwise, let V = V1 ⊕ V2 with V1 and V2 two non-zero submodules.
Then proceed by induction.
Recall that in Section 7.2 we studied the radical rad(V ) of a module V . The following
lemma shows that V / rad(V ) is well behaved if V is of finite length:
Lemma 10.9. Let V be a module of finite length. Then V / rad(V ) is semisimple.
Proof. Assume that l(V / rad(V )) = n. Inductively T we look for maximal submodules
U1 , . . . , Un of V such that for 1 ≤ t ≤ n and Vt := ti=1 Ui we have
t
M
V /Vt ∼
= V /Ui
i=1
and l(V /Vt ) = t. Note that V /Ui is simple for all i.

For t = 1 there is nothing to show. If U1 , . . . , Ut are already constructed and if t < n,
then rad(V ) ⊂ Vt . Thus there exists a maximal submodule Ut+1 with Vt ∩ Ut+1 ⊂ Vt .
Since Vt 6⊆ Ut+1 , we know that Ut+1 ⊂ Vt + Ut+1 . The maximality of Ut+1 implies
that Vt + Ut+1 = V . Set Vt+1 := Vt ∩ Ut+1 .
V FF
{{ FF
{{{ FF
FF
{ F
{{
Vt B Ut+1
B
BB x
BB xx
BB xxx
xx
Vt+1
Thus we obtain
V /Vt+1 = Vt /Vt+1 ⊕ Ut+1 /Vt+1
∼
= V /Ut+1 ⊕ V /Vt
t
M
∼
= V /Ut+1 ⊕ V /Ui .
i=1
The last of these isomorphisms comes from the induction assumption.
10.2. The Fitting Lemma.

Lemma 10.10 (Fitting). Let V be a module of finite length, say l(V ) = n, and let
f ∈ End(V ). Then we have
V = Im(f n ) ⊕ Ker(f n ).
In particular, if V is indecomposable, then Im(f n ) = 0 or Ker(f n ) = 0.
Proof. We have
0 = Ker(f 0 ) ⊆ Ker(f 1 ) ⊆ Ker(f 2 ) ⊆ · · · .
(For x ∈ Ker(f i ) we get f i (x) = 0 and therefore f i+1 (x) = 0.)
Assume that Ker(f i−1 ) = Ker(f i ) for some i. It follows that Ker(f i ) = Ker(f i+1 ).
(Assume f i+1 (x) = 0. Then f i (f (x)) = 0 and therefore f (x) ∈ Ker(f i ) = Ker(f i−1 ).
This implies f i(x) = f i−1 (f (x)) = 0. Thus Ker(f i+1 ) ⊆ Ker(f i).)
If
0 = Ker(f 0 ) ⊂ Ker(f 1 ) ⊂ · · · ⊂ Ker(f i ),
then l(Ker(f i )) ≥ i. This implies i ≤ n, and therefore Ker(f m ) = Ker(f n ) for all
m ≥ n.
We have
· · · ⊆ Im(f 2 ) ⊆ Im(f ) ⊆ Im(f 0 ) = V.
(For x ∈ Im(f i ) we get x = f i (y) = f i−1 (f (y)) for some y ∈ V . Thus x ∈ Im(f i−1 ).)
Assume that Im(f i−1 ) = Im(f i ). Then Im(f i ) = Im(f i+1 ). (For every y ∈ V
there exists some z with f i−1 (y) = f i (z). This implies f i(y) = f i+1 (z). Thus
Im(f i ) ⊆ Im(f i+1 ).)
If
Im(f i ) ⊂ · · · ⊂ Im(f 1 ) ⊂ Im(f 0 ) = V,
then l(Im(f i)) ≤ n − i, which implies i ≤ n. Thus Im(f m ) = Im(f n ) for all m ≥ n.
So we proved that
Ker(f n ) = Ker(f 2n ) and Im(f n ) = Im(f 2n ).
We claim that Im(f n ) ∩Ker(f n ) = 0: Let x ∈ Im(f n ) ∩Ker(f n ). Then x = f n (y) for
some y and also f n (x) = 0, which implies f 2n (y) = 0. Thus we get y ∈ Ker(f 2n ) =
Ker(f n ) and x = f n (y) = 0.
Next, we show that Im(f n ) + Ker(f n ) = V : Let v ∈ V . Then there is some w with
f n (v) = f 2n (w). This is equivalent to f n (v −f n (w)) = 0. Thus v −f n (w) ∈ Ker(f n ).
Now v = f n (w) + (v − f n (w)). This finishes the proof.
Corollary 10.11. Let V be an indecomposable module of finite length n, and let
f ∈ End(V ). Then either f is an isomorphism, or f is nilpotent (i.e. f n = 0).
Proof. If Im(f n ) = 0, then f n = 0, in particular f is nilpotent. Now assume that

Ker(f n ) = 0. Then f n is injective, which implies that f is injective (f (x) = 0 =⇒
f n (x) = 0 =⇒ x = 0). Thus f is an isomorphism.
Combining the above with Lemma 9.3 we obtain the following important result:
Corollary 10.12. Let V be an indecomposable module of finite length. Then End(V )
is a local ring.
Let V be a module, and let R = End(V ) be the endomorphism ring of V . Assume

that V = V1 ⊕ V2 be a direct decomposition of V . Then the map e : V → V defined
by e(v1 , v2 ) = (v1 , 0) is an idempotent in End(V ). Now e = 1 if and only if V2 = 0,
and e = 0 if and only if V1 = 0.
It follows that the endomorphism ring of any decomposable module contains idem-
potent which are not 0 or 1.
Example: The 1-module V = (K[T ], T ·) is indecomposable, but its endomorphism
ring End(V ) ∼
= K[T ] is not local.
Lemma 10.13. Let V be a module. If End(V ) is a local ring, then V is indecom-
posable.
Proof. If a ring R is local, then its only idempotents are 0 and 1. Then the result
follows from the discussion above.
10.3. The Harada-Sai Lemma.

Lemma 10.14 (Harada-Sai). Let Vi be indecomposable modules of length at most n
where 1 ≤ i ≤ m = 2n , and let fi : Vi → Vi+1 where 1 ≤ i < m be homomorphisms.
If fm−1 · · · f2 f1 6= 0, then at least one of the homomorphisms fi is an isomorphism.
Proof. We show this by induction on a: Let a ≤ n. Let Vi , 1 ≤ i ≤ m = 2a be

modules of length at most n, and fi : Vi → Vi+1 , 1 ≤ i < m homomorphisms. If
l(Im(fm−1 · · · f2 f1 )) > n − a,
then at least one of the homomorphisms is an isomorphism.
If a = 1, then there is just one homomorphism, namely f1 : V1 → V2 . If l(Im(f1 )) >

n − 1, then f1 is an isomorphism. Remember that by our assumption both modules
V1 and V2 have length at most n.
Assume the statement holds for a < n. Define m = 2a . Let Vi be indecomposable

modules with 1 ≤ i ≤ 2m, and for 1 ≤ i < 2m let fi : Vi → Vi+1 be homomorphisms.
Let f = fm−1 · · · f1 , g = fm and h = f2m−1 · · · fm+1 . Thus
f g h
V1 −
→ Vm −
→ Vm+1 −
→ V2m .
Assume l(Im(hgf )) > n − (a + 1). We can assume that l(Im(f )) ≤ n − a and
l(Im(h)) ≤ n − a, otherwise we know by induction that one of the homomorphisms
fi is an isomorphism.
Since
l(Im(f )) ≥ l(Im(gf )) ≥ l(Im(hgf )) > n − (a + 1)
and
l(Im(h)) ≥ l(Im(hgf )),
it follows that l(Im(f )) = n − a = l(Im(h)) and therefore l(Im(hgf )) = n − a.
Since Im(f ) and Im(hgf ) have the same length, we get Im(f ) ∩ Ker(hg) = 0. Now
Im(f ) has length n − a, and Ker(hg) has length l(Vm ) − l(Im(hg)). This implies
l(Im(hg)) = n − a, because
l(Im(hgf )) ≤ l(Im(hg)) ≤ l(Im(h)).
So we see that Im(f ) + Ker(hg) = Vm . In this way, we obtained a direct decompo-
sition
Vm = Im(f ) ⊕ Ker(hg).
But Vm is indecomposable, and Im(f ) 6= 0. It follows that Ker(hg) = 0. In other
words, hg is injective, and so g is injective.
In a similar way, we can show that g is also surjective: Namely
Vm+1 = Im(gf ) ⊕ Ker(h) :
Since Im(gf ) and Im(hgf ) have the same length, we get
Im(gf ) ∩ Ker(h) = 0.
On the other hand, the length of Ker(h) is
l(Vm+1 ) − l(Im(h)) = l(Vm+1 ) − (n − a).
Since Vm+1 is indecomposable, Im(gf ) 6= 0 implies Vm+1 = Im(gf ). Thus gf is
surjective, which yields that g is surjective as well.
Thus we have shown that g = fm is an isomorphism.
Corollary 10.15. If V is an indecomposable module of finite length n, and if I
denotes the radical of End(V ), then I n = 0.
Proof. LetPS be a subset of End(V ), and let SV be the set of all (finite) sums of
the form i fi (vi ) with fi ∈ S and vi ∈ V . This is a submodule of V . (It follows
from the definition, that SV is closed under addition. Since all fi are linear maps,
SV is also closed under scalar multiplication. Finally, for V = (V, φj )j we have

φj (SV ) ⊆ SV , since
!
X X
φj fi (vi ) = fi (φj (vi )),
i i
because all the fi are homomorphisms.)

For i ≥ 0 we can look at the submodule I i V of V . Thus
· · · ⊆ I 2 V ⊆ IV ⊆ I 0 V = V.
If I i−1 V = I i V , then I i V = I i+1 V .
The Harada-Sai Lemma implies I m = 0 for m = 2n − 1, thus also I m V = 0. Thus
there exists some t with
0 = I t V ⊂ · · · ⊂ I 2 V ⊂ IV ⊂ I 0 V = V.
This is a filtration of the module V , and since V has length n, we conclude t ≤ n.
This implies I n V = 0 and therefore I n = 0.
References
[ED] D. Eisenbud, J.A. de la Peña, Chains of maps between indecomposable modules, Journal für
die Reine und Angewandte Mathematik 504 (1998), 29–35.
[F] H. Fitting, Über direkte Produktzerlegungen einer Gruppe in unzerlegbare, Mathematische
Zeitschrift 39 (1934), 16–30.
[HS] M. Harada, Y. Sai, On categories of indecomposable modules I, Osaka Journal of Mathematics
7 (1970), 323–344.
[Ri] C.M. Ringel, Report on the Brauer-Thrall conjecture, In: Representation Theory II, Springer
Lecture Notes in Mathematics 832 (1980), 104–135.
10.4. Exercises. 1: Find the original references for Schreier’s Theorem, the Jordan-
Hölder Theorem, the Fitting Lemma and the Harada-Sai Lemma.
2: Let V be a module with a simple submodule S, such that S is contained in every

non-zero submodule of V . Assume that every endomorphism of S occurs as the
restriction of an endomorphism of V . Show: The endomorphism ring of V is local,
and its radical factor ring is isomorphic to the endomorphism ring of S.
In particular: The endomorphism ring of N(∞) is a local ring with radical factor
ring isomorphic to the ground field K.
3: Let V = (K[T ], T ·). Show that V is indecomposable and that End(V ) is not a
local ring.
————————————————————————————-
11. Direct summands of finite direct sums
11.1. The Exchange Theorem. If the endomorphism ring of a module V is local,

then V is indecomposable. In representation theory we are often interested in the
indecomposable direct summands of a module. Then one can ask if these direct
summands are in some sense uniquely determined (at least up to isomorphism).
Lemma 11.1. For i = 1, 2 let hi : V → Yi be homomorphisms. Let Y = Y1 ⊕ Y2 and
f = t [h1 , h2 ] : V → Y.
If h1 is an isomorphism, then
Y = Im(f ) ⊕ Y2 .
Proof. For y ∈ Y , write y = y1 + y2 with y1 ∈ Y1 and y2 ∈ Y2 . Since h1 is surjective,

there is some v ∈ V with h1 (v) = y1 . We get
y = y1 + y2 = h1 (v) + y2 = h1 (v) + h2 (v) − h2 (v) + y2 = f (v) + (−h2 (v) + y2 ).
Now f (v) ∈ Im(f ) and −h2 (v) + y2 ∈ Y2 . So we proved that Im(f ) + Y2 = Y .
For y ∈ Im(f ) ∩ Y2 , there is some v ∈ V with y = f (v). Furthermore, y = f (v) =
h1 (v) + h2 (v). Since y ∈ Y2 , we get h1 (v) = y − h2 (v) ∈ Y1 ∩ Y2 = 0. Since h1 is
injective, h1 (v) = 0 implies v = 0. Thus y = f (v) = f (0) = 0.
Theorem 11.2 (Exchange Theorem). Let V, W1 , . . . , Wm be modules, and define
m
M
W = Wj .
j=1
Let f : V → W be a split monomorphism. If the endomorphism ring of V is local,

then there exists some t with 1 ≤ t ≤ m and a direct decomposition of Wt of the
form Wt = V ′ ⊕ Wt′ such that
M
W = Im(f ) ⊕ Wt′ ⊕ Wj and V ′ ∼ = V.
j6=t
If we know additionally that W = Im(f ) ⊕ W1 ⊕ C for some submodule C of W ,

then we can assume 2 ≤ t ≤ m.
Proof. Since f is a split monomorphism, there is a homomorphism g : W → V

with gf = 1V . Write f = t [f1 , . . . , fm ] and g = [g1 , . . . , gm ] with homomorphisms
fj : V → Wj and gj : Wj → V . Thus we have
m
X
gf = gj fj = 1V .
j=1
Since End(V ) is a local ring, there is some t with 1 ≤ t ≤ m such that gt ft is

invertible. Without loss of generality assume that g1 f1 is invertible.
Since g1 f1 is invertible, f1 is a split monomorphism, thus f1 : V → W1 is injective

and Im(f1 ) ⊕ W1′ = W1 for some submodule W1′ of W1 . Let h : V → Im(f1 ) be
defined by h(v) = f1 (v) for all v ∈ V . Thus we can write
f1 : V → W1 = Im(f1 ) ⊕ W1′
in the form f1 = t [h, 0]. Thus
f = t [h, 0, f2 , . . . , fm ] : V → Im(f1 ) ⊕ W1′ ⊕ W2 ⊕ · · · ⊕ Wm .
Since h is an isomorphism, the result follows from Lemma 11.1. (Choose h1 = h,
Y1 = Im(f1 ), h2 = [0, f1 , . . . , fm ] and Y2 = W1′ ⊕ W2 ⊕ · · · ⊕ Wm .)
Finally, we assume that W = Im(f ) ⊕ W1 ⊕ C for some submodule C of W . Let

g : W → Im(f ) be the projection from W onto Im(f ) with kernel W1 ⊕ C followed
by the isomorphism f −1 : Im(f ) → V defined by f −1 (f (v)) = v. It follows that
gf = 1V .
We can write g = [g1 , . . . , gm ] with homomorphisms gj : Wj → V , thus gj is just the

restriction of g to Wj . By assumption g1 = 0 since W1 lies in the kernel of g. In the
first part of the proof we have chosen some 1 ≤ t ≤ m such that gt ft is invertible in
End(V ). Since g1 = 0, we see that t > 1.
11.2. Consequences of the Exchange Theorem.

Corollary 11.3. Let V, X, W1 , . . . , Wm be modules, and let
m
M
V ⊕X ∼
= W := Wj .
j=1
If End(V ) is a local ring, then there exists some t with 1 ≤ t ≤ m and a direct
decomposition Wt = V ′ ⊕ Wt′ with V ′ ∼
= V and
M
X∼= Wt′ ⊕ Wj .
j6=t
Proof. The composition

Lm of the inclusion ι : V → V ⊕ X and of an isomorphism
i : V ⊕ X → j=1 Wj is a split monomorphism
m
M
f: V → Wj ,
j=1
and the cokernel Cok(f ) is isomorphic to X.

ι
0 / V / V ⊕X / X / 0

i
f
0 / V / W / Cok(f ) / 0
The Exchange Theorem provides a t with 1 ≤ t ≤ m and a direct decomposition

Wt = V ′ ⊕ Wt′ with V ′ ∼
= V such that
M
W = Im(f ) ⊕ Wt′ ⊕ Wj .
j6=t
This directL
decomposition of W shows that the cokernel of f is also isomorphic to
Z = Wt ⊕ j6=t Wj . This implies X ∼
′
= Z. In particular, we have Wt = V ′ ⊕ Wt′ ∼=
′
V ⊕ Wt .
Corollary 11.4 (Cancellation Theorem). Let V, X1 , X2 be modules with V ⊕ X1 ∼

=
V ⊕ X2 . If End(V ) is a local ring, then X1 ∼
= X2 .
Proof. We apply Corollary 11.3 with X = X1 , W1 = V and W2 = X2 . There are

two cases: In the first case there is a direct decomposition V = W1 = V ′ ⊕ W1′ with
V′ ∼
= V and X1 ∼ = W1′ ⊕W2 . Since V is indecomposable, V ∼ = V ′ ⊕W1′ implies W1′ = 0.
Therefore X1 ∼ = W2 = X2 . In the second case, there is a direct decomposition
X2 = W2 = V ′ ⊕ W2′ with V ′ ∼ = V and X1 ∼ = V ⊕ W2′ , thus X2 ∼= V ⊕ W2′ ∼
= X1 .
Corollary 11.5 (Krull-Remak-Schmidt Theorem). Let V1 , . . . , Vn be modules with
local endomorphism rings, and let W1 , . . . , Wm be indecomposable modules. If
n
M m
M
Vi ∼
= Wj
i=1 j=1
then n = m and there exists a permutation π such that Vi ∼

= Wπ(i) for all 1 ≤ i ≤ n.
Proof. We proof this via induction

L on n: For n = 0 there is nothing to show. Thus
let n ≥ 1. Set V = V1 and X = ni=2 Vi . By Corollary 11.3 there is some 1L
≤t≤m
′ ′ ′ ∼ ∼ ′
and a direct decomposition Wt = V1 ⊕ Wt with V1 = V1 and X = Wt ⊕ j6=t Wj .
The indecomposability of Wt implies Wt′ = 0. This implies
n
M M
Vi ∼
= Wj .
i=2 j6=t
By induction n − 1 = m − 1, and there exists a bijection
π : {2, . . . , n} → {1, . . . , m} \ {t}
such that Vi ∼
= Wπ(i) for all 2 ≤ i ≤ n. Now just set π(1) = t.
Rema(r)k: In the literature the Krull-Remak-Schmidt Theorem is often called

Krull-Schmidt Theorem. But in fact Remak was the first to prove such a result in
the context of finite groups, which Krull then generalized to modules. The result
was part of Robert Remaks Doctoral Dissertation which he published in 1911. He
was born in 1888 and murdered in Auschwitz in 1942.
References
[K] W. Krull, Über verallgemeinerte endliche Abelsche Gruppen, Mathematische Zeitschrift 23
(1925), 161–196.
[R] R. Remak, Über die Zerlegung der endlichen Gruppen in direkte unzerlegbare Faktoren, Journal
für die Reine und Angewandte Mathematik (1911), 293–308.
[S1] O. Schmidt, Sur les produits directs, Bulletin de la S.M.F. 41 (1913), 161–164.
[S2] O. Schmidt, Über unendliche Gruppen mit endlicher Kette, Mathematische Zeitschrift 29
(1928), 34–41.
Corollary 11.6. Let V1 , . .L

. , Vn be modules with local endomorphism ring, and let
U be a direct summand of ni=1 Vi . Then there exists a subset I ⊆ {1, . . . , n} such
that M
U∼
= Vi .
i∈I
Proof. We prove this via induction

L on n: For n = 0 there is nothing to show. Thus
1. Set V = V1 , X = ni=2 Vi and W1 = U. Let W2 be a direct complement
let n ≥ L
of U in ni=1 Vi . Thus
V ⊕ X = W1 ⊕ W2 .
There are two cases: In the first case there is a direct decomposition W1 = U =
= V and X ∼
V ′ ⊕ U ′ with V ′ ∼ = U ′ ⊕ W2 . Since U ′ is isomorphic toLa direct summand
′ ′ ∼
of X, induction yields a subset I ⊆ {2, . . . , n} such that U = i∈I ′ Vi . Thus with
I := I ′ ∪ {1} we get M
U = V ′ ⊕ U′ ∼
= V1 ⊕ U ′ ∼ = Vi .
i∈I
In the second case there is a direct decomposition W2 = V ′ ⊕ W2′ with V ′ ∼

= V and
∼ ′
X = U ⊕ W2 . Thus U is also isomorphic L to a direct summand of X. Therefore there
is a subset I ⊆ {2, . . . , n} with U ∼
= i∈I Vi .
11.3. Examples. We present some examples which show what happens if we work
with indecomposable direct summands, whose endomorphism ring is not local.
Assume |J| = 2, thus M = (K[T1 , T2 ], T1 ·, T2 ·) is a J-module. Let U1 and U2 be
non-zero submodules of M. We claim that U1 ∩ U2 6= 0: Let u1 ∈ U1 and u2 ∈ U2
be non-zero elements. Then we get u1 u2 ∈ U1 ∩ U2 , and we have u1 u2 6= 0.
In other words, the module M is uniform. (Recall that a module V is called uniform
if for all non-zero submodules U1 and U2 of V we have U1 ∩ U2 6= 0.) This implies
that every submodule of M is indecomposable.
The submodules U of M are the ideals of K[T1 , T2 ]. If U is generated by elements

p1 , . . . , pt , we write U = I(p1 , . . . , pt ). (One can show that every ideal in K[T1 , T2 ]
is finitely generated, but we do not need this here.)
Now let U1 , U2 be ideals with U1 + U2 = K[T1 , T2 ]. This yields an exact sequence
f g
0 → U1 ∩ U2 −
→ U1 ⊕ U2 −
→ M → 0,
where f = t [ι, −ι] and g = [ι, ι]. Here we denote all inclusion homomorphisms just
by ι.
This sequence splits: Since g is surjective, there is some u1 ∈ U1 and u2 ∈ U2 with

g(u1, u2) = 1. If we define
h : M → U1 ⊕ U2
by h(p) = (pu1 , pu2 ) for p ∈ K[T1 , T2 ], then this is a homomorphism, and we have
gh = 1M . This implies
M ⊕ (U1 ∩ U2 ) ∼ = U1 ⊕ U2 .
This setup allows us to construct some interesting examples:
Example 1: Let U1 = I(T1 , T2 ) and U2 = I(T1 − 1, T2 ). We obtain
M ⊕ (U1 ∩ U2 ) ∼ = I(T1 , T2 ) ⊕ I(T1 − 1, T2 ).
Now M is a cyclic module, but I(T1 , T2 ) and I(T1 − 1, T2 ) are not. Thus I(T1 , T2 ) ⊕
I(T1 −1, T2 ) contains a cyclic direct summand, but none of the indecomposable direct
summands I(T1 , T2 ) and I(T1 −1, T2 ) is cyclic. (We have U1 ∩U2 = I(T1 (T1 −1), T2 ),
but this fact is not used here.)
Example 2: Let U1 = I(T1 ) and U2 = I(T12 − 1, T1 T2 ). We obtain U1 ∩ U2 =
I(T13 − T1 , T1 T2 ) and
M ⊕ I(T 3 − T1 , T1 T2 ) ∼
1 = I(T1 ) ⊕ I(T 2 − 1, T1 T2 ).
1
The map f 7→ T1 f yields an isomorphism M → I(T1 ), but the modules I(T13 −

T1 , T1 T2 ) and I(T12 − 1, T1 T2 ) are not isomorphic. Thus in this situation there is no
cancellation rule.
Example 3: Here is another (trivial) example for the failure of the cancellation
rule: Let J = ∅, and let V be an infinite dimensional K-vector space. Then we have
V ⊕K ∼ =V ∼ = V ⊕ 0.
Thus we cannot cancel V . On the other hand, in contrast to Example 2, V is not
an indecomposable module.
11.4. Exercises. 1: Let V = (K[T ], T ·). Show:
(a): The direct summands of V ⊕ V are 0, V ⊕ V and all the submodules of the
form
Uf,g := {(hf, hg) | h ∈ K[T ]}
where f and g are polynomials with greatest common divisor 1.
(b): There exist direct summands U of V ⊕ V such that none of the modules 0,
V ⊕ V , U1,0 = V ⊕ 0 and U0,1 = 0 ⊕ V are a direct complement of U in V ⊕ V .
2: Let M1 , . . . , Mt be pairwise non-isomorphic modules of finite length, and let
mi ≥ 1 for 1 ≤ i ≤ t. Define
Mt
V = Mimi ,
i=1
and let R = End(V ) be the endomorphism ringLof V . Show: There exists an

idempotent e in R such that e(V ) is isomorphic to ti=1 Mi , and we have R = ReR.
***********************************************************************
Part 3. Modules II: A-Modules
12. Modules over algebras
12.1. Representations of an algebra. Let A and B be K-algebras. A map

η : A → B is a K-algebra homomorphism if η is a ring homomorphism which is
also K-linear. In other words, for all a1 , a2 ∈ A and all λ ∈ K the map η satisfies
the following:
η(a1 + a2 ) = η(a1 ) + η(a2 ),
η(λa) = λη(a),
η(a1 a2 ) = η(a1 )η(a2 ),
η(1A ) = 1B .
An example of an algebra is the endomorphism ring EndK (V ) of a K-vector space

V . The underlying set of EndK (V ) is the set of K-linear maps f : V → V . Addition
and scalar multiplication are defined pointwise, and the multiplication is given by
the composition of maps. Thus we have
(f1 + f2 )(v) = f1 (v) + f2 (v),
(λf )(v) = λ(f (v)) = f (λv),
(f1 f2 )(v) = f1 (f2 (v))
for all f, f1 , f2 ∈ EndK (V ), λ ∈ K and v ∈ V .
Similarly, the set Mn (K) of n × n-matrices with entries in K forms naturally a
K-algebra.
From the point of view of representation theory, these algebras are very boring (they
are “semisimple”). We will meet more interesting algebras later on.
A representation of a K-algebra A is a K-algebra homomorphism
η : A → EndK (V )
where V is a K-vector space. We want to write down explicitely what this means:
To every a ∈ A we associate a map η(a) : V → V such that the following hold:
(R1 ) η(a)(v1 + v2 ) = η(a)(v1 ) + η(a)(v2 ),
(R2 ) η(a)(λv) = λ(η(a)(v)),
(R3 ) η(a1 + a2 )(v) = η(a1 )(v) + η(a2 )(v),
(R4 ) η(λa)(v) = λ(η(a)(v)),
(R5 ) η(a1 a2 )(v) = η(a1 )(η(a2 )(v)),
(R6 ) η(1A )(v) = v
for all a, a1 , a2 ∈ A, v, v1 , v2 ∈ V and λ ∈ K. The conditions (R1 ) and (R2 ) just
mean that for every a ∈ A the map η(a) : V → V is K-linear. The other rules show
that η is an algebra homomorphism: (R3 ) and (R4 ) say that η is K-linear, (R5 )
means that η is compatible with the multiplication, and (R6 ) shows that the unit
element of A is mapped to the unit element of EndK (V ).
12.2. Modules over an algebra. An A-module structure on V (or more pre-

cisely, a left A-module structure on V ) is a map
σ: A × V → V
(where we write a·v or av instead of σ(a, v)) such that for all a, a1 , a2 ∈ A, v, v1 , v2 ∈
V and λ ∈ K the following hold:
(M1 ) a(v1 + v2 ) = av1 + av2 ,
(M2 ) a(λv) = λ(av),
(M3 ) (a1 + a2 )v = a1 v + a2 v,
(M4 ) (λa)v = λ(av),
(M5 ) (a1 a2 )v = a1 (a2 v),
(M6 ) 1A v = v.
The conditions (M1 ) and (M2 ) are the K-linearity in the second variable, and (M3 )
and (M4 ) are the K-linearity in the first variable. Condition (M5 ) gives the com-
patibility with the multiplication, and (M6 ) ensures that 1A acts as the identity on
V . The map σ is sometimes called scalar multiplication. An A-module (left
A-module) is a vector space V together with an A-module structure on V .
Thus an A-module V has two scalar multiplications: the one coming from V as a
vector space over our ground field K, and the other one from the A-module structure.
In the latter case, the scalars are elements of A. The scalar multiplication with
elements of K is just a special case of the scalar multiplication with elements of the
algebra, because λ · v = (λ · 1A ) · v for all λ ∈ K and v ∈ V .
12.3. Modules and representations. Let Abb(V, V ) be the set of all (set theo-
retic) maps V → V . If we have any (set theoretic) map η : A → Abb(V, V ), then
we can define a map η : A × V → V by
η(a, v) = η(a)(v).
This defines a bijection between the set of all maps A → Abb(V, V ) and the set of
maps A × V → V .
Lemma 12.1. Let A be a K-algebra, and let V be a K-vector space. If η : A →
EndK (V ) is a map, then η is a representation of A if and only if η : A × V → V is
an A-module structure on V .
Proof. If η : A → EndK (V ) is a representation, we obtain a map

η: A × V → V
which is defined by η((a, v)) := η(a)(v). Then η defines an A-module structure on
V.
Vice versa, let σ : A × V → V be an A-module structure. Then the map σ : A →

EndK (V ) which is defined by σ(a)(v) := σ(a, v) is a representation of A.
Now it is easy to match the conditions (Ri ) and (Mi ) for 1 ≤ i ≤ 6.
Let V be an A-module. We often write V = A V and say “V is a left A-module”.

Often we are a bit sloppy and just say: “V is an A-module”, “V is a module over A”,
“V is a module” (if it is clear which A is meant), or “V is a representation” without
distinguishing between the two concepts of a “module” and a “representation”.
12.4. A-modules and |A|-modules. Let η : A → EndK (V ) be a representation of

A. Since we associated to every a ∈ A an endomorphism η(a) of a vector space V ,
we see immediately that each representation of A gives us an |A|-module, namely
(V, η(a))a∈|A| .
By |A| we just mean the underlying set of the algebra A. (Of course |A| is just A
itself, but we can forget about the extra structure (like multiplication etc.) which
turns the set A into an algebra.) But note that the endomorphisms η(a) are not
just arbitrary and cannot be chosen independently of each other: They satisfy very
strong extra conditions which are given by (R3 ), . . . , (R6 ).
So we see that every A-module is an |A|-module.
This means that we can use the terminology and theory of modules which we de-
veloped in the previous chapters in the context of A-modules. (We just interpret
them as |A|-modules.) The |A|-modules are the maps A × V → V which satisfy
the axioms (M1 ) and (M2 ), and the A-modules are exactly the |A|-modules which
additionally satisfy (M3 ), . . . , (M6 ).
If A V is an A-module, then every submodule and every factor module (in the sense
of the general moduleLdefinition) is again an A-module.
Q If A Vi , i ∈ I are A-modules,
then the direct sum i∈I A Vi and the product i∈I A Vi are again A-modules.
As suggested in the considerations above, if A V and A W are A-modules, then a
map f : V → W is an A-module homomorphism if it is a homomorphism of
|A|-modules. In other words, for all v, v1 , v2 ∈ V and a ∈ A we have
f (v1 + v2 ) = f (v1 ) + f (v2 ),
f (av) = af (v).
We write HomA (V, W ) or just Hom(V, W ) for the set of homomorphisms A V → A W .
Recall that HomA (V, W ) is a K-vector space (with respect to addition and scalar
multiplication). Similarly, let EndA (V ) or End(V ) be the endomorphism ring of V .
By Mod(A) we denote the K-linear category with all A-modules as objects, and
with A-module homomorphisms as morphisms. We call Mod(A) the category of
(left) A-modules. By mod(A) we denote the category of all finite-dimensional
A-modules. This is a full subcategory of Mod(A).
12.5. Free modules. Let V be an A-module. A subset U of V is a submodule if

and only if U is closed under addition and scalar multiplication with scalars from
A.
If X is a subset of V , then the Pnsubmodule U(X) generated by X is the set of
all (finite) linear combinations P i=1 ai xi with x1 , . . . , xn ∈ X and a1 , . . . , an ∈ A:
Clearly the elements of the form ni=1 ai xi have to belong to U(X). On the other
hand, the set of all elements, which can be written in such a way, is closed under
addition and scalar multiplication. Thus they form a submodule and this submodule
contains X.
For x ∈ V let Ax = {ax | a ∈ A}. Thus Ax is the submodule of V generated by x.
Similarly, for all subsets X ⊆ V we have
X
U(X) = Ax.
x∈X
If A is an algebra, then the multiplication map µ : A × A → A satisfies all properties

of an A-module structure, where V = A as a vector space. Thus by our convention
we denote this A-module by A A. The corresponding representation
A → EndK (A)
with a 7→ λa is the regular representation. Here for a ∈ A the map λa : A → A
is defined by λa (x) = ax, thus λa is the left multiplication map with a.
A free A-module is by definition a module V which is isomorphic to a (possibly

infinite) direct sum of copies of A A.
If V is an A-module, then a subset X of V is a free generating set if the following
two conditions are satisfied:
P
• X is a generating set of V , i.e. V = x∈X Ax;
• If x1 , . . . , xn are pairwise different elements in X and a1 , . . . , an are arbitrary
elements in A with
X n
ai xi = 0,
i=1
then ai = 0 for all 1 ≤ i ≤ n.
(Compare the definition of a free generating set with the definition of a basis of a
vector space, and with the definition of a linearly independent set of vectors.)
Lemma 12.2. An A-module is free if and only if it has a free generating set.
L
Proof. Let W be a direct sum of copies of A A, say W = i∈I Wi with Wi = A A
for all i ∈ I. By ei we denote the 1-element of Wi . (In coordinate notation: All
coefficients of ei are 0, except the ith coefficient is the element 1A ∈ A A = Wi .)
Thus the set {ei | i ∈ I} is a free generating set of W .
If f : W → V is an isomorphism of A-modules, and if X is a free generating set of

W , then f (X) is a free generating set of V .
Vice versa, we want to show that every A-module V with a free generating set X
is isomorphic to a free module. We take a direct
L sum of copies of A A, which are
indexed by the elements in X. Thus W = x∈X Wx where Wx = A A for all x.
As before, let ex be the 1-element
P of Wx = A A. Then every element in W can be
written as a (finite) sum x∈X ax ex with ax ∈ A for all x ∈ X, and ax = 0 for
almost all (i.e. all but finitely many) x ∈ X. We define a map f : W → V by
!
X X
f ax ex = ax x.
x∈X x∈X
It is easy to check that f is an A-module homomorphism which is surjective and

injective, thus it is an isomorphism of A-modules.
If F is a free A-module with free generating set X, then the cardinality of X is

called the rank of F . Thus F has finite rank, if X is a finite set.
Let F be a free A-module, and let W be an arbitrary A-module. If X is a free

generating set of F , and if we choose for every x ∈ X an element wx ∈ W , then
there exists exactly one A-module homomorphism f : F → W such that f (x) = wx
for all x ∈ X. Namely, !
X X
f ax x = ax wx
x∈X x∈X
for all x ∈ X and all ax ∈ A. If the set {wx | x ∈ X} is a generating set of
the A-module W , then the homomorphism f is surjective. Thus in this case W is
isomorphic to a factor module of F . So we proved the following result:
Theorem 12.3. Every A-module is isomorphic to a factor module of a free A-
module.
Inside the category of all |A|-modules, we can now characterize the A-modules as
follows: They are exactly the modules which are isomorphic to some factor module of
some free A-module. Thus up to isomorphism one obtains all A-modules by starting
with A A, taking direct sums of copies of A A and then taking all factor modules of
these direct sums.
Every finitely generated A-module is isomorphic to a factor module of a free module

of finite rank. In particular, each simple A-module is isomorphic to a factor module
of a free module of rank one. Thus, we get the following:
Lemma 12.4. Let A be a finite-dimensional K-algebra. For an A-module M the
following are equivalent:
(i) M is finitely generated;

(ii) M is finite-dimensional as a K-vector space;
(iii) l(M) < ∞.
12.6. The opposite algebra. If A is a K-algebra, then we denote the opposite

algebra of A by Aop . Here we work with the same underlying vector space, but the
multiplication map is changed: To avoid confusion, we denote the multiplication of
Aop by ⋆, which is defined as
a1 ⋆ a2 = a2 · a1 = a2 a1
for all a1 , a2 ∈ A (where · is the multiplication of A). This defines again an algebra.
Of course we have (Aop )op = A.
Lemma 12.5. If A is an algebra, then EndA (A A) ∼

= Aop .
Proof. As before let λa : A → A be the left multiplication with a ∈ A, i.e. λa (x) = ax

for all x ∈ A. Similarly, let ρa : A → A be the right multiplication with a ∈ A, i.e.
ρa (x) = xa for all x ∈ A. It is straightforward to check that the map
ρ : Aop → EndA (A A)
defined by ρ(a) = ρa is an algebra homomorphism. In particular we have
ρ(a1 ⋆ a2 )(x) = ρ(a2 a1 )(x)

= x(a2 a1 ) = (xa2 )a1
= (ρ(a2 )(x)) · a1
= ρ(a1 )ρ(a2 )(x)
for all a1 , a2 , x ∈ A. The map ρ is injective: If a ∈ |A|, then ρ(a)(1) = 1 · a = a.

Thus ρ(a) = 0 implies a = ρ(a)(1) = 0.
We know that
λa ρb = ρb λa
for all a, b ∈ A. (This follows directly from the associativity of the multiplication in
A.) In other words, the vector space endomorphisms ρb are endomorphisms of the
A-module A A, and ρ yields an embedding of Aop into EndA (A A).
It remains to show that every endomorphism f of A A is a right multiplication: Let
f (1) = b. We claim that f = ρb : For a ∈ A we have
f (a) = f (a · 1) = a · f (1) = a · b = ρb (a).
12.7. Right A-modules. A right A-module structure on V is a map
ρ: V × A → V
(where we write v ·a or va instead of ρ(v, a)) such that for all a, a1 , a2 ∈ A, v, v1 , v2 ∈

V and λ ∈ K the following hold:
(M1′ ) (v1 + v2 )a = v1 a + v2 a,
(M2′ ) (λv)a = λ(va),
(M3′ ) v(a1 + a2 ) = va1 + va2 ,
(M4′ ) v(λa) = λ(va),
(M5′ ) v(a1 a2 ) = (va1 )a2 ,
(M6′ ) v1A = v.
A right A-module is a vector space V together with a right A-module structure
on V . We often write V = VA and say “V is a right A-module”.
12.8. Examples. For A-modules V and W the homomorphism space HomA (V, W )
carries a (left) EndA (W )-module structure defined by
EndA (W ) × HomA (V, W ) → HomA (V, W ), (f, g) 7→ f g,
and HomA (V, W ) has a right EndA (V )-module structure given by
HomA (V, W ) × EndA (V ) → HomA (V, W ), (g, f ) 7→ gf.
One can also turn A V into a module over EndA (V ) by
EndA (V ) × V → V, (f, v) 7→ f (v).
12.9. Direct decompositions of the regular representation. Let A be a K-

algebra, and let
M
AA = Pi
i∈I
P modules Pi 6= 0 for all

be a direct decomposition of the regular module A A with
i ∈ I. Thus every element a ∈ A is of the form a = i∈I ai with ai ∈ Pi . (Only
finitely many of the ai are allowed to be non-zero.) In particular let
X
1 = 1A = ei
i∈I
with ei ∈ Pi .
Lemma 12.6. For all i ∈ I we have Pi = Aei .
Proof. Since Pj is a submodule of A A and ej ∈ Pj , we know that Aej ⊆ Pj . Vice

versa, let x ∈ Pj . We have
X
x=x·1 = xei .
i∈I
Since x belongs to Pj , and since A A is the direct sum of the submodules Pi , we get
x = xej (and xei = 0 for all i 6= j). In particular, x ∈ Aej .
Lemma 12.7. The index set I is finite.
Proof. Only finitely many of the ei are different from 0. If ei = 0, then Pi = Aei = 0,
a contradiction to our assumption.
A set {fi | i ∈ I} ⊆ R of idempotents in a ring R is a set of pairwise orthogonal

idempotents if fi fjP= 0 for all i 6= j. Such a set of pairwise orthogonal idempotents
is complete if 1 = i∈I fi .
Lemma 12.8. The set {ei | i ∈ I} defined above is a complete set of pairwise
orthogonal idempotents.
Proof. We have X
ej = ej · 1 = ej ei .
i∈I
As in the proof of Lemma 12.6, the unicity of the decomposition of an element in a
direct sum yields that ej = ej ej and ej ei = 0 for all i 6= j.
L
Warning: Given a direct decomposition A A = i∈I Pi . If we choose idempotents
ei ∈ Pi with Pi = Aei , then these idempotents do not have to be orthogonal to each
other. For example, let A = M2 (K) be the algebra of 2 × 2-matrices with entries in
K. Take
1 0 0 0
e1 = and e2 = ,
1 0 0 1
and define Pi = Aei . We obtain A A = P1 ⊕ P2 . The elements e1 and e2 are
idempotents, but they are not orthogonal.
Lemma 12.8 shows that any direct decomposition of A A yields a complete set of
orthogonal idempotents in A. Vice versa, assume that fi , i ∈ I is a complete set of
orthogonal idempotents in an algebra A, then
M
A A = Afi
is a direct decomposition of A A.
Example: Let B be an algebra, and let A = Mn (B) be the algebra of n×n-matrices

with entries in B for some n ∈ N1 . Let eij be the n × n-matrix with entry 1 at the
position (i, j) and all other entries 0. For brevity write ei = eii . The diagonal
matrices ei , 1 ≤ i ≤ n form a complete set of orthogonal idempotents in A. Note
that Aei contains exactly the matrices whose only non-zero entries are in the ith
column. It follows immediately that
M n
AA = Aei .
i=1
Note also that the modules Aei are isomorphic to each other: We get an isomorphism
Aei → Aej via right multiplication with eij .
Instead of working with this isomorphism, we could also argue like this: Let X = B n
be the vector space of n-tupels with coefficients in B. We interpret these n-tupels
as n × 1-matrices. So matrix multiplication yields an A-module structure on X. It
is clear that X and Aei have to be isomorphic: X and Aei only differ by the fact
that Aei contains some additional 0-columns.
Warning: The above direct decomposition of Mn (B) is for n ≥ 2 of course not the
only possible decomposition. For example for n = 2 and any x ∈ B the matrices

1 x 0 −x
and
0 0 0 1
form also a complete set of orthogonal idempotents in M2 (B). In this case

1 x 0 −x
M2 (B) = M2 (B) ⊕ M2 (B) ,
0 0 0 1
where
1 x
M2 (B)
0 0
consists of the matrices of the form

b1 b1 x
b2 b2 x
with b1 , b2 ∈ B, and
0 −x
M2 (B)
0 1
consists of the matrices whose only non-zero entries are in the second column.
12.10. Modules defined by idempotents. As before, let A be a K-algebra.

Lemma 12.9. Let e be an idempotent in A. The endomorphism ring EndA (Ae) of
the A-module Ae is isomorphic to (eAe)op . In particular, EndA (A A) is isomorphic
to Aop . We obtain an isomorphism
η : EndA (Ae) → (eAe)op
which maps f ∈ EndA (Ae) to f (e). Vice versa, for each a ∈ A, the inverse η −1 (eae)
is the right multiplication with eae.
Proof. Let f ∈ EndA (Ae), and let a = f (e) ∈ Ae. Then a = ae because a belongs
to Ae. Since f is a homomorphism, and e is an idempotent we have a = f (e) =
f (e2 ) = ef (e) = ea. Thus a = eae ∈ eAe. Clearly, the map defined by η(f ) = f (e)
is K-linear.
Let f1 , f2 ∈ EndA (Ae), and let η(fi ) = fi (e) = ai for i = 1, 2. We get

η(f1 f2 ) = (f1 f2 )(e) = f1 (f2 (e)) = f1 (a2 ) = f1 (a2 e) = a2 f1 (e) = a2 a1 .
Thus η yields an algebra homomorphism η : EndA (Ae) → (eAe)op . (Note that the
unit element of (eAe)op is e.)
The algebra homomorphism η is injective: If η(f ) = 0, then f (e) = 0 and therefore
f (ae) = af (e) = 0 for all a ∈ A. Thus f = 0.
The map η is also surjective: For every a ∈ A let ρeae : Ae → Ae be the right
multiplication with eae defined by ρeae (be) = beae where b ∈ A. This map is
obviously an endomorphism of the A-module Ae, and we have η(ρeae ) = ρeae (e) =
eae.
Thus we have shown that η is bijective. In particular, the inverse η −1 (eae) is the
right multiplication with eae.
Lemma 12.10. If X is an A-module, then HomA (Ae, X) ∼
= eX as vector spaces.
Proof. Let η : HomA (Ae, X) → eX be the map defined by η(f ) = f (e). Since
f (e) = f (e2 ) = ef (e), we have f (e) ∈ eX, thus this is well defined. It is also clear
that η is K-linear.
If f1 , f2 ∈ HomA (Ae, X), then η(f1 ) = η(f2 ) implies f1 (e) = f2 (e), and therefore
f1 (ae) = af1 (e) = af2 (e) = f2 (ae)
for all a ∈ A. So f1 = f2 . This proves that η is injective.
Next, let ex ∈ eX. Define fx : Ae → X by fx (ae) = aex. It follows that fx (a1 a2 e) =

a1 fx (a2 e) for all a1 , a2 ∈ A. Thus fx ∈ HomA (Ae, X) and η(fx ) = fx (e) = ex. So η
is surjective.
An idempotent e 6= 0 in a ring R is called primitive if e is the only non-zero

idempotent in eRe.
Lemma 12.11. A non-zero idempotent e in a ring R is primitive if and only if the
following hold: Let e = e1 + e2 with e1 and e2 orthogonal idempotents, then e1 = 0
or e2 = 0.
Proof. Let e1 and e2 be orthogonal idempotents with e = e1 + e2 . Then ee1 e = e1

and ee2 e = e2 . Thus e1 and e2 belong to eRe.
Vice versa, if e′ is an idempotent in eRe, then e′ and e − e′ is a pair of orthogonal
idempotents with sum equal to e.
Lemma 12.12. Let e, e′ be idempotents in A. Then the following are equivalent:
(i) The modules Ae and Ae′ are isomorphic;

(ii) There exist some x ∈ eAe′ and y ∈ e′ Ae such that xy = e and yx = e′ .
Proof. We can identify HomA (Ae, M) with eM: We just map f : Ae → M to f (e).
Since e = e2 we get f (e) = f (e2 ) = ef (e), thus f (e) ∈ eM. Thus the homomor-
phisms f ∈ HomA (Ae, Ae′ ) correspond to the elements in eAe′ .
(i) =⇒ (ii): If Ae and Ae′ are isomorphic, there exist homomorphisms f : Ae → Ae′
and g : Ae′ → Ae such that gf = 1Ae . Set x = f (e) and y = g(e′ ). Thus x ∈ eAe′ ,
y ∈ e′ Ae, xy = e and yx = e′ .
(ii) =⇒ (i): Assume there exist elements x ∈ eAe′ and y ∈ e′ Ae with xy = e and
yx = e′ . Let f : Ae → Ae′ be the right multiplication with x, and let g : Ae′ → Ae be
the right multiplication with y. Then f and g are A-module homomorphisms, and we
have gf = 1Ae and f g = 1Ae′ . Thus the A-modules Ae and Ae′ are isomorphic.
The statement (ii) in the above lemma is left-right symmetric. Thus (i) and (ii) are
also equivalent to
(iii) The Aop -modules eA and e′ A are isomorphic.
We want to compare A-modules and eAe-modules. If M is an A-module, then eM

is an eAe-module.
Lemma 12.13. Let e be an idempotent in A. If S is a simple A-module with eS 6= 0,
then eS is a simple eAe-module.
Proof. We have to show that every element x 6= 0 in eS generates the eAe-module

eS. Since x ∈ S, we get Ax = S. Thus eAex = eAx = eS. Here we used that
x = ex for every element x ∈ eS.
Recall that a module V is called local if it contains a maximal submodule U, which

contains all proper submodules of V .
Lemma 12.14. Let e be an idempotent in A. Then Ae is a local module if and only
if eAe is a local ring.
Proof. Let Ae be a local module, and let M be the maximal submodule of Ae. For
every element x ∈ Ae we have x = xe, thus M = Me. We have eM = eAe ∩ M.
(Clearly, eM ⊆ eAe ∩ M. The other inclusion follows from the fact that e is an
idempotent: If a ∈ A and eae ∈ M, then eae = e(eae) ∈ eM.)
In particular we have eM = eMe ⊆ eAe. We have e ∈ eAe, but e does not belong
to M or eM. Thus eMe ⊂ eAe.
We claim that eMe is an ideal in eAe: Clearly, eAe · eMe ⊆ eMe. Since the right
multiplications with the elements from eAe are the endomorphisms of Ae, we have
Me · eAe ⊆ Me. (Me is the radical of the module Ae.) Thus eMe · eAe ⊆ eMe.
If x ∈ eAe \ eMe, then x ∈ Ae and x ∈ / M. (Note that exe = x.) Thus x generates
the local module Ae. It follows that there exists some y ∈ A with yx = e. Because
x = ex, we have
eye · x = eyx = e2 = e.
Thus x is left-invertible in eAe, and eye is right-invertible in eAe.
The element eye does not belong to eM, since eM is closed under right multiplication
with elements from eAe, and e ∈/ eM. So we get eye ∈ eAe \ eMe.
Thus also the element eye has a left inverse in eAe. This proves that eye is invertible
in eAe. It follows that exe is invertible in eAe: Namely, we have
(eye)−1 · eye · x = (eye)−1 e.
Multiplying both sides of this equation from the right with eye yields x · eye = e.
We have shown that all elements in eAe \ eMe are invertible, thus eAe is a local
ring.
Vice versa, assume that eAe is a local ring. Then Ae is a non-zero cyclic module,
thus it has maximal submodules. Let M1 be a maximal submodule, and let M2 be
any proper submodule of Ae. Suppose M2 is not contained in M1 . This implies
Ae = M1 + M2 , thus e = x1 + x2 with xi ∈ Mi . We have e = ee = ex1 + ex2 .
Since eAe is a local ring, one of the elements exi = exi e, i = 1, 2 is invertible in
eAe. If e = yexi , then e belongs to Axi ⊆ Mi , thus Ae = Mi . By assumption both
modules M1 and M2 are proper submodules of Ae. This contradiction shows that
M1 contains all proper submodules of Ae, thus Ae is a local module.
12.11. Modules over factor algebras. Let A be a K-algebra, and let I be an

ideal in A. Define B = A/I. If M = B M is a B-module, then we can turn M
into an A-module by defining a · m := am for all a ∈ A and m ∈ M. We write
ιA
B (M) or A M for this A-module. (But often we just write M.) For this A-module
M = ιA B (M) we obviously have I · M = 0. We say that M is annihilated by I.
Vice versa, if X is an A-module with I · X = 0, then we can interpret X as a

B-module: For b ∈ B and x ∈ X, we write b = a + I and then define b · x := ax.
This is well defined since I · X = 0. It is easy to check that this turns X into an
B-module.
These two constructions are inverse to each other. Thus we can identify the B-
modules with the A-modules, which are annihilated by I.
The following is obviously also true: If M1 and M2 are A-modules, which are anni-
hilated by I, then a map M1 → M2 is A-linear if and only if it is B-linear. Thus we
get HomA (M1 , M2 ) = HomB (M1 , M2 ).
Proposition 12.15. Let I be an ideal in a K-algebra A, and let B = A/I. If we
associate to each B-module M the A-module ιA B (M), then we obtain an embedding of
the category of B-modules into the category of A-modules. The image of the functor
consists of all A-modules, which are annihilated by I.
12.12. Modules over products of algebras. Let R and S be rings. Recall that
the product R × S of R and S is again a ring with componentwise addition and
multiplication. Thus (r, s) + (r ′, s′ ) = (r + r ′ , s + s′ ) and (r, s) · (r ′ , s′ ) = (rr ′ , ss′ ).
Similarly, if A and B are algebras, then A×B is again an algebra. In this case, define
eA = (1, 0) and eB = (0, 1). These form a complete set of orthogonal idempotents.
We have (A × B)eA = A × 0 and (A × B)eB = 0 × B. These are ideals in A × B, and
we can identify the factor algebra (A × B)/(A × 0) with B, and (A × B)/(0 × B)

with A.
Let C = A × B, and let M be a C-module. We get M = eA M ⊕ eB M as a direct

sum of vector spaces, and the subspaces eA M and eB M are in fact submodules of
M. The submodule eA M is annihilated by 0 × B = (A × B)eB , thus eA M can be
seen as a module over (A × B)/(0 × B) and therefore as a module over A: For a ∈ A
and m ∈ M define a · eA m = (a, 0)eA m = (a, 0)m. Similarly, eB M is a B-module.
Thus we wrote M as a direct sum of an A-module and a B-module.
Vice versa, if M1 is an A-module and M2 is a B-module, then the direct sum M1 ⊕M2
of vector spaces becomes an (A×B)-module by defining (a, b)·(m1 , m2 ) = (am1 , bm2 )
for a ∈ A, b ∈ B, m1 ∈ M1 and m2 ∈ M2 . In particular, we can interpret all A-
modules and all B-modules as (A × B)-modules: If M is an A-module, just define
A×B
(a, b)m = am for a ∈ A, b ∈ B and m ∈ M. (This is the same as applying ιA to
M.) We call an (A × B)-modules, which is annihilated by 0 × B just an A-module,
and an (A × B)-modules, which is annihilated by A × 0 is just a B-module.
Thus we proved the following result:
Proposition 12.16. Let A and B be algebras. Then each (A × B)-module is the
direct sum of an A-module and a B-module.
In particular, indecomposable modules over A × B are either A-modules or B-

modules.
Warning: If A = B, we have to be careful. If we say that an A-module M can be

seen as a (A × B)-module, we have to make clear which copy of A we mean, thus if
we regard M as a module over A × 0 or 0 × A.
12.13. Bimodules. Let A and B be K-algebras. An A-B-bimodule V is a K-

vector space V together with two module structures
µA : A × V → V and µB : B × V → V
such that for all a ∈ A, b ∈ B and v ∈ V we have
µA (a, µB (b, v)) = µB (b, µA (a, v)).
Using our short notation av for µA (a, v) and bv instead of µB (b, v), we can write
this as
a(bv) = b(av).
Note that for all λ ∈ K and v ∈ V we have
µA (λ · 1A , v) = λv = µB (λ · 1B , v).
Warning: In many books, our A-B-bimodules are called A-B op -bimodules.

Assume that M is an A-B-bimodule. We get a canonical map c : B → EndA (M)
which sends b ∈ B to the scalar multiplication b· : M → M which maps m to bm.
It is easy to check that the image of c lies in EndA (M): We have

c(b)(am) = b(am) = a(bm) = a(c(b)(m))
for all a ∈ A, b ∈ B and m ∈ M.
Example: Let M be an A-module, and let B := EndA (M) be its endomorphism
algebra. Then M is an A-B-bimodule. Namely, M becomes a B-module by
µB (f, m) = f (m)
for all f ∈ EndA (M) and m ∈ M. But we also have f (am) = af (m).
The next result shows that bimodule structures allow us to see homomorphism
spaces again as modules.
Lemma 12.17. Let M be an A-B-bimodule, and let N an A-C-bimodule. Then
HomA (M, N) is an B op -C-bimodule via
b(c(f (m))) = c(f (bm))
for all b ∈ B, c ∈ C, f ∈ HomA (M, N) and m ∈ M.
Proof. Let ⋆ be the multiplication in B op , and set H := HomA (M, N). It is clear
that the two maps B op × H → H, (b, f ) 7→ (bf : m 7→ f (bm)) and C × H → H,
(c, f ) 7→ (cf : m 7→ cf (m)) are bilinear. We also have 1B · f = f and 1C · f = f for
all f ∈ H.
For b1 , b2 ∈ B and f ∈ H we have
((b1 ⋆ b2 )f )(m) = f ((b1 ⋆ b2 )m)
= f ((b2 b1 )m)
= f (b2 (b1 m))
= (b2 f )(b1 m)
= (b1 (b2 f ))(m).
This shows that
(b1 ⋆ b2 )f = b1 (b2 (f )).
Similarly,
((c1 c2 )f )(m) = (c1 c2 )(f (m))
= c1 (c2 (f (m)))
= c1 ((c2 f )(m))
= (c1 (c2 f ))(m).
shows that (c1 c2 )f = c1 (c2 f ) for all c1 , c2 ∈ C and f ∈ H.
Let M be an A-B-bimodule. This gives a covariant functor

HomA (M, −) : Mod(A) → Mod(B op ).
Similarly, if N is an A-C-bimodule we get a contravariant functor

HomA (−, N) : Mod(A) → Mod(C).
12.14. Modules over tensor products of algebras. Let A and B be K-algebras.

Then A ⊗K B is again a K-algebra with multiplication
(a1 ⊗ b1 ) · (a2 ⊗ b2 ) := (a1 a2 ⊗ b1 b2 ).
(One has to check that this is well defined and that one gets indeed a K-algebra.)
Proposition 12.18. The category of A-B-bimodules is equivalent to the category of
A ⊗K B-modules.
Sketch of proof. Let M be an A-B-bimodule. This becomes an A ⊗K B-module via

(a ⊗ b)m := abm
for all a ∈ A, b ∈ B and m ∈ M. The same rule applied the other way round turns
an A ⊗K B-module into an A-B-bimodule.
12.15. Exercises. 1: Let A = KhX1 , . . . , Xn i be the K-algebra of polynomials in n

non-commuting variables X1 , . . . , Xn , and let J = {1, . . . , n}. Show: The category
of J-modules is equivalent to Mod(A).
In particular, Mod(K[T ]) is equivalent to the category of 1-modules.
2: Let A be a K-algebra. Show that the category of left A-modules is equivalent to
the category of right Aop -modules.
————————————————————————————-
13. Semisimple algebras
13.1. Semisimple algebras and their modules.

Theorem 13.1. Let A be an algebra. Then the following are equivalent:
(i) The module A A is semisimple;

(ii) Every A-module is semisimple;
(iii) There exist K-skew fields Di and natural numbers ni where 1 ≤ i ≤ s such
that
Ys
A=∼ Mni (Di ).
i=1
An algebra A is called semisimple if one of the equivalent conditions in the above

theorem is satisfied.
The opposite algebra Aop of a semisimple algebra A is again semisimple. This follows
directly from Condition (iii): If D is a skew field, then D op is also a skew field. For
an arbitrary ring R there is an isomorphism
Mn (R)op → Mn (Rop )
which maps a matrix Φ to its transpose t Φ.
Proof of Theorem 13.1. The implication (ii) =⇒ (i) is trivial.

(i) =⇒ (ii): Let A A be a semisimple module. Since direct sums of semisimple
modules are again semisimple, we know that all free A-modules are semisimple.
But each A-modules is a factor module of a free module, and factor modules of
semisimple modules are semisimple. Thus all A-modules are semisimple.
(i) =⇒ (iii): Since A A is semisimple, we know that A A is a direct sum of simple
modules. By the results in Section 12.9 this direct sum has to be finite. Thus
EndA (A A) is a finite product of matrix rings over K-skew fields. We know that
EndA (A A) ∼
= Aop , thus Aop is a finite product of matrix rings over K-skew fields.
Thus the same holds for A.
(iii) =⇒ (i): Let A be a product of s matrix rings over K-skew fields. We want to
show that A A is semisimple. It is enough to study the case s = 1: If A = B × C,
then the modules B B and C C are semisimple, and therefore A A is also semisimple.
Let A = Mn (D) for some K-skew field D and some n ∈ N1 . Let S = D n be the set
of column vectors of length n with entries in D. It is easy to show that S is a simple
Mn (D)-module. (One only has to show that if x 6= 0 is some non-zero column vector
in S, then Mn (D)x = S.) On the other hand, we can write A A as a direct sum of n
copies of S. Thus A A is semisimple.
Let A = Mn (D) for some K-skew field D and some n ∈ N1 . We have shown that A A
is a direct sum of n copies of the simple module S consisting of column vectors of
length n with entries in D. It follows that every A-module is a direct sum of copies
of S. (Each free module is a direct sum of copies of S, and each module is a factor
module of a free module. If a simple A-module T is isomorphic to a factor module
of a free A-module, we obtain a non-zero homomorphism S → T . Thus T ∼ = S by
Schur’s Lemma.) If
Ys
A= ∼ Mni (Di ),
i=1
then there are exactly s isomorphism classes of simple A-modules.
Proposition 13.2. Let K be an algebraically closed field. If A is a finite-dimensional
semisimple K-algebra, then
Y s
A∼ = Mni (K)
i=1
for some natural numbers ni , 1 ≤ i ≤ s.
Proof. First, we look at the special case A = D, where D is a K-skew field: Let
d ∈ D. Since D is finite-dimensional, the powers di with i ∈ N0 cannot be linearly
independent. Thus there exists a non-zero polynomial p in K[T ] such that p(d) = 0.
We can assume that p is monic. Since K is algebraically closed, we can write it as
a product of linear factors, say p = (T − c1 ) · · · (T − cn ) with ci ∈ K. Thus in D we
have (d − c1 ) · · · (d − cn ) = 0. Since D has no zero divisors, we get d − ci = 0 for
some i, and therefore d = ci ∈ K.
Now we investigate the general case: We know that A is isomorphic to a product
of matrix rings of the form Mni (Di ) with K-skew fields Di and ni ∈ N1 . Since A
is finite-dimensional, every K-skew field Di must be finite-dimensional over K. But
since K is algebraically closed, and the Di are finite-dimensional K-skew fields we
get Di = K.
The centre of a ring R is by definition the set of elements c ∈ R such that cr = rc

for all r ∈ R. We denote the centre of R by C(R). If R and S are rings, then
C(R × S) = C(R) × C(S).
Qs
Lemma 13.3. If A ∼ = Mn (K), then the centre of A is s-dimensional.
i=1 i
Proof. It is easy to show that the centre of a matrix ring Mn (K) is just the set of
scalar matrices. Thus we get
s
! s s
Y Y Y
C Mni (K) = C(Mni (K)) = ∼ K.
i=1 i=1 i=1

13.2. Examples: Group algebras. Let G be a group, and let K[G] be a K-vector
space with basis {eg | g ∈ G}. Define
eg eh := egh .
Extending this linearly turns the vector space K[G] into a K-algebra. One calls
K[G] the group algebra of G over K. Clearly, K[G] is finite-dimensional if and
only if G is a finite group.
A representation of G over K is a group homomorphism
ρ : G → GL(V )
where V is a K-vector space. In the obvious way one can define homomorphisms
of representations. It turns out that the category of representations of G over K is
equivalent to the category Mod(K[G]) of modules over the group algebra K[G]. If
V is a K[G]-module, then for g ∈ G and v ∈ V we often write gv instead of eg v.
The representation theory of G depends very much on the field K, in particular, the
characteristic of K plays an important role.
Theorem 13.4 (Maschke). Let G be a finite group, and let K be a field such that
the characteristic of K does not divide the order of G. Then every K[G]-module is
semisimple.
Proof. It is enough to show that every finite-dimensional K[G]-module is semisimple.

Let U be a submodule of a finite-dimensional K[G]-module V . Write
V =U ⊕W
with W a subspace of V . But note that W is not necessarily a submodule.
Let θ : V → V be the projection onto U. So θ(u) = u and θ(w) = 0 for all u ∈ U
and w ∈ W . Define f : V → V by
1 X −1
f (v) = g θ(gv).
|G| g∈G
Here we use our assumption on the characteristic of K, otherwise we would divide

by 0, which is forbidden in mathematics.
We claim that f ∈ EndK[G] (V ): Clearly, f is a linear map. For h ∈ G we have

1 X −1
f (hv) = g θ(ghv).
|G| g∈G
Set xg := gh. Thus g −1 = hx−1

g . So we get
1 X −1
f (hv) = hx θ(xg v) = hf (v).
|G| g∈G g
Thus f is an endomorphism of V .
We have Im(f ) = U: Namely, Im(f ) ⊆ U since each term in the sum is in U. If
u ∈ U, then
1 X −1 1 X −1 1 X
f (u) = g θ(gu) = g gu = u = u.
|G| g∈G |G| g∈G |G| g∈G
Clearly, this implies U ∩ Ker(f ) = 0: Namely, if 0 6= u ∈ U ∩ Ker(f ), then f (u) =

u = 0, a contradiction.
We have dim Ker(f ) + dim Im(f ) = dim V . This implies

V = Im(f ) ⊕ Ker(f ) = U ⊕ Ker(f ),
and Ker(f ) is a submodule of V . Now let U be a simple submodule of V . We
get V = U ⊕ Ker(f ). By induction on dim V , Ker(f ) is semisimple, thus V is
semisimple.
13.3. Remarks. Let G be a finite group, and let K be a field. If char(K) does not
divide the order of G, then K[G] is semisimple. In this case, from our point of view,
the representation theory of K[G] is very boring. (But be careful: If you say this at
the wrong place and wrong time, you will be crucified.)
More interesting is the modular representation theory of G, i.e. the study of
representations of G over K where char(K) does divide |G|.
For example, if G = Sn is the symmetric group of bijective maps {1, . . . , n} →

{1, . . . , n}, then one can parametrize the simple K[G]-modules by certain partitions.
But in the modular case, it is not even known which dimensions these simple modules
have.
Another interesting question is the following: Given two finite groups G and H,
and let K be a field. When are the module categories mod(K[G]) and mod(K[H])
derived equivalent? (We will learn about derived categories and derived equivalences
later on.)
Again let G be a finite group, let K be a field such that char(K) divides |G|,
and let S1 , . . . , Sn be a set of representatives of the isomorphism classes of simple
K[G]-modules. We define a quiver Γ associated to K[G] as follows: Its vertices are
S1 , . . . , Sn . There is an arrow Si → Sj if and only if there exists a non-split short
exact sequence
0 → Sj → E → Si → 0
of K[G]-modules. (In fact, one should work with multiple arrows here, namely the
number of arrows Si → Sj should be equal to dim Ext1K[G](Si , Sj ), but we did not
introduce Ext-groups yet...)
The connected components of Γ parametrize the “blocks of K[G]”: A K-algebra A is
connected if it cannot be written as a product A = A1 × A2 of non-zero K-algebras
A1 and A2 . Now write K[G] as a product
K[G] = B1 × · · · × Bt
of connected algebras Bi . It turns out that the Bi are uniquely determined up
to isomorphism and reordering. They are called the blocks of K[G]. The simple
representations of a block Bi correspond to the vertices of a connected component
Γi of Γ. To understand the representation theory of K[G] is now the same as
understanding the representation theory of each of the blocks. Such blocks are in
general not any longer group algebras. Thus to understand group algebras, one
is forced to study larger classes of finite-dimensional algebras. Each block is a
“selfinjective algebra”.
Q
13.4. Exercises. 1: Prove that the K-algebra A := i∈I K is semisimple if and
only if the index set I is finite. If I is infinite, construct a submodule U of the
regular representation A A which does not have a direct complement in A A.
2: Let G = Z2 be the group with two elements, and let K be a field.
(a) Assume char(K) 6= 2. Show: Up to isomorphism there are exactly two simple
K[G]-modules.
(b) Assume char(K) = 2. Show: Up to isomorphism there are exactly two
indecomposable K[G]-modules, and one of them is not simple.
(c) Assume that K is an infinite field with char(K) = 2. Construct an infinite
number of 2-dimensional pairwise non-isomorphic representations of K[G ×
G].
————————————————————————————-
14. The Jacobson radical of an algebra
In this section let A be a K-algebra.
14.1. The radical of an algebra. The radical of A is defined as

J(A) := rad(A A).
In other words, J(A) is the intersection of all maximal left ideals of A. Often one
calls J(A) the Jacobson radical of A.
Since the A-module A A is finitely generated (it is cyclic), we know that A A contains
maximal submodules, provided A 6= 0. In particular, J(A) = A if and only if A = 0.
Lemma 14.1. The radical J(A) is a two-sided ideal.
Proof. As an intersection of left ideals, J(A) is a left ideal. It remains to show that
J(A) is closed with respect to right multiplication. Let a ∈ A, then the right multi-
plication with a is a homomorphism A A → A A, and it maps rad(A A) to rad(A A).
Lemma 14.2. If A is semisimple, then J(A) = 0.
Proof. Obvious. (Why?)

Lemma 14.3. Let x ∈ A. The following statements are equivalent:
(i) x ∈ J(A);
(ii) For all a1 , a2 ∈ A, the element 1 + a1 xa2 has an inverse;
(iii) For all a ∈ A, the element 1 + ax has a left inverse;
(iv) For all a ∈ A, the element 1 + xa has a right inverse.
Proof. (i) =⇒ (ii): Let x ∈ J(A). We have to show that 1 + x is invertible. Since
x ∈ J(A), we know that x belongs to all maximal left ideals. This implies that 1 + x
does not belong to any maximal left ideal (because 1 is not contained in any proper
ideal).
We claim that A(1 + x) = A: The module A A is finitely generated. Assume that
A(1 + x) is a proper submodule of A A. Then Corollary 7.18 implies that A(1 + x)
must be contained in a maximal submodule of A A, a contradiction.
Therefore there exists some a ∈ A with a(1 + x) = 1. Let y = a − 1. We have
a = 1 + y, thus (1 + y)(1 + x) = 1, which implies y + x + yx = 0. This implies
y = (−1 − y)x ∈ Ax ⊆ J(A). Thus also 1 + y has a left inverse. We see that 1 + y
is left invertible and also right invertible. Thus its right inverse 1 + x is also its left
inverse. Since J(A) is an ideal, also a1 xa2 belongs to J(A) for all a1 , a2 ∈ A. Thus
all elements of the form 1 + a1 xa2 are invertible.
(ii) =⇒ (iii): Obvious.
(iii) =⇒ (i): If x ∈
/ J(A), then there exists a maximal left ideal M, which does not
contain x. This implies A = M + Ax, thus 1 = y − ax for some y ∈ M and a ∈ A.
We get 1 + ax = y, and since y belongs to the maximal left ideal M, y cannot have
a left inverse.
(iii) ⇐⇒ (iv): Condition (ii) is left-right symmetric.
Corollary 14.4. The radical J(A) of A is the intersection of all maximal right
ideals.
Proof. Condition (ii) in Lemma 14.3 is left-right symmetric.

Lemma 14.5. If I is a left ideal or a right ideal of A, which consists only of nilpotent
elements, then I is contained in J(A).
Proof. Let I be a left ideal of A, and assume all elements in I are nilpotent. It is
enough to show that for all x ∈ I the element 1 + x is left-invertible. (If a ∈ A, then
ax ∈ I.) Since x is nilpotent, we can define
X
z= (−1)i xi = 1 − x + x2 − x3 + · · · .
i≥0
We get (1 + x)z = 1 = z(1 + x). The left-right symmetry shows that every right
ideal, which consists only of nilpotent elements is contained in the radical.
Warning: Nilpotent elements do not have to belong to the radical, as the following
example shows: Let A = M2 (K). Then A is a semisimple algebra, thus J(A) = 0.
But of course A contains many nilpotent elements, for example

0 1
x= .
0 0
But observe that there are elements y in Ax which are not nilpotent. In other words
1 + y is not invertible. For example

0 0
e=
0 −1
is in Ax and 1 + e is not invertible. We can also construct maximal left ideals of A
which do not contain x.
Proposition 14.6. Let a ∈ A. Then a ∈ J(A) if and only if aS = 0 for all simple
A-modules S.
Proof. Let T be a simple A-module, and let x be a non-zero element in T . The map
f : A A → T defined by f (b) = bx is an A-module homomorphism. Since x 6= 0, we
have f 6= 0. Since T is simple, f is surjective and the kernel of f is a maximal left
ideal. It follows from the definition of J(A) that J(A) is contained in the kernel of
f . Thus J(A)x = 0, and therefore J(A)T = 0.
Vice versa, assume aS = 0 for all simple A-modules S. We assume that a does not
belong to J(A). Since J(A) is the intersection of all maximal left ideals, there exists
a maximal left ideal I with a ∈/ I. We know that SI := A A/I is a simple A-module.

For b ∈ A set b = b + I. It follows that a 6= 0. Since
a · 1 = a · 1 = a 6= 0
we have aSI 6= 0. This contradiction shows that a ∈ J(A).
In other words, the radical J(A) is the intersection of the annihilators of the simple
A-modules. (Given an A-module M, the annihilator of M in A is the set of all
a ∈ A such that aM = 0.)
Corollary 14.7. For every A-module M we have J(A)M ⊆ rad(M).
Proof. If M ′ is a maximal submodule of M, then M/M ′ is a simple A-module, thus

J(A)(M/M ′ ) = 0. This implies J(A)M ⊆ M ′ . Since J(A)M is contained in all
maximal submodules of M, it is also contained in the intersection of all maximal
submodules of M.
Warning: In general, we do not have an equality J(A)M = rad(M): Let A = K[T ].

Then J(K[T ]) = 0, and therefore J(K[T ])M = 0 for all K[T ]-modules M. But for
the K[T ]-module N(2) we have rad(N(2)) ∼= N(1) 6= 0.
Corollary 14.8. If M is a finitely generated A-module, M ′ is a submodule of M
and M ′ + J(A)M = M, then M ′ = M.
Proof. Assume M is finitely generated and M ′ is a submodule of M with M ′ +

J(A)M = M. By Corollary 14.7 we know that J(A)M ⊆ rad(M). Thus M ′ +
rad(M) = M. Since M is finitely generated Corollary 7.17 implies that rad(M) is
small in M. Thus we get M ′ = M.
Corollary 14.9 (Nakayama Lemma). If M is a finitely generated A-module such
that J(A)M = M, then M = 0.
Proof. In Corollary 14.8 take M ′ = 0.

Lemma 14.10. The algebra A is a local ring if and only if A/J(A) is a skew field.
Proof. If A is a local ring, then J(A) is a maximal left ideal. Thus A/J(A) is a ring
which contains only one proper left ideal, namely the zero ideal. Thus A/J(A) is a
skew field.
Vice versa, if A/J(A) is a skew field, then J(A) is a maximal left ideal. We have to
show that J(A) contains every proper left ideal: Let L be a left ideal, which is not
contained in J(A). Thus J(A) + L = A. Now J(A) = rad(A A) is a small submodule
of A A, since A A is finitely generated. Thus L = A.
Theorem 14.11. If A/J(A) is semisimple, then for all A-modules M we have
J(A)M = rad(M).
Proof. We have seen that J(A)M ⊆ rad(M). On the other hand, M/J(A)M is
annihilated by J(A), thus it is an A/J(A)-module. Since A/J(A) is semisimple, each
A/J(A)-module is a semisimple A/J(A)-module, thus also a semisimple A-module.
But if M/J(A)M is semisimple, then rad(M) has to be contained in J(A)M.
Examples: If A = K[T ], then A/J(A) = A/0 = A. So A/J(A) is not semisimple.

If A is an algebra with l(A A) < ∞ (for example if A is finite-dimensional), then
A/J(A) = A A/ rad(A A) is semisimple.
Lemma 14.12. If e is an idempotent in A, then J(eAe) = eJ(A)e = J(A) ∩ eAe.
Proof. We have J(A) ∩ eAe ⊆ eJ(A)e, since x ∈ eAe implies x = exe. Thus, if
additionally x ∈ J(A), then x = exe belongs to eJ(A)e.
Next we show that eJ(A)e ⊆ J(eAe): Let x ∈ J(A). If a ∈ A, then 1 + eae · x · e
is invertible, thus there exists some y ∈ A with y(1 + eaexe) = 1. This implies
eye(e + eaexe) = ey(1 + eaexe)e = e. Thus all elements in e + eAe(exe) are left-
invertible. This shows that exe belongs to J(eAe).
Finally, we show that J(eAe) ⊆ J(A) ∩ eAe: Clearly, J(eAe) ⊆ eAe, thus we have
to show J(eAe) ⊆ J(A). Let S be a simple A-module. Then eS = 0, or eS is a
simple eAe-module. Thus J(eAe)eS = 0, and therefore J(eAe)S = 0, which implies
J(eAe) ⊆ J(A).
14.2. Exercises. 1: Let Q be a quiver. Show that the radical J(KQ) has as a basis
the set of all paths from i to j such that there is no path from j to i, where i and j
run through the set of vertices of Q.
————————————————————————————-
15. Quivers and path algebras
Path algebras are an extremely important class of algebras. In fact, one of our main
aims is to obtain a better understanding of their beautiful representation theory and
also of the numerous links between representation theory of path algebras and other
areas of mathematics.
Several parts of this section are taken from Crawley-Boevey’s excellent lecture notes
on representation theory of quivers.
15.1. Quivers and path algebras. Recall: A quiver is a quadruple

Q = (Q0 , Q1 , s, t)
where Q0 and Q1 are finite sets, and s, t : Q1 → Q0 are maps. The elements in Q0
are the vertices of Q, and the elements in Q1 the arrows. For an arrow a ∈ Q1
we call s(a) the starting vertex and t(a) the terminal vertex of a.
Thus we can think of Q as a finite directed graph. But note that multiple arrows
and loops (a loop is an arrow a with s(a) = t(a)) are allowed.
Let Q = (Q0 , Q1 , s, t) be a quiver. A sequence
a = (a1 , a2 , . . . , am )
of arrows ai ∈ Q1 is a path in Q if s(ai ) = t(ai+1 ) for all 1 ≤ i ≤ m − 1. Such a path

has length m, we write l(a) = m. Furthermore set s(a) = s(am ) and t(a) = t(a1 ).
Instead of (a1 , a2 , . . . , am ) we often just write a1 a2 · · · am .
Additionally there is a path ei of length 0 for each vertex i ∈ Q0 , and we set
s(ei ) = t(ei ) = i.
The path algebra KQ of Q over K is the K-algebra with basis the set of all paths
in Q. The multiplication of paths a and b is defined as follows:
If a = ei is of length 0, then
(
b if t(b) = i,
ab := a · b :=
0 otherwise.
If b = ei , then
(
a if s(a) = i,
ab := a · b :=
0 otherwise.
Finally, assume that a = (a1 , . . . , al ) and b = (b1 , . . . , bm ) are paths of length l, m ≥

1. Then
(
(a1 , . . . , al , b1 , . . . , bm ) if s(al ) = t(b1 ),
ab := a · b :=
0 else.
15.2. Examples. 1: Let Q be the following quiver:
a b
2o 4 / 5
c d e
f
1 / 3
Then the paths in Q are
e1 , e2 , e3 , e4 , e5 , a, b, c, d, e, f, ca, da, f c, f d, f da, f ca.

Thus, KQ is a 17-dimensional K-algebra. Here are some examples of multiplications

of paths:
e3 · e4 = 0,
f c · a = f ca,
a · f c = 0,
b · e4 = b,
e4 · b = 0,
e5 · b = b.
The algebra KQ has a unit element, namely 1 := e1 + e2 + e3 + e4 + e5 .
2: Let Q be the quiver

Q: 8 1
Then KQ is isomorphic to the polynomial ring K[T ] in one variable T .
Q: 8 1f
Then KQ is isomorphic to the free algebra KhX, Y i in two non-commuting variables
X and Y .
15.3. Idempotents in path algebras. Let A = KQ for some quiver Q, and

assume that Q0 = {1, . . . , n}.
Then the ei are orthogonal idempotents, in other words e2i = ei and ei ej = 0 for all
i 6= j. Clearly,
Xn
1= ei
i=1
is the identity of A. The vector spaces Aei and ej A have as bases the set of paths
starting in i and the set of paths ending in j, respectively. Furthermore, ej Aei has
as a basis the set of paths starting in i and ending in j. We have
Mn
A= Aei .
i=1
Clearly, each Aei is a left A-module. So this is a direct decomposition of the regular
representation A A.
Lemma 15.1. If 0 6= x ∈ Aei and 0 6= y ∈ ei A, then xy 6= 0.
Proof. Look at the longest paths p and q involved in x and y, respectively. In the
product xy the coefficient of pq cannot be zero.
Lemma 15.2. The ei are primitive idempotents.
Proof. If EndA (Aei ) ∼

= (ei Aei )op contains an idempotent f , then f 2 = f = f ei . This
implies f (ei − f ) = 0. Now use Lemma 15.1.
Corollary 15.3. The A-modules Aei are indecomposable.
Proof. The only idempotents in EndA (Aei ) are 0 and 1.

Lemma 15.4. If ei ∈ Aej A, then i = j.
Proof. The vector space Aej A has as a basis the paths passing through the vertex
j.
Lemma 15.5. If i 6= j, then Aei ∼
6= Aej .
Proof. Assume i 6= j and that there exists an isomorphism f : Aei → Aej . Set
y = f (ei ). It follows from Lemma 12.10 that y ∈ ei Aej . Let g = f −1 , and let
x = g(ej ). This implies
(gf )(ei) = g(y) = g(yej ) = yg(ej ) = yx = ei .
A similar calculation shows that xy = ej . But y ∈ ei Aej and x ∈ Aei . Thus
y = ei yej and x = xei . This implies ej = xy = xei yej ∈ Aei A, a contradiction to
Lemma 15.4.
15.4. Representations of quivers. A representation (or more precisely a K-

representation) of a quiver Q = (Q0 , Q1 , s, t) is given by a K-vector space Vi for
each vertex i ∈ Q0 and a linear map
Va : Vs(a) → Vt(a)
for each arrow a ∈ Q1 . Such a representation is called finite-dimensional if Vi is
finite-dimensional for all i. In this case,
X
dim V := dim Vi
i∈Q0
is the dimension of the representation V .
For a path p = (a1 , . . . , am ) of length m ≥ 1 in Q, define

Vp := Va1 ◦ Va2 ◦ · · · ◦ Vam : Vs(p) → Vt(p) .
A morphism
θ: V → W
between representations V = (Vi , Va )i,a and W = (Wi , Wa )i,a is given by linear maps
θi : Vi → Wi , i ∈ Q0 such that the diagram
θs(a)
Vs(a) / Ws(a)
Va Wa
θt(a)
Vt(a) / Wt(a)
commutes for each a ∈ Q1 . The vector space of homomorphisms from V to W is
denoted by Hom(V, W ), or more precisely by HomQ (V, W ).
A morphism θ = (θi )i : V → W is an isomorphism if each θi is an isomorphism.

In this case, we write V ∼
= W.
The composition ψ ◦ θ of two morphisms θ : V → W and ψ : W → X is given by
(ψ ◦ θ)i = ψi ◦ θi .
The K-representations form a K-category denoted by Rep(Q) = RepK (Q). The full
subcategory of finite-dimensional representations is denoted by rep(Q) = repK (Q).
A subrepresentation of a representation (Vi , Va )i,a is given by a tuple (Ui )i of

subspaces Ui of Vi such that
Va (Us(a) ) ⊆ Ut(a)
for all a ∈ Q1 . In this case, we obtain a representation (Ui , Ua )i,a where Ua : Us(a) →
Ut(a) is defined by Ua (u) = Va (u) for all u ∈ Us(a) .
The direct sum of representations V = (Vi , Va )i,a and W = (Wi , Wa )i,a is defined
in the obvious way, just take V ⊕ W := (Vi ⊕ Wi , Va ⊕ Wa )i,a .
Now we can speak about simple representations and indecomposable repre-

sentations. (As for modules, it is part of the definition of a simple and of an
indecomposable representation V that V 6= 0.)
15.5. Examples. 1: For i ∈ Q0 let Si be the representation with

(
K if i = j,
(Si )j =
0 else.
for all j ∈ Q0 , and set (Si )a = 0 for all a ∈ Q1 . Obviously, Si is a simple represen-
tation.
2: For λ ∈ K let Vλ be the representation
λ /
K K
of the quiver 1 2 . Then Vλ ∼
/ = Vµ if and only if λ = 0 = µ or λ 6= 0 6= µ. We
have 

0 if λ 6= 0 and µ = 0,
dim Hom(Vλ , Vµ ) = 1 if µ 6= 0,

2 if λ = 0 = µ.
3: For λ1 , λ2 let Vλ1 ,λ2 be the representation

λ1
/
K / K
λ2
/ 2 . Then Vλ1 ,λ2 ∼

/
of the quiver 1 = Vµ1 ,µ2 if and only if there exists some c 6= 0
with c(λ1 , λ2 ) = (µ1 , µ2 ): Assume there exists an isomorphism
θ = (θ1 , θ2 ) : Vλ1 ,λ2 → Vµ1 ,µ2 .
Thus θ = (a, b) for some a, b ∈ K ∗ . We obtain a diagram

a /
K K
λ1 λ2 µ1 µ2
b

K / K
satisfying bλ1 = µ1 a and bλ2 = µ2 a. Set c = a−1 b. It follows that c(λ1 , λ2 ) =
(µ1 , µ2 ).
4: For λ ∈ K let Vλ be the representation
λ 5 K
of the 1-loop quiver. Then Vλ ∼
= Vµ if and only if λ = µ.
5: Let V be the representation
[ 10 ]
/
K / K2
[ 01 ]
/
of the quiver 1 / 2 . The subrepresentations of V are (K, K 2 ) and (0, U) where
U runs through all subspaces of K 2 . It is easy to check that none of these subrepre-
sentations is a direct summand of V . Thus V is an indecomposable representation.
15.6. Representations of quivers and modules over path algebras. Let V =

(Vi , Va )i,a be a representation. Let
M M
η : KQ × Vi → Vi
i∈Q0 i∈Q0
be the map defined by

( (
vi if i = j, Vp (vi ) if i = s(p),
η(ej , vi ) = and η(p, vi ) =
0 otherwise 0 otherwise
where the ej are the paths of length 0, p runs through the set of paths of length at
least one and vi ∈ Vi . Then we extend these rules linearly.
Vice versa, let V be a KQ-module, i.e. there is a KQ-module structure

η : KQ × V → V
on the K-vector space V . For each path ei of length 0 define Vi := ei V , which is
clearly a K-vectorspace. It follows that
M
V = Vi .
i∈Q0
For each arrow a ∈ Q1 define a linear map

Va : Vs(a) → Vt(a)
by Va (v) := η(a, v) for all v ∈ Vs(a) . Then (Vi , Va )i,a is obviously a representation of
Q.
We leave it as an exercise to show that these constructions yield equivalences of

K-categories between RepK (Q) and Mod(KQ).
So from now on we can use all the terminology and the results we obtained for
modules over algebras also for representations of quivers. In particular, we get a
Jordan-Hölder and a Krull-Remak-Schmidt Theorem, we can ask for Auslander-
Reiten sequences of quiver representations, etc. We will often not distinguish any
longer between a representation of Q and a module over KQ.
If V = (Vi , Va )i,a is a representation of Q let
d = dim(V ) = (dim Vi )i∈Q0
be its dimension vector. If V is a finite-dimensional indecomposable representa-
tion, then dim(V ) ∈ NQ0 is called a root of Q. A root d is a Schur root if there
exists a representation V with EndQ (V ) ∼ = K and dim(V ) = d. Assume that d is
a root. If there exists a unique (up to isomorphism) indecomposable representation
V with dim(V ) = d, then d is called a real root. Otherwise, d is an imaginary
root.
A representation V of Q is rigid (or exceptional) if each short exact sequence
0→V →W →V →0
splits, i.e. if W ∼
= V ⊕V.
Here are some typical problems appearing in representation theory of quivers:
(i) Classify all indecomposable representations of Q. (This is usually very hard

and can only be achieved for very few quivers.)
(ii) Determine all roots of Q. Determine the real roots and the Schur roots of
Q.
(iii) Classify all rigid representations of Q.
(iv) Compute the Auslander-Reiten quiver of mod(KQ), or at least try to de-
scribe the shape of its connected components.
(v) How does the representation theory of a quiver Q change, if we change the
orientation of an arrow of Q?
15.7. Exercises. 1: Let Q be a quiver. Show that KQ is finite-dimensional if and

only if Q has no oriented cycles.
1 1 1
2: Let V = (K ←−K − → K) and W = (K ← −K− → 0) be representations of the quiver
1←−2− → 3. Show that HomQ (V, W ) is one-dimensional, and that HomQ (W, V ) = 0.
1−
→2−
→ ··· −
→ n.
Show that KQ is isomorphic to the subalgebra
A := {M ∈ Mn (K) | mij = 0 if there is no path from j to i}
of Mn (K).
4: Let Q be any quiver. Determine the centre of KQ. (Reminder: The centre
C(A) of an algebra A is defined as C(A) = {a ∈ A | ab = ba for all b ∈ A}.)
5: Let Q be a quiver with n vertices. Show that there are n isomorphism classes of
simple KQ-modules if and only if Q has no oriented cycles.
6: Let Q be a quiver. Show that the categories RepK (Q) and Mod(KQ) are equiv-
alent.
7: Construct an indecomposable representation of the quiver

◦

◦ / ◦ / ◦o ◦o ◦
with dimension vector
1
1 2 3 2 1
8: Show: If V = (Vi , Va )i∈Q0 ,a∈Q1 is an indecomposable representation of the quiver

Q: ◦ / ◦ / ◦ / ◦ / ◦
then dim Vi ≤ 1 for all i ∈ Q0 .
Construct the Auslander-Reiten quiver of Q.
9: Let Q be the following quiver:

◦

◦ / ◦ / ◦ / ◦ / ◦ / ◦
Let A = KQ. Write A A as a direct sum of indecomposable representations and
compute the dimension of the indecomposable direct summands.
10: Let
K[T ]/(T 2 ) 0
A= .
K[T ]/(T 2 ) K
This gives a K-algebra via the usual matrix multiplication. (The elements of A are
of the form
a 0
b c
where a, b ∈ K[T ]/(T 2 ) and c ∈ K.) Show that A is isomorphic to KQ/I where Q
is the quiver
9◦
α /◦
and I is the ideal in KQ generated by the path α2 := (α, α).
————————————————————————————
16. Digression: Classification problems in Linear Algebra
Many problems in Linear Algebra can be reformulated using quivers. In this section,
we give some examples of this kind.
16.1. Classification of endomorphisms.

Q: 8 1
Let K be an algebraically closed field, and let V be a finite-dimensional K-vector

space of dimension n. By EndK (V ) we denote the set of K-linear maps V → V ,
and by G = GL(V ) the set of invertible K-linear maps V → V . For f ∈ EndK (V )
let
Gf = {g −1f g | g ∈ G} ⊆ EndK (V )
be the G-orbit of f . One easily checks that for f1 , f2 ∈ EndK (V ) we have either
Gf1 = Gf2 or Gf1 ∩ Gf2 = ∅.
Question 16.1. Can we classify all G-orbits?
Answer: Of course we can, since we paid attention in our Linear Algebra lectures.
For n = 0 everything is trivial, there is just one orbit containing only the zero
map. Thus assume n ≥ 1. Fix a basis B of V . Now each map f ∈ EndK (V ) is
(with respect to B) given by a particular matrix, which (via conjugation) can be
transformed to a Jordan normal form. It follows that each orbit Gf is uniquely
determined by a set
{(n1 , λ1 ), (n2 , λ2 ), . . . , (nt , λt )}
where the ni are positive integers with n1 + · · · + nt = n, and the λi are elements in
K. Here (ni , λi ) stands for a Jordan block of size ni with Eigenvalue λi .
16.2. Classification of homomorphisms.

Q: 1 / 2
Let K be any field, and let V1 and V2 be finite-dimensional K-vector spaces of

dimension n1 and n2 , respectively. By HomK (V1 , V2 ) we denote the set of K-linear
maps V1 → V2 , and let G = GL(V1 ) × GL(V2 ). For f ∈ HomK (V1 , V2 ) let
Gf = {h−1 f g | (g, h) ∈ G} ⊆ HomK (V1 , V2 )
be the G-orbit of f .
Answer: Of course we can. This is even easier than the previous problem: Fix
bases B1 and B2 of V1 and V2 , respectively. Then each f ∈ HomK (V1 , V2 ) is given
by a matrix with respect to B1 and B2 . Now using row and column transformations
(which can be expressed in terms of matrix multiplication from the left and right)
we can transform the matrix of f to a matrix of the form

Er 0
0 0
where Er is the r × r-unit matrix and the zeros are matrices with only zero entries.
Here r is the rank of the matrix of f .
It turns out that there are 1+min{n1 , n2 } different G-orbits where ni is the dimension
of Vi .
16.3. The Kronecker problem.

Q: //
1 2
Let K be an algebraically closed field, and let V1 and V2 be finite-dimensional

K-vector spaces. Let G = GL(V1 ) × GL(V2 ). For (f1 , f2 ) ∈ HomK (V1 , V2 ) ×
HomK (V1 , V2 ) let
G(f1 , f2 ) = {(h−1 f1 g, h−1f2 g) | (g, h) ∈ G} ⊆ HomK (V1 , V2 ) × HomK (V1 , V2 )
be the G-orbit of (f1 , f2 ).
Answer: Yes, we can do that, by we will need a bit of theory here. As you can
see, the problem became more complicated, because we simultaneously transform
the matrices of f1 and f2 with respect to some fixed bases of V1 and V2 .
1 λ 1 µ
Example: The orbits G(K −
→ K, K −
→ K) and G(K −
→ K, K −
→ K) are equal if
and only if λ = µ.
16.4. The n-subspace problem.

Q: 1 FF 2 ··· x n
FF xx
FF xx
FF x
FF xxx
# {x
0
An n-subspace configuration is just an n + 1-tuple (V, V1 , . . . , Vn ) where V is a vector

space and the Vi are subspaces of V . We call
dim(V, V1 , . . . , Vn ) = (dim V, dim V1 , . . . , dim Vn )
the dimension vector of the n-subspace configuration (V, V1 , . . . , Vn ).
We say that two n-subspace configurations (V, V1 , . . . , Vn ) and (W, W1 , . . . , Wn ) are
isomorphic if there exists an isomorphism (= bijective linear map) f : V → W such
that the following hold:
• f (Vi ) ⊆ Wi ;
• The linear maps fi : Vi → Wi defined by fi (vi ) = f (vi ) where 1 ≤ i ≤ n and

vi ∈ Vi are isomorphisms.
In particular, two isomorphic n-subspace configurations have the same dimension

vector.
Problem 16.4. Classify all n-subspace configurations up to isomorphism.
We can reformulate this problem as follows: Let V, V1 , . . . , Vn be vector spaces such

that dim Vi ≤ dim V for all i. Set
Z = Inj(V1 , V ) × · · · × Inj(Vn , V )
where Inj(Vi , V ) denotes the set of injective linear maps from Vi → V . Let G =
GL(V ) × GL(V1 ) × · · · × GL(Vn ). Each element (f1 , . . . , fn ) can be thought of as an
n-subspace configuration given by (V, Im(f1 ), . . . , Im(fn )).
Then G acts on Z as follows: For (f1 , . . . , fn ) and g = (g0 , g1 , . . . , gn ) ∈ G define
g · (f1 , . . . , fn ) = (g0−1f1 g1 , . . . , g0−1fn gn )
and let
G(f1 , . . . , fn ) = {g · (f1 , . . . , fn ) | g ∈ G}
be the G-orbit of (f1 , . . . , fn ). Classifying all n-subspace configurations with dimen-
sion vector (dim V, dim V1 , . . . , dim Vn ) up to isomorphism corresponds to classifying
the G-orbits in Z.
It turns out that Problem 16.4 is much too hard for large n. But for small n one
can solve it.
Given two n-subspace configurations (V, V1 , . . . , Vn ) and (W, W1 , . . . , Wn ), we define
their direct sum by
(V, V1 , . . . , Vn ) ⊕ (W, W1 , . . . , Wn ) = (V ⊕ W, V1 ⊕ W1 , . . . , Vn ⊕ Wn ).
It follows that (V, V1 , . . . , Vn )⊕(W, W1 , . . . , Wn ) is again an n-subspace configuration.
An n-subspace configuration (V, V1 , . . . , Vn ) is indecomposable if it is not isomor-
phic to the direct sum of two non-zero n-subspace configurations. (We say that an
n-subspace configuration (V, V1 , . . . , Vn ) is zero, if V = 0.)
One can prove that any n-subspace configuration can be written (in a “unique way”)
as a direct sum of indecomposable n-subspace configurations. Thus to classify all
n-subspace configurations, it is enough to classify the indecomposable ones.
We will see for which n there are only finitely many indecomposable n-subspace
configurations.
Instead of asking for the classification of all n-subspace configurations, we might ask
the following easier question:
Problem 16.5. Classify the dimension vectors of the indecomposable n-subspace
configurations.
It turns out that there is a complete answer to Problem 16.5.
16.5. Exercises. 1: Classify all indecomposable 3-subspace configurations. Does

the result depend on the field K?
2: Solve the Kronecker problem as described above for V1 = V2 = K 2 where K is
an algebraically closed field.
3: Find the publication of Kronecker where he solves the Kronecker problem.
***********************************************************************
Part 4. Projective modules
17. Projective modules
In this section let A be a K-algebra. As before, let Mod(A) be the category of left
A-modules, and let mod(A) be the full subcategory of finitely generated A-modules.
17.1. Definition and basic properties. An A-module P is called projective if

for any epimorphism g : M → N and any homomorphism h : P → N there exists a
homomorphism h′ : P → M such that g◦h′ = h. This is called the “lifting property”.
M
}>
h′ } g
}
} h

P / N
An A-module P is projective if and only if for every epimorphism g : M → N of
A-modules the induced map
HomA (P, g) : HomA (P, M) → HomA (P, N)
is surjective. For every A-module X, the functor HomA (X, −) is left exact. Thus a
module P is projective if and only if HomA (P, −) is exact.
(A functor F : Mod(A) → Mod(B) is exact if for every short exact sequence 0 →

U → V → W → 0 in Mod(A) the sequence 0 → F (U) → F (V ) → F (W ) → 0 is
exact in Mod(B).)
Recall that an A-module F is free if F is isomorphic to a direct sum of copies of
the regular representation A A.
Lemma 17.1. Free modules are projective.
Proof. Let F be a free A-module with free generating set X. Let g : M → N be an

epimorphism of A-modules, and let h : F → N be any homomorphism of A-modules.
For every x ∈ X we look at the image h(x). Since g is surjective there exists some
mx ∈ M with g(mx ) = h(x). Define a homomorphism h′ : F → M by h′ (x) = mx .
Since X is a free generating set of X, there exists exactly one such homomorphism
h′ . For every x ∈ X we have (g ◦ h′ )(x) = h(x), and this implies gh′ = h, since X is
a generating set of F .
The map h′ constructed in the proof of the above lemma is in general not uniquely
determined. There can be many different maps h′ with the property gh′ = h:
For example, for F = A A the set {1A } is a free generating set. Let M be an
A-module and let U be a submodule of M. By g : M → M/U we denote the
corresponding projection map. This is a typical epimorphism. For x ∈ M let x =
x+U be the corresponding residue class in M/U. Let h : A A → M/U be an arbitrary
homomorphism. Then h(1A ) = x for some x ∈ M. Now the homomorphisms
h′ : A A → M such that gh′ = h correspond to the elements in x = x + U = {x + u |
u ∈ U}, namely we can take any x + u ∈ x + U and then define h′ (1A ) = x + u.

Thus if U 6= 0, then there are many such homomorphisms h′ .
Lemma 17.2. A direct sum of modules is projective if and only if each direct sum-
mand is projective.
L
Proof. Let g : M → N be an epimorphism. First let P = i∈I Pi with Pi projective
for all I. For a homomorphism h : P → N let hi : Pi → N be its restriction to Pi .
For every hi there exists aLlifting, i.e. there exists a homomorphism h′i : Pi → M
with gh′i = hi . Define h′ : ′
i∈I Pi → M such that the restriction of h to Pi is just
h′i . This implies gh′ = h.
Vice versa, let P be a projective module, and let P = P1 ⊕ P2 be a direct decompo-
sition of P . For a homomorphism h1 : P1 → N let h = [h1 , 0] : P1 ⊕ P2 → N. Since
P is projective, there exists a homomorphism h′ : P → M with gh′ = h. We can
write h′ = [h′1 , h′2 ] with h′i : Pi → M. It follows gh′1 = h1 .
Lemma 17.3. For a module P the following are equivalent:
(i) P is projective;
(ii) Every epimorphism M → P splits;
(iii) P is isomorphic to a direct summand of a free module.
Furthermore, a projective module P is a direct summand of a free module of rank c

if and only if P has a generating set of cardinality c.
Proof. (i) =⇒ (ii): Let M be an A-module, and let g : M → P be an epimorphism.

For the identity map 1P : P → P the lifting property gives a homomorphism h′ : P →
M with g ◦ h′ = 1P . Thus g is a split epimorphism.
(ii) =⇒ (iii): There exists an epimorphism f : F → P where F is a free A-module.
Since f splits, P is isomorphic to a direct summand of a free module.
(iii) =⇒ (i): The class of projective modules contains all free modules and is closed
under direct summands.
Now we proof the last statement of the Lemma: If P has a generating set X of
cardinality c, then let F be a free module of rank c. Thus F has a generating set of
cardinality c. We get an epimorphism F → P which has to split.
Vice versa, if P is a direct summand of a free module F of rank c, then P has a
generating set of cardinality c: We choose an epimorphism f : F → P , and if X is
a generating set of F , then f (X) is a generating set of P .
Thus an A-module P is finitely generated and projective if and only if P is a direct

summand of a free module of finite rank.
Let Proj(A) be the full subcategory of Mod(A) of all projective A-modules, and set
proj(A) := Proj(A) ∩ mod(A).
Warning: There exist algebras A such that A A is isomorphic to A A ⊕ A A. Thus

the free module A A has rank n for any positive integer n. For example, take as A
the endomorphism algebra of an infinite dimensional vector space.
Corollary 17.4. If P and Q are indecomposable projective modules, and if p : P →
Q an epimorphism, then p is an isomorphism.
Proof. Since Q is projective, p is a split epimorphism. But P is indecomposable and

Q 6= 0. Thus p has to be an isomorphism.
17.2. The radical of a projective module. Recall that a submodule U of a

module M is small in M if U + U ′ ⊂ M for all proper submodules U ′ of M.
As before, by J(A) we denote the radical of an algebra A.
Lemma 17.5. If P is a projective A-module, then
J(EndA (P )) = {f ∈ EndA (P ) | Im(f ) is small in P }.
Proof. Let J := J(EndA (P )). Let f : P → P be an endomorphism such that the

image Im(f ) is small in P . If g ∈ EndA (P ) is an arbitrary endomorphism, then
Im(f g) ⊆ Im(f ), thus Im(f g) is also small in P . Clearly, we have
P = Im(1P ) = Im(1P + f g) + Im(f g).
Since Im(f g) is small, we get that 1P + f g is surjective. But P is projective,
therefore 1P + f g is a split epimorphism. Thus there exists some h ∈ EndA (P ) with
(1P + f g)h = 1P . We have shown that the element 1P + f g has a right inverse for
all g ∈ EndA (P ). Thus f belongs to J.
Vice versa, assume f ∈ J. Let U be a submodule of P with P = Im(f ) + U, and let
p : P → P/U be the projection. Since P = Im(f )+U, we know that pf is surjective.
But P is projective, therefore there exists some p′ : P → P with p = pf p′ . Now
1P − f p′ is invertible, because f ∈ J. Since p(1P − f p′ ) = 0 we get p = 0. This
implies U = P . It follows that Im(f ) is small in P .
Corollary 17.6. Let P be a projective A-module. If rad(P ) is small in P , then
J(EndA (P )) = {f ∈ EndA (P ) | Im(f ) ⊆ rad(P )}.
Proof. Each small submodule of a module M is contained in rad(M). If rad(M)

is small in M, then the small submodules of M are exactly the submodules of
rad(M).
Lemma 17.7. If P is a projective A-module, then rad(P ) = J(A)P .
Proof. By definition J(A) = rad(A A) and J(A)A = J(A). This shows that the
statement is true for P = A A. Now let Mi , i ∈ I be a family of modules. We have
! !
M M M M
J(A) Mi = J(A)Mi and rad Mi = rad(Mi ).
i∈I i∈I i∈I i∈I
We know that J(A)M ⊆ rad(M) for all modules M. Thus we get

! !
M M M M
J(A) Mi = J(A)Mi ⊆ rad(Mi ) = rad Mi .
i∈I i∈I i∈I i∈I
L i ⊂ rad(M
This is a proper inclusion only if there existsLsome i with J(A)M i ). Thus,
if J(A)Mi = rad(Mi ) for all i, then J(A) i∈I M i = rad i∈I Mi . This shows
that the statement is true for free modules.
L L
Vice versa, if J(A) i∈I Mi = rad i∈I M i , then J(A)Mi = rad(Mi ) for all i.
Since projective modules are direct summands of free modules, and since we proved
the statement already for free modules, we obtain it for all projective modules.
Lemma 17.8. Let U be a submodule of a projective module P such that for every
endomorphism f of P we have f (U) ⊆ U. Define
f∗ : P/U → P/U
by f∗ (x + U) := f (x) + U. Then the following hold:
(i) f∗ is an endomorphism of P/U;

(ii) The map f 7→ f∗ defines a surjective algebra homomorphism
EndA (P ) → EndA (P/U)
with kernel {f ∈ EndA (P ) | Im(f ) ⊆ U}.
Proof. Let p : P → P/U be the projection. Thus f∗ is defined via p ◦ f = f∗ ◦ p. It

is easy to show that this is really an A-module homomorphism, and that f 7→ f∗
defines an algebra homomorphism. The description of the kernel is also obvious.
It remains to show the surjectivity: Here we use that P is projective. If g is an

endomorphism of P/U, there exists a lifting of g ◦ p : P → P/U. In other words
there exists a homomorphism g ′ : P → P such that p ◦ g ′ = g ◦ p.
l5 P
g′ l l l
l p
l l l
l l
P / P/U / P/U
p g
Thus we get g∗′ = g.
Let M be an A-module. For all f ∈ EndA (M) we have f (rad(M)) ⊆ rad(M). Thus
the above lemma implies that for any projective module P there is a surjective
algebra homomorphism EndA (P ) → EndA (P/ rad(P )), and the kernel is the set of
all endomorphisms of P whose image is contained in rad(P ).
We have shown already: If rad(P ) is a small submodule of P , then the set of all
endomorphisms of P whose image is contained in rad(P ) is exactly the radical of
EndA (P ). Thus, we proved the following:
Corollary 17.9. Let P be a projective A-module. If rad(P ) is small in P , then

EndA (P )/J(EndA (P )) ∼
= EndA (P/ rad(P )).
17.3. Cyclic projective modules.

Lemma 17.10. Let P be an A-module. Then the following are equivalent:
(i) P is cyclic and projective;

(ii) P is isomorphic to a direct summand of A A;
(iii) P is isomorphic to a module of the form Ae for some idempotent e ∈ A.
Proof. We have shown before that a submodule U of A A is a direct summand of A A

if and only if there exists an idempotent e ∈ A such that U = Ae.
If e is any idempotent in A, then A A = Ae ⊕ A(1 − e) is a direct decomposition.

Thus Ae is a direct summand of A A, and Ae is projective and cyclic.
Vice versa, let U be a direct summand of A A, say A A = U ⊕ U ′ . Write 1 = e + e′
with e ∈ U and e′ ∈ U ′ . This implies U = Ae and U ′ = Ae′ , and one checks easily
that e, e′ form a complete set of orthogonal idempotents.
Lemma 17.11. Let P be a projective A-module. If P is local, then EndA (P ) is a
local ring.
Proof. If P is local, then P is obviously cyclic. A cyclic projective module is of the

form Ae for some idempotent e ∈ A, and its endomorphism ring is (eAe)op . We
have seen that Ae is a local module if and only if eAe is a local ring. Furthermore,
we know that eAe is a local ring if and only if (eAe)op is a local ring.
The converse of Lemma 17.11 is also true. We will not use this result, so we skip
the proof.
17.4. Projective covers. Let M be an A-module. A homomorphism p : P → M

is a projective cover of M if the following hold:
• P is projective;
• p is an epimorphism;
• Ker(p) is a small submodule of P .
In this situation one often calls the module P itself a projective cover of M and
writes P = P (M).
Lemma 17.12. Let P be a finitely generated projective module. Then the projection
map P → P/ rad(P ) is a projective cover.
Proof. The projection map is surjective and its kernel is rad(P ). By assumption P
is projective. For every finitely generated module M the radical rad(M) is a small
submodule of M. Thus rad(P ) is small in P .
Warning: If P is an arbitrary projective A-module, then rad(P ) is not necessarily

small in P : For example, let A be the subring of K(T ) consisting of all fractions
of the form f /g such that g is not divisible by T . This is a local ring. Now let
P be a free A-module of countable rank, for example the module of all sequences
(a0 , a1 , . . .) with ai ∈ A for all i such that only finitely many of the ai are non-zero.
The radical U = rad(P ) consists of all such sequences with ai divisible by T for all
i. We define a homomorphism f : P → A K(T ) by
X a1 a2
f (a0 , a1 , . . .) = T −i ai = a0 + + 2 +··· .
i≥0
T T
Let W be the kernel of f . Since f 6= 0, W is a proper submodule of P . On the
other hand we will show that U + W = P . Thus U = rad(P ) is not small in P . Let
a = (a0 , a1 , . . .) be a sequence in P and choose n such that aj = 0 for all j > n.
Define b = (b0 , b1 , . . .) by
n
X
bn+1 = T n−i+1ai = a0 T n+1 + a1 T n + · · · + an T
i=0
and bj = 0 for all j 6= n + 1. Since b belongs to T A, we know that b is in U. On the

other hand f (b − a) = 0, thus b − a ∈ W . We see that a = b − (b − a) belongs to
U + W.
Given two projective covers pi : Pi → Mi , i = 1, 2, then the direct sum
p1 ⊕ p2 : P1 ⊕ P2 → M1 ⊕ M2
is a projective cover. Here
p1 0
p1 ⊕ p2 = .
0 p2
The map p1 ⊕ p2 is obviously an epimorphism and its kernel is Ker(p1 ) ⊕ Ker(p2 ).
By assumption Ker(pi ) is small in Pi , thus Ker(p1 ) ⊕ Ker(p2 ) is small in P1 ⊕ P2 .
Warning:
L Given infinitely many projective
L modules Pi with small submodules Ui ,
then i∈I Ui is not necessarily small in i∈I Pi .
Lemma 17.13 (Projective covers are unique). Let p1 : P1 → M be a projective cover,
and let p2 : P2 → M be an epimorphism with P2 projective. Then the following hold:
• There exists a homomorphism f : P2 → P1 such that p1 ◦ f = p2 ;

• Each homomorphism f : P2 → P1 with p1 ◦ f = p2 is a split epimorphism;
• If p2 is also a projective cover, then every homomorphism f : P2 → P1 with
p1 ◦ f = p2 is an isomorphism.
Proof. Since p1 is an epimorphism, and since P2 is projective, there exists a homo-

morphism f with p1 f = p2 .
P
> 1
f }
} p1
}
} p2
P2 / M
We have to show that each such f is a split epimorphism: We show that

Ker(p1 ) + Im(f ) = P1 .
For x ∈ P1 we have p1 (x) ∈ M. Since p2 is surjective, there exists some x′ ∈ P2
such that p1 (x) = p2 (x′ ) = (p1 f )(x′ ). Thus p1 (x − f (x′ )) = 0. We see that x − f (x′ )
belongs to Ker(p1 ), thus x = (x−f (x′ ))+f (x′ ) lies in Ker(p1 )+Im(f ). Now Ker(p1 )
is small in P1 , which implies Im(f ) = P1 . We have shown that f is surjective. But
each epimorphism to a projective module is a split epimorphism.
Now we assume that p2 is also a projective cover. Again let f : P2 → P1 be a homo-
morphism with p1 f = p2 . Since f is a split epimorphism, there exists a submodule
U of P2 with Ker(f ) ⊕ U = P2 . We show that
Ker(p2 ) + U = P2 .
If y ∈ P2 , then there exists some y ′ ∈ P1 with p2 (y) = p1 (y ′). We know that
f : P2 → P1 is surjective, thus y ′ has a preimage in P2 . Since P2 = Ker(f ) ⊕ U, we
can find this preimage in U. Thus there is some u ∈ U with f (u) = y ′. Summarizing,
we get p2 (y) = p1 (y ′) = (p1 f )(u) = p2 (u). We see that y − u belongs to Ker(p2 ),
thus y = (y − u) + u ∈ Ker(p2 ) + U. Since Ker(p2 ) is small in P2 we get U = P2 and
therefore Ker(f ) = 0. Thus f is also injective.
So projective covers are (up to isomorphism) uniquely determined. Also, if p : P →
M is a projective cover, and f : P → P is a homomorphism with p ◦ f = p, then f
is an isomorphism.
Corollary 17.14. Let P and Q be finitely generated projective modules. Then P ∼
=
Q if and only if P/ rad(P ) ∼
= Q/ rad(Q).
Proof. Since P and Q are finitely generated projective modules, the projections
p : P → P/ rad(P ) and q : Q → Q/ rad(Q) are projective covers. If f : P/ rad(P ) →
Q/ rad(Q) is an isomorphism, then f ◦ p : P → Q/ rad(Q) is a projective cover. The
uniqueness of projective covers yields P ∼
= Q. The other direction is obvious.
Corollary 17.15. Let P be a direct sum of local projective modules. If U is a
submodule of P which is not contained in rad(P ), then there exists an indecomposable
direct summand P ′ of P which is contained in U.
L
Proof. Let P = i∈I Pi with local projective modules PiL . Let U be a submodule of
P which is not contained in rad(P ). We have rad(P ) = i∈I rad(Pi ) and therefore
M
P/ rad(P ) = Pi / rad(Pi ).
i∈I
Let u : U → P be the inclusion map, and let p : P → P/ rad(P ) be the projection.

Finally, for every i ∈ I let πi : P/ rad(P ) → Pi / rad(Pi ) also be the projection. The
composition pu is not the zero map. Thus there exists some i ∈ I with πi pu 6= 0.
Since Pi / rad(Pi ) is a simple module, πi pu is surjective. Let pi : Pi → Pi / rad(Pi )
be the projection. Since Pi is a local projective module, pi is a projective cover.
By the surjectivity of πi pu the lifting property of Pi yields an f : Pi → U such that
πi puf = pi . Now we use that pi is an epimorphism: The lifting property of P gives

us a homomorphism g : P → Pi with pi g = πi p.
U
C
u

P
{
f { p
{
{
{
{ P/ rad(P )
{ g
{ { πi
{
}{ pi

Pi / Pi / rad(Pi )
Thus we have
pi guf = πi puf = pi .
Since pi is a projective cover, guf must be an isomorphism. Thus we see that uf is

a split monomorphism whose image P ′ := Im(uf ) is a direct summand of P which
is isomorphic to Pi . Clearly, P ′ as the image of uf is contained in U = Im(f ).
Lemma 17.16. Let P be a finitely generated projective A-module, and let M be

a finitely generated module. For a homomorphism p : P → M the following are
equivalent:
(i) p is a projective cover;

(ii) p is surjective and Ker(p) ⊆ rad(P );
(iii) p induces an isomorphism P/ rad(P ) → M/ rad(M).
Proof. (i) =⇒ (ii): Small submodules of a module are always contained in the
radical.
(ii) =⇒ (iii): Since Ker(p) ⊆ rad(P ) we have rad(P/ Ker(p)) = rad(P )/ Ker(p).
Now p induces an isomorphism P/ Ker(p) → M which maps rad(P/ Ker(p)) onto
rad(M) and induces an isomorphism P/ rad(P ) → M/ rad(M).
(iii) =⇒ (i): We assume that p : P → M induces an isomorphism p∗ : P/ rad(P ) →
M/ rad(M). This implies rad(M) + Im(p) = M. Since M is a finitely generated
module, its radical is a small submodule. Thus Im(p) = M. We see that p is an
epimorphism. Since p∗ is injective, the kernel of p must be contained in rad(P ). The
radical rad(P ) is small in P because P is finitely generated. Now Ker(p) ⊆ rad(P )
implies that Ker(p) is small in P .
————————————————————————————-
18. Injective modules
18.1. Definition and basic properties. A module I is called injective if the

following is satisfied: For any monomorphism f : X → Y , and any homomorphism
h : X → I there exists a homomorphism g : Y → I such that gf = h.
Y
g O
f

Io h
X
Lemma 18.1. The following are equivalent:
(i) I is injective;
(ii) The functor HomA (−, I) is exact;
(iii) Every monomorphism I → N splits;
Proof. (i) ⇐⇒ (ii): By (i) we know that for all monomorphisms f : X → Y

the map HomA (f, I) : HomA (Y, I) → HomA (X, I) is surjective. This implies that
HomA (−, I) is an exact contravariant functor. The converse is also true.
(i) =⇒ (iii): Let f : I → N be a monomorphism. Thus there exists some g : N → I
such that the diagram
NO
g
f

Io 1I
I
commutes. Thus f is a split monomorphism.
(iii) =⇒ (i): Let f : X → Y be a monomorphism, and let h : X → I be an arbitrary

homomorphism. Taking the pushout along h we obtain a commutative diagram
f
0 / X / Y / Cok(f ) / 0
h h′
f′
0 / I / E / Cok(f ) / 0
with exact rows. By (iii) we know that f ′ is a split monomorphism. Thus there
exists some f ′′ : E → I with f ′′ ◦ f ′ = 1I . Observe that Im(h′ ◦ f ) ⊆ Im(f ′ ). Set
g := f ′′ ◦ h′ . This implies g ◦ f = h. In other words, I is injective.
Lemma 18.2. For an algebra A the following are equivalent:
(i) A is semisimple;
(ii) Every A-module is projective;
(iii) Every A-module is injective.
Proof. Recall that A is semisimple if and only if all A-modules are semisimple. A
module M is semisimple if and only if every submodule of M is a direct summand.
Thus A is semisimple if and only if each short exact sequence

0→X→Y →Z→0
of A-modules splits. Now the lemma follows from the basic properties of projective
and injective modules.
For any left A-module A M let D(A M) = HomK (A M, K) be the dual module of
op
A M. This is a right A-module, or equivalently, a left A -module: For α ∈ D(A M),
a ∈ Aop and x ∈ A M define (aα)(x) := α(ax). It follows that ((ab)α)(x) = α(abx) =
(aα)(bx) = (b(aα))(x). Thus (b ⋆ a)α = (ab)α = b(aα) for all x ∈ M and a, b ∈ A.
Similarly, let MA now be a right A-module. Then D(MA ) becomes an A-module

as follows: For α ∈ D(MA ) and a ∈ A set (aα)(x) := α(xa). Thus we have
((ab)α)(x) = α(xab) = (bα)(xa) = (a(bα))(x) for all x ∈ M and a, b ∈ A.
Lemma 18.3. The A-module D(AA ) = D(Aop A) is injective.
Proof. Let f : X → Y be a monomorphism of A-modules, and let

e : HomK (AA , K) → K
be the map defined by α 7→ α(1). Clearly, e is K-linear, but in general it will not
be A-linear. Let h : X → HomK (AA , K) be a homomorphism of A-modules.
Let us now just think of K-linear maps: There exists a K-linear map e′ : Y → K
such that e′ ◦ f = e ◦ h. Define a map h′ : Y → HomK (AA , K) by h′ (y)(a) := e′ (ay)
for all y ∈ Y and a ∈ A.
h e
X / HomA (AA , K) hhhh4/ K
q8
hh hhhh
h′ q
f
q q hhhhhhhh
hh e ′
qhqhhhhh
h
Y
It is easy to see that h′ is K-linear. We want to show that h′ is A-linear. (In other
words, h′ is a homomorphism of A-modules.)
For y ∈ Y and a, b ∈ A we have h′ (by)(a) = e′ (aby). Furthermore, (bh′ (y))(a) =
h′ (y)(ab) = e′ (aby). This finishes the proof.
Lemma 18.4. There are natural isomorphisms
!
Y Y
HomA −, Mi ∼= HomA (−, Mi )
i∈I i∈I
and !
M Y
HomA Mi , − ∼
= HomA (Mi , −).
i∈I i∈I
Proof. Exercise.
Lemma 18.5. The following hold:
(i) Direct summands of injective modules are injective;

(ii) Direct products of injective modules are injective;
(iii) Finite direct sums of injective modules are injective.
Proof. Let I = I1 ⊕ I2 be a direct sum decomposition of an injective A-module

I, and let f : X → Y be a monomorphism. If h : X → I1 is a homomorphism,
then [ h0 ] : X → I1 ⊕ I2 is a homomorphism, and since I is injective, we get some
g = [ gg12 ] : Y → I1 ⊕ I2 such that

g ◦ f = gg12 ff ◦ f = [ h0 ] .
Thus g1 ◦ f = h and therefore I1 is injective. This proves (i).
Let Ii , i ∈ I be injective
Q A-modules, let f : X → Y be a monomorphism, and
suppose that h : X → i∈I Ii is any homomorphism. Clearly, Q h = (hi )i∈I where
hi is obtained by composing h with the obvious projection i∈I Ii → Ii . Since
Q exists a homomorphism gi : Y → Ii with gi ◦ f = hi . Set
Ii is injective, there
g := (gi )i∈I : Y → i∈I Ii . It follows that g ◦ f = h. This proves (ii).
The statement (iii) follows obviously form (ii).
Warning: Infinite direct sums of injective modules are often not injective. The
reason is that in general we have
!
M M M
HomA (−, Mi ) ∼
6= HomA (Mi , −) 6∼
= HomA Mi , − .
i∈I i∈I i∈I
Lemma 18.6. If PA is a projective Aop -module, then D(PA ) is an injective A-

module.
L
Proof. First assume that PA = i∈I AA is a free Aop -module. We know already by
Lemma 18.3 that D(AA ) is an injective A-module. By Lemma 18.4 we have
!
M Y Y
D(PA ) = HomK AA , K ∼ = HomK (AA , K) = D(AA ).
i∈I i∈I i∈I
Now Lemma 18.5 (ii) implies that D(PA ) is projective. Any projective module is
a direct summand of a free module. Thus Lemma 18.5 (i) yields that D(PA ) is an
injective A-module for all projective Aop -module PA .
18.2. Injective envelopes.

Lemma 18.7. Every A-module can be embedded into an injective A-module.
Proof. Let A M be an A-module. There exists a projective Aop -module PA and

an epimorphism PA → D(A M). Applying the duality D = HomK (−, K) gives a
monomorphism DD(A M) → D(PA ). Lemma 18.6 says that D(PA ) is an injective
A-module. It is also clear that there exists a monomorphism A M → DD(A M). This
finishes the proof.
One can now define injective resolutions, and develop Homological Algebra with
injective instead of projective modules.
Recall that a submodule U of a module M is called large if for any non-zero sub-
module V of M the intersection U ∩ V is non-zero.
A homomorphism f : M → I is called an injective envelope if the following hold:
(i) I is injective;
(ii) f is a monomorphism;
(iii) f (M) is a large submodule of I.
Lemma 18.8. Let I be an injective module, and let U and V be submodules of I
such that U ∩ V = 0. Assume that U and V are maximal with this property (i.e.
if U ⊆ U ′ with U ′ ∩ V = 0, then U = U ′ , and if V ⊆ V ′ with U ∩ V ′ = 0, then
V = V ′ ). Then I = U ⊕ V .
Proof. It is easy to check that the map

f : I → I/U ⊕ I/V
defined by m 7→ (m + U, m + V ) is a monomorphism: Namely, m ∈ Ker(f ) implies
m ∈ U ∩ V = 0.
There is an embedding (U + V )/U → I/U. We claim that (U + V )/U is large in
I/U: Let U ′ /U be a submodule of I/U (thus U ⊆ U ′ ⊆ I) with
(U + V )/U ∩ (U ′ /U) = 0 = U/U.
In other words, (U + V ) ∩ U ′ = U + (V ∩ U ′ ) = U. This implies (V ∩ U ′ ) ⊆ U and
(obviously) (V ∩ U ′ ) ⊆ V . Thus V ∩ U ′ = 0. By the maximality of U we get U = U ′
and therefore U ′ /U = 0.
Similarly one shows that (U + V )/V is a large submodule of I/V .
We get
(U + V )/U ⊕ (U + V )/V ∼
= V ⊕ U ⊆ M ⊆ M/U ⊕ M/V.
By Lemma 7.13 we know that M is large in M/U ⊕ M/V . But M is injective and
therefore a direct summand of M/U ⊕ M/V . Thus M ⊕ C = M/U ⊕ M/V for some
C. Since M is large, we get C = 0. So M = M/U ⊕ M/V . By the maximality of U
and V we get V = M/U and U = M/V and therefore U ⊕ V = M.
The dual statement for projective modules is also true:

Lemma 18.9. Let P be a projective module, and let U and V be submodules of P
such that U + V = P . Assume that U and V are minimal with this property (i.e.
if U ′ ⊆ U with U ′ + V = P , then U = U ′ , and if V ′ ⊆ V with U + V ′ = P , then
V = V ′ ). Then P = U ⊕ V .
Lemma 18.10. Let U be a submodule of a module M. Then there exists a submodule
V of M which is maximal with the property U ∩ V = 0.
Proof. Let
V := {W ⊆ M | U ∩ W = 0}.
Take S
a chain (Vi )i∈J in V. (Thus for all Vi and Vj we have Vi ⊆ Vj or Vj ⊆ Vi .) Set
V = i Vi . We get
!
[ [
U ∩V =U ∩ Vi = (U ∩ Vi ) = 0.
i i
Now the claim follows from Zorn’s Lemma.
Warning: For a submodule U of a module M there does not necessarily exist a

minimal V such that U + V = M.
Example: Let M = K[T ] and U = (T ). Then for each n ≥ 1 we have (T ) + (T +

1)n = M.
Theorem 18.11. Every A-module has an injective envelope.
Proof. Let X be an A-module, and let X → I be a monomorphism with I injective.

Let V be a submodule of I with X ∩ V = 0 and we assume that V is maximal with
this property. Such a V exists by the previous lemma.
Next, let
U := {U ⊆ I | U ∩ V = 0 and X ⊆ U}.
Again, by Zorn’s Lemma we obtain a submodule U of I which is maximal with
U ∩ V = 0 and X ⊆ U.
Thus, U and V are as in the assumptions of the previous lemma, and we obtain
I = U ⊕ V and X ⊆ U. We know that U is injective, and we have our embedding
X → U.
We claim that X is a large submodule of U:

Let U ′ be a submodule of U with X ∩ U ′ = 0. We have to show that U ′ = 0. We
have X ∩ (U ′ ⊕ V ) = 0: If x = u′ + v, then x − u′ = v and therefore v = 0. Thus
x = u′ ∈ X ∩U ′ = 0. By the maximality of V we have U ′ ⊕V = V . Thus U ′ = 0.
Warning: Projective covers do not exist in general.
If X is an A-module, we denote its injective hull by I(X).

Lemma 18.12. Injective envelopes are uniquely determined up to isomorphism.
Proof. Exercise.
Recall that a module M is uniform, if for all non-zero submodules U and V of M

we have U ∩ V 6= 0.
Lemma 18.13. Let I be an indecomposable injective A-module. Then the following
hold:
(i) I is uniform (i.e. if U and V are non-zero submodules of I, then U ∩V 6= 0);

(ii) Each injective endomorphism of I is an automorphism;
(iii) If f, g ∈ EndA (I) are both not invertible, then f + g is not invertible;
(iv) EndA (I) is a local ring.
Proof. (i): Let U and V be non-zero submodules of I. Assume U ∩ V = 0. Let U ′

and V ′ be submodules which are maximal with the properties U ⊆ U ′ , V ⊆ V ′ and
U ′ ∩ V ′ = 0. Lemma 18.8 implies that I = U ′ ⊕ V ′ . But I is indecomposable, a
contradiction.
(ii): Let f : I → I be an injective homomorphism. Since I is injective, f is a split

monomorphism. Thus I = f (I) ⊕ Cok(f ). Since I is indecomposable and f (I) 6= 0,
we get Cok(f ) = 0. Thus f is also surjective and therefore an automorphism.
(iii): Let f and g be non-invertible elements in EndA (I). by (ii) we know that f and
g are not injective. Thus Ker(f ) 6= 0 6= Ker(g). By (i) we get Ker(f ) ∩ Ker(g) 6= 0.
This implies Ker(f + g) 6= 0.
We know already from the theory of local rings that (iii) and (iv) are equivalent
statements.
————————————————————————————-
19. Digression: The stable module category
Let C be a K-linear category. An ideal I in C is defined as follows: To each

pair (C, C ′) of objects C, C ′ ∈ C there is a subspace I(C, C ′ ) of HomC (C, C ′ ) such
that for arbitrary morphisms f : D → C, h : C ′ → D ′ and g ∈ I(C, C ′ ) we have
h ◦ g ◦ f ∈ I(D, D ′ ).
If I is an ideal in C we can define the factor category C/I as follows: The objects
are the same as in C and
HomC/I (C, C ′) := HomC (C, C ′)/I(C, C ′ ).
The composition of morphisms is defined in the obvious way.
If X is a class of objects in C which is closed under finite direct sums, then we say
that f : C → C ′ factors through X , if f = f2 ◦ f1 with f1 : C → X, f2 : X → C ′
and X ∈ X . Let I(X )(C, C ′ ) be the set of morphisms C → C ′ which factor through
X . In this way we obtain an ideal I(X ) in C.
Now let A be an arbitrary K-algebra, and as before let Proj(A) be the full subcate-
gory of projective A-modules. The stable module category of Mod(A) is defined
as
Mod(A) = Mod(A)/I(Proj(A)).
Define
Hom(M, N) := HomMod(A)/Proj(A) (M, N) = HomA (M, N)/I(Proj(A))(M, N).
Similarly, we define mod(A) = mod(A)/I(proj(A)).
Thus the objects of Mod(A) are the same ones as in Mod(A), namely just the
A-modules. But it follows that all projective A-modules become zero objects in
Mod(A): If P is a projective A-module, then 1P lies in I(Proj(A))(P, P ). Thus 1P
becomes zero in Mod(A). Vice versa, if a module M is a zero object in Mod(A),
then M is a projective A-module: If 1M factors through a projective A-module, then
M is a direct summand of a projective module and therefore also projective.
Now Schanuel’s Lemma implies the following: If M is an arbitrary module, and if
p : P → M and p′ : P ′ → M are epimorphisms with P and P ′ projective, then the
kernels Ker(p) and Ker(p′ ) are isomorphic in the category Mod(A).
If we now choose for every module M an epimorphism p : P → M with P projective,
then M 7→ Ker(p) yields a functor Mod(A) → Mod(A). If we change the choice of
P and p, then the isomorphism class of Ker(p) in Mod(A) does not change.
So it makes sense to work with an explicit construction of a projective module P
and an epimorphism p : P → M. Let M be a module, and let F (M) be the free
module with free generating set |M| (the underlying set of the vector space M).
Define
p(M) : F (M) → M
by m 7→ m. In this way we obtain a functor F : Mod(A) → Mod(A)
Let Ω(M) be the kernel of p(M), and let

u(M) : Ω(M) → F (M)
be the corresponding inclusion. Then Ω : Mod(A) → Mod(A) is a functor and
u : Ω → F is a natural transformation. We obtain a short exact sequence
0 → Ω(M) → F (M) → M → 0
with F (M) a free (and thus projective) module. One calls Ω the loop functor or
syzygy functor. This is a functor but it is not at all additive.
For example, if M = 0, then F (M) = A A and Ω(M) = A A.

Future: Let A be a finite-dimensional K-algebra. We will meet stable homomor-
phism spaces in Auslander-Reiten theory, for example the Auslander-Reiten formula
reads
Ext1A (N, τ (M)) ∼
= DHomA (M, N),
for all finite-dimensional A-modules M and N. If A has finite global dimension, we
have
mod(A)b ∼= D b (A)
where A b is the repetitive algebra of A and D b (A) is the derived category of bounded
complexes of A-modules.
————————————————————————————-
20. Projective modules over finite-dimensional algebras
There is a beautiful general theory on projective modules, however one can cut
this short and concentrate on finite-dimensional projective modules over finite-
dimensional algebras. The results in this section can be generalized considerably.
The general theory is developed in Sections 22 and 23.
Theorem 20.1 (Special case of Theorem 23.1). Let A be a finite-dimensional alge-
bra. Then A/J(A) is semisimple.
Proof. The module A A is a finite direct sum of local modules, thus A A/ rad(A A) is
a finite sum of simple modules and therefore semisimple.
bra. If
A A = P1 ⊕ · · · ⊕ Pn
is a direct decomposition of the regular representation into indecomposable modules

Pi , then each finite-dimensional indecomposable projective A-module is isomorphic
to one of the Pi .
Proof. For each finite-dimensional indecomposable projective A-module P there ex-

ists an epimorphism F → P with F a free A-module of finite rank. In particular F
is finite-dimensional. Since P is projective, this epimorphism splits. Then we use
the Krull-Remak-Schmidt Theorem.
bra. The map P 7→ P/ rad(P ) yields a bijection between the isomorphism classes of
finite-dimensional indecomposable projective A-modules and the isomorphism classes
of simple A-modules.
Proof. If P and Q are isomorphic modules, then P/ rad(P ) and Q/ rad(Q) are also
isomorphic. Therefore P 7→ P/ rad(P ) yields a well defined map.
Ln
The map is surjective: Let S be a simple module. We write A A = i=1 Pi with
indecomposable modules Pi . Since Pi is of finite length and indecomposable, we
know that EndA (Pi ) is a local ring. Furthermore, Pi = Aei for some idempotent ei ∈
A. Since EndA (Pi ) ∼
= (ei Aei )op we know that ei Aei is also a local ring. Therefore,
Lemma 12.14 implies that Aei is a local module.
There exists a non-zero homomorphism A A → S, and therefore for at least one index
i we get a non-zero map f : Pi → S. Since S is simple, we know that f is surjective.
Furthermore, the kernel of f is rad(Pi ) because Pi is local. Thus S is isomorphic to
Pi / rad(Pi ).
The map is injective: Let P and Q be finite-dimensional indecomposable projective

modules such that P/ rad(P ) ∼ = Q/ rad(Q). Then Corollary 17.14 implies that
∼
P = Q.
If P is a local projective module, then S := P/ rad(P ) is a simple module and

P (S) := P is the projective cover of S.
Theorem 20.4 (Special case of Theorem 23.4). Let A be a finite-dimensional al-
gebra, and let P be a finite-dimensional indecomposable projective A-module. Set
S := P/ rad(P ). Then the following hold:
(i)P is local;
(ii)EndA (P ) is a local ring;
(iii)J(EndA (P )) = {f ∈ EndA (P ) | Im(f ) ⊆ rad(P )};
(iv) Each endomorphism of P induces an endomorphism of S, and we obtain an
algebra isomorphism EndA (P )/J(EndA (P )) → EndA (S); Lm
(v) The multiplicity of P in a direct sum decomposition A A = i=1 Pi with
indecomposable modules Pi is exactly the dimension of S as an EndA (S)-
module.
Proof. We have shown already that each finite-dimensional indecomposable module

is local and has a local endomorphism ring. Since P is finitely generated, rad(P )
is small in P . Now (iii) and (iv) followL
from Lemma 17.8 and Corollary 17.9. It
m
remains to prove (v): We write A A = i=1 Pi with indecomposable modules Pi .
Then
m
M Mm
J(A) = rad(A A) = rad(Pi ) and A A/J(A) = Pi / rad(Pi ).
i=1 i=1
The multiplicity of P in this decomposition (in other words, the number of direct
summands Pi which are isomorphicLto P ) is equal to the multiplicity of S in the
m
direct decomposition A A/J(A) = i=1 Pi / rad(Pi ). But this multiplicity of S is
the dimension of S as an EndA (S)-module. (Recall that A/J(A) is a semisimple
algebra, and that EndA (S) is a skew field.)
Theorem 20.5. Let A be a finite-dimensional algebra. Then every finitely generated
module has a projective cover.
Proof. Let M be a finitely generated A-module. There exists a finitely

Lngenerated
projective module P and an epimorphism p : P → M. We write P = i=1 Pi with
indecomposable modules Pi . We can assume that P is chosen such that n is minimal.
We want to show that Ker(p) ⊆ rad(P ):
Assume Ker(p) is not a submodule of rad(P ). Then there exists an indecomposable
direct summand P ′ of P which is contained in Ker(p), see Corollary 17.15. But then
we can factorize p through P/P ′ and obtain an epimorphism P/P ′ → M. Since
P ′ is an indecomposable direct summand of P , the Krull-Remak-Schmidt Theorem
implies that P/P ′ is a direct sum of n − 1 indecomposable modules, which is a
contradiction to the minimality of n. Thus we have shown that Ker(p) ⊆ rad(P ).
Since M is finitely generated, rad(M) is small in M, and therefore every submodule

U ⊆ rad(M) is small in M.
Let A be a finite-dimensional algebra, and let M be a finitely generated A-module.

How do we “construct” a projective cover of M?
Let ε : M → M/ rad(M) be the canonical projection. The module M/ rad(M) is a
finitely generated A/J(A)-module. Since A/J(A) is semisimple, also M/ rad(M) is
semisimple. So M/ rad(M) can be written as a direct sum of finitely many simple
modules Si , say
Mn
M/ rad(M) = Si .
i=1
Ln
For each module Si we choose a projective cover qi : P (Si) → Si . Set P = i=1 P (Si)
and
Mn
q= qi : P → M/ rad(M).
i=1
Since P is projective there exists a lifting p : P → M of q, i.e. p is a homomorphism
with ε ◦ p = q. Thus we get a commutative diagram
s9 M
p s
s ε
s
s
s
P q
/ M/ rad(M)
Since P and M are finitely generated we see that p is a projective cover.
————————————————————————————-
21. Projective modules over basic algebras
A K-algebra A is a basic algebra if the following hold:
• A is finite-dimensional;
• A/J(A) ∼ =K| ×K× {z· · · × K}.
n times
In this case, there are n isomorphism classes of simple A-modules, and each simple
A-module is 1-dimensional.
Let Q = (Q0 , Q1 , s, t) be a quiver. By KQ+ we denote the subspace of KQ generated

by all paths of length at least one. Clearly, KQ+ is an ideal in KQ.
An ideal I in KQ is an admissible ideal if there exists some m ≥ 2 such that
(KQ+ )m ⊆ I ⊆ (KQ+ )2 .
It follows that A := KQ/I is a finite-dimensional K-algebra.
Theorem 21.1 (Gabriel). A K-algebra A is basic if and only if A ∼

= KQ/I where
Q is a quiver and I is an admissible ideal.
Proof. Later.
Theorem 21.2 (Gabriel). Let A be a finite-dimensional K-algebra with K alge-
braically closed. Then there exists a uniquely determined quiver Q and an admissible
ideal I in KQ such that the categories mod(A) and mod(KQ/I) are equivalent.
Proof. Later.
We will actually not use Theorems 21.1 and 21.2 very often. But of course these
results are still of central importance, because they tell us that path algebras and
their quotients by admissible ideals are not at all exotic. They are hidden behind
every finite-dimensional algebra over an algebraically closed field.
Assume now that I is an admissible ideal in a path algebra KQ and set A := KQ/I.
For each i ∈ Q0 let Si be a 1-dimensional K-vector space, and let

η : A × S i → Si
be the A-module structure defined by
(
s if a = ei ,
η(a, s) =
0 otherwise
for all s ∈ Si and all paths a in Q. It is clear that the modules Si , i ∈ Q0 are
pairwise non-isomorphic 1-dimensional (and therefore simple) A-modules.
Define A+ = KQ+ /I. This is an ideal in A. The algebra A is (as a vector space)
generated by the residue classes p = p + I of all paths p in Q.
Lemma 21.3. A+ is an ideal in A, and all elements in A+ are nilpotent.
Proof. Clearly, A+ is (as a vector space) generated by the residue classes p = p + I

of all paths p in Q with l(p) ≥ 1. Now it is obvious that A+ is an ideal.
Since A is finite-dimensional, we get that every element in A+ is nilpotent. (When

we take powers of a linear combination of residue classes of paths of length at least
one, we get linear combinations of residue classes of strictly longer paths, which
eventually have to be zero for dimension reasons.)
Corollary 21.4. A+ ⊆ J(A).
Proof. By Lemma 14.5 an ideal consisting just of nilpotent elements is contained in

the radical J(A).
Lemma 21.5. {ei + I | i ∈ Q0 } is a linearly independent subset of A.
Proof. This follows because I ⊆ (KQ+ )2 ⊆ KQ+ .

Corollary 21.6. dim A+ = dim A − |Q0 |.
By abuse of notation we denote the residue class ei + I also just by ei .

Lemma 21.7. dim J(A) ≤ dim A − |Q0 |.
Proof. We know that J(A) consists of all elements x ∈ A such that xS = 0 for all
6 0 for all i ∈ Q0 .
simple A-modules S, see Proposition 14.6. By definition ei Si =
Thus none of the ei belongs to J(A).
Thus for dimension reasons, we obtain the following result:

Lemma 21.8. We have A+ = J(A) and dim J(A) = dim A − |Q0 |.
Lemma 21.9. ei Aei is a local ring for all i ∈ Q0 .
Proof. As a vector space, ei Aei is generated by all residue classes of paths p in Q

with s(p) = t(p) = i. By Lemma 14.12 we know that J(ei Aei ) = ei J(A)ei . We
proved already that A+ = J(A). It follows that J(ei Aei ) is (as a vector space)
generated by all paths p with s(p) = t(p) = i and l(p) ≥ 1. Thus
dim ei Aei /J(ei Aei ) = 1.
Therefore ei Aei is a local ring.
Theorem 21.10. Let A = KQ/I where I is an admissible ideal in a path algebra
KQ. Then the following hold:
L
(i) A A = i∈Q0 Aei is a direct decomposition of the regular representation into
indecomposables;
(ii) Each finite-dimensional indecomposable projective A-module is isomorphic to
one of the Aei ;
(iii) Aei is a local module with top(Aei ) := Aei / rad(Aei ) ∼
= Si ;
∼
(iv) Aei = Aej if and only if i = j;
(v) The Si are L the only simple A-modules;
∼
(vi) A/J(A) = i∈Q0 Si ;
(vii) A is a basic algebra.
Proof. There exists a non-zero homomorphism πi : Aei → Si defined by πi (aei ) =

aei · 1. (Recall that the underlying vector space of Si is just our field K.) It follows
that πi is an epimorphism.
Since ei Aei is a local ring, we know that the modules Aei are local (and indecom-
posable). This implies
Aei / rad(Aei ) ∼
= Si .
The rest of the theorem follows from results we proved before for arbitrary finite-
dimensional algebras.
————————————————————————————-
22. Direct summands of infinite direct sums
22.1. The General Exchange Theorem.

Theorem 22.1 (General Exchange Theorem). Let M be a module with direct de-
compositions
Mm
M =U⊕ Mi = U ⊕ N ⊕ V.
i=1
Ln
We assume that N = j=1 Nj such that the endomorphism rings of the Nj are local.
Then for 1 ≤ i ≤ m there exist direct decompositions Mi = Mi′ ⊕ Mi′′ such that
m
M m
M
M = U ⊕N ⊕ Mi′ and N∼
= Mi′′ .
i=1 i=1
Proof. We prove the theorem by induction on n. For n = 0 there is nothing to show:

We can choose Mi′ = Mi for all i.
Let
m
M n
M
M =U⊕ Mi = U ⊕ Nj ⊕ V
i=1 j=1
be direct decompositions of M, and assume that the endomorphism rings of the
modules Nj are local. Take
M = U ⊕ N ′ ⊕ (Nn ⊕ V )
Ln−1
where N ′ = j=1 Nj . By the induction assumption there are direct decompositions
Mi = Xi ⊕ Yi such that
Mm Mm
M = U ⊕ N′ ⊕ Xi and N ′ ∼ = Yi .
i=1 i=1
Now we look at the direct decomposition

m
M
′
M = (U ⊕ N ) ⊕ Xi
i=1
and the inclusion homomorphism from Nn into M. Then we apply the Exchange
Theorem (see Skript 1) to this situation: We use that Nn ⊕ (U ⊕ N ′ ) is a direct
summand of M. For 1 ≤ i ≤ m we obtain a direct decomposition Xi = Mi′ ⊕ Xi′
such that
Mm
′
M = (U ⊕ N ) ⊕ Nn ⊕ Mi′
i=1
Lm ′ ∼
with i=1 Xi = Nn . Note that only one of the modules Xi′ is non-zero. Define
Mi′′ = ′
Xi ⊕ Yi . This implies
Mi = Xi ⊕ Yi = Mi′ ⊕ Xi′ ⊕ Yi = Mi′ ⊕ Mi′′
and m m m
M M M
Mi′′ = Xi′ ⊕ Yi ∼
= N ′ ⊕ Nn = N.
i=1 i=1 i=1
L
If M = i∈I Mi is a direct sum of modules Mi , and J is a subset of the index set
I, we define M
MJ := Mi .
i∈J
We want to study modules which can be written as direct sums of modules with
local endomorphism ring. The key result in this situation is the following:
L
Theorem 22.2. Let M = i∈I Mi be a direct sum of modules Mi with local endo-
morphism rings, and let U be a direct summand of M. Then the following hold:
(a) For every element u ∈ U there exists a direct decomposition U = U ′ ⊕ U ′′

and a finite subset J ⊆ I such that u ∈ U ′ and U ′ ∼
= MJ ;
(b) If M/U is indecomposable, then there exists some i ∈ I with M = U ⊕ Mi .
Proof. For u ∈ U there exists a finite subset I ′ of I such that u ∈ MI ′ . Since U is a

direct summand of M, we can choose a direct decomposition M = U ⊕ C. By the
General Exchange Theorem 22.1 there exist submodules U ′′ ⊆ U and C ′′ ⊆ C such
that M = MI ′ ⊕ U ′′ ⊕ C ′′ . Define
U ′ = (MI ′ ⊕ C ′′ ) ∩ U and C ′ = (MI ′ ⊕ U ′′ ) ∩ C.
We claim that
U = U ′ ⊕ U ′′ and C = C ′ ⊕ C ′′ .
It is enough to show the first equality: Of course we have U ′ ∩ U ′′ = 0, since
(MI ′ ⊕ C ′′ ) ∩ U ′′ = 0. Since U ′′ ⊂ U, we get by modularity
U = M ∩ U = (U ′′ ⊕ MI ′ ⊕ C ′′ ) ∩ U
= U ′′ + ((MI ′ ⊕ C ′′ ) ∩ U)
= U ′′ + U ′ .
We see that
(1) U ′ ⊕ U ′′ ⊕ C ′ ⊕ C ′′ = U ⊕ C = M = MI ′ ⊕ U ′′ ⊕ C ′′
and therefore
= M/(U ′′ ⊕ C ′′ ) ∼
U ′ ⊕ C′ ∼ = MI ′ .
By the Krull-Remak-Schmidt Theorem there exists a subset J ⊆ I ′ with U ′ ∼ =
MJ . Of course u belongs to U ′ = (MI ′ ⊕ C ′′ ) ∩ U. Thus we constructed a direct
decomposition U = U ′ ⊕ U ′′ with u ∈ U ′ and U ′ ∼ = MJ with J a finite subset of I.
This proves part (a) of the theorem.
We started with an arbitrary direct decomposition M = U ⊕ C, and now we want

to prove (b) for the direct summand C (and not for U). Thus we assume that M/C
is indecomposable. Since U ∼ = M/C, we know that U is indecomposable. Let u be
a non-zero element in U. As before we obtain a direct decomposition U = U ′ ⊕ U ′′

with u ∈ U ′ . We see that U ′′ = 0. Thus Equation 1 reduces to
M
U ′ ⊕ C ′ ⊕ C ′′ = M = Mi ⊕ C ′′ .
i∈I ′
′
Now C is isomorphic to a direct summand of MI ′ , thus by the Krull-Remak-Schmidt
Theorem it is a finite direct sum of modules with local endomorphism rings. Thus
we can apply the General Exchange Theorem 22.1 and obtain direct decompositions
Mi = Mi′ ⊕ Mi′′ for i ∈ I ′ such that
M M
M = C′ ⊕ Mi′ ⊕ C ′′ = C ⊕ Mi′ .
i∈I ′ i∈I ′
Since M/C is indecomposable, we know that exactly one of the modules Mi′ , say
Mi′0 , is non-zero. On the other hand, Mi0 = Mi′0 ⊕ Mi′′0 is indecomposable, and
therefore Mi′0 = Mi0 . Thus M = C ⊕ Mi0 . This proves part (b) for the direct
summand C of M.
L
Corollary 22.3. Let M = i∈I Mi be a direct sum of modules Mi with local endo-
morphism rings. Then every non-zero direct summand of M has a direct summand
which is isomorphic to one of the Mi ’s.
Proof. If U is a non-zero direct summand of M, then choose some 0 6= u ∈ U. Then

part (a) of Theorem 22.2 yields a direct decomposition U = U ′ ⊕ U ′′ with u ∈ U ′
and U ′ ∼
= MJ for some finite index set J ⊆ I. Since 0 6= u ∈ U ′ , we know that J is
non-empty. If i ∈ J, then U has a direct summand isomorphic to Mi .
L
Corollary 22.4. Let M = i∈I Mi be a direct sum of modules Mi with local endo-
morphism rings. If U is an indecomposable direct summand of M, then U ∼ = Mi for
some i ∈ I.
Proof. Choose 0 6= u ∈ U. We get a direct decomposition U = U ′ ⊕ U ′′ and a finite

non-empty index set J ⊆ I with u ∈ U ′ and U ′ ∼
= MJ . Since U is indecomposable,
′ ∼
U = U = Mi with i ∈ J.
22.2. The Krull-Remak-Schmidt-Azumaya Theorem.

L
Theorem 22.5 (Azumaya). Let M L = i∈I Mi be a direct sum of modules Mi with
local endomorphism rings. Let U = j∈J Uj be a direct summand of M. For every
indecomposable module N let I(N) be the set of indices i ∈ I with Mi ∼
= N, and let
∼
J(N) be the set of indices j ∈ J with Uj = N. Then we have
|J(N)| ≤ |I(N)|.
Proof. First, let J(N) be finite and non-empty, and let j0 ∈ J(N). Corollary 22.4
yields that there exists some i0 ∈ I with Mi0 ∼= Uj0 . The Cancellation Theorem
implies that M M
Mi ∼
= Uj .
i∈I\{i0 } j∈J\{j0 }
By induction we obtain |J(N)| ≤ |I(N)|.

L
Next, assume that J(N) is infinite. For t ∈ J define Ut′ = j6=t Uj . Let i ∈ I(N),
and let Ji be the set of all t ∈ J with M = Mi ⊕ Ut′ . Obviously, Ji is a subset of
= Mi ∼
J(N), because Mi ⊕ Ut′ = Ut ⊕ Ut′ implies Ut ∼ = N.
On the other hand, if t ∈ J(N), then Ut′ is a maximal direct summand of M. Thus
there exists some i ∈ I with Ut′ ⊕ Mi = M, and we see that t ∈ Ji . We proved that
[
Ji = J(N).
i∈I(N )
We claim that every set of the form Ji is finite: LetL x 6= 0 be an element in Mi .

There exists a finite subset J(x) ⊆ J such that x ∈ j∈J(x) Uj . If t ∈ / J(x), then
L ′ ′
j∈J(x) Uj ⊆ Ut , and therefore x ∈ Mi ∩ Ut which implies t ∈/ Ji . We see that Ji is
a subset of the finite set J(x).
S
Since J(N) is infinite, I(N) has to be infinite as well. The cardinality of i∈I(N ) Ji
is at most |I(N)|, thus |J(N)| ≤ |I(N)|.
L
Corollary 22.6. Let M = i∈I Mi be a direct sum of modules Mi with L local en-
domorphism rings. Let U be a direct summand of M such that U = j∈J Uj with
Uj indecomposable for all j ∈ J. Then there exists an injective map σ : J → I such
that Uj ∼ = Mσ(j) for all j ∈ J. In particular, U is isomorphic to MI ′ for some subset
′
I of I.
L
Proof. Let U = j∈J Uj with indecomposable module Uj . We choose a direct
complement C of U, thus
M
M = U ⊕C = Uj ⊕ C.
j∈J
To this decomposition we apply the above Theorem 22.5. Thus for any indecompos-
able module N we have |J(N)| ≤ |I(N)|. Thus there is an injective map σ : J → I
such that Uj ∼
= Mσ(j) for all j. We can identify J with a subset I ′ of I, and we
∼
obtain U = MI ′ .
L
Corollary 22.7 (Krull-Remak-Schmidt-Azumaya). Assume that M = L i∈I Mi is
a direct sum of modules Mi with local endomorphism rings. Let M = j∈J Uj
with indecomposable modules Uj . Then there exists a bijection σ : I → J such that
Mi ∼
= Uσ(i) .
Proof. By Corollary 22.6 there is an injective map σ : I → J with Mi ∼

= Uσ(i) for all i.
By Corollaries 22.4 and 22.6 we know that the modules Uj have local endomorphism
rings. Thus for every indecomposable module N we have not only |J(N)| ≤ |I(N)|,
but also the reverse |I(N)| ≤ |J(N)|, which implies |J(N)| = |I(N)|. Thus we can
construct a bijection σ : I → J with Mi ∼ = Uσ(i) for all i.
22.3. The Crawley-Jønsson-Warfield Theorem.

L
Theorem 22.8 (Crawley-Jønsson-Warfield). Let M = i∈I Mi be a direct sum of
modules Mi with local endomorphism rings. If U is a countably generated direct
summand of M, then there exists a subset J of I with
M
U∼ = MJ := Mj .
j∈J
Proof. Let U be a countably generated direct summand of M, and let u1 , u2 , . . . be

a countable generating set of U. Inductively we construct submodules Ut and Vt of
U with
U = U1 ⊕ · · · ⊕ Ut ⊕ Vt
Lt
such that u1 , . . . , ut ∈ i=1 Ui , and such that each Ut is a direct sum of indecom-
posable modules.
As a start of our induction we take t = 0, and there is nothing to show. Now assume
that we constructed alread U1 , . . . , Ut , Vt with the mentioned properties.
Lt
Let ut+1 = xt+1 + yt+1 with xt+1 ∈ i=1 Ui and yt+1 ∈ Vt . Now Vt is a direct
summand of M and vt+1 ∈ Vt . Thus by Theorem 22.2, (a) there exists a direct
decomposition Vt = Ut+1 ⊕ Vt+1 with yt+1 ∈ Ut+1 such that Ut+1 is a direct sum of
Lt+1 of the form Mi . Since yt+1 ∈ Ut+1 , we know that ut+1 = xt+1 + yt+1 belongs
modules
to i=1 Ui .
We obtain the direct decomposition
U1 ⊕ · · · ⊕ Ut ⊕ Ut+1 ⊕ Vt+1 ,
and the modules Ui with 1 ≤ i ≤ t + 1 have the desired form.
By construction, the submodules Ui with i ∈ N1 form a direct sum. This direct sum
L ui with i ∈ N1 . Since these
is a submodule of U, and it also contains all elements
elements form a generating set for U, we get U = i Ui . Thus we wrote U as a
direct sum of indecomposable modules. Corollary 22.6 shows now that U is of the
desired form.
22.4. Kaplansky’s Theorem. Let U be a submodule of a module M, and let

M = M1 ⊕ M2 be a direct decomposition of M. We call U compatible with the
direct decomposition M = M1 ⊕ M2 if U = (U ∩ M1 ) + (U ∩ M2 ).
By ℵ0 we denote the first infinite cardinal number. (For example Q is of cardinality
ℵ0 .) Let c be a cardinal number. A module M is called c-generated if M has a
generating set of cardinality at most c, and M is countably generated if M is
ℵ0 -generated.
Theorem 22.9 (Kaplansky). Let c ≥ ℵ0 be a cardinal number. The class of mod-
ules, which are direct sums of c-generated modules, is closed under direct summands.
L
Proof. For each i ∈ I let Mi be a c-generated module, and let M = i∈I Mi . We
can assume Mi 6= 0 for all i. Let M = X ⊕ Y . We want to show that X is a direct
sum of c-generated submodules. Set e = e(X, Y ). If U is a submodule of M, define
σ(U) = eU + (1 − e)U. We call a subset J ⊆ I compatible if MJ is compatible with
the decomposition X ⊕ Y . A set with cardinality at most c is called a c-set.
We start with some preliminary considerations:
(1) If J ⊆ I is a c-set, then MJ is c-generated.

S
Proof: For every i ∈ I choose a c-set Xi which generates Mi . Then L set X J = i∈J Xi
which is a generating set of MJ . (Here we assume that MJ = i∈J Mi is an inner
direct sum, so the Mi are indeed submodules of MJ .) Since J and also all the Xi
are c-sets, the cardinality of XJ is at most c2 . Since c is an infinite cardinal, we have
c2 = c.
(2) If U is a c-generated submodule of M, then eU, (1−e)U and σ(U) are c-generated.
Proof: If X is a generating set of U, then eX = {ex | x ∈ X} is a generating set of
eU. Similarly, (1 − e)X is a generating set of (1 − e)U.
(3) For every c-generated submodule U of M there exists a c-set J ⊆ I such that
U ⊆ MJ .
Proof: Let X be a generating set of U.SFor every x ∈ X there exists a finite subset
J(x) ⊆ I with x ∈ MJ(x) . Define J = x∈X J(x). Now all sets J(x) are finite, X is
a c-set and c is an infinite cardinal number, thus we conclude that J is a c-set. By
construction X is contained in the submodule MJ of M. Since U is (as a module)
generated by X, we know that U is a submodule of MJ .
(4) For every c-generated submodule U of M there exists a compatible c-set J ⊆ I
such that U ⊆ MJ .
Proof: Let U be a c-generated submodule of M. By (3) we can find a c-set J(1) ⊆ I
such that U ⊆ MJ(1) . We can form σ(MJ(1) ).
Inductively we construct c-sets J(1) ⊆ J(2) ⊆ · · · ⊆ I such that σ(M S J(t) ) ⊆
S t ≥ 1. (Here we use (1), (2) and (3).) Define J = t≥1 J(t).
σ(MJ(t+1) ) for all
We have MJ = t≥1 MJ(t) . Since J(t) is a c-set, and since c is an infinite cardinal
number, the set J is also a c-set. It remains to show that MJ is compatible with the
decomposition M = X ⊕ Y , in other words, we have to show that for every x ∈ MJ
also ex belongs to MJ : Since x ∈ MJ we have x ∈ MJ(t) for some t. Therefore
ex ∈ eMJ(t) ⊆ σ(MJ(t) ) ⊆ MJ(t+1) ⊆ MJ .
S
(5) If I(j) is a compatible subset of I, then eMI(j) ⊆ MI(j) . Set J = I(j). If
eMI(j) ⊆ MI(j) for every j, then eMJ ⊆ MJ .
Now we can start with the proof of the theorem:
Let I(α) be an ordered chain of compatible subsets of I with the following properties:
(i) The cardinality of I(α + 1) \ I(α) isSat most c;

(ii) If λ is a limit
S number, then I(λ) = α<λ I(α);
(iii) We have α I(α) = I.
Here I(α) is defined inductively: Let I(0) = 0. If α is an ordinal number with

I(α) ⊂ I, choose some x ∈ MI \ MI(α) . Let Ux be the submodule generated by x.
By (4) there exists a compatible subset J(x) of I with cardinality at most c such
that Ux is contained in MJ(x) . Define I(α + 1) = I(α) ∪ J(x). By (5) this is again a
compatible set. For a limit number λ define I(λ) as in (ii). It follows from (5) that
I(λ) is compatible.
Since I(α) is compatible, we get a decomposition
MI(α) = X(α) ⊕ Y (α)
with X(α) ⊆ X and Y (α) ⊆ Y . Let us stress that the submodules X(α) and
Y (α) are direct summands of M. We have X(α) ⊆ X(α + 1), and X(α) is a direct
summand of X(α + 1),Ssay X(α + 1) = X(α) ⊕ U(α + 1). If λ is a limit ordinal
number, then X(α) = α<λ X(α). We get
M
X= U(α)
α
and
U(α + 1) = X(α + 1)/X(α) = eMI(α+1) /eMI(α) ∼ = eMI(α+1)\I(α) .
By (i), (1) and (2) we obtain that U(α) is c-generated.
L
Corollary 22.10. Let M = i∈I Mi be a direct sum of countably generated modules
Mi with local endomorphism rings. If U is a direct summand of M, then there exists
a subset J ⊆ I such that U ∼= MJ .
Proof. By Theorem 22.9 every direct summand U of M is a direct sum of countably

generated modules Uj . On the other hand, we know that every countably generated
direct summand Uj of M is a direct sum of indecomposable modules, thus U ist a
direct sum of indecomposable modules. Finally, we use Corollary 22.6.
Question 22.11. Is the class of modules, which are direct sums of modules with
local enomorphism rings, closed under direct summands?
The following is a direct consequence of Theorem 22.9:

Theorem 22.12 (Kaplansky). Every projective module is a direct sum of countably
generated projective modules.
Proof. Each projective A-module is a direct summand of a direct sum of modules

of the form A A. The module A A is cyclic, in particular it is countably generated.
Thus we can apply Theorem 22.9, where we set c = ℵ0 .
————————————————————————————-
23. Projective modules over semiperfect algebras
In this section we generalize the results from Section 20.

An algebra A is semiperfect if the following equivalent conditions are satisfied:
(i) A A is a (finite) direct sum of local modules;

(ii) A A is a (finite) direct sum of modules with local endomorphism ring;
(iii) The identity 1A is a sum of pairwise orthogonal idempotents ei such that the
rings ei Aei are local.
If A is semiperfect, then Aop is semiperfect as well. This follows since condition (iii)
is left-right symmetric.
Examples:
(a) Finite-dimensional algebras are semiperfect.

(b) Let M1 , . . . , Mn be indecomposable A-modules with local endomorphism
rings, and let M = M1 ⊕ · · · ⊕ Mn . Then EndA (M) is semiperfect, since
condition (iii) is obviously satisfied.
Theorem 23.1. Let A be a semiperfect algebra. Then A/J(A) is semisimple.
Proof. The module A A is a finite direct sum of local modules, thus A A/ rad(A A) is
a finite sum of simple modules and therefore semisimple.
Warning: The converse of the above theorem does not hold: By K(T ) we denote
the field of rational functions in one variable T . Thus K(T ) consists of fractions
f /g of polynomials f, g ∈ K[T ] where g 6= 0. Now let A be the subring of K(T )
consisting of all rational functions f /g such that neither T nor T − 1 divide g. The
radical J(A) of A is the ideal generated by T (T − 1), and the corresponding factor
ring A/J(A) is isomorphic to K × K, in particular it is semisimple. Note that A
has no zero divisors, but A/J(A) contains the two orthogonal idempotents −T + 1
and T . For example
(−T + 1)2 = T 2 − 2T + 1 = (T 2 − T ) − T + 1,
and modulo T 2 − T this is equal to −T + 1.
Theorem 23.2. Let A be a semiperfect algebra. Then the following hold:
• Each projective A-module is a direct sum of indecomposable modules;

• Each indecomposable projective module is local and has a local endomophism
ring;
• If A A = P1 ⊕ · · · ⊕ Pn is a direct decomposition of the regular represen-
tation into indecomposable modules Pi , then each indecomposable projective
A-module is isomorphic to one of the Pi .
Proof. Since A is a semiperfect algebra, the module AA is a direct sum of local

modules. Thus let
Mm
AA = Qi
i=1
with local modules Qi . As a direct summand of A A each Qi is of the form Aei
for some idempotent ei . In particular Qi is cyclic. The endomorphism ring of an
A-module of the form Ae (where e is an idempotent) is (eAe)op , and if Ae is local,
then so is eAe. Thus also (eAe)op is a local ring.
Let P be a projective A-module. Thus P is a direct summand of a free A-module F .
We know that F is a direct sum of modules with local endomorphism ring, namely
of copies of the Qi . Kaplansky’s Theorem implies that
M
P = Pj
j∈J
is a direct sum of countably generated modules Pj . By the Crawley-Jønsson-Warfield

Theorem each Pj (and therefore also P ) is a direct sum of modules of the form Qi .
So each projective module is a direct sum of indecomposable modules, and each

indecomposable projective module is of the form Qi , in particular it is local.
L
If A A = nk=1 Pk is another direct decomposition with indecomposable modules Pk ,
then by the Krull-Remak-Schmidt Theorem we get m = n, and each Pk is isomorphic
to some Qi .
Theorem 23.3. Let A be a semiperfect algebra. The map P 7→ P/ rad(P ) yields
a bijection between the isomorphism classes of indecomposable projective A-modules
and the isomorphism classes of simple A-modules.
Proof. By Theorem 23.2 we know that each indecomposable projective A-module is

local and isomorphic to a direct summand of A A. Now we can continue just as in
the proof of Theorem 20.3.
Theorem 23.4. Let A be a semiperfect algebra, and let P be an indecomposable
projective A-module. Set S := P/ rad(P ). Then the following hold:
(i)P is local;
(ii)EndA (P ) is a local ring;
(iii)J(EndA (P )) = {f ∈ EndA (P ) | Im(f ) ⊆ rad(P )};
(iv) Each endomorphism of P induces an endomorphism of S, and we obtain an
algebra isomorphism EndA (P )/J(EndA (P )) → EndA (S); Lm
(v) The multiplicity of P in a direct sum decomposition A A = i=1 Pi with
indecomposable modules Pi is exactly the dimension of S as an EndA (S)-
module.
Proof. We have shown already that each indecomposable projective A-module P is

local and isomorphic to a direct summand of A A. Therefore EndA (P ) is local. In
particular, rad(P ) is small in P . Now copy the proof of Theorem 20.4.
————————————————————————————-
24. Digression: Projective modules in other areas of mathematics
...
Part 5. Homological Algebra I: Resolutions and extension groups
25. Pushout and pullback
25.1. Pushout. Let U, V1 , V2 be modules, and let f1 : U → V1 and f2 : U → V2 be

homomorphisms.
V1
f1 ~>
~~~
~
~~
U@
@ @@
@@
f2 @
V2
Define
W = V1 ⊕ V2 /{(f1 (u), −f2 (u)) | u ∈ U}
and gi : Vi → W where g1 (v1 ) = (v1 , 0) and g2 (v2 ) = (0, v2 ). Thus gi is the composi-
tion of the inclusion Vi → V1 ⊕ V2 followed by the projection V1 ⊕ V2 → W .
One calls W (or more precisely W together with the homomorphisms g1 and g2 ) the
pushout (or fibre sum) of f1 and f2 .
So W is the cokernel of the homomorphism t [f1 , −f2 ] : U → V1 ⊕V2 , and [g1 , g2 ] : V1 ⊕

V2 → W is the corresponding projection. We get an exact sequence
t [f
1 ,−f2 ] [g1 ,g2 ]
U −−−−−→ V1 ⊕ V2 −−−→ W → 0.
Obviously,
t [f
1 ,f2 ] [g1 ,−g2 ]
U −−−−→ V1 ⊕ V2 −−−−→ W → 0
is also an exact sequence.
Proposition 25.1 (Universal property of the pushout). For the module W and
the homomorphisms g1 : V1 → W and g2 : V2 → W as defined above the following
hold: We have g1 f1 = g2 f2 , and for every module X together with a pair (h1 : V1 →
X, h2 : V2 → X) of homomorphisms such that h1 f1 = h2 f2 there exists a uniquely
determined homomorphism h : W → X such that h1 = hg1 and h2 = hg2 .
>
V1 ATTTT
f1 ~~ AA TTTTT h
~ AA TTT1 T
~~ g1 AAA TTTT
~~ TTT
U@ W _ _ _h _ _ _Tj*/4 X
@ > } j jjj
@@ g2 }} jjjj
@@ }}} jjjjjj
f2 @ }}jjjjj h2
j
V2
Proof. Of course g1 f1 = g2 f2 . If we have h1 f1 = h2 f2 for some homomorphisms

hi : Vi → X, then we can write this as

f1
[h1 , h2 ] = 0.
−f2
This implies that the homomorphism [h1 , h2 ] factorizes through the cokernel of
t
[f1 , −f2 ]. In other words there is a homomorphism h : W → X such that
[h1 , h2 ] = h[g1 , g2 ].
But this means that h1 = hg1 and h2 = hg2 . The factorization through the cokernel
is unique, thus h is uniquely determined.
More generally, let f1 : U → V1 , f2 : U → V2 be homomorphisms. Then a pair

(g1 : V1 → W, g2 : V2 → W ) is called a pushout of (f1 , f2 ), if the following hold:
• g1 f1 = g2 f2 ;
• For all homomorphisms h1 : V1 → X, h2 : V2 → X such that h1 f1 = h2 f2
there exists a unique(!) homomorphism h : W → X such that hg1 = h1 and
hg2 = h2 .
Lemma 25.2. Let f1 : U → V1 , f2 : U → V2 be homomorphisms, and assume that
(g1 : V1 → W, g2 : V2 → W ) and also (g1′ : V1 → W ′ , g2′ : V2 → W ′ ) are pushouts of
(f1 , f2 ). Then there exists an isomorphism h : W → W ′ such that hg1 = g1′ and
hg2 = g2′ . In particular, W ∼= W ′.
Proof. Exercise.
25.2. Pullback. Let V1 , V2 , W be modules, and let g1 : V1 → W and g2 : V2 → W

be homomorphisms.
V1 A
AA g1
AA
AA
A
W
}}>
}}
}}
}} g2
V2
Define
U = {(v1 , v2 ) ∈ V1 ⊕ V2 | g1 (v1 ) = g2 (v2 )}.
One easily checks that U is a submodule of V1 ⊕V2 . Define fi : U → Vi by fi (v1 , v2 ) =
vi . Thus fi is the composition of the inclusion U → V1 ⊕V2 followed by the projection
V1 ⊕ V2 → Vi . One calls U (or more precisely U together with the homomorphisms
f1 and f2 ) the pullback (or fibre product) of g1 and g2 . So U is the kernel
of the homomorphism [g1 , −g2 ] : V1 ⊕ V2 → W and t [f1 , f2 ] : U → V1 ⊕ V2 is the
corresponding inclusion. We get an exact sequence
t [f
1 ,f2 ] [g1 ,−g2 ]
0 → U −−−−→ V1 ⊕ V2 −−−−→ W.
Of course, also
t [f
1 ,−f2 ] [g1 ,g2 ]
0 → U −−−−−→ V1 ⊕ V2 −−−→ W.
is exact.
Proposition 25.3 (Universal property of the pullback). For the module U and
the homomorphisms f1 : U → V1 and f2 : U → V2 as defined above the following
hold: We have g1 f1 = g2 f2 , and for every module Y together with a pair (h1 : Y →
V1 , h2 : Y → V2 ) of homomorphisms such that g1 h1 = g2 h2 there exists a uniquely
determined homomorphism h : Y → U such that h1 = f1 h and h2 = f2 h.
j5 V1 A
h1 jjjjj
jjjj~~> AA g1
~ AA
jjj ~
~ f1 AA
j jjjjj ~~ A
j j h
Y T_TT_T _ _ _ _/ U @ W
TTTT
TTTT @@ f }}>
2
TTTT @@@ }}
}}
h2 TTTT @
T) }} g2
V2
Proof. Exercise.
More generally, let g1 : V1 → W , g2 : V2 → W be homomorphisms. Then a pair

(f1 : U → V1 , f2 : U → V2 ) is called a pullback of (g1 , g2 ), if the following hold:
• g1 f1 = g2 f2 ;
• For all homomorphisms h1 : Y → V1 , h2 : Y → V2 such that g1 h1 = g2 h2
there exists a unique(!) homomorphism h : Y → U such that f1 h = h1 and
f2 h = h2 .
Lemma 25.4. Let g1 : V1 → W , g2 : V2 → W be homomorphisms, and assume that
(f1 : U → V1 , f2 : U → V2 ) and also (f1′ : U ′ → V1 , f2′ : U ′ → V2 ) are pullbacks of
(g1 , g2 ). Then there exists an isomorphism h : U ′ → U such that f1 h = f1′ and
f2 h = f2′ . In particular, U ∼
= U ′.
Proof. Exercise.
Since the pushout of a pair (f1 : U → V1 , f2 : U → V2 ) (resp. the pullback of a pair

(g1 : V1 → W, g2 : V2 → W )) is uniquely determined up to a canonical isomorphism,
we speak of “the pushout” of (f1 , f2 ) (resp. “the pullback” of (g1 , g2 )).
25.3. Properties of pushout and pullback.

Lemma 25.5. Let (g1 : V1 → W, g2 : V2 → W ) be the pushout of homomorphisms
(f1 : U → V1 , f2 : U → V2 ), and let (f1′ : U ′ → V1 , f2′ : U ′ → V2 ) be the pullback of
(g1 , g2 ). Then the uniquely determined homomorphism h : U → U ′ with f1 = f1′ h
and f2 = f2′ h is surjective. If t [f1 , f2 ] is injective, then h is an isomorphism, and
(f1 , f2 ) is a pullback of (g1 , g2 ).
j5 V1 A
f1jjj jjjj}}> AA g1
j jj }} AA
jjjj } f′ AA
jj j }} 1
A
j jj h
U T_TT_T _ _ _ _/ U ′ >W
TTTT AA ′
TTTT AAf2 }}}
TTTT AA }
}}
f2 TTTT A
T) }} g2
V2
Proof. Exercise.
Lemma 25.6. Let (f1 : U → V1 , f2 : U → V2 ) be the pullback of homomorphisms
(g1 : V1 → W, g2 : V2 → W ), and let (g1′ : V1 → W ′ , g2′ : V2 → W ′ ) be the pushout of
(f1 , f2 ). Then the uniquely determined homomorphism h : W ′ → W with g1 = hg1′
and g2 = hg2′ is injective. If [g1 , g2 ] is surjective, then h is an isomorphism, and
(g1 , g2 ) is the pushout of (f1 , f2 ).
V1 TBTTT
f1 ~? BB TTTTT g1
~~~ BB
B
TTTT
TTTT
~ g1′ BB!
~~ TTTT
T
′ _ _ _h _ _ _/*4
U@ >W jjjj
W
@ | j
@@ g2′ || jjjj
@@ ||| jjjjjgjj
f2 @ || jj 2
jjjj
V2
Proof. Exercise.
Lemma 25.7. Let (g1 : V1 → W, g2 : V2 → W ) be the pushout of a pair (f1 : U →
V1 , f2 : U → V2 ). If f1 is injective, then g2 is also injective.
Proof. Assume g2 (v2 ) = 0 for some v2 ∈ V2 . By definition g2 (v2 ) is the residue class
of (0, v2 ) in W , thus there exists some u ∈ U with (0, v2 ) = (f1 (u), −f2 (u)). If we
assume that f1 is injective, then 0 = f1 (u) implies u = 0. Thus v2 = −f2 (u) = 0.
Lemma 25.8. Let (f1 : U → V1 , f2 : U → V2 ) be the pullback of a pair (g1 : V1 →
W, g2 : V2 → W ). If g1 is surjective, then f2 is also surjective.
Proof. Let v2 ∈ V2 . If we assume that g1 is surjective, then for g2 (v2 ) ∈ W there

exists some v1 ∈ V1 such that g1 (v1 ) = g2 (v2 ). But then u = (v1 , v2 ) belongs to U,
and therefore f2 (u) = v2 .
Pushouts are often used to construct bigger modules from given modules. If V1 , V2
are modules, and if U is a submodule of V1 and of V2 , then we can construct the
pushout of the inclusions f1 : U → V1 , f2 : U → V2 . We obtain a module W and
homomorphisms g1 : V1 → W , g2 : V2 → W with g1 f1 = g2 f2 .
>
V1 A
~~
f1 AA g1
~ AA
~~ AA
~~ A
U@ >W
@ @@ }}
@@ }}
}}
f2 @ }} g2
V2
Since f1 and f2 are both injective, also g1 and g2 are injective. Also (up to canonical
isomorphism) (f1 , f2 ) is the pullback of (g1 , g2 ).
25.4. Induced exact sequences. Let

f g
0→U −
→V −
→W →0
be a short exact sequence, and let a : U → X be any homomorphism. We construct

the pushout (a′ : V → P, f ′ : X → P ) of (f : U → V, a : U → X). Since the
homomorphisms g : V → W and 0 : X → W satisfy the equation gf = 0 = 0a, there
is a homomorphism g ′ : P → W with g ′a′ = g and g ′ f ′ = 0. Thus we obtain the
commutative diagram
f g
0 / U / V / W / 0
a a′
f′ g′
0 / X / P / W / 0
and we claim that (f ′ , g ′) is again a short exact sequence, which we call the (short
exact) sequence induced by a. We write a∗ (f, g) = (f ′ , g ′ ).
Proof. Since f is injective, we know that f ′ is also injective. Since g = g ′a′ is

surjective, also g ′ is surjective. By construction g ′f ′ = 0, thus Im(f ′ ) ⊆ Ker(g ′ ). We
have to show that also the other inclusion holds: Let (v, x) ∈ Ker(g ′ ) where v ∈ V
and x ∈ X. Thus
0 = g ′((v, x)) = g ′ (a′ (v) + f ′ (x)) = g ′ (a′ (v)) = g(v).
Since (f, g) is an exact sequence, there is some u ∈ U with f (u) = v. This implies
f ′ (x + a(u)) = (0, x + a(u)) = (v, x),
because (v, x) − (0, x + a(u)) = (v, −a(u)) = (f (u), −a(u)).
Dually, let b : Y → W be any homomorphism. We take the pullback (g ′′ : Q →

Y, b′′ : Q → Y ) of (b : Y → W, g : V → W ). Since the homomorphisms 0 : U → Y
and f : U → V satisfy b0 = 0 = gf , there exists a homomorphism f ′′ : U → Q with
g ′′f ′′ = 0 and b′′ f ′′ = f . Again we get a commutative diagram

f ′′ g ′′
0 / U / Q / Y / 0
b′′ b
f g
0 / U / V / W / 0,
and similarly as before we can show that (f ′′ , g ′′ ) is again a short exact sequence.
We write b∗ (f, g) = (f ′′ , g ′′), and call this the (short exact) sequence induced
by b.
Lemma 25.9. Let

f g
0→U −
→V −
→W →0
be a short exact sequence. Then the following hold:
(i) If a : U → X is a homomorphism, then there exists a homomorphism
a′′ : V → X
with a = a′′ f if and only if the induced sequence a∗ (f, g) splits;

(ii) If b : Y → W is a homomorphism, then there exists a homomorphism
b′′ : Y → V
with b = gb′′ if and only if the induced sequence b∗ (f, g) splits.
Y
}
b′′ }
} b
f ~} g
0 / U /V / W / 0
~
a ~
~ a′′
~~
X
Proof. Let a : U → X be a homomorphism. We obtain a commutative diagram with

exact rows:
f g
0 / U / V / W / 0
a a′
f′ g′
0 / X / P / W / 0
The lower sequence is by definition a∗ (f, g). If this sequence splits, then there
exists some f ′′ : P → X such that f ′′ f ′ = 1X . Define a′′ = f ′′ a′ . Then a′′ f =
f ′′ a′ f = f ′′ f ′ a = a. Vice versa, let a′′ : V → X be a homomorphism with a′′ f = a.
Since a′′ f = 1X a, the universal property of the pushout shows that there exists a
homomorphism h : P → X such that a′′ = ha′ , 1X = hf ′ .
> V @T@TTTTTT
f ~~~ @@ TTTTa′′
~~ @@ TTTT
~~ a′ @ TTTT
h_ _ T_
T/)
U @@ > P _ _ _
jj j
j 5 X
@@ f ′ ~~~ jjj
jjj
@
a @@ ~ ~~ jjjjj1jX
~jjj jj
X
In particular, f ′ is a split monomorphism. Thus the sequence (f ′ , g ′ ) = a∗ (f, g)
splits.
The second part of the lemma is proved dually.
Lemma 25.10. Let
f g
0 / U / V / W / 0
a a′′
f ′′ g ′′
0 / X / P′ / W / 0
be a commutative diagram with exact rows. Then the pair (a′′ , f ′′ ) is a pushout of
(f, a).
Proof. We construct the induced exact sequence a∗ (f, g) = (f ′ , g ′): Let (a′ : V →
P, f ′ : X → P ) be the pushout of (f, a). For g ′ : P → W we have g = g ′ a′ .
Since a′′ f = f ′′ a there exists some homomorphism h : P → P ′ with a′′ = ha′
and f ′′ = hf ′ . We claim that g ′ = g ′′ h: This follows from the uniqueness of the
factorization through a pushout, because we know that
g ′ a′ = g = g ′′ a′′ = g ′′ ha′
and
g ′ f ′ = 0 = g ′′ f ′′ = g ′′ hf ′ .
Thus we have seen that h yields an equivalence of the short exact sequences (f ′ , g ′)
and (f ′′ , g ′′ ), In particular, h has to be an isomorphism. But if h is an isomorphism,
then the pair (a′′ = ha′ , f ′′ = hf ′ ) is a pushout of (f, a), since by assumption (a′ , f ′ )
is a pushout of (f, a).
We leave it as an exercise to prove the corresponding dual of the above lemma:

Lemma 25.11. Let
f ′′′ g ′′′
0 / U / Q′ / Y / 0
b′′′ b
f g
0 / U / V / W / 0
be a commutative diagram with exact rows. Then the pair (b′′′ , g ′′′) is a pullback of
(g, b).
25.5. Examples. Let

f g
0 → N(2) −
→ N(3) −
→ N(1) → 0
be a short exact sequence, and let h : N(2) → N(1) be a homomorphism. As before,
N(m) is the m-dimensional 1-module (K m , φ) with basis e1 , . . . , em , φ(e1 ) = 0 and
φ(ei ) = ei−1 for all 2 ≤ i ≤ m. We will fix such bases for each m, and display the
homomorphisms f , g and h as matrices: For example, let
 
1 0
f = 0 1 and g = [0, 0, 1].
0 0
For h = [0, 1], the induced sequence h∗ (f, g) is of the form
0 → N(1) → N(2) → N(1) → 0.
For h = [0, 0], the induced sequence h∗ (f, g) is of the form
0 → N(1) → N(1) ⊕ N(1) → N(1) → 0.
Similarly as above let

f g
0 → N(3) −
→ N(4) −
→ N(1) → 0
be the obvious canonical short exact sequence, and let h : N(3) → N(2) be the
homomorphism given by the matrix

0 1 0
.
0 0 1
Thus h is the canonical epimorphism from N(3) onto N(2). Now one can check that
the pushout of (f, h) is isomorphic to N(3).
On the other hand, if h is given by the matrix

0 0 1
.
0 0 0
Then the pushout of (f, h) is isomorphic to N((2, 1)) = N(2) ⊕ N(1).
25.6. Schanuel’s Lemma. Let A be an algebra, and let

0 → M′ → P → M → 0
be a short exact sequence of A-modules. If M is an arbitrary module, then such a
sequence exists, because every module is factor module of a projective module. If
we fix M and look at all possible sequence of the above form, then the modules M ′
are similar to each other, in fact they are “stably equivalent”.
Every module can be written as a factor module of a projective module, but the
submodules of projective modules are in general a very special class of modules.
For example there are algebras, where submodules of projective modules are always
projective.
An A-module M is called torsion free if M is isomorphic to a submodule of a

projective module.
Lemma 25.12 (Schanuel). Let P and Q be projective modules, let U be a submodule
= Q/V , then U ⊕ Q ∼
of P and V a submodule of Q. If P/U ∼ = V ⊕ P.
Proof. Let M := P/U, and let p : P → M be the projection map. Similarly, let
q : Q → M be the epimorphism with kernel V (since Q/V is isomorphic to M such
a q exists).
We construct the pullback of (p, q) and obtain a commutative diagram
p′
E / Q
q′ q
p
P / M
where p is an epimorphism with kernel isomorphic to U = Ker(p), and q ′ is an
′
epimorphism with kernel isomorphic to V = Ker(q): Set U ′ := Ker(p′ ) and V ′ :=

Ker(q ′ ). We get a diagram
0 0
iV
V ′ _ _ _/ V
p′
0 / U ′ / E / Q / 0
iU
q′ q

p
0 / U / P / M / 0

0 0
We can assume E = {(v, w) ∈ P ⊕ Q | p(v) = q(w)}, q ′ (v, w) = v and p′ (v, w) = w
for all (v, w) ∈ E. Now it is easy to define homomorphisms iU and iV such that
everything commutes, and then ones shows that iU and iV are in fact isomorphisms.
Since Q is projective, p′ is a split epimorphism, which implies E ∼= U ⊕ Q. Since P

is projective as well, q ′ is a split epimorphism, thus E ∼
= V ⊕ P.
f
25.7. Short exact sequences with projective middle term. Let η : 0 → X − →
g ′ ′
′ ′ f ′ g
Y −→ Z → 0 and η : 0 → X − →Y − → Z → 0 be short exact sequences. We say
that η induces η if there exist homomorphisms h and h′ such that the diagram
′
f g
0 / X / Y / Z / 0
h h′
f′ g′
0 / X′ / Y′ / Z / 0
commutes.
Lemma 25.13. Let
u p
0→U − →P − →W →0
be a short exact sequence of A-modules with P a projective module. Then this se-
quence induces every short exact sequence which ends in W .
Proof. Let
f g
0 → U′ −
→ V′ −
→W →0
be an arbitrary short exact sequence of A-modules. Since g is an epimorphism, the
lifting property of the projective module P yields a homomorphism p′ : P → V ′ such
that gp′ = p. This implies gp′ u = pu = 0. Thus p′ u can be factorized through the
kernel of g. So there exists some h : U → U ′ such that p′ u = f h. Thus we obtain
the following commutative diagram with exact rows:
u p
0 / U / P / W / 0
h p′
f g
0 / U′ / V′ / W / 0
This shows that (f, g) is the short exact sequence induced by h.
25.8. Exercises. 1: Let

f1 g1
0→U −
→ V1 −
→W →0
and
f2 g2
0→U − → V2 −→W →0
be equivalent short exact sequences of J-modules, and let a : U → X be a homo-
morphism. Show that the two short exact sequences a∗ (f1 , g1 ) and a∗ (f2 , g2 ) are
equivalent.
2: Recall: For any partition λ, we defined a 1-module N(λ). Let
f1 g1
0 → N(n) −
→ N(2n) −
→ N(n) → 0
be the short exact sequence with f1 the canonical inclusion and g1 the canonical
projection, and let
f2 g2
η: 0 → N(n) −
→ N(λ) −
→ N(n) → 0
be a short exact sequence with λ = (λ1 , λ2 ).
Show: There exists some homomorphism a : N(n) → N(n) such that a∗ (f1 , g1 ) = η.
3: Let
f g
0→U − →V − →W →0
be a short exact sequence of J-modules, and let a : U → X, a′ : X → X ′ , b : Y → W ,
b′ : Y ′ → Y be homomorphisms of J-modules. Show:
• The induced sequences (a′ a)∗ (f, g) and a′∗ (a∗ (f, g)) are equivalent;
• The induced sequences (bb′ )∗ (f, g) and (b′ )∗ (b∗ (f, g)) are equivalent;
• The induced sequences a∗ (b∗ (f, g)) and b∗ (a∗ (f, g)) are equivalent.
————————————————————————————-
26. Homological Algebra
26.1. The Snake Lemma.

Theorem 26.1 (Snake Lemma). Given the following commutative diagram of ho-
momorphisms
f1 g1
U1 / V1 / W1 / 0
a b c
f2 g2
0 / U2 / V2 / W2
such that the two rows are exact. Taken kernels and cokernels of the homomorphisms
a, b, c we obtain a commutative diagram
0 0 0
f0 g0
U0 / V0 / W0
a0 b0 c0
f1 g1
U1 / V1 / W1 / 0
a b c
f2 g2
0 / U2 / V2 / W2
a2 b2 c2
f3 g3
U3 / V3 / W3

0 0 0
with exact rows and columns. Then
δ(x) := (a2 ◦ f2−1 ◦ b ◦ g1−1 ◦ c0 )(x)
defines a homomorphism (the “connecting homomorphism”)
δ : Ker(c) → Cok(a)
such that the sequence
f0 g0 δ f3 g3
Ker(a) −
→ Ker(b) −
→ Ker(c) −
→ Cok(a) −
→ Cok(b) −
→ Cok(c)
is exact.
Proof. The proof is divided into two steps: First, we define the map δ, second we
verify the exactness.
Relations
We need some preliminary remarks on relations: Let V and W be modules. A

submodule ρ ⊆ V × W is called a relation. If f : V → W is a homomorphism, then
the graph
Γ(f ) = {(v, f (v)) | v ∈ V }
of f is a relation. Vice versa, a relation ρ ⊆ V ×W is the graph of a homomorphism,
if for every v ∈ V there exists exactly one w ∈ W such that (v, w) ∈ ρ.
If ρ ⊆ V × W is a relation, then the opposite relation is defined as ρ−1 = {(w, v) |
(v, w) ∈ ρ}. Obviously this is a submodule again, namely of W × V .
If V1 , V2 , V3 are modules and ρ ⊆ V1 × V2 and σ ⊆ V2 × V3 are relations, then
σ ◦ ρ := {(v1 , v3 ) ∈ V1 × V3 | there exists some v2 ∈ V2 with (v1 , v2 ) ∈ ρ, (v2 , v3 ) ∈ σ}
is the composition of ρ and σ. It is easy to check that σ ◦ ρ is a submodule of
V1 × V3 .
For homomorphisms f : V1 → V2 and g : V2 → V3 we have Γ(g) ◦ Γ(f ) = Γ(gf ).

The composition of relations is associative: If ρ ⊆ V1 ×V2 , σ ⊆ V2 ×V3 and τ ⊆ V3 ×V4
are relations, then (τ ◦ σ) ◦ ρ = τ ◦ (σ ◦ ρ).
Let ρ ⊆ V × W be a relation. For a subset X of V define ρ(X) = {w ∈ W | (x, w) ∈
ρ for some x ∈ X}. If x ∈ V , then set ρ(x) = ρ({x}).
For example, if f : V → W is a homomorphism and X a subset of V , then

(Γ(f ))(X) = f (X).
Similarly, (Γ(f )−1 )(Y ) = f −1 (Y ) for any subset Y of W .
Thus in our situation, a2 f2−1 bg1−1 c0 stands for

Γ(a2 ) ◦ Γ(f2 )−1 ◦ Γ(b) ◦ Γ(g1 )−1 ◦ Γ(c0 ).
First, we claim that this is indeed the graph of some homomorphism δ.
δ is a homomorphism
We show that a2 f2−1 bg1−1 c0 is a homomorphism: Let S be the set of tuples

(w0 , w1 , v1 , v2 , u2 , u3) ∈ W0 × W1 × V1 × V2 × U2 × U3
such that
w1 = c0 (w0 ) = g1 (v1 ),
v2 = b(v1 ) = f2 (u2 ),
u3 = a2 (u2 ).
w_0
c0
g1
v_1 / w1
b
f2
u_2 / v2
a2

u3
We have to show that for every w0 ∈ W0 there exists a tuple
(w0 , w1 , v1 , v2 , u2 , u3)
in S, and that for two tuples (w0 , w1 , v1 , v2 , u2 , u3 ) and (w0′ , w1′ , v1′ , v2′ , u′2 , u′3) with
w0 = w0′ we always have u3 = u′3 .
Thus, let w ∈ W0 . Since g1 is surjective, there exists some v ∈ V1 with g1 (v) = c0 (w).
We have
g2 b(v) = cg1 (v) = cc0 (w) = 0.
Therefore b(v) belongs to the kernel of g2 and also to the image of f2 . Thus there
exists some u ∈ U2 with f2 (u) = b(v). So we see that
(w, c0 (w), v, b(v), u, a2(u)) ∈ S.
Now let (w, c0(w), v ′, b(v ′ ), u′ , x) also be in S. We get
g1 (v − v ′ ) = c0 (w) − c0 (w) = 0.
Thus v − v ′ belongs to the kernel of g1 , and therefore to the image of f1 . So there
exists some y ∈ U1 with f1 (y) = v − v ′ . This implies
f2 (u − u′ ) = b(v − v ′ ) = bf1 (y) = f2 a(y).
Since f2 is injective, we get u − u′ = a(y). But this yields
a2 (u) − x = a2 (u − u′) = a2 a(y) = 0.
Thus we see that a2 (u) = x, and this implies that δ is a homomorphism.
Exactness
Next, we want to show that Ker(δ) = Im(g0 ): Let x ∈ V0 . To compute δg0 (x)
we need a tuple (g0 (x), w1 , v1 , v2 , u2 , u3) ∈ S. Since g1 b0 = c0 g0 and bb0 = 0 we
can choose (g0 (x), c0 g0 (x), b0 (x), 0, 0, 0). This implies δg0 (x) = 0. Vice versa, let
w ∈ Ker(δ). So there exists some (w, w1, v1 , v2 , u2, 0) ∈ S. Since u2 belongs to the
kernel of a2 and therefore to the image of a, there exists some y ∈ U1 with a(y) = u2 .
We have
bf1 (y) = f2 a(y) = f2 (u2 ) = b(v1 ).
Thus v1 − f1 (y) is contained in Ker(b). This implies that there exists some x ∈ V0
with b0 (x) = v1 − f1 (y). We get
c0 g0 (x) = g1 b0 (x) = g1 (v1 − f1 (y)) = g1 (v1 ) = c0 (w).
Since c0 is injective, we have g0 (x) = w. So we see that w belongs to the image of

g0 .
Finally, we want to show that Ker(f3 ) = Im(δ): Let (w0 , w1, v1 , v2 , u2 , u3) ∈ S, in
other words δ(w0 ) = u3 . We have
f3 (u3 ) = f3 a2 (u2 ) = b2 f2 (u2 ).
Since f2 (u2) = v2 = b(v1 ), we get b2 f2 (u2) = b2 b(v1 ) = 0. This shows that the
image of δ is contained in the kernel of f3 . Vice versa, let u3 be an element in U3 ,
which belongs to the kernel of f3 . Since a2 is surjective, there exists some u2 ∈ U2
with a2 (u2 ) = u3 . We have b2 f2 (u2 ) = f3 a2 (u2 ) = f3 (u3 ) = 0, and therefore f2 (u2 )
belongs to the kernel of b2 and also to the image of b. Let f2 (u2 ) = b(v1 ) =: v2 .
This implies cg1 (v1 ) = g2 b(v1 ) = g2 f2 (u2 ) = 0. We see that g1 (v1 ) is in the kernel
of c and therefore in the image of c0 . So there exists some w0 ∈ W0 with c0 (w0 ) =
g1 (v1 ). Altogether, we constructed a tuple (w0 , w1 , v1 , v2 , u2 , u3 ) in S. This implies
u3 = δ(w0 ). This finishes the proof of the Snake Lemma.
Next, we want to show that the connecting homomorphism is “natural”: Assume

we have two commutative diagrams with exact rows:
f1 g1
U1 / V1 / W1 / 0
a b c
f2 g2
0 / U2 / V2 / W2 ,
f1′ g1′
U1′ / V1′ / W1′ / 0
a′ b′ c′
f2′ g2′
0 / U2′ / V2′ / W2′ .
Let δ : Ker(c) → Cok(a) and δ ′ : Ker(c′ ) → Cok(a′ ) be the corresponding connecting
homomorphisms.
Additionally, for i = 1, 2 let pi : Ui → Ui′ , qi : Vi → Vi′ and ri : Wi → Wi′ be homo-

morphisms such that the following diagram is commutative:
U1 A / V1 @ / W1 D / 0
AAp1 @@q1 DDr1
A @ D!
U1′ / V′
1
/ W′
1
/ 0

0 / U2 A / V2 @ / W2 D
AAp2 @@q2 DDr2
A @ D!
0 / U′ / V′ / W′
2 2 2
The homomorphisms pi : Ui → Ui′ induce a homomorphism p3 : Cok(a) → Cok(a′ ),

and the homomorphisms ri : Wi → Wi′ induce a homomorphism r0 : Ker(c) →
Ker(c′ ).
Lemma 26.2. The diagram

δ
Ker(c) / Cok(a)
r0 p3

δ′
Ker(c′ ) / Cok(a′ )
is commutative.
Proof. Again, let S be the set of tuples

(w0 , w1 , v1 , v2 , u2 , u3) ∈ W0 × W1 × V1 × V2 × U2 × U3
such that
w1 = c0 (w0 ) = g1 (v1 ),
v2 = b(v1 ) = f2 (u2 ),
u3 = a2 (u2 ),
and let S ′ be the correspondingly defined subset of W0′ × W1′ × V1′ × V2′ × U2′ × U3′ .
Now one easily checks that for a tuple (w0 , w1 , v1 , v2 , u2 , u3 ) in S the tuple
(r0 (w0 ), r1 (w1 ), q1 (v1 ), q2 (v2 ), p2 (u2 ), p3 (u3))
belongs to S ′ . The claim follows.
26.2. Complexes. A complex of A-modules is a tuple C• = (Cn , dn )n∈Z (we often

just write (Cn , dn )n or (Cn , dn )) where the Cn are A-modules and the dn : Cn → Cn−1
are homomorphisms such that
Im(dn ) ⊆ Ker(dn−1 )
for all n, or equivalently, such that dn−1dn = 0 for all n.
dn+2 dn+1
n d dn−1
· · · −−−
→ Cn+1 −−−
→ Cn −→ Cn−1 −−−→ · · ·
A cocomplex is a tuple C • = (C n , dn )n∈Z where the C n are A-modules and the

dn : C n → C n+1 are homomorphisms such that dn+1 dn = 0 for all n.
dn−2 dn−1 dn dn+1
· · · −−−→ C n−1 −−−→ C n −→ C n+1 −−−→ · · ·
Remark: We will mainly formulate results and definitions by using complexes, but
there are always corresponding results and definitions for cocomplexes. We leave it
to the reader to perform the necessary reformulations.
In this lecture course we will deal only with (co)complexes of modules over a K-
algebra A and with (co)complexes of vector spaces over the field K.
A complex C• = (Cn , dn )n∈Z is an exact sequence of A-modules if
Im(dn ) = Ker(dn−1)
for all n. In this case, for a > b we also call

a d da−1 db+1
Ca −→ Ca−1 −→ · · · −→ Cb ,
db+2 db+1
· · · −→ Cb+1 −→ Cb ,
a d da−1
Ca −→ Ca−1 −→ · · ·
exact sequences. An exact sequence of the form
f g
0 −→ X −→ Y −→ Z → 0
is a short exact sequence. We denote such a sequence by (f, g). Note that this
implies that f is a monomorphism and g is an epimorphism.
Example: Let M be an A-module, and let C• = (Cn , dn )n∈Z be a complex of
A-modules. Then
HomA (M, C• ) = (HomA (M, Cn ), HomA (M, dn ))n∈Z
is a complex of K-vector spaces and
HomA (C• , M) = (HomA (Cn , M), HomA (dn+1 , M))n∈Z
is a cocomplex of K-vector spaces. (Of course, K is a K-algebra, and the K-modules
are just the K-vector spaces.)
Given two complexes C• = (Cn , dn )n∈Z and C•′ = (Cn′ , d′n )n∈Z , a homomorphism
of complexes (or just map of complexes) is given by f• = (fn )n∈Z : C• → C•′
where the fn : Cn → Cn′ are homomorphisms with d′n fn = fn−1 dn for all n.
dn+1 dn
··· / Cn+1 / Cn / Cn−1 / ···
fn+1 fn fn−1
d′n+1 d′n

′ ′
··· / Cn+1 / Cn′ / Cn−1 / ···
The maps C• → C•′ of complexes form a vector space: Let f• , g• : C• → C•′ be such
maps, and let λ ∈ K. Define f• + g• := (fn + gn )n∈Z , and let λf• := (λfn )n∈Z .
If f• = (fn )n : C• → C•′ and g• = (gn )n : C•′ → C•′′ are maps of complexes, then the
composition
g• f• = g• ◦ f• : C• → C•′′
is defined by g• f• := (gn fn )n .
Let C• = (Cn , dn )n be a complex. A subcomplex C•′ = (Cn′ , d′n )n of C• is given by
submodules Cn′ ⊆ Cn such that d′n is obtain via the restriction of dn to Cn′ . (Thus we
require that dn (Cn′ ) ⊆ Cn−1
′
for all n.) The corresponding factor complex C• /C•′
is of the form (Cn /Cn′ , d′′n )n where d′′n is the homomorphism Cn /Cn′ → Cn−1 /Cn−1
′
induced by dn .
Let f• = (fn )n : C•′ → C• and g• = (gn )n : C• → C•′′ be homomorphisms of com-
plexes. Then
f• g•
0 → C•′ − → C•′′ → 0
→ C• −
is a short exact sequence of complexes provided

fn gn
0 → Cn′ −→ Cn −→ Cn′′ → 0
is a short exact sequence for all n.
26.3. From complexes to modules. We can interpret complexes of J-modules

(here we use our terminology from the first part of the lecture course) as J ′ -modules
where
J ′ := J ∪ Z ∪ {d}.
(We assume that J, Z and {d} are pairwise disjoint sets.)
If C• = (Cn , dn )n is a complex of J-modules, then we consider the J-module
M
C := Cn .
n∈Z
We add some further endomorphisms of the vector space C, namely for n ∈ Z

take the projection φn : C → C onto Cn and additionally take φd : C → C whose
restriction to Cn is just dn . This converts C into a J ′ -module.
Now if f• = (fn )n : C• → C•′ is a homomorphism of complexes, then
M M M
fn : Cn → Cn′
n∈Z n∈Z n∈Z
defines a homomorphism of J ′ -modules, and one obtains all homomorphisms of J ′ -

modules in such a way.
We can use this identification of complexes of J-modules with J ′ -modules for trans-
ferring the terminology we developed for modules to complexes: For example sub-
complexes or factor complexes can be defined as J ′ -submodules or J ′ -factor modules.
26.4. Homology of complexes. Given a complex C• = (Cn , dn )n define
Hn (C• ) = Ker(dn )/ Im(dn+1 ),
the nth homology module (or homology group) of C• . Set H• (C• ) = (Hn (C• ))n .
Similarly, for a cocomplex C • = (C n , dn ) let
H n (C • ) = Ker(dn )/ Im(dn−1 )
be the nth cohomology group of C • .

Each homomorphism f• : C• → C•′ of complexes induces homomorphisms
Hn (f• ) : Hn (C• ) → Hn (C•′ ).

(One has to check that fn (Im(dn+1 )) ⊆ Im(d′n+1 ) and fn (Ker(dn )) ⊆ Ker(d′n ).)
fn+1 ′
Cn+1 / Cn+1
dn+1 d′n+1
fn

Cn / Cn′
dn d′n
fn−1

′
Cn−1 / Cn−1
It follows that Hn defines a functor from the category of complexes of A-modules to

the category of A-modules.
Let C• = (Cn , dn ) be a complex. We consider the homomorphisms
dn+1 n d dn−1
Cn+1 −−−
→ Cn −→ Cn−1 −−−→ Cn−2 .
By assumption we have Im(di+1 ) ⊆ Ker(di) for all i.
The following picture illustrates the situation. Observe that the homology groups
Hi (C• ) = Ker(di )/ Im(di+1 )
are highlighted by the thick vertical lines. The marked regions indicate which parts
of Ci and Ci−1 get identified by the map di . Namely di induces an isomorphism
Ci / Ker(di) → Im(di ).
dn+1 - Cn
dn -
dn−1 -
Cn+1 Cn−1 Cn−2
Ci rH rH rH r
HH HH HH
HH HH HH
Ker(di ) r
HH HH
r
HH HH
r
HH HH r
HH HH HH HH HH HH
HH HH HH HH HH HH
Im(di+1 ) r
HH
r
HH
r
HH
r
HH HH HH
0 r Hr Hr Hr
The map dn factors through Ker(dn−1 ) and the map Cn → Ker(dn−1 ) factors through
Cok(dn+1 ). Thus we get an induced homomorphism dn : Cok(dn+1 ) → Ker(dn−1 ).
The following picture describes the situation:
dn - Ker(dn−1 )
Cok(dn+1)
HH
r
HH
rH HH
HH
r
HH
HH HH
r HH Hr
HH
HH
r
So we obtain a commutative diagram

dn
Cn / Cn−1
O

dn
Cok(dn+1) / Ker(dn−1 )
The kernel of dn is just Hn (C• ) and its cokernel is Hn−1 (C• ). Thus we obtain an
exact sequence
n iC n d pC
n−1
0 → Hn (C• ) −→ Cok(dn+1 ) −→ Ker(dn−1 ) −−−→ Hn−1 (C• ) → 0
where iC C
n and pn−1 denote the inclusion and the projection, respectively. The inclu-
sion Ker(dC C
n ) → Cn is denoted by un .
26.5. Homotopy of morphisms of complexes. Let C• = (Cn , dn ) and C•′ =

(Cn′ , d′n ) be complexes, and let f• , g• : C• → C•′ be homomorphisms of complexes.
Then f• and g• are called homotopic if for all n ∈ Z there exist homomorphisms
′
sn : Cn → Cn+1 such that
hn := fn − gn = d′n+1sn + sn−1 dn .
In this case we write f• ∼ g• . (This defines an equivalence relation.) The sequence

s = (sn )n is a homotopy from f• to g• .
dn+1 dn
··· / Cn+1 /C
n
/ Cn−1 / ···
z sn−1 zzz
sn zzz
z h z
z n
zz
}zz }zz
′ / C′ / C′
··· / Cn+1 n n−1
/ ···
d′n+1
The morphism f• : C• → C•′ is zero homotopic if f• and the zero homomorphism

0 : C• → C•′ are homotopic. The class of zero homotopic homomorphisms forms an
ideal in the category of complexes of A-modules.
Proposition 26.3. If f• , g• : C• → C•′ are homomorphisms of complexes such that

f• and g• are homotopic, then Hn (f• ) = Hn (g• ) for all n ∈ Z.
Proof. Let C• = (Cn , dn ) and C•′ = (Cn′ , d′n ), and let x ∈ Ker(dn ). We get
fn (x) − gn (x) = (fn − gn )(x)

= (d′n+1 sn + sn−1 dn )(x)
= d′n+1 sn (x)
since dn (x) = 0. This shows that fn (x) and gn (x) only differ by an element in
Im(d′n+1 ). Thus they belong to the same residue class modulo Im(d′n+1 ).
Corollary 26.4. Let f• : C• → C•′ be a homomorphism of complexes. Then the

following hold:
(i) If f• is zero homotopic, then Hn (f• ) = 0 for all n;

(ii) If there exists a homomorphism g• : C•′ → C• such that g• f• ∼ 1C• and
f• g• ∼ 1C•′ , then Hn (f• ) is an isomorphism for all n.
Proof. As in the proof of Proposition 26.3 we show that fn (x) ∈ Im(d′n+1). This
implies (i). We have Hn (g• )Hn (f• ) = Hn (g• f• ) = Hn (1C• ) and Hn (f• )Hn (g• ) =
Hn (f• g• ) = Hn (1′C• ). Thus Hn (f• ) is an isomorphism.
26.6. The long exact homology sequence. Let

f• g•
0 → A• −
→ B• −
→ C• → 0
be a short exact sequence of complexes. We would like to construct a homomorphism
δn : Hn (C• ) → Hn−1 (A• ).
Recall that the elements in Hn (C• ) are residue classes of the form x + Im(dC
n+1 ) with
x ∈ Ker(dn ). Here we write A• = (An , dn ), B• = (Bn , dn ) and C• = (Cn , dC
C A B
n ).
For x ∈ Ker(dC
n ) set
δn (x + Im(dC A
n+1 )) := z + Im(dn )
−1 B −1
where z ∈ (fn−1 dn gn )(x).
Theorem 26.5 (Long Exact Homology Sequence). With the notation above, we
obtain a well defined homomorphism
δn : Hn (C• ) → Hn−1 (A• )
and the sequence

δn+1 Hn (f• ) Hn (g• ) n δ Hn−1 (f• )
· · · −−→ Hn (A• ) −−−−→ Hn (B• ) −−−−→ Hn (C• ) −→ Hn−1 (A• ) −−−−−→ · · ·
is exact.
Proof. Taking kernels and cokernels of the maps dA B B

n , dn and dn we obtain the fol-
lowing commutative diagram with exact rows and columns:
0 0 0
fn′
′
gn

0 / Ker(dA
n)
/ Ker(dB
n)
/ Ker(dC
n)
fn gn
0 / An / Bn / Cn / 0
dA
n dB
n dC
n
fn−1 gn−1
0 / An−1 / Bn−1 / Cn−1 / 0
′′
fn−1 ′′
gn−1
Cok(dA
n)
/ Cok(dB
n)
/ Cok(dC
n)
/ 0

0 0 0
(The arrows without label are just the canonical inclusions and projections, re-
spectively. By fn′ , gn′ and fn−1
′′ ′′
, gn−1 we denote the induced homomorphisms on the
kernels and cokernels of the maps dA B C
n , dn and dn , respectively.)
The map fn′ is a restriction of the monomorphism fn , thus fn′ is also a monomor-
phism. Since gn−1 is an epimorphism and gn−1 (Im(dB C
n )) ⊆ Im(dn ), we know that
′′
gn−1 is an epimorphism as well.
We have seen above that the homomorphism dA

n : An → An−1 induces a homomor-
phism
A A
a = dA
n : Cok(dn+1 ) → Ker(dn−1 ).
Similarly, we obtain b = dB C
n and c = dn . The kernels and cokernels of these homo-
morphisms are homology groups. We obtain the following commutative diagram:
0 0 0
Hn (f• ) Hn (g• )
Hn (A• ) / Hn (B• ) / Hn (C• )
iA
n iB
n iC
n
fn′′
′′
gn

Cok(dA
n+1 )
/ Cok(dB
n+1 )
/ Cok(dC
n+1 )
/ 0
a b c
′
fn−1 ′
gn−1
0 / Ker(dA
n−1 )
/ Ker(dB
n−1 )
/ Ker(dC
n−1 )
pA
n−1 pB
n−1 pC
n−1
Hn−1 (f• ) Hn−1 (g• )
Hn−1 (A• ) / Hn−1 (B• ) / Hn−1 (C• )

0 0 0
Now we can apply the Snake Lemma: For our n we obtain a connecting homomor-
phism
δ : Hn (C• ) → Hn−1 (A• )
which yields the required exact sequence. It remains to show that δ = δn .
Let T be the set of all triples (x, y, z) with x ∈ Ker(dC

n ), y ∈ Bn , z ∈ An−1 such that
gn (y) = x and fn−1 (z) = dB n (y).
(1) For every x ∈ Ker(dC
n ) there exists a triple (x, y, z) ∈ T :
Let x ∈ Ker(dC n ). Since gn is surjective, there exists some y ∈ Bn with gn (y) = x.

We have
gn−1dB C C
n (y) = dn gn (y) = dn (x) = 0.
Thus dBn (y) belongs to the kernel of gn−1 and therefore to the image of fn−1 . Thus
there exists some z ∈ An−1 with fn−1 (z) = dB n (y).
(2) If (x, y1 , z1 , ), (x, y2, z2 ) ∈ T , then z1 − z2 ∈ Im(dA

n ):
We have gn (y1 − y2 ) = x − x = 0. Since Ker(gn ) = Im(fn ) there exists some an ∈ An

such that fn (an ) = y1 − y2 . It follows that
fn−1 dA B B
n (an ) = dn fn (an ) = dn (y1 − y2 ) = fn−1 (z1 − z2 ).
Since fn−1 is a monomorphism, we get dA A

n (an ) = z1 − z2 . Thus z1 − z2 ∈ Im(dn ).
(3) If (x, y, z) ∈ T and x ∈ Im(dC A

n+1 ), then z ∈ Im(dn ):
Let x = dCn+1 (r) for some r ∈ Cn+1 . Since gn+1 is surjective there exists some
s ∈ Bn+1 with gn+1 (s) = r. We have
gn (y) = x = dC C B
n+1 (r) = dn+1 gn+1 (s) = gn dn+1 (s).
Therefore y − dBn+1 (s) is an element in Ker(gn ) and thus also in the image of fn . Let
y − dB
n+1 (s) = fn (t) for some t ∈ An . We get
fn−1 dA B B B B B
n (t) = dn fn (t) = dn (y) − dn dn+1 (s) = dn (y) = fn−1 (z).
Since fn−1 is injective, this implies dA A

n (t) = z. Thus z is an element in Im(dn ).
(4) If (x, y, z) ∈ T , then z ∈ Ker(dA

n−1 ):
We have
fn−2 dA B B B
n−1 (z) = dn−1 fn−1 (z) = dn−1 dn (y) = 0.
Since fn−2 is injective, we get dA n−1 (z) = 0.
Combining (1),(2),(3) and (4) yields a homomorphism δn : Hn (C• ) → Hn−1 (A• )

defined by
δn (x + Im(dC A
n+1 )) := z + Im(dn )
for each (x, y, z) ∈ T .
The set of all pairs (pC A

n (x), pn−1 (z)) such that there exists a triple (x, y, z) ∈ T is
given by the relation
Γ(pA A
n−1 ) ◦ Γ(un−1 )
−1
◦ Γ(fn−1 )−1 ◦ Γ(dB
n ) ◦ Γ(gn )
−1
◦ Γ(uC C −1
n ) ◦ Γ(pn ) .
This is the graph of our homomorphism δn .

pC
n
Ker(dC
n)
/ Hn (C• )
uC
n
gn
Bn / Cn
dB
n
pA
n−1 uA
n−1 fn−1
Hn−1 (A• ) o Ker(dA
n−1 )
/ An−1 / Bn−1
Now it is not difficult to show that this relation coincides with the relation
Γ(pA ′
n−1 ) ◦ Γ(fn−1 )
−1
◦ Γ(b) ◦ Γ(gn′′ )−1 ◦ Γ(iC
n)
which is the graph of δ.

Hn (C• )
iC
n
′′
gn

Cok(dB
n+1 )
/ Cok(dC
n+1 )
b
′
fn−1
Ker(dA
n−1 )
/ Ker(dB
n−1 )
pA
n−1

Hn−1 (A• )
This implies δ = δn .
The exact sequence in the above theorem is called the long exact homology
sequence associated to the given short exact sequence of complexes. The homo-
morphisms δn are called connecting homomorphisms.
The connecting homomorphisms are “natural”: Let
f• g•
0 / A• / B• / C• / 0
p• q• r•
f•′ g•′
0 / A′• / B•′ / C•′ / 0
be a commutative diagram with exact rows. Then the diagram
δn
Hn (C• ) / Hn−1 (A• )
Hn (r• ) Hn−1 (p• )
′
δn

Hn (C•′ ) / Hn−1 (A′• )
commutes, where δn and δn′ are the connecting homomorphisms coming from the
two exact rows.
————————————————————————————-
27. Projective resolutions and extension groups
27.1. Projective resolutions. Let Pi , i ≥ 0 be projective modules, and let M be

an arbitrary module. Let pi : Pi → Pi−1 , i ≥ 1 and ε : P0 → M be homomorphisms
such that
pi+1 pi p2 p1 ε
· · · → Pi+1 −−→ Pi −
→ ··· − → P1 −→ P0 − →M →0
is an exact sequence. Then we call
pi+1 pi p2 p1
P• := (· · · → Pi+1 −−→ Pi −
→ ··· −
→ P1 −
→ P0 )
a projective resolution of M. We think of P• as a complex of A-modules: Just
set Pi = 0 and pi+1 = 0 for all i < 0.
Define
ΩP• (M) := Ω1P• (M) := Ker(ε),
and let ΩiP• (M) = Ker(pi−1 ), i ≥ 2. These are called the syzygy modules of M
with respect to P• . Note that they depend on the chosen projective resolution.
If all Pi are free modules, we call P• a free resolution of M.
The resolution P• is a minimal projective resolution of M if the homomorphisms
Pi → ΩiP• (M), i ≥ 1 and also ε : P0 → M are projective covers. In this case, we call
Ωn (M) := ΩnP• (M)
the nth syzygy module of M. This does not depend on the chosen minimal
projective resolution.
Lemma 27.1. If
0→U →P →M →0
is a short exact sequence of A-modules with P projective, then U ∼
= Ω(M) ⊕ P ′ for
′
some projective module P .
Proof. Exercise.
Sometimes we are a bit sloppy when we deal with syzygy modules: If there exists
a short exact sequence 0 → U → P → M → 0 with P projective, we just write
Ω(M) = U, knowing that this is not at all well defined and depends on the choice
of P .
Lemma 27.2. For every module M there is a projective resolution.
Proof. Define the modules Pi inductively. Let ε = ε0 : P0 → M be an epimorphism

with P0 a projective module. Such an epimorphism exists, since every module is
isomorphic to a factor module of a free module. Let µ1 : Ker(ε0 ) → P0 be the
inclusion. Let ε1 : P1 → Ker(ε0 ) be an epimorphism with P1 projective, and define
p1 = µ1 ◦ ε1 : P1 → P0 . Now let ε2 : P2 → Ker(ε1 ) be an epimorphism with P2
projective, etc.
The first row of the resulting diagram
p3 p2 p1 ε
··· /P
> 2 GGG
/P
1 G
/P
0
/ M / 0
} ww; GG ww;
} GG w GG w
}
}}µ3 GG www GG ww
ε2 GG µ ε GG ww
ww µ1
} w 2 1
}} # ww #
··· Ker(ε1 ) Ker(ε0 )
is exact, and we get a projective resolution
p3 p2 p1
··· −
→ P2 −
→ P1 −
→ P0
of Cok(p1 ) = M.
Theorem 27.3. Given a diagram of homomorphisms with exact rows
p3 p2 p1 ε
··· / P2 / P1 / P0 / M / 0
f
p′3 p′2 p′1 ε′

··· / P2′ / P1′ / P0′ / N / 0
where the Pi and Pi′ are projective. Then the following hold:
(i) There exists a “lifting” of f , i.e. there are homomorphisms fi : Pi → Pi′ such
that
p′i fi = fi−1 pi and ε′ f0 = f ε
for all i;
(ii) Any two liftings f• = (fi )i≥0 and f•′ = (fi′ )i≥0 are homotopic.
Proof. (i): The map ε′ : P0′ → N is an epimorphism, and the composition f ε : P0 →

N is a homomorphism starting in a projective module. Thus there exists a homo-
morphism f0 : P0 → P0′ such that ε′ f0 = f ε.
We have Im(p1 ) = Ker(ε) and Im(p′1 ) = Ker(ε′ ). So we obtain a diagram with exact
rows of the following form:
p3 p2 p1
··· / P2 / P1 / Im(p1 ) / 0
fe0
p′3 p′2 p′1
··· / P2′ / P1′ / Im(p′1 ) / 0
The homomorphism fe0 is obtained from f0 by restriction to Im(p1 ). Since P1 is
projective, and since p′1 is an epimorphism there exists a homomorphism f1 : P1 → P1′
such that p′1 f1 = fe0 p1 , and this implies p′1 f1 = f0 p1 . Now we continue inductively
to obtain the required lifting (fi )i≥0 .
(ii): Assume we have two liftings, say f• = (fi )i≥0 and f•′ = (fi′ )i≥0 . This implies
f ε = ε′ f0 = ε′ f0′
and therefore ε′ (f0 − f0′ ) = 0.
Let ιi : Im(p′i ) → Pi−1
′
be the inclusion and let πi : Pi′ → Im(p′i ) be the obvious
projection. Thus p′i = ιi ◦ πi .
The image of f0 − f0′ clearly is contained in Ker(ε′ ) = Im(p′1 ). Now let s′0 : P0 →
Im(p′1 ) be the map defined by s′0 (m) = (f0 −f0′ )(m). The map π1 is an epimorphism,
and s′0 is a map from a projective module to Im(p′1 ). Thus by the projectivity of P0
there exists a homomorphism s0 : P0 → P1′ such that π1 ◦ s0 = s′0 .
We obtain the following commutative diagram:
ε /
k kk P0 M
k x
s0 kkkkk xxx
kk x f0 −f0′ 0
kkkk xxx s′
kkk {
x 0
u
k
P1′ / Im(p′ )
π1
/ P′
1
/N
ι1 0 ε′
Now assume si−1 : Pi−1 → Pi′ is already defined such that

′
fi−1 − fi−1 = p′i si−1 + si−2 pi−1 .
We claim that p′i (fi − fi′ − si−1 pi ) = 0: We have
p′i (fi − fi′ − si−1 pi ) = p′i fi − p′i fi′ − p′i si−1 pi
′
= fi−1 pi − fi−1 pi − p′i si−1 pi
′
= (fi−1 − fi−1 )pi − p′i si−1 pi
= (p′i si−1 + si−2 pi−1 )pi − p′i si−1 pi
= si−2 pi−1 pi
=0
(since pi−1 pi = 0).

pi pi−1
Pi / Pi−2/ Pi−1
| y
|| yy
fi
|||si−1 fi−1yyysi−2 fi−2
}|| |yy
′ ′ ′
P i
/ P / P i−1 i−1
p′i p′i−1
Therefore
Im(fi − fi′ − si−1 pi ) ⊆ Ker(p′i ) = Im(p′i+1 ).
Let s′i : Pi → Im(p′i+1 ) be defined by s′i (m) = (fi − fi′ − si−1 pi )(m).
′
Since Pi is projective there exists a homomorphism si : Pi → Pi+1 such that πi+1 ◦si =
′
si . Thus we get a commutative diagram
j Pi
jjjjjvjvv
j si
jjj vvv fi −fi′ −si−1 pi
jjjjjjj {
v vv s′
u jj /
j i
′
Pi+1 πi+1
Im(p′ ) / P′
i+1 ιi+1 i
Thus p′i+1 si = fi − fi′ − si−1 pi and therefore fi − fi′ = p′i+1 si + si−1 pi , as required.
This shows that f• − f•′ is zero homotopic. Therefore f• = (fi )i and f•′ = (fi′ )i are
homotopic.
27.2. Ext. Let

pn+1 pn p2 p1
P• = (· · · −−→ Pn −→ · · · −
→ P1 −
→ P0 )
be a projective resolution of M = Cok(p1 ), and let N be any A-module. Define
ExtnA (M, N) := H n (HomA (P• , N)),
the nth cohomology group of extensions of M and N. This definition does not
depend on the projective resolution we started with:
Lemma 27.4. If P• and P•′ are projective resolutions of M, then for all modules N
we have
H n (HomA (P• , N)) ∼= H n (HomA (P•′ , N)).
Proof. Let f• = (fi )i≥0 and g• = (gi )i≥0 be liftings associated to

p3 p2 p1 ε
··· / P2 / P1 / P0 / M / 0
1M
p′3 p′2 p′1 ε′

··· / P2′ / P1′ / P0′ / M / 0
and
p′3 p′2 p′1 ε′
··· / P2′ / P1′ / P0′ / M / 0
1M
p3 p2 p1 ε

··· / P2 / P1 / P0 / M / 0.
By Theorem 27.3 these liftings exist and we have g• f• ∼ 1P• and f• g• ∼ 1P•′ . Thus,
we get a diagram
p3 p 2 p 1
··· /P
2
/P
1
/P
0
~~ ~~ ~~
~ g2 f2 −1P2 ~ g1 f1 −1P1 ~ g0 f0 −1P0
~~ ~~ ~~
~~~ ~~~ ~~~
~ ~ ~
~~~ s2 ~~~ s1 ~~~ s0
~ ~~ ~~
~~ ~~ ~~
~~~ p3 p2 p1
··· / P2 / P1 / P0
such that gi fi − 1Pi = pi+1 si + si−1 pi for all i. (Again we think of P• as a complex
with Pi = 0 for all i < 0.)
Next we apply HomA (−, N) to all maps in the previous diagram and get
HomA (g• f• , N) ∼ HomA (1P• , N).
Similarly, one can show that HomA (f• g• , N) ∼ HomA (1P•′ , N). Now Corollary 26.4
tells us that H n (HomA (g• f• , N)) = H n (HomA (1P• , N)) and H n (HomA (f• g• , N)) =
H n (HomA (1P•′ , N)). Thus
H n (HomA (f• , N)) : H n (HomA (P•′ , N)) → H n (HomA (P• , N))
is an isomorphism.
27.3. Induced maps between extension groups. Let P• be a projective resolu-

tion of a module M, and let g : N → N ′ be a homomorphism. Then we obtain an
induced map
ExtnA (M, g) : H n (HomA (P• , N)) → H n (HomA (P• , N ′ ))
defined by [α] 7→ [g ◦ α]. Here α : Pn → N is a homomorphism with α ◦ pn+1 = 0.
There is also a contravariant version of this: Let f : M → M ′ be a homomorphism,

and let P• and P•′ be projective resolutions of M and M ′ , respectively. Then for any
module N we obtain an induced map
ExtnA (f, N) : H n (HomA (P•′ , N)) → H n (HomA (P• , N))
defined by [β] 7→ [β ◦ fn ]. Here β : Pn′ → N is a homomorphism with β ◦ p′n+1 = 0
and fn : Pn → Pn′ is part of a lifting of f .
27.4. Some properties of extension groups. Obviously, we have ExtnA (M, N) =

0 for all n < 0.
Lemma 27.5. Ext0A (M, N) = HomA (M, N).
Proof. The sequence P1 → P0 → M → 0 is exact. Applying HomA (−, N) yields an

exact sequence
HomA (p1 ,N )
0 → HomA (M, N) → HomA (P0 , N) −−−−−−−→ HomA (P1 , N).
By definition Ext0A (M, N) = Ker(HomA (p1 , N)) = HomA (M, N).
Let M be a module and

µ1 ε
0 → Ω(M) −→ P0 −→ M → 0
a short exact sequences with P0 projective.
Lemma 27.6. Ext1A (M, N) ∼
= HomA (Ω(M), N)/{s ◦ µ1 | s : P0 → N}.
Proof. It is easy to check that HomA (Ω(M), N) ∼

= Ker(HomA (p2 , N)) and {s ◦ µ1 |
s : P0 → N} ∼ = Im(Hom (p
A 1 , N)).
Lemma 27.7. For all n ≥ 1 we have Extn+1 ∼ n
A (M, N) = ExtA (ΩM, N).
p3 p2
Proof. If P• = (Pi , pi)i≥0 is a projective resolution of M, then · · · P3 −
→ P2 −
→ P1 is
a projective resolution of Ω(M).
27.5. Long exact Ext-sequences. Let

0→X→Y →Z→0
be a short exact sequence of A-modules, and let M be any module and P• a projective
resolution of M. Then there exists an exact sequence of cocomplexes
0 → HomA (P• , X) → HomA (P• , Y ) → HomA (P• , Z) → 0.
This induces an exact sequence

0 / HomA (M, X) / HomA (M, Y ) / HomA (M, Z)
ffff ff
f
fffff
ff fffffff
sfffff
Ext1A (M, X) / Ext1 (M, Y )
A f
/ Ext1 (M, Z)
A
f f
ffff ffff
f
fffff
fffffffff
sf
Ext2A (M, X) / Ext2 (M, Y )
A f
/ Ext2 (M, Z)
A
f f
ffff ffff
f
fffff
fffffffff
sf
Ext3A (M, X) / ···
which is called a long exact Ext-sequence.

To obtain a “contravariant long exact Ext-sequence”, we need the following result:
Lemma 27.8 (Horseshoe Lemma). Let
f g
0→X−
→Y −
→Z→0
be a short exact sequence of A-module. Then there exists a short exact sequence of
complexes
η : 0 → P•′ → P• → P•′′ → 0
where P•′ , P• and P•′′ are projective resolutions of X, Y and Z, respectively. We
also have P• ∼= P•′ ⊕ P•′ .
Proof. ...
Let N be any A-module. In the situation of the above lemma, we can apply
HomA (−, N) to the exact sequence η. Since η splits, we obtain an exact sequence
of cocomplexes
0 → HomA (P•′′ , N) → HomA (P• , N) → HomA (P•′ , N) → 0.
Thus we get an exact sequence
0 / HomA (Z, N) / HomA (Y, N) / HomA (X, N)

fff fff
f
fffff
fff ffffff
sffff
Ext1A (Z, N) / Ext1 (Y, N)
A f
/ Ext1 (X, N)
A
f ff
ffff fff
fffff
ffff fffff
s
f
Ext2A (Z, N) / Ext2 (Y, N)
A f
/ Ext2 (X, N)
A
f ff
ffff fff
fffff
fffff
sfffff
Ext3A (Z, N) / ···
which is again called a (contravariant) long exact Ext-sequence.
27.6. Short exact sequences and the first extension group. Let M and N be
modules, and let
pn+1 pn p2 p1
P• = (· · · −→ Pn −→ · · · −→ P1 −→ P0 )
ε
be a projective resolution of M = Cok(p1 ). Let P0 −→ M be the cokernel map of
p1 , i.e.
p1 ε
P1 −→ P0 −→ M −→ 0
is an exact sequence.
We have
H n (HomA (P• , N)) := Ker(HomA (pn+1 , N))/ Im(HomA (pn , N)).
Let [α] := α + Im(HomA (pn , N)) be the residue class of some homomorphism
α : Pn → N with α ◦ pn+1 = 0.
Clearly, we have
Im(HomA (pn , N)) = {s ◦ pn | s : Pn−1 → N} ⊆ HomA (Pn , N).
For an exact sequence

f g
0→N −
→E−
→M →0
let
ψ(f, g)
be the set of homomorphisms α : P1 → N such that there exists some β : P0 → E

with f ◦ α = β ◦ p1 and g ◦ β = ε.
p2 p1 ε
P2 / P1 / P 0 / M / 0
α β

f g
0 / N / E / M / 0
Observe that ψ(f, g) ⊆ HomA (P1 , N).
Lemma 27.9. The set ψ(f, g) is a cohomology class, i.e. it is the residue class of
some element α ∈ Ker(HomA (p2 , N)) modulo Im(HomA (p1 , N)).
Proof. (a): If α ∈ ψ(f, g), then α ∈ Ker(HomA (p2 , N)):

We have
f ◦ α ◦ p2 = β ◦ p1 ◦ p2 = 0.
Since f is a monomorphism, this implies α ◦ p2 = 0.
(b): Next, let α, α′ ∈ ψ(f, g). We have to show that α − α′ ∈ Im(HomA (p1 , N)):
There exist β and β ′ with g ◦ β = ε = g ◦ β ′ , f ◦ α = β ◦ p1 and f ◦ α′ = β ′ ◦ p1 . This
implies g(β − β ′ ) = 0. Since P0 is projective and Im(f ) = Ker(g), there exists some
s : P0 → N with f ◦ s = β − β ′ . We get
f (α − α′ ) = (β − β ′ )p1 = f ◦ s ◦ p1 .
Since f is a monomorphism, this implies α − α′ = s ◦ p1 . In other words, α − α′ ∈
Im(HomA (p1 , N)).
(c): Again, let α ∈ ψ(f, g), and let γ ∈ Im(HomA (p1 , N)). We claim that α + γ ∈
ψ(f, g):
Clearly, γ = s ◦ p1 for some homomorphism s : P0 → N. There exists some β with
g ◦ β = ε and f ◦ α = β ◦ p1 . This implies
f (α + γ) = βp1 + f sp1 = (β + f s)p1 .
Set β ′ := β + f s. We get
gβ ′ = g(β + f s) = gβ + gf s = gβ = ε.
Here we used that g ◦ f = 0. Thus α + γ ∈ ψ(f, g).
Theorem 27.10. The map
ψ : {0 → N → ⋆ → M → 0}/∼ −→ Ext1A (M, N)
(f, g) 7→ ψ(f, g)
defines a bijection between the set of equivalence classes of short exact sequences
f g
0→N −
→⋆−
→M →0
and Ext1A (M, N).
Proof. First we show that ψ is surjective: Let α : P1 → N be a homomorphism with

α ◦ p2 = 0. Let
p2 p1 ε
P2 −
→ P1 −
→ P0 −
→M →0
be a projective presentation of M. Set Ω(M) := Ker(ε).
Thus p1 = µ1 ◦ ε1 where ε1 : P1 → Ω(M) is the projection, and µ1 : Ω(M) → P0 is
the inclusion. Since α ◦ p2 = 0, there exists some α′ : Ω(M) → N with α = α′ ◦ ε1 .
Let (f, g) := α∗′ (µ1 , ε) be the short exact sequence induced by α′ . Thus we have a
commutative diagram
p2 p1 ε
P2 / P1 / P0 / M / 0
α ε1
µ1 ε
0 / Ω(M) / P0 / M / 0
α′ β
f g
0 / N / E / M / 0
This implies α ∈ ψ(f, g).

Next, we will show that ψ is injective: Assume that ψ(f1 , g1 ) = ψ(f2 , g2 ) for two
short exact sequence (f1 , g1 ) and (f2 , g2 ), and let α ∈ ψ(f1 , g1 ). Let α′ : Ω(M) → N
and µ1 : Ω(M) → P0 be as before. the restriction of α to Ω(M) and let p′′1 : Ω(M) →
P0 be the obvious inclusion.
We obtain a diagram
µ1 ε
0 / Ω(M) / P0 / M / 0
α′ β1 β2
f1 g1
0 / N / E 1 / M / 0
γ

f2 g2
0 / N / E2 / M / 0
with exact rows and where all squares made from solid arrows commute.
By the universal property of the pushout there is a homomorphism γ : E1 → E2

with γ ◦ f1 = f2 and γ ◦ β1 = β2 . Now as in the proof of Skript 1, Lemma 10.10
we also get g2 ◦ γ = g1 . Thus the sequences (f1 , g1) and (f2 , g2 ) are equivalent. This
finishes the proof.
f g
Let 0 → X −→Y −→ Z → 0 be a short exact sequence, and let M and N be modules.
Then the connecting homomorphism
HomA (M, Z) → Ext1A (M, X)

is given by h 7→ [η] where η is the short exact sequence h∗ (f, g) induced by h via a
pullback.
η: 0 /X /⋆

/M /0

h

f g
0 / X / Y / Z / 0
Similarly, the connecting homomorphism

HomA (X, N) → Ext1A (Z, N)
is given by h 7→ [η] and where η is the short exact sequence h∗ (f, g) induced by h
via a pushout.
f g
0 / X / Y / Z / 0
h

η: 0 / N / ⋆ / Z / 0
If (f, g) is a split short exact sequence, then ψ(f, g) = 0 + Im(HomA (p1 , N)) is the
zero element in Ext1A (M, N): Obviously, the diagram
p1 ε
P 1 / P 0 / M / 0

0
[ 0ε ]
[ 10 ] [0 1]
0 / N / N ⊕M / M / 0
is commutative. This implies
ψ([ 10 ] , [ 0 1 ]) = 0 + Im(HomA (p1 , N)).
In fact, Ext1A (M, N) is a K-vector space and ψ is an isomorphism of K-vector spaces.
So we obtain the following fact:
Lemma 27.11. For an A-module M we have Ext1A (M, M) = 0 if and only if each
short exact sequence
0→M →E→M →0
splits. In other words, E ∼
= M ⊕ M.
27.7. The vector space structure on the first extension group. Let
ηM : 0 → Ω(M) → P0 → M → 0
be a short exact sequence with P0 projective. For i = 1, 2 let
fi gi
ηi : 0 → N −
→ Ei −
→M →0
be short exact sequences.
Take the direct sum η1 ⊕η2 and construct the pullback along the diagonal embedding
M → M ⊕ M. This yields a short exact sequence η ′ .
We know that every short exact sequence 0 → X → ⋆ → M → 0 is induced by ηM .

Thus we get a homomorphism [ αα12 ] : Ω(M) → N ⊕ N such that the diagram
u ε
ηM : 0 / Ω(M) / P0 / M / 0
[ αα12 ]

η′ : 0 / N ⊕N / E′ / M / 0
[ 11 ]

η1 ⊕ η2 : 0 / N ⊕ N hf / iE1 ⊕ E2 h g /
iM ⊕M / 0
1 0 1 0
0 f2 0 g2
commutes. Taking the pushout of η ′ along [1, 1] : N ⊕ N → N we get the following

commutative diagram:
u ε
ηM : 0 / Ω(M) / P0 / M / 0
[ αα12 ]

η′ : 0 / N ⊕N / E′ / M / 0
[1,1]

η ′′ : 0 / N / E ′′ / M / 0
In other words,
η ′ = [ αα12 ]∗ (ηM ),
η ′′ = [1, 1]∗ (η ′ ).
This implies η ′′ = (α1 + α2 )∗ (ηM ). Define
η1 + η2 := η ′′ .
Note that there exists some βi , i = 1, 2 such that the diagram
u ε
ηM : 0 / Ω(M) / P0 / M / 0
αi βi
fi gi
ηi : 0 / N / Ei / M / 0
commutes. Thus ηi = (αi )∗ (ηM ).
Similarly, let η : 0 → N → E → M → 0 be a short exact sequence. For λ ∈ K let

Let η ′ := (λ·)∗ (η) be the short exact sequence induced by the multiplication map
with λ. We also know that there exists some α : Ω(M) → N which induces η. Thus
we obtain a commutative diagram

ηM : 0 / Ω(M) / P0 / M / 0
α

η: 0 / N / E / M / 0
λ·

η′ : 0 / N / E′ / M / 0
Define λη := η ′ .
Thus, we defined an addition and a scalar multiplication on the set of equivalence
classes of short exact sequences. We leave it as an (easy) exercice to show that this
really defines a K-vector space structure on Ext1A (M, N).
27.8. Injective resolutions. injective resolution
...
minimal injective resolution
...
Theorem 27.12. Let I • be an injective resolution of an A-module N. Then for any
A-module M we have an isomorphism
Extn (M, N) ∼
A = H n (HomA (M, I • )).
which is “natural in M and N”.
Proof. Exercise.
————————————————————————————-
28. Digression: Homological dimensions
28.1. Projective, injective and global dimension. Let A be a K-algebra. For

an A-module M let proj. dim(M) be the minimal j ≥ 0 such that there exists a
projective resolution (Pi , di)i of M with Pj = 0, if such a minimum exists, and
define proj. dim(M) = ∞, otherwise.
We call proj. dim(M) the projective dimension of M. The global dimension of

A is by definition
gl. dim(A) = sup{proj. dim(M) | M ∈ mod(A)}.
Here sup denote the supremum of a set.
It often happens that the global dimension of an algebra A is infinite, for example
if we take A = K[X]/(X 2 ). One proves this by constructing the minimal projective
resolution of the simple A-module S. Inductively one shows that Ωi (S) ∼ = S for all
i ≥ 1.
Proposition 28.1. Assume that A is finite-dimensional. Then we have
gl. dim(A) = max{proj. dim(S) | S a simple A-module}.
Proof. Use the Horseshoe Lemma.
Similarly, let inj. dim(M) be the minimal j ≥ 0 such that there exists an injec-
tive resolution (Ii , di )i of M with Ij = 0, if such a minimum exists, and define
inj. dim(M) = ∞, otherwise.
We call inj. dim(M) the injective dimension of M.
Theorem 28.2 (No loop conjecture). Let A be a finite-dimensional K-algebra. If
Ext1A (S, S) 6= 0 for some simple A-module S, then gl. dim(A) = ∞.
Conjecture 28.3 (Strong no loop conjecture). Let A be a finite-dimensional K-
algebra. If Ext1A (S, S) 6= 0 for some simple A-module S, then proj. dim(S) = ∞.
28.2. Hereditary algebras. A K-algebra A is hereditary if gl. dim(A) ≤ 1.
28.3. Selfinjective algebras.
28.4. Finitistic dimension. For an algebra A let

fin.dim(A) := sup{proj. dim(M) | M ∈ mod(A), proj. dim(M) < ∞}
be the finitistic dimension of A. The following conjecture is unsolved for several
decades and remains wide open:
Conjecture 28.4 (Finitistic dimension conjecture). If A is finite-dimensional, then
fin.dim(A) < ∞.
28.5. Representation dimension. The representation dimension of a finite-

dimensional K-algebra A is the infimum over all gl. dim(C) where C is a generator-
cogenerator of A, i.e. each indecomposable projective module and each indecom-
posable injective module occurs (up to isomorphism) as a direct summand of C.
Theorem 28.5 (Auslander). For a finite-dimensional K-algebra A the following
hold:
(i) rep.dim(A) = 0 if and only if A is semisimple;

6 1;
(ii) rep.dim(A) =
(iii) rep.dim(A) = 2 if and only if A is representation-finite, but not semisimple.
Theorem 28.6 (Iyama). If A is a finite-dimensional algebra, then rep.dim(A) < ∞.
Theorem 28.7 (Rouquier). For each n ≥ 3 there exists a finite-dimensional algebra
A with rep.dim(A) = n.
28.6. Dominant dimension. dominant dimension of A
28.7. Auslander algebras. Let A be a finite-dimensional representation-finite K-

algebra. The Auslander algebra of A is defined as EndA (M) where M is the
direct sum of a complete set of representatives of isomorphism classes of the inde-
composable A-modules.
Theorem 28.8 (Auslander). ...
28.8. Gorenstein algebras.
————————————————————————————-
29. Tensor products, adjunction formulas and Tor-functors
29.1. Tensor products of modules. Let A be a K-algebra. Let X be an Aop -

module, and let Y be an A-module. Recall that X can be seen as a right A-module
as well. For x ∈ X and a ∈ A we denote the action of Aop and A on X by a⋆x = x·a.
By V (X, Y ) we denote a K-vector space with basis
X × Y = {(x, y) | x ∈ X, y ∈ Y }.
Let R(X, Y ) be the subspace of V (X, Y ) which is generated by all vectors of the
form
(1) ((x + x′ ), y) − (x, y) − (x′ , y),

(2) (x, (y + y ′)) − (x, y) − (x, y ′ ),
(3) (xa, y) − (x, ay),
(4) λ(x, y) − (λx, y).
where x ∈ X, y ∈ Y , a ∈ A and λ ∈ K. The vector space

X ⊗A Y := V (X, Y )/R(X, Y )
is the tensor product of XA and A Y . The elements z in X ⊗A Y are of the form
Xm
xi ⊗ yi ,
i=1
where x ⊗ y := (x, y) + R(X, Y ). But note that this expression of z is in general not
unique.
Warning
From here on there are only fragments, incomplete proofs or no proofs
at all, typos, wrong statements and other horrible things...
A map β : X × Y → V where V is a vector space is called balanced if for all

x, x′ ∈ X, y, y ′ ∈ Y , a ∈ A and λ ∈ K the following hold:
(1) β(x + x′ , y) = β(x, y) + β(x′ , y),

(2) β(x, y + y ′ ) = β(x, y) + β(x, y ′),
(3) β(xa, y) = β(x, ay),
(4) β(λx, y) = λβ(x, y).
In particular, a balanced map is K-bilinear.
For example, the map

ω : X × Y → X ⊗A Y
defined by (x, y) 7→ x⊗y is balanced. This map has the following universal property:
Lemma 29.1. For each balanced map β : X ×Y → V there exists a unique K-linear
map γ : X ⊗A Y → V with β = γ ◦ ω.
ω
X ×Y / X ⊗A Y
r
β r rr
γ
xr r
V
Furthermore, this property characterizes X ⊗A Y up to isomorphism.
Proof. We can extend β and ω (uniquely) to K-linear maps β ′ : V (X, Y ) → V and

ω ′ : V (X, Y ) → X ⊗A Y , respectively. We have R(X, Y ) ⊆ Ker(β ′ ), since β is
balanced. Let ι : R(X, Y ) → Ker(β ′ ) be the inclusion map. Now it follows easily
that there is a unique K-linear map γ : X ⊗A Y → V with β = γ ◦ ω and β ′ = γ ◦ ω ′.
ω′
0 / R(X, Y ) / V (X, Y ) / X ⊗ A Y / 0
ι γ

β′
0 / Ker(β ′ ) / V (X, Y ) / V / 0

Let A, B, C be K-algebras, and let A XB be an A-B op -bimodule and B YC a B-C op -

bimodule. We claim that X ⊗B Y is an A-C op -bimodule with the bimodule structure
defined by
a(x ⊗ y) = (ax) ⊗ y,
(x ⊗ y)c = x ⊗ (yc)
where a ∈ A, c ∈ C and x ⊗ y ∈ X ⊗B Y : One has to check that everything is
well defined. It is clear that we obtain an A-module structure and a C op -module
structure. Furthermore, we have
(a(x ⊗ y))c = ((ax) ⊗ y)c = (ax) ⊗ (yc) = a((x ⊗ y)c).
Thus we get a bimodule structure on X ⊗B Y .
Lemma 29.2. For any A-module M, we have

∼
A AA ⊗A M = M
as A-modules.
Proof. The A-module homomorphisms η : A ⊗A M → M, a ⊗ m 7→ am and φ : M →

A ⊗A M, m 7→ 1 ⊗ m are mutual inverses.
Let f : XA → XA′ and g : A Y → A Y ′ be homomorphisms. Then the map β : X ×Y →

X ′ ⊗A Y ′ defined by (x, y) 7→ f (x) ⊗ g(y) is balanced. Thus there exists a unique
K-linear map
f ⊗ g : X ⊗A Y → X ′ ⊗A Y ′
with (f ⊗ g)(x ⊗ y) = f (x) ⊗ g(y).
ω
X ×Y / X ⊗A Y
q
β q qq
xq q f ⊗g
′ ′
X ⊗A Y
Now let f = 1X , and let g be as above. We obtain a K-linear map

X ⊗ g := 1X ⊗ g : X ⊗A Y → X ⊗A Y ′
Lemma 29.3. (i) For any right A-module XA we get an additive right exact
functor
X ⊗A − : Mod(A) → Mod(K)
defined by Y 7→ X ⊗A Y and g 7→ X ⊗ g.
(ii) For any A-module A Y we get an additive right exact functor
− ⊗A Y : Mod(A) → Mod(K)
defined by X 7→ X ⊗A Y and f 7→ f ⊗ Y .
Proof. We just prove (i) and leave (ii) as an exercise. Clearly, X ⊗A − is a functor:
We have X ⊗A (g ◦ f ) = (X ⊗A g) ◦ (X ⊗A f ). In particular, X ⊗A 1Y = 1X⊗A Y .
Additivity:
(X ⊗A (f + g))(x ⊗ y) = x ⊗ (f + g)(y)
= x ⊗ (f (y) + g(y))
= (x ⊗ f (y)) + (x ⊗ g(y))
= (X ⊗ f )(x ⊗ y) + (X ⊗ g)(x ⊗ y).
Right exactness:
...

Lemma 29.4. (i) Let XA be a right A-module. If (Yi )i is a family of A-modules,

then !
M M
X ⊗A Yi ∼= (X ⊗A Yi )
i i
where an isomorphism is defined by x ⊗ (yi)i 7→ (x ⊗ yi)i .
(ii) Let A Y be an A-module. If (Xi )i is a family of right A-modules, then
!
M M
Xi ⊗A Y ∼ = (Xi ⊗A Y )
i i
where an isomorphism is defined by (xi )i ⊗ y 7→ (xi ⊗ y)i .
Proof. Again, we just prove (i).

...

Corollary 29.5. If PA is a projective right A-module and A Q a projective left A-
module, then
P ⊗A − : Mod(A) → Mod(K)
and
− ⊗A Q : Mod(Aop ) → Mod(K)
are exact functor.
L
L that A ⊗A − is exact. It follows that i A ⊗A − is exact. Since
Proof. We know
PA ⊕ QA = i A for some QA , we use the additivity of ⊗ and get that PA ⊗ − is
exact as well. The exactness of − ⊗A Q is proved in the same way.
Lemma 29.6. Let A be a finite-dimensional algebra, and let XA be a right A-module.
If X ⊗A − is exact, then XA is projective.
Proof. Exercise.
29.2. Adjoint functors. Let A and B be categories, and let F : A → B and

G : B → A be functors. If
HomB (F (X), Y )) ∼
= HomA (X, G(Y ))
for all X ∈ A and Y ∈ B and if this isomorphism is “natural”, then F and G are
adjoint functors. One calls F the left adjoint of G, and G is the right adjoint
of F .
Theorem 29.7 (Adjunction formula). Let A and B be K-algebras, let A XB be an A-
B op -bimodule, B Y a B-module and A Z an A-module. Then there is an isomorphism
Adj := η : HomA (X ⊗B Y, Z) → HomB (Y, HomA (X, Z))
where η is defined by η(f )(y)(x) := f (x⊗y). Furthermore, η is “natural in X, Y, Z”.
Proof. ...

29.3. Tor. We will not need any Tor-functors, but at least we will define them and
acknowledge their existence.
Let P• be a projective resolution of A Y , and let XA be a right A-module. This yields
a complex
· · · → X ⊗A P1 → X ⊗A P0 → X ⊗A 0 → · · ·
For n ∈ Z define
TorA
n (X, Y ) := Hn (X ⊗A P• ).
Let P• be a projective resolution of a right A-module XA . Then one can show that
for all A-modules A Y we have
TorA ∼
n (X, Y ) = Hn (P• ⊗A Y ).
Similarly as for Ext1A (−, −) one can prove that TorA

n (X, Y ) does not depend on the
choice of the projective resolution of Y .
The following hold:
(i) TorAn (X, Y ) = 0 for all n < 0;

(ii) TorA0 (X, Y ) = X ⊗A Y ;
(iii) If A P is projective, then TorAn (X, P ) = 0 for all n ≥ 1.
(iv)
Again, similarly as for Ext1A (−, −) we get long exact Tor-sequences:
(i) Let
η : 0 → XA′ → XA → XA′′ → 0
be a short exact sequence of right A-modules. For every A-module A Y this induces
an exact sequence
··· / TorA (X ′′ , Y )
g 2
ggggggg
ggggg
gg gg ggggg
sgg
TorA
1 (X ′
, Y ) / TorA (X, Y )
1 g
/ TorA (X ′′ , Y )
1
gggg
ggggggg
gggg
ggggg
sggggg
X′ ⊗ Y A
/X⊗ Y
A
/ X ′′ ⊗ Y
A
/ 0
(ii) Let
η : 0 → A Y ′ → A Y → A Y ′′ → 0
be a short exact sequence of A-modules. For every right A-module XA this induces
an exact sequence
··· / TorA (X, Y ′′ )
g 2
g gggggg
ggggg
gg gg ggggg
sgg
TorA
1 (X, Y ′
) / TorA (X, Y )
1 g
/ TorA (X, Y ′′ )
1
gggg
gggg ggg
gggg
ggggg
sggggg
X ⊗ Y′ A
/ X ⊗A Y / X ⊗ Y ′′
A
/ 0
Note that the bifunctor TorA

n (−, −) is covariant in both arguments. This is not true
n
for ExtA (−, −).
Theorem 29.8 (General adjunction formula). Let A and B be K-algebras, let A XB
be an A-B op -bimodule, B Y a B-module and A Z an A-module. If A Z is injective,
then there is an isomorphism
HomA (TorB (X, Y ), Z) ∼
n = Extn (Y, HomA (X, Z))
B
for all n ≥ 1.
***********************************************************************
Part 6. Homological Algebra II: Auslander-Reiten Theory
30. Irreducible homomorphisms and Auslander-Reiten sequences
30.1. Irreducible homomorphisms. Let M be a module category. A homomor-

phism f : V → W in M is irreducible (in M) if the following hold:
• f is not a split monomorphism;

• f is not a split epimorphism;
• For any factorization f = f2 f1 in M, f1 is a split monomorphism or f2 is a
split epimorphism.
Note that any homomorphism f : V → W has many factorizations f = f2 f1 with

f1 a split monomorphism or f2 a split epimorphism: Let C be any module in M,
and let g : V → C and h : C → W be arbitrary homomorphisms. Define f1 =
t
[1, 0] : V → V ⊕ C and f2 = [f, h] : V ⊕ C → W . Then f = f2 f1 with f1 a split
monomorphism.
Similarly, define f1′ = t [f, g] : V → W ⊕ C and f2′ = [1, 0] : W ⊕ C → W . Then

f = f2′ f1′ with f2′ a split epimorphism. We could even factorize f as f = f2′′ f1′′ with
f1′′ a split monomorphism and f2′′ a split epimorphism: Take f1′′ = t [1, 1, 0] : V →
V ⊕ V ⊕ W and f2′′ = [0, f, 1] : V ⊕ V ⊕ W → W .
Thus the main point is that the third condition in the above definition applies to
ALL factorizations f = f2 f1 of f .
The notion of an irreducible homomorphism makes only sense if we talk about a
certain fixed module category M.
Examples: Let V and W be non-zero modules in a module category M. Examples
of homomorphisms which are NOT irreducible are 0 → 0, 0 → W , V → 0, 0 : V →
W , 1V : V → V . (Recall that the submodule 0 of a module is always a direct
summand.)
Lemma 30.1. Assume that M is a module category which is closed under images
(i.e. if f : U → V is a homomorphism in M, then Im(f ) is in M). Then every
irreducible homomorphism in M is either injective or surjective.
Proof. Assume that f : U → V is a homomorphism in M which is neither injective

nor surjective, and let f = f2 f1 where f1 : U → Im(f ) is the homomorphism defined
by f1 (u) = f (u) for all u ∈ U, and f2 : Im(f ) → V is the inclusion homomorphism.
Then f1 is not a split monomorphism (it is not even injective), and f2 is not a split
epimorphism (it is not even surjective).
30.2. Auslander-Reiten sequences and Auslander-Reiten quivers. Again let

M be a module category. An exact sequence
f g
0→U −
→V −
→W →0
with U, V, W ∈ M is an Auslander-Reiten sequence in M if the following hold:
(i) The homomorphisms f and g are irreducible in M;

(ii) Both modules U and W are indecomposable.
(We will see that for many module cateogories assumption (ii) is not necessary: If M
is closed under kernels of surjective homomorphisms and f : U → V is an injective
homomorphism which is irreducible in M, then Cok(f ) is indecomposable. Similarly,
if M is closed under cokernels of injective homomorphisms and g : V → W is a
surjective homomorphism which is irreducible in M, then Ker(g) is indecomposable.)
Let (Γ0 , Γ1 ) and (Γ0 , Γ2 ) be two “quivers” with the same set Γ0 of vertices, but with
disjoint sets Γ1 and Γ2 of arrows. Then (Γ0 , Γ1 , Γ2 ) is called a biquiver. The arrows
in Γ1 are the 1-arrows and the arrows in Γ2 the 2-arrows. To distinguish these two
types of arrows, we usually draw dotted arrows for the 2-arrows. (Thus a biquiver
Γ is just an oriented graph with two types of arrows: The set of vertices is denoted
by Γ0 , the “1-arrows” are denoted by Γ1 and the “2-arrows” by Γ2 .)
Let M be a module category, which is closed under direct summands. Then the
Auslander-Reiten quiver of M is a biquiver ΓM which is defined as follows: The
vertices are the isomorphism classes of indecomposable modules in M. For a module
V we often write [V ] for its isomorphism class. There is a 1-arrow [V ] → [W ] if
and only if there exists an irreducible homomorphism V → W in M, and there is a
2-arrow from [W ] to [U] if and only if there exists an Auslander-Reiten sequence
0 → U → V → W → 0.
The Auslander-Reiten quiver is an important tool which helps to understand the
structure of a given module category.
Later we will modify the above definition of an Auslander-Reiten quiver and also
allow more than one arrow between two given vertices.
30.3. Properties of irreducible homomorphisms. We want to study irreducible

homomorphisms in a module category M in more detail.
For this we assume that M is closed under kernels of surjective homomorphisms,
that is for every surjective homomorphism g : V → W in M, the kernel Ker(g)
belongs to M. In particular, this implies the following: If g1 : V1 → W , g2 : V2 → W
are in M, and if at least one of these homomorphisms gi is surjective, then also the
pullback of (g1 , g2 ) is in M.
Lemma 30.2 (Bottleneck Lemma). Let M be a module category which is closed
under kernels of surjective homomorphisms. Let
f g
0→U −
→V −
→W →0
be a short exact sequence in M, and assume that f is irreducible in M. If g ′ : V ′ →
W is any homomorphism, then there exists a homomorphism b1 : V ′ → V with
gb1 = g ′ , or there exists a homomorphism b2 : V → V ′ with g ′ b2 = g.
V′ or V ′
b1 } b2 }>
} g′ } g′
} }
f ~} g f } g
0 / U / V / W / 0 0 / U / V / W / 0
The name “bottleneck” is motivated by the following: Any homomorphism with

target W either factors through g or g factors through it. So everything has to pass
through the “bottleneck” g.
Proof. The induced sequence (g ′)∗ (f, g) looks as follows:

f1 g1
0 / U / P / V′ / 0
f2 g′
f g
0 / U / V / W / 0
The module P is the pullback of (g, g ′), thus P belongs to M. We obtain a factor-
ization f = f2 f1 in M. By our assumption, f is irreducible in M, thus f1 is a split
monomorphism or f2 is a split epimorphism. In the second case, there exists some
f2′ : V → P such that f2 f2′ = 1V . Therefore for b2 := g1 f2′ we get
g ′b2 = g ′ g1 f2′ = gf2 f2′ = g1V = g.
On the other hand, if f1 is a split monomorphism, then the short exact sequence
(f1 , g1 ) splits, and it follows that g1 is a split epimorphism. We obtain a homomor-
phism g1′ : V ′ → P with g1 g1′ = 1V ′ . For b1 := f2 g1′ we get
gb1 = gf2g1′ = g ′g1 g1′ = g ′1V ′ = g ′.

Corollary 30.3. Let M be a module category which is closed under kernels of
surjective homomorphisms. If
f g
0→U −
→V −
→W →0
is a short exact sequence in M with f irreducible in M, then W is indecomposable.
Proof. Let W = W1 ⊕ W2 , and let ιi : Wi → W be the inclusions. We assume that

W1 6= 0 6= W2 . Thus none of these two inclusions is surjective. By the Bottleneck
Lemma, there exist homomorphisms ci : Wi → V with gci = ιi . (If there were
homomorphisms c′i : V → Wi with ιi c′i = g, then g and therefore also ιi would be
surjective, a contradiction.)
Let C = Im(c1 ) + Im(c2 ) ⊆ V . We have Im(f ) ∩ C = 0: If f (u) = c1 (w1 ) + c2 (w2 )
for some u ∈ U and wi ∈ Wi , then
0 = gf (u) = gc1(w1 ) + gc2(w2 ) = ι1 (w1 ) + ι2 (w2 )
and therefore w1 = 0 = w2 .
On the other hand, we have Im(f ) + C = V : If v ∈ V , then g(v) = ι1 (w1′ ) + ι2 (w2′ )

for some wi′ ∈ Wi . This implies g(v) = gc1 (w1′ ) + gc2 (w2′ ), thus v − c1 (w1′ ) − c2 (w2′ )
belongs to Ker(g) and therefore to Im(f ). If we write this element in the form f (u′)
for some u′ ∈ U, then v = f (u′ ) + c1 (w1′ ) + c2 (w2′ ).
Altogether, we see that Im(f ) is a direct summand of V , a contradiction since we
assumed f to be irreducible.
Corollary 30.4. Let M be a module category which is closed under kernels of
surjective homomorphisms. If
f1 g1
0 → U1 −
→ V1 −
→W →0
f2 g2
0 → U2 −
→ V2 −
→W →0
are two Auslander-Reiten sequences in M, then there exists a commutative diagram
f1 g1
0 / U1 / V1 / W / 0
a b
f2 g2
0 / U2 / V2 / W / 0
with a and b isomorphisms.
Proof. Since f1 is irreducible, there exists a homomorphism b : V1 → V2 with g1 = g2 b

or a homomorphism b′ : V2 → V1 with g2 = g1 b′ . For reasons of symmetry, it
is enough to consider only one of these cases. Let us assume that there exists
b : V1 → V2 with g1 = g2 b. This implies the existence of a homomorphism a : U1 → U2
with bf1 = f2 a. (Since g2 bf1 = 0, we can factorize bf1 through the kernel of g2 .)
Thus we constructed already a commutative diagram as in the statement of the
corollary.
It remains to show that a and b are isomorphisms: Since g1 is irreducible, and
since g2 is not a split epimorphism, the equality g1 = g2 b implies that b is a split
monomorphism. Thus there is some b′ : V2 → V1 with b′ b = 1V1 . We have b′ f2 a =
b′ bf1 = f1 , and since f1 is irreducible, a is a split monomorphism or b′ f2 is a split
epimorphism. Assume b′ f2 : U2 → V1 is a split epimorphism. We know that U2 is
indecomposable and that V1 6= 0, thus b′ f2 has to be an isomorphism. This implies
that f2 is a split monomorphism, a contradiction. So we conclude that a : U2 → U1
is a split monomorphism. Now U2 is indecomposable and U1 6= 0, thus a is an
isomorphism. This implies that b also has to be an isomorphism.
Corollary 30.5. Let M be a module category which is closed under direct summands
and under kernels of surjective homomorphisms. Let
f g
0→U −
→V −
→W →0
be an Auslander-Reiten sequence in M. If Y is a module in M which can be writ-
ten as a finite direct sum of indecomposable modules, and if h : Y → W is a ho-
momorphism which is not a split epimorphism, then there exists a homomorphism
h′ : Y → V with gh′ = h.
Proof. We first assume that Y is indecomposable. By the Bottleneck Lemma, instead

of h′ there could exist a homomorphism g ′ : V → Y with g = hg ′ . But g is irreducible
and h is not a split epimorphism. Thus g ′ must be a split monomorphism. Since
Y is indecomposable and V 6= 0, this implies that g ′ is an isomorphism. Thus
h = g(g ′)−1 .
Lt
Now let Y =
Lt i=1 Yi with Yi indecomposable for all i. As usual let ιs : Ys →
i=1 Yi be the inclusion homomorphisms. Set hi = hιi . By our assumptions, h is
not a split epimorphism, thus the same is true for hi . Thus we know that there
are homomorphisms h′i : Yi → V with gh′i = hi . Then h′ = [h′1 , . . . , h′t ] satisfies
gh′ = h.
Now we prove the converse of the Bottleneck Lemma (but note the different assump-
tion on M):
Lemma 30.6 (Converse Bottleneck Lemma). Let M be a module category which is

closed under cokernels of injective homomorphisms. Let
f g
0→U −
→V −
→W →0
be a non-split short exact sequence in M such that the following hold: For every
homomorphism g ′ : V ′ → W in M there exists a homomorphism b1 : V ′ → V with
gb1 = g ′ , or there exists a homomorphism b2 : V → V ′ with g ′b2 = g. Then it follows
that f is irreducible in M.
Proof. Let f = f2 f1 be a factorization of f in M. Thus f1 : U → V ′ for some V ′ in

M. The injectivity of f implies that f1 is injective as well. Let g1 : V ′ → W ′ be the
cokernel map of f1 . By assumption W ′ belongs to M. Since gf2f1 = gf = 0, we can
factorize gf2 through g1 . Thus we obtain g ′ : W ′ → W with g ′ g1 = gf2 . Altogether
we constructed the following commutative diagram:
f g
0 / U / VO / WO / 0
f2 g′
f1 g1
0 / U / V′ / W′ / 0
It follows that the pair (f2 , g1 ) is the pullback of (g, g ′). Our assumption implies that
for g ′ there exists a homomorphism b1 : W ′ → V with gb1 = g ′ or a homomorphism
b2 : V → W ′ with g ′ b2 = g.
If b1 exists with gb1 = g ′ = g ′1W ′ , then the pullback property yields a homomorphism
h : W ′ → V ′ with b1 = f2 h and 1W ′ = g1 h. In particular we see that g1 is a split
epimorphism, and therefore f1 is a split monomorphism.

i4 V C
i
b1
i i iiii{{= CC g
iii ii {{{ CC
CC
ii iiii {{{ f2 C!
i ii h
W′ T_TT_T _ _ _ _/ V ′ C W
TTTT CC g {=
TTTT CC 1 {{{
TTTT C {{ ′
1W ′ TTTTCC! {{ g
T*
′
W
In the second case, if b2 exists with g1V = g = g ′b2 , we obtain a homomorphism
h′ : V → V ′ with 1V = f2 h′ and b2 = g1 h′ . Thus f2 is a split epimorphism.
j4 V C
1V jjj jjjj{{= CC g
j jjj {{ CC
j { CC
jj jjjj {{{ f2 C!
j jj h ′
V T_TT_T _ _ _ _/ V ′ W
TTTT
TTTT
CC
CC g1 {{=
TTTT CCC {
{{
b2 TTTT C! {{{ g′
T*
′
W

30.4. Dual statements. Let us formulate the corresponding dual statements:

Lemma 30.7 (Bottleneck Lemma). Let M be a module category which is closed
under cokernels of injective homomorphisms. Let
f g
0→U −
→V −
→W →0
be a short exact sequence in M, and assume that g is irreducible in M. If f ′ : U →
V ′ is any homomorphism, then there exists a homomorphism a1 : V → V ′ with
a1 f = f ′ , or there exists a homomorphism a2 : V ′ → V with a2 f ′ = f .
f g f g
0 / U /V / W / 0 or 0 / U /V / W / 0
} }>
f′
} f′
}
} a1 } a2
~} }
′ ′
V V
Corollary 30.8. Let M be a module category which is closed under cokernels of
injective homomorphisms. If
f g
0→U −
→V −
→W →0
is a short exact sequence in M, and if g is irreducible in M, then U is indecompos-
able.
Corollary 30.9. Let M be a module category which is closed under cokernels of
injective homomorphisms. If
f1 g1
0→U −
→ V1 −
→ W1 → 0
f2 g2
0→U −
→ V2 −
→ W2 → 0
are two Auslander-Reiten sequences in M, then there exists a commutative diagram

f1 g1
0 / U / VO 1 / WO 1 / 0
b c
f2 g2
0 / U / V2 / W2 / 0
with b and c isomorphisms.
Corollary 30.10. Let M be a module category which is closed under direct sum-
mands and under cokernels of injective homomorphisms. Let
f g
0→U −
→V −
→W →0
be an Auslander-Reiten sequence in M. If X is a module in M which can be
written as a finite direct sum of indecomposable modules, and if h : U → X is a ho-
momorphism which is not a split monomorphism, then there exists a homomorphism
h′ : V → X with h′ f = h.
Lemma 30.11 (Converse Bottleneck Lemma). Let M be a module category which
is closed under kernels of surjective homomorphisms. Let
f g
0→U −
→V −
→W →0
be a non-split short exact sequence in M such that the following hold: For every
homomorphism f ′ : U → V ′ in M there exists a homomorphism a1 : V → V ′ with
a1 f = f ′ , or there exists a homomorphism a2 : V ′ → V with a2 f ′ = f . Then it
follows that g is irreducible in M.
The proofs of these dual statements are an exercise.
30.5. Examples: Irreducible maps in N f.d. . In this section let

M := N f.d.
be the module category of all 1-modules (V, φ) with V finite-dimensional and φ
nilpotent.
Recall that we denoted the indecomposable modules in M by N(n) where n ≥ 1.

Let us also fix basis vectors e1 , . . . , en of N(n) = (V, φ) such that φ(ei ) = ei−1 for
2 ≤ i ≤ n and φ(e1 ) = 0.
By
ιn : N(n) → N(n + 1)
we denote the canonical inclusion (defined by ιn (en ) = en ), and let
πn+1 : N(n + 1) → N(n)
be the canonical projection (defined by πn+1 (en+1 ) = en ). For n > t let
πn,t := πt+1 ◦ · · · ◦ πn+1 ◦ πn : N(n) → N(t),
and for t < m set
ιt,m := ιm−1 ◦ · · · ◦ ιt+1 ◦ ιt : N(t) → N(m).
Finally, let πn,n = ιn,n = 1N (n) .

Lemma 30.12. For m, n ≥ 1 the following hold:
(i) Every injective homomorphism N(n) → N(n + 1) is irreducible (in M);

(ii) Every surjective homomorphism N(n + 1) → N(n) is irreducible (in M).
(iii) If f : N(n) → N(m) is irreducible (in M), then either m = n + 1 or n =
m + 1, and f is either injective or surjective.
Proof. Let h : N(n) → N(n+1) be an injective homomorphism. Clearly, h is neither

a split monomorphism nor a split epimorphism. Let h = gf where f : N(n) → N(λ)
and g : N(λ) → N(n + 1) are homomorphisms with λ = (λ1 , . . . , λt ) a partition.
(Recall that the isomorphism classes of objects in M are parametrized by partitions
of natural numbers.) Thus
t
M
t
f = [f1 , . . . , ft ] : N(n) → N(λi )
i=1
and
t
M
g = [g1 , . . . , gt ] : N(λi ) → N(n + 1)
i=1
with fi : N(n) → N(λi ) and gi : N(λi ) → N(n + 1) homomorphisms and
t
X
h = gf = gi fi .
i=1
Since h is injective, we have h(e1 ) 6= 0. Thus there exists some i with gi fi (e1 ) 6= 0.
This implies that gi fi is injective, and therefore fi is injective.
If λi > n + 1, then gi (e1 ) = 0, a contradiction. (Note that gi fi (e1 ) 6= 0 implies
gi (e1 ) 6= 0).) Thus λi is either n or n + 1. If λi = n, then fi is an isomorphism, if
λi = n + 1, then gi is an isomorphism. In the first case, set
f ′ = [0, . . . , 0, fi−1, 0, . . . , 0] : N(λ) → N(n).
We get f ′ f = 1N (n) , thus f is a split monomorphism.
In the second case, set
g ′ = t [0, . . . , 0, gi−1, 0, . . . , 0] : N(n + 1) → N(λ).
It follows that gg ′ = 1N (n+1) , so g is a split epimorphism. This proves part (i). Part
(ii) is proved similarly.
Next, let f : N(n) → N(m) be an irreducible homomorphism. We proved already
before that every irreducible homomorphism has to be either injective or surjective.
If m ≥ n + 2, then f factors through N(n + 1) as f = f2 f1 where f1 is injective but
not split, and f2 is not surjective, a contradiction. Similarly, if m ≤ n − 2, then f
factors through N(n − 1) as f = f2 f1 where f1 is not injective, and f2 is surjective
but not split, again a contradiction. This proves (iii).
Lemma 30.13. For m, n ≥ 1 the following hold:
• Every non-invertible homomorphism N(n) → N(m) in M is a linear com-

bination of compositions of irreducible homomorphisms;
• Every endomorphism N(n) → N(n) in M is a linear combination of 1N (n)
and of compositions of irreducible homomorphisms.
Proof. Let f : N(n) → N(m) be a homomorphism with

t
X
f (en ) = ai ei
i=1
with at 6= 0. It follows that n ≥ t and m ≥ t, and dim Im(f ) = t. Let
g = ιt,m ◦ πn,t : N(n) → N(m).
Now it is easy to check that
dim Im(f − at g) ≤ t − 1.
We see that f − at g is not an isomorphism, thus by induction assumption it is a

linear combination of compositions of irreducible homomorphisms in M. Also, g is
either 1N (n) (in case n = m) or it is a composition of irreducible homomorphisms.
Thus f = at g + (f − at g) is of the required form.
Thus we determined all irreducible homomorphisms between indecomposable mod-

ules in M. So we know how the 1-arrows of the Auslander-Reiten quiver of M look
like. We still have to determine the Auslander-Reiten sequences in M in order to
get the 2-arrows as well.
30.6. Exercises. Use the Converse Bottleneck Lemma to show that for n ≥ 1 the
short exact sequence
[ πιnn ] [πn+1 ,−ιn−1 ]

0 → N(n) −−−→ N(n + 1) ⊕ N(n − 1) −−−−−−−→ N(n) → 0
is an Auslander-Reiten sequence in N f.d. . (We set N(0) = 0.)
————————————————————————————-
31. Auslander-Reiten Theory
31.1. The transpose of a module. ...

31.2. The Auslander-Reiten formula. An A-module M is finitely presented

if there exists an exact sequence
p q
P1 −
→ P0 −
→M →0
with P0 and P1 are finitely generated projective A-modules. Our aim is to prove the
following result:
Theorem 31.1 (Auslander-Reiten formula). For a finitely presented A-module M
we have
Ext1A (N, τ (M)) ∼
= DHomA (M, N).
Before we can prove Theorem 31.1 we need some preparatory results:

p
Lemma 31.2. Let X → Y −
→ Z → 0 be exact, and let
p
X / Y / Z / 0
ξx ξy ξ
f g
X′ / Y′ / Z′
be a commutative diagram where ξx and ξy are isomorphisms and Im(f ) ⊆ Ker(g).
Then
Ker(g)/ Im(f ) ∼
= Ker(ξ).
Proof. ...

Lemma 31.3. Let f : X → Y be a homomorphism, and let u : Y → Z be a
monomorphism. Then
Ker(HomA (N, f )) = Ker(HomA (N, u ◦ f )).
Proof. Let h : N → X be a homomorphism. Then h ∈ Ker(HomA (N, f )) if and only

if f ◦ h = 0. This is equivalent to u ◦ f ◦ h = 0, since u is injective. Furthermore
u ◦ f ◦ h = 0 if and only if h ∈ Ker(HomA (N, u ◦ f )).
Let A be a K-algebra, and let X be an A-module. Set

X ∗ := HomA (X, A A).
Observe that X ∗ is a right A-module.
For an A-module Y define
ηXY : X ∗ ⊗A Y → HomA (X, Y )
by (α ⊗ y)(x) := α(x) · y. In other words
ηXY (α ⊗ y) := ρy ◦ α
where ρy is the right multiplication with y.
α ρy
→ A A −→ Y
X−
Clearly, X ∗ is a right A-module: For α ∈ X ∗ and a ∈ A set (α · a)(x) := α(x) · a.

The map X ∗ × Y → HomA (X, Y ), (α, y) 7→ ρy ◦ α is bilinear, and we have
(αa, y) 7→ ρy ◦ (αa)
(α, ay) 7→ ρay ◦ α.
We also know that
(ρy ◦ (αa))(x) = ρy (α(x) · a) = α(x) · ay = (ρay ◦ α)(x).
In other words, the map (α, y) 7→ ρy ◦ α is balanced.
ω / X∗
X∗ × Y ⊗A Y
nnn
nnnnn
wnnn ηXY
HomA (X, Y )
Lemma 31.4. The image of ηXY consists of the homomorphisms X → Y which
factor through finitely generated projective modules.
Proof. We have
n
! n
X X
ηXY αi ⊗ yi = ηXY (αi ⊗ yi)
i=1 i=1
Xn
= ρyi ◦ αi .
i=1
2 α1 3
.. 5
4
. n
M [ρy1 ,...,ρyn ]
αn
X −−−→ A A −−−−−−→ Y
i=1
To prove the other direction, let P be a finitely generated projective module, and
assume h = g ◦ f for some homomorphisms h : X → Y , f : X → P and g : P → Y .
There exists a module C such that P ⊕ C is a free module of finite rank. Thus
without loss of generality we can assume that PP is free of finite rank. Let e1 , . . . , en
be a free generating set of P . Then f (x) = i αi (x)ei for some αi (x) ∈ A. This
defines some homomorphisms αi : X → A A. Set yi := g(ei). It follows that
!
X X
ηXY αi ⊗ yi (x) = αi (x)yi
i i
X
= αi (x)g(ei )
i
!
X
=g αi (x)ei
i
= (g ◦ f )(x) = h(x).
Lemma 31.5. Assume that X is finitely generated, and let f : X → Y be a homo-

morphism. Then the following are equivalent:
(i) f factors through a projective module;

(ii) f factors through a finitely generated projective module;
(iii) f factors through a free module of finite rank.
Proof. Exercise.
Let HomA (X, Y )P := PA (X, Y ) be the set of homomorphisms X → Y which factor

through a projective module. Clearly, this is a subspace of HomA (X, Y ). As before,
define
HomA (X, Y ) := HomA (X, Y )/PA (X, Y ).
Lemma 31.6. If X is a finitely generated projective A-module, then ηXY is bijective.
Proof. It is enough to show that

ηA A,Y : (A A)∗ ⊗A Y → HomA (A A, Y )
is bijective. (Note that ηX⊕X ′ ,Y is bijective if and only if ηXY and ηX ′ Y are bijective.)
= A Y and HomA (A A, A Y ) ∼
= AA , AA ⊗A Y ∼
Recall that (A A)∗ = HomA (A A, A A) ∼ =
A Y .
Thus we have isomorphisms AA ⊗A Y → Y , α ⊗ y 7→ α(1)y and Y → HomA (A A, Y ),
y 7→ ρy . Composing these yields a map α ⊗ y 7→ ρα(1)y = ρy ◦ α. We have
ρα(1)y (a) = aα(1)y = α(a)y = (ρy ◦ α)(a).

31.3. The Nakayama functor. Let

ν : Mod(A) → Mod(A)
be the Nakayama functor defined by
ν(X) := D(X ∗ ) = HomK (X ∗ , K) = HomK (HomA (X, A A), K).
Since X ∗ is a right A-module, we know that ν(X) is an A-module.
Lemma 31.7. The functor ν is right exact, and it maps finitely generated projective
modules to injective modules.
Proof. We know that for all modules N the functor HomA (−, N) is left exact. It is
also clear that D is contravariant and exact. Thus ν is right exact.
Now let P be finitely generated projective. It follows that D(P ∗ ) is injective: With-
out loss of generality assume P = A A. Then P ∗ = AA and HomK (AA , K) is injec-
tive.
Set ν −1 := HomA (D(AA ), −).

31.4. Proof of the Auslander-Reiten formula. Now we can prove Theorem

31.1: Let M be a finitely presented module. Thus there exists an exact sequence
p q
P1 −
→ P0 −
→M →0
where P0 and P1 are finitely generated projective modules. Applying ν yields an
exact sequence
ν(p) ν(q)
ν(P1 ) −−→ ν(P0 ) −−→ ν(M) → 0
where ν(P0 ) and ν(P1 ) are now injective modules. Define
τ (M) := Ker(ν(p)).
We obtain an exact sequence

ν(p) ν(q)
0 → τ (M) → ν(P1 ) −−→ ν(P0 ) −−→ ν(M) → 0.
Warning: τ (M) is not uniquely determined by M, since it depends on the chosen

projective presentation of M. But if Mod(A) has projective covers, then we take a
minimal projective presentation of M. In this case, τ (M) is uniquely determined
up to isomorphism.
f g
Notation: If X −
→Y −
→ Z are homomorphisms with Im(f ) ⊆ Ker(g), then set
f g
H(X −
→Y −
→ Z) := Ker(g)/ Im(f ).
We know that Ext1A (N, τ (M)) is equal to

HomA (N,ν(p)) HomA (N,ν(q))

H HomA (N, ν(P1 )) −−−−−−−−→ HomA (N, ν(P0 )) −−−−−−−−→ HomA (N, ν(M)) .
Let u : ν(M) → I be a monomorphism where I is injective. We get

Ext1A (N, τ (M)) = Ker(HomA (N, ν(q)))/ Im(HomA (N, ν(p)))
= Ker(HomA (N, u) ◦ HomA (N, ν(q)))/ Im(HomA (N, ν(p))).
For the last equality we used Lemma 31.3.
Define a map
ξXY := i ◦ D(ηXY : D HomA (X, Y ) → HomA (Y, ν(X))
by
D(ηXY )
D HomA (X, Y ) / D(X ∗ ⊗A Y ) HomK (X ∗ ⊗A Y, K)
TTTT
TTTT
TTTT i
TTTT
TTTT
TTTT
TTTT HomA (Y, HomK (X ∗ , K))
ξXY TTTT
TTTT
TTTT
TTT*
HomA (Y, ν(X))
where i := Adj is the isomorphism given by the adjunction formula Theorem 29.7.
We know by Lemma 31.6 that ξXY is bijective, provided X is finitely generated
projective.
Using this, we obtain a commutative diagram
D HomA (P1 , N) / D HomA (P0 , N) / D HomA (M, N) / 0
ξP1 N ξP0 N ξM N

µ: HomA (N, ν(P1 )) / HomA (N, ν(P0 )) / HomA (N, ν(M))
whose first row is exact and whose second row is a complex. This is based on the
facts that the functor D is exact, and the functor HomA (−, N) is left exact.
Thus we can apply Lemma 31.2 to this situation and obtain
H(µ) = Ker(ξM N )
= Ker(D(ηM N ))
= {α ∈ D HomA (M, N) | α(Im(ηM N )) = 0}.
(If f : V → W is a K-linear map, then the kernel of f ∗ : DW → DV consists of all
g : W → K such that g ◦ f = 0. This is equivalent to g(Im(f )) = 0.)
Recall that
ξM N = Adj ◦ D(ηM N ).
If M is finitely generated, then Lemma 31.4 and Lemma 31.5 yield that
Im(ηM N ) = HomA (M, N)P .
This implies
{α ∈ D HomA (M, N) | α(Im(ηM N )) = 0} = DHomA (M, N).
This finishes the proof of Theorem 31.1.
The isomorphism
DHomA (M, N) → Ext1A (N, τ (M))
is “natural in M and N”:
Let M be a finitely presented A-module, and let f : M → M ′ be a homomorphism.
This yields a map
D HomA (f, N) : D HomA (M, N) → D HomA (M ′ , N)
and a homomorphism τ (f ) : τ (M) → τ (M ′ ). Now one easily checks that the diagram
Ext1A (N, τ (M)) o DHomA (M, N)

Ext1A (N,τ (f )) DHomA (f,N )

Ext1A (N, τ (M ′ )) o DHomA (M ′ , N)
commutes, and that Ext1A (N, τ (f )) is uniquely determined by f .

Similarly, if g : N → N ′ is a homomorphism, we get a commutative diagram

Ext1A (N, τ (M)) o DHomA (M, N)
O
O
Ext1A (g,τ (M )) DHomA (M,g)
Ext1A (N ′ , τ (M)) o DHomA (M, N ′ )
Explicit construction of the isomorphism

φM N : DHomA (M, N) → Ext1A (N, τ (M)).
...
31.5. Existence of Auslander-Reiten sequences. Now we use the Auslander-

Reiten formula to prove the existence of Auslander-Reiten sequences:
Let M = N be a finitely presented A-module, and assume that EndA (M) is a local
ring. We have EndA (M) := HomA (M, M) = EndA (M)/I where
I := EndA (M)P := {f ∈ EndA (M) | f factors through a projective module}.
If M is projective, then HomA (M, M) = 0. Thus, assume M is not projective.
The identity 1M does not factor through a projective module: If 1M = g ◦ f for
some homomorphisms f : M → P and g : P → M with P projective, then f is a
split monomorphism. Since M is indecomposable, it follows that M is projective, a
contradiction.
Note that EndA (M)P is an ideal in EndA (M). It follows that
EndA (M)P ⊆ rad(EndA (M)).
Thus we get a surjective homomorphism of rings

HomA (M, M) → EndA (M)/ rad(EndA (M)).
Recall that EndA (M)/ rad(EndA (M)) is a skew field.
Set
U := {α ∈ DEndA (M) | α(rad(EndA (M))) = 0},
and let ε be a non-zero element in U.
Now our isomorphism
φM M : DHomA (M, M) → Ext1A (M, τ (M))
sends ε to a non-split short exact sequence
f g
η : 0 → τ (M) −
→Y −
→ M → 0.
Let
f g
0→X−
→Y −
→Z→0
be a short exact sequence of A-modules. Then g is a right almost split homo-

morphism if for every homomorphism h : N → Z which is not a split epimorphism
there exists some h′ : N → Y with g ◦ h′ = h.
N
~
h′ ~
~ h
f ~~ g
0 / X / Y / Z / 0
Dually, f is a left almost split homomorphism if for every homomorphism
h : X → M which is not a split monomorphism there exists some h′ : Y → M with
h′ ◦ f = h.
f g
0 / X /Y / Z / 0
}
h
}
} h′
~}
M
Now let
f g
η : 0 → τ (M) −
→Y −
→ M → 0.
be the short exact sequence we constructed above.
Lemma 31.8. g is a right almost split homomorphism.
Proof. Let h : N → M be a homomorphism, which is not a split epimorphism. We

have to show that there exists some h′ : N → Y such that gh′ = h, or equivalently
that the induced short exact sequence h∗ (f, g) splits.
Since h is not a split epimorphism, the map
HomA (M, h) : HomA (M, N) → HomA (M, M)
defined by f 7→ hf is not surjective: If hf = 1M , then h is a split epimorphism, a
contradiction.
The induced map
HomA (M, h) : HomA (M, N) → HomA (M, M)
is also not surjective, since its image is contained in rad(EndA (M)). We obtain a
commutative diagram
φM M
DHomA (M, M) / Ext1A (M, τ (M))
DHomA (M,h) Ext1A (h,τ (M ))

φM N
DHomA (M, N) / Ext1A (N, τ (M))
where φM M (ε) = η and DHomA (M, h)(ε) = 0. This implies Ext1A (h, τ (M))(η) = 0.
Note that the map Ext1A (h, τ (M)) sends a short exact sequence ψ to the short exact
sequence h∗ (ψ) induced by h via a pullback.
So we get h∗ (η) = 0 for all h : N → M which are not split epimorphisms. In other
words, g is a right almost split morphism.
f g
Lemma 31.9. Let 0 → X − → Y − → Z → 0 be a non-split short exact sequence.
Assume that g is right almost split and that EndA (X) is a local ring. Then f is left
almost split.
Proof. Let h : X → X ′ be a homomorphism which is not a split monomorphism.

Taking the pushout we obtain a commutative diagram
f g
0 / X / Y / Z / 0
h h′
f′ g′
ψ: 0 / X′ / Y′ / Z / 0
whose rows are exact. Assume ψ does not split. Thus g ′ is not a split epimorphism.
Since g is right almost split, there exists some g ′′ : Y ′ → Y with g ◦ g ′′ = g ′ . It
follows that g(g ′′f ′ ) = g ′ f ′ = 0.
g
0 / X / Y `A / ZO / 0
A
A g′
g ′′ A
Y′
Since Im(f ) = Ker(g) this implies g ′′ f ′ = f f ′′ for some homomorphism f ′′ : X ′ → X.

Thus
g(g ′′h′ ) = (gg ′′)h′ = g ′ h′ = g.
In other words, g(g ′′h′ − 1Y ) = 0. Again, since Im(f ) = Ker(g), there exists some
p : Y → X with g ′′h′ − 1Y = f p. This implies
f f ′′ h = g ′′ f ′ h
= g ′′ h′ f
= (f p + 1Y )f
= f pf + f
and therefore f (f ′′h − pf − 1X ) = 0. Since f is injective, f ′′ h − pf − 1X = 0. In
other words, 1X = f ′′ h − pf . By assumption, EndA (X) is a local ring. So f ′′ h
or pf is invertible in EndA (X). Thus f is a split monomorphism or h is a split
monomorphism. In both cases, we have a contradiction.
Recall the following result:

Lemma 31.10 (Fitting Lemma). Let M be a module of length m, and let h ∈
EndA (M). Then M = Im(hm ) ⊕ Ker(hm ).
A homomorphism g : M → N is right minimal if all h ∈ EndA (M) with gh = g

are automorphisms. Dually, a homomorphism f : M → N is left minimal if all
h ∈ EndA (N) with hf = f are automorphisms.
Lemma 31.11. Let g : M → N be a homomorphism, and assume that M has

length m. Then there exists a decomposition M = M1 ⊕ M2 with g(M2 ) = 0, and
the restriction g : M1 → N is right minimal.
Proof. Let M = M1 ⊕ M2 with M2 ⊆ Ker(g) and M2 is of maximal length with this

property. If now M1 = M1′ ⊕ M1′′ with M1′′ ⊆ Ker(g), then M1′′ ⊕ M2 ⊆ Ker(g). Thus
M1′′ = 0.
So without loss of generality assume that g(M ′ ) 6= 0 for each non-zero direct sum-
mand M ′ of M. Assume that gh = g for some h ∈ EndA (M).
By the Fitting Lemma we have M = Im(hm ) ⊕ Ker(hm ) for some m. If Ker(hm ) 6=
0, then g(Ker(hm )) 6= 0, and therefore there exists some 0 6= x ∈ Ker(hm ) with
g(x) 6= 0. We get g(x) = ghm (x) = 0, a contradiction. Thus Ker(hm ) = 0. This
implies M = Im(hm ), which implies that h is surjective. It follows that h is an
isomorphism.
f g
Lemma 31.12. Let 0 → X − →Y − → Z → 0 be a non-split short exact sequence. If
X is indecomposable, then g is right minimal.
Proof. Without loss of generality we assume that f is an inclusion map. By Lemma

31.11 We have a decomposition Y = Y1 ⊕Y2 such that Y2 ⊆ Ker(g) and the restriction
g : Y1 → Z is right minimal. It follows that X = Ker(g) = (Ker(g) ∩ Y1 ) ⊕ Y2 .
Case 1: Ker(g) ∩ Y1 = 0. This implies X = Y2 , thus f is a split monomorphism, a
contradiction since our sequence does not split.
Case 2: Y2 = 0. Then Y = Y1 and the restriction g : Y1 → Z coincides with g.
We leave it as an exercise to formulate and prove the dual statements of Lemma

31.11 and 31.12.
Theorem 31.13. Let
f g
0→X−
→Y −
→Z→0
be a short exact sequence of A-modules. Then the following are equivalent:
(i) g is right almost split, and X is indecomposable;

(ii) f is left almost split, and Z is indecomposable;
(iii) f and g are irreducible.
Proof. Use Skript 1, Cor. 11.5 and the dual statement Cor. 11.10 and
Skript 1, Lemma 11.6 (Converse Bottleneck Lemma) and the dual state-
ment Lemma 11.11. Furthermore, we need Skript 1, Cor. 11.3 and Cor.
11.8.
31.6. Properties of τ , Tr and ν.

Lemma 31.14. For any indecomposable A-module M we have

ν −1 (τ (M)) ∼
= Ω2 (M).
Proof. Let P1 → P0 → M → 0 be a minimal projective presentation of M. Thus we

get ab exact sequence
p
0 → Ω2 (M) → P1 −
→ P0 → M → 0.
Applying ν yields an exact sequence
ν(p)
0 → τ (M) → ν(P1 ) −−→ ν(P0 ).
Now we apply ν −1 and obtain an exact sequence
p
0 → ν −1 (τ (M)) → P1 −
→ P0 .
Here we use that ν −1 (ν(P )) ∼
= P , which comes from the fact that ν induces an equiv-
alence between the category of projective A-modules and the category of injective
A-modules. This implies ν −1 (τ (M)) ∼ = Ω2 (M).
Here is the dual statement:

Lemma 31.15. For any indecomposable A-module M we have
ν(τ −1 (M)) ∼
= Σ2 (M).
(i) proj. dim(M) ≤ 1;

(ii) For each injective A-module I we have HomA (I, τ (M)) = 0.
Proof. Clearly, proj. dim(M) ≤ 1 if and only if Ω2 (M) = 0. By the Lemma above
this is equivalent to HomA (D(AA ), τ (M)) = 0. But we know that each indecom-
posable injective A-module is isomorphic to a direct summand of D(AA ). (Let I be
an indecomposable injective A-module. Then D(I) is an indecomposable projective
right A-module. It follows that D(I) is isomorphic to a direct summand of AA . Thus
I∼= DD(I) is a direct summand of D(AA ).) This finishes the proof.
Here is the dual statement, which can be proved accordingly:

(i) inj. dim(M) ≤ 1;

(ii) For each projective A-module P we have HomA (τ −1 (M), P ) = 0.
31.7. Properties of Auslander-Reiten sequences. Let A be a finite-dimensional

K-algebra. In this section, by a “module” we mean a finite-dimensional module. A
homomorphism f : X → Y is a source map for X if the following hold:
(i) f is not a split monomorphism;

(ii) For each homomorphism f ′ : X → Y ′ which is not a split monomorphism
there exists a homomorphism f ′′ : Y → Y ′ with f ′ = f ′′ ◦ f ;
f
X /Y
}
f′
}
} f ′′
~}
′
Y
(iii) If h : Y → Y is a homomorphism with f = h ◦ f , then h is an isomorphism.
f
X / Y h h
Dually, a homomorphism g : Y → Z is a sink map for Z if the following hold:
(i)∗ g is not a split epimorphism;

(ii)∗ For each homomorphism g ′ : Y ′ → Z which is not a split epimorphism there
exists a homomorphism g ′′ : Y ′ → Y with g ′ = g ◦ g ′′ ;
Y′
g ′′ }
} g′
}
~} g
Y / Z
(iii)∗ If h : Y → Y is a homomorphism with g = g ◦ h, then h is an isomorphism.
g
h Y / Z
6
We know already the following facts:
• If
f g
0→X− →Y − →Z→0
is an Auslander-Reiten sequence, then f is a source map for X, and g is a
sink map for Z.
• If X is an indecomposable module which is not injective, then there exists a
source map for X.
• If Z is an indecomposable module which is not projective, then there exists
a sink map for Z.
Lemma 31.18. (i) If f : X → Y is a source map, then X is indecomposable;
(ii) If g : Y → Z is a sink map, then Z is indecomposable.
Proof. We just prove (i): Let X = X1 ⊕ X2 with X1 6= 0 6= X2 , and let π : X → Xi ,

i = 1, 2 be the projection. Clearly, πi is not a split monomorphism, thus there exists
some gi Y → Xi with gi ◦ f = πi . This implies 1X = [π1 , π2 ]t = [g1 , g2t ] ◦ f . Thus f
is a split monomorphism, a contradiction.
Lemma 31.19. Let P be an indecomposable projective module. Then the embedding

rad(P ) → P
is a sink map.
Proof. Denote the embedding rad(P ) → P by g. Clearly, g is not a split epimor-

phism. This proves (i)∗ . Let g ′ : Y ′ → P be a homomorphism which is not a split
epimorphism. Since P is projective, we can conclude that g ′ is not an epimorphism.
Thus Im(g ′) ⊂ P which implies Im(g ′) ⊆ rad(P ). Here we use that P is a local
module. So we proved (ii)∗ . Finally, assume g = gh for some h ∈ EndA (rad(P )).
Since g is injective, this implies that h is the identity 1rad(P ) . This proves (iii)∗ .
Lemma 31.20. Let I be an indecomposable injective module. Then the projection
Q → Q/ soc(Q)
is a source map.
Proof. Dualize the proof of Lemma 31.19.

Corollary 31.21. There a source map and a sink map for every indecomposable
module.
Lemma 31.22. Let f : X → Y be a source map, and let f ′ : X → Y ′ be an arbitrary
homomorphism. Then the following are equivalent:
(i) There exists a homomorphism f ′′ : X → Y ′′ and an isomorphism h : Y →

Y ′ ⊕ Y ′′ such that the diagram
f
X /Y
» – vvv
f′ vv
f ′′ vvv h
zv
Y ′ ⊕ Y ′′
commutes.
(ii) f ′ is irreducible or Y ′ = 0.
Proof. (ii) =⇒ (i): If Y ′ = 0, then choose f ′′ = f . Thus, let f ′ be irreducible. It

follows that f ′ is not a split monomorphism. Thus there exists some h′ : Y → Y ′
with f ′ = h′ f .
f
X /Y
}
f′
}
} h′
~}
Y′
Now f ′ is irreducible and f is not a split monomorphism. Thus h′ is a split epimor-
phism. Let Y ′′ = Ker(h′ ). This is a direct summand of of Y . Let p : Y → Y ′′ be the
corresponding projection. We obtain a commutative diagram

f
X II /Y
II h ′i
I h
» ′ –III p
f I$
f ′′ Y ⊕ Y ′′
′
h′
Clearly, p is an isomorphism. Now set f ′′ := pf .
h i
f′
(i) =⇒ (ii): Without loss of generality we assume h = 1. Thus f = f ′′ :X →
Y = Y ′ ⊕ Y ′′ . We have to show: If Y ′ 6= 0, then f ′ is irreducible.
(a): f ′ is not a split monomorphism: Otherwise f would be a split monomorphism,
a contradiction.
(b): f ′ is not a split epimorphism: We know that Y ′ 6= 0 and X is indecomposable.
If f ′ is a split epimorphism, we get that f ′ is an isomorphism and therefore a split
monomorphism, a contradiction.
(c): Let f ′ = hg.
f′
X / Y′
} >
}
g }}}
}
}} h
C
h i
f′
There is a source map f ′′ : X → Y ′ ⊕ Y ′′ . Assume g is not a split monomorphism.
Then there exists some [g ′ , g ′′ ] : Y ′ ⊕ Y ′′ such that the diagram
» –
f′
f ′′
X / Y ′ ⊕ Y ′′
uu
g uuu
u ′ ′′
uz uu [g ,g ]
C
commutes. Thus g = g ′f ′ + g ′′ f ′′ . It follows that the diagram
» –
f′
f ′′
X / Y ′ ⊕ Y ′′
» – q
f′ qqq
f ′′ qhqq ′ ′′ i
xqqq hg hg
0 1
Y′⊕Y ′′
h ′i ′ ′′
commutes. Since ff′′ is left minimal, the map hg0 hg1 is an automorphism. Thus
hg ′ is an automorphism. This implies that h is a split epimorphism. So we have
shown that f ′ is irreducible.
Corollary 31.23. Let f : X → Y be a source map, and let h : Y → M be a split
epimorphism. Then h ◦ f : X → M is irreducible.
Here is the dual statement which is proved accordingly:

Lemma 31.24. Let g : Y → Z be a sink map, and let g ′ : Y ′ → Z be an arbitrary

homomorphism. Then the following are equivalent:
(i) There exists a homomorphism g ′′ : Y ′′ → Z and an isomorphism h : Y ′ ⊕

Y ′′ → Y such that the diagram
Y ′ ⊕ Y ′′
v » –
h vvv g′
v
vv g ′′
vz v g
Y / Z
commutes.
(ii) g ′ is irreducible or Y ′ = 0.
Corollary 31.25. Let g : Y → Z be a sink map, and let h : M → Y be a split
monomorphism. Then g ◦ h : M → Z is irreducible.
Here is again the (preliminary) definition of the Auslander-Reiten quiver ΓA

of A: The vertices are the isomorphism classes of indecomposable A-modules, and
there is an arrow [X] → [Y ] if and only if there exists an irreducible map X →
Y . Furthermore, we draw a dotted arrow [τ (X)] o_ _ _ [X] for each non-projective
indecomposable A-module X.
A (connected) component of ΓA is a full subquiver Γ = (Γ0 , Γ1 ) of ΓA such that

the following hold:
(i) For each arrow [X] → [Y ] in ΓA with {[X], [Y ]}∩Γ0 6= ∅ we have {[X], [Y ]} ⊆
Γ0 ;
(ii) If [X] and [Y ] are vertices in Γ, then there exists a sequence
([X1 ], [X2 ], . . . , [Xt ])
of vertices in Γ with [X] = [X1 ], [Y ] = [Xt ], and for each 1 ≤ i ≤ t − 1 there
is an arrow [Xi ] → [Xi+1 ] or an arrow [Xi+1 ] → [Xi ].
L
Corollary 31.26. Let X → Y be a source map, and let Y = ti=1 Yini where Yi is
indecomposable, ni ≥ 1 and Yi 6∼
= Yj for all i 6= j. Then there are precisely t arrows
in ΓA starting at [X], namely [X] → [Yi ], 1 ≤ i ≤ t.
Lemma 31.27. A vertex [X] is a source in ΓA if and only if X is simple projective.
Proof. Assume P is a simple projective module. Then any non-zero homomorphism

X → P is a split epimorphism. So [P ] has to be a source in ΓA . Now assume P
is projective, but not simple. Then the embedding rad(P ) → P is a non-zero sink
map. It follows that [P ] cannot be a source in ΓA . Finally, if Z is an indecomposable
non-projective A-module, then again there exists a non-zero sink map Y → Z. So
[Z] cannot be a source. This finishes the proof.
Lemma 31.28. A source map X → Y is not a monomorphism if and only if X is
injective.
We leave it to the reader to formulate the dual statements.

Corollary 31.29. ΓA is a locally finite quiver.
f g
Let 0 → X − → Y − → Z → 0 be an Auslander-Reiten sequence in mod(A). Thus,
by definition f and g are irreducible. We proved already that X and Z have to be
indecomposable (Skript 1). It follows that we get a commutative diagram
f′ g′
0 / X / E / τ −1 (X) / 0
h h′
f g
0 / X / Y / Z / 0
′
where h and h are isomorphisms.
Here τ −1 (X) := TrD(X).
Source maps are unique in the following sense: Let X be an indecomposable A-
module which is not injective, and let f : X → Y and f : X → Y ′ be source maps.
By g : Y → Z and g ′ : Y ′ → Z ′ we denote the projections onto the cokernel of f and
f ′ , respectively. Then we get a cimmutative diagram
f′ g′
0 / X / Y′ / Z′ / 0
h h′
f g
0 / X / Y / Z / 0
where h and h′ are isomorphisms.
Dually, sink maps are unique as well.
31.8. Digression: The Brauer-Thrall Conjectures. Assume that A is a finite-

dimensional K-algebra, and let S1 , . . . , Sn be a set of representatives of isomorphism
classes of simple A-modules. Then the quiver of A has vertices 1, . . . , n and there
are exactly dim Ext1A (Si , Sj ) arrows from i to j.
The algebra A is connected if the quiver of A is connected.

Lemma 31.30. For a finite-dimensional algebra A the following are equivalent:
(i) A is connected;
(ii) For any indecomposable projective A-modules P 6∼ = P ′ there exists a tu-
ple (P1 , P2 , . . . , Pm ) of indecomposable projective modules such that P1 =
P , Pm = P ′ and for each 1 ≤ i ≤ m − 1 we have HomA (Pi , Pi+1 ) ⊕
HomA (Pi+1 , Pi) 6= 0;
(iii) For any simple A-modules S and S ′ there exists a tuple (S1 , S2 , . . . , Sm ) of
simple modules such that S1 = S, Sm = S ′ and for each 1 ≤ i ≤ m − 1 we
have Ext1A (Si , Si+1 ) ⊕ Ext1A (Si+1 , Si ) 6= 0;
(iv) If A = A1 × A2 then A1 = 0 or A2 = 0;
(v) 0 and 1 are the only central idempotents in A.
Proof. Exercise. Hint: If ExtA (Si , Sj ) 6= 0, then there exists a non-split short exact
sequence
f g
0 → Sj −
→E− → Si → 0.
Then there exists an epimorphism pi : Pi → Si . This yields a homomorphism
p′i : Pi → E such that gp′i = pi . Clearly, h′ has to be an epimorphism. (Why?)
Let pj : Pj → Sj be the obvious epimorphism. Then there exists an epimorphism
p′j : Pj → E such that f pj = p′j . Next, there exists a non-zero homomorphism
q : Pj → Pi such that pi q = f pj .
Theorem 31.31 (Auslander). Let A be a finite-dimensional connected K-algebra,
and let C be a component of the Auslander-Reiten quiver of A. Assume that there
exists some b such that all indecomposable modules in C have length at most b. Then
C is a finite component and it contains all indecomposable A-modules. In particular,
A is representation-finite.
Proof. (a): Let X be an indecomposable A-module such that there exists a non-zero
homomorphism h : X → Y for some [Y ] ∈ C. We claim that [X] ∈ C: Let
t1
M
(1) (1) (1) (1)
g = [g1 , . . . , gt1 ] : Yi →Y
i=1
(1)
be the sink map ending in Y , where is indecomposable for all 1 ≤ i ≤ t1 . If h
Yi
is a split epimorphism, then h is an isomorphism and we are done. Thus, assume
h0 := h is not a split epimorphism. It follows that there exists a homomorphism
 (1) 
f1 Mt1
(1)  ..  (1)
f = . :X→ Yi
(1) i=1
ft1
such that
t1
X
(1) (1) (1) (1)
h0 = g f = gi fi : X → Y.
i=1
(1) (1) (1)
Since h0 6= 0, there exists some 1 ≤ i1 ≤ t1 such that gi1 ◦ fi1 6= 0. Set h1 := fi1
(1)
and h′1 := gi1 . Next, assume that for each 1 ≤ k ≤ n − 1 we already constructed a
non-invertible homomorphism
(k) (k−1)
h′k : Yik → Yik−1 ,
(k) (0)
where [Yik ] ∈ C and Yi0 := Y , and a homomorphism
(k)
hk : X → Yik
such that h′1 ◦ · · · ◦ h′k ◦ hk 6= 0. So we get the following diagram:
X X ··· X X
hn−1 hn−2 h1 h0

(n−1) (n−2) (1)
Yin−1 / Yin−2 / ··· / Yi1 / Y
h′n−1 h′n−2 h′2 h′1
with h′1 ◦ h′2 ◦ · · · ◦ h′n−1 ◦ hn−1 6= 0.
If hn−1 is an isomorphism, then X ∼

(n−1)
= Yin−1 and therefore [X] ∈ C.
(n−1)
Thus assume that hn−1 : X → Yin−1 is non-invertible. Let
tn
M
(n) (n) (n) (n−1)
g (n) = [g1 , . . . , gtn ] : Yi → Yin−1
i=1
(n−1) (n)
be the sink map ending in Yin−1 ,
where is indecomposable for all 1 ≤ i ≤ tn .
Yi
Since hn−1 is not a split epimorphism, there exists a homomorphism
 (n) 
f1 Mtn
(n)  ..  (n)
f = . :X → Yi
(n) i=1
ftn
such that
tn
X (n) (n) (n−1)
hn−1 = g (n) f (n) = gi fi : X → Yin−1 .
i=1
Since h′1 ◦ h′2 ◦ · · · ◦ h′n−1 ◦ hn−1 6= 0, there exists some 1 ≤ in ≤ tn such that
(n) (n)
h′1 ◦ h′2 ◦ · · · ◦ h′n−1 ◦ gin ◦ fin 6= 0.
(n) (n)
Set hn := fin and h′n := gin . Thus
h′1 ◦ h′2 ◦ · · · ◦ h′n−1 ◦ h′n ◦ hn 6= 0.
Clearly, h′n is non-invertible, since h′n is irreducible.
If n ≥ 2b − 2 we know by the Harada-Sai Lemma that hn has to be an isomorphism.

This finishes the proof of (a).
(b): Dually, if Z is an indecomposable A-module such that there exists a non-zero
homomorphism Y → Z for some [Y ] ∈ C, then [Z] ∈ C.
(c): Let Y be an indecomposable A-module with [Y ] ∈ C, and let S be a composition

factor of Y . Then there exists a non-zero homomorphism PS → Y where PS is
the indecomposable projective module with top S. By (a) we know that [PS ] ∈
C. Now we use Lemma 31.30, (iii) in combination with (a) and (b) to show that
all indecomposable projective A-modules lie in C. Finally, if Z is an arbitrary
indecomposable A-module, then again there exists an indecomposable projective
module P and a non-zero homomorphism P → Z. Now (b) implies that [Z] ∈ C.
It follows that C = (ΓA , dA ). By the proof of (a) and (b) we know that there is a
path of length at most 2b − 2 in C which starts in [P ] and ends in [Z]. It is also
clear that C has only finitely many vertices: Since ΓA is a locally finite quiver, for
each projective vertex [P ] there are only finitely many paths of length at most 2b − 2
starting in [P ].
Corollary 31.32 (1st Brauer-Thrall Conjecture). Let A be a finite-dimensional K-
algebra. Assume there exists some b such that all indecomposable A-modules have
length at most b. Then A is representation-finite.
Thus the 1st Brauer-Thrall Conjecture says that bounded representation type implies
finite representation type. There exists a completely different proof of the 1st Brauer-
Thrall conjecture due to Roiter, using the Gabriel-Roiter measure.
Conjecture 31.33 (2nd Brauer-Thrall Conjeture). Let A be a finite-dimensional
algebra over an infinite field K. If A is representation-infinite, then there exists
some d ∈ N such that the following hold: For each n ≥ 1 there are infinitely many
isomorphism classes of indecomposable A-modules of dimension nd.
Theorem 31.34 (Smalø). Let A be a finite-dimensional algebra over an infinite field
K. Assume there exists some d ∈ N such that there are infinitely many isomorphism
classes of indecomposable A-modules of dimension d. Then for each n ≥ 1 there are
infinitely many isomorphism classes of indecomposable A-modules of dimension nd.
Thus to prove Conjecture 31.33, the induction step is already known by Theorem
31.34. Just the beginning of the induction is missing...
Conjecture 31.33 is true if K is algebraically closed. This was proved by Bautista
using the well developed theory of representation-finite algebras over algebraically
closed fields.
31.9. The bimodule of irreducible morphisms. Let A be a finite-dimensional

K-algebra, and as before let mod(A) be the category of finitely generated A-modules.
All modules are assumed to be finitely generated.
For indecomposable A-modules X and Y let
radA (X, Y ) := {f ∈ HomA (X, Y ) | f is not invertible}.
In particular, if X 6∼
= Y , then radA (X, Y ) = HomA (X, Y ). If X = Y , then
radA (X, X) = rad(EndA (X)) := J(EndA (X)).
L L
Now let X = si=1 Xi and Y = tj=1 Yj be A-modules with Xi and Yj indecompos-
able for all i and j. Recall that we can think of an endomorphism f : X → Y as a
matrix  
f11 · · · fs1
f =  ... .. 
.
f1t · · · fst
where fij : Xi → Yj is an homomorphism for all i and j. Set
 
radA (X1 , Y1 ) · · · radA (Xs , Y1)
radA (X, Y ) :=  .. .. .
. .
radA (X1 , Yt ) · · · radA (Xs , Yt )
Thus radA (X, Y ) ⊆ HomA (X, Y ).
Lemma 31.35. For A-modules X and Y we have f ∈ / radA (X, Y ) if and only if
there exists a split monomorphism u : X ′ → X and a split epimorphism p : Y → Y ′
such that p ◦ f ◦ u : X ′ → Y ′ is an isomorphism and X ′ 6= 0.
Proof. Exercise.
For A-modules X and Y let rad2A (X, Y ) be the set of homomorphisms f : X → Y

with f = h ◦ g for some g ∈ radA (X, M), h ∈ radA (M, Y ) and M.
Lemma 31.36. Let X and Y be indecomposable A-modules. For a homomorphism
f : X → Y the following are equivalent:
(i) f is irreducible;
(ii) f ∈ radA (X, Y ) \ rad2A (X, Y ).
Proof. Assume f : X → Y is irreducible. Since X and Y are indecomposable we

know that f is an isomorphism if and only if f is a split monomorphism if and only
if f is a split epimorphism. Thus f ∈ radA (X, Y ). Assume f ∈ rad2A (X, Y ).
...

For indecomposable A-modules X and Y define

IrrA (X, Y ) := radA (X, Y )/ rad2A (X, Y ).
We call IrrA (X, Y ) the bimodule of irreducible maps from X to Y .
Set F (X) := EndA (X)/ rad(EndA (X)) and F (Y ) := EndA (Y )/ rad(EndA (X)).
Since X and Y are indecomposable, we know that F (X) and F (Y ) are skew fields.
Lemma 31.37. IrrA (X, Y ) is an F (X)op -F (Y )-bimodule.
Proof. Let f ∈ IrrA (X, Y ), g ∈ F (X) and h ∈ F (Y ), where f ∈ radA (X, Y ),

g ∈ EndA (X) and h ∈ EndA (Y ). Define
g ⋆ f := f g,
h · f := hf.
We have to check that this is well defined: We have a map
EndA (Y ) × HomA (X, Y ) × EndA (X) → HomA (X, Y )
defined by (h, f, g) 7→ hf g. Clearly, if f ∈ radA (X, Y ), then hf and f g are in
radA (X, Y ). It follows that radA (X, Y ) is an EndA (X)op -EndA (Y )-bimodule. It is
also clear that rad2A (X, Y ) is a subbimodule: Let f = f2 f1 ∈ rad2A (X, Y ) where f1 ∈
radA (X, C) and f2 ∈ radA (C, Y ) for some C. Then hf = (hf2 )f1 and f g = f2 (f1 g),
so they are both in rad2A (X, Y ). Furthermore, the images of the maps radA (X, Y ) ×
rad(EndA (X)) → radA (X, Y ), (f, g) 7→ f g and radA (X, Y ) × rad(EndA (Y )) →
radA (X, Y ), (h, f ) 7→ hf are both contained in rad2A (X, Y ). Thus IrrA (X, Y ) is
annihilated by rad(EndA (X)op ) and rad(EndA (Y )). This implies that IrrA (X, Y ) is
an F (X)op -F (Y )-bimodule.
Lemma 31.38. Let Z be an indecomposable non-projective A-module. Then F (Z) ∼ =
F (τ (Z)).
Proof. Exercise.
Lemma 31.39. Assume K is algebraically closed. If X is an indecomposable A-
module, then F (X) ∼
= K.
Proof. Exercise.
Theorem 31.40. Let M and N be indecomposable A-modules. Let g : Y → N be a
sink map for N. Write
Y = Mt ⊕ Y ′
with t maximal. Thus g = [g1 , . . . , gt , g ′] where gi : M → N, 1 ≤ i ≤ t and g ′ : Y ′ →
N are homomorphisms. Then the following hold:
(i) The residue classes of g1 , . . . , gt in IrrA (M, N) form a basis of the F (M)op -
vector space IrrA (M, N);
(ii) We have
dimK (IrrA (M, N))
t = dimF (M )op (IrrA (M, N)) = .
dimK (F (M))
Dually, let f : M → X be a source map for M. Write

X = Ns ⊕ X′
with s maximal. Thus f = t [f1 , . . . , fs , f ′ ] where fi : M → N, 1 ≤ i ≤ s and
f ′ : M → X ′ are homomorphisms. Then the following hold:
(iii) The residue classes of f1 , . . . , fs in IrrA (M, N) form a basis of the F (N)-
vector space IrrA (M, N);
(iv) We have
dimK (IrrA (M, N))
s = dimF (N ) (IrrA (M, N)) = .
dimK (F (N))
We have s = t if and only if dimK (F (M)) = dimK (F (N)) or s = t = 0.
Proof. (a): First we show that the set {g1 , . . . , gt } is linearly independent in the
F (M)op -vector space IrrA (M, N):
Assume
t
X
(2) λi ⋆ g i = 0
i=1
where λi ∈ EndA (M), gi ∈ radA (M, N), λi = λi + rad(EndA (M)), gi = gi +

rad2A (M, N) and 0 = 0 + rad2A (M, N). By definition λi ⋆ gi = gi λi . We have to show
that λi = 0, i.e. λi ∈ rad(EndA (M)) for all i.
Assume λ1 ∈
/ rad(EndA (M)). In other words, λ1 is invertible. We get
t
" λ1 # " λ1 #
X . .
gi λi = [g1 , . . . , gt , g ′] ◦ .. = g ◦ .. : M → N.
λt λt
i=1 0 0
By Equation (2) we know that this map is contained in rad2A (M, N).
" λ1 #
.
Clearly, .. is a split monomorphism, since
λt
0
" λ1 #
..
[λ−1
1 , 0, . . . , 0] ◦ . = 1M .
λt
0
P
Using Lemma 31.24 this implies that ti=1 giλi is irreducible and can therefore not
be contained in rad2A (M, N), a contradiction.
(b): Next, we show that {g1 , . . . , gt } generates the F (M)op -vector space IrrA (M, N):
Let u : M → N be a homomorphism with u ∈ radA (M, N). We have to show that
u := u + rad2A (M, N) is a linear combination of g1 , . . . , gt .
Since g is a sink map and u is not a split epimorphism, we get a commutative
diagram
2 u1 3
6 .. 7
4 .mM
5
ut
m m m
u′
m u
vm m
Mt ⊕Y′ /N
[g1 ,...,gt ,g ′ ]
Pt
such that u = i=1 gi u i + g ′ u ′ .
We know that g ′ ∈ radA (Y ′ , N), since g ′ is just the restriction of the sink map g
to a direct summand Y ′ of Y . Thus g ′ is irreducible or g ′ = 0. Furthermore, M is
indecomposable and Y ′ does not contain any direct summand isomorphic to M. So
u′ ∈ radA (M, Y ′ ). Thus implies g ′ u′ ∈ rad2A and therefore g ′u′ = 0. It follows that
t
X t
X
u= u i ⋆ gi + g ′ u′ = u i ⋆ gi .
i=1 i=1
The second part of the theorem is proved dually.

Corollary 31.41. Let
0 → τ (Z) → Y → Z → 0
be an Auslander-Reiten sequence, and let M be indecomposable. Then
dim K IrrA (M, Z) = dim K IrrA (τ (Z), M).
Proof. Let t be maximal such that Y = M t ⊕ Y ′ for some module Y ′ . Then we get
dimK IrrA (M, Z) dimK IrrA (τ (Z), M)
t= = .
dimK F (M) dimK F (M)

It is often quite difficult to construct Auslander-Reiten sequences. But if there exists

a projective-injective module, one gets one such sequence for free:
Lemma 31.42. Let I be an indecomposable projective-injective A-module, and as-

sume that I is not simple. Then there is an Auslander-Reiten sequence of the form
0 → rad(I) → rad(I)/ soc(I) ⊕ I → I/ soc(I) → 0.
Proof. ...

31.10. Translation quivers and mesh categories. Let Γ = (Γ0 , Γ1 , s, t) be a

quiver (now we allow Γ0 and Γ1 to be infinite sets).
We call Γ locally finite if for each vertex y there are at most finitely many arrows
ending at y and there are are most finitely many arrows starting at y.
If there is an arrow x → y then x is called a direct predecessor of y, and if there

is an arrow y → z then z is a direct successor of y.
Let y − be the set of direct predecessors of y, and let y + be the set of direct successors
of y. Note that we do not assume that y − and y + are disjoint.
A path of length n ≥ 1 in Γ is of the form w = (α1 , . . . , αn ) where the αi are arrows
such that s(αi ) = t(αi+1 ) for 1 ≤ i ≤ n − 1. We say that w starts in s(w) := s(αn ),
and w ends in t(w) := t(α1 ). In this case, s(w) is a predecessor of t(w), and t(w)
is a successor of s(w).
Additionally, for each vertex x of Γ there is a path 1x of length 0 with s(1x ) =
t(1x ) = x. For vertices x and y let W (x, y) be the set of paths from x to y. If a
path w in Γ starts in x and ends in y, we say that x is a predecessor of y, and y is a
successor of x. If w = (α1 , . . . , αn ) has length n ≥ 1, and if s(w) = t(w), then w is
called a cycle in Γ. In this case, we say that s(α1 ), . . . , s(αn ) lie on the cycle w.
A vertex x in a quiver Γ is reachable if there are just finitely many paths in Γ

which end in x.
It follows immediately that a vertex x is reachable if and only if x has only finitely
many predecessors and none of these lies on a cycle. Of course, every predecessor of
a reachable vertex is again reachable. We define a chain
∅ = −1 Γ ⊆ 0 Γ ⊆ · · · ⊆ n−1 Γ ⊆ n Γ ⊆ · · ·
of subsets of Γ0 .
By definition −1 Γ = ∅. For n ≥ 0, if n−1 Γ is already defined, then let n Γ be the set
of all vertices z of Γ such that z − ⊆ n−1 Γ.
By n Γ we denote the full subquiver of Γ with vertices n Γ. Set

[ [
∞ Γ := nΓ and ∞ Γ := n Γ.
n≥0 n≥0
Clearly, ∞Γ is the set of all reachable vertices of Γ.

Now let K be a field. We define the path category KΓ as follows:
The objects in KΓ are the vertices of Γ. For vertices x, y ∈ Γ0 , we take as morphism
set HomKΓ (x, y), the K-vector space with basis W (x, y).
The composition of morphisms is by definition K-bilinear, so it is enough to define
the composition of two basis elements: First, the path 1x of length 0 is the unit
element for the object x. Next, if w = (α1 , . . . , αn ) ∈ W (x, y) and v = (β1 , . . . , βm ) ∈
W (y, z), then define
vw := v · w := (β1 , . . . , βm , α1 , . . . , αn ) ∈ W (x, z).
This is again a path since s(βm ) = t(α1 ).
We call Γ = (Γ0 , Γ1 , s, t, τ, σ) a translation quiver if the following hold:
(T1) (Γ0 , Γ1 , s, t) is a locally finite quiver without loops;

(T2) τ : Γ′0 → Γ0 is an injective map where Γ′0 is a subset of Γ0 , and for all z ∈ Γ′0
and every y ∈ Γ0 the number of arrows y → z equals the number of arrows
τ (z) → y;
(T3) σ : Γ′1 → Γ1 is an injective map with σ(α) : τ (z) → y for each α : y → z,
where Γ′1 is the set of all arrows α : y → z with z ∈ Γ′0 .
Note that a translation quiver can have multiple arrows between two vertices.
If Γ = (Γ0 , Γ1 , s, t, τ, σ) is a translation quiver, then τ is called the translation of Γ.
The vertices in Γ0 \Γ′0 are the projective vertices, and Γ0 \τ (Γ′0 ) are the injective
vertices. If Γ does not have any projective or injective vertices, then Γ is stable.
A translation quiver Γ is preprojective if the following hold:
(P1) There are no oriented cycles in Γ;

(P2) If z is non-projective vertex, then z − 6= ∅;
(P3) For each vertex z there exists some n ≥ 0 such that τ n (z) is a projective
vertex.
A translation quiver Γ is preinjective if the following hold:
(I1) There are no oriented cycles in Γ;

(I2) If z is non-injective vertex, then z + 6= ∅;
(I3) For each vertex z there exists some n ≥ 0 such that τ −n (z) is an injective
vertex.
Again, let Γ be a translation quiver. A function f : Γ0 → Z is additive if for all

non-projective vertices z we have
X
f (τ (z)) + f (z) = f (y).
y∈z −
For example, if C is a component of the Auslander-Reiten quiver of an algebra A

with dimK IrrA (X, Y ) ≤ 1 for all X, Y ∈ C, then f ([X]) := l(X) is an additive
function on the translation quiver C.
We will often investigate translation quivers without multiple arrows. In this case,
we do not mention the map σ, since it is uniquely determined by the other data.
By condition (T2) we know that each non-projective vertex z of Γ yields a subquiver

of the form
y
= 1=
σ(α1 ) || ==
==α1
|||
| ==
|| =
τ (z) .. z
BB . @
BB
BB
σ(αn ) BB αn
!
yn
Such a subquiver is called a mesh in Γ. (Recall that there could be more than one
arrow from τ (z) to yi and therefore also from yi to z. In this case, the map σ yields
a bijection between the set of arrows yi → z and the set of arrows τ (z) → yi .)
Now let K be a field, and let Γ = (Γ0 , Γ1 , s, t, τ, σ) be a translation quiver. We
look at the path category KΓ := K(Γ0 , Γ1 , s, t) of the quiver (Γ0 , Γ1 , s, t). For each
non-projective vertex z we call the linear combination
X
ρz := α · σ(α)
α : y→z
the mesh relation associated to z, where the sum runs over all arrows ending in
z. This is an element in the path category KΓ.
The mesh category KhΓi of the translation quiver Γ is by definition the factor
category of KΓ modulo the ideal generated by all mesh relations ρz where z runs
through the set Γ′0 of all non-projective vertices of Γ.
Example: Let Γ be the following translation quiver:
w
β ~~? @@@
~ @@δ
~~~ @@
~~ @
o_ _ _ _ _ _ _ y
? v @@ ~? ??? ξ
α @@ γ ǫ ~~~ ??
@@ ~
@@ ~~ ??
~
u o
_ _ _ _ _ _ _ x o
_ _ _ _ _ _ _z
This is a translation quiver without multiple arrows. The dashed arrows describe τ ,
they start in some z and end in τ (z). Thus we have three projective vertices u, v, w
and three injective vertices w, y, z. The mesh relations are

γα = 0,
δβ + ǫγ = 0,
ξǫ = 0.
For example, in the path category KΓ we have dim HomKΓ(u, y) = 2. But in the
mesh category KhΓi, we obtain HomKhΓi (u, y) = 0.
Assume that Γ = (Γ0 , Γ1 , s, t, τ, σ) is a translation quiver without multiple arrows.
A function
d : Γ0 ∪ Γ1 → N 1
is a valuation for Γ if the following hold:
(V1) If α : x → y is an arrow, then d(x) and d(y) divide d(α);

(V2) We have d(τ (z)) = d(z) and d(τ (z) → y) = d(y → z) for every non-projective
vertex z and every arrow y → z.
If d is a valuation for Γ, then we call (Γ, d) a valued translation quiver. If d is a

valuation for Γ with d(x) = 1 for all vertices x of Γ, then d is a split valuation.
Our main and most important examples of valued translation quivers are the fol-
lowing:
Let A be a finite-dimensional K-algebra. For an A-module X denote its isomorphism

class by [X]. If X and Y are indecomposable A-modules, then as before define
F (X) := EndA (X)/ rad(EndA (X))
and
IrrA (X, Y ) := radA (X, Y )/ rad2A (X, Y ).
Let τA be the Auslander-Reiten translation of A.
The Auslander-Reiten quiver ΓA of A has as vertices the isomorphism classes
of indecomposable A-modules. If X and Y are indecomposable A-modules, there
is an arrow [X] / [Y ] if and only if IrrA (X, Y ) 6= 0. Define τ ([Z]) := [τA (Z)]
if Z is indecomposable and non-projective. In this case, we draw a dotted arrow
[τA (Z)] o_ _ _ [Z].
For each vertex [X] of ΓA define

dX := dA ([X]) := dimK F (X),
and for each arrow [X] → [Y ] let
dXY := dA ([X] → [Y ]) := dimK IrrA (X, Y ).
XY d
When we display arrows in ΓA we often write [X] −−→ [Y ].
For an indecomposable projective module P and an indecomposable module X let

rXP be the multiplicity of X in a direct sum decompositions of rad(P ) into inde-
composables, i.e.
rad(P ) = X rXP ⊕ C
for some module C and rXP is maximal with this property.
Lemma 31.43. For a finite-dimensional K-algebra the following hold:
(i) Γ(A) := (ΓA , dA ) is a translation quiver;

(ii) The valuation dA is split if and only if for each indecomposable A-module X
we have EndA (X)/ rad(EndA (X)) ∼ = K (For example, if K is algebraically
closed, then dA is a split valuation.);
(iii) A vertex [X] of (Γ, dA ) is projective (resp. injective) if and only if X is
projective (resp. injective).
Proof. We have IrrA (X, X) = 0 for every indecomposable A-module X. (Recall that
every irreducible map between indecomposable modules is either a monomorphism
or an epimorphism.) Thus the quiver ΓA does not have any loops. If Z is an
indecomposable non-projective module, then the skew fields F (τA (Z)) and F (Z) are
isomorphic, and dimK IrrA (τA (Z), Y ) = dimK IrrA (Y, Z) for each indecomposable
module Y . This shows that ΓA is locally finite, and that the conditions (T1), (T2),
(T3) and (V2) are satisfied. Since IrrA (X, Y ) is an F (X)op -F (Y )-bimodule, also
(V1) holds.
...
If C is a connected component of (ΓA , dA ) such that C is a preprojective (resp. prein-

jective) translation quiver, then C is called a preprojective (resp. preinjective)
component of ΓA .
An indecomposable A-module X is preprojective (resp. preinjective) if [X] lies
in a preprojective (resp. preinjective) component of ΓA .
Let Γ be a translation quiver with a split valuation d. Then we define the expansion
(Γ, d)e of Γ as follows:
The quiver (Γ, d)e has the same vertices as (Γ, d), and also the same translation τ .
For every arrow α : x → y in Γ, we get a sequence of d(x → y) arrows αi : x → y
where 1 ≤ i ≤ d(α). (Thus the arrows in (Γ, d)e starting in x and ending in y are
enumerated, there is a first arrow, a second arrow, etc.) Now σ sends the ith arrow
y → z to the ith arrow τ (z) → y provided z is a non-projective vertex.
31.11. Examples of Auslander-Reiten quivers. (a): Let K = R and set

R C
A= ⊂ M2 (C).
0 C
Clearly, A is a 5-dimensional K-algebra. Let e11 = ( 10 00 ) and e22 = ( 00 01 ). Set

M = Ae11 = [ R0 ] and N = Ae22 = [ CC ] .
These are the indecomposable projective A-modules, and we have A A = M ⊕ N.
We can identify HomA (M, N) with C since
= e11 Ae22 ∼
HomA (M, N) = radA (M, N) ∼ = C.
Next, we observe that rad(M) = 0 and rad(N) = [ C0 ] = [ R0 ] ⊕ [ R0 ]. It follows that

the obvious map M ⊕ M → N is a sink map. Furthermore, it is easy to check that
EndA (M) ∼= R, F (M) ∼ = C and F (N) ∼
= R, EndA (N) ∼ = C.
We have
dimK IrrA (M, N) dimK IrrA (M, N)
2 = rM N = =
dimK F (M) 1
This implies dimK IrrA (M, N) = 2. Thus M → N is a source map.
C/R We get an
Auslander-Reiten sequence 0 → M → N → Q → 0 where Q = C .
Next, we look for the source map starting in N: We have dimK IrrA (N, Q) =
dimK IrrA (M, N) = 2 and dimK F (Q) = 1. Thus N → Q ⊕ Q is a source map.
We obtain an Auslander-Reiten sequence 0 → N → Q ⊕ Q → R → 0 where
R = [ C0 ].
The modules τ −1 (M) and τ −1 (N) are injective, thus the following is the Auslander-
Reiten quiver of A:
dN = 2 [ C ] o_ _ _ _ _ _ _ _ _ [ C0 ]
>} C DD z<
2 }}} DD 2 2 zzz
DDD z
}} D" zz
}} C/R z
dM = 1 [ R ] o_ _ _ _ _ _ _ _
0 C
So there are just four indecomposable A-modules up to isomorphism. Using dimen-

sion vectors it looks as follows:
2 o_ _ _ _ _ _ _ 0
2 @ 1 << 1
<< @
<
<<
<
1 o_ _ _ _ _ _ _ 1
1 0 1
Note that the valuation of the vertices remains constant on τ -orbits (and τ −1 -orbits),
so it is enough to display them only once per orbit.
(b): Next, let
k K
A= ⊂ M2 (K)
0 K
√
where k ⊂ K is a field extension of dimension three, e.g. k = Q and K = Q( 3 2).
The indecomposable projective A-modules are
M = Ae11 = [ k0 ] and N = Ae22 = [ K
K].
In this case there are 6 indecomposable A-modules, and the Auslander-Reiten quiver
ΓA looks like this:
3 o_ _ _ _ _ _ _ 3 0
dN = 3 @ 1 << @ 2
o_ _ _ _ _ _ _
<< @ 1
<< 3 << 3
3 << 3 << 3
<< <<

1 o_ _ _ _ _ _ _ 2 o_ _ _ _ _ _ _ 1
dM = 1 0 1 1
(c): Here is the Auslander-Reiten quiver of the algebra A = KQ/I where Q is the
quiver
1=
==
a
==c
==

2 == 3
==
=
b ==
d
4
and I is the ideal generated by ba − dc:
0
1 0 o_ _ _ _ _ _ _ _ 0 0 1 o_ _ _ _ _ _ _ _ 1 1 0
> 1 BB > 0 BB > 0 BB
||| BB ||| BB ||| BB
|| BB || BB || BB
| B B | BB | BB
|| || ||
0 0 1 1
0 0 o_ _ _ _ _ _ _ _ 1 1 o_ _ _ _ _ _ _ _ 1 1 o_ _ _ _ _ _ _ _ 0 0
1 1
F 00 BB 0 0
00 |> F 00 F
00 00 BBB |
|
| 00
00 00 BBB || 00
00 0 || 00
00 000 1 1 1 00
00 00 1 00
00 00 00
00 00 00
0 0 0
0 0 1
0 1 o_ _ _ _ _ _ _ _ 1 0 o_ _ _ _ _ _ _ _ 0 1
1 0 0
31.12. Knitting preprojective components. Let A be a finite-dimensional K-

algebra.
Basic idea: Let X be an indecomposable A-module. Whenever the sink map ending
in X is known, we can construct the source map starting in X. In Γ(A) = (ΓA , dA )
the situation around the vertex [X] looks like this:
[Y1 ] o_ _ _ _ _ _ _ _ [τA−1 (Y1 )]

++ H
++
++
++
.. +
..
. ++
.
++
++
++
[Yr ] o_ _ _+++ _ _ __ _ [τA−1 (Yr )]
BB :
BB +++ vvvv
BB + v
BB + vvv
! v
[X]
{= I . III
{ {{ ... IIII
{{ .. II
{{ ..
I$
[I1 ] .. [P1 ]
..
..
..
..
.. .. ...
. ..

..
.

[Is ] [Pt ]
Here the Yi are non-injective modules, the Ii are injective, and the Pi are projective.
The sink map ending in X is of the form Y → X where
r
M s
M
dY X /dYi dIi X /dIi
Y = Yi i ⊕ Ii .
i=1 i=1
To get the source map X → Z, we have to translate the non-injective modules Yi

using τA−1 . Note that
dXτ −1 (Yi ) = dYi X and dτ −1 (Yi ) = dYi

A A
for all i. Furthermore, we have to check if X occurs as a direct summand of rad(P )

where P runs through the set of indecomposable projective modules. In this case,
there is an arrow [X] → [P ] with valuation
dXP = dimK IrrA (X, P ) = rXP · dimK F (X).
We get
r
M t
M
dXτ −1 (Y ) /dτ −1 (Y dXPi /dPi
Z= τA−1 (Yi) A i A i) ⊕ Pi .
i=1 i=1
If X is non-injective, we get a mesh

[τA−1 (Y1 )]
H 1
111
11
11

.. 11
. 11
11
11
11
−1 11

[τA (Yr )] 11
v : LLL 11
v
vvv LLL
L 1
v LLL 11
vvv &
[X] Io_ _ _ _ _ _ _ _ _ _ _ [τA−1 (X)]
.. II r8 F
.. III
I r r rrr
.. II rr
.. I$ rrr
.. [P1 ]
..
..
..
..
.. ..
.. .
..
..

[Pt ]
in the Auslander-Reiten quiver Γ(A) of A. We have
dτ −1 (Yi )τ −1 (X) = dXτ −1 (Yi ) and dτ −1 (X) = dX .
A A A A
Knitting preparations
(i) Determine all indecomposable projectives P1 , . . . , Pn and all indecomposable

injectives I1 , . . . , In .
(ii) For each 1 ≤ i ≤ n determine rad(Pi ) and decompose it into indecomposable
modules, say
ri
M r
rad(Pi ) = Rijij
j=1
where rij ≥ 1, and the Rij are indecomposable such that Rik ∼
= Ril if and
only if k = l.
(iii) For each 1 ≤ i ≤ n determine dPi = dimK F (Pi ).
Note that
dRij Pi = dimK IrrA (Rij , Pi ) = rij · dRij
where rij = rRij Pi . Furthermore, we know that
= EndA (Pi / rad(Pi )) ∼
F (Pi ) = EndA (Pi )/ rad(EndA (Pi )) ∼ = EndA (Si )
where Si is the simple A-module with Si ∼
= Pi / rad(Pi ).
The knitting algorithm

Let −1 ∆ be the empty quiver.

We define inductively quivers n ∆, n ∆′ , n ∆′′ , n ≥ 0 which are subquivers of (ΓA , dA ).
For all n ≥ 1 these quivers will satisfy
n−1 ∆ ⊆ n ∆ ⊆ n−1 ∆′′ ⊆ n ∆′ ⊆ n ∆′′ .
By n ∆, n ∆′ , n ∆′′ , we denote the set of vertices of n ∆, n ∆′ , n ∆′′ , respectively.
(a0 ) Define 0 ∆: Let 0 ∆ be the quiver (without arrows) with vertices [S] where
S is simple projective.
(b0 ) Add projectives: For each [S] ∈ 0 ∆, if S ∼ = Rij for some i, j, then (if it
wasn’t added already) add the vertex [Pi ] with valuation dPi , and add an
arrow [S] → [Pi ] with valuation dSPi = rSPi · dS . We denote the resulting
quiver by 0 ∆′ .
(c0 ) Translate the non-injectives in 0 ∆: For each [S] ∈ 0 ∆ with S non-
injective, add the vertex [τA−1 (S)] to 0 ∆′ with valuation dτ −1 (S) = dS , and for
A
each arrow [S] → [Y ] constructed so far add an arrow [Y ] → [τA−1 (S)] to 0 ∆′
with valuation dY τ −1 (S) = dSY . We denote the resulting quiver by 0 ∆′′ .
A
Note that any source map starting in a simple projective module S is of the form
S → P where P is projective. (Proof: Assume there is an indecomposable non-
projective module X and an arrow [S] → [X]. Then there has to be an arrow
[τA (X)] → [S], a contradiction since [S] is a source in (ΓA , dA ).) Thus we get P
from the data collected in (i), (ii) and (iii). More precisely, we have
n
M dSPi /dPi
P = Pi ,
i=1
and we know that dSPi = rSPi · dS .

Now assume that for n ≥ 1 the quivers n−1 ∆, n−1 ∆′ and n−1 ∆′′ are already defined.
We also assume that for each vertex [X] ∈ n−1 ∆′′ and each arrow [X] → [Y ] in
′′
n−1 ∆ we defined valuations dX and dXY , respectively.
(an ) Define n ∆: Let n ∆ be the full subquiver of n−1 ∆′′ with vertices [X] such
that all direct predecessors of [X] in n−1 ∆′′ are contained in n−1 ∆, and if
[X] is a vertex with X ∼ = Pi projective, then we require additionally that
[Rij ] ∈ n−1 ∆ for all j.
(bn ) Add projectives: For each [X] ∈ n ∆, if X ∼ = Rij for some i, j, then (if
it wasn’t added already) add the vertex [Pi ] to n−1 ∆′′ with valuation dPi ,
and add an arrow [X] → [Pi ] to n−1 ∆′′ with valuation dXPi = rXP · dX . We
denote the resulting quiver by n ∆′ .
(cn ) Translate the non-injectives in n ∆\ n−1 ∆: For each [X] ∈ n ∆\ n−1 ∆ with
X non-injective, add the vertex [τA−1 (X)] to n ∆′ with valuation dτ −1 (X) = dX ,
A
and for each arrow [X] → [Y ] constructed to far add an arrow [Y ] → [τA−1 (X)]
to n ∆′ with valuation dY τ −1 (X) = dXY . We denote the resulting quiver by

A
′′
n∆ .
The algorithm stops if n ∆ \ n−1 ∆ is empty for some n. It can happen that the
algorithm never stops.
Define [ [
∞∆ = n∆ and ∞∆ = n ∆.
n≥0 n≥0
Let [X] ∈ n ∆, and let [X] → [Zi ], 1 ≤ i ≤ t be the arrows in n ∆′ starting in [X].
Then the corresponding homomorphism
t
M dXZi /dZi
X→ Zi
i=1
is a source map. Similarly, let [Yi ] → [X], 1 ≤ i ≤ s be the arrows in n ∆ ending in

[X]. Then the corresponding homomorphism
M s
dY X /dYi
Yi i →X
i=1
is a sink map. The following lemma is now easy to prove:
Lemma 31.44. For all n ≥ −1 we have
n∆ = n (ΓA ).
In particular, ∞∆ = ∞ (ΓA ).
Clearly, ∞ ∆ is a full subquiver of (ΓA , dA ). One easily checks that ∞ ∆ is a translation

subquiver of (ΓA , dA ) in the obvious sense.
The number of connected components of ∞∆ is bounded by the number of simple
projective A-modules.
If we know the dimension vectors dim(Pi ) and dim(Rij ) for all i, j, then our knitting
algorithm yields an algorithm to determine dim(X) for any vertex [X] ∈ ∞ ∆:
Let [X] be a vertex in n ∆ \ n−1 ∆, and let [X] → [Zi ], 1 ≤ i ≤ t be the arrows in
′
n ∆ starting in [X]. Then X is non-injective if and only if
t
X
l(X) < dXZi · l(Zi ).
i=1
In this case, we have
t
X
−1
dim(τ (X)) = −dim(X) + dXZi · dim(Zi).
i=1
These considerations provide a knitting algorithm which is only based on dimension

vectors. We will prove the following result:
Theorem 31.45. Let [X], [Y ] ∈ ∞ ∆. Then [X] = [Y ] if and only if dim(X) =

dim(Y ).
Lemma 31.46. Let C be a connected component of (ΓA , dA ). If
C ⊆ ∞ ∆,
then C is a preprojective component of (ΓA , dA ).
Proof. (a): By construction, for each [X] ∈ n ∆′′ we have τAn (X) is projective for
some n ≥ 0.
(b): The quiver n ∆ has no oriented cycles: One shows by induction on n that if
[X] → [Y ] is an arrow in n ∆, then there exists a unique t ≤ n such that [Y ] ∈
t ∆ \ t−1 ∆ and [X] ∈ t−1 ∆. The result follows.
(c): Let [X] ∈ n ∆. Then [X] has a direct predecessor in n ∆ if and only if X is not
in 0 ∆.
Often knitting does not work. For example, we cannot even start with the knitting
procedure, if there is no simple projective module. Furthermore, if an indecompos-
able projective module Pi is inserted such that an indecomposable direct summand
of rad(Pi ) does not show up in some step of the knitting prodedure, then we are
doomed and cannot continue.
But the good news is that in many interesting situations knitting does work. Here
are the two most important situations: Path algebras and directed algebras. In fact,
using covering theory, one can use knitting to construct the Auslander-Reiten quiver
of any representation-finite algebra (provided the characteristic of the ground field
is not two).
The dual situation: Obviously, there is also a “dual knitting algorithm” by starting
with the simple injective A-modules. As a knitting preparation one needs to de-
compose Ii / soc(Ii ) into a direct sum of indecomposables, and one needs the values
dIi = dimK F (Ii). If C is a component of Γ(A) which is obtained by the dual knitting
algorithm, then C is a preinjective component.
Lemma 31.47. Let Q be a finite connected quiver without oriented cycles. Then
the following hold:
(i) Γ(KQ) has a unique preprojective component P and a unique preinjective

component I;
(ii) P = I if and only if KQ is representation-finite.
Proof. Exercise.
31.13. More examples of Auslander-Reiten quivers. (a): Let Q be the quiver

2= 3 4
==
==
==

1
and let A = KQ. Using the dimension vector notation, ΓA looks as follows:
1 0 0 o_ _ _ _ _ _ _ _ 0 1 1 o_ _ _ _ _ _ _ _ 1 0 0
<x 1 FF < 1 FF x<
0
x x FF x xx FF xx
x FFF x FFF x
xx F" xx F" xx
xx xx xx
0 0 0 o_ _ _ _ _ _ _ _ 1 1 1 o_ _ _ _ _ _ _ _ 1 1 1
1 33FF < E 2 3F < E 1 3F
33 FFF x xx 33 FFF x xx 33 FFF
33 FFF x 33 FF x 33 FF
xx 33 FF" xx 33 FF"
33 " xx xx
3
33 0 1 0 o_ _ _ _ _ _ _33 _ 1 0 1 o_ _ _ _ _ _ _333 _ 0
1 0
33 1 33 1 33 0
33 33 33
33 33 33

0 0 1 o_ _ _ _ _ _ _ _ 1 1 0 o_ _ _ _ _ _ _ _ 0 0 1
1 1 0
Here is an interesting question: What happens with the Auslander-Reiten quiver of

KQ if we change the orientation of an arrow in Q?
For example, the path algebra of the quiver
2= 3 4
== @
==
==

1
has the following Auslander-Reiten quiver:
1 0 1 o_ _ _ _ _ _ _ _ 0 1 0 o_ _ _ _ _ _ _ _ 1 0 0
E13
3
1
E 333 E 0
3 3 3
33
3 33
3
0 1 1 o_ 3_3 _ _ _ _ _ _ 1 0 0 o_ 3_3 _ _ _ _ _ _ 0 1 0
1 F 3 1 F 3 0
xx< FF 333 xx< FF 333 xx<
x FFF 33 x FFF 33 xx
xxx x
FF 3 xxx FF 3 xxx
x x " x " x
0 0 1 o_ _ _ _ _ _ _ _ 1 1 1 o_ _ _ _ _ _ _ _ 1 1 0
< 1 FF < 2 FF < 1
xxx FF x xx FF x xx
x FF x FF x
xx FF xx FF xx
xx " xx " xx
0 0 1 o_ _ _ _ _ _ _ _ 0 0 0 o_ _ _ _ _ _ _ _ 1 1 1
0 1 1
(b): Let Q be the quiver

b
1o 2f a
and let A = KQ/I where I is generated by the path aa. Clearly, A is finite-
dimensional, and has two simple modules, whch we denote by 1 and 2. The
Auslander-Reiten quiver of A looks like this:
o_ _ _ _ _ _ _ _ _ 2
z= 2 GGG > 1 <<
zzz GG | || <<
G ||
z zz GG
G | <<
<<
zz G# ||
z
2 o_ _ _ _ _ _ _ _ _ 2 o_ _ _ _ _ _ _ 2
1 BB < 1 2?? A
BB yy ??
BB y y ??
BB yy ??
yy ?
2
1 2 o_ _ _ _ _ _ _ _ 2
2
? 1D
D @
D

DD
DD
D
DD
!
1 o_ _ _ _ _ _ _ _ _ _ 22
1
Note that this time, we did not display the dimension vectors of each indecomposable
module. Instead we used the composition factors 1 and 2 to indicate how the modules
look like. For example, the 4-dimensional A-module
2
1 2
1
has a simple top 2, its socle is isomorphic to 1 ⊕ 1. Note also that one has to identify
the two vertices on the upper left with the two vertices on the upper right. Thus ΓA
has in fact just 7 vertices. Sometimes one displays certain vertices more than once,
in order to obtain a nicer and easier to understand picture.
Clearly, ΓA does not contain a preprojective component. We have a simple projective
module, namely 1. So 0 ∆ = {1}. But then we see that 1 ∆ \ 0 ∆ = ∅. So there is just
one reachable vertex in ΓA .
We constructed ΓA “by hand”. In other words, our methods are not yet developed
enough to prove that this is really the Auslander-Reiten quiver of A.
(c): Let A be the path algebra of the quiver
3

2

1
Then there is an infinite preprojective component in (ΓA , dA ), which can be ob-

tained from the following picture by identifying the vertices in the first with the
corresponding vertices in the fourth row:
1
1 o_ _ _ _ _ _ _ _ 3 2 o_ _ _ _ _ _ _ _ 4 4 o_ _ _ _ _ _ _ · · ·
2 3 5 A
@ == @ == @ <<
=== === <
== == <<

= = <<
<<
0 2 3
1 o
_ _ _ _ _ _ _ _ 2 o
_ _ _ _ _ _ _ _ 4 o
_ _ _ _ _ _ _ _ ·@ · ·<
@ 1 == @ 3 == @ 4 ==
<<
== == ==
<<
<<
== == ==
== == == <<
<<
0
0 o_ _ _ _ _ _ _ _ 2 1 o_ _ _ _ _ _ _ _ 3 3 o_ _ _ _ _ _ _ _ 5 4 o_ _ _ _ _ _ _ · · ·
1 2 4 5 A
== @ == @ == @ <<
== === === <
== == == <<

== = = <<
<<
1
1 o_ _ _ _ _ _ _ _ 3 2 o_ _ _ _ _ _ _ _ 4 4 o_ _ _ _ _ _ _ _ · · ·
2 3 5
Exercise: Determine n ∆ for all n ≥ 0.

(d): Let

R C
A= ⊂ M2 (C).
0 R
Using the dimension vector notation, we obtain an infinite preprojective component

of (ΓA , dA ):
2 o_ _ _ _ _ _ _ 4 o_ _ _ _ _ _ _ 6 o_ _ _ _ _ _ _ · · ·
1 @ 1 << @ 3 << @ 5 >> >
<< 2 << 2 >> 2 ~~
2 << 2 << 2 >> 2 ~~
<< << >> ~~
> ~~~
1 o_ _ _ _ _ _ _ 3 o_ _ _ _ _ _ _ 5 o_ _ _ _ _ _ _ ···
1 0 2 4
(e): Let
 
R C C
A =  0 C C ⊂ M2 (C).
0 0 R
Again using the dimension vector notation we get an infinite preprojective compo-
nent:
2
1 1 o_ _ _ _ _ 22 o_ _ _ _ _ 43 o_ _ _ _ _ · · ·
1 B
2
C 77
772 2
C 1 777 2 2
C 3 777 2 2

77 777 777

2 4 6 8
2 1 o_ _ _ _ _ 3 o_ _ _ _ _ 5 o_ _ _ _ _ 7
0 7
2
C 77 C 2 777 2 C 4 777 2 C 6 999 2
772 2
777 2
777 2
99
7 99
1
1 0 o_ _ _ _ _ 11 o_ _ _ _ _ 32 o_ _ _ _ _ 33 o_ _ _ _ _ · · ·
0 0 2 2
(f): Let A = KQ/I where Q is the quiver

f
6o 7
e
g
5o 8
d
c b a
4 / 3 / 2 / 1
and the ideal I is generated by abcdef and cdg. It turns out that (ΓA , DA ) consists
of a single preprojective component:
10 11
10 10
111 110
|> BB
BB |> BB
BB
||| BB ||| BB
|| ||
00
10 o_ _ _ _ _ _ _ 11 0
0 o_ _ _ _ _ _ _ 11 10 o_ _ _ _ _ _ _ 01 01 o_ _ _ _ _ _ _ 10 0
0 o_ _ _ _ _ _ _ 00 1
0
111 > 1 10 > 10 0 > 00 0 > 0 00 > 0 00
|> BB BB BB BB BB
REPRESENTATION THEORY OF ALGEBRAS I: MODULES

|| BB ||| BB ||| BB ||| BB ||| BB |||
| BB | BB | BB | BB | BB |
|| || || || || ||
00
00 o_ _ _ _ _ _ _ 01 00 o_ _ _ _ _ _ _ 11 00 o_ _ _ _ _ _ _ 12 11 o_ _ _ _ _ _ _ 11 01 o_ _ _ _ _ _ _ 10 10
> 111BB > 1 1 0 BB > 1 0 0 BB > 1 0 0 BB > 0 0 0 BB > 00 0
||| BB ||| BB ||| BB ||| BB ||| BB |||
| BB | BB | BB | BB | BB |
|| || || || || ||
00 00 00 10 21 11
00 o_ _ _ _ _ _ _ 0 0 o_ _ _ _ _ _ _ 1 0 o_ _ _ _ _ _ _ 2 1 o_ _ _ _ _ _ _ 2 1 o_ _ _ _ _ _ _ 1 1
>| 0 1 1 BB > 1 1 0 BB > 1 0 000 BB > 1F 0 000 BB > 1F 0 000 BB > 0F 0 0 BB
|| BB ||| BB ||| 0 BB ||| 0 BB ||| 0 BB ||| BB
| BB | BB | 00 B B || 00 B B || 00 B B || BB
|| || || 0 | 0 | 0 |
00
00 o_ _ _ _ _ _ _ 00 00 o_ _ _ _ _ _ _ 00 00 o_ _ _ _ _ _000_ 01 00 o_ _ _ _ _ _000_ 11 01 o_ _ _ _ _ _000_ 11 10 o_ _ _ _ _ _ _ 00 01
001 010 100 00 0 0 0 00 1 0 0 00 0 0 0 00 0
00 0 0
0 00

00

00
11 o_ _ _ _ _ _ _ 11 00 o_ _ _ _ _ _ _ 11 11
100 000 100
237
(g): Let A = KQ/I where Q is the quiver
3
a

2
b c

1
and I is the ideal generated by ba. The indecomposable projective A-modules are
3
of the form P1 = 1 , P2 = 1 2 1 , P3 = 2 . Then ∞ ∆ consists of a preprojective
1
component
0
1 o_ _ _ _ _ _ _ 03 o_ _ _ _ _ _ _ 05 o_ _ _ _ _ _ _ · · ·
2 B 4 A

B 88
88
88
88 B 6 :::
2 2 2 2 :::2
88 882
88 88 ::
::
0
0 o_ _ _ _ _ _ _ 02 o_ _ _ _ _ _ _ 04 o_ _ _ _ _ _ _ · · ·
1 3 5
which does not contain P3 .
(h): Let A = KQ/I where Q is the quiver
3
a

2 d
b c

1
and I is the ideal generated by ba. The indecomposable projective A-modules are of
3
the form P1 = 1 , P2 = 1 2 1 , P3 = 2 1 . Then ∞ ∆ consists of two points, namely
1
P1 and P2 :
0 4
1 o_ _ _ _ _ _ _ 7
B 2 88 B 12
88
2 2
88 2
88

0 2
0 4
1 88 B 7
88
882 2
88

1
1
2
Note that one of the direct summands of the radical of P3 does not show up in the
course of the knitting algorithm. So we get 2 ∆ \ 1 ∆ = ∅.
(i): Let A = KQ/I where Q is the quiver
3=
==
a ==d
==

2 4
b c

1
and I is the ideal generated by ba. The indecomposable projective A-modules are
3
of the form P1 = 1 , P2 = 1 2 1 , P3 = 2 4 , P4 = 4 . Then ∞ ∆ has two connected
1
components, one is an (infinite) preprojective component, and the other one consists
just of the vertex P4 :
0
1 0 o_ _ _ _ _ _ _ _ 03 0 o_ _ _ _ _ _ _ _ 05 0 o_ _ _ _ _ _ _ · · ·
2 @ 4 @ 6 A
@ ==
== 2
==
== 2
<<
2 == 2 == 2 << 2 2
<<
== == <<
<
0
0 0 o_ _ _ _ _ _ _ _ 02 0 o_ _ _ _ _ _ _ _ 04 0 o_ _ _ _ _ _ _ _ ···
1 3 5
0 1
0 1 o_ _ _ _ _ _ _ _ 1 0
0 1
==
== @
==
==

1
1 1
1
(j): Let A = KQ/I where Q is the quiver
3
b a

2
c

1
and I is the ideal generated by ca and cb. The indecomposable projective A-modules
are of the form P1 = 1 , P2 = 1 2 1 , P3 = 2 3 2 . Then ∞ ∆ consists of an infinite
preprojective component containing an injective module:
0
1
B 1 88
88
88
88

0 0
0 o_ _ _ _ _ _ _ 1 o_ _ _ _ _ _ _ 23 o_ _ _ _ _ _ _ 45 o_ _ _ _ _ _ _ · · ·
1 0 88 B 0 88 B 0 88 A <<<
88 88 88 <<
882 2 882 2 882 2
<<2
88 88 88 <<
<<

1
2 o_ _ _ _ _ _ _ 34 o_ _ _ _ 5
_ _ _ 6 o_ _ _ _ _ _ _ · · ·
0 0 0
(l): Let A = K[T ]/(T 4). There is just one simple A-modules S, and all indecom-
posable A-modules are uniserial. The Auslander-Reiten quiver looks like this:
S
S
S
S
O

S W

S hg
S
O
W
S
S
O hg
W
S hg
The only indecomposable projective A-module has length 4. For the other three
indecomposables we have τA (X) ∼
= X. For example, the obvious sequence of the
form
S S S
0→ S → S ⊕ S → S →0
S
is an Auslander-Reiten sequence.
(m): Let Q be the quiver
4=
==
==a
==

2 == 3
==
==

= b
1
and set A = KQ/(ba).

Using the socle series notation the Auslander-Reiten quiver of A looks as follows:
3 o_ _ _ _ _ _ _ _ _ _ _ 2 o_ _ _ _ _ _ _ _ _ _ _ 4
? 1 FF tt: JJ
JJ x< 3 ??
FF t

FF tt
tt JJ
JJ xx ??
F tt JJ xxx ??
F" tt J% xx ??

1 o_ _ _ _ _ _ _ _ 2 3 o_ _ _ _ _ _ _ _ _ _ _ 4 o_ _ _ _ _ _ _ _ 4
== 1 3 2 @
== zz= HH
HH v vv: DD
D
== z HH v D DD
== zz HH vv
= zzz H$ vvv DD
D
z !
2 o_ _ _ _ _ _ _ _ _ 3 4 2 3 o_ _ _ _ _ _ _ _ _ 4
1 B 1 G > 2
BB
BB ww; GG |||
BB ww GG ||
BB ww GG ||
ww G# ||
4 4
2 3 o_ _ _ _ _ _ _ _ _ _ 2 3
>1 GG ; 1 B
||| GG www BB
BB
|| GG w w BB
| | GG w w BB
| G# ww B
| |
4
3 o_ _ _ _ _ _ _ _ _ _ 2 o_ _ _ _ _ _ _ _ _ _ 3
1
(n): Let Q be the quiver

a / b /
1o c
2o 3
d
and let A = KQ/I where I is generated by ba, cd, ac − db. The (ΓA , dA ) looks as
follows (one has to identify the three modules on the left with the three modules on
the right):
3
2
? 3 ??
??
??
??

1 o_ _ _ _ _ _ _ _ _ 2 o_ _ _ _ _ _ _ _ 3 o_ _ _ _ _ _ _ _ _ 1
DD > 3 ? ?2 BB =
DD || ?? BB zz
DD | ?? BB zzz
DD || ?? BB z
DD || ?? B zz
! || zz
2 2
1 3 / 2 o_ _ _ _ _ _ _ _ 2 o_ _ _ _ _ _ _ _ 1 3 / 1 3
2 = 1 3 BB ? ?? > 2 DD 2
zzz BB ?? || DDD
zz BB ?? || DD
zz BB ?? ||| DD
zz B ? || D!
z
3 o_ _ _ _ _ _ _ _ _ 2 o_ _ _ _ _ _ _ _ 1 o_ _ _ _ _ _ _ _ _ 3
1 ? ? 2
??
??
??
?
1
2
1
Note that A is a selfinjective algebra, i.e. an A-module is projective if and only if it

is injective.
(o): Let Q be the quiver
a b c
1o 2 / 3 / 4 / 5o 6
and let A = KQ/I where I is generated by cba. Then (ΓA , dA ) looks as follows:
6 o_ _ _ _ _ _ _ 4 o_ _ _ _ _ _ _ 3 Ho_ _ _ _ _ _ _ 1 2
? 5 DD v; FFF xx
; HH z< ??
DD v vv FF xx HH
HH zz ??
D" v FF x z ??
vv # x x $ z
5 o_ _ _ _ _ _ 4 5 6 o_ _ _ _ _ _ _ 3 4 o_ _ _ _ _ _ _ 1 2 3 o_ _ _ _ _ _ 2
== = F < DD ;x BB @
==
| || FFF
zzz DD xx BB
== | F F# D" x B
|| zz xx B!
3 2
4
5
o
_ _ _ _ _ _ 4 6 1 3 o_ _ _ _ _ _ 2 3
@@ < 5 BB 4 G
@@
@@ z zz BB F 222
BB 22
zz
3
B!
22
4 o
_ _ _ _ _ _ _ 6 22
5
22
22

2
1 o_ _ _ _ _ _ _ 3
4
Next, we display the Auslander-Reiten quiver of KQ:

6 o_ _ _ _ _ _ _ 4 o_ _ _ _ _ _ _ _ _ _ 3 o_ _ _ _ _ _ _ 2
? 5 DD u uu: MMMM
M qq qq8 IIII z< 1 ??
DD uu MMM qq II zz ??
D" uu MM& qqq I$ zz ??
q
5 o_ _ _ _ _ _ 4 5 6 o_ _ _ _ _ _ _ _ _ 3 4 o_ _ _ _ _ _ _ _ _ 1 2 3 o_ _ _ _ _ _ 2
== 9 K @
== ||= HH
HHH ssss KKK
K vvv
; BB
BB
== | s KKK% v BB
||
| H# sss vv !
3 2
4
5
o
_ _ _ _ _ _ 4 6 o
_ _ _ _ _ _ _ _ _ 1 3 o
_ _ _ _ _ _ 2
>> < 5 HH : 4
DD @ 3
>> zz HH v vv D
>> zz HH v DDD
zz $ vv "
3 2
4 o_ _ _ _ _ _ _ 1 3 4 6 o_ _ _ _ _ _ _ 2 3
5 4
BB
BB ww ; 5
GGG ||=
BB w GGG ||
! www # ||
2 2
1 o_ _ _ _ _ _ _ _ 3 4 6
3
4
5 5
=| GG ; BB
|| GG www BB
| GG# w BB
||| ww BB
| 2 !
1 o_ _ _ _ _ _ _ _ _ 3 4 o_ _ _ _ _ _ _ _ _ 6
5
————————————————————————————-
32. Grothendieck group and Ringel form
32.1. Grothendieck group. As before, let A be a finite-dimensional K-algebra,

and let S1 , . . . , Sn be a complete set of representatives of isomorphism classes of the
simple A-modules. For a finite-dimensional module M let
dim(M) := ([M : S1 ], . . . , [M : Sn ])
be its dimension vector. Here [M : Si ] is the Jordan-Hölder multiplicity of Si in M.
Note that dim(M) ∈ Nn0 ⊂ Zn . Set ei := dim(Si ). Then
G(A) := K0 (A) := Zn
is the Grothendieck group of mod(A), and e1 , . . . , en is a free generating set of

the abelian group G(A).
We can see dim as a map

dim : {A-modules}/ ∼
= −→ G(A)
which associates to each modules M, or more precisely to each isomorphism class
[M], the dimension vector dim(M).
Note that n
X
[M : Si ] = l(X).
i=1
Furthermore, dim is additive on short exact sequences , i.e. if 0 → X → Y →
Z → 0 is a short exact sequence, then dim(Y ) = dim(X) + dim(Z).
Lemma 32.1. If
f : {A-modules}/ ∼
= −→ H
is a map which is additive on short exact sequences and H is an abelian group, then
there exists a unique group homomorphism f ′ : G(A) → H such that the diagram
dim
{A-modules}/ ∼
= / G(A)
j j j
j
f
j j j′
uj j j f
H
commutes.
Proof. Define a group homomorphism f ′ : G(A) → H by f ′ (ei ) := f (Si) for 1 ≤ i ≤

n. We have to show that f ′ (dim(M)) = f (M) for all finite-dimensional A-modules
M. We proof this by induction on the length l(M) of M. If l(M) = 1, then M ∼ = Si
and we are done, since f ′ (dim(M)) = f ′ (ei ) = f (Si ).
Next, assume l(M) > 1. Then there exists a submodule U of M such that U 6= 0 6=
M/U. We obtain a short exact sequence
0 → U → M → M/U → 0.
Clearly, l(U) < l(M) and l(M/U) < l(M). Thus by induction f ′ (dim(U)) = f (U)
and f ′ (dim(M/U)) = f (M/U). Since f is additive on short exact sequences, we get
f (M) = f (U) + f (M/U) = f ′ (dim(U)) + f ′ (dim(M/U)) = f ′ (dim(M)).
It is obvious that f ′ is unique. This finishes the proof.
Here is an alternative construction of G(A): Let F (A) be the free abelian group
with generators the isomorphism classes of finite-dimensional A-modules. Let U(A)
be the subgroup of F (A) which is generated by the elements of the form
[X] − [Y ] + [Z]
if there is a short exact sequence 0 → X → Y → Z → 0. Define
G(A) := F (A)/U(A).
For an A-module M set [M] := [M] + U(A). It follows that G(A) is isomorphic to
Zn with generators [Si ], 1 ≤ i ≤ n. By induction on l(M) one shows that
n
X
[M] = [M : Si ] · [Si ].
i=1
32.2. The Ringel form. We assume now that A is a finite-dimensional K-algebra

with gl. dim(A) = d < ∞. In other words, we assume Extd+1A (X, Y ) = 0 for all
A-modules X and Y and d is minimal with this property.
Define
d
X
hX, Y iA := (−1)t dim ExttA (X, Y ).
t=0
(If gl. dim(A) = ∞,
P but proj. dim(X) < ∞ or inj. dim(Y ) < ∞, then we can still
define hX, Y iA := t≥0 (−1)t dim ExttA (X, Y ).)
Recall that Ext0A (X, Y ) = HomA (X, Y ). We know that for each short exact sequence
0 → X ′ → X → X ′′ → 0
and an A-module Y we get a long exact sequence
0 / Ext0A (X ′′ , Y ) / Ext0A (X, Y ) g
/ Ext0A (X ′ , Y )
gg
ggggg
ggg gggggg
gg
sggggg
Ext1A (X ′′ , Y ) / Ext1 (X, Y )
A
/ Ext1A (X ′ , Y )
gggggg
ggggg
gg ggggggg
sgggg
Ext2A (X ′′ , Y ) / Ext2 (X, Y )
A
/ Ext2A (X ′ , Y )
gggggg
ggggg
gg ggggggg
sgggg
Ext3A (X ′′ , Y ) / ···
Now one easily checks that this implies

d
X d
X
t
(−1) dim ExttA (X ′′ , Y )− (−1)t dim ExttA (X, Y )
t=0 t=0
d
X
+ (−1)t dim ExttA (X ′ , Y ) = 0.
t=0
In other words,
hX ′′ , Y iA − hX, Y iA + hX, Y iA = 0.
It follows that
h−, Y iA : {A-modules}/ ∼
=→Z
is a map which is additive (on short exact sequences). Thus hdim(X), Y iA :=
hX, Y iA is well defined.
Similarly, we get that

hX, Y ′ iA − hX, Y iA + hX, Y ′′ iA = 0.
if 0 → Y ′ → Y → Y ′′ → 0 is a short exact sequence.
Thus hdim(M), dim(N)iA := hM, NiA is well defined, and we obtain a bilinear map
h−, −iA : G(A) × G(A) → Z.
This map is determined by the values
d
X
hei , ej iA = (−1)t dim ExttA (Si , Sj )
t=0
Pn
since dim(M) = i=1 [M : Si ]ei .
————————————————————————————-
33. Reachable and directing modules
Let K be a field, and let A be a finite-dimensional K-algebra. By M = M(A) we

denote the category mod(A) of all finite-dimensional A-modules.
33.1. Reachable modules. A path of length n ≥ 0 in M is a finite sequence

([X0 ], [X1 ], . . . , [Xn ]) of isomorphism classes of indecomposable A-modules Xi such
that for all 1 ≤ i ≤ n there exists a homomorphism Xi−1 → Xi which is non-zero
and not an isomorphism, in other words we assume radA (Xi−1 , Xi ) 6= 0. We say
that such a path ([X0 ], [X1 ], . . . , [Xn ]) starts in X0 and ends in Xn . If n ≥ 1 and
[X0 ] = [Xn ], then ([X0 ], [X1 ], . . . , [Xn ]) is a cycle in M. In this case, we say that
the modules X0 , . . . , Xn−1 lie on a cycle.
If X and Y are indecomposable A-modules, we write X Y if there exists a path
from X to Y , and we write X ≺ Y if there is such a path of length n ≥ 1.
An indecomposable module X in M is reachable if there are only finitely many

paths in M which end in X. Let
E(A)
be the subcategory of reachable modules in M.
Furthermore, we call X directing if X does not lie on a cycle, or equivalently, if
X 6≺ X.
The following two statements are obvious:
Lemma 33.1. Every reachable module is directing.
Lemma 33.2. If X is a directing module, then rad(EndA (X)) = 0.
Examples: (a): Let A = K[T ]/(T m ) for some m ≥ 2. Then none of the indecom-
posable A-modules is directing.
(b): If A is the path algebra of a quiver of type A2 , then each indecomposable

A-module is directing.
Let Γ(A) = (ΓA , dA ) be the Auslander-Reiten quiver of A. If Y is a reachable A-
module, and [X] is a predecessor of [Y ] in Γ(A), then by definition there exists a
path from [X] to [Y ] in ΓA . Thus, we also get a path from X to Y in M. This
implies that X is a reachable module as well. In particular, if Z is a reachable non-
projective module, then τA (Z) is reachable. So the Auslander-Reiten translation
maps the set of isomorphism classes of reachable modules into itself.
We define classes
∅ = −1 M ⊆ 0 M ⊆ · · · ⊆ n−1 M ⊆ n M ⊆ · · ·
of indecomposable modules as follows: Set −1 M = ∅. Let n ≥ 0 and assume
that n−1 M is already defined. Then let n M be the subcategory of all indecompos-
able modules M in M with the following property: If N is indecomposable with
radA (N, M) 6= 0, then N ∈ n−1 M.
Let [
∞M = nM
n≥0
be the full subcategory of M containing all M ∈ n M, n ≥ 0.
Then the following hold:
(a) n−1 M ⊆ n M (Proof by induction on n ≥ 0);

(b) 0 M is the class of simple projective modules;
(c) 1 M contains additionally all indecomposable projective modules P such that
rad(P ) is semisimple and projective;
(d) 2 M can contain non-projective modules (e.g. if A is the path algebra of a
quiver of type A2 );
(e) n M is closed under indecomposable submodules;
(f) If g : Y → Z is a sink map, and
t
M
Y = Yi
i=1
a direct sum decomposition with Yi indecomposable and Yi ∈ n−1 M for
all i, then Z ∈ n M; (Proof: Let N be indecomposable, and let 0 6= h ∈
radA (N, Z). Then there exists some h′ : N → Y with h = g ◦ h′ .
N
~
~ h
~ h′
~~ g
Y / Z
Thus we can find some 0 6= h′i : N → Yi. If h′i is an isomorphism, then
N∼= Yi ∈ n−1 M. If h′i is not an isomorphism, then N ∈ n−2 M ⊆ n−1 M.)
(g) If Z ∈ n M is non-projective, then τA (Z) ∈ n−2 M;

(h) We have
E(A) = ∞ M.
Lemma 33.3. Let A be a finite-dimensional K-algebra. If Z is an indecomposable
A-module, then Z ∈ n M if and only if [Z] ∈ n (ΓA ).
Proof. The staatement is correct for n = −1. Thus assume n ≥ 0. If Z ∈ n M and

t
M
Yi → Z
i=1
is a sink map with Yi indecomposable for all i, then Yi ∈ n−1 M for all i. Thus
by induction assumption [Yi ] ∈ n−1 (ΓA ), and therefore [Z] ∈ n (ΓA ). Vice versa, if
[Z] ∈ n (ΓA ), then [Yi ] ∈ n−1 (ΓA ). Thus Yi ∈ n−1 M. Using (f) we get Z ∈ n M.
Let
E(A)
be the full subquiver of all vertices [X] of ΓA such that X is a reachable module.
One easily checks that E(A) is again a valued translation quiver.
Summarizing our results and notation, we obtain
E(A) = ∞ (ΓA ) = ∞ ∆, and E(A) = ∞ M.
Furthermore, E(A) is the full subcategory of all A-modules X such that [X] ∈ E(A).
We say that K is a splitting field for A if EndA (S) ∼
= K for all simple A-modules
S.
Examples: If K is algebraically closed, then K is a splitting field for K. Also, if
A = KQ is a finite-dimensional path algebra, then K is a splitting field for A.
Roughly speaking, if K is a splitting field for A, then there are more combinatorial
tools available, which help to understand (parts of) mod(A). The most common
tools are mesh categories and integral quadratic forms.
Theorem 33.4. Let A be a finite-dimensional K-algebra, and assume that K is a
splitting field for A. Then the valuation for E(A) splits, and there is an equivalence
of categories
η : KhE(A)e i → E(A).
Proof. Let I be a complete set of indecomposable A-modules (thus we take exactly

one module from each isomorphism class). Set
nI = I ∩ nM and ∞I = I ∩ E(A).
For X, Y ∈ ∞ I we want to construct homomorphisms

aiXY ∈ HomA (X, Y )
with 1 ≤ i ≤ dXY := dimK IrrA (X, Y ).
If Y = P is projective, we choose a direct decomposition

M
rad(P ) = X dXP .
X∈I
We know that dXP = dimK IrrA (X, P ). Let

aiXP : X → P
with 1 ≤ i ≤ dXP be the inclusion maps.
By induction we assume that for all X, Y ∈ n I we have chosen homomorphisms
aiXY : X → Y where 1 ≤ i ≤ dXY .
Let Z ∈ n+1 I be non-projective, and let
f
M g
0→X − → Y dXY −
→Z→0
Y ∈n I
be the Auslander-Reiten sequence ending in Z, where the dXY component maps

X → Y of f are given by aiXY , 1 ≤ i ≤ dXY . Now g together with the direct sum
decomposition M
Y dXY
Y ∈n I
yields homomorphisms aiY Z

: Y → Z, 1 ≤ i ≤ dXY = dY Z . These homomorphisms
obviously satisfy the equation
X dX
XY
aiY Z aiXY = 0.
Y ∈n I i=1
Denote the corresponding arrows from [X] to [Y ] in

Γ := E(A)e
i
by αXY where 1 ≤ i ≤ dXY .
We obtain a functor
η : KhΓi → E(A)
as follows: For X ∈ ∞ I define
i

η([X]) := X and η αXY := aiXY .
This yields a functor KhΓi → E(A), since by the equation above the mesh relations
are mapped to 0.
Now we will show that η is bijective on the homomorphism spaces.
Before we start, note that EndA (X) ∼
= K for all X ∈ E(A). (Proof: A reachable
module X does not lie on a cycle in M(A), thus rad(EndA (X)) = 0. This shows
that F (X) ∼
= EndA (X). Let X ∈ ∞ M = E(A). If X = P is projective, then
F (X) ∼= EndA (P/ rad(P )) ∼
= EndA (S) ∼
=K
where S is the simple A-module isomorphic to P/ rad(P ). Here we used that K is
a splitting field for A. If X is non-projective, then F (X) ∼
= F (τA (X)). Furthermore
we know that τAn (X) is projective for some n ≥ 1. Thus by induction we get
F (X) ∼
= EndA (X) ∼= K.)
Surjectivity of η: Let h : M → Z be a homomorphism in ∞ I, and let Z ∈ n I. We
use induction on n. If M = Z, then h = c · 1M for some c ∈ K. Thus h = η(c · 1[M ] ).
Assume now that M 6= Z. This implies that h is not an isomorphism. The sink
map ending in Z is M
g = (aiY Z )Y,i : Y dY Z → Z.
Y ∈n−1 I
We get X
h= aiY Z hY,i.
Y,i
By induction the homomorphisms hY,i : M → Y are in the image of η, and by the
construction of η also the homomorphisms aiY Z are contained in the image of η.
Thus h lies in the image of η
Injectivity of η: Let R be the mesh ideal in the path category KΓ. We investigate
the kernel K of
η : KΓ → ∞ I.
Clearly, R ⊆ K. Next, let ω ∈ K. Thus ω ∈ HomKΓ([M], [Z]) for some [M] and
[Z]. We have to show that ω ∈ R. Assume [Z] ∈ n I. We use induction on n.
Additionally, we can assume that ω 6= 0. Thus there exists a path from [M] to [Z].
If [M] = [Z], then ω = c · 1[M ] and η(ω) = c · 1M = 0. This implies c = 0 and
therefore ω = 0.
Thus we assume that [M] 6= [Z]. Now ω is a linear combination of paths from [M]
to [Z], i.e. ω is of the form X
ω= αYi Z ωY,i
Y,i
where the ωY,i are elements in HomKΓ([M], [Y ]). Note that [Y ] ∈ n−1 I. Applying η
we obtain X
0 = η(ω) = aiY Z η(ωY,i).
Y,i
If Z is projective, then each aiY Z : Y → Z is an inclusion map, and we have

Im(aiY11 Z ) ∩ Im(aiY22 ,Z ) 6= 0
if and only if Y1 = Y2 and i1 = i2 . This implies aiY Z η(ωY,i) = 0 for all Y, i. Since aiY Z
is injective, we get η(ωY,i) = 0. Thus by induction ωY,i ∈ R and therefore ω ∈ R.
Thus assume Z is not projective. Then we know the kernel of the map
M
g = (aiY Z )Y,i : Y dY Z → Z
Y ∈n−1 I
namely M
f = (aiXY )Y,i : X → Y dY Z .
Y ∈n−1 I
Thus the map

M
h := (η(ωY,i))Y,i : M → Y dY Z
Y ∈n−1 I
factorizes through f , since g ◦ h = 0. So we obtain a homomorphism h′ : M → X

such that
(aiXY )Y,i ◦ h′ = (η(ωY,i))Y,i
and therefore aiXY ◦ h′ = η(ωY,i).
By the surjectivity of η there exists some ω ′ : [M] → [X] such that η(ω ′) = h′ . Thus
we see that
i

η αXY ω ′ = aiXY ◦ h′ = η(ωY,i).
i
In other words, η (ωY,i − αXY ω ′) = 0. By induction ωY,i − αXY
i
ω ′ belongs to the
mesh ideal. Thus also
X
ω= αYi Z ωY,i
Y,i
X X i ′
= i
αYi Z ωY,i − αXY ω′ + i
αY Z αXY ω
Y,i Y,i
is contained in the mesh ideal. This finishes the proof.
33.2. Computations in the mesh category. Let M and X be non-isomorphic

indecomposable A-modules such that X is non-projective. Let 0 → τA (X) → E →
X → 0 be the Auslander-Reiten sequence ending in X. Then
0 → HomA (M, τA (X)) → HomA (M, E) → HomA (M, X) → 0
is exact.
Let Γ = (ΓA , dA ). If [X] and [Z] are vertices in E(A) such that none of the paths
in Γ starting in [X] and ending in [Z] contains a subpath of the form [Y ] → [E] →
[τA−1 (Y )], then we have
HomKhE(A)e i ([X], [Z]) = HomKΓ ([X], [Z]).
Using this and the considerations above, we can now calculate dimensions of homo-
morphism spaces using in the mesh category KhE(A)e i.
Let Q be the quiver

2o 5

1O o 3
4o 6
and let A = KQ. Here is the Auslander-Reiten quiver of A, using the dimension
vector notation:
11 00 10 11 00 00
10 11 10 11 00 00
00 10 11 00 10 01
@ == @ == @ == @ == @ == @
== == == == ==

10 11 10 21 11 0 0
10 21 21 21 11 0 0
00 10 21 11 10 1 1
@ == @ == @ == @ == @ == @
== == == == ==

00 00 10 10 21 11 21 10 21 11 11 0 0
10 / 11 / 21 / 10 / 31 / 21 / 32 / 11 / 21 / 10 / 11 / 0 1
00 00 10 10 21 11 21 10 21 11 11 0 0
== @ == @ == @ == @ == @ ==
== == == == == ==

00 10 21 11 10 1 1
10 21 21 21 11 0 0
10 11 10 21 11 0 0
== @ == @ == @ == @ == @ ==
== == == == == ==

00 10 11 00 10 01
10 11 10 11 00 00
11 00 10 11 00 00
Here we display the locations of the indecomposable projective and the indecom-
posable injective A-modules:
P5 > ◦ ◦ ◦ ◦ I6
> >> A ;;; A ;;; A ;;; @ ???? ?
}} >> ;; ;; ;; ?
}}
2 P A ? ◦ ;; A◦; A◦; A◦= I4
}}> AA ;; ;;; ;;; ==== ?
}} AA ; ; ;
P1 /P /◦ /◦ /◦ /◦ /◦ /◦ /◦ /◦ /I
1 ? I3
/
3
AA }}> >>
>> A ;;
;; A ;;
;; A ;;
;; @ ?
AA ??
}}
} > ; ; ;
P4 A ? ◦ ;; A ◦ ;; A ◦ ;; A ◦ == I2
AA ;; ;; ;; == ? ??
??
A ; ; ; =
P6 ◦ ◦ ◦ ◦ I5
The following pictures show how to compute dim HomA (Pi , −) for all indecom-
posable projective A-modules Pi . Note that the cases P2 and P4 , and also P5
and P6 are dual to each other. We marked the vertices [Z] by a where a =
dim HomA (Pi , Z), provided none of the paths in E(A) starting in [Pi ] and ending
in [Z] contains a subpath of the form [Y ] → [E] → [τA−1 (Y )]. Of course, we can
compute dim HomA (X, −) for any indecomposable A-module.
dim HomA (P1 , −):
1 1 1 1 0 0
@ ;; B 888 B 888 B 888 B 888 B
;;
;;
88
88
88
88
88
88
88
88

1 2 2 2 1 0
@ >> A 888 B 888 B 888 B 888 B
>> 8 8 8 8
>> 88 88 88 88
> 8 8 8 8
1 / 1 /2 /1 /3 /2 /3 /1 /2 /1 /1 /0
>> @ ;;
;; B 88
8 B 88
8 B 88
8 B 88
8
>>

;; 88 88 88 88
>>
; 88 88 88 88

1 2
A 8 2
B 8 2
B 8 1
B 8 B08
>> 888 888 888 888 888
>> 88 88 88 88 88
>>
8 8 8 8 8
1 1 1 1 0 0
1 0 1 1 0 0
@ >> @ ;;; B 888 B 888 B 888 B
>> ;; 88 88 88

>>
> ;; 88 88 88

1 1 A 1 88 B 2 88 B 1 88 0
A >> @ >>
>> 8 8 8 B
>>

>> 88 88 88
>>
>> 88 88 88

0; /0 / 1 / 1 /2 /1 /2 /1 /2 /1 /1 /0
@ ;; B 88 B 88 B 88
;; @ >>
>>
;; 88 88 88
;;
;; >> ;; 88 88 88
; 8 8 8

0 >> 1 2 1 1 1
>> @ >> A 888 B 888 B 888 B 888
>> 88 88 88 88
>> >> 88

88

88

88
>>

0 1 1 0 1 0
@0 >> @
1 B08 B18 B08 B0
>>
;;
;; 888 888 888
>> 88 88 88
>
;; 8

8

8

A0 >> @
1
>> A18 B18 B18 B0
>>
>> >> 888 888 888
88 88 88
>
>> 8

8

8

>
0 / 1 / 1 / 0 /1 /1 /2 /1 /1 /0 /1 /1
;;
;; @ >> @ ;;
;; B 888 B 888 B 888
>> 88 88 88
;;
>> ;; 88

88

88
; ;
0 >> 1 A 1 88 B 1 88 B 1 88 B08
>> @ >>
>> 88 88 88 888
>> >> 88 88 88 88
>>

8 8 8 8
0 1 0 1 0 0

1 @0> @0 ;; B18 B08 B0

A >>
>> >>> ;; 888 888
>> >> ;; 88 88
>> ; 8 8

B0 ;; 1 @0 >> A18 B18 B0
;; @ >>
>> >> 888 888
;; >>

88 88
; >>
>> 8

8

08 /0 /0
>>
/0 / 1 / 1 /
@ 1
/0 /1 /1 /1 /0
88 A >> @ >> ;; B 888 B 888
88

>> ;; 88 88
88
>>
>> ;; 88

88
>>
;
0 ;; @ 0 >> 1 A 1 88 B 0 88 B 1 88
>> @ >>
;; >>
>>

88

88

88
;; >> >> 88 88 88
; 8 8 8
0 0 1 0 0 1
33.3. Directing modules.

Lemma 33.5. Let X be a directing A-module, then EndA (X) is a skew-field, and
we have ExtiA (X, X) = 0 for all i ≥ 1.
Proof. Since rad(EndA (X)) = 0, we know that EndA (X) is a skew-field. It is also
clear that Ext1A (X, X) = 0: If 0 → X → M → X → 0 is a short exact sequence
which does not split, then we immediately get a cycle (X, Mi , X) where Mi is an
indecomposable direct summand of M.
Let C be the class of indecomposable A-modules M with M X. We will show by
induction that ExtjA (M, X) = 0 for all M ∈ C and all j ≥ 1:
The statement is clear for j = 1. Namely, if Ext1A (M, X) 6= 0, then any non-split
short exact sequence M
0→X→ Yi → M → 0
i
yields X ≺ M X, a contradiction.
Next, assume j > 1. Without loss of generality assume M is not projective. Let
ε
0 → Ω(M) → P0 − → M → 0 be a short exact sequence where ε : P0 → M is a
projective cover of M. We get
Extj (M, X) ∼ j−1
= Ext (Ω(M), X). A A
IfExtjA (M, X) 6= 0, then there exists an indecomposable direct summand M ′ of

j−1
Ω(M) such that ExtA (M ′ , X) 6= 0. But for some indecomposable direct summand
P of P0 we have M P ≺ M X, and therefore M ′ ∈ C. This is a contradiction
′
to our induction assumption.

Corollary 33.6. Assume gl. dim(A) < ∞, and let X be a directing A-module. Then
the following hold:
(i) χA (X) = hX, XiA = dimK EndA (X);

(ii) If K is algebraically closed, then χA (X) = 1;

(iii) If K is a splitting field for A, and if X is preprojective or preinjective, then
χA (X) = 1.
As before, let A be a finite-dimensional K-algebra. An A-module M is sincere if

each simple A-module occurs as a composition factor of M.
We call the algebra A sincere if there exists an indecomposable sincere A-module.
Lemma 33.7. For an A-module M the following are equivalent:
(i) M is sincere;
(ii) For each simple A-module S we have [M : S] 6= 0;
(iii) If e is a non-zero idempotent in A, then eM 6= 0;
(iv) For each indecomposable projective A-module P we have HomA (P, M) 6= 0;
(v) For each indecomposable injective A-module I we have HomA (M, I) 6= 0
Proof. Exercise.
Theorem 33.8. Let M be a sincere directing A-module. Then the following hold:
(i) proj. dim(M) ≤ 1;

(ii) inj. dim(M) ≤ 1;
(iii) gl. dim(A) ≤ 2.
Proof. (i): We can assume that M is not projective. Assume there exists an inde-
composable injective A-module I with HomA (I, τ (M)) 6= 0. Since M is sincere, we
have HomA (M, I) 6= 0. This yields M I ≺ τ (M) ≺ M, a contradiction. Thus
proj. dim(M) ≤ 1.
(ii): This is similar to (i).

(iii): Assume gl. dim(A) > 2. Thus there are indecomposable A-modules with
ε
Ext3A (U, V ) 6= 0. Let 0 → Ω(U) → P0 − → U → 0 be a short exact sequence with
ε : P0 → U a projective cover. It follows that Ext2A (Ω(U), V ) ∼ = Ext3A (U, V ) 6= 0.
Thus proj. dim(Ω(U)) ≥ 2. Let U ′ be an indecomposable direct summand of Ω(U)
with proj. dim(U ′ ) ≥ 2. This implies HomA (I, τA (U ′ )) 6= 0 for some indecomposable
injective A-module I. It follows that
M I ≺ τA (U ′ ) ≺ U ′ ≺ P M
where P is an indecomposable direct summand of P0 , a contradiction. The first and
the last inequality follows from our assumption that M is sincere. This finishes the
proof.
Theorem 33.9. Let X and Y be indecomposable finite-dimensional A-modules with
dim(X) = dim(Y ). If X is a directing module, then X ∼
= Y.
Proof. (a): Without loss of generality we can assume that X and Y are sincere:
Assume X is not sincere. Then let R be the two-sided ideal in A which is gen-
erated by all primitive idempotents e ∈ A such that eX = 0. It follows that
R ⊆ AnnA (X) := {a ∈ A | aX = 0} and R ⊆ AnnA (Y ) := {a ∈ A | aY = 0}.
Clearly, eX = 0 if and only eY = 0, since dim(X) = dim(Y ). We also know that
AnnA (X) is a two-sided ideal: If a1 X = 0 and a2 X = 0, then (a1 + a2 )X = 0.
Furthermore, if aX = 0, then a′ aX = 0 and also aa′′ X ⊆ aX = 0 for all a′ , a′′ ∈ A.
It follows that X and Y are indecomposable sincere A/R-modules. Furthermore, X
is also directing as an A/R-module, since a path in mod(A/R) can also be seen as
a path in mod(A). Thus from now on assume that X and Y are sincere.
(b): Since X is directing, we get proj. dim(X) ≤ 1, inj. dim(X) ≤ 1 and gl. dim(A) ≤
2. Furthermore, we know that hdim(X), dim(X)iA = dimK EndA (X) > 0, and
therefore
hdim(X), dim(X)iA = hdim(X), dim(Y )iA

= dim HomA (X, Y ) − dim Ext1A (X, Y ) + dim Ext2A (X, Y ).
We have Ext2A (X, Y ) = 0 since proj. dim(X) ≤ 1. It follows that HomA (X, Y ) 6= 0.
Similarly,
hdim(X), dim(X)iA = hdim(Y ), dim(X)iA = dim HomA (Y, X) − Ext1A (Y, X)
since inj. dim(X) ≤ 1. This implies HomA (Y, X) 6= 0. Thus, if X 6∼
= Y , we get
X ≺ Y ≺ X, a contradiction.
Motivated by the previous theorem, we say that an indecomposable A-module X is

determined by composition factors if X ∼ = Y for all indecomposable A-modules
Y with dim(X) = dim(Y ).
Summary
Let A be a finite-dimensional K-algebra. By mod(A) we denote the category of

finite-dimensional left A-modules. Let ind(A) be the subcategory of mod(A) con-
taining all indecomposable A-modules.
The two general problems are these:

Problem 33.10. Classify all modules in ind(A).
Problem 33.11. Describe HomA (X, Y ) for all modules X, Y ∈ ind(A).
Note that we do not specify what “classify” and “describe” should exactly mean.
(a) Let E(A) be the subcategory of ind(A) containing all reachable A-modules.
For all X ∈ E(A) and all Y ∈ ind(A) we have dim(X) = dim(Y ) if and only
if X ∼
=Y.
(b) The knitting algorithm gives ∞ ∆ = ∞ (ΓA ) = E(A), and for each [X] ∈ E(A)
we can compute dim(X).
(c) For X ∈ ind(A) we have [X] ∈ E(A) if and only if X ∈ E(A).
(d) If K is a splitting field for A (for example, if K is algebraically closed), then

the mesh category KhE(A)e i is equivalent to E(A).
(e) We can use the mesh category of compute dim HomA (X, Y ) for all X, Y ∈
E(A).
We cannot hope to solve Problems 33.10 and 33.11 in general, but for the subcat-
egory E(A) ⊆ ind(A) of reachable A-modules, we get a complete classification of
reachable A-modules (the isomorphism classes of reachable modules are in bijection
with the dimension vectors obtained by the knitting algorithm), and we know a lot
of things about the morphism spaces between them.
Keep in mind that there is also a dual theory, using “coreachable modules” etc.
Furthermore, for some classes of algebras we have E(A) = ind(A), for example
if A is a representation-finite path algebra, or more generally if ΓA is a union of
preprojective components.
33.4. The quiver of an algebra. Let A be a finite-dimensional K-algebra. The

valued quiver QA of A has vertices 1, . . . , n, and there is an arrow i → j if and
only if dimK Ext1A (Si , Sj ) 6= 0. In this case, the arrow has valuation
dij := dimK Ext1A (Si , Sj ).
Each vertex i of QA has valuation di := dimK EndA (Si ).
Let Qop
A be the opposite quiver of A, which is obtained from QA by reversing all
arrows. The valuation of arrows and vertices stays the same.
Note that QA and Qop A can be seen as valued translation quivers, where all vertices
are projective and injective.
Special case: Assume that A is hereditary. Then we have
dPj Pi = dij and d P i = d Si = d i .
Thus, the subquiver PA of preprojective components of (ΓA , dA ) is (as a valued
translation quiver) isomorphic to NQop
A.
We define the valued graph QA of A as follows: The vertices are again 1, . . . , n.

There is a (non-oriented) edge between i and j if and only if
Ext1A (Si , Sj ) ⊕ Ext1A (Sj , Si ) 6= 0.
Such an edge has as a valuation the pair
(dimEndA (Sj ) Ext1A (Si , Sj ), dimEndA (Si )op Ext1A (Si , Sj )) = (dij /dj , dij /di).
Example of a valued graph:

(2,1)
· · · ·
The representation-finite hereditary algebras can be characterized as follows:

Theorem 33.12. A hereditary algebra A is representation-finite if and only if QA

is a Dynkin graph.
The list of Dynkin graphs can be found in Skript 3. Note that non-isomorphic
hereditary algebras can have the same valued graph.
33.5. Exercises. 1: Let A be an algebra with gl. dim(A) ≥ d. Show that there
exist indecomposable A-modules X and Y with ExtdA (X, Y ) 6= 0.
————————————————————————————-
34. Cartan and Coxeter matrix
Let A be a finite-dimensional K-algebra. We use the usual notation:
• P1 , . . . , Pn are the indecomposable projective A-modules;

• I1 , . . . , In are the indecomposable injective A-modules;
• S1 , . . . , Sn are the simple A-modules;
• Si ∼= top(Pi ) ∼ = soc(Ii ).
(Of course, the modules Pi , Ii and Si are just sets of representatives of isomorphism
classes of projective, injective and simple A-modules, respectively.)
Let X and Y be A-modules.
If proj. dim(X) < ∞ or inj. dim(Y ) < ∞, then

X
hX, Y iA := hdim(X), dim(Y )iA := (−1)t dimK ExtiA (X, Y )
t≥0
is the Ringel form of A. This defines a (not necessarily symmetric) bilinear form
h−, −iA : Zn × Zn → Z.
If proj. dim(X) < ∞ or inj. dim(X) < ∞, then set
X
χA (X) := χA (dim(X)) := hX, XiA = (−1)t dimK ExtiA (X, X).
t≥0
This defines a quadratic form χA (−) : Zn → Z.
34.1. Coxeter matrix.
We did all the missing proofs in this section in the lectures. But you
also find them in Ringel’s book.
If dim(P1 ), . . . , dim(Pn ) are linearly independent, then define the Coxeter matrix
ΦA of A by
dim(Pi )ΦA = −dim(Ii )
for 1 ≤ i ≤ n. It follows that ΦA ∈ Mn (Q).

Lemma 34.1. If gl. dim(A) < ∞, then dim(P1 ), . . . , dim(Pn ) are linearly indepen-
dent.
Proof. We know that gl. dim(A) < ∞ if and only if proj. dim(S) < ∞ for all simple
A-modules S. Furthermore {dim(Si ) | 1 ≤ i ≤ n} are a free generating set of the
Grothendieck group G(A). Let
0 → P (d) → · · · → P (1) → P (0) → S → 0
be a minimal projective resolution of a simple A-module S. This implies
d
X
(−1)i dim(P (i) ) = dim(S).
i=0
Thus the vectors dim(Pi ) generate Zn . The result follows.
Dually, if gl. dim(A) < ∞, then dim(I1 ), . . . , dim(In ) are also linearly independent.
So ΦA is invertible in this case.
By the definition of ΦA , for each P ∈ proj(A) we have
(3) dim(P )ΦA = −dim(ν(P )).
p
Let M be an A-module, and let P (1) −
→ P (0) → M → 0 be a minimal projective
presentation of M. Thus we obtain an exact sequence
(4) 0 → M ′′ → P (1) → P (0) → M → 0
where M ′′ = Ker(p) = Ω2 (M). We also get an exact sequence
νA (p)
(5) 0 → τA (M) → νA (P (1) ) −−−→ νA (P (0) ) → νA (M) → 0
since the Nakajama functor νA is right exact.
There is the dual construction of τA−1 : For an A-module N let
q
(6) 0 → N → I (0) −
→ I (1) → N ′′ → 0
q
be an exact sequence where 0 → N → I (0) −
→ I (1) is a minimal injective presentation
of N.
Applying νA−1 yields an exact sequence
ν −1 (q)
(7) 0 → νA−1 /N) → νA−1 (I (0) ) −−
A
−→ νA−1 (I (1) ) → τA−1 (N) → 0
Lemma 34.2. We have
(8) dim(τA (M)) = dim(M)ΦA − dim(M ′′ )ΦA + dim(νA (M)).
Proof. From Equation (4) we get

−dim(P (1) ) + dim(P (0) ) = dim(M) − dim(M ′′ ).
Applying ΦA to this sequence, and using dim(P )ΦA = −dim(νA (P )) for all projec-
tive modules P , we get
dim(νA (P (1) )) − dim(νA (P (0) )) = dim(M)ΦA − dim(M ′′ )ΦA .
From the injective presentation of τA (M) (see in Equation (5)) we get
dim(τA (M)) = dim(νA (P (1) )) − dim(νA (P (0) )) + dim(νA (M))
= dim(M)ΦA − dim(M ′′ )ΦA + dim(νA (M)).

Lemma 34.3. If proj. dim(M) ≤ 2, then
(9) dim(τA (M)) ≥ dim(M)ΦA .
If proj. dim(M) ≤ 2 and inj. dim(τA (M)) ≤ 2, then
(10) dim(τA (M)) − dim(M)ΦA = dim(I)
for some injective module I.
Proof. If proj. dim(M) ≤ 2, then M ′′ is projective, which implies dim(M ′′ )ΦA =

−dim(νA (M ′′ )). Therefore
dim(τA (M)) − dim(M)ΦA = dim(νA (M ′′ ) ⊕ νA (M)),
and therefore this vector is non-negative. Note that νA (M ′′ ) is injective. If we
assume additionally that inj. dim(τA (M)) ≤ 2, then νA (M) is also injective, since it
is the cokernel of the homomorphism
νA (p) : νA (P (1) ) → νA (P (0) )
with νA (P (1) ) and νA (P (0) ) being injective.
Lemma 34.4. If proj. dim(M) ≤ 1 and HomA (M, A A) = 0, then
(11) dim(τA (M)) = dim(M)ΦA .
Proof. If proj. dim(M) ≤ 1, then M ′′ = 0, since Equation (4) gives a minimal

projective presentation of M. By assumption νA (M) = D HomA (M, A A) = 0. Thus
the result follows directly from Equation (8).
Note that Equation (11) has many consequences and applications. For example, if
A is a hereditary algebra, then each A-module M satisfies proj. dim(M) ≤ 1, and if
M is non-projective, then HomA (M, A A) = 0.
Lemma 34.5. Assume proj. dim(M) ≤ 2. If dim(τA (M)) = dim(M)ΦA , then
proj. dim(M) ≤ 1 and HomA (M, A A) = 0.
Proof. Clearly, dim(τA (M)) = dim(M)ΦA implies νA (M ′′ )oplusνA (M) = 0. Since

M ′′ is projective, we have νA (M ′′ ) = 0 if and only if M ′′ = 0.
Using the notations from Equation (6) and (7) we obtain the following dual state-
ments:
(i) We have
dim(τA−1 (N)) = dim(N)φ−1 ′′ −1 −1
A − dim(N )ΦA + dim(νA (N)).
(ii) If inj. dim(N) ≤ 2, then

dim(τA−1 (N)) ≥ dim(N)Φ−1
A .
If inj. dim(N) ≤ 2 and proj. dim(τA−1 (N)) ≤ 2, then

dim(τA−1 (N)) − dim(N)Φ−1
A = dim(P )
for some projective module P .

(iii) If inj. dim(N) ≤ 1 and HomA (D(AA ), N) = 0, then
dim(τA−1 (N)) = dim(N)Φ−1
A .
Lemma 34.6. If 0 → U → X → V → 0 is a non-split short exact sequence of

A-modules, then
dim EndA (X) < dim EndA (U ⊕ V ).
Proof. Applying HomA (−, U), HomA (−, X) and HomA (−, V ) we obtain the com-
mutative diagram
0 0 0

δ
0 / HomA (V, U) / HomA (X, U) / HomA (U, U) / Ext1A (V, U)

0 / HomA (V, X) / HomA (X, X) / HomA (U, X)

0 / HomA (V, V ) / HomA (X, V ) / HomA (U, V )
with exact rows and columns. Since η does not split, we know that the connecting
homomorphism δ is non-zero. This implies
dim HomA (X, U) ≤ dim HomA (V, U) + dim HomA (U, U) − 1.
Thus we get
dim HomA (X, X) ≤ dim HomA (X, U) + dim HomA (X, V )
≤ dim HomA (V, U) + dim HomA (U, U) − 1
+ dim HomA (V, V ) + dim HomA (U, V )
= dim EndA (U ⊕ V ) − 1.
Recall that for an indecomposable A-module X we defined

F (X) = EndA (X)/ rad(EndA (X)),
which is a K-skew field. If K is algebraically closed, then F (X) ∼
= K for all indecom-
posables X. If K is a splitting field for K, then F (τ (Pi )) = K and F (τ n (Ii )) ∼
−n ∼ =K
for all n ≥ 0.
An algebra A is directed if every indecomposable A-module is directing.
Let A be of finite-global dimension. Then we call the quadratic form χA weakly
positive if χA (x) > 0 for all x > 0 in Zn . If x ∈ Zn with χA (x) = 1, then x is called
a root of χA .
Theorem 34.7. Let A be a finite-dimensional directed algebra. If gl. dim(A) ≤ 2,
then the following hold:
(i) χA is weakly positive;

(ii) If K is algebraically closed, then dim yields a bijection between the set of
isomorphism classes of indecomposable A-modules and the set of positive
roots of χA .
Proof. (i): Let x > 0 in G(A) = Zn . Thus x = dim(X) for some non-zero A-module
X. We choose X such that dim EndA (X) is minimal. In other words, if Y is another
module with dim(Y ) = x, then dim EndA (X) ≤ dim EndA (Y ).
Let X = X1 ⊕ · · · ⊕ Xt with Xi indecomposable for all i. It follows from Lemma

34.6 that Ext1A (Xi , Xj ) = 0 for all i 6= j. (Without loss of generality assume
Ext1A (X2 , X1 ) 6= 0. Then there exists a non-split short exact sequence
t
M
0 → X1 → Y → Xi → 0
i=2
and Lemma 34.6 implies that dim EndA (Y ) < dim EndA (X), a contradiction.) Fur-
thermore, since Xi is directing, we have Ext1A (Xi , Xi) = 0 for all i. Thus we get
Ext1A (X, X) = 0. Since gl. dim(A) ≤ 2, we have
χA (x) = χA (dim(X)) = dim EndA (X) + dim Ext2A (X, X) > 0.
Thus χA is weakly positive.
(ii): If Y is an indecomposable A-module, then we know that
χA (Y ) = dim EndA (Y ),
since Y is directing. We also know that EndA (Y ) is a skew field, which implies
F (Y ) ∼
= EndA (Y ). Thus, χA (Y ) = 1 in case F (Y ) ∼
= K.
Furthermore, we know that any two non-isomorphic indecomposable A-modules Y
and Z satisfy dim(Y ) 6= dim(Z). So the map dim is injective.
Assume additionally that x is a root of χA . Now
1 = χA (x) = dim EndA (X) + dim Ext2A (X, X)
shows that EndA (X) ∼

= K. This implies that X is indecomposable.
It follows that the map dim from the set of isomorphism classes of indecomposable
A-modules to the set of positive roots is surjective.
Note that a sincere directed algebra A always satisfies gl. dim(A) ≤ 2.

Corollary 34.8. If Q is a representation-finite quiver, then χKQ is weakly positive.
Proof. If KQ is representation-finite, then ΓKQ consists of a union of preprojective

components. Therefore all KQ-modules are directed. Furthermore, gl. dim(KQ) ≤
1. Now one can apply the above theorem.
Proposition 34.9 (Drozd). A weakly positive integral quadratic form χ has only
finitely many positive roots.
Proof. Use partial derivations of χ and some standard results from Analysis. For
details we refer to [Ri1].
From now on we assume that K is a splitting field for A.
34.2. Cartan matrix. As before, we denote the transpose of a matrix M by M T .

For a ring or field R we denote the elements in Rn as row vectors.
The Cartan matrix CA = (cij )ij of A is the n × n-matrix with ijth entry equal to
cij := [Pj : Si ] = dim(Pj )i .
Thus the jth column of CA is given by dim(Pj )T .
Recall that the Nakayama functor ν = νA = D HomA (−, A A) induces an equivalence
ν : proj(A) → inj(A)
where ν(Pi ) = Ii . It follows that
dim(Ii )j = dim HomA (Ii , Ij ) = dim HomA (Pi , Pj ) = cij .
(Here we used our assumption that K is a splitting field for A.)
Thus the ith row of CA is equal to dim(Ii ). So we get
(12) dim(Pi ) = ei CAT and dim(Ii ) = ei CA .
Lemma 34.10. If gl. dim(A) < ∞, then CA is invertible over Z.
Proof. Copy the proof of Lemma 34.1.
But note that there are algebras A where CA is invertible over Q, but not over Z,
for example if A is a local algebra with non-zero radical.
Assume now that the Cartan matrix CA of A is invertible. We get a (not necesssarily
symmetric) bilinear form
h−, −i′A : Qn × Qn → Q
defined by
hx, yi′A := xCA−T y T .
Here CA−T denote the inverse of the transpose CAT of C. Furthermore, we define a
symmetric bilinear form
(−, −)′A : Qn × Qn → Q
by
(x, y)′A := hx, yi′A + hy, xi′A = x(CA−1 + CA−T )y T .
Set χ′A (x) := hx, xi′A . This defines a quadratic form
χ′A : Qn → Q.
It follows that
(x, y)′A = χ′A (x + y) − χ′A (x) − χ′A (y).
The radical of χ′A is defined by

rad(χ′A ) = {w ∈ Qn | (w, −)′A = 0}.
The following lemma shows that the form h−, −i′A we just defined using the Cartan
matrix, coincides with the Ringel form we defined earlier:
Lemma 34.11. Assume that CA is invertible. If X and Y are A-modules with
proj. dim(X) < ∞ or inj. dim(Y ) < ∞, then
X
hdim(X), dim(Y )i′A = hX, Y iA = (−1)t dim ExttA (X, Y ).
t≥0
In particular, χ′A (dim(X)) = χA (X).
Proof. Assume proj. dim(X) = d < ∞. (The case inj. dim(Y ) < ∞ is done dually.)
We use induction on d.
If d = 0, then X is projective. Without loss of generality we assume that X is
indecomposable. Thus X = Pi for some i. Let y = dim(Y ). We get
hdim(X), dim(Y )i′A = hdim(Pi ), yi′A = dim(Pi )CA−T y T = ei y T = dim HomA (Pi , Y ).
Furthermore, we have ExttA (Pi , Y ) = 0 for all t > 0.
Next, let d > 0. Let P → X be a projective cover of X and let X ′ be its kernel. It
follows that proj. dim(X ′ ) = d − 1. We apply HomA (−, Y ) to the exact sequence
0 → X ′ → P → X → 0.
Using the long exact homology sequence we obtain

X X X
(−1)i dim ExttA (X, Y ) = (−1)i dim ExttA (P, Y ) − (−1)i dim ExttA (X ′ , Y )
t≥0 t≥0 t≥0
= hdim(P ), Y i′A − hdim(X ), dim(Y )i′A

′
= hdim(X), dim(Y )i′A .

Here the second equality is obtained by induction. This finishes the proof.
Let δij be the Kronecker function.

Corollary 34.12. If A is hereditary, then
(
1 if i = j,
hei , ej iA =
−dim Ext1A (Si , Sj ) otherwise.
Proof. This holds since gl. dim(A) ≤ 1 and since K is a splitting field for A.
Lemma 34.13. Let A = KQ be a finite-dimensional path algebra. Then for any
simple A-module Si and Sj we have dim Ext1A (Si , Sj ) is equal to the number of
arrows i → j in Q.
Proof. Let aij be the number of arrows i → j. Since A is finite-dimensional we have

aii = 0 for all i. The minimal projective resolution of the simple A-module Si is of
the form
Mn
a
0→ Pj ij → Pi → Si → 0
j=1
Applying HomA (−, Sj ) yields an exact sequence

a
0 → HomA (Si , Sj ) → HomA (Pi , Sj ) → HomA (Pj ij , Sj ) → Ext1A (Si , Sj ) → 0.
Thus dim Ext1A (Si , Sj ) = aij .
Corollary 34.14. Let A = KQ be a finite-dimensional path algebra, and let X and
Y be A-modules with dim(X) = α and dim(Y ) = β. Then
X X
hX, Y iKQ = hα, βiKQ = αi βi − αs(a) βt(a)
i∈Q0 a∈Q1
and
n
X X
χKQ (X) = hα, αiKQ = αi2 − qij αi αj
i=1 i<j
where qij is the number of arrows a ∈ Q1 with {s(a), t(a)} = {i, j}.
Lemma 34.15. Assume that CA is invertible. Then
ΦA = −CA−T CA .
Proof. For each 1 ≤ i ≤ n we have to show that

(13) dim(Pi )ΦA = −dim(Ii ).
We have
dim(Pi )(−CA−T CA ) = −dim(Ii ) if and only if − dim(Ii )T = −CAT CA−1 dim(Pi )T .
Clearly, CA−1 dim(Pi )T = eTi , and −CAT eTi = −dim(Ii )T .
Example: Let Q be the quiver

2 NNN 3 4 pp 5
NNN === pppp
NNN == p
NNN == ppp
NN' wpppp
1
and let A = KQ. Then !
1 1 1 1 1
0 1 0 0 0
CA = 0 0 1 0 0
0 0 0 1 0
0 0 0 0 1
and !
−1 −1 −1 −1 −1
1 0 1 1 1
ΦA = 1 1 0 1 1 .
1 1 1 0 1
1 1 1 1 0
Here are some calculations:
• (3, 1, 1, 1, 1)ΦA = (1, 0, 0, 0, 0) and (3, 1, 1, 1, 1)Φ2A = −(1, 1, 1, 1, 1),

• (1, 1, 1, 0, 0)ΦA = (1, 0, 0, 1, 1) and (1, 1, 1, 0, 0)Φ2A = (1, 1, 1, 0, 0),
• (2, 1, 1, 1, 1)ΦA = (2, 1, 1, 1, 1).
Lemma 34.16. For all x, y ∈ Qn we have
hx, yi′A = −hy, xΦA i′A = hxΦA , yΦA i′A .
Proof. We have
hx, yi′A = xCA−T y T = (xCA−T y T )T = yCA−1 xT
= yCA−T CAT CA−1 xT = −yCA−T ΦTA xT = −hy, xΦA i′A .
This proves the first equality. Repeating this calculation we obtain the second
equality.
Lemma 34.17. If there exists some x > 0 such that xΦA = x, then χA is not weakly
positive.
Proof. We have (x, y)′A = 0 for all y if and only if x(CA−1 + CA−T ) = 0 if and only if
xCA−1 = −xCA−T if and only if xΦA = x.
Corollary 34.18. If there exists some x > 0 such that xΦA = x, then χ′A is not
weakly positive.
Proof. If x ∈ rad(χ′A ), then χ′A (x) = 0.

m
Assume there exists an indecomposable KQ-module X with τKQ (X) ∼
= X and
assume m ≥ 1 is minimal with this property. Set
m
M
i
Y = τKQ (X).
i=1
Then τKQ (Y ) ∼
= Y which implies
dim(Y ) = dim(Y )ΦKQ .
We get
(Y, Z)KQ = hY, ZiKQ + hZ, Y iKQ
= −hdim(Z), dim(Y )ΦKQ i − hdim(Y )Φ−1
KQ , dim(Z)i
= −(hY, ZiKQ + hZ, Y iKQ ).
This implies dim(Y ) ∈ rad(χKQ ).
Lemma 34.19. For an A-module M the following hold:
(i) If proj. dim(M) ≤ 1, then

τA (M) ∼
= D Ext1A (M, A A).
(ii) If inj. dim(M) ≤ 1, then
τA−1 (M) ∼
= Ext1Aop (D(M), AA ).
Proof. Assume proj. dim(M) ≤ 1. Then in Equation (4) we have M ′′ = 0. Applying

HomA (−, A A) yields an exact sequence
0 HomA (M, A A) → HomA (P (0) , A A) → HomA (P (1) , A A) → Ext1A (M, A A) → 0
of right A-modules. Keeping in mind that νA = D HomA (−, A A) we dualize the
above sequence get an exact sequence
0D Ext1A (M, A A) → νA (P (1) ) → νA (P (0) ) → νA (M) → 0.
This implies (i). Part (ii) is proved dually.
34.3. Exercises. 1: Show the following: If the Cartan matrix CA is an upper

triangular matrix, then CA is invertible over Q. In this case, CA is invertible over Z
if and only if EndA (Pi ) ∼
= K for all i.
————————————————————————————-
35. Representation theory of quivers
Parts of this section are copied from Crawley-Boevey’s lecture notes “Lectures on
representations of quivers”, which you can find on his homepage.
35.1. Bilinear and quadratic forms. Let Q = (Q0 , Q1 , s, t) be a finite quiver

with vertices Q0 = {1, . . . , n}, and let A = KQ be the path algebra of Q.
For vertices i, j ∈ Q0 let qij = qji be the number of arrows a ∈ Q1 with {s(a), t(a)} =
{i, j}. Note that the numbers qij do not depend on the orientation of Q.
For α = (α1 , . . . , αn ) ∈ Zn define
n
X X
qQ (α) := αi2 − qij αi αj .
i=1 i≤j
n
We call the quadratic form qQ : Z → Z the Tits form of Q.
The symmetric bilinear form (−, −)Q : Zn × Zn → Z of Q is defined by
(
−qij if i 6= j,
(ei , ej )Q :=
2 − 2qii otherwise.
As before, ei denotes the canonical basis vector of Zn with ith entry 1 and all other
entries 0.
We have
(α, α)Q = 2qQ (α),
(α, β)Q = qQ (α + β) − qQ (α) − qQ (β).
Note that qQ and (−, −)Q do not depend on the orientation of the quiver Q.
For α, β ∈ Zn define
X X
hα, β, iQ := αi βi − αs(a) βt(a) .
i∈Q0 a∈Q1
This defines a (not necessarily symmetric) bilinear form

h−, −iQ : Zn × Zn → Z
which is called the Euler form of Q. Clearly, we have
qQ (α) = hα, αiQ ,
(α, β)Q = hα, βiQ + hβ, αiQ.
The bilinear form h−, −iQ does depend on the orientation of Q.
The Tits form qQ is positive definite if qQ (α) > 0 for all 0 6= α ∈ Zn , and qQ is
positive semi-definite if qQ (α) ≥ 0 for all α ∈ Zn .
The radical of q is defined by
rad(qQ ) = {α ∈ Zn | (α, −)Q = 0}.
For α, β ∈ Zn set β ≥ α if β − α ∈ Nn . This defines a partial ordering on Zn .
An element α = (α1 , . . . , αn ) ∈ Zn is sincere if αi 6= 0 for all i. We write α ≥ 0 if
αi ≥ 0 for all i, and α > 0 if α ≥ 0 and αi > 0 for some i.
Let S1 , . . . , Sn be the simple KQ-modules corresponding to the vertices of Q. (These

are the only simple KQ-modules if and only if Q has no oriented cycles.) It is easy
to check that dim Ext1KQ (Si , Sj ) equals the number of arrows i → j in Q. (Just
construct the minimal projective resolution
M a
0→ Pj ij → Pi → Si → 0
j∈Q0
of Si , where aij is the number of arrows i → j in Q. Then apply the functor

HomKQ (−, Sj ).)
Lemma 35.1. Let Q be a connected quiver, and let β ≥ 0 be a non-zero element in
rad(qQ ). Then the following hold:
(i) β is sincere;
(ii) qQ is positive semi-definite;
(iii) For α ∈ Zn the following are equivalent:
(a) qQ (α) = 0;
(b) α ∈ Qβ;
(c) α ∈ rad(qQ ).
Proof. (a): By assumption we have

X
(β, ei)Q = (2 − 2qii )βi − qij βj = 0.
j6=i
If βi = 0, then X
qij βj = 0,
j6=i
and since qij ≥ 0 for all i, j and β ≥ 0, we get βj = 0 whenever qij > 0. Since Q is
connected, we get β = 0, a contradiction. Thus we proved that β is sincere.
(b): The following calculation shows that qQ is positive semi-definite:

X βi βj αi αj 2 X βj 2 X X βi 2
qij − = qij αi − qij αi αj + qij αj
i<j
2 βi β j i<j
2βi i<j i<j
2β j
X βj 2 X
= qij α − qij αi αj
i6=j
2βi i i<j
X 1 2 X
= (2 − 2qii )βi αi − qij αi αj = qQ (α).
i
2β i i<j
For the last equality we used n times the equation

X
(2 − 2qii )βi = qij βj .
j6=i
(c): If qQ (α) = 0, then the calculation above shows that αi /βi = αj /βj whenever
qij > 0. Since Q is connected it follows that α ∈ Qβ.
(d): If α ∈ Qβ, then α ∈ rad(qQ ), since β ∈ rad(qQ ).
(e): Clearly, if α ∈ rad(qQ ), then qQ (α) = 0.

Theorem 35.2. Suppose that Q is connected.
(i) If Q is a Dynkin quiver, then qQ is positive definite;

(ii) If Q is an Euclidean quiver, then qQ is positive semi-definite and rad(qQ ) =
Zδ, where δ is the dimension vector for Q listed in Figure 2;
(iii) If Q is not a Dynkin and not an Euclidean quiver, then there exists some
α ≥ 0 in Zn with qQ (α) < 0 and (α, ei)Q ≤ 0 for all i.
Proof. (ii): It is easy to check that δ ∈ rad(qQ ): If there are no loops or multiple
edges we have to check that for all vertices i we have
X
2δi = δj
j
where j runs over the set of neighbours of i in Q. By Lemma 35.1 this implies that
qQ is positive semi-definite.
In each case there exists some vertex i such that δi = 1. Thus rad(qQ ) = Qδ ∩ Zn =
Zδ.
(i): Any Dynkin quiver Q with n vertices can be seen as a full subquiver of some
Euclidean quiver Q e with n + 1 vertices. We have q e (x) > 0 for all non-sincere
Q
elements in Zn+1 , since the x with qQe (x) = 0 are all multiples of the sincere element
δ. So qQ is positive definite. (The form qQ is obtained from qQe via restriction to the
e
subquiver Q of Q.)
(iii): Let Q be a quiver which is not Dynkin and not Euclidean. Then Q contains
a (not necessarily full) subquiver Q′ such that Q′ is a Euclidean quiver. Note that
any dimension vector of Q′ can be seen as a dimension vector of Q by just adding
some zeros in case Q has more vertices than Q′ .
Let δ be the radical vector associated to Q′ . If the vertex sets of Q′ and Q coincide,
then α := δ satisfies qQ (α) < 0.
Otherwise, if i is a vertex of Q which is not a vertex of Q′ but which is connected
to a vertex in Q′ by an edge, then α := 2δ + ei satisfies qQ (α) < 0.
Let Q be a Euclidean quiver. If i is a vertex of Q with δi = 1, then i is called

an extending vertex. Observe that there always exists such an extending vertex.
Furthermore, if we delete an extending vertex (and the arrows attached to it), then
we will obtain a corresponding Dynkin diagram.
For Q a Dynkin or an Euclidean quiver, let

∆Q := {α ∈ Zn | α 6= 0, qQ (α) ≤ 1}
be the set of roots of Q.
A root α of Q is real if qQ (α) = 1. Otherwise, if qQ (α) = 0, it is called an imaginary

root. Let ∆re im
Q and ∆Q be the set of real and imaginary roots, respectively.
Proposition 35.3. Let Q be a Dynkin or a Euclidean quiver. Then the following

hold:
(i) Each ei is a root;

(ii) If α ∈ ∆Q ∪ {0}, then −α and α + β are in ∆Q ∪ {0} where β ∈ rad(qQ );
(iii) We have
(
∅ if Q is Dynkin,
∆im
Q =
{rδ | 0 6= r ∈ Z} if Q is Euclidean;
(iv) Every root α ∈ ∆Q is either positive or negative;
(v) If Q is Euclidean, then the set (∆Q ∪ {0})/Zδ of residue classes modulo Zδ
is finite;
(vi) If Q is Dynkin, then ∆Q is finite.
Proof. (i): Clearly, we have qQ (ei ) = 1, so ei is a root.

(ii): Let α ∈ ∆Q ∪ {0} and β ∈ rad(qQ ). Since (β, α)Q = 0 = qQ (β), we have
qQ (α) = qQ (β + α) = qQ (β) + qQ (α) + (β, α)Q
= qQ (β − α) = qQ (β) + qQ (α) − (β, α)Q
Thus −α and α + β are in ∆Q ∪ {0}. (The case β = 0 yields qQ (−α) = qQ (α).)
(iii): This follows directly from Lemma 35.1.
(iv): Let α be a root. So we can write α = α+ − α− where α+ , α− ≥ 0 and have

disjoint supports. Assume that both α+ and α− are non-zero. It follows immediately
that (α+ , α− )Q ≤ 0. This implies
1 ≥ qQ (α) = qQ (α+ ) + qQ (α− ) − (α+ , α− )Q ≥ qQ (α+ ) + qQ (α− ).
Thus one of α+ and α− is an imaginary root and is therefore sincere. So the other
one is zero, a contradiction.
(v): Let Q be an Euclidean quiver, and let e be an extending vertex of Q. If α is a
root with αe = 0, then δ − α and δ + α are roots which are positive at the vertex e.
Thus both are positive roots. This implies
{α ∈ ∆ ∪ {0} | αe = 0} ⊆ {α ∈ Zn | −δ ≤ α ≤ δ},
and obviously this is a finite set.
If β ∈ ∆ ∪ {0}, then β − βe δ belongs to the finite set
{α ∈ ∆ ∪ {0} | αe = 0}.
(vi): If Q is a Dynkin quiver, we can consider Q as a full subquiver of the cor-

responding Euclidean quiver Qe with extending vertex e. (Thus, we obtain Q by
e We can now view a root α of Q as a root of Q

deleting e from Q.) e with αe = 0.
Thus by the proof of (v) we get that ∆ is a finite set.
35.2. Gabriel’s Theorem. Combining our results obtained so far, we obtain the
following famous theorem:
Theorem 35.4 (Gabriel). Let Q be a connected quiver. Then KQ is representation-
finite if and only if Q is a Dynkin quiver. In this case dim yields a bijection between
the set of isomorphism classes of indecomposable KQ-modules and the set of positive
roots of qQ .
Proof. (a): We know that there is a unique preprojective component PKQ of the
Auslander-Reiten quiver ΓKQ .
(b): We have χKQ (X) = qQ (dim(X)) for all KQ-modules X.
(c): Assume KQ is representation-finite. This is the case if and only if PKQ =
ΓKQ. Since all indecomposable preprojective modules are directed, we know that
KQ is a directed algebra. Furthermore, we have gl. dim(KQ) ≤ 1 ≤ 2. So we
can apply Theorem xx and obtain a bijection between the isomorphism classes of
indecomposable KQ-modules and the set of positive roots of χKQ . Furthermore, an
element α ∈ Nn is a positive root of χKQ if and only if α ∈ ∆Q . We also know that
χKQ = qQ is weakly positive. But this implies that Q has to be a Dynkin quiver.
(For all quivers Q which are not Dynkin we found some α > 0 with qQ (α) ≤ 0.)
(d): If KQ is representation-infinite, the component PKQ is infinite. Each indecom-
posable module X in PKQ is directed, and K is a splitting field for KQ. Thus
χKQ (X) = qQ (dim(X)) = 1.
Furthermore, we know that there is no other indecomposable KQ-module Y with
dim(X) = dim(Y ). So we found infinitely many α ∈ Zn with qQ (α) = 1.
Suppose that Q is a Dynkin quiver. Then
∆Q = {α ∈ Zn | qQ (α) = 1}
is a finite set, a contradiction.
————————————————————————————-
36. Cartan matrices and (sub)additive functions
In Figure 1 we define a set of valued graphs called Dynkin graphs. By definition

each of the graphs An , Bn , Cn and Dn has n vertices. The graphs An , Dn , E6 , E7
and E8 are the simply laced Dynkin graphs.
An · · ··· · · n≥1
(1,2)
Bn · · ··· · · n≥2
(2,1)
Cn · · ··· · · n≥3
Dn · · · ··· · n≥4
E6 · · · · ·
E7 · · · · · ·
E8 · · · · · · ·
(1,2)
F4 · · · ·
(1,,3)
G2 · ·
Figure 1. Dynkin graphs
In Figure 2 we define a set of valued graphs called Euclidean graphs. By definition

each of the graphs A en , B
en , C
en , D
e n , BC
g n , BD
g n and CD
g n has n + 1 vertices. The
e e e e e
graphs An , Dn , E6 , E7 and E8 are the simply laced Euclidean graphs. By
definition the graph Ae0 has one vertex and one loop, and Ae1 has two vertices joined
by two edges. Our table of Euclidean graphs does not only contain the graphs
themselves, but for each graph we also display a dimension vector which we will
denote by δ.
A quiver Q is a Dynkin quiver or an Euclidean quiver of the underlying graph
of Q (replace each arrow of Q by a non-oriented edge) is a simply laced Dynkin
graph or a simply laced Euclidean graph, respectively.
***********************************************************************
en
A 1 1 1 ··· 1 1 1
(1,2) (2,1)
en
B 1 1 1 ··· 1 1 1
(2,1) (1,2)
en
C 1 2 2 ··· 2 2 1
1 1
en
D 1 2 2 ··· 2 2 1
e6
E 1 2 3 2 1
e7
E 1 2 3 4 3 2 1
e8
E 2 4 6 5 4 3 2 1
(1,4) (2,2)
e11
A 2 1 e12
A 1 1
(1,2) (1,2)
gn
BC 2 2 2 ··· 2 2 1
(2,1)
gn
BD 1 2 2 ··· 2 2 2
(1,2)
gn
CD 1 2 2 ··· 2 2 1
(1,2) (2,1)
Fe41 1 2 3 2 1 Fe42 1 2 3 4 2
(1,3) (3,1)
e21
G 1 2 1 e22
G 1 2 3
Figure 2. Euclidean graphs and additive functions δ

Part 7. Extras
37. Classes of modules
simple modules
serial modules
uniserial modules
cyclic modules
cocyclic modules
indecomposable modules
projective modules
injective modules
preprojective modules (which should really be called postprojective modules)
preinjective modules
regular modules
bricks
stones
exceptional modules
Schur modules
tree modules (2 different definitions)

string modules
band modules
(generalized) tilting modules
(generalized) partial tilting modules

torsion modules
torsion free modules
In the world of infinite dimensional modules we find names like the following:
Prüfer modules
p-adic modules
generic modules
pure-injective modules
algebraically compact module
Classifications of modules
For some algebras of infinite representation type, a complete classification of inde-

composable modules is known. We list some of these classes of algebras:
Solved:
tame hereditary algebras
tubular algebras
Gelfand-Ponomarev algebras
dihedral 2-group algebras

quaternion algebra
special biserial algebras
clannish algebras
multicoil algebras
Open:
biserial algebras
However, one still has to be careful what it means to have a classification of all
indecomposable modules over an algebra. For example for tubular algebras, one can
parametrize all indecomposable modules by roots of a quadratic form. But given a
root, it is still very difficult to write down explicitely the corresponding indecom-
posable module(s). In fact, for tubular algebras this remains an open problem.
38. Classes of algebras
We list some names of classes of mostly finite-dimensional algebras which were stud-
ied in the literature:
Basic algebras
Biserial algebras
Canonical algebras
Clannish algebras
Cluster-tilted algebra
Directed algebras
Dynkin algebras
Euclidean algebras
Gentle algebras
Group algebras
Hereditary algebras
Multicoil algebras
Nakayama algebras
Path algebras
Poset algebras
Preprojective algebras
Quasi-hereditary algebras
Quasi-tilted algebras
Representation-finite algebras
Selfinjective algebras
Semisimple algebras
Simply connected algebras

Special biserial algebras
String algebras
Strongly simply connected algebras
Symmetric algebras
Tame algebras
Tilted algebras
Tree algebras
Triangular algebras
Trivial extension algebras

Tubular algebras
Wild algebras
Here are some classes of algebras, which are not finite-dimensional, but linked to
the finite-dimensional world:
Repetitive algebras
Enveloping algebras of Lie algebras

Quantized enveloping algebras
Ringel-Hall algebras
Cluster algebras
Hecke algebras
39. Dimensions
The concept of “dimension” occurs frequently and with different meanings in the
representation theory of algebras. Here just some of the most common dimensions:
dimension of a module as a vector space
projective dimension of a module
injective dimension of a modules

global dimension of an algebra
finitistic dimension of an algebra
dominant dimension of an algebra
representation dimension of an algebra

Krull-Gabriel dimension of an algebra
Krull-dimension of a commutative ring
dimension of a variety
***********************************************************************
References
[ASS] I. Assem, D. Simson, A. Skowroński, Elements of the Representation Theory of Associative
Algebras. LMS Student Texts 65.
[ARS] M. Auslander, I. Reiten, S.Smalø, Representation theory of Artin algebras. Corrected
reprint of the 1995 original. Cambridge Studies in Advanced Mathematics, 36. Cambridge
University Press, Cambridge, 1997. xiv+425 pp.
[B1] D.J. Benson, Representations and cohomology. I. Basic representation theory of finite groups
and associative algebras. Second edition. Cambridge Studies in Advanced Mathematics, 30.
Cambridge University Press, Cambridge, 1998. xii+246 pp.
[B2] D.J. Benson, Representations and cohomology. II. Cohomology of groups and modules. Sec-
ond edition. Cambridge Studies in Advanced Mathematics, 31. Cambridge University Press,
Cambridge, 1998. xii+279 pp.
[CE] H. Cartan, S. Eilenberg, Homological algebra. With an appendix by David A. Buchsbaum.
Reprint of the 1956 original. Princeton Landmarks in Mathematics. Princeton University
Press, Princeton, NJ, 1999. xvi+390 pp.
[DK] Y.A. Drozd, V. Kirichenko, Finite-dimensional algebras. Translated from the 1980 Russian
original and with an appendix by Vlastimil Dlab. Springer-Verlag, Berlin, 1994. xiv+249 pp.
[G] P. Gabriel, Auslander-Reiten sequences and representation-finite algebras. Representation
theory, I (Proc. Workshop, Carleton Univ., Ottawa, Ont., 1979), pp. 1–71, Lecture Notes in
Math., 831, Springer, Berlin, 1980.
[GR] P. Gabriel, A.V. Roiter, Representations of finite-dimensional algebras. Translated from the
Russian. With a chapter by B. Keller. Reprint of the 1992 English translation. Springer-
Verlag, Berlin, 1997. iv+177 pp.
[HSt] Hilton, Stammbach
[P] R.S. Pierce, Associative algebras. Graduate Texts in Mathematics, 88. Studies in the History
of Modern Science, 9. Springer-Verlag, New York-Berlin, 1982. xii+436 pp.
[ML] S. Mac Lane, Homology. Reprint of the 1975 edition. Classics in Mathematics. Springer-
Verlag, Berlin, 1995. x+422 pp.
[Ri1] C.M. Ringel, Tame algebras and integral quadratic forms. Lecture Notes in Mathematics,
1099. Springer-Verlag, Berlin, 1984. xiii+376 pp.
[Ri2] C.M. Ringel, Representation theory of finite-dimensional algebras. Representations of alge-
bras (Durham, 1985), 7–79, London Math. Soc. Lecture Note Ser., 116, Cambridge Univ.
Press, Cambridge, 1986.
Index
A-module, 86 Category, 37
A-module homomorphism, 87 Centre of a ring, 101
J-module, 12 Chinese Reminder Theorem, 44
J-module homomorphism, 23 Cocomplex of A-modules, 164
J(A), 121 Cohomology group, 166
K-algebra, 25 Cokernel of a homomorphism, 24
K-algebra homomorphism, 85 Column finite matrix, 53
K-linear category, 38 Complete lattice, 15
K-linear functor, 39 Complete set of pairwise orthogonal idempo-
N (∞), 40 tents, 92
N (λ), 20 Complex of A-modules, 164
U (X), 13 Component of an Auslander-Reiten quiver,
IrrA (X, Y ), 218 213
mod(A), 87 Composition factors, 66
radA (X, Y ), 217 Composition of relations, 161
rep(Q) = repK (Q), 111 Composition series, 66
Hom(M, N ), 135 Connected algebra, 103, 214
Mod(A), 135 Connecting homomorphism, 160, 172
mod(A), 135 Contravariant functor, 38
a∗ (f, g), 155 Converse Bottleneck Lemma, 195
b∗ (f, g), 156 Countably generated, 146
c-generated module, 13 Countably generated module, 14
f ⊗ g, 187 Covariant functor, 38
n-subspace problem, 116 Coxeter matrix, 257
Mod(A), 87 Crawley-Jønsson-Warfield Theorem, 145
Rep(Q) = RepK (Q), 111 Cycle (in a module category), 245
Cycle in a quiver, 221
Additive function on a translation quiver, 223 Cyclic module, 14
Additive on short exact sequences, 243
Adjoint functors, 188 Decomposable module, 18
Adjunction formula, 188 Dedekind Lemma, 16
Admissible ideal, 139 Degree of a polynomial, 41
Algebra, 25 Dense functor, 39
Algebra homomorphism, 85 Determined by composition factors, 255
Annihilator of a module, 106 Dimension of a J-module, 12
Auslander algebra, 184 Dimension of a representation, 110
Auslander-Reiten formula, 200 Dimension vector of a finite length module,
Auslander-Reiten quiver, 213, 224 67
Auslander-Reiten quiver (of M), 192 Dimension vector of a quiver representation,
Auslander-Reiten sequence (in M), 192 113
Direct complement of a submodule, 18
Balanced map, 185 Direct decomposition of a module, 18
Basic algebra, 139 Direct sum of modules, 18
Bimodule, 97 Direct sum of quiver representations, 111
Bimodule of irreducible maps, 218 Direct summand of a module, 18
Biquiver, 192 directed algebra, 261
Block of a group algebra, 103 Directing module, 245
Bottleneck Lemma, 192 Dominant dimension of an algebra, 184
Butterfly Lemma, 63 Dynkin quiver, 272
Bézout’s Theorem, 44
Endomorphism, 24
Cancellation Theorem, 81 Endomorphism algebra, 24
Cartan matrix, 262 Epimorphism, 23
279
Equivalence of categories, 39 Indecomposable module, 18

Equivalence of short exact sequences, 33 Induced short exact sequence, 155
Euclidean graphs, 272 Injective dimension, 183
Euclidean quiver, 272 Injective resolution, 134
Euler form of a quiver, 267 Injective vertex of a translation quiver, 222
Exact sequence, 31 Invertible element in a ring, 68
Exact sequence of A-modules, 164 Irreducible homomorphism (in M), 191
Exceptional representation of a quiver, 113 Irreducible module, 48
Exchange Theorem, 79 Isomorphic filtrations, 63
expansion of a translation quiver, 225 Isomorphism, 23
extending vertex of an Euclidean quiver, 269 Isomorphism of categories, 39
Isomorphism of quiver representations, 111
Factor category, 135 Isotypical component of a module, 51
Factor complex, 165
Factor module, 15 Jacobson radical of a local ring, 71
Factors of a filtration, 63 Jacobson radical of an algebra, 104
Faithful functor, 39 Jordan-Hölder multiplicity, 66
Fibre product, 152 Jordan-Hölder Theorem, 66
Fibre sum, 151
Filtration of a module, 63 Kernel of a homomorphism, 24
Finitely generated module, 13 Kronecker problem, 116
Finitely presented module, 200 Krull-Remak-Schmidt Theorem, 81
Finitistic dimension of an algebra, 184 Krull-Remak-Schmidt-Azumaya Theorem, 145
First Isomorphism Theorem, 26
Fitting Lemma, 75 Large submodule, 57
Free A-module, 88 Lattice, 15
Free generating set of a module, 88 Left A-module, 86
Full functor, 39 Left adjoint functor, 188
Full subcategory, 38 Left almost split morphism, 206
Functor, 38 Left ideal, 69
Left inverse, 68
General adjunction formula, 190 Left minimal homomorphism, 207
General Exchange Theorem, 141 Left-invertible element in a ring, 68
Generating set of a submodule, 13 Length of a composition series, 66
Global dimension, 183 Length of a filtration, 63
Grothendieck group, 243 Length of a module, 66
Group algebra, 101 Local module, 71
Local ring, 68
Harada-Sai Lemma, 76 Locally finite quiver, 221
Hasse diagram, 15 Long exact Ext-sequence, 177, 178
Hereditary algebra, 184 Long exact homology sequence, 169
Homology group, 166 Long exact Tor-sequence, 189
Homomorphism of A-modules, 87 Loop functor, 136
Homomorphism of J-modules, 23
Homomorphism of complexes, 165 Map of complexes, 165
Homomorphism Theorem, 25 Maschke’s Theorem, 101
Homotopic morphisms of complexes, 167 Maximal complement of a submodule, 58
Homotopy, 168 Maximal left ideal, 69
Maximal right ideal, 69
Ideal, 69 Maximal submodule, 48
Ideal of a category, 135 Mesh category, 223
Idempotent in a ring, 29 Mesh relation, 223
Image of a homomorphism, 24 Minimal generating set, 14
imaginary root (Euclidean case), 270 Minimal injective resolution, 134
Imaginary root of a quiver, 113 Modular lattice, 15
Module category, 39 Representation of a quiver, 110

Module homomorphism, 87 Right A-module, 91
Monic polynomial, 41 Right adjoint functor, 188
Monomorphism, 23 Right almost split morphism, 206
Morphism of quiver representations, 110 Right ideal, 69
Right inverse, 68
Nakayama functor, 202 Right minimal homomorphism, 207
Nakayama Lemma, 106 Right-invertible element in a ring, 68
Nilpotent element in a ring, 68 Rigid representation of a quiver, 113
Ring, 24
Opposite algebra, 90
Ring of rational functions, 46
Opposite relation, 160
root (Dynkin and Euclidean case), 269
Orthogonal idempotents, 92
Root of a quiver, 113
Partially ordered set (poset), 15
Partition, 20 Schanuel’s Lemma, 123
Path (in a module category), 245 Schreier’s Theorem, 63
Path algebra, 108 Schur root of a quiver, 113
Path in a quiver, 108, 221 Schur’s Lemma, 52
preinjective component, 225 Second Isomorphism Theorem, 26
preinjective module, 225 Semiperfect algebra, 148
Preinjective translation quiver, 222 Semisimple algebra, 99
preprojective component, 225 Semisimple module, 48
preprojective module, 225 Short exact sequence, 33
Preprojective translation quiver, 222 Short exact sequence of complexes, 165
Primitive idempotent, 94 Simple module, 48
Principal ideal in K[T ], 43 sincere algebra, 254
Product of ideals, 69 Sincere module, 254
Product of modules, 19 Sink map, 210
Product of rings, 53 Skew field, 52
Projective cover, 125 Small submodule, 57
Projective dimension, 183 Snake Lemma, 159
Projective module, 119 Socle of a module, 54
Projective vertex of a translation quiver, 222 Source map, 210
Proper submodule, 12 Split epimorphism, 30
Pullback, 152 Split exact sequence, 34
Pushout, 151 Split monomorphism, 30
Split valuation for a translation quiver, 224
Quiver, 107 Splitting field, 247
Quiver of an algebra, 214 Stable module category, 135
Stable translation quiver, 222
Radical factor ring, 71 Subcategory, 38
Radical of a local ring, 71 Subcomplex, 165
Radical of a module, 56 Submodule, 12
Radical of an algebra, 104, 121 Submodule generated by a set X, 13
Rank of a free module, 89 Subrepresentation of a quiver representation,
Reachable module, 245 111
Reachable vertex in a quiver, 221 Symmetric bilinear form of a quiver, 267
real root (Dynkin and Euclidean case), 270 Syzygy functor, 136
Real root of a quiver, 113
Refinement of a filtration, 63 Tensor product of algebras, 99
Regular representation of an algebra, 88 Tensor product of modules, 185
Relation, 160 Tits form of a quiver, 267
Representation dimension of an algebra, 184 Top of a module, 56
Representation of a K-algebra, 85 Torsion free module, 123
Representation of a group, 101 Translation quiver, 222
Uniform module, 41, 134

Universal property of the pullback, 153
Universal property of the pushout, 151
Valuation of a translation quiver, 224

valued graph of an hereditary algebra, 256
valued quiver of an algebra, 256
Valued translation quiver, 224
weakly positive form, 261
Young diagram, 20
Zero homotopic, 168

Zorn’s Lemma, 48
Claus Michael Ringel

Fakultät für Mathematik
Universität Bielefeld
Postfach 100131
D-33501 Bielefeld
Germany
E-mail address: [email protected]
Jan Schröer
Mathematisches Institut
Universität Bonn
Beringstr. 1
D-53115 Bonn
Germany
E-mail address: [email protected]

Claus Michael Ringel, Jan Schröer - Representation Theory of Algebras I - Modules (Preliminary Version Feb 10, 2009) (2009) PDF

Uploaded by

Copyright:

Available Formats

Claus Michael Ringel, Jan Schröer - Representation Theory of Algebras I - Modules (Preliminary Version Feb 10, 2009) (2009) PDF

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

Claus Michael Ringel, Jan Schröer - Representation Theory of Algebras I - Modules (Preliminary Version Feb 10, 2009) (2009) PDF

Uploaded by

Copyright:

Available Formats

February 10, 2009

REPRESENTATION THEORY OF ALGEBRAS I: MODULES

CLAUS MICHAEL RINGEL AND JAN SCHRÖER

Abstract. This is the first part of a planned book “Introduction to Representa-

Part 1. Modules I: J-Modules 12

2.7. Decompositons and direct sums of modules 18

3.5. Homomorphisms between direct sums 27

4.3. Equivalences of categories 39

5. Examples of infinite dimensional 1-modules 40

5.4. The module (K[T ], T ·) 43

6. Semisimple modules and their endomorphism rings 48

6.2. Endomorphism rings of semisimple modules 52

7. Socle and radical of a module 54

Part 2. Modules of finite length 63

9. Digression: Local rings 67

10.1. Some length formulas for modules of finite length 72

Part 3. Modules II: A-Modules 85

12.1. Representations of an algebra 85

12.5. Free modules 88

12.9. Direct decompositions of the regular representation 91

13.1. Semisimple algebras and their modules 99

14. The Jacobson radical of an algebra 104

15.1. Quivers and path algebras 107

15.2. Examples 108

15.4. Representations of quivers 110

16. Digression: Classification problems in Linear Algebra 115

16.4. The n-subspace problem 116

Part 4. Projective modules 119

17.3. Cyclic projective modules 123

18.2. Injective envelopes 129

22. Direct summands of infinite direct sums 139

22.3. The Crawley-Jønsson-Warfield Theorem 143

23. Projective modules over semiperfect algebras 146

Part 5. Homological Algebra I: Resolutions and extension groups 149

25.3. Properties of pushout and pullback 151

25.7. Short exact sequences with projective middle term 157

26.2. Complexes 163

26.6. The long exact homology sequence 168

27.3. Induced maps between extension groups 176

27.8. Injective resolutions 183

28.3. Selfinjective algebras 184

28.7. Auslander algebras 185

29.2. Adjoint functors 188

Part 6. Homological Algebra II: Auslander-Reiten Theory 191

30.3. Properties of irreducible homomorphisms 192

31. Auslander-Reiten Theory 199

31.3. The Nakayama functor 202

31.5. Existence of Auslander-Reiten sequences 205

31.9. The bimodule of irreducible morphisms 217

31.13. More examples of Auslander-Reiten quivers 233

33. Reachable and directing modules 245

33.4. The quiver of an algebra 256

34.2. Cartan matrix 262

35.2. Gabriel’s Theorem 271

36. Cartan matrices and (sub)additive functions 271

Part 7. Extras 274

38. Classes of algebras 275

• Representation theory of quivers and finite-dimensional algebras

• Decomposition theory for irreducible components of varieties of modules