REPRESENTATIONS OF THE KRONECKER QUIVER

ANDREW HUBERY

1. The Kronecker quiver

We consider the Kronecker quiver K : 1 2 .

A representation X = (U, V ; A, B) of K over a field k is given by a pair of (finite
dimensional) vector spaces U, V and a pair of linear maps A, B : A → B; we also
A
write this graphically as X: U V . A morphism X → X 0 between two
B
such representations if given by a pair of linear maps f : U → U 0 and g : V → V 0
such that both squares below commute
A B
U V U V
f g and f g

A0 B0
U0 V0 U0 V0

These form an abelian category denoted repk K, which is equivalent to the category
of left modules over the path algebra
 
k 0
kK :=
k2 k
There are two simple objects: the simple injective S1 = I(0) = (k, 0; 0, 0) and the
simple projective S2 = P (0) = (0, k; 0, 0). The Grothendieck group of the category
is therefore isomorphic to Z2 , with basis e1 = dim S1 and e2 = dim S2 . Given a
representation X = (U, V ; A, B) we write dim X = (dim U, dim V ) for its image in
the Grothendieck group.
The category repk K is hereditary, so Exti (X, Y ) = 0 for all X, Y and all i ≥ 2.
Thus the Euler form of the category is given by
hX, Y i := dim Hom(X, Y ) − dim Ext1 (X, Y ),
and this descends to a bilinear form on the Grothendieck group. With respect to
the standard basis this is represented by the matrix
 
1 −2
.
0 1

We also have the symmetric bilinear form (x, y) = hx, yi + hy, xi on Z2 . Thus

x = (a, b) implies (x, x) = s(a − b)2 ,

so this symmetric bilinear form is positive semi-definite, with radical generated by
δ := (1, 1).
A useful concept for a representation is then the defect ∂, where
∂(X) = hδ, dim Xi.
1
2 ANDREW HUBERY

1.1. Duality. The vector space duality D = Homk (−, k) induces a duality on the
category repk K
A D(A)
D: U V 7→ D(V ) D(U )
B D(B)

This swaps the entries of the dimension vector, and hence changes the sign of the
defect
dim X = (a, b) ⇒ dim D(X) = (b, a), and ∂(D(X)) = −∂(X).
1.2. Reflection functors. We introduce two endofunctors S ± of the category of
representations. The functor S + is given by a pull-back construction:
A0
T U
A A0
S+ : U V 7→ T U given by the pull-back B 0
B
B B0
A
U V
Dually the functor S − is given by a push-out construction:
A
U V
A A0
S− : U V 7→ V W given by the push-out B B0
B B0
0
A
V W

Lemma 1.1. We have DS + ∼

= S − D. 
Theorem 1.2. The reflection functors form an adjoint pair, so we have an iso-
morphism
Hom(S − X, Y ) ∼
= Hom(X, S + Y )
which is natural in both X and Y .
Moreover, the unit of the adjunction X → S + S − X is an epimorphism, with
kernel some number of copies of the simple injective I(0), and the counit of the
adjunction S − S + X → X is a monomorphism, with cokernel some number of copies
of the simple projective P (0).
Proof. That we have an adjoint pair follows immediately from the universal proper-
ties of the pull-back and push-out. To compute the unit, consider a representation
A A00
X: U V . Then S + S − (X) = U V 0 is given by the commutative
B B 00
diagram
A0
T U

B 00
B0 B

A 00 V0 g

U A
V
The map g is necessarily injective, so (1, g) defines a monomorphism X → S + S − X.
The cokernel is given by
0 V /V 0
which is easily seen to be isomorphic to P (0)n where n = dim V /V 0 . The result for
the counit is dual. 
Corollary 1.3. Let X be indecomposable and set dim X = (a, b).
 REPRESENTATIONS OF THE KRONECKER QUIVER 3

(1a) If X is simple projective, then S + (X) = 0.

(1b) If X is not simple projective, then
(i) S + (X) is indecomposable.
(ii) dim S + (X) = (2a − b, a).
(iii) X ∼= S − S + X.
Dually we have
(2a) If X is simple injective, then S − (X) = 0.
(2b) If X is not simple injective, then
(i) S − (X) is indecomposable.
(ii) dim S − (X) = (b, 2b − a).
(iii) X ∼= S + S − X.
We consider the full additive subcategory repk K generated by all indecompos-
ables except the simple projective P (0). Dually let repk K be the full additive
subcategory generated by all indecomposables except the simple injective I(0).
Proposition 1.4. The subcategory repk K is closed under extensions and taking
quotients; dually repk K is closed under extensions and taking subobjects. Moreover,
the reflection functors restrict to give a mutually inverse equivalances
∼ ∼
S + : repk K −
→ repk K and S − : repk K −
→ repk K.
Proof. Since P (0) is simple projective, we have
repk K = {X : Hom(X, P (0)) = 0}.
It is then clear that repk K is closed under extensions and quotients. Moreover,
repk K is the essential image of S − . For, given Y ∈ repk K we have
Hom(S − Y, P (0)) ∼
= Hom(Y, S + P (0)) = 0,
so S − Y ∈ repk k, and conversely if X ∈ repk K, then S − S + X ∼= X.
Dually, X ∈ repk K if and only if Hom(I(0), X) = 0, so this subcategory is closed
under extensions and subrepresentations, and is the essential image of S + .
It is now clear that S ± induce mutually inverse equivalences between repk K and
repk K, and so in particular preserve exact sequences. 

2. Classification of representations
By the Krull-Remak-Schmidt Theorem, every representation is a finite direct
sum of indecomposable representations in an essentially unique way. We therefore
wish to classify the indecomposable representations
L over a field k.
Consider the ring of polynomials k[s, t] = d≥0 Vd , graded according to total
degree. For convenience we also set V−1 = 0.
For d ≥ 0 we define
s
P (d) : Vd−1 Vd dim P (d) = (d, d + 1), ∂(P (d)) = −1
t

where the two maps are multiplication by s and t; we also set I(d) := D(P (d)), so
D(s)
I(d) : D(Vd ) D(Vd−1 ) dim I(d) = (d + 1, d), ∂(I(d)) = 1
D(t)

finally for 0 6= f ∈ Vd we define

s
R(f ) : Vd−1 Vd /(f ) dim R(f ) = (d, d), ∂(R(f )) = 0
t
4 ANDREW HUBERY

We will see that the P (d) and I(d) are all indecomposable, as are the R(f ) for
f a power of an irreducible polynomial, and that these yield a classification of all
indecomposable representations.

2.1. Non-zero defect.

Proposition 2.1. We have S −d (P (0)) ∼ = P (d). Moreover, if X is indecomposable
of negative defect, then X ∼
= P (d) for some d ≥ 0.
Dually, S +d (I(0)) ∼
= I(d). Moreover, if X is indecomposable of positive defect,
then X ∼= I(d) for some d ≥ 0.
Proof. It is easy to see that we have a push-out diagram
s
Vd−1 Vd
t t

s
Vd Vd+1

so we have S − (P (d)) = P (d + 1). Dually S + (I(d)) = I(d + 1).

Now suppose that X is indecomposable, and set dim X = (a, b). Assume ∂(X) <
0, so a < b. If X is simple projective, then X ∼ = P (0). Otherwise S + (X) is again
+
indecomposable and dim S (X) = (2a − b, a). Thus 2a − b < a < b, so by induction
on the dimension vector we must have S + (X) ∼ = P (d) for some d ≥ 0. It follows
that X ∼= S − S + (X) ∼
= S − (P (d)) = P (d + 1). Dually for ∂(X) > 0. 

Proposition 2.2. For e ≥ 0 we have Hom(P (d), P (d + e)) ∼

= Ve , and every non-
zero homomorphism is injective.
Proof. The result is clear when d = 0, and using the reflection functor S −d we have
that
Hom(P (d), P (d + e)) ∼
= Hom(P (0), P (e)) ∼= Ve .
Moreover, if f ∈ Ve , the corresponding homomorphism P (0) → P (e) is given by
multiplication by f ; applying S −d we then see that the induced homomorphism
P (d) → P (d + e) is also given by multiplication by f . In particular, if f is non-zero,
then this homomorphism is injective. 

Corollary 2.3. We have

Hom(P (e + 1), P (d)) = 0 and Ext1 (P (d), P (e)) = 0 for all e ≥ d.
Proof. Take a homomorphism θ : P (e) → P (d). Composing with an injective map
f : P (d) → P (e) we obtain an endomorphism f θ of P (e), so necessarily a scalar. It
cannot be an isomorphism for dimension reasons, so has to be zero. Thus θ = 0.
For the second statement we know that dim Ext1 (P (d), P (e)) equals
dim Hom(P (d), P (e)) − hP (d), P (e)i = (e − d + 1) − (e − d + 1) = 0. 

We next want to compute homomorphisms and extensions between the repre-

sentations P (d) and R(f ). For this, we first observe that the representations R(f )
are unchanged by the reflection functors.
Lemma 2.4. Let f ∈ Vn be non-zero. Then S ± R(f ) ∼
= R(f ), and hence we have
a short exact sequence
f
0 → P (d) −
→ P (d + n) → R(f ) → 0.
 REPRESENTATIONS OF THE KRONECKER QUIVER 5

∼
Proof. Fix any vector space isomorphism θ : Vd−1 − → Vd /(f ), and set α := θ−1 ◦
−1
s and β := θ ◦ t. It follows that R(f ) is isomorphic to the representation
α
Vd−1 Vd−1 , and hence S ± (R(f )) ∼
= R(f ) follows from the push-out (and
β
pull-back) diagram
α
Vd−1 Vd−1
β β 
α
Vd−1 Vd−1

It follows that we can realise R(f ) as the representation

s
R(f ) ∼
= Vd+n−1 /(f Vd−1 ) Vd+n /(f Vd )
t

Corollary 2.5. For f ∈ Vn non-zero and m ≥ 0 we have

Hom(P (d), R(f )) ∼

= Vn+m /(f Vm ),

whereas
Hom(R(f ), P (d)) = 0 and Ext1 (P (d), R(f )) = 0.

Proof. Consider the short exact sequence

f
0 → P (d + m) −
→ P (d + m + n) → R(f ) → 0.

Applying Hom(P (d), −) and using that Ext1 (P (d), P (d + m)) = 0 yields the short
exact sequence
f
0 → Vm −
→ Vm+n → Hom(P (d), R(f )) → 0.

Now use the Euler form to compute that Ext1 (P (d), R(f )) = 0.
On the other hand, applying Hom(−, P (d)) yields that

Hom(R(f ), P (d)) ≤ Hom(P (d + n), P (d)) = 0. 

Finally we compute homomorphisms and extensions between P (d) and I(e).

Lemma 2.6. We have

Hom(P (d), I(e)) ∼

= D(Vd+e−1 ) and Ext1 (P (d), I(e)) = 0.

Also Hom(I(e), P (d)) = 0 and dim Ext1 (I(e), P (d)) = d + e + 2.

Proof. Applying S −e we have

Hom(P (d), I(e)) ∼

= Hom(P (d + e), I(0)) ∼
= D(Vd+e−1 ).

Similarly, it is easy to check that Hom(I(0), P (d)) = 0 for all d. Then by the Euler
form we have

dim Ext1 (P (d), I(e)) = dim Hom(P (d), I(e)) − hP (d), I(e)i = (d + e) − (d + e) = 0,

and similarly dim Ext1 (I(e), P (d)) = d + e + 2. 

6 ANDREW HUBERY

2.2. Zero defect. We now want to describe the indecomposable representations of

defect zero. We will show that if f is irreducible, then R(f m ) is indecomposable for
all m ≥ 1; moreover any indecomposable representation of defect zero is isomorphic
to one of this form (with the irreducible polynomial uniquely determined up to
scalar). More generally, if f factors up to scalar as pm mr
1 · · · pr , where the pi are
1

irreducible and pairwise coprime, then

R(f ) ∼
= R(pm mr
1 ) ⊕ · · · ⊕ R(pr ).
1

Denote by rep0k K the full additive subcategory generated by the indecomposable

representations of defect zero, and by rep0,f
k K for f irreducible the full additive
subcategory generated by all the R(f m ). We will show that rep0k K and each
rep0,f 1
k K are thick abelian subcategories of repk K, and

rep0k K ∼
a
= rep0,f
k K,

the coproduct being taken over all irreducible homogeneous polynomials up to

scalar.
Finally we will show that for f irreducible we have

rep0,f ∼
k K = mod Of ,
b

where O bf is a complete DVR over k, depending on f . We can write Of as the

degree zero part of the graded localisation k[s, t](f ) . Explicitly, if s does not divide
f , then this is isomorphic to k[u](f (1,u)) , whereas if t does not divide f , then it is
isomorphic to k[u]f (u,1) .

2.2.1. Modules over k[u]. We begin by reviewing the module theory for the principal
ideal domain k[u]. A finite-dimensional module is determined by a pair (V ; φ),
where V is a finite-dimensional vector space, and φ ∈ End(V ) gives the action on
u. We can regard such pairs as a k-representation for the Jordan quiver

Q: ·

and in fact we obtain an equivalence (even an isomorphism) of categories

mod k[u] ∼
= repk Q.

On the other hand, the structure theorem for finitely-generated modules over a
principal ideal domain implies that every finite-dimensional indecomposable k[u]-
module is isomorphic to k[u]/(pn ) for some monic irreducible polynomial p. In
this case the corresponding representation of the Jordan quiver has vector space
k[u]/(pn ) and endomorphism corresponding to multiplication by u. More generally,
if f is any monic polynomial, then we can factorise f = pm mr
1 · · · pr
1
into a product
of distinct monic irreducible polynomials p1 , . . . , pr , in which case the cyclic module
k[u]/(f ) is isomorphic to the direct sum

k[u]/(f ) ∼
= k[u]/(pm mr



1 ) ⊕ · · · ⊕ k[u]/(pr ) .
1

1 A thick abelian subcategory is one which is closed under kernels, cokernels and extensions.
 REPRESENTATIONS OF THE KRONECKER QUIVER 7

If we take the basis {1, u, u2 , . . . , ud−1 } where deg f = d, then multiplication

by u is represented by the companion matrix C(f ).
If f = g m with deg g = r, then we may take the basis
{ui g j : 0 ≤ i < r, 0 ≤ j < m},
in which case multiplication by u is represented by a matrix in block form,
with C(g) on the diagonal and the elementary matrix E1r on the lower
diagonal. In particular, if g = u − λ is linear, then this specialises to the
(transpose of the) Jordan matrix Jm (λ).

Thus the structure theorem can be expressed in the following form.

Theorem 2.7. We have an equivalence of categories

mod k[u] ∼
a
= mod Obf

where the product is taken over all monic irreducible polynomials f ∈ k[u], and
Of := k[u](f ) is a DVR.

If the residue field κ(f ) := k[u]/(f ) is separable over k, then the Cohen
Structure Theorem tells us that O bf ∼= κ(f )[[u]]. In general we always have
such an isomorphism as rings, but when the residue field is inseparable over
k, the k-algebra structure is not the obvious one coming from k → κ(f ) →
κ(f )[[u]].

2.2.2. Two exact embeddings.

Proposition 2.8. We have exact embeddings
φ
F0 : mod k[u] → rep0k K, (V ; φ) 7→ V V
1

and
1
F∞ : mod k[u] → rep0k K, (V ; φ) 7→ V V
φ

In particular, F0 (k; 0) ∼
= R(s) and F∞ (k; 0) ∼
= R(t).
Proof. Write F for either F0 or F∞ . It is straightforward to check that F is a
fully-faithful and exact functor from mod k[u] to repk K. It remains to show that
its image lies in the subcategory rep0k K.
Clearly F (V ; φ) cannot have any I(d) as a direct summand, since the two linear
maps D(s) and D(t) used in I(d) have non-trivial kernel. Since F (V ; φ) has zero
defect and has no summand of positive defect, it also cannot have a summand of
negative defect, so F (V ; φ) ∈ rep0k K. 

It follows from the proposition that the essential image of F0 is the full subcat-
A
egory consisting of those representations U V such that B is an isomor-
B
phism; it is then clear that this is a thick abelian subcategory of repk K. Similarly
for F∞ .
Lemma 2.9. Let f ∈ Vn be non-zero, and assume that t does not divide f . Set
f¯ := f (u, 1) ∈ k[u]. Then the evaluation map ev(u,1) : k[s, t] → k[u] induces an
8 ANDREW HUBERY

isomorphism
s
R(f ) Vn−1 Vn /(f )
t
o ev(u,1) ev(u,1)

u
F0 (k[u]/(f¯)) k[u]/(f¯) k[u]/(f¯)
1

Similarly, if s does not divide f , then the evaluation ev(1,u) induces an isomorphism
R(f ) ∼
= F∞ (k[u]/(f (1, u)). 
Lemma 2.10. Let X be indecomposable of defect zero. Then X lies in the essential
image of either F0 or F∞ .
A
Proof. Write the representation X as V V . We need to show that one of
B
A or B is an isomorphism. Assume therefore that A is not an isomorphism, and
take 0 6= v ∈ Ker(A). If Bv = 0, then we can use v to define a monomorphism
S1 → X. Since S1 is injective, this must be a split monomorphism, contradicting
the fact that X is indecomposable of defect zero. Thus Bv 6= 0 and we have a
monomorphism
0
R(t) k k
1
v Bv
A
X V V
B

Let Y be the cokernel, which again has zero defect. Moreover, every indecom-
posable summand of Y has zero defect, so Y ∈ rep0k K. For, we know that
Ext1 (P (d), R(t)) = 0, so X indecomposable implies that Y has no summand of
negative defect, and hence also no summand of positive defect.
A0
Consider an indecomposable summand of the cokernel, say X 0 : V 0 V0 .
B0
Suppose that A0 is an isomorphism. Then it is clear that Hom(X 0 , R(t)) = 0,
and so by the Euler form also Ext1 (X 0 , R(t)) = 0. Thus the split monomorphism
X 0 → Y lifts to a monomorphism X 0 → X, and hence X 0 is a direct summand of
X, a contradiction. By induction on dimension vector we deduce that X 0 lies in the
essential image of F0 . Since this is true for every direct summand of Y , we deduce
that Y , and hence also X, lies in the essential image of F0 . 

Theorem 2.11. We have

rep0k K ∼
a
= mod O
bf

where the product is taken over all irreducible homogeneous polynomials up to scalar,
and that this is a thick abelian subcategory of repk K.
2.2.3. Duality. We next show that D(R(f )) ∼ = R(f ) for all homogeneous polyno-
mials f . To do this, we recall that a finite-dimensional commutative k-algebra A
is called a Frobenius algebra provided there is a linear functional π : A → k whose
kernel contains no non-zero ideal of A. Each Frobenius algebra is self-injective; in
fact, the map π induces an isomorphism of A-modules
∼
A−
→ D(A), a 7→ (b 7→ π(ab)).
Lemma 2.12. Let f ∈ k[u] be a monic irreducible polynomial. Then k[u]/(f m+1 )
is a Frobenius algebra.
 REPRESENTATIONS OF THE KRONECKER QUIVER 9

Proof. Consider the basis A given by ea,b := f a ub for 0 ≤ a ≤ m and 0 ≤ b < deg f .
We take π = δm,0 to be the dual basis element corresponding to em,0 . Now take
0 6= g ∈ k[u]/(f m+1 ). Write g = f a ḡ with ḡ 6∈ (f ). Since k[u]/(f ) is a field, we can
find h̄ such that ḡ h̄ = 1 ∈ k[u]/(f ). Now set h = f m−a h̄, so that gh = f m and
hence π(gh) = 1. Thus Ker(π) cannot contain any non-zero ideal. 
Proposition 2.13. We have D(R) ∼
= R for all R ∈ rep0k K.
Proof. It is enough to prove this for every indecomposable. By Lemma 2.9 we
can pass to the indecomposable k[u]-module k[u]/(f m ), where f ∈ k[u] is monic
irreducible, and by the previous lemma we know that this is isomorphic to its
dual. 
Corollary 2.14. For each non-zero homogeneous polynomial f ∈ Vd we have a
short exact sequence
D(f )
0 → R(f ) → I(d + e) −−−→ I(e) → 0.
Proof. Apply the duality to the short exact sequence
f
0 → P (e) −
→ P (d + e) → R(f ) → 0. 
2.2.4. Computation of homomorphisms. Since we have decomposed K into a rep0k
coproduct, it is easy to compute homomorphisms between indecomposables, and
hence between arbitrary representations. The following version is still useful, how-
ever.
Lemma 2.15. Let f, g be homogeneous, and write h = gcd(f, g) ∈ Vn . Then
Hom(R(f ), R(g)) ∼= Vn /(h) ∼
= Ext1 (R(f ), R(g)).
Proof. Let f and g have degrees d and e respectively, giving the short exact sequence
f
0 → P (0) −
→ P (d) → R(f ) → 0.
Applying Hom(−, R(g)) and using Corollary 2.5 we have the map
f
Ve /(g) ∼ → Hom(P (0), R(g)) ∼
= Hom(P (d), R(g)) − = Vd+e /(gVd ),
whose kernel is Hom(R(f ), R(g)) and whose cokernel is Ext1 (R(f ), R(g)).
Now write f = f¯h and g = ḡh. Then the kernel is ḡVe−n /(g) and the cokernel is
Vd+e /(h(f¯Ve−n + ḡVd−n )) ∼
= Vd+e /(hVd+e−n ), where we have used that f¯ and ḡ are
coprime. Finally, for all m we have an isomorphism Vn /(h) ∼ = Vm+n /(hVm ). 
2.3. Computations of some extensions.
Lemma 2.16. Let f ∈ Vd and g ∈ Ve be coprime. If n ≥ d + e, then we have a
short exact sequence
g
(−f ) (f,g)
0 P (n − d − e) P (n − d) ⊕ P (n − e) P (n) 0

whereas if n < d + e, then we have a short exact sequence

(f,g)
0 P (n − d) ⊕ P (n − e) P (n) I(d + e − n − 1) 0

Proof. Suppose first that n = d+e. Since f and
g are coprime, the map (f, g) : P (e)⊕
g
P (d) → P (d + e) has kernel spanned by −f , as required. Applying S − yields the
first sequence when n ≥ d + e.
Suppose instead that d + e = n + 1. Then the map (f, g) : P (e − 1) ⊕ P (d − 1) →
P (d + e − 1) is injective, and the cokernel has dimension vector (1, 0), so must be
isomorphic to I(0). Applying S + yields the second sequence when n < d + e. 
10 ANDREW HUBERY

Proposition 2.17. We have

Ext1 (I(e), P (d)) ∼
= Vd+e+1 .
In fact, for each non-zero f ∈ Vd+e+1 , the corresponding extension is of the form
0 → P (d) → R(f ) → I(e) → 0.
Proof. Applying S −e , we may assume that e = 0. Now consider the projective
resolution
(s,t)
0 P (0)2 P (1) I(0) 0.

Applying Hom(−, P (d)) yields the short exact sequence

(st)
0 Vd−1 Vd2 Ext1 (I(0), P (d)) 0,

so that Ext1 (I(0), P (d)) ∼

= Vd+1 , and this isomorphism is given explicitly by forming
the push-out along P (0)2 → P (d).
Suppose first that the field k is infinite. Given g, h ∈ Vd , set f := gt − hs. Take
p ∈ V1 coprime to f , write p = at − bs with a, b ∈ k and set q := bg − ah. If f 6= 0,
then we have a commutative square
(gs ht )
P (0)2 P (1) ⊕ P (d)
(a,b) (q,p)
f
P (0) P (d + 1)

Note that if p divides q, then it necessarily divides af and bf , so divides f , a

contradiction. Thus each vertical map is surjective by the previous lemma, and has
kernel P (0). Moreover, the induced endomorphism of P (0) is non-zero, so is an
automorphism. It follows that the cokernel of the top horizontal map is isomorphic
to the cokernel R(f ) of the bottom horizontal map.
Suppose instead that the field k is finite, so there are irreducible polynomials of
arbitrarily high degree. Take a homogeneous irreducible polynomial p ∈ Vn+1 with
n > d and write p = at − bs for some a, b ∈ Vd . Given g, h ∈ Vd , set q := bg − ah
and f := gt − hs. If f 6= 0, then we have a commutative square
(gs ht )
P (0)2 P (1) ⊕ P (d)
(a,b) (q,p)
f
P (n) P (d + n + 1)

Moreover, since p is irreducible, a and b must be coprime. Also, if p divides q, then it

necessarily divides af and bf , so divides f , a contradiction since n > d. Thus p and
q are also coprime. From the previous lemma we deduce that the cokernel of each
f
vertical map is I(n−1). Moreover, the composition P (n) − → P (n+d+1) → I(n−1)
does not vanish, since it cannot factor through (q, p) : P (1) ⊕ P (d) → P (n + d + 1).
Thus the induced endomorphism of I(n−1) is non-zero, so is an automorphism. We
conclude that the cokernel of the top horizontal map is isomorphic to the cokernel
R(f ) of the bottom horizontal map.
In all cases we have shown that the push-out of the projective resolution for I(0)
along the map (g, h) : P (0)2 → P (d) is isomorphic to R(gt − hs). 

This result also yields the important factorisation result.

 REPRESENTATIONS OF THE KRONECKER QUIVER 11

Corollary 2.18. Let f be irreducible. Then every homomorphism P (d) → I(e)

factors through some R(f m ).
Proof. Consider the short exact sequence
0 → P (d) → R(f m ) → I(e0 ) → 0,
where e0 depends on m. For m, and hence e0 , sufficiently large we know that
Ext1 (I(e0 ), I(e)) = 0, and so the map
Hom(R(f m ), I(e)) → Hom(P (d), I(e))
is surjective.