THE KNITTING ALGORITHM FOR AD TYPE QUIVERS

NEELAM VENKATA PRASAD AKULA

Abstract. We present an expository note on the theory of quiver represen-

tations along with a description of the knitting algorithm, which is used to
compute the Auslander-Reiten quiver ΓQ for AD type quivers. We focus on
constructing ΓQ utilizing the Coxeter matrix Φ, highlighting the benefit of this
method with regards to direct computation.

Contents
1. Introduction 1
2. Quiver Representations 2
2.1. Representations of Quivers 2
2.2. The Category rep Q 5
3. The Auslander-Reiten Quiver ΓQ 8
4. The Knitting Algorithm 9
4.1. Coxeter Matrix Φ 11
4.2. Examples of the Knitting Algorithm 12
References 17

1. Introduction
A quiver is a directed graph, defined using a tuple of the form Q = (Q0 , Q1 , s, t)
where Q0 is a set of vertices, Q1 a set of arrows between vertices, and s, t : Q1 → Q0
are mappings which map arrows to their source and target vertex respectively. First
introduced in [Gab72], quivers have integrated themselves into various research
fields across mathematics. From areas in complex geometry and mathematical
physics [CG97, Soi19, Gin09] to more combinatorial flavored topics [DW11, FM+ 17]
and recently even machine learning [GW22, AJ21].
Introduced in [ARS97], Auslander-Reiten theory studies the representation of Ar-
tinian rings amongst other things. One tool utilized in this theory is the Auslander-
Reiten quiver, denoted as ΓQ , a map which allows the contemporary mathematician
to traverse the category of representations of a quiver rep Q. The goal of this paper
is to explain the knitting algorithm, focusing on the construction of ΓQ utilizing
the Coxeter functor and matrix Φ.
Fortunately quivers, as they are presented in this paper, are relatively easy to
understand. Requiring only a familiarity with linear algebra. Further material on
quiver representations and Auslander-Reiten theory can be found in [Bri08, Uni12,

2010 Mathematics Subject Classification. Primary 16G70; Secondary 16G20.

Key words and phrases. Auslander-Reiten theory, Auslander-Reiten quiver, representation the-
ory of finite-dimensional algebras.
1
2 NEELAM VENKATA PRASAD AKULA

DW17, DW05], we highlight [Sch14] as much of the material in this paper was
derived from it.
In this paper we take k to be an arbitrary algebraically closed field, for intuitive
purposes it often suffices to think of k as C. For a quiver Q, we label the vertices
with positive integers and the arrows with greek letters.
In section 2 we introduce quiver representations and some initial results provid-
ing examples to illustrate the ideas more readily, also included is a brief detour
to categories and finite-dimensional algebras providing further context into what
motivates quiver theory in subsection 2.2. Section 3 introduces the Auslander-
Reiten quiver ΓQ formally, and illustrates how one can traverse it. Lastly, section
4 discusses the aforementioned knitting algorithm.

2. Quiver Representations
We begin with an example of an elementary quiver, before remarking on some
properties of quivers setting the stage for the rest of this section.
Example 2.1. Let Q = (Q0 , Q1 , s, t) be the quiver as shown below.

3
β1 β2

α
1 2 5
γ2 γ1
4
It follows that Q0 = {1, 2, 3, 4, 5}, Q1 = {α, β1 , β2 , γ1 , γ2 }, and s(γ1 ) = 5 and
t(γ1 ) = 4.
Quivers can be quite expressive, we allow for (self) loops and multiple arrows
between the same two vertices. If Q0 and Q1 are both finite sets, then we say that
Q itself is finite. For this paper we assume all quivers discussed are finite. Other
ideas from graph theory (connected, cycle, paths) apply to quivers as well, for a
primer on graph theory we recommend [Har11].

2.1. Representations of Quivers. The broad goal of representation theory is to

take an algebraic structure and use linear algebra to gain a new perspective of sorts.
The motivation for this arises since linear algebra is a well-studied and understood
field. By using representation theoretic techniques we can gain, at the very least,
a new perspective on the particular problem at hand, often resulting in a stronger
understanding.
Remark. A large branch of representation theory is the representation theory of
finite groups, in this subject a representation is a group homomorphism of the form
ρ : G → V where V = GLm (k) is an m-dimensional k-vector space. Quivers hold
use in this subject by means of the McKay quiver, a quiver which encodes the data
of a group representation in its structure, introduced in [McK80].
For the representations of quivers, the idea is quite simple, vertices become vector
spaces and arrows become linear maps. Formally, we define a representation of a
quiver as follows.
 THE KNITTING ALGORITHM FOR AD TYPE QUIVERS 3

Definition 2.2. Let Q = (Q0 , Q1 , s, t) be a quiver, a representation

M = (Mi , ϕα )i∈Q0 ,α∈Q1
is a collection of k-vector spaces Mi one for each vertex, and linear maps
ϕα : Ms(α) → Mt(α)
one for each arrow.
A representation M is finite-dimensional if each vector space Mi is also finite-
dimensional. In this case we define the dimension vector of M to be dimM =
(dim Mi )i∈Q0 an n-tuple of the dimensions of each vector space in M where n is
the number of vertices in Q. An element of M is another n-tuple defined as (mi )i∈Q0
with mi ∈ Mi .
Definition 2.3. Let Q be a quiver with two representations M = (Mi , ϕα ) and
N = (Ni , ψα ). A morphism of representations f : M → N is a collection of linear
α
maps f = (fi ), fi : Mi → Ni such that for each arrow i −
→ j ∈ Q1 the following
diagram commutes.
ϕα
Mi Mj

fi fj

ψα
Mi′ Mj′

That is, fj ◦ ϕα (m) = ψα ◦ fi (m) for all m ∈ Mi .

We now introduce direct sums.
Definition 2.4. Let Q be a quiver with representations M = (Mi , ϕα ) and N =
(Ni , ψα ). Then,
󰀕 󰀗 󰀘󰀖
ϕα 0
M ⊕ N = Mi ⊕ N i ,
0 ψα i∈Q ,α∈Q0 1

is another representation of Q called the direct sum of M and N .

If M ∕= 0 is a representation of Q, then M is said to be indecomposable if M
cannot be written as the direct sum of two non-zero representations. Indecompos-
able representations are analogous to primes to the integers, acting as a building
block for all other representations.
The following example, adapted from [Sch14, Example 1.7], illustrates the three
previous concepts.
Example 2.5. Suppose we have Q as in ??. We first define two representations of
Q as follows;
T S
L: k2 k2 k

1 0
M: k k 0
4 NEELAM VENKATA PRASAD AKULA

with maps T and S given as

󰀗 󰀘 󰀗 󰀘
2 0 1
T = = 2I S= .
0 2 1
We can now consider a morphism of representations f : L → M , illustrated below
as dashed arrows.
T S
L k2 k2 k

f f1 f2 f3

1 0
M k k 0
󰀅 󰀆
It follows
󰀅 then,
󰀆 that f1 : k 2 → k is of the form a b , likewise f2 : k 2 → k has
form c d , and lastly f3 : k → 0 is just the zero map. To preserve commutativity
we then get the relations a + b = c + d = 0 and thus b = −a and d = −c. However
we also have the added condition that a = 2c, this follows from the need to have
the diagram commute, as T = 2I and we need to preserve the commutativity.
Since one choice of scalar a ∈ k completely determines the morphism we see that
Hom(L, M ) ∼ = k.
Now consider a new representation N ∼ = L ⊕ M , first we compute L ⊕ M
󰀵 󰀶 󰀵 󰀶
2 0 0 1 0
󰀹 󰀺 󰀹 󰀺
󰀹0 2 0󰀺 󰀹1 0󰀺
󰀷 󰀸 󰀷 󰀸
0 0 1 0 0
L⊕M: k2 ⊕ k k2 ⊕ k k⊕0 .
Thus, we see that N is the representation
󰀵 󰀶 󰀵 󰀶
2 0 0 1
󰀹 󰀺 󰀹 󰀺
󰀹0 2 0󰀺 󰀹 1󰀺
󰀷 󰀸 󰀷 󰀸
0 0 1 0
N: k3 k3 k .
A broad goal of representation theory (of quivers) is to classify all representations
of a quiver Q and morphisms between them up to isomorphism. The well-known
Krull-Schmidt theorem shows that it suffices to classify the indecomposable repre-
sentations and the morphisms between them. We list the theorem here, but omit
the proof. Proving existence is straightforward but uniqueness is less so, requiring
a series of prior results, we reference [Hun12] for a complete proof.
Theorem 2.1 (Krull-Schmidt). Let Q be a quiver and M a representation of Q,
then
M∼ = M1 ⊕ M2 ⊕ · · · ⊕ Mt
where the Mi are indecomposable representations of Q that are unique up to order.
The last notion on representations we need to cover are projective, injective, and
simple representations. Their use will come up in the later sections, but we will
define them here. Formally, we have to define the dimensions of each vector space
in the collection and then define the linear maps. In our case, we focus strictly on
the dimension of each vector space, omitting the details of the maps as they are not
 THE KNITTING ALGORITHM FOR AD TYPE QUIVERS 5

necessarily relevant towards the Auslander-Reiten quiver. Let Q be the following

quiver.

1 3 4
(2.1)
2
A simple representation S(i) of Q is determined by having each vector space in
its collection be 0 except for the ith vertex, which is of one dimension. That is,
󰀫
k j=i
S(i)j := .
0 j ∕= i
It follows then that the dimension vector of S(i) is equal to 0 everywhere and 1 at the
ith index, e.g. if Q is as above then S(3) ∈ rep Q is such that dimS(3) = (0, 0, 1, 0).
A projective representation P(i) of Q is determined by the number of paths (see
Definition 2.8) from vertices i to j, for all vertices j. It may be easier to consider
a working example, let P(1) ∈ rep Q. Then the dimension of P(1)1 is just 1 as
there is just the trivial path going from vertex 1 → 1. The dimension of P(1)2 is
likewise just 1 as there is only one path from 1 → 2. The dimension of P(1)3 is 2
since there are two paths from 1 → 3; the first being 1 → 3 and the second being
1 → 2 → 3. The dimension of P(1)4 is 3 since there are three paths from 1 → 4;
the first 1 → 4, second 1 → 3 → 4, and last 1 → 2 → 3 → 4. We see then that
dimP(1) = (1, 1, 2, 3).
An injective representation I(i) of Q is similar to projective representations, but
instead determined by the number of paths j to i, for all vertices j. We consider
I(4) ∈ rep Q. The dimension of I(4)1 is 3 since there are three paths from 1 → 4,
the ones described above for P(1)4 . The dimension of I(4)2 is 1 since there is just
the path 2 → 3 → 4. Likewise, the dimensions of I(4)3 and I(4)4 are both just 1.
So dimI(4) = (3, 1, 1, 1).
Remark. If S(i) = P(i) (resp. S(i) = I(i)) then Q has a sink (resp. source) at
vertex i.
We list all S, P, and I representations in Figure 1. Moreover, each S(i), P(i),
or I(i) are all indecomposable.
We further note that it suffices to describe representations in terms of its cor-
responding dimension vector, this allows us to quickly present a representation
without having to draw it graphically or explicitly define its linear maps.
2.2. The Category rep Q. Without detailing too much ring theory we recall a few
foundational definitions.
Definition 2.6. Let A be a k-vector space equipped with an additional binary
operation · : A × A → A. Then A is a k-algebra if for all a, b, c ∈ A and λ, µ ∈ k:
(1) (a + b) · c = a · c + b · c,
(2) c · (a + b) = c · a + c · b,
(3) (λa) · (µb) = (λµ)(a · b).
Alternatively, one can think of A as a ring with unity such that A has a k-vector
space structure.
6 NEELAM VENKATA PRASAD AKULA

i S(i) P(i) I(i)

k 0 0 k k2 k3 k 0 0
1
0 k 0

0 0 0 0 k k k 0 0
2
k k k

0 k 0 0 k k k2 k 0
3
0 0 k

0 0 k 0 0 k k3 k k
4
0 0 k

Figure 1. S, P, and I representations for Q as in (2.1).

Definition 2.7. Let R be a ring with unity, a (right) R-module M is an abelian

group equipped with a binary operation called a (right) R-action
M ×R→M (m, r) 󰀁→ mr
such that for all m, n ∈ M and r, s ∈ R:
(1) (m + n)r = mr + nr,
(2) m(r + s) = mr + ms,
(3) m(rs) = (mr)s,
(4) m1 = m.
Fix a quiver Q, the set of all representations of Q over our fixed field k is denoted
as rep Q. Informally, a category is a collection of objects and morphisms between
objects such that two properties are satisfied; the first being associativity, and the
second being identity. In the case of rep Q our objects are representations of Q and
the morphisms between objects are morphisms between representations.
To give further context into why quivers are a valuable field of study we must
define the path algebra of Q. We first define the natural notion of a path on Q and
then the path algebra kQ.
Definition 2.8. Let Q = (Q0 , Q1 , s, t) be a quiver with i, j ∈ Q0 . A path c from i
to j with length l in Q is a sequence
c = (i | α1 , α2 , . . . , αl | j)
where αh ∈ Q1 such that s(α1 ) = i, s(αh ) = t(αh−1 ) for 2 ≤ h ≤ l, and t(αl ) = j.
Definition 2.9. Let Q be a quiver, the path algebra kQ of Q is a k-algebra whose
basis is the set of all paths in Q and multiplication is defined on two basis elements
c and d as follows 󰀫
c · d s(d) = t(s)
cd =
0 s(d) ∕= t(s)
Where c·d is the concatenation of paths, i.e. if c = (i | {α} | j) and d = (j | {β} | k)
then c · d = (i | {α}, {β} | k).
 THE KNITTING ALGORITHM FOR AD TYPE QUIVERS 7

An 1 2 3 ··· n−1 n

Dn 1 2 3 ··· n−2 n−1

E6 1 2 3 4 5

6
E7 1 2 3 4 5 6

7
E6 1 2 3 4 5 6 7

Figure 2. Dynkin diagrams An , Dn and E6,7,8 .

This then sets us up for the following theorem.

Theorem 2.2. The category of finite-dimensional representations, rep Q, is equiv-
alent to the category of finite-dimensional (right) kQ-modules, mod kQ.
The proof of this is quite straightforward, but does require a somewhat intimate
understanding of category theory. The idea is to construct two functors F and G
and prove that their composition is the desired identity. An often cited proof can
be found in [CB92].
This result, is often paired with the well-celebrated Gabriel’s theorem. We first
note that an underlying Dynkin diagram of a quiver Q, denoted ∆Q , is achieved
by removing arrows from Q and replacing them with edges, i.e. removing any
orientation of arrows. We list the relevant Dynkin diagrams in Figure 2.
Theorem 2.3 (Gabriel 1972). Let Q be a connected quiver.
(1) Then Q is of finite representation type if and only if the underlying Dynkin
diagram, ∆Q , of Q is of Dynkin type A, D, or E.
(2) If Q has underlying Dynkin type A, D, or E then the dimension vector
induces a bijection from isoclasses of indecomposable representations of Q
to the set of positive roots Φ (of the relevant quadratic form):
Ψ : ind Q → Φ+ Ψ : M 󰀁→ dimM.
The implications of Gabriel’s theorem are quite profound, it bridges the study
of finite-dimensional algebras to that of quivers. Furthermore, the set of indecom-
posable modules of kQ are in a one-to-one correspondence with the set of indecom-
posable representations of Q, we explicitly note that simple representations map
to simple modules and the same applies towards projective and injective represen-
tations. Gabriel’s theorem sparked much interest in quivers and has led to the
development of the field as a whole. It directly led to Kac’s theorem [Kac82], which
extended itself to the study of Lie algebras. Once again, we omit a proof here as
it is quite lengthy and requires some more background on algebraic varieties and
8 NEELAM VENKATA PRASAD AKULA

quadratic forms. There are many expository notes which lead up to the proof of
Gabriel’s theorem [Bri08, Cum11, Len19, Hal21, Aku22].

3. The Auslander-Reiten Quiver ΓQ

As we have seen, by the Krull-Schmidt theorem, the goal of representation the-
ory is to classify all indecomposable representations and the morphisms between
them. The so-called Auslander-Reiten quiver ΓQ gives an approximation of rep Q,
and in the case of Q being Dynkin type A, D, or E (that is when Q has finite rep-
resentation type) provides complete information about rep Q. Broadly, ΓQ encodes
three aspects of rep Q; indecomposable representations, “irreducible” morphisms,
and exact sequences. In this paper we omit any detailed mention of exact sequences
(and almost split sequences, see [AR75]) as they require more background in homo-
logical algebra. For further detail on Auslander-Reiten quivers see [Zel05]. Lastly,
for the duration of this paper we assume that Q is either Dynkin type A or D,
unless otherwise stated.
For a fixed quiver Q, ΓQ is a quiver whose vertices are the indecomposable
representations in rep Q and the arrows between vertices are the irreducible mor-
phisms (informally, these are morphisms between indecomposable representations
which cannot be factored through another representation). We present the following
adapted from [Sch14, Example 1.14] and [Zel05, Example 1.3].
Example 3.1. Suppose we have the same elementary quiver Q as in Example
2.5. Note that ∆Q = A3 , and thus ΓQ will provide full information about rep Q.
One can directly compute the isoclasses of indecomposable representations in rep Q
using the technique discussed in Section 2. We see that there are precisely six such
representations;

S(1) : k 0 0 S(2) : 0 k 0

S(3) : 0 0 k I(2) : k k k

P(1) : k k 0 P(3) : 0 k k
The Auslander-Reiten quiver is of the form:

P(3) S(1)

ΓQ : S(2) I(2)

P(1) S(3)

For representations L, M , and N in Example 2.5 we get the decompositions;

L∼
= P(1) ⊕ I(2), M∼
= P(1), N∼
=L⊕M ∼
= P(1) ⊕ I(2) ⊕ P(1).
As another example, consider Q is such that ∆Q = An with left-to-right arrow
orientation, i.e.
Q = 1 → · · · → n.
 THE KNITTING ALGORITHM FOR AD TYPE QUIVERS 9

Ringel [Rin96, Appendix 2] presents the Auslander-Reiten quiver of Q, shown in

Figure 3. Here, Mi,j+1 := S(i) ⊕ S(i + 1) ⊕ · · · ⊕ S(j). We also have the set of
projective and injective representations given as;
P = {M1,n+1 , M2,n+1 , M3,n+1 , . . . , Mn−1,n+1 , Mn,n+1 } ,

I = {M1,n+1 , M1,n , M1,n−1 , . . . , M1,4 , M1,3 , M1,2 } .

Naturally, we see that the bottom-most level of ΓQ are the simple representations,
the leading leftmost diagonal are the projective representations, and the rightmost
diagonal (the last elements in each level) are the injective representations.
The last notion we cover here are the meshes which occur in the Auslander-
Reiten quiver. In Figure 4 we see the four different meshes which can occur in an
Auslander-Reiten quiver. These meshes represent the aforementioned almost split
sequences and furthermore play an important role in the knitting algorithm.

4. The Knitting Algorithm

Here we present an algorithm to construct the Auslander-Reiten quiver for quiv-
ers such that ∆Q is type A or D. We encourage, as part of the expository nature
of this paper, that readers interested in this material should explore QPA [22]—
a package for GAP [22]. The following material (and much more) can be found
within the QPA package, and in our case QPA can accelerate the computing of the
Auslander-Reiten quiver by computing our representations. We will highlight this
in Example 4.3.
To begin, we introduce the knitting algorithm itself:
The Knitting Algorithm. Let Q be a quiver, to construct the Auslander-Reiten
quiver ΓQ we follow the procedure;
(1) Compute the indecomposable projective representations
P(1), P(2), . . . , P(n).
(2) Draw an arrow P(i) → P(j) in ΓQ if there exists an arrow j → i ∈ Q1 .
Position the P(i) to be on their own level.
(3) (Knit) Complete the appropriate mesh such that
2
󰁛
−1
dimL + dimτ L= dimMi .
i=1

(4) Repeat (3) until there are negative integers in the dimension vector.
It is easy to see that the real difficulty of this algorithm lies within step (3), we
focus now on explaining what it means to complete the mesh. First we recognize
that the available meshes depend on ∆Q . If ∆Q = An then there are three meshes
(the first three shown in Figure 4), and completing them is illustrated below.

M1 M1
(4.1)

L L τ −1 L
 10

M1,n+1

M2,n+1 M1,n

M3,n+1 M2,n M1,n−1

M1,4

Mn−1,n+1 M2,4 M1,3

Mn,n+1 Mn−1,n M3,4 M2,3 M1,2

NEELAM VENKATA PRASAD AKULA

Figure 3. Auslander-Reiten quiver of Q = 1 → · · · → n.

 THE KNITTING ALGORITHM FOR AD TYPE QUIVERS 11

• •
• • •
• • • • •
• • •
• •

Figure 4. The four different types of meshes.

L L τ −1 L
(4.2)
M2 M2

M1 M1

(4.3) L L τ −1 L

M2 M2
If ∆Q = Dn then there are four meshes, with the first three the same as type An .
The fourth mesh and its completion are shown below.
M1 M1

(4.4) L M2 L M2 τ −1 L

M1 M3
It follows that in the case of the fourth mesh, we must change the stopping criterion
in step (3) of the algorithm. We modify it to be
3
󰁛
dimL + dimτ −1 L = dimMi .
i=1

4.1. Coxeter Matrix Φ. Regardless of which mesh we wish to complete we must

compute a new representation τ −1 L (the symbol τ −1 L is given here since the tra-
ditional method of computing utilizes the Auslander-Reiten translation, which can
be found in [AR75]). As noted at the end of subsection 2.1, it suffices to describe
a representation in terms of its dimension vector. In this section we introduce and
define the Coxeter Matrix Φ which will allow us to compute the dimension vector
of τ −1 L and thus let us complete the mesh.
To do so, we need the Cartan matrix C ∈ Mn (Z) defined as follows;
i. The ith column is the dimension vector of the indecomposable projective rep-
resentation P(i).
ii. The ith row is the dimension vector of the indecomposable injective represen-
tation I(i).
12 NEELAM VENKATA PRASAD AKULA

Alternatively, for C = (cij )1≤i,j≤n , cij is the number of paths from j to i. Note
that since in step (1) of the algorithm we compute all the indecomposable projective
representations and thus constructing C becomes trivial.
Since Q is only of type A or D, we have that it has no oriented cycles. Then we
can renumber the vertices of Q such that, if there is a path from j to i, then i ≤ j.
Such an enumeration of vertices results in C being upper triangular (with diagonal
entries given as just 1) and thus C is invertible.
We then define the Coxeter matrix Φ such that Φ = −C T C −1 , with natural
inverse Φ−1 = −C(C −1 )T . Then for an arbitrary representation M ,
ΦdimM = dimτ M, if M is not projective and ΦdimP(i) = −dimI(i).
And
Φ−1 dimM = dimτ −1 M, if M is not injective and Φ−1 dimI(i) = −dimP(i).
Example 4.1. Suppose Q is as described in Example 2.5 with ΓQ given in Example
3.1. Our three projective representations are given as
dimP(1) = (1, 1, 0), dimP(2) = (0, 1, 0), dimP(3) = (0, 1, 1).
Then, the Cartan matrix is
󰀵 󰀶
󰀅 1
󰀆 0 0
C = P(1) P(2) P(3) = 󰀷1 1 1󰀸
0 0 1
with inverse 󰀵 󰀶
1 0 0
C −1 = 󰀷−1 1 −1󰀸 .
0 0 1
Thus the Coxeter matrix is
󰀵 󰀶
0 −1 1
Φ = −C T C −1 = 󰀷1 −1 1󰀸
1 −1 0
with inverse 󰀵 󰀶
−1 1 0
Φ−1 = −C(C −1 )T = 󰀷−1 1 −1󰀸
0 1 −1
Suppose that we want to compute dimτ −1 P(2) (note that this is the same as
computing dimτ −1 S(2) since P(2) = S(2)), then we have Φ−1 (0, 1, 0) = (1, 1, 1)
which is in fact dimI(2), where I(2) is the representation which completes the
appropriate mesh.
4.2. Examples of the Knitting Algorithm. Here we provide two examples of
using the Coxeter matrix in the knitting algorithm to compute the Auslander-Reiten
quiver for two quivers; one of type A and the other of type D. Both examples are
taken from [Sch14, Chapter 3], but have been rewritten to illustrate the algorithm
and use of the Coxeter matrix more clearly.
Example 4.2. Let Q be the quiver
1 2 3 4 5 .
 THE KNITTING ALGORITHM FOR AD TYPE QUIVERS 13

Note that ∆Q = A5 . We have the indecomposable projective representations;

dimP(1) = (1, 0, 0, 0, 0), dimP(2) = (1, 1, 0, 0, 0), dimP(3) = (1, 1, 1, 1, 0),

dimP(4) = (0, 0, 0, 1, 0), dimP(5) = (0, 0, 0, 1, 1).

This completes step (1) of the algorithm, we now need to place the P(i) in ΓQ
along with their arrows. This gives us

P(5)

P(4)

P(3)

P(2)

P(1)

In the same manner as above, we compute the Coxeter matrix (and its inverse)
to be 󰀵 󰀶 󰀵 󰀶
−1 1 0 0 0 0 0 −1 1 0
󰀹−1 0 1 0 0󰀺 󰀹1 0 −1 1 0󰀺
󰀹 󰀺 󰀹 󰀺
Φ=󰀹󰀹−1 0 1 −1 1󰀺
󰀺 Φ−1 =󰀹
󰀹0 1 −1 1 0󰀺󰀺
󰀷 0 0 1 −1 1󰀸 󰀷0 1 −1 1 −1󰀸
0 0 1 −1 0 0 0 0 1 −1
We now start completing the mesh, beginning with the mesh consisting of P(1)
and P(2) corresponding to (4.1). This gives dimτ −1 P(1) = (0, 1, 0, 0, 0) with
(1, 0, 0, 0, 0) + (0, 1, 0, 0, 0) = (1, 1, 0, 0, 0) = dimP(2) so the mesh is complete. Like-
wise, completing the appropriate meshes for P(2) and P(4) (meshes (4.1) and (4.3)
respectively) gives us dimτ −1 P(2) = (0, 1, 1, 1, 0) and dimτ −1 P(4) = (1, 1, 1, 1, 1),
one can easily see that the meshes are complete. These first three iterations give us
an updated Auslander-Reiten quiver, we indicate the new meshes as dashed arrows.

P(5)

P(4) (1, 1, 1, 1, 1)

P(3)

P(2) (0, 1, 1, 1, 0)

P(1) (0, 1, 0, 0, 0)
14 NEELAM VENKATA PRASAD AKULA

P(5) I(1)

P(4) (1, 1, 1, 1, 1) I(2)

P(3) (0, 1, 1, 1, 1) I(3)

P(2) (0, 1, 1, 1, 0) I(4)

P(1) S(2) (0, 0, 1, 1, 0) I(5)

Figure 5. Auslander-Reiten quiver for Q as in Example 4.2.

Continuing on with this process we get the Auslander-Reiten quiver of Q to be

as shown in Figure 5.
We know where to stop knitting when we reach injective representations, for
instance if we tried to compute dimτ −1 I(4) we would get (0, 0, 0, −1, 0) and there
is a negative in the dimension vector, thus I(4) is the last element in that level.
Example 4.3. Let Q be the quiver
4

1 2 3

5
Note that ∆Q = D5 . We have the indecomposable projective representations;
dimP(1) = (1, 1, 0, 0, 0), dimP(2) = (0, 1, 0, 0, 0), dimP(3) = (0, 1, 1, 0, 1),
dimP(4) = (0, 1, 1, 1, 1), dimP(5) = (0, 0, 0, 0, 1).
We now place the P(i) in ΓQ , noting that P(4) is placed on the same level as P(3).

P(5)

P(3) P(4)

P(2)

P(1)
 THE KNITTING ALGORITHM FOR AD TYPE QUIVERS 15

As mentioned earlier, much of this work can be accelerated by using the QPA
package in GAP. To illustrate some use, we include some GAP code which mirrors
the work we need to do.
gap> Q := Quiver(5, [ [1,2], [3,2], [3,5], [4,3] ]);;
gap> kQ := PathAlgebra(Rationals, Q);;
gap> P := IndecProjectiveModules(kQ);
[ <[ 1, 1, 0, 0, 0 ]>,
<[ 0, 1, 0, 0, 0 ]>,
<[ 0, 1, 1, 0, 1 ]>,
<[ 0, 1, 1, 1, 1 ]>,
<[ 0, 0, 0, 0, 1 ]> ]
gap> CoxeterInverse := Inverse(TransposedMat(CoxeterMatrix(kQ)));
[ [ -1, 1, 0, 0, 0 ],
[ -1, 1, 0, -1, 1 ],
[ 0, 1, 0, -1, 1 ],
[ 0, 0, 1, -1, 0 ],
[ 0, 1, 0, -1, 0 ] ]
gap> # compute dim t^-1P(2)
gap> CoxeterInverse * [0,1,0,0,0];
[ 1, 1, 1, 0, 1 ]
gap> # compute dim t^-1P(5)
gap> CoxeterInverse * [0,0,0,0,5];
[ 0, 1, 1, 0, 0 ]
This reveals the first two new elements in ΓQ (P(2) is mesh type (4.3) and P(5)
is mesh type (4.2)). Then from here we can compute dimτ −1 P(3) which is our first
example of mesh (4.4).
gap> # compute dim t^-1P(3)
gap> CoxeterInverse * [0,1,1,0,1];
[ 1, 2, 2, 1, 1 ]
We illustrate the first iteration of the algorithm in dashed arrows and the second
in squiggly arrows.

P(5) (0, 1, 1, 0, 0)

P(3) P(4) (1, 2, 2, 1, 1)

P(2) (1, 1, 1, 0, 1)

P(1) (0, 0, 1, 0, 1)

Finishing the algorithm then results in the Auslander-Reiten quiver of Q being

as shown in Figure 6.
 16

P(5) (0, 1, 1, 0, 0) (1, 1, 1, 1, 1) (0, 0, 1, 0, 0) (0, 0, 0, 1, 0)

P(3) P(4) (1, 2, 2, 1, 1) (1, 1, 1, 0, 0) (1, 1, 2, 1, 1) (0, 0, 1, 1, 1) (0, 0, 1, 1, 0)

P(2) (1, 1, 1, 0, 1) (0, 1, 2, 1, 1) (1, 1, 1, 1, 0)

P(1) (0, 0, 1, 0, 1) (0, 1, 1, 1, 0) (1, 0, 0, 0, 0)

Figure 6. Auslander-Reiten quiver for Q as in Example 4.3.

NEELAM VENKATA PRASAD AKULA
 THE KNITTING ALGORITHM FOR AD TYPE QUIVERS 17

References
[AJ21] Marco Armenta and Pierre-Marc Jodoin, The representation theory of neural networks,
Mathematics 9 (2021), no. 24, 3216.
[Aku22] Neelam Venkata Prasad Akula, A proof of gabriel’s theorem, 2022. Online available at:
https://math.umd.edu/~nakula/papers/gabriels-theorem.pdf.
[AR75] Maurice Auslander and Idun Reiten, Representation theory of artin algebras iii almost
split sequences, Communications in Algebra 3 (1975), no. 3, 239–294.
[ARS97] Maurice Auslander, Idun Reiten, and Sverre O. Smalø, Representation theory of artin
algebras, Bulletin of the London Mathematical Society 29 (1997), no. 2, 243–254.
[Bri08] Michel Brion, Representations of quivers, 2008. Online available at: http://
www-fourier.univ-grenoble-alpes.fr/~mbrion/notes_quivers_rev.pdf.
[CB92] William Crawley-Boevey, Lectures on representations of quivers, 1992. Online available
at: https://www1.maths.leeds.ac.uk/pure/staff/crawley_b/quivlecs.pdf.
[CG97] Neil Chriss and Victor Ginzburg, Representation theory and complex geometry, Vol. 42,
Springer, 1997.
[Cum11] Emma Cummin, Representations of quivers & gabriel’s theorem, 2011. Online available
at: https://people.bath.ac.uk/ac886/students/emmaCummin.pdf.
[DW05] Harm Derksen and Jerzy Weyman, Quiver representations, Notices of the AMS 52
(2005), no. 2, 200–206.
[DW11] , The combinatorics of quiver representations, Annales de l’institut fourier, 2011,
pp. 1061–1131.
[DW17] , An introduction to quiver representations, Vol. 184, American Mathematical
Soc., 2017.
[FM+ 17] Shigeyuki Fujii, Satoshi Minabe, et al., A combinatorial study on quiver varieties,
SIGMA. Symmetry, Integrability and Geometry: Methods and Applications 13 (2017),
052.
[Gab72] Peter Gabriel, Unzerlegbare darstellungen i, Manuscripta mathematica 6 (1972), no. 1,
71–103.
[22] GAP – Groups, Algorithms, and Programming, Version 4.12.2, The GAP Group, 2022.
[Gin09] Victor Ginzburg, Lectures on nakajima’s quiver varieties, arXiv preprint
arXiv:0905.0686 (2009).
[GW22] Iordan Ganev and Robin Walters, Quiver neural networks, arXiv preprint
arXiv:2207.12773 (2022).
[Hal21] Henry Hale, An exposition of a proof of gabriel’s theorem, 2021. Online available at:
http://math.uchicago.edu/~may/REU2021/REUPapers/Hale.pdf.
[Har11] Tero Harju, Lecture notes on graph theory (2011).
[Hun12] Thomas W Hungerford, Algebra, Vol. 73, Springer Science & Business Media, 2012.
[Kac82] Victor G Kac, Infinite root systems, representations of graphs and invariant theory, ii,
Journal of algebra 78 (1982), no. 1, 141–162.
[Len19] Emma Lennen, Quiver representations: Gabriel’s theorem and kac’s theorem, 2019. On-
line available at: https://math.uchicago.edu/~may/REU2019/REUPapers/Lennen.pdf.
[McK80] John McKay, Graphs singularities and finite groups, Proc. of 1979 Santa Cruz group
theory conference 37 (1980), 183–186.
[22] QPA – Quivers, Path Algebras, and Representations - a GAP package, Version 1.33,
The QPA Team, 2022.
[Rin96] Claus Michael Ringel, Pbw-bases of quantum groups. (1996).
[Sch14] Ralf Schiffler, Quiver representations, Vol. 1, Springer, 2014.
[Soi19] Alexander Soibelman, Lecture notes on quiver representations and moduli problems in
algebraic geometry, arXiv preprint arXiv:1909.03509 (2019).
[Uni12] Karlstad University, Introduction to the representation theory of quivers, 2012. Online
available at: https://www.math.uni-bielefeld.de/~sek/kau/.
[Zel05] Shmuel Zelikson, Auslander–reiten quivers and the coxeter complex, Algebras and rep-
resentation theory 8 (2005), no. 1, 35–55.

Department of Mathematics, University of Maryland

Email address: [email protected]
URL: www.math.umd.edu/~nakula