Lecture Notes

Introduction to Cluster Algebra

Ivan C.H. Ip∗
Updated: April 14, 2017

§1 Examples
1.1 Conway-Coxeter frieze pattern
Frieze is the wide central section part of an entablature, often seen in Greek temples,
usually with a horizontal repeating pattern.
Example 1.1 (Frieze of height n = 3). Fill in the blanks with numbers, such that
b
whenever we have 4 numbers arranged as a d , we have ad − bc = 1.
c

1 1 1 1 1 1
1 · · · · ·
1 · · · · ·
1 · · · ·
1 1 1 1 1

bc+1
It is easy to solve by putting d = a , and fill in the blanks accordingly, we end
up with:

1 1 1 1 1 1
1 2 2 2 1 4
1 3 3 1 3 ···
1 4 1 2 ···
1 1 1 1 1

∗ Center for the Promotion of Interdisciplinary Education and Research/

Department of Mathematics, Graduate School of Science, Kyoto University, Japan

Email: [Link]@[Link]

1
 We note that the pattern starts repeating itself (with the fundamental domain
highlighted in red)
Let’s look at a more complicated example by changing the shape of the left
boundary
Example 1.2 (Frieze of height n = 5). .

1 1 1 1 1 1
1 3 2 2 2 1 ···
1 2 5 3 3 1 ···
1 3 7 4 1 ···
1 4 9 1 ···
1 5 2 1 ···
1 1 1 1

Again we return to a row of 1’s! Furthermore, all the entries are positive inte-
gers.
Now let us look at a more general pattern in the case of height n = 2:

1 1 1 1
x1 x3 x5 x7 ···
x2 x4 x6 ···
1 1 1

Example 1.3. We can solve for the variables and obtain

x2 + 1
x3 =
x1
x3 + 1 x1 + x2 + 1
x4 = =
x2 x1 x2
x4 + 1 x1 + 1
x5 = =
x3 x2
x5 + 1
x6 = = x1
x4
x6 + 1
x7 = = x2
x5
We see that we return to the initial variables, and the pattern repeat itself. In
general, for arbitrary height n,
(1) For k > n, the variables xk can be expressed as Laurent polynomial of the
variables x1 , ..., xn

2
 (2) The variables x1 , ..., xn align themselves again, and the pattern repeat
(3) The denominators of the Laurent polynomials are all different

Example 1.4. Let us consider again frieze of height n = 2, but change the rules
of the game to
 d1
xk + 1 k is even
xk+1 xk−1 =
xdk2 + 1 k is odd

Previously we have studied the case (d1 , d2 ) = (1, 1). In (d1 , d2 ) = (1, 2), we obtain
x2 + 1
x3 =
x1
2
x +1 (x2 + 1)2 + x21
x4 = 3 =
x2 x1 x2
x4 + 1 x21 + x2 + 1
x5 = =
x3 x1 x2
x21 + 1
x6 =
x2
x7 = x1
x8 = x2

again we return to the original variables. Also the denominators of the Laurent
polynomials are all distinct.
If we take (d1 , d2 ) = (1, 3), and for simplicity we let x1 = x2 = 1, then we obtain
the sequence
(xn ) = 1, 1, 2, 9, 5, 14, 3, 2, 1, 1, ...
again return to the initial numbers.
However, this phenomenon does not always hold. Take for example (d1 , d2 ) =
(1, 4), we obtain
(xn ) = 1, 1, 2, 17, 9, 386, 43, 8857, 206, ...
and for (d1 , d2 ) = (2, 2)

(xn ) = 1, 1, 2, 5, 13, 34, 89, 223, 610, 1597, ...

In fact, it is known that all xk can be expressed as Laurent polynomials in x1 , x2 ,

but we have the periodicity property only in the case when (d1 , d2 ) = (1, 1), (1, 2)
and (1, 3).
In fact the frieze pattern is closely related to triangulations of polygons. More
precisely,

3
Theorem 1.5. A frieze pattern of height n corresponds to a triangulation of n + 3-
gon, such that the top row represents the number of triangles incident to the vertex
of the polygons in the clockwise direction.

(1, 2, 2, 2, 1, 4, ...) −→

1.2 Triangulations
The rules for the original frieze pattern is closely related to an ancient Greek result
about Euclidean geometry known as the Ptolemy’s Theorem:
Theorem 1.6 (Ptolemy). If ABCD is inscribed in a circle, then

AC · BD = AB · CD + AD · BC

A B

If we consider a regular pentagon with side 1 on the circle, then the diagonals
of different triangulations will satisfy the relation same as before! (Of course on
Euclidean circle, x1 and x2 will already be fixed... but we can leviate this freedom
by considering certain hyperbolic models to be discussed in later lectures.)

4
 x2 x1

x2 x1
x3 x5

x4 x4

x3 x5

The relations among different triangulations here is known as the “pentagon re-
lation”, where different triangulations are related by “flip” of triangles, i.e. changing
the diagonals. The pentagon, on the other hand, is also known as the (type A2 ) as-
sociahedron because it is related to the usual associativity of products in 4 variables,
and one can find a one-to-one correspondence between the two.

b c

a d

(ab)(cd)

Then each flip of triangulation is related by the associativity relation (ab)c =

a(bc)

5
 (ab)(cd)

((ab)c)d a(b(cd))

(a(bc))d a((bc)d)

For example, when n = 3, we obtain the following associahedron, where each

vertex is 3-regular.

Figure 1: The A3 associahedron

1.3 Grassmannian
Another appearance of the Ptolemy relation is the coordinate rings of the Grass-
mannian, which also serve a large number of exampels of cluster algebra. Let us

6
recall the basic definition from linear algebra.
Let us consider vector spaces over C.
Definition 1.7. The Grassmannian variety Gr(k, n) is the space of all k-dimensional
linear subspace of Cn .
For example, Gr(1, n) = {lines in Cn } = projective space Pn .
In our motivation, we will stick to the example Gr(2, n + 3). What are the
points q ∈ Gr(2, n + 3)? Recall that every 2-dimensional vector subspace V can
be described by a basis v1 , v2 ∈ Cn+3 . Hence we can write q as a collection of row
vectors:  
−v1 −
q= ∈ M at2,n+3 (C)/{rank(q) = 2}
−v2 −
such that spanhv1 , v2 i = V .
Of course, this choice is not unique: we can always choose another basis by
applying a linear transform by GL2 (C), acting on the left. i.e.,

q1 = q2 ∈ Gr(2, n + 3)
⇐⇒q1 = A · q2 , A ∈ GL2 (C)

In fact, Gr(2, n + 3) is a projective variety, we have an embedding

Gr(2, n + 3) ,→ P(Λ2 Cn+3 ) ' PN

given by  
−v1 −
q= 7→ [v1 ∧ v2 ]
−v2 −
 
2 n+3 n+3
where N := dim Λ C = .
2
In terms of coordinate, we see that the embedding is given by “minors”, i.e.
determinants of 2 × 2 submatrix of q:
 
q1,1 q1,2 · · · q1,n+3
q= 7→ [∆12 (q) : · · · : ∆n+2,n+3 (q)]
q2,1 q2,2 · · · q2,n+3

where ∆i,j (q) = q1,i q2,j − q1,j q2,i is the 2 × 2 minor.

Proposition 1.8. This map is well-defined.
Proof. First, at least one of ∆i,j (q) 6= 0 because q has rank=2. Furthermore, if
q1 = q2 ∈ Gr(2, n + 3), then q1 = A · q2 for some A ∈ GL2 (C), and we have
∆i,j (q1 ) = det(A)∆i,j (q2 ), hence all the coordinates are rescaled by the same factor
det(A) 6= 0, hence determined the same point in P(Λ2 Cn+3 ).
In fact, Gr(2, n + 3) is a smooth projective variety, cut out by certain quadratic
equations, called the “Plücker relations”

∆i,k ∆j,l = ∆i,j ∆k,l + ∆i,l ∆j,k , 1≤i<j <k <l ≤n+3

7
which is essentially the same as Ptolemy relation!

i j

Hence understanding the (homogeneous coordinate) ring of functions on Gr(2, n+

3), is essentially the same as studying the algebra generated by the minors ∆ij

\
A := C[Gr(2, n + 3)] := C[∆ij ]i6=j /(Plücker relations)

where
\
Gr(2, n + 3) := θ−1 (Gr(2, n + 3)) ∪ {0} ⊂ CN
denote the affine cone, where θ : CN −→ PN is the projection.
Formally, one can consider the (n + 3)-gon, and associate ∆ij to the diagonals
joining i − j. This will give us the analogue of the variables xn discussed in the
previous examples. Since there is a relation between the overlapping variables,
we see that A will locally be described by different triangulations of the (n + 3)-
gon, where the corresponding minors are non-zero. Hence, for each triangulations,
the corresponding n minors defines the cluster variables of the algebra A. More
precisely, the non-overlapping diagonals form a linear basis of A:
X
A= K[∆ij ]∆ij is diagonal of T
T ∈triangulations

where K = C[∆12 , ∆23 , ..., ∆n+3,1 ] is generated by the sides of the (n + 3)-gon.
As a side note, for each triangulations, one can naturally associate to it a
“quiver” by putting an arrow between the edges (usually not an edge of the polygon)
of the triangles in the following way:

8
 Then for example, a change of triangulation will induce a change of quivers,
called the “quiver mutaiton”, that will be describe in more detail later.

−→

One can then rewrite the Plücker relation as

Y Y
∆K ∆0K = ∆I + ∆I
∆I −→∆K ∆K −→∆I

where ∆K and ∆0K are the minors corresponding to the diagonal of the two pictures.
We will see that this is a general way of writing mutations in cluster algebra associate
with quivers.