Dugas’ Restricted Quiver Mutation Conjecture

Problem B

The restricted mutation classes were computed through a modified sage. A general pattern that I noted while playing with examples is that quivers with a higher density of oriented $3$-cycles seem to have larger restricted mutation classes (e.g. the $j \to j - 2 \pmod n$ and triangular quivers below). Those that don’t tend to be have relatively small restricted mutation classes in comparison.

Below, RMCS is short for $|[Q]_\text{res}|$.

Vertices = $[n]$ and arrows $i \to i + 1 \pmod n$ of multiplicity $2$

[proof=RMCS vs. $n$] Seem to have $[Q]_\mathrm{res}$ that grows exponentially in cardinality in $n$. Here is sage code that plots these quivers:


Code used to find the restricted mutation classes (requires this version of sage):


Here is a table consisting of $n$ and the corresponding $[Q]_\mathrm{res}$ cardinalities.

$n$$|[Q]_\mathrm{res}|$
$3$$1$
$4$$2$
$5$$3$
$6$$5$
$7$$5$
$8$$8$
$9$$10$
$10$$15$
$11$$19$
$12$$31$
$13$$41$
$14$$64$
$15$$94$
$16$$143$
$17$$211$
$18$$329$
$19$$493$
$20$$766$
$21$$1170$
$22$$1811$
$23$$2787$
$24$$4341$
$25$$6713$
$26$$10462$
$27$$16274$
$28$$25415$

[/proof]

Vertices = $[n]$ and arrows $i \to i + 1 \pmod n$ and $j \to j+2\pmod n$.

[proof=RMCS vs. $n$] Seem to have $[Q]_\mathrm{res}$ that grows exponentially in cardinality in $n$. Here is sage code that plots these quivers:


Code used to find the restricted mutation classes (requires this version of sage):


Here is a table consisting of $n$ and the corresponding $[Q]_\mathrm{res}$ cardinalities.

$n$$|[Q]_\mathrm{res}|$
$5$$6$
$6$$6$
$7$$4$
$8$$4$
$9$$5$
$10$$6$
$11$$7$
$12$$11$
$13$$12$
$14$$17$
$15$$23$
$16$$31$
$17$$40$
$18$$58$
$19$$76$
$20$$109$
$21$$149$
$22$$208$
$23$$287$
$24$$410$
$25$$567$
$26$$803$
$27$$1128$
$28$$1599$
$29$$2249$
$30$$3201$

[/proof]

Vertices = $[n]$ and arrows $i \to i + 1 \pmod n$ and $j \to j - 2 \pmod n$

[proof=RMCS vs. $n$] Seem to have $[Q]_\mathrm{res}$ that grows exponentially in cardinality in $n$. Here is sage code that plots these quivers:


Code used to find the restricted mutation classes (requires this version of sage):


Here is a table consisting of $n$ and the corresponding $[Q]_\mathrm{res}$ cardinalities.

$n$$|[Q]_\mathrm{res}|$
$5$$6$
$6$$2$
$7$$28$
$8$$42$
$9$$48$
$10$$440$
$11$$4000$
$12$$768$
$13$$50585$
$14$$62630$
$15$$86369$

[/proof]

[proof=Growth rate for $n = 5$ (finite)] For $n = 5$, there are finitely many restricted cluster variables. Code used to find the restricted cluster variables (requires this version of sage):


[/proof]

[proof=Growth rate for $n = 6$ (exponential?)] Computed up to depth 15. Code used to find the restricted cluster variables (requires this version of sage):


Scatter plot of data:


[/proof]

[proof=Growth rate for $n = 7$ (finite)] For $n = 7$, there are finitely many restricted cluster variables. [/proof]

[proof=Growth rate for $n = 8$ (finite)] For $n = 8$, there are finitely many restricted cluster variables. [/proof]

[proof=Growth rate for $n = 9$ (cubic?)] For $n = 9$, we seem to have cubic growth rate.


[/proof]

[proof=Growth rate for $n = 10$ (finite)] For $n = 10$, there are finitely many restricted cluster variables. [/proof]

Triangular quivers

[proof=RMCS vs. $n$] Seem to have $[Q]_\mathrm{res}$ that grows rapidly in cardinality in the side length. Here is sage code that plots these quivers:


Code used to find the restricted mutation classes (requires this version of sage):


Here is a table consisting of $n$ and the corresponding $[Q]_\mathrm{res}$ cardinalities.

Side length$|[Q]_\mathrm{res}|$
$0$$1$
$1$$1$
$2$$2$
$3$$19$
$4$$17340$

For the case of a side length of $5$, the restricted mutation class appears to be extremely large (already $854{,}236$ quivers at depth $11$) yet none of the resultant quivers have edge multiplicity larger than $2$. These types of quivers may be good to check as an extreme case.

[/proof]

[proof=Growth rate for side length 2 (finite)] There are finitely many cluster variables. [/proof]

[proof=Growth rate for side length 3 (exponential?)] Code used to find the restricted cluster variables (requires this version of sage):


Scatter plot of data:


[/proof]

[proof=Growth rate for side length 4 (quintic?)]


[/proof]

Problem C

Many balanced and connected quivers generate infinitely many cluster variables, even if we only allow for sequences of mutations at $4$-valent vertices.

Markov quiver

The Markov quiver is the quiver on $[3]$ where with arrows $1 \to 2$, $2 \to 3$, and $3 \to 1$ all of multiplicity $2$. It is so named because it the cluster variables it generates can be used to construct the solutions to the Markov equation (see 3.4 of https://arxiv.org/abs/1608.05735).


For this quiver, mutating at any vertex simply reverses all the arrows so restricted mutation sequences are just the usual non-restricted mutation sequences. Since this quiver is not mutation-equivalent to a Dynkin ADE diagram, it generates infinitely-many cluster variables.

Multiplicity 2 square

The multiplicity two square is the quiver on $[4]$ with arrows $i \to i + 1$ modulo $4$ of multiplicity two. Here is code to plot these quivers:


One can produce infinitely many cluster variables by performing the sequence of mutations at the vertices

\[1 \to 3 \to 2 \to 4 \to 1 \to 3 \to 2 \to 4 \to \cdots.\]

Here is code that implements this mutation sequence and displays the resultant cluster seeds (note: gets messy very fast).


[proof=Proof of infinitely many cluster variables] Let $Q$ be our quiver as described above and $(Q, \{x_1, \ldots, x_4\})$ be our initial seed. We will specialize by letting $x_1 = x_3 = x$ and $x_2 = x_4 = y$. By direct computation, one obtains

\[\mu_4 \mu_2 \mu_3 \mu_1(Q, \{x, y, x, y\}) = \left(Q, \left\{\frac{2y^2}{x}, \frac{8y^3}{x^2}, \frac{2y^2}{x}, \frac{8y^3}{x^2}\right\}\right).\]

Under this specialization, we define

\[X_n = \frac{2^{n(2n - 1)} y^{2n}}{x^{2n - 1}} \quad\text{ and }\quad Y_n = \frac{2^{n(2n + 1)}y^{2n + 1}}{x^{2n}}.\]

for $n \ge 1$ and $X_0 = x$ and $Y_0 = y$. By induction and the formula we computed for $\mu_4 \mu_2 \mu_3 \mu_1(Q, {x, y, x, y})$, one can show that

\[(\mu_4 \mu_2 \mu_3 \mu_1)^n(Q, \{x, y, x, y\}) = (Q, \{X_n, Y_n, X_n, Y_n\})\]

for all $n \ge 0$. Thus, $Q$ cannot produce finitely many cluster variables. [/proof]

Square with multiplicity 2 diagonal

Code to plot the quiver in question:


Mutating this quiver at the vertex $0$ or vertex $1$ gives a quiver that is isomorphic to the original quiver. In fact, the resultant quiver is the original quiver with all its arrows reversed.


Since vertices $0$ and $1$ are the only vertices that can be mutated at, this construction is equivalent to an ice quiver of type $\widetilde{A}_1$. So this quiver produces infinitely many cluster variables.

Surfaces of type A

The quivers of surface type $A$ are constructed by taking a triangulation of a regular $(n+3)$-gon adding vertices to all edges of the polygon and adding arrows so that there is a clockwise-oriented triangle within each triangle of the triangulation. There is an example of this in Example 2.2, Figure 3 of Lauren William’s survey article. The $4$-valent vertices are precisely the vertices of the quiver corresponding to the diagonals of the triangulation. These quivers are known to be both restricted-mutation finite as well as produce finitely-many cluster variables.

Surfaces of type D

The quivers of surface type $D$ are constructed similarly to type $A$ but instead take place within a regular $n$-gon with one of the triangulation vertices sitting in the interior of the $n$-gon. See Figure 5.8 of Introduction to Cluster Algebras. Chapters 4-5. We can also add the arrows between the frozen vertices of the quiver without changing the cluster algebra to get a balanced quiver. The resultant quiver will be restricted mutation-finite and the cluster variables generated through sequences of restricted mutations give rise to the type $D$ cluster algebras (with coefficients).