Not really an answer, but a comment about the related question: > Can one reach every cluster in that way ? It is rather easy to check (using [sagemath][1] or [Keller's applet][2]) that for the cyclic quiver in type $D_n$, not every cluster can be reached in that way. For example, only 13 among the 14 clusters in type $D_3$ are reached. A conjecture for the number of reached clusters for this initial quiver in type $D_n$ can be found in this [OEIS sequence][3]. **EDIT:** Even in type $A_4$, not all clusters are reached if the initial cluster is not the equioriented Dynkin quiver. Here is a small sage program that generates the set of reached clusters: def recursive_comb(carquois): """ EXAMPLES:: sage: Q = ClusterQuiver(DiGraph({0:[1]})) sage: len(list(recursive_comb(Q))) 5 """ n = carquois.n() seed = ClusterSeed(carquois) initial_cluster = seed.cluster() def voisins(s): c = s.cluster() for i in range(n): if c[i] in initial_cluster: yield s.mutate(i, inplace=False) return RecursivelyEnumeratedSet([seed], voisins, structure='graded') **EDIT:** Using the same program, one can see that some cluster variables are missing already for some choices of initial quiver in type $D_4$: sage: Q1 = ClusterQuiver(DiGraph({'b':['c'],'c':['d','e']})) sage: Q2 = ClusterQuiver(DiGraph({'c':['b','d','e']})) sage: len(set(v for c in recursive_comb(Q1) for v in c.cluster())) 16 sage: len(set(v for c in recursive_comb(Q2) for v in c.cluster())) 15 **EDIT**: Even in type $A$, some clusters can be out of reach: sage: Q1 = ClusterQuiver(DiGraph({'b':['c','a'],'c':['d']})) sage: len(list(recursive_comb(Q1))) 37 instead of 42, the full number of clusters. [1]: https://www.sagemath.org/ [2]: https://webusers.imj-prg.fr/~bernhard.keller/quivermutation/ [3]: https://oeis.org/A165205