Rainbow Cycles in Flip Graphs

Stefan Felsner, Linda Kleist, Torsten Mütze, Leon Sering

Institut für Mathematik

TU Berlin, 10623 Berlin, Germany

{felsner,kleist,muetze,sering}@math.tu-berlin.de

Abstract. The flip graph of triangulations has as vertices all triangulations of a convex -gon, and an edge between any two triangulations that differ in exactly one edge. An -rainbow cycle in this graph is a cycle in which every inner edge of the triangulation appears exactly times. This notion of a rainbow cycle extends in a natural way to other flip graphs. In this paper we investigate the existence of -rainbow cycles for three different flip graphs on geometric classes of objects: the aforementioned flip graph of triangulations of a convex -gon, the flip graph of plane trees on an arbitrary set of points, and the flip graph of non-crossing perfect matchings on a set of points in convex position. In addition, we consider two flip graphs on classes of non-geometric objects: the flip graph of permutations of and the flip graph of -element subsets of . In each of the five settings, we prove the existence and non-existence of rainbow cycles for different values of , and .

## 1. Introduction

Flip graphs are fundamental structures associated with families of geometric objects such as triangulations, plane spanning trees, non-crossing matchings, partitions or dissections. A classical example is the flip graph of triangulations. The vertices of this graph are the triangulations of a convex -gon, and two triangulations are adjacent whenever they differ by exactly one edge. In other words, moving along an edge of corresponds to flipping the diagonal of a convex quadrilateral formed by two triangles. Figure 1 shows the graph .

A question that has received considerable attention is to determine the diameter of , i.e., the number of flips that is necessary and sufficient to transform any triangulation into any other, see the survey [MR2455502]. In a landmark paper [MR928904], Sleator, Tarjan and Thurston proved that the diameter of is for sufficiently large . Recently, Pournin [MR3197650] gave a combinatorial proof that the diameter is for all . A challenging algorithmic problem in this direction is to efficiently compute a minimal sequence of flips that transforms two given triangulations into each other, see [MR1744193, DBLP:conf/cocoon/LiZ98]. These questions involving the diameter of the flip graph become even harder when the points are not in convex, but in general position, see e.g. [MR1640808, MR1706610, MR2770955]. Moreover, apart from the diameter, many other properties of the flip graph have been investigated, e.g., its realizability as a convex polytope [MR3437894], its automorphism group [MR1022776], the vertex-connectivity [MR1723053], and the chromatic number [MR2535071].

Another property of major interest is the existence of a Hamilton cycle in . This was first established by Lucas [MR920505] and a very nice and concise proof was given by Hurtado and Noy [MR1723053]. The reason for the interest in Hamilton cycles is that a Hamilton cycle in corresponds to a so-called Gray code, i.e., an algorithm that allows to generate each triangulation exactly once, by performing only a single flip operation when moving to the next triangulation. In general, the task of a Gray code algorithm is to generate all objects in a particular combinatorial class, each object exactly once, by applying only a small transformation in each step, such as a flip in a triangulation. Combinatorial classes of interest include geometric configurations such as triangulations, plane spanning trees or non-crossing perfect matchings, but also classes without geometric information such as permutations, combinations, bitstrings etc. This fundamental topic is covered in depth in the most recent volume of Knuth’s seminal series The Art of Computer Programming [MR3444818], and in the classical books by Nijenhuis and Wilf [MR510047, MR993775]. Here are some important Gray code results in the geometric realm: Hernando, Hurtado and Noy [MR1939072] proved the existence of a Hamilton cycle in the flip graph of non-crossing perfect matchings on a set of points in convex position for every even . Aichholzer et al. [MR2346418] described Hamilton cycles in the flip graphs of plane graphs on a general point set, for plane and connected graphs and for plane spanning trees on a general point set. Huemer et al. [MR2510231] constructed Hamilton cycles in the flip graphs of non-crossing partitions of a point set in convex position, and for the dissections of a convex polygon by a fixed number of non-crossing diagonals.

As mentioned before, a Hamilton cycle in a flip graph corresponds to a cyclic listing of all objects in some combinatorial class, such that each object is encountered exactly once, by performing a single flip in each step. In this work we consider the dual problem: we are interested in a cyclic enumeration of some of the combinatorial objects, such that each flip operation is encountered exactly once. For instance, in the flip graph of triangulations , we ask for the existence of a cycle with the property that each inner edge of the triangulation appears (and disappears) exactly once. An example of such a cycle is shown in Figure 1. This idea can be formalized as follows. Consider two triangulations and that differ in flipping the diagonal of a convex quadrilateral, i.e., is obtained from by removing the diagonal and inserting the other diagonal . We view the edge between and in the flip graph as two arcs in opposite directions, where the arc from to receives the label , and the arc from to receives the label , so the label corresponds to the edge of the triangulation that enters in this flip; see the right hand side of Figure 1. Interpreting the labels as colors, we are thus interested in a directed cycle in the flip graph in which each color appears exactly once, and we refer to such a cycle as a rainbow cycle. More generally, for any integer , an -rainbow cycle in is a cycle in which each edge of the triangulation appears (and disappears) exactly times. Note that a rainbow cycle does not need to visit all vertices of the flip graph. Clearly, this notion of rainbow cycles extends in a natural way to all the other flip graphs discussed before, see Figure 2.

### 1.1. Our results

In this work we initiate the investigation of rainbow cycles in flip graphs for five popular classes of combinatorial objects. We consider three geometric classes: triangulations of a convex polygon, plane spanning trees on point sets in general position, and non-crossing perfect matchings on point sets in convex position. In addition, we consider two classes without geometric information: permutations of the set , and -element subsets of . We proceed to present our results in these five settings in the order they were just mentioned. For the reader’s convenience, all results are summarized in Table 1.1.

geometric |
\adl@mkpreamc\@addtopreamble\@arstrut\@preamble | \adl@mkpreamc\@addtopreamble\@arstrut\@preamble | |||||
---|---|---|---|---|---|---|---|

Vertices | Arcs/edges | Yes | No | ||||

]24cmtriangulations of convex -gon | edge flip | Thm. LABEL:thm:triang | |||||

]24cmplane spanning trees on point set in general position | edge flip | Thm. LABEL:thm:trees | |||||

]34cmnon-crossing perfect matchings on points in convex position | ]22cmtwo edge flip | Thm. LABEL:thm:match | |||||

]-22.4cmodd , | |||||||