Region crossing change on planar trivalent graphs

A hypergraph $H$ consists of a finite set $V=\{v_{1},\cdots,v_{n}\}$ and a family of hyperedges $E=\{e_{1},\cdots,e_{m}\}$, where each $e_{i}$ is a subset of $V$. For a given hypergraph $H$, the incidence matrix of $H$ is a $m\times n$ matrix $M_{H}=(a_{ij})_{m\times n}$, where

$a_{ij}=\begin{cases}1&\text{if $v_{j}\in e_{i}$};\\ 0&\text{otherwise.}\end{cases}$

In this note, we are concerned with the rank of this matrix. Throughout this note, all the matrices are considered as elements of $M_{m\times n}(\mathbb{Z}_{2})$, which denotes the collection of all $m\times n$ matrices over $\mathbb{Z}_{2}=\mathbb{Z}/2\mathbb{Z}$. Here we have two examples.

The purpose of this note is to investigate the effect of region crossing change on planar trivalent graphs. Suppose we are given a planar trivalent graph $G$ embedded on the plane. For simplicity, let us assume that $G$ is connected and it contains no bridge. As a corollary, $G$ is loopless and locally the three regions around each vertex are mutually distinct. Therefore each region is a $n$-gon for some positive integer $n$, and one of them is unbounded. If $n$ is even, then we say this region is an even region, otherwise we call it an odd region. By a state of $G$, we mean a function $s:V(G)=\{v_{1},\cdots,v_{n}\}\to\{0,1\}$, thus a state can be represented by a row vector $(s(v_{1}),\cdots,s(v_{n}))$. Similar to Example 1.2, a region crossing change on a region defines a new state by switching $s(v_{i})$ for all the vertices $v_{i}$ lying on the boundary of this region. We say two states are equivalent if they are related by a sequence of region crossing changes.

As a natural question, we want to know for a fixed planar trivalent graph $G$, how many non-equivalent states does $G$ have. In this paper, we establish the following result, which provides an answer to this question.

The definition of the group homomorphism $\phi:\mathcal{G}\to S_{3}$ can be found in Section 2. The assumption $n\geq 4$ excludes the possibility that $G$ is the $\theta$-graph, which obviously has two non-equivalent states, although all the three regions are even. The reason why we exclude the $\theta$-graph is, in this case, $m=3>2=n$. Actually, since $2|E(G)|=3|V(G)|=3n$, the Euler Identity implies that $m=\frac{n}{2}+2$. It follows that $m\geq 4$ if and only if $n\geq 4$ and in this case we always have $n\geq m$.

As an analogue of Example 1.1 and Example 1.2, the calculation of $s(G)$ can be translated into the problem of calculating the rank of an incidence matrix $M_{H}$ for a certain hypergraph $H$. Let us use $\{e_{1},\cdots,e_{l}\},\{R_{1},\cdots,R_{m}\}$ and $\{v_{1},\cdots,v_{n}\}$ to denote the set of edges, the set of regions and the set of vertices of $G$, respectively. As we mentioned above, here $l,m$ are both determined by $n$, i.e. $l=\frac{3n}{2}$ and $m=\frac{n}{2}+2$. Now we construct a hypergraph $H$ such that $V(H)=V(G)$ and each region $R_{i}$ corresponds to a hyperedge $r_{i}$, where $r_{i}=\{v_{i_{1}},\cdots,v_{i_{k}}\}$ if $v_{i_{1}},\cdots,v_{i_{k}}$ lie on the boundary of $R_{i}$. We will often abuse our notation by using $r_{i}$ to denote both a hyperedge of $H$ and the corresponding row vector of $M_{H}$. Since a state $s$ of $G$ corresponds to a row vector $s=(s(v_{1}),\cdots,s(v_{n}))$, taking region crossing changes on $\{R_{i_{1}},\cdots,R_{i_{k}}\}$ yields a new state $s+\sum\limits_{j=1}^{k}r_{i_{j}}$. Therefore, for a given state of $G$, there are totally $2^{\text{rank}(M_{H})}$ distinct states can be obtained from the given one by taking region crossing changes. It follows that the number of non-equivalent states of $G$ equals $2^{n-\text{rank}(M_{H})}$. As a result, in order to prove Theorem 1.3, it suffices to show that

$\text{rank}(M_{H})=\begin{cases}m-2&\text{if all the regions are even;}\\ m-1&\text{if there exist odd regions but no adjacent odd regions, and }\phi(\mathcal{G})\subseteq\mathcal{H};\\ m&\text{otherwise.}\end{cases}$

The followings are some examples.

Here we have three planar trivalent graphs $G_{1},G_{2},G_{3}$ illustrated in Figure 1, which correspond to the three cases mentioned in Theorem 1.3. According to our definition, the incidence matrices of $H_{1},H_{2}$ and $H_{3}$ are listed below.

$M_{H_{1}}=\begin{pmatrix}1&1&0&0&1&1&0&0\\ 0&1&1&0&0&1&1&0\\ 0&0&1&1&0&0&1&1\\ 1&0&0&1&1&0&0&1\\ 0&0&0&0&1&1&1&1\\ 1&1&1&1&0&0&0&0\end{pmatrix}M_{H_{2}}=\begin{pmatrix}1&1&0&0&1&0\\ 0&1&1&0&1&1\\ 0&0&1&1&0&1\\ 1&0&0&1&1&1\\ 1&1&1&1&0&0\end{pmatrix}M_{H_{3}}=\begin{pmatrix}1&1&1&0\\ 1&1&0&1\\ 0&1&1&1\\ 1&0&1&1\end{pmatrix}$ Direct calculation shows that rank$(M_{H_{1}})=$ rank$(M_{H_{2}})=$ rank$(M_{H_{3}})=4$.

Note that $M_{H}$ is a matrix of size $m\times n$ $(m\leq n)$, and 1 appears three times in each column of $M_{H}$, since $G$ is trivalent.

The outline of the paper is as follows. In Section 2, we study the so-called $\mathbb{Z}_{3}$-coloring of planar trivalent graphs and its reduction to a $\mathbb{Z}_{2}$-coloring. Section 3 is devoted to give a proof of Theorem 1.3, based on a calculation of rank$(M_{H})$. In Section 4, we provide a criterion for triangulating $S^{2}$ with exactly two odd vertices.

Let $G$ be a planar trivalent graph embedded on the plane. As before, let us use $\{v_{1},\cdots,v_{n}\}$, $\{e_{1},\cdots,e_{l}\}$ and $\{R_{1},\cdots,R_{m}\}$ to denote $V(G),E(G)$ and the set of all regions respectively. Note that the dual graph $G^{\prime}$ is a triangulation of the plane, of which one triangle is unbounded. Equivalently, $G^{\prime}$ also can be regarded as a triangulation of $S^{2}$. Let $\{v_{1}^{\prime},\cdots,v_{m}^{\prime}\},\{e_{1}^{\prime},\cdots.e_{l}^{\prime}\}$ and $\{R_{1}^{\prime},\cdots,R_{n}^{\prime}\}$ be $V(G^{\prime}),E(G^{\prime})$ and the region set of $G^{\prime}$, respectively. Throughout this paper, we will frequently switch our focus from $G$ to $G^{\prime}$ and vice verse, which will not cause confusion since they determine each other mutually.

By a $\mathbb{Z}_{3}$-coloring of the triangulation $G^{\prime}$, we mean a function $c$ from $\{v_{1}^{\prime},\cdots,v_{m}^{\prime}\}$ to a 3-element set of colors $\{1,2,3\}$ such that adjacent vertices always receive distinct colors. In other words, the three vertices of each triangle are precisely colored by 1, 2 and 3 respectively. It is worthy to point out that, unlike the Tait coloring, which assigns a color from $\{1,2,3\}$ to each edge such that the edges of different colors are incident at each vertex, our $\mathbb{Z}_{3}$-coloring does not always exist. For example, if the degree of a vertex $v_{i}^{\prime}$ is odd (we simply say $v_{i}^{\prime}$ is odd), then such a $\mathbb{Z}_{3}$-coloring does not exist. It is not difficult to observe that converse also holds, i. e. a triangulation $G^{\prime}$ admits a $\mathbb{Z}_{3}$-coloring if and only if all the vertices of it are even. In order to see this, it is sufficient to notice that if a triangle $R_{i}^{\prime}$ has been colored properly, the coloring of any other triangle $R_{j}^{\prime}$ which shares an edge with $R_{i}^{\prime}$ has the only option. Therefore, by an arbitrary choice of a coloring on a triangle, one can extend this colorings to all other triangles uniquely. The key point here is, the two coloring extending around an even vertex are consistent. Hence there is no conflict in the extending. Figure 2 explains what will happen if one slides a path over an even vertex or an odd vertex.

When some vertices of $G^{\prime}$ are odd, there exists no global $\mathbb{Z}_{3}$-coloring. However, we can still discuss the $\mathbb{Z}_{3}$-coloring locally. Consider a sequence of triangles $R_{i_{1}}^{\prime},\cdots,R_{i_{k}}^{\prime}$ such that $R_{i_{j}}^{\prime}$ and $R_{i_{j+1}}^{\prime}$ are adjacent, here $1\leq j\leq k-1$. Then a coloring of $R_{i_{1}}^{\prime}$ induces a unique coloring of $R_{i_{k}}^{\prime}$ with respect to this sequence. In particular, if $R_{i_{1}}^{\prime}$=$R_{i_{k}}^{\prime}$, i. e. $R_{i_{1}}^{\prime},\cdots,R_{i_{k}}^{\prime}$ are dual to a circuit in $G$, then this sequence induces a recoloring of $R_{i_{1}}^{\prime}$.

For a given planar trivalent graph $G$, without loss of generality, we assume that the regions $R_{1},\cdots,R_{k}$ $(k\geq 2)$ are odd and other regions $R_{k+1},\cdots,R_{m}$ are even. Furthermore, we also assume that any two odd regions are not adjacent. Suppose one of the three regions around $v_{1}$ is odd, say $R_{1}$. Note that since no pair of odd regions are adjacent, the other two regions must be even, say $R_{i},R_{j}$ ($i,j\geq k+1$). Denote the vertices of $G^{\prime}$ corresponding to $R_{1},\cdots,R_{m}$ by $v_{1}^{\prime},\cdots,v_{m}^{\prime}$, then a circuit $l:v_{1}=v_{i_{1}},\cdots,v_{i_{k}}=v_{1}$ that begins and ends at $v_{1}$ corresponds to a sequence of adjacent triangles $R_{1}^{\prime}=R_{i_{1}}^{\prime},\cdots,R_{i_{k}}^{\prime}=R_{1}^{\prime}$, which begins and ends at $R_{1}^{\prime}$. Fix a $\mathbb{Z}_{3}$-coloring of the triangle $R_{1}^{\prime}$ such that the vertex $v_{1}^{\prime}$ is colored by 2. Now suppose we are given a circuit that begins and ends at $v_{1}$, the corresponding sequence of adjacent triangles induces a recoloring of $R_{1}^{\prime}$. As we have seen, such a recoloring is preserved if one slides this sequence over an even vertex of $G^{\prime}$. It follows immediately that, with a given coloring of $R_{1}^{\prime}$ such that the vertex $v_{1}^{\prime}$ is colored by 2, the recoloring defines a group homomorphism

$\phi:\mathcal{G}=\pi_{1}(S^{2}\setminus\{v_{1}^{\prime},\cdots,v_{k}^{\prime}\})\to S_{3}$,

here $S_{3}$ denotes the symmetric group of degree 3. This homomorphism was named as even obstruction map in [Fis1977], but here we would like use the name coloring monodromy, following [Izm2015].

Now we turn to the $\mathbb{Z}_{2}$-coloring of the triangulation $G^{\prime}$. Roughly speaking, a $\mathbb{Z}_{2}$-coloring can be obtained from a $\mathbb{Z}_{3}$-coloring by replacing 1 and 3 with 1 and replacing 2 with 0. More precisely, a $\mathbb{Z}_{2}$-coloring of the triangulation $G^{\prime}$ is a map from the vertex set $V(G^{\prime})$ to $\mathbb{Z}_{2}=\{0,1\}$, such that for each triangle, the sum of the colors of the three vertices equals 0 (mod 2). It is evident that each $\mathbb{Z}_{3}$-coloring corresponds to a $\mathbb{Z}_{2}$-coloring. However, the converse is not true. As an example, the dual graph of the trivalent planar graph $G_{2}$ in Figure 1 does not admit a $\mathbb{Z}_{3}$-coloring. However, the coloring which assigns 0 to $R_{1},R_{3}$ and 1 to the rest regions suggests that it admits a $\mathbb{Z}_{2}$-coloring.

We first explain how to relate the $\mathbb{Z}_{2}$-coloring of $G^{\prime}$ to the region crossing change problem on $G$. Recall that a $\mathbb{Z}_{2}$-coloring assigns 0 or 1 to each vertex of $G^{\prime}$ such that the three vertices of each triangle are colored by either $\{0,0,0\}$ or $\{0,1,1\}$. Obviously, if the vertices of one triangle of $G^{\prime}$ is colored by $\{0,0,0\}$, then all vertices of $G^{\prime}$ must be colored by 0. We call this kind of coloring the trivial coloring, which always exists. What we concern about is the number of nontrivial colorings.

According to the coloring rule, for a nontrivial coloring, if one vertex in $G^{\prime}$ is colored by 0, then all neighbors of it have to be colored by 1, otherwise one obtains the trivial coloring. On the other hand, if a vertex in $G^{\prime}$ has color 1, then the neighbors of it must be colored by 0 and 1 alternatively. It follows immediately that if the degree of a vertex in $G^{\prime}$ is odd, it only can be assigned with 0. For example, if there exist two adjacent odd vertices in $G^{\prime}$, then $G^{\prime}$ admits no nontrivial $\mathbb{Z}_{2}$-coloring.

Now we give the proof of Theorem 1.3.

Now we consider the special case that $V(G^{\prime})$ contains exactly two odd vertices, say $v_{1}^{\prime}$ and $v_{2}^{\prime}$. According to our discussion above, if $c:\{v_{1}^{\prime},\cdots,v_{m}^{\prime}\}\to\mathbb{Z}_{2}$ is a nontrivial coloring then $c(v_{1}^{\prime})=c(v_{2}^{\prime})=0$. Let us assign 0 to $v_{1}^{\prime}$, then any path $p$ connecting $v_{1}^{\prime}$ and $v_{2}^{\prime}$ induces a color for $v_{2}^{\prime}$, denoted by $c_{p}(v_{2}^{\prime})$. If this induced color $c_{p}(v_{2}^{\prime})=0$, then we say $v_{1}^{\prime}$ and $v_{2}^{\prime}$ are compatible with respect to $p$. Otherwise, we say $v_{1}^{\prime}$ is incompatible with $v_{2}^{\prime}$ with respect to $p$. Since there are only two odd vertices, this induced color $c_{p}(v_{2}^{\prime})$ actually does not depend on the choice of $p$. If $v_{1}^{\prime},v_{2}^{\prime}$ are incompatible, then $G^{\prime}$ admits no nontrivial $\mathbb{Z}_{2}$-coloring. The following result tells us that this never happens.

As a corollary, we have the following result.

Based on his previous work [Fis1973], Fisk noticed that if a surface admits a triangulation with two odd vertices and they are adjacent, then this triangulation has no four coloring [Fis1978, Lemma 1]. Thanks to the celebrated work of Appel and Haken [App1977] (see also [RSS1997]), we know that the four-color theorem holds for $S^{2}$. Therefore, if a triangulation of $S^{2}$ has only two odd vertices, then they cannot be adjacent. Recently, this result was reproved by Izmestiev [Izm2015] using coloring monodromy. The main idea of Izmestiev’s proof can be sketched as follows. Consider $\mathbb{Z}_{3}$-colorings of the triangulation $G^{\prime}$, there exists a homomorphism from the fundamental group of the complement of the set of odd vertices to $S_{3}$. If the two odd vertices $\{v_{1}^{\prime},v_{k}^{\prime}\}$ are adjacent, we choose a triangle containing these two odd vertices as the beginning triangle, then the images of the two loops around these two odd vertices generates $S_{3}$. However, this is impossible since $\pi_{1}(S^{2}\backslash\{v_{1}^{\prime},v_{k}^{\prime}\})\cong\mathbb{Z}$, which is abelian.

Proposition 4.1 tells us that if a triangulation of $S^{2}$ has exactly two odd vertices, then these two odd vertices not only can not be adjacent, actually they must be compatible. It is interesting to point out that the converse of this result also holds.

More precisely, let us consider a convex $(2m+1)$-gon centered at $u_{0}^{\prime}$ and a convex $(2n+1)$-gon centered at $v_{0}^{\prime}$ on $S^{2}$, and label the vertices by $\{u_{1}^{\prime},\cdots,u_{2m+1}^{\prime}\}$ and $\{v_{1}^{\prime},\cdots,v_{2n+1}^{\prime}\}$ respectively. Adding edges $u_{0}^{\prime}u_{i}^{\prime}$ $(1\leq i\leq 2m+1)$ and edges $v_{0}^{\prime}v_{j}^{\prime}$ $(1\leq j\leq 2n+1)$ divides these two polygons into $2m+1$ and $2n+1$ triangles. Suppose these two polygons are connected by a sequence of adjacent triangles such that consecutive triangles share an edge, and $u_{0}^{\prime}$ and $v_{0}^{\prime}$ are compatible with respect to this sequence of triangles. See Figure 4 for an example, where all vertices with color 0 are indicated. Now, we obtain a triangulation $T$ of a polygon with only two interior vertices, both of which are of odd degree.

Zhiyun Cheng is supported by NSFC 12071034.