This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Two-connected signed graphs with maximum nullity at most two

Marina Arav Frank J. Hall Zhongshan Li Hein van der Holst111Project sponsored by the National Security Agency under Grant Number H98230-14-1-0152. The United States Government is authorized to reproduce and distribute reprints notwithstanding any copyright notation herein. Department of Mathematics and Statistics
Georgia State University
Atlanta, GA 30303, USA
Abstract

A signed graph is a pair (G,Σ)(G,\Sigma), where G=(V,E)G=(V,E) is a graph (in which parallel edges are permitted, but loops are not) with V={1,,n}V=\{1,\ldots,n\} and ΣE\Sigma\subseteq E. The edges in Σ\Sigma are called odd and the other edges of EE even. By S(G,Σ)S(G,\Sigma) we denote the set of all symmetric n×nn\times n matrices A=[ai,j]A=[a_{i,j}] with ai,j<0a_{i,j}<0 if ii and jj are adjacent and connected by only even edges, ai,j>0a_{i,j}>0 if ii and jj are adjacent and connected by only odd edges, ai,ja_{i,j}\in\mathbb{R} if ii and jj are connected by both even and odd edges, ai,j=0a_{i,j}=0 if iji\not=j and ii and jj are non-adjacent, and ai,ia_{i,i}\in\mathbb{R} for all vertices ii. The parameters M(G,Σ)M(G,\Sigma) and ξ(G,Σ)\xi(G,\Sigma) of a signed graph (G,Σ)(G,\Sigma) are the largest nullity of any matrix AS(G,Σ)A\in S(G,\Sigma) and the largest nullity of any matrix AS(G,Σ)A\in S(G,\Sigma) that has the Strong Arnold Property, respectively. In a previous paper, we gave a characterization of signed graphs (G,Σ)(G,\Sigma) with M(G,Σ)1M(G,\Sigma)\leq 1 and of signed graphs with ξ(G,Σ)1\xi(G,\Sigma)\leq 1. In this paper, we characterize the 22-connected signed graphs (G,Σ)(G,\Sigma) with M(G,Σ)2M(G,\Sigma)\leq 2 and the 22-connected signed graphs (G,Σ)(G,\Sigma) with ξ(G,Σ)2\xi(G,\Sigma)\leq 2.

keywords:
signed graph, nullity, symmetric
MSC:
05C22, 05C50, 15A03

1 Introduction

A signed graph is a pair (G,Σ)(G,\Sigma), where G=(V,E)G=(V,E) is a graph (in which parallel edges are permitted, but loops are not) and ΣE\Sigma\subseteq E. (We refer to [3] for the notions and concepts in graph theory.) The edges in Σ\Sigma are called odd and the other edges even. If V={1,2,,n}V=\{1,2,\ldots,n\}, we denote by S(G,Σ)S(G,\Sigma) the set of all real symmetric n×nn\times n matrices A=[ai,j]A=[a_{i,j}] with

  • 1.

    ai,j<0a_{i,j}<0 if ii and jj are adjacent and all edges between ii and jj are even,

  • 2.

    ai,j>0a_{i,j}>0 if ii and jj are adjacent and all edges between ii and jj are odd,

  • 3.

    ai,ja_{i,j}\in\mathbb{R} if ii and jj are connected by odd and even edges,

  • 4.

    ai,j=0a_{i,j}=0 if iji\not=j and ii and jj are non-adjacent, and

  • 5.

    ai,ia_{i,i}\in\mathbb{R} for all vertices ii.

In [1] we introduced for any signed graph (G,Σ)(G,\Sigma), among other parameters, the signed graph parameters MM and ξ\xi. For a signed graph (G,Σ)(G,\Sigma), M(G,Σ)M(G,\Sigma) is the maximum of the nullities of the matrices in S(G,Σ)S(G,\Sigma). In order to describe the parameter ξ\xi we need the notion of Strong Arnold Property (SAP for short). A matrix A=[ai,j]S(G,Σ)A=[a_{i,j}]\in S(G,\Sigma) has the SAP if X=0X=0 is the only symmetric matrix X=[xi,j]X=[x_{i,j}] such that xi,j=0x_{i,j}=0 if ii and jj are adjacent vertices or i=ji=j, and AX=0AX=0. Then ξ(G,Σ)\xi(G,\Sigma) is defined as the largest nullity of any matrix AS(G,Σ)A\in S(G,\Sigma) satisfying the SAP. It is clear that ξ(G,Σ)M(G,Σ)\xi(G,\Sigma)\leq M(G,\Sigma) for any signed graph (G,Σ)(G,\Sigma). This signed graph parameter ξ\xi is analogous to the parameter ξ\xi for simple graphs introduced by Barioli, Fallat, and Hogben [2].

If GG is a graph and UV(G)U\subseteq V(G), then δ(U)\delta(U) denotes the set of edges of GG with one end in UU and the other end in V(G)UV(G)-U. The symmetric difference of two sets AA and BB is the set AΔB=ABBAA\Delta B=A\setminus B\cup B\setminus A. If (G,Σ)(G,\Sigma) is a signed graph and UV(G)U\subseteq V(G), we say that (G,Σ)(G,\Sigma) and (G,ΣΔδ(U))(G,\Sigma\Delta\delta(U)) are sign-equivalent and call the operation ΣΣΔδ(U)\Sigma\to\Sigma\Delta\delta(U) re-signing on UU. Re-signing on UU amounts to performing a diagonal similarity on the matrices in S(G,Σ)S(G,\Sigma), and hence it does not affect M(G,Σ)M(G,\Sigma) and ξ(G,Σ)\xi(G,\Sigma).

Let (G,Σ)(G,\Sigma) be a signed graph. If HH is a subgraph of GG, then we say that HH is odd if ΣE(H)\Sigma\cap E(H) has an odd number of elements, otherwise we call HH even. Zaslavsky showed in [7] that two signed graphs are sign-equivalent if and only if they have the same set of odd cycles. Thus, two signed graphs (G,Σ)(G,\Sigma) and (G,Σ)(G,\Sigma^{\prime}) that have the same set of odd cycles have ξ(G,Σ)=ξ(G,Σ)\xi(G,\Sigma)=\xi(G,\Sigma^{\prime}).

In [1], we showed that a signed graph (G,Σ)(G,\Sigma) has M(G,Σ)1M(G,\Sigma)\leq 1 if and only if (G,Σ)(G,\Sigma) is sign-equivalent to a signed graph (H,)(H,\emptyset), where HH is a graph whose underlying simple graph is a path. Furthermore, we showed that a signed graph (G,Σ)(G,\Sigma) has ξ(G,Σ)1\xi(G,\Sigma)\leq 1 if and only if (G,Σ)(G,\Sigma) is sign-equivalent to a signed graph (H,)(H,\emptyset), where HH is a graph whose underlying simple graph is a disjoint union of paths. Observe that in case the signed graph (G,Σ)(G,\Sigma) is connected, M(G,Σ)1M(G,\Sigma)\leq 1 if and only if ξ(G,Σ)1\xi(G,\Sigma)\leq 1. In this paper, we characterize the class of 22-connected signed graphs (G,Σ)(G,\Sigma) with M(G,Σ)2M(G,\Sigma)\leq 2. We will see that this class coincides with the class of signed graphs (G,Σ)(G,\Sigma) with ξ(G,Σ)2\xi(G,\Sigma)\leq 2.

The above characterizations are extensions of results known for simple graphs to signed graphs. For a simple graph GG, denote by S(G)S(G) the set of all real symmetric n×nn\times n matrices A=[ai,j]A=[a_{i,j}] with ai,j0a_{i,j}\not=0 if ii and jj are connected by an edge, ai,j=0a_{i,j}=0 if iji\not=j and ii and jj are non-adjacent, and ai,ia_{i,i}\in\mathbb{R} for all vertices ii. The maximum nullity M(G)M(G) of a simple graph GG is the maximum of the nullities of the matrices in S(G)S(G). Fiedler [4] proves that a simple graph GG has M(G)1M(G)\leq 1 if and only if GG is a path. In [6], Johnson, Loewy, and Smith characterize the class of simple graphs GG with M(G)2M(G)\leq 2. Barioli, Fallat, and Hogben introduced in [2] the parameter ξ\xi. For a simple graph GG, ξ(G)\xi(G) is defined as the largest nullity of any matrix AS(G)A\in S(G) satisfying the SAP. In [2], they prove that a graph GG has ξ(G)1\xi(G)\leq 1 if and only if GG is a subgraph of a path. In [5], Hogben and van der Holst give a characterization of the class of simple graphs GG with ξ(G)2\xi(G)\leq 2.

2 The maximum nullity of some signed graphs

Contracting an edge ee with ends uu and vv in a graph GG means deleting ee and identifying the vertices uu and vv. A signed graph (H,Γ)(H,\Gamma) is a weak minor of a signed graph (G,Σ)(G,\Sigma) if (H,Γ)(H,\Gamma) can be obtained from (G,Σ)(G,\Sigma) by deleting edges and vertices, contracting edges, and re-signing around vertices. We use weak minor to distinguish it from minor in which only even edges are allowed to be contracted (possibly after re-signing around vertices). The parameter ξ\xi has the nice property that if (H,Γ)(H,\Gamma) is a weak minor of the signed graph (G,Σ)(G,\Sigma), then ξ(H,Γ)ξ(G,Σ)\xi(H,\Gamma)\leq\xi(G,\Sigma).

Let us now introduce some signed graphs. By KneK_{n}^{e} and KnoK_{n}^{o} we denote the signed graphs (Kn,)(K_{n},\emptyset) and (Kn,E(Kn))(K_{n},E(K_{n})), respectively. By Kn=K_{n}^{=} we denote the signed graph (G,Σ)(G,\Sigma), where GG is the graph obtained from KnK_{n} by adding to each edge an edge in parallel, and where Σ\Sigma is the set of edges of KnK_{n}. By K4iK_{4}^{i} we denote the signed graph (K4,{e})(K_{4},\{e\}), where ee is an edge of K4K_{4}. By K2,3eK_{2,3}^{e} and K2,3iK_{2,3}^{i}, we denote the signed graphs (K2,3,)(K_{2,3},\emptyset) and (K2,3,{e})(K_{2,3},\{e\}), where ee is an edge of K2,3K_{2,3}, respectively.

The following lemma follows from Proposition 4 in [1].

Lemma 1.

M(Kn=)=ξ(Kn=)=nM(K_{n}^{=})=\xi(K_{n}^{=})=n.

From Proposition 8 in [1], the following lemma follows.

Lemma 2.

M(Kne)=ξ(Kne)=n1M(K_{n}^{e})=\xi(K_{n}^{e})=n-1 and M(Kno)=ξ(Kno)=n1M(K_{n}^{o})=\xi(K_{n}^{o})=n-1.

The following lemma follows from Proposition 34 in [1].

Lemma 3.

M(K4i)=ξ(K4i)=2M(K_{4}^{i})=\xi(K_{4}^{i})=2.

The following lemma follows from Proposition 35 in [1].

Lemma 4.

M(K2,3e)=ξ(K2,3e)=3M(K_{2,3}^{e})=\xi(K_{2,3}^{e})=3 and M(K2,3i)=ξ(K2,3i)=2M(K_{2,3}^{i})=\xi(K_{2,3}^{i})=2.

From Lemmas 2 and 4, it follows that signed graphs (G,Σ)(G,\Sigma) with ξ(G,Σ)2\xi(G,\Sigma)\leq 2 cannot have a weak K4eK_{4}^{e}-, K4oK_{4}^{o}-, or K2,3eK_{2,3}^{e}-minor. If a signed graph (G,Σ)(G,\Sigma) has no weak K4eK_{4}^{e}-, K4oK_{4}^{o}-, or K2,3eK_{2,3}^{e}-minor, the graph GG can still have a K4K_{4}- or K2,3K_{2,3}-minor. However, these minors force the signed graph to have additional structure. We will study this in the next section.

By W4W_{4} we denote the graph obtained from C4C_{4} by adding a new vertex vv, called the hub, and connecting it to each vertex of C4C_{4}. The subgraph C4C_{4} in W4W_{4} is called the rim of W4W_{4}. Any edge between vv and a vertex of the rim of W4W_{4} is called a spoke. Let e1,e2e_{1},e_{2} be two nonadjacent edges of the C4C_{4} in W4W_{4}. By W4oW_{4}^{o}, we denote the signed graph (W4,{e1,e2})(W_{4},\{e_{1},e_{2}\}). See Figure 1 for a picture of W4oW_{4}^{o}; here a bold edge is an odd edge and a thin edge an even edge. This signed graph appears as a special case in the characterization of 22-connected signed graphs (G,Σ)(G,\Sigma) with M(G,Σ)2M(G,\Sigma)\leq 2.

Refer to caption
Figure 1: The signed four-wheel.
Lemma 5.

M(W4o)=ξ(W4o)=2M(W_{4}^{o})=\xi(W_{4}^{o})=2

Proof.

Let v1,v2,v3,v4v_{1},v_{2},v_{3},v_{4} be the vertices on the rim of W4W_{4} in this cyclic order, and let vv be the hub of W4W_{4}. We assume that the edges between v1v_{1} and v2v_{2} and between v3v_{3} and v4v_{4} are even. Suppose for a contradiction that M(W4o)3M(W_{4}^{o})\geq 3. Then there exists a matrix A=[ai,j]S(W4o)A=[a_{i,j}]\in S(W_{4}^{o}) with nullity 3\geq 3. Since nullity(A)3\mbox{nullity}(A)\geq 3, there exist nonzero vectors x,yker(A)x,y\in\ker(A) with xv1=xv2=0x_{v_{1}}=x_{v_{2}}=0 and yv1=yv4=0y_{v_{1}}=y_{v_{4}}=0. If xv=0x_{v}=0, then from Ax=0Ax=0 it would follow that xv4=xv3=0x_{v_{4}}=x_{v_{3}}=0. This contradiction shows that xv0x_{v}\not=0. We may assume that xv>0x_{v}>0. Then, since the edge between vv and v1v_{1} is even, the edge between v1v_{1} and v4v_{4} is odd, and Ax=0Ax=0, it follows that xv4>0x_{v_{4}}>0. Similarly, xv3>0x_{v_{3}}>0. Let ava_{v} denote the row of AA corresponding to vv. Then avx=0a_{v}x=0, and, since xv>0x_{v}>0, xv4>0x_{v_{4}}>0, xv3>0x_{v_{3}}>0, it follows that av,v>0a_{v,v}>0.

We will now do the same with the vector yy. If yv=0y_{v}=0, then it would follow that yv2=yv3=0y_{v_{2}}=y_{v_{3}}=0. This contradiction shows that yv0y_{v}\not=0. We may assume that yv>0y_{v}>0. Then, since the edge between v1v_{1} and v2v_{2} is even, the edge between v1v_{1} and vv is even, and Ay=0Ay=0, it follows that yv2<0y_{v_{2}}<0. Similarly, yv3<0y_{v_{3}}<0. Since avy=0a_{v}y=0 and yv>0y_{v}>0, yv2<0y_{v_{2}}<0, yv3<0y_{v_{3}}<0, it follows that av,v<0a_{v,v}<0. We have arrived at a contradiction, and we can conclude that ξ(W4o)M(W4o)2\xi(W_{4}^{o})\leq M(W_{4}^{o})\leq 2. Since W4oW_{4}^{o} contains an odd cycle, M(W4o)ξ(W4o)ξ(K2=)=2M(W_{4}^{o})\geq\xi(W_{4}^{o})\geq\xi(K_{2}^{=})=2, and hence M(W4o)=ξ(W4o)=2M(W_{4}^{o})=\xi(W_{4}^{o})=2. ∎

3 Wide separations

Let (G,Σ)(G,\Sigma) be a signed graph. A pair [G1,G2][G_{1},G_{2}] of subgraphs of GG is a wide separation of (G,Σ)(G,\Sigma) if there exists an odd 44-cycle C4C_{4} such that G1C4G2=GG_{1}\cup C_{4}\cup G_{2}=G, E(G1)E(C4)=E(G_{1})\cap E(C_{4})=\emptyset, E(G2)E(C4)=E(G_{2})\cap E(C_{4})=\emptyset, V(G1)V(G2)=V(G_{1})\cap V(G_{2})=\emptyset, V(G1)V(C4)={r1,r2}V(G_{1})\cap V(C_{4})=\{r_{1},r_{2}\} and V(G2)V(C4)={s1,s2}V(G_{2})\cap V(C_{4})=\{s_{1},s_{2}\}, where r1r_{1} and r2r_{2} are nonadjacent vertices of C4C_{4} and s1s_{1} and s2s_{2} are nonadjacent vertices of C4C_{4}. We call r1,r2r_{1},r_{2} the vertices of attachment of G1G_{1} and s1,s2s_{1},s_{2} the vertices of attachment of G2G_{2} in the wide separation. See Figure 2 for an illustration. Here a bold edge is an odd edge and a thin edge an even edge.

Refer to caption
Figure 2: A wide separation [G1,G2][G_{1},G_{2}].
Lemma 6.

Let (G,Σ)(G,\Sigma) be a signed graph with no weak minor isomorphic to K4eK_{4}^{e}, K4oK_{4}^{o}, or K2,3iK_{2,3}^{i}. If GG has a K4K_{4}-minor, but no W4W_{4}-minor, then (G,Σ)(G,\Sigma) has a wide separation.

Proof.

Since GG has a K4K_{4}-minor and all vertices of K4K_{4} have degree three, GG has a subgraph isomorphic to a subdivision of K4K_{4}. Hence there are distinct vertices v1,v2,v3,v4v_{1},v_{2},v_{3},v_{4} and openly disjoint paths P1,,P6P_{1},\ldots,P_{6} of length 1\geq 1 in GG, where P1P_{1} has ends v1v_{1} and v2v_{2}, P2P_{2} has ends v1v_{1} and v3v_{3}, P3P_{3} has ends v1v_{1} and v4v_{4}, P4P_{4} has ends v2v_{2} and v3v_{3}, P5P_{5} has ends v2v_{2} and v4v_{4}, and P6P_{6} has ends v3v_{3} and v4v_{4}. Let HH be the subgraph spanned by v1,v2,v3,v4v_{1},v_{2},v_{3},v_{4} and P1,,P6P_{1},\ldots,P_{6}. Since (G,Σ)(G,\Sigma) has no weak K4eK_{4}^{e}- or K4oK_{4}^{o}-minor, (H,ΣE(H))(H,\Sigma\cap E(H)) has no weak K4eK_{4}^{e}- or K4oK_{4}^{o}-minor.

By re-signing if needed, we may assume that, in (G,Σ)(G,\Sigma), P1P_{1} is an odd path and that P2,,P6P_{2},\ldots,P_{6} are even paths. We can see that the paths P1P_{1}, P2P_{2}, P5P_{5}, and P6P_{6} consist of single edges as follows. P6P_{6} is a single edge, for otherwise, possibly after re-signing, we contract an odd edge in P6P_{6} and obtain a signed graph that contains K4oK_{4}^{o} as a weak minor. P1P_{1} is a single edge, for otherwise, we contract an odd edge in P1P_{1} and obtain a signed graph that contains K4eK_{4}^{e} as a weak minor. If both P3P_{3} and P5P_{5} have length 2\geq 2, then, possibly after re-signing, we contract an odd edge in P3P_{3} and in P5P_{5}, and obtain a signed graph that has K4oK_{4}^{o} as a weak minor. Hence at least one of P3P_{3} and P5P_{5} consists of a single edge. In the same way, at least one of P2P_{2} and P4P_{4} consists of a single edge. If both P2P_{2} and P3P_{3} have length 2\geq 2, then possibly after re-signing, we contract an odd edge in P2P_{2} and in P3P_{3}, and obtain a signed graph that has K4oK_{4}^{o} as a weak minor. Hence at least one of P2P_{2} and P3P_{3} consists of a single edge. In the same way, at least one of P4P_{4} and P5P_{5} consists of a single edge. If one of P2P_{2} and P5P_{5} has length 2\geq 2, then P3P_{3} and P4P_{4} consist each of a single edge. Hence, we can conclude that both P2P_{2} and P5P_{5} consist of single edges, or that both P3P_{3} and P4P_{4} consist of single edges. By symmetry we may assume that P2P_{2} and P5P_{5} consist of single edges.

Since GG has no W4W_{4}-minor, each path connecting a vertex of P3P_{3} to a vertex of P4P_{4} must contain at least one vertex of {v1,v4}\{v_{1},v_{4}\} and at least one vertex of {v2,v3}\{v_{2},v_{3}\}. Thus (G,Σ)(G,\Sigma) has a wide separation. ∎

The graph with two vertices and no edges is denoted by K2cK_{2}^{c}.

Lemma 7.

Let (G,Σ)(G,\Sigma) be a graph with no weak K2,3eK_{2,3}^{e}-minor. If GG has a K2,3K_{2,3}-minor but no K4K_{4}-minor, then (G,Σ)(G,\Sigma) has a wide separation [G1,G2][G_{1},G_{2}], where G1G_{1} is isomorphic to K2cK_{2}^{c}.

Proof.

Since GG has a K2,3K_{2,3}-minor, GG has a subgraph HH isomorphic to a subdivision of K2,3K_{2,3}. Hence there are vertices v1,v2,v3,v4,v5v_{1},v_{2},v_{3},v_{4},v_{5} and openly disjoint paths P1,,P6P_{1},\ldots,P_{6} of length 1\geq 1 in HH, where P1P_{1} has ends v1v_{1} and v2v_{2}, P2P_{2} has ends v1v_{1} and v3v_{3}, P3P_{3} has ends v1v_{1} and v4v_{4}, P4P_{4} has ends v2v_{2} and v5v_{5}, P5P_{5} has ends v3v_{3} and v5v_{5}, and P6P_{6} has ends v4v_{4} and v5v_{5}. We now view the paths P1,,P6P_{1},\ldots,P_{6} as paths in the signed graph (H,ΣE(H))(H,\Sigma\cap E(H)). As (G,Σ)(G,\Sigma) has no weak K2,3eK_{2,3}^{e}-minor, (G,Σ)(G,\Sigma) has a weak K2,3iK_{2,3}^{i}-minor. Hence we may re-sign (H,ΣE(H))(H,\Sigma\cap E(H)) such that P1P_{1} is odd and P2,,P6P_{2},\ldots,P_{6} are even.

P1P_{1} is a single edge, for otherwise, possibly after re-signing, we contract an odd edge in P1P_{1} and obtain a signed graph that contains K2,3eK_{2,3}^{e} as a weak minor. Similarly, P4P_{4} is a single edge. If both P2P_{2} and P3P_{3} have length 2\geq 2, then, possibly after re-signing, we contract an odd edge in P2P_{2} and in P3P_{3}, and obtain a signed graph that has K2,3eK_{2,3}^{e} as a weak minor. Hence, at least one of P2P_{2} and P3P_{3} consists of a single edge. Similarly, at least one of P5P_{5} and P6P_{6} consists of a single edge, at least one of P2P_{2} and P6P_{6} consists of a single edge, and at least one of P3P_{3} and P5P_{5} consists of a single edge. Hence at most one path of P2,P3,P5,P6P_{2},P_{3},P_{5},P_{6} has length 2\geq 2. By symmetry, we may assume that each path of P2,P3,P5P_{2},P_{3},P_{5} consists of a single edge. Let QQ be the concatenation of P3P_{3} and P6P_{6}. Since GG has no K4K_{4}-minor, there are no paths in GG connecting v2v_{2} and v3v_{3}, there are no paths in GG connecting v2v_{2} and an internal vertex of QQ, and there are no paths in GG connecting v3v_{3} and an internal vertex of QQ. Hence, (G,Σ)(G,\Sigma) has a wide separation [G1,G2][G_{1},G_{2}], where G1G_{1} is isomorphic to K2cK_{2}^{c}. ∎

4 The signed four-wheel

In this section we show that if (G,Σ)(G,\Sigma) is a signed graph with no weak K4eK_{4}^{e}-, K4oK_{4}^{o}-, or K2,3eK_{2,3}^{e}-minor, but GG has a W4W_{4}-minor, then the edges in each parallel class of (G,Σ)(G,\Sigma) have the same parity and, after removing all but one edge from each parallel class, we obtain W4oW_{4}^{o}.

Lemma 8.

Let (W4,Σ)(W_{4},\Sigma) be a signed graph with no weak K4eK_{4}^{e}-, K4oK_{4}^{o}-, or K2,3eK_{2,3}^{e}-minor. Then (W4,Σ)(W_{4},\Sigma) is sign-equivalent to W4oW_{4}^{o}.

Proof.

If at most one triangle of (W4,Σ)(W_{4},\Sigma) is even, then (W4,Σ)(W_{4},\Sigma) has a weak K4oK_{4}^{o}-minor. If at most one triangle of (W4,Σ)(W_{4},\Sigma) is odd, then (W,Σ)(W_{,}\Sigma) has a weak K4eK_{4}^{e}-minor. So we may assume that (W4,Σ)(W_{4},\Sigma) has exactly two odd triangles. If they share an edge, then (W4,Σ)(W_{4},\Sigma) has a weak K2,3eK_{2,3}^{e}-minor. If they do not share an edge, then (W4,Σ)(W_{4},\Sigma) is sign-equivalent to W4oW_{4}^{o}. ∎

Lemma 9.

Let (G,Σ)(G,\Sigma) be obtained from W4oW_{4}^{o} by adding an odd or even edge between nonadjacent vertices. Then (G,Σ)(G,\Sigma) has a weak K4oK_{4}^{o}- or K4eK_{4}^{e}-minor.

Proof.

Let v1,v2,v3,v4v_{1},v_{2},v_{3},v_{4} be the vertices on the rim of W4W_{4} in this order. Up to symmetry there is only one possibility to add an edge between two nonadjacent vertices in W4W_{4}. We may assume that we add the edge ee between v1v_{1} and v3v_{3}. If ee is an even edge, then the resulting signed graph has a weak K4oK_{4}^{o}-minor. If ee is an odd edge, then the resulting graph has a weak K4eK_{4}^{e}-minor. ∎

Lemma 10.

Let (G,Σ)(G,\Sigma) be a signed graph which has an edge whose contraction yields W4oW_{4}^{o}. Then (G,Σ)(G,\Sigma) has a weak K4oK_{4}^{o}- or K4eK_{4}^{e}-minor.

Proof.

If GG has a vertex of degree two, then GG arises from W4W_{4} by inserting a new vertex on an edge ee of W4W_{4}, which results in a path PP of length two in GG. If ee is an even edge on the rim of W4W_{4}, then, possibly after re-signing, we contract all but one odd edge of PP. The resulting signed graph contains a weak K4oK_{4}^{o}-minor. If ee is an odd edge on the rim of W4W_{4}, then, possibly after re-signing, we contract all but one even edge of PP. The resulting signed graph contains a weak K4eK_{4}^{e}-minor. If ee is a spoke of W4W_{4}, then, possibly after re-signing, we contract all but one odd edge of PP. The resulting signed graph contains a weak K4oK_{4}^{o}-minor.

So we may assume that GG has no vertex of degree two. Then GG is isomorphic either to the prism, C6cC_{6}^{c} (see Figure 3), or to K3,3K_{3,3}. Let ee be the edge in GG such that contracting ee yields W4W_{4}.

If GG is isomorphic to K3,3K_{3,3}, then (G,Σ)e(G,\Sigma)-e contains a weak K4oK_{4}^{o}-minor. Suppose next that GG is isomorphic to the prism. Then (G,Σ)(G,\Sigma) has either two odd triangles or two even triangles. If (G,Σ)(G,\Sigma) has two odd triangles, then, possibly after re-signing, we contract ee and another edge not on the two odd triangles, and obtain a signed graph that contains a K4oK_{4}^{o} as a signed subgraph. If (G,Σ)(G,\Sigma) has two even triangles, then, possibly after re-signing, we contract ee and another edge not on the two even triangles, and obtain a signed graph that contains a K4eK_{4}^{e} as a signed subgraph. ∎

Refer to caption
Figure 3: The prism with the edge ee
Lemma 11.

Let (G,Σ)(G,\Sigma) be a signed graph. If GG has a W4W_{4}-minor, then at least one of the following holds:

  1. 1.

    (G,Σ)(G,\Sigma) has a weak K4oK_{4}^{o}-, K4eK_{4}^{e}-, or K2,3eK_{2,3}^{e}-minor, or

  2. 2.

    the edges in each parallel class of (G,Σ)(G,\Sigma) have the same parity and, after removing from each parallel class all but one edge, we obtain W4oW_{4}^{o}.

Proof.

Suppose for a contradiction that there exists a signed graph (G,Σ)(G,\Sigma) such that GG has a W4W_{4}-minor, but (G,Σ)(G,\Sigma) has no weak K4oK_{4}^{o}-, K4eK_{4}^{e}-, or K2,3eK_{2,3}^{e}-minor and it is not the case that the edges in each parallel class of (G,Σ)(G,\Sigma) have the same parity and, after removing from each parallel class all but one edge, we obtain W4oW_{4}^{o}. We take (G,Σ)(G,\Sigma) with |V(G)|+|E(G)||V(G)|+|E(G)| as small as possible.

If the underlying simple graph of GG is isomorphic to W4W_{4}, then there must be parallel edges of different parity in (G,Σ)(G,\Sigma). In this case (G,Σ)(G,\Sigma) has a weak K4oK_{4}^{o}-, K4eK_{4}^{e}-, or K2,3eK_{2,3}^{e}-minor by Lemma 8. So we may assume that the underlying simple graph of GG is not isomorphic to W4W_{4}. Since GG has a W4W_{4}-minor and GG is connected, there exists a signed graph (H,Ω)(H,\Omega) such that (H,Ω)(H,\Omega) is a minor of (G,Σ)(G,\Sigma), the underlying simple graph of HH is not isomorphic to W4W_{4}, and W4oW_{4}^{o} can be obtained from (H,Ω)(H,\Omega) by deleting or contracting one edge. If W4oW_{4}^{o} can be obtained from (H,Ω)(H,\Omega) by deleting an edge ee, then ee connects nonadjacent vertices of W4oW_{4}^{o}. In this case, (H,Ω)(H,\Omega) has a weak K4oK_{4}^{o}- or K4eK_{4}^{e}-minor by Lemma 9, and hence (G,Σ)(G,\Sigma) has a K4oK_{4}^{o}- or K4eK_{4}^{e}-minor. If W4oW_{4}^{o} can be obtained from (H,Ω)(H,\Omega) by contracting an edge ee, then (H,Ω)(H,\Omega) has a weak K4oK_{4}^{o}- or K4eK_{4}^{e}-minor by Lemma 10, and hence (G,Σ)(G,\Sigma) has a weak K4oK_{4}^{o}- or K4eK_{4}^{e}-minor. ∎

5 Partial wide 22-path

In this section, we first make some new definitions. A sided 22-path is defined recursively as follows:

  1. 1.

    Let TT be a triangle and let \mathcal{F} be a set of two distinct edges in this triangle. Then (T,)(T,\mathcal{F}) is a sided 22-path.

  2. 2.

    If (G,)(G,\mathcal{F}) is a sided 22-path and HH is obtained from GG by adding edges parallel to the edges in \mathcal{F}, then (H,)(H,\mathcal{F}) is a sided 22-path.

  3. 3.

    Let (G,)(G,\mathcal{F}) be a sided 22-path and let ee and ff be distinct edges in a disjoint triangle TT. If HH is obtained from GG by identifying the edge ff of TT with an edge hh in \mathcal{F}, then (H,({h}){e})(H,(\mathcal{F}\setminus\{h\})\cup\{e\}) is a sided 22-path.

The edges in \mathcal{F} are called the sides of the sided 22-path. A 22-path is a graph GG for which there exists a set \mathcal{F} of two distinct edges of GG such that (G,)(G,\mathcal{F}) is a sided 22-path. A partial 22-path is a subgraph of a 22-path. A 22-connected partial 22-path with no parallel edges is the same as a linear singly edge articulated cycle graph (LSEAC), a type of graph introduced by Johnson et al. [6], and it is the same as a linear 2-tree, a type of graph introduced by Hogben and van der Holst [5].

Lemma 12.

Let GG be a 22-connected graph with no K4K_{4}-, K2,3K_{2,3}-, or K3=K_{3}^{=}-minor. Then GG is a partial 22-path.

Proof.

Since GG has no K4K_{4}- and no K2,3K_{2,3}-minor, GG is outerplanar. Hence GG can be embedded in the plane such that all its vertices are incident to the infinite face. Add edges such that all finite faces are either triangles or cycles with exactly two edges, and let the resulting graph be HH. Construct the following tree RR . The vertices of RR are all finite faces of the plane embedding. Connect two vertices of the tree if the corresponding faces have an edge in common. Then RR is a path. For if not, there would be a face that has edges in common with at least three other faces. Such a graph has a K3=K_{3}^{=}-minor. By induction it can now be shown that HH is a 22-path. Hence GG is a partial 22-path. ∎

A pair {e,f}\{e,f\} of nonadjacent edges in K4iK_{4}^{i} is called split if both ee and ff belong to an even and an odd triangle.

A sided wide 22-path [(G,Σ),][(G,\Sigma),\mathcal{F}] is defined recursively as follows:

  1. 1.

    Let (G,Σ)(G,\Sigma) be an even or odd triangle or a K4iK_{4}^{i}. If (G,Σ)(G,\Sigma) is a triangle, let \mathcal{F} be two distinct edges in this triangle. If (G,Σ)=K4i(G,\Sigma)=K_{4}^{i}, let \mathcal{F} be a split pair of edges in K4iK_{4}^{i}. Then [(G,Σ),][(G,\Sigma),\mathcal{F}] is a sided wide 22-path.

  2. 2.

    If [(G,Σ),][(G,\Sigma),\mathcal{F}] is a sided wide 22-path and (H,Ω)(H,\Omega) is obtained from (G,Σ)(G,\Sigma) by adding odd and even edges parallel to edges in \mathcal{F}, then [(H,Ω),][(H,\Omega),\mathcal{F}] is a sided wide 22-path.

  3. 3.

    Let [(G,Σ),][(G,\Sigma),\mathcal{F}] be a sided wide 22-path and let ee and ff be distinct edges in an even or odd triangle TT. If (H,Ω)(H,\Omega) is obtained from (G,Σ)(G,\Sigma) by identifying the edge ff of TT with an edge hh in \mathcal{F}, then [(H,Ω),({h}){e}][(H,\Omega),(\mathcal{F}\setminus\{h\})\cup\{e\}] is a sided wide 22-path.

  4. 4.

    Let [(G,Σ),][(G,\Sigma),\mathcal{F}] be a sided wide 22-path and let {e,f}\{e,f\} be a split pair of edges in K4iK_{4}^{i}. If (H,Ω)(H,\Omega) is obtained from (G,Σ)(G,\Sigma) by identifying the edge ff of K4iK_{4}^{i} with an edge hh in \mathcal{F}, then [(H,Ω),({h}){e}][(H,\Omega),(\mathcal{F}\setminus\{h\})\cup\{e\}] is a sided wide 22-path.

The edges in \mathcal{F} are called the sides of the sided wide 22-path. A wide 22-path is a signed graph (G,Σ)(G,\Sigma) for which there exists a set \mathcal{F} of two distinct edges of (G,Σ)(G,\Sigma) such that [(G,Σ),][(G,\Sigma),\mathcal{F}] is a sided wide 22-path. A signed graph (G,Σ)(G,\Sigma) is a partial wide 22-path if it is a spanning subgraph of a wide 22-path. Observe that if GG is a partial 22-path, then (G,Σ)(G,\Sigma) is a partial wide 22-path.

Lemma 13.

Let (G,Σ)(G,\Sigma) be a 22-connected signed graph. If (G,Σ)(G,\Sigma) has no minor isomorphic to K4eK_{4}^{e}-, K4oK_{4}^{o}-, K2,3eK_{2,3}^{e}, or K3=K_{3}^{=}, then, after removing in each parallel class all but one edge of the same parity, (G,Σ)(G,\Sigma) is either isomorphic to W4oW_{4}^{o} or (G,Σ)(G,\Sigma) is a partial wide 22-path.

Proof.

If GG has a W4W_{4}-minor, then (G,Σ)(G,\Sigma) is isomorphic to W4oW_{4}^{o}, by Lemma 11. We may therefore assume that GG has no W4W_{4}-minor.

Suppose GG has a K4K_{4}-minor. Then, by Lemma 6, (G,Σ)(G,\Sigma) has a wide separation [G1,G2][G_{1},G_{2}]. For i=1,2i=1,2, let (Hi,Ωi)(H_{i},\Omega_{i}) be obtained from (Gi,E(Gi)Σ)(G_{i},E(G_{i})\cap\Sigma) by adding between the vertices of attachment of (Gi,E(Gi)Σ)(G_{i},E(G_{i})\cap\Sigma) in the wide separation [G1,G2][G_{1},G_{2}] an odd and even edge in parallel. Then (Hi,Ωi)(H_{i},\Omega_{i}), i=1,2i=1,2, contains no weak minor isomorphic to K4eK_{4}^{e}, K4oK_{4}^{o}, K2,3eK_{2,3}^{e}, K3=K_{3}^{=}, or W4oW_{4}^{o}, for otherwise (G,Σ)(G,\Sigma) would contain a weak minor isomorphic to K4eK_{4}^{e}, K4oK_{4}^{o}, K2,3eK_{2,3}^{e}, K3=K_{3}^{=}, or W4oW_{4}^{o}.

Let uu and vv be the vertices of attachment of (G1,E(G1)Σ)(G_{1},E(G_{1})\cap\Sigma) in the wide separation [G1.G2][G_{1}.G_{2}]. Suppose G1{u,v}G_{1}-\{u,v\} contains more than one component; let C1,,CkC_{1},\ldots,C_{k} be the components. Then, as GG is 22-connected, each G1[V(Ci){u,v}]G_{1}[V(C_{i})\cup\{u,v\}] contains a path of length 2\geq 2 connecting uu and vv. If there are components CjC_{j} and CkC_{k} such that both G1[V(Cj){u,v}]G_{1}[V(C_{j})\cup\{u,v\}] and G1[V(Ck){u,v}]G_{1}[V(C_{k})\cup\{u,v\}] contain an even path connecting uu and vv, then (G,Σ)(G,\Sigma) has a K2,3eK_{2,3}^{e}-minor, a contradiction. Similarly, there are no two components CjC_{j} and CkC_{k} such that both G1[V(Cj){u,v}]G_{1}[V(C_{j})\cup\{u,v\}] and G1[V(Ck){u,v}]G_{1}[V(C_{k})\cup\{u,v\}] contain an odd path connecting uu and vv. Hence G1{u,v}G_{1}-\{u,v\} has exactly two components C1C_{1} and C2C_{2}. If G[V(C1){u,v}]G[V(C_{1})\cup\{u,v\}] contains a path PP of length 3\geq 3 connecting uu and vv and G[V(C2){u,v}]G[V(C_{2})\cup\{u,v\}] contains an even path connecting uu and vv, then, possibly after re-signing, we contract an edge of PP to obtain an even path. Then (G,Σ)(G,\Sigma) has a K2,3eK_{2,3}^{e}-minor. The cases where G[V(C2){u,v}]G[V(C_{2})\cup\{u,v\}] has an odd path connecting uu and vv, and where G[V(C2){u,v}]G[V(C_{2})\cup\{u,v\}] has a path of length 3\geq 3 connecting uu and vv are similar.

If G[V(C1){u,v}]G[V(C_{1})\cup\{u,v\}] has parallel edges whose ends are not uu and vv, then G[V(C1){u,v}]G[V(C_{1})\cup\{u,v\}] and G[V(C2){u,v}]G[V(C_{2})\cup\{u,v\}] contain paths connecting uu and vv of equal parity. Then (G,Σ)(G,\Sigma) has a K2,3eK_{2,3}^{e}. A similar statement holds for G[V(C2){u,v}]G[V(C_{2})\cup\{u,v\}]. Hence G[V(C1){u,v}]G[V(C_{1})\cup\{u,v\}] and G[V(C2){u,v}]G[V(C_{2})\cup\{u,v\}] are paths of length 22 and have different parity. Thus (H1,Ω1)(H_{1},\Omega_{1}) is a subgraph of a sided wide 22-path where one of the parallel edges between uu and vv is a side.

Suppose G1{u,v}G_{1}-\{u,v\} contains exactly one component. By induction (H1,Ω1)(H_{1},\Omega_{1}) is a partial wide 22-path. Since, in the construction of a wide 22-path, parallel edges appear only parallel to the edges in \mathcal{F} of a sided wide 22-path and (H1,Ω1){u,v}(H_{1},\Omega_{1})-\{u,v\} has exactly one component, there exists a set 1\mathcal{F}_{1} of two distinct edges, one of which is between uu and vv, of (H1,Ω1)(H_{1},\Omega_{1}) such that [(H1,Ω1),1][(H_{1},\Omega_{1}),\mathcal{F}_{1}] is a sided wide 22-path.

Similarly there exists a set 2\mathcal{F}_{2} of two distinct edge, one of which is between the vertices of attachments of (G2,Σ2)(G_{2},\Sigma_{2}) such that [(H2,Ω2),2][(H_{2},\Omega_{2}),\mathcal{F}_{2}] is a sided wide 22-path. Then (G,Σ)(G,\Sigma) is a partial wide 22-path. We may therefore assume that GG has no K4K_{4}-minor.

Suppose GG has a K2,3K_{2,3}-minor. Then, by Lemma 7, (G,Σ)(G,\Sigma) has a wide separation [G1,G2][G_{1},G_{2}], where G1G_{1} is isomorphic to K2cK_{2}^{c}. Let (H2,Ω2)(H_{2},\Omega_{2}) be obtained from (G2,E(G2)Σ)(G_{2},E(G_{2})\cap\Sigma) by adding between the vertices attachment of (G2,E(G2)Σ)(G_{2},E(G_{2})\cap\Sigma) in the wide separation [G1,G2][G_{1},G_{2}] an odd and even edge in parallel. Then (H2,Ω2)(H_{2},\Omega_{2}) contains no minor isomorphic to K4eK_{4}^{e}, K4oK_{4}^{o}, K2,3eK_{2,3}^{e}, K3=K_{3}^{=}, or W4oW_{4}^{o}. By induction (H2,Ω2)(H_{2},\Omega_{2}) is a partial wide 22-path. Similar as above, there is a sided wide 22-path [(J2,Δ2),2][(J_{2},\Delta_{2}),\mathcal{F}_{2}] such that (H2,Ω)(H_{2},\Omega) is a subgraph of (J2,Δ2)(J_{2},\Delta_{2}) and one of the edges of 2\mathcal{F}_{2} is an edge between the attachments of (G2,E(G2)Σ)(G_{2},E(G_{2})\cap\Sigma) in the wide separation [G1,G2][G_{1},G_{2}]. Then (G,Σ)(G,\Sigma) is a partial wide 22-path. We may therefore assume that GG has no K2,3K_{2,3}-minor.

Since GG has no K4K_{4}-, K2,3K_{2,3}-, or K3=K_{3}^{=}-minor, GG is a partial 22-path, and so (G,Σ)(G,\Sigma) is a partial wide 22-path. ∎

Lemma 14.

If (G,Σ)(G,\Sigma) is a 22-connected partial wide 22-path, then M(G,Σ)2M(G,\Sigma)\leq 2.

Proof.

Suppose for a contradiction that M(G,Σ)>2M(G,\Sigma)>2. Then there exists a matrix A=[ai,j]S(G,Σ)A=[a_{i,j}]\in S(G,\Sigma) with nullity(A)>2\mbox{nullity}(A)>2. Since (G,Σ)(G,\Sigma) is a partial wide 22-path, (G,Σ)(G,\Sigma) is a spanning subgraph of a wide 22-path (H,Ω)(H,\Omega).

If a wide separation [H1,H2][H_{1},H_{2}] in (H,Ω)(H,\Omega) does not yield a wide separation [G[V(H1)],G[V(H2)]][G[V(H_{1})],G[V(H_{2})]] in (G,Σ)(G,\Sigma), then we replace the K4iK_{4}^{i} in (H,Ω)(H,\Omega) by two adjacent triangles. We may therefore assume that for each wide separation [H1,H2][H_{1},H_{2}] in (H,Ω)(H,\Omega), [G[V(H1)],G[V(H2)]][G[V(H_{1})],G[V(H_{2})]] is a wide separation in (G,Σ)(G,\Sigma). For a vertex vv of (G,Σ)(G,\Sigma), we denote by ava_{v} the vvth row of AA.

Let s1,s2s_{1},s_{2} be the two ends of an edge ee in the set \mathcal{F} of the wide 22-path (H,Ω)(H,\Omega). As nullity(A)>2\mbox{nullity}(A)>2, there exists a nonzero vector xker(A)x\in\ker(A) with xs1=xs2=0x_{s_{1}}=x_{s_{2}}=0. If ee belongs to a triangle, let ff be the edge distinct from ee in the construction of the wide 22-path (H,Ω)(H,\Omega). Exactly one end r1r_{1} of ff belongs to {s1,s2}\{s_{1},s_{2}\}, let’s say r1=s1r_{1}=s_{1}, while the other end, r2r_{2}, is adjacent in (G,Σ)(G,\Sigma) to s2s_{2}. From as2x=0a_{s_{2}}x=0 it follows that xr2=0x_{r_{2}}=0.

If ee belongs to a K4iK_{4}^{i}, then ee belongs to an odd and even triangle of K4iK_{4}^{i}. Let r1r_{1} and r2r_{2} be the vertices of this K4iK_{4}^{i} distinct from s1s_{1} and s2s_{2}. By symmetry, we may assume that the edges s1r1s_{1}r_{1}, s2r1s_{2}r_{1}, and s2r2s_{2}r_{2} are even and that the edge s1r2s_{1}r_{2} is odd. Suppose xr1>0x_{r_{1}}>0. From as1x=0a_{s_{1}}x=0, it follows that xr2>0x_{r_{2}}>0. From as2x=0a_{s_{2}}x=0, it follows that xr1<0x_{r_{1}}<0. This contradiction shows that xr10x_{r_{1}}\leq 0. In the same way, it is not possible that xr1<0x_{r_{1}}<0. Hence xr1=0x_{r_{1}}=0. From as1x=0a_{s_{1}}x=0, it then follows that xr2=0x_{r_{2}}=0.

Repeating the above shows that x=0x=0, which contradicts that xx is a nonzero vector in ker(A)\ker(A). Thus M(G,Σ)2M(G,\Sigma)\leq 2. ∎

We now arrive at our main result.

Theorem 15.

Let (G,Σ)(G,\Sigma) be a 22-connected signed graph. Then the following are equivalent:

  1. (i)

    M(G,Σ)2M(G,\Sigma)\leq 2,

  2. (ii)

    ξ(G,Σ)2\xi(G,\Sigma)\leq 2,

  3. (iii)

    (G,Σ)(G,\Sigma) has no minor isomorphic to K3=K_{3}^{=}, K4eK_{4}^{e}, K4oK_{4}^{o}, or K2,3eK_{2,3}^{e}.

  4. (iv)

    (G,Σ)(G,\Sigma) is a partial wide 22-path or is isomorphic to W4oW_{4}^{o}.

Proof.

Since ξ(G,Σ)M(G,Σ)\xi(G,\Sigma)\leq M(G,\Sigma), it is clear that (i) implies (ii).

Suppose (G,Σ)(G,\Sigma) is a signed graph with ξ(G,Σ)2\xi(G,\Sigma)\leq 2. Since ξ(K3=)=ξ(K4e)=ξ(K4o)=ξ(K2,3e)=3\xi(K_{3}^{=})=\xi(K_{4}^{e})=\xi(K_{4}^{o})=\xi(K_{2,3}^{e})=3, (G,Σ)(G,\Sigma) has no minor isomorphic to K3=K_{3}^{=}, K4eK_{4}^{e}, K4oK_{4}^{o}, or K2,3eK_{2,3}^{e}. Hence (ii) implies (iii).

Suppose the signed graph (G,Σ)(G,\Sigma) has no minor isomorphic to K3=K_{3}^{=}, K4eK_{4}^{e}, K4oK_{4}^{o}, or K2,3eK_{2,3}^{e}. Then, by Lemma 13, (G,Σ)(G,\Sigma) is either isomorphic to W4oW_{4}^{o} or (G,Σ)(G,\Sigma) is a partial wide 22-path.

If (G,Σ)(G,\Sigma) is a partial wide 22-path, then, by Lemma 14, M(G,Σ)2M(G,\Sigma)\leq 2. Since also M(W4o)2M(W_{4}^{o})\leq 2, by Lemma 5, (iv) implies (i). ∎

6 References

References

  • [1] M. Arav, F. Hall, Z. Li, and H. van der Holst. The inertia set of a signed graph. Linear Algebra Appl., 439(5):1506–1529, 2013.
  • [2] F. Barioli, S. Fallat, and L. Hogben. A variant on the graph parameters of Colin de Verdière: implications to the minimum rank of graphs. Electron. J. Linear Algebra, 13:387–404, 2005.
  • [3] R. Diestel. Graph Theory. Springer-Verlag, New York, second edition, 2000.
  • [4] M. Fiedler. A characterization of tridiagonal matrices. Linear Algebra and Appl., 2:191–197, 1969.
  • [5] L. Hogben and H. van der Holst. Forbidden minors for the class of graphs GG with ξ(G)2\xi(G)\leq 2. Linear Algebra Appl., 423(1):42–52, 2007.
  • [6] C. R. Johnson, R. Loewy, and P. A. Smith. The graphs for which the maximum multiplicity of an eigenvalue is two. Linear Multilinear Algebra, 57(7):713–736, 2009.
  • [7] T. Zaslavsky. Signed graphs. Discrete Appl. Math., 4(1):47–74, 1982.