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

Regular graphs with a complete bipartite graph as a star complementthanks: This work is supported by the National Natural Science Foundation of China (Grant No. 11971180), the Guangdong Provincial Natural Science Foundation (Grant No. 2019A1515012052).

Xiaona Fanga, Lihua Youa, Rangwei Wua, Yufei Huangb
aSchool of Mathematical Sciences, South China Normal University,
Guangzhou, 510631, P. R. China
bDepartment of Mathematics Teaching, Guangzhou Civil Aviation College,
Guangzhou, 510403, P. R. China
Corresponding author: [email protected]

Abstract: Let GG be a graph of order nn and μ\mu be an adjacency eigenvalue of GG with multiplicity k1k\geq 1. A star complement HH for μ\mu in GG is an induced subgraph of GG of order nkn-k with no eigenvalue μ\mu, and the vertex subset X=V(GH)X=V(G-H) is called a star set for μ\mu in GG. The study of star complements and star sets provides a strong link between graph structure and linear algebra. In this paper, we study the regular graphs with Kt,s(st2)K_{t,s}\ (s\geq t\geq 2) as a star complement for an eigenvalue μ\mu, especially, characterize the case of t=3t=3 completely, obtain some properties when t=st=s, and propose some problems for further study.

Keywords: Adjacency eigenvalue; Star set; Star complement; Regular graph.

AMS classification: 05C50

1 Introduction

Let GG be a simple graph with vertex set V(G)={1,2,,n}=[n]V(G)=\left\{1,2,\dots,n\right\}=\left[n\right] and edge set E(G)E(G). The adjacency matrix of GG is an n×nn\times n matrix A(G)=(aij)A(G)=(a_{ij}), where aij=1a_{ij}=1 if vertex ii is adjacency to vertex jj, and 0 otherwise. We use the notation iji\sim j (iji\nsim j) to indicate that i,ji,j are adjacent (not-adjacent) and the notation dG(i)d_{G}(i) to indicate the degree of vertex ii in GG. The adjacency eigenvalues of GG are just the eigenvalues of A(G)A(G). For more details on graph spectra, see [6].

Let μ\mu be an eigenvalue of GG with multiplicity kk. A star set for μ\mu in GG is a subset XX of V(G)V(G) such that |X|=k\left|X\right|=k and μ\mu is not an eigenvalue of GXG-X, where GXG-X is the subgraph of GG induced by X¯=V(G)X\overline{X}=V(G)\setminus X. In this situation H=GXH=G-X is called a star complement corresponding to μ\mu. Star sets and star complements exist for any eigenvalue of a graph, and they need not to be unique. The basic properties of star sets are established in Chapter 77 of [7].

There is another equivalent geometric definition for star sets and star complements. Let GG be a graph with vertex set V(G)={1,,n}V(G)=\left\{1,\dots,n\right\} and adjacency matrix AA. Let {e1,,en}\left\{e_{1},\dots,e_{n}\right\} be the standard orthonormal basis of n\mathbb{R}^{n}, μ\mu be an eigenvalue of GG, and PP be the matrix which represents the orthogonal projection of n\mathbb{R}^{n} onto the eigenspace (μ)={xn:A(G)x=μx}\mathcal{E}(\mu)=\left\{x\in\mathbb{R}^{n}:A(G)x=\mu x\ \right\} of AA with respect to {e1,,en}\left\{e_{1},\dots,e_{n}\right\}. Since (μ)\mathcal{E}(\mu) is spanned by the vectors Pej(j=1,,n)Pe_{j}(j=1,\dots,n), there exists XV(G)X\subseteq V(G) such that the vectors Pej(jX)Pe_{j}(j\in X) form a basis for (μ)\mathcal{E}(\mu). Such a subset XX of V(G)V(G) is called a star set for μ\mu in GG. In this situation H=GXH=G-X is called a star complement for μ\mu.

For any graph GG of order nn with distinct eigenvalues λ1,,λm\lambda_{1},\dots,\lambda_{m}, there exists a partition V(G)=X1XmV(G)=X_{1}\bigcup\cdots\bigcup X_{m} such that XiX_{i} is a star set for eigenvalue λi(i=1,,m)\lambda_{i}\ (i=1,\dots,m). Such a partition is called a star partition of GG. For any graph GG, there exists at least one star partition ([10]). Each star partition determines a basis for n\mathbb{R}^{n} consisting of eigenvectors of an adjacency matrix. It provides a strong link between graph structure and linear algebra.

There are a lot of literatures about using star complements to construct and characterize certain graphs([1, 2, 3, 8, 12, 14, 16, 18, 19, 20, 21, 22, 23, 24, 25]), especially, regular graphs with a prescribed graph such as K1,sK_{1,s}, K1hKqK_{1}\triangledown hK_{q}, K2,5K_{2,5}, K2,sK_{2,s}, K1,1,tK_{1,1,t}, K1,1,1,tK_{1,1,1,t}, sK1Kt¯\overline{sK_{1}\cup K_{t}}, Pt(μ=1)P_{t}\ (\mu=1), Kr,r,r(μ=1)K_{r,r,r}\ (\mu=1) or Kr,s+tK1(μ=1)K_{r,s}+tK_{1}\ (\mu=1) as a star complement were well studied in the literature. Motivated by the above research, in this paper, we introduce the fundamental properties of the theory of star complements in Section 2, study the regular graphs with the bipartite graph Kt,s(st1)K_{t,s}\ (s\geq t\geq 1) as a star complement for an eigenvalue μ\mu in Section 3, completely characterize the regular graphs with K3,s(s3)K_{3,s}\ (s\geq 3) as a star complement for an eigenvalue μ\mu in Section 4, study some properties of Ks,sK_{s,s} in Section 5, and propose some problems for further research.

2 Preliminaries

In this section, we introduce some results of star sets and star complements that will be required in the sequel. The following fundamental result combines the Reconstruction Theorem ([7, Theorem 7.4.1]) with its converse ([7, Theorem 7.4.4]).

Theorem 2.1.

([7]) Let μ\mu be an eigenvalue of GG with multiplicity kk, XX be a set of vertices in the graph GG. Suppose that GG has adjacency matrix

(AXBTBC),\left(\begin{matrix}A_{X}&B^{T}\\ B&C\end{matrix}\right),

where AXA_{X} is the adjacency matrix of the subgraph induced by XX. Then XX is a star set for μ\mu in GG if and only if μ\mu is not an eigenvalue of CC and

μIAX=BT(μIC)1B.\mu I-A_{X}=B^{T}(\mu I-C)^{-1}B. (2.1)

In this situation, (μ)\mathcal{E}(\mu) consists of the vectors

(x(μIC)1Bx),\left(\begin{matrix}\textbf{x}\\ (\mu I-C)^{-1}B\textbf{x}\end{matrix}\right), (2.2)

where xk\textbf{x}\in\mathbb{R}^{k}.

Note that if XX is a star set for μ\mu, then the corresponding star complement H(=GX)H(=G-X) has adjacency matrix CC, and (2.1) tells us that GG can be determined by μ\mu, HH and the HH-neighbourhood of vertices in XX, where the HH-neighbourhood of the vertex uXu\in X, denoted by NH(u)N_{H}(u), is defined as NH(u)={vvu,vV(H)}N_{H}(u)=\left\{v\mid v\sim u,v\in V(H)\right\}.

It is usually convenient to apply (2.1) in the form

m(μ)(μIAX)=BTm(μ)(μIC)1B,m(\mu)(\mu I-A_{X})=B^{T}m(\mu)(\mu I-C)^{-1}B,

where m(x)m(x) is the minimal polynomial of CC. This is because m(μ)(μIC)1m(\mu)(\mu I-C)^{-1} is given explicitly as follows.

Proposition 2.2.

([8], Proposition 0.2) Let CC be a square matrix with minimal polynomial

m(x)=xd+1+cdxd+cd1xd1++c1x+c0.m(x)=x^{d+1}+c_{d}x^{d}+c_{d-1}x^{d-1}+\cdots+c_{1}x+c_{0}.

If μ\mu is not an eigenvalue of CC, then

m(μ)(μIC)1=adCd+ad1Cd1++a1C+a0I,m(\mu)(\mu I-C)^{-1}=a_{d}C^{d}+a_{d-1}C^{d-1}+\cdots+a_{1}C+a_{0}I,

where ad=1a_{d}=1 and for 0<id0<i\leq d, adi=μi+cdμi1+cd1μi2++cdi+1.a_{d-i}=\mu^{i}+c_{d}\mu^{i-1}+c_{d-1}\mu^{i-2}+\cdots+c_{d-i+1}.

In order to find all the graphs with a prescribed star complement HH for μ\mu, we need to find all solution AXA_{X}, BB for given μ\mu and CC. For any x,yq\textbf{x},\textbf{y}\in\mathbb{R}^{q}, where q=|V(H)|q=\left|V(H)\right|, let

x,y=xT(μIC)1y.\left\langle\textbf{x},\textbf{y}\right\rangle=\textbf{x}^{T}(\mu I-C)^{-1}\textbf{y}. (2.3)

Let bu\textbf{b}_{u} be the column of BB for any uXu\in X. By Theorem 2.1, we have

Corollary 2.3.

([10], Corollary 5.1.8 ) Suppose that μ\mu is not an eigenvalue of the graph HH, where |V(H)|=q\left|V(H)\right|=q. There exists a graph GG with a star set X={u1,u2,,uk}X=\left\{u_{1},u_{2},\dots,u_{k}\right\} for μ\mu such that GX=HG-X=H if and only if there exist (0,1)(0,1)-vectors bu1,bu2,,buk\textbf{b}_{u_{1}},\textbf{b}_{u_{2}},\dots,\textbf{b}_{u_{k}} in q\mathbb{R}^{q} such that
(1) bu,bu=μ\left\langle\textbf{b}_{u},\textbf{b}_{u}\right\rangle=\mu for all uXu\in X, and
(2) bu,bv={1,uv0,uv\left\langle\textbf{b}_{u},\textbf{b}_{v}\right\rangle=\left\{\begin{matrix}-1,&u\sim v\\ 0,&u\nsim v\end{matrix}\right. for all pairs u,vu,v in XX.

In view of the two equations in the above corollary, we have

Lemma 2.4.

([7]) Let XX be a star set for μ\mu in GG, and H=GXH=G-X.
(1) If μ0\mu\neq 0, then V(H)V(H) is a dominating set for GG, that is, the HH-neighbourhood of any vertex in XX are non-empty;
(2) If μ{1,0}\mu\notin\left\{-1,0\right\}, then V(H)V(H) is a location-dominating set for GG, that is, the HH-neighbourhood of distinct vertices in XX are distinct and non-empty.

It follows from (2)(2) of Lemma 2.4 that there are only finitely regular graphs with a prescribed star complement for μ{1,0}\mu\notin\left\{-1,0\right\}. If μ=0\mu=0 and XX has distinct vertices uu and vv with the same neighbourhood in GG, then uu and vv are called duplicate vertices. If μ=1\mu=-1 and XX has distinct vertices uu and vv with the same closed neighbourhood in GG, then uu and vv are called co-duplicate vertices (see [11]).

Recall that if the eigenspace (μ)\mathcal{E}(\mu) is orthogonal to the all-1 vector j then μ\mu is called a non-main eigenvalue. From (2.2), we have the following result.

Lemma 2.5.

([8], Proposition 0.3) The eigenvalue μ\mu is a non-main eigenvalue if and only if

bu,j=1 for all uX,\left\langle\textbf{b}_{u},\textbf{j}\ \right\rangle=-1\ \mbox{ for all }u\in X, (2.4)

where j is the all-1 vector.

Lemma 2.6.

([10], Corollary 3.9.12) In an rr-regular graph, all eigenvalues other than rr are non-main.

In the rest of this paper, we let HKt,s(st1)H\cong K_{t,s}\ (s\geq t\geq 1), (V,W)(V,W) be a bipartition of the graph Kt,sK_{t,s} with V={v1,v2,,vt}V=\left\{v_{1},v_{2},\dots,v_{t}\right\}, W={w1,w2,,ws}W=\left\{w_{1},w_{2},\dots,w_{s}\right\}. We say that a vertex uXu\in X is of type (a,b)(a,b) if it has aa neighbours in VV and bb neighbours in WW. Clearly (a,b)(0,0)(a,b)\neq(0,0) and 0at0\leq a\leq t, 0bs0\leq b\leq s.

Let CC be the adjacency matrix of HH, then CC has minimal polynomial m(x)=x(x2ts).m(x)=x(x^{2}-ts). Since μ\mu is not an eigenvalue of CC, we have μ0\mu\neq 0 and μ2ts\mu^{2}\neq ts. From Proposition 2.2, we have

m(μ)(μIC)1=C2+μC+(μ2ts)I.m(\mu)(\mu I-C)^{-1}=C^{2}+\mu C+(\mu^{2}-ts)I. (2.5)

If μ\mu is a non-main eigenvalue of GG, then by (2.4) and (2.5) we have

μ2(a+b)+μ(as+tb)=μ(μ2ts).\mu^{2}(a+b)+\mu(as+tb)=-\mu(\mu^{2}-ts). (2.6)

Using (2.5) to compute bu,bu=μ\left\langle\textbf{b}_{u},\textbf{b}_{u}\right\rangle=\mu, we obtain the following equation

(μ2ts)(a+b)+a2s+tb2+2abμ=μ2(μ2ts).(\mu^{2}-ts)(a+b)+a^{2}s+tb^{2}+2ab\mu=\mu^{2}(\mu^{2}-ts). (2.7)

Let u,vu,v be distinct vertices in XX of type (a,b)(a,b), (c,d)(c,d), respectively. Let ρuv=|NH(u)NH(v)|\rho_{uv}=\left|N_{H}(u)\cap N_{H}(v)\right|, auv=1a_{uv}=1 or 0 according as uvu\sim v or uvu\nsim v. Using (2.5) to compute bu,bv=auv\left\langle\textbf{b}_{u},\textbf{b}_{v}\right\rangle=-a_{uv}, we have

(μ2ts)ρuv+acs+bdt+μ(ad+bc)=μ(μ2ts)auv.(\mu^{2}-ts)\rho_{uv}+acs+bdt+\mu(ad+bc)=-\mu(\mu^{2}-ts)a_{uv}. (2.8)

3 Regular graphs with Kt,sK_{t,s} as a star complement

An rr-regular graph GG with nn vertices is said to be strongly regular with parameters (n,r,e,f)(n,r,e,f) if every two adjacent vertices in GG have ee common neighbours and every two non-adjacent vertices have ff common neighbours. For example, Petersen graph is strongly regular with parameters (10,3,0,1)(10,3,0,1).

For the regular graphs with the complete bipartite graph Kt,sK_{t,s} as a star complement, the case of t=1t=1 was solved by Rowlinson and Tayfeh-Rezaie in 2010 ([19]), the case of t=2,s=5t=2,\ s=5 was solved by Rowlinson and Jackson in 1999 ([18]), the case of t=2,s5t=2,\ s\neq 5 was solved by Yuan, Zhao, Liu and Chen in 2018 ([25]), and the conclusions are listed below.

Theorem 3.1.

([19]) If the rr-regular graph GG has K1,s(s>1)K_{1,s}\ (s>1) as a star complement for an eigenvalue μ\mu, then one of the following holds:
(1) μ=±2,r=s=2\mu=\pm 2,r=s=2 and GK2,2G\cong K_{2,2};
(2) μ=12(1±5),r=s=2\mu=\frac{1}{2}(-1\pm\sqrt{5}),r=s=2 and GG is a 5-cycle;
(3) μ+,r=s\mu\in\mathbb{N}_{+},r=s and GG is strongly regular with parameters ((μ2+3μ)2,μ(μ2+3μ+1),0,μ(μ+1))\left(\left(\mu^{2}+3\mu\right)^{2},\mu\left(\mu^{2}+3\mu+1\right),0,\mu(\mu+1)\right).

Theorem 3.2.

([18, 25]) Let s2s\geq 2. If the rr-regular graph GG has K2,sK_{2,s} as a star complement for an eigenvalue μ\mu, then one of the following holds:
(1) μ=±3,r=s=3\mu=\pm 3,r=s=3 and GK3,3G\cong K_{3,3};
(2) 1μ+,r=s1\neq\mu\in\mathbb{N}_{+},r=s and GG is an rr-regular graph of order (μ4+10μ3+27μ2+10μ)/4\left(\mu^{4}+10\mu^{3}+27\mu^{2}+10\mu\right)/4, where r=μ(μ+1)(μ+4)/2r=\mu(\mu+1)(\mu+4)/2.
(3) μ=1\mu=1, s=5s=5 and either GSch10G\cong Sch_{10} or GG is isomorphic to one of the eleven induced regular subgraphs of Sch10Sch_{10}.
(4) μ=1\mu=-1, r1(mod 2s1)r\equiv-1\ ({\rm mod}\ 2s-1) and GG(r)G\cong G^{\prime}(r) (see [25] for specific definitions).

In this section, we consider the general case. We prove that there is no regular graph GG with Kt,s(st1)K_{t,s}\ (s\geq t\geq 1) as a star complement for μ=t\mu=-t, characterize the graph GG when μ=r\mu=r, μ=1\mu=-1, and the case with all vertices in XX of type (0,b)(0,b) for μ{t,r,1}\mu\notin\{-t,r,-1\}. Furthermore, we propose a question for further research.

Proposition 3.3.

There is not an rr-regular graph GG with Kt,s(st1)K_{t,s}\ (s\geq t\geq 1) as a star complement for μ=t\mu=-t.

Proof. Let μ=t\mu=-t. Since μ2ts\mu^{2}\neq ts, we have sts\neq t and then s>ts>t. Let uXu\in X be a vertex of type (a,b)(a,b), thus (a,b)(0,0)(a,b)\neq(0,0) and 0at0\leq a\leq t, 0bs0\leq b\leq s.

If t=1t=1, from Theorem 2.2 of [19], there is no rr-regular graph GG with K1,sK_{1,s} as a star complement for μ=1\mu=-1.

If t2t\geq 2, by Lemma 2.6, we know that μ=t\mu=-t is a non-main eigenvalue of GG, and by (2.6), we have

t(ts)(at)=0.t(t-s)(a-t)=0. (3.1)

Since s>ts>t and t2t\geq 2, (3.1) implies that a=ta=t, and thus s(bt2)=b2tbt3+t2s(b-t^{2})=b^{2}-tb-t^{3}+t^{2} by (2.7).

If b=t2b=t^{2}, then t42t3+t2=0t^{4}-2t^{3}+t^{2}=0, thus t=0t=0 or 11, a contradiction.

If bt2b\neq t^{2}, then s=b2tbt3+t2bt2=bt2+t2(t1)2bt2+2t2ts=\frac{b^{2}-tb-t^{3}+t^{2}}{b-t^{2}}=b-t^{2}+\frac{t^{2}(t-1)^{2}}{b-t^{2}}+2t^{2}-t. If b<t2b<t^{2}, then s2t2(t1)2+2t2t=ts\leq-2\sqrt{t^{2}(t-1)^{2}}+2t^{2}-t=t, which contradicts with s>ts>t. Thus b>t2b>t^{2} and s(b+t1)=b(t1)2bt2>0s-(b+t-1)=\frac{b(t-1)^{2}}{b-t^{2}}>0. Considering degrees, we have

dG(v1)=dG(v2)==dG(vt)=s+|X|,d_{G}(v_{1})=d_{G}(v_{2})=\cdots=d_{G}(v_{t})=s+\left|X\right|,

and

dG(u)a+b+|X|1=b+t1+|X|,uX.d_{G}(u)\leq a+b+\left|X\right|-1=b+t-1+|X|,\ u\in X.

Hence, dG(v1)=dG(v2)==dG(vt)>dG(u)d_{G}(v_{1})=d_{G}(v_{2})=\cdots=d_{G}(v_{t})>d_{G}(u) which contradicts to the regularity of GG.

Combining the above arguments, there is not an rr-regular graph GG with Kt,s(st1)K_{t,s}\ (s\geq t\geq 1) as a star complement for μ=t\mu=-t.\quad\square

Theorem 3.4.

If the rr-regular graph GG has Kt,s(st1)K_{t,s}\ (s\geq t\geq 1) as a star complement for an eigenvalue μ=r\mu=r, then s=t=1s=t=1, GC3G\cong C_{3} or r=s=t+1r=s=t+1, GKt+1,t+1G\cong K_{t+1,t+1}.

Proof. Since μ\mu is not an eigenvalue of HKt,s(st1)H\cong K_{t,s}\ (s\geq t\geq 1), we have μ0\mu\neq 0 and μ2ts\mu^{2}\neq ts. By Lemma 2.4, V(Kt,s)V(K_{t,s}) is a location-dominating set, and so GG is connected.

By GG is rr-regular and connected, μ=r\mu=r, we know k=1k=1 and then |X|=1\left|X\right|=1. Since GG is regular, we have dG(v1)=dG(v2)==dG(vt)d_{G}(v_{1})=d_{G}(v_{2})=\cdots=d_{G}(v_{t}). Let X={u}X=\left\{u\right\}. Then either uv1u\sim v_{1}, uv2u\sim v_{2}, …,uvtu\sim v_{t}, or uv1u\nsim v_{1}, uv2u\nsim v_{2},…, uvtu\nsim v_{t}.

If uv1u\sim v_{1}, uv2u\sim v_{2}, …,uvtu\sim v_{t}, then dG(v1)=dG(v2)==dG(vt)=s+1,d_{G}(v_{1})=d_{G}(v_{2})=\cdots=d_{G}(v_{t})=s+1, which implies that dG(u)=s+1d_{G}(u)=s+1. It follows that the vertex uu is adjacent to st+1s-t+1(1\geq 1) vertices of WW, and thus t=1t=1 by dG(w1)=dG(w2)==dG(ws)d_{G}(w_{1})=d_{G}(w_{2})=\cdots=d_{G}(w_{s}). Since dG(w1)=t+1=dG(v1)d_{G}(w_{1})=t+1=d_{G}(v_{1}), we have s=t=1s=t=1 and GC3G\cong C_{3}.

If uv1u\nsim v_{1}, uv2u\nsim v_{2}, …, uvtu\nsim v_{t}, in view of the regularity, we have dG(u)=dG(v1)=dG(v2)==dG(vt)=s,d_{G}(u)=d_{G}(v_{1})=d_{G}(v_{2})=\cdots=d_{G}(v_{t})=s, and then dG(w1)=dG(w2)==dG(ws)=t+1d_{G}(w_{1})=d_{G}(w_{2})=\cdots=d_{G}(w_{s})=t+1. Hence we have s=r=t+1s=r=t+1 and GKt+1,t+1G\cong K_{t+1,t+1}.\square

Let HKt,sH\cong K_{t,s}, (V,W)(V,W) be a bipartition of the graph Kt,sK_{t,s} with V={v1,v2,,vt}V=\left\{v_{1},v_{2},\dots,v_{t}\right\}, W={w1,w2,,ws}W=\left\{w_{1},w_{2},\dots,w_{s}\right\}. We obtain an rr-regular graph G(r)G(r) with V(G(r))=XV(H)V(G(r))=X\cup V(H), X=V1VtW1WsX=V_{1}\cup\cdots\cup V_{t}\cup W_{1}\cup\cdots\cup W_{s} where ViV_{i} is the set of vertices of type (1,s)(1,s) adjacent to viVv_{i}\in V with |Vi|=(r+1)(s1)/(ts1)1|V_{i}|=(r+1)(s-1)/(ts-1)-1, ViV_{i} induces a clique for 1it1\leq i\leq t, WiW_{i} is the set of vertices of type (t,1)(t,1) adjacent to wiWw_{i}\in W with |Wi|=(r+1)(t1)/(ts1)1|W_{i}|=(r+1)(t-1)/(ts-1)-1, WiW_{i} induces a clique for 1is1\leq i\leq s. For any ii, jj, each vertex in ViV_{i} is adjacent to all vertices in WjW_{j}.

The greatest common divisor of aa and bb is denoted by gcd(a,b)\gcd(a,b). For μ=1\mu=-1, we have the following theorem.

Theorem 3.5.

If GG is an rr-regular graph with H=Kt,s(st2)H=K_{t,s}\ (s\geq t\geq 2) as a star complement for an eigenvalue 1-1, then r1(modts1gcd(s1,t1))r\equiv-1\ ({\rm mod}\frac{ts-1}{\gcd(s-1,t-1)}) and GG(r)G\cong G(r).

Proof. Since Kt,sK_{t,s} is connected and V(Kt,s)V(K_{t,s}) is a dominating set (see Lemma 2.4), we know GG is connected. Let HKt,sH\cong K_{t,s}, (V,W)(V,W) be a bipartition of the graph Kt,sK_{t,s} as above. Let uXu\in X be a vertex of type (a,b)(a,b), thus (a,b)(0,0)(a,b)\neq(0,0) and 0at0\leq a\leq t, 0bs0\leq b\leq s. Let μ=1\mu=-1 in (2.7), so that

(1ts)(a+b1)+a2s+tb22ab=0.(1-ts)(a+b-1)+a^{2}s+tb^{2}-2ab=0. (3.2)

By Lemma 2.6, we know that μ=1\mu=-1 is a non-main eigenvalue of GG, thus from (2.6), we have

1ts+a(s1)+b(t1)=0.1-ts+a(s-1)+b(t-1)=0. (3.3)

Combining (3.2) and (3.3), if a=1a=1, then b=sb=s; if a=ta=t, then b=1b=1; if 1<a<t1<a<t, then b=at+1,s=2tb=a-t+1,\ s=2-t or b=at,s=1tb=\frac{a}{t},\ s=\frac{1}{t}. It is obvious that s=2ts=2-t or s=1ts=\frac{1}{t} contradicts with s+s\in\mathbb{Z}_{+}. Therefore, the possible types of vertices in XX are (1,s)(1,s), (t,1)(t,1), and the feasible solution of (2.8) are shown in Table 1.

(a,b)(a,b) (c,d)(c,d) auva_{uv} ρuv\rho_{uv}
(1,s)(1,s) (1,s)(1,s) 0 ss
(1,s)(1,s) (1,s)(1,s) 11 s+1s+1
(t,1)(t,1) (t,1)(t,1) 0 tt
(t,1)(t,1) (t,1)(t,1) 11 t+1t+1
(1,s)(1,s) (t,1)(t,1) 11 22
Table 1: The feasible solution of (2.8)

We observe that when u,vu,v are of different types, they must be adjacent; when u,vu,v are of the same type, uvu\sim v if and only if they have the same HH-neighbourhoods, and thus u,vu,v are co-duplicate vertices. We can add arbitrarily many co-duplicate vertices when constructing graphs with a prescribed star complement for 1-1.

Now we partition the vertices in XX. Let ViV_{i} be the set of vertices of type (1,s)(1,s) in XX adjacent to viVv_{i}\in V, WiW_{i} be the set of vertices of type (t,1)(t,1) in XX adjacent to wiWw_{i}\in W. Then any two vertices in Vi(Wi)V_{i}\ (W_{i}) are co-duplicate vertices. We do not exclude the possibility that some of the sets ViV_{i}, WiW_{i} are empty. Then for any viVv_{i}\in V, we have dG(vi)=s+|Vi|+i=1s|Wi|;d_{G}(v_{i})=s+|V_{i}|+\sum\limits_{i=1}^{s}|W_{i}|; and for any wiWw_{i}\in W, we have dG(wi)=t+i=1t|Vi|+|Wi|.d_{G}(w_{i})=t+\sum\limits_{i=1}^{t}|V_{i}|+|W_{i}|.

Since GG is rr-regular, we have |V1|=|V2|==|Vt||V_{1}|=|V_{2}|=\cdots=|V_{t}| by dG(v1)=dG(v2)==dG(vt)d_{G}(v_{1})=d_{G}(v_{2})=\cdots=d_{G}(v_{t}) and |W1|=|W2|==|Ws||W_{1}|=|W_{2}|=\cdots=|W_{s}| by dG(w1)=dG(w2)==dG(ws)d_{G}(w_{1})=d_{G}(w_{2})=\cdots=d_{G}(w_{s}). Then we have

r=dG(v1)=s+|V1|+s|W1|andr=dG(w1)=t+t|V1|+|W1|.r=d_{G}(v_{1})=s+|V_{1}|+s\cdot|W_{1}|\ \mbox{and}\ r=d_{G}(w_{1})=t+t\cdot|V_{1}|+|W_{1}|.

It turns out that

|V1|=(s1)(r+1)ts11,|W1|=(t1)(r+1)ts11.|V_{1}|=\frac{(s-1)(r+1)}{ts-1}-1,\ |W_{1}|=\frac{(t-1)(r+1)}{ts-1}-1.

Since |V1||V_{1}|\in\mathbb{N}, |W1||W_{1}|\in\mathbb{N} and

gcd(t1,ts1)=gcd(s1,ts1)=gcd(t1,s1),\gcd(t-1,ts-1)=\gcd(s-1,ts-1)=\gcd(t-1,s-1),

we have r1(modts1gcd(s1,t1))r\equiv-1\ ({\rm mod}\ \frac{ts-1}{\gcd(s-1,t-1)}). Consequently we obtain an rr-regular graph G(r)G(r).\quad\square

Remark 3.6.

Note that if Vi=Vi{vi},viV\ V_{i}^{*}=V_{i}\cup\left\{v_{i}\right\},\ v_{i}\in V and Wi=Wi{wi},wiWW_{i}^{*}=W_{i}\cup\left\{w_{i}\right\},\ w_{i}\in W, then each of sets ViV_{i}^{*}, WiW_{i}^{*} induces a clique in G(r)G(r).

Next, we consider the case μ{1,t,r}\mu\notin\left\{-1,-t,r\right\}. The following lemma lists all possible types of vertices in XX.

Lemma 3.7.

Let GG be a graph with H=Kt,s(st1)H=K_{t,s}\ (s\geq t\geq 1) as a star complement for μ\mu. If μ\mu is a non-main eigenvalue of GG and μ{1,t}\mu\notin\left\{-1,-t\right\}, then the possible types of vertices in the star set XX are (a,μ3+tμ2ta+a2μ+a)(a,\frac{\mu^{3}+t\mu^{2}-ta+a^{2}}{\mu+a}), where 0at10\leq a\leq t-1 and aμa\neq-\mu.

Proof. Let uXu\in X be a vertex of type (a,b)(a,b), thus (a,b)(0,0)(a,b)\neq(0,0) and 0at0\leq a\leq t, 0bs0\leq b\leq s. By (2.6), (2.7) and μ{0,1,t}\mu\notin\{0,-1,-t\}, we have: (1) a=ta=t, b=μb=-\mu, s=μ2ts=\frac{\mu^{2}}{t}; (2) a=μa=-\mu, b=μ2tb=\frac{\mu^{2}}{t}, s=μ2ts=\frac{\mu^{2}}{t}; (3) 0at10\leq a\leq t-1, aμa\neq-\mu, {b=aμt,s=μ2t,\left\{\begin{matrix}[l]b=\frac{-a\mu}{t},\\ s=\frac{\mu^{2}}{t},\end{matrix}\right. or {b=μ3+tμ2ta+a2μ+a,s=μ4+(2t+1)μ3+(2a+t2)μ2+(2a2at)μ+a2tat2(ta)μ+ata2.\left\{\begin{matrix}[l]b=\frac{\mu^{3}+t\mu^{2}-ta+a^{2}}{\mu+a},\\ s=\frac{\mu^{4}+(2t+1)\mu^{3}+(2a+t^{2})\mu^{2}+(2a^{2}-at)\mu+a^{2}t-at^{2}}{(t-a)\mu+at-a^{2}}.\end{matrix}\right. Since μ\mu is not an eigenvalue of HH, we have μ2ts\mu^{2}\neq ts. Thus the possible types of vertices in the star set XX are (a,μ3+tμ2ta+a2μ+a)(a,\frac{\mu^{3}+t\mu^{2}-ta+a^{2}}{\mu+a}).\qquad\square

For μ{1,t,r}\mu\notin\{-1,-t,r\}, we consider the case a=0a=0 in the following by Lemma 3.7.

Theorem 3.8.

If the rr-regular graph GG has Kt,s(st1)K_{t,s}\ (s\geq t\geq 1) as a star complement for μ{1,t,r}\mu\notin\left\{-1,-t,r\right\} and all vertices in XX are of type (0,b)(0,b), then one of the following holds:
(1) μ=r,r=s=t+1\mu=-r,r=s=t+1 and GKt+1,t+1G\cong K_{t+1,t+1};
(2) μ=12(1±5),t=1,r=s=2\mu=\frac{1}{2}(-1\pm\sqrt{5}),t=1,r=s=2 and GG is a 55-cycle;
(3) μ+,r=s\mu\in\mathbb{N}_{+},r=s and GG is an rr-regular graph of order μ(μ+2t+1)(μ2+2tμ+μ+t2t)/t2\mu\left(\mu+2t+1\right)\left(\mu^{2}+2t\mu+\mu+t^{2}-t\right)/t^{2}, where r=μ(μ2+2tμ+μ+t2)/tr=\mu\left(\mu^{2}+2t\mu+\mu+t^{2}\right)/t.

Proof. Since μ\mu is not an eigenvalue of Kt,sK_{t,s}, we have μ0\mu\neq 0 and μ2ts\mu^{2}\neq ts. By Lemma 2.4, V(Kt,s)V(K_{t,s}) is a location-dominating set, and so GG is connected. Then by a=0a=0 and Lemma 3.7, we have

{b=μ2+tμ,s=μ(μ2+2tμ+μ+t2)/t.\left\{\begin{matrix}b=\mu^{2}+t\mu,\\ s=\mu\left(\mu^{2}+2t\mu+\mu+t^{2}\right)/t.\end{matrix}\right.

Now we consider H=Kt,μ(μ2+2tμ+μ+t2)/tH=K_{t,\mu\left(\mu^{2}+2t\mu+\mu+t^{2}\right)/t} and all vertices in XX are of type (0,μ2+tμ)(0,\mu^{2}+t\mu). Then r=sr=s. Counting the edges between XX and V(H)V(H), we have |X|(μ2+tμ)=s(rt).\left|X\right|(\mu^{2}+t\mu)=s(r-t). Thus

|X|=s(rt)(μ2+tμ)=1t2(μ2+2tμ+μ+t2)(μ2+tμ+μt).\left|X\right|=\frac{s(r-t)}{(\mu^{2}+t\mu)}=\frac{1}{t^{2}}(\mu^{2}+2t\mu+\mu+t^{2})(\mu^{2}+t\mu+\mu-t). (3.4)

Case 1: |X|=1\left|X\right|=1.

Then (μ+t+1)(μ3+(2t+1)μ2+(t2t)μt2)=0(\mu+t+1)\left(\mu^{3}+(2t+1)\mu^{2}+(t^{2}-t)\mu-t^{2}\right)=0 from (3.4). When μ3+(2t+1)μ2+(t2t)μt2=0\mu^{3}+(2t+1)\mu^{2}+(t^{2}-t)\mu-t^{2}=0, then μ\mu\notin\mathbb{Z} and thus r=μt(μ2+2tμ+μ+t2)1t(μ3+(2t+1)μ2+(t2t)μt2)=μ+tr=\frac{\mu}{t}\left(\mu^{2}+2t\mu+\mu+t^{2}\right)-\frac{1}{t}(\mu^{3}+(2t+1)\mu^{2}+(t^{2}-t)\mu-t^{2})=\mu+t\notin\mathbb{Z}, a contradiction. When μ=(t+1)\mu=-(t+1), then r=s=t+1r=s=t+1 and GKt+1,t+1G\cong K_{t+1,t+1}.

Case 2: |X|2\left|X\right|\geq 2.

We apply the compatibility condition (2.8) to vertices u,vu,v in XX, we find that

ρuv={tμ,uv,(t1)μ,uv.\rho_{uv}=\left\{\begin{matrix}t\mu,&u\nsim v,\\ (t-1)\mu,&u\sim v.\end{matrix}\right. (3.5)

If t=1t=1 and XX induces a clique then |X|1=rμ2μ|X|-1=r-\mu^{2}-\mu, whence (μ+1)(μ+2)(μ2+μ1)=0(\mu+1)(\mu+2)(\mu^{2}+\mu-1)=0. Thus either μ=2\mu=-2, r=s=t+1=2r=s=t+1=2 and GK2,2G\cong K_{2,2} which belongs to case (1), or μ=12(1±5)\mu=\frac{1}{2}(-1\pm\sqrt{5}) and we have case (2) ([19]).

Otherwise, it follows from (3.5) that μ+\mu\in\mathbb{N}_{+} and GG is μ(μ2+2tμ+μ+t2)/t\mu\left(\mu^{2}+2t\mu+\mu+t^{2}\right)/t-regular of order μ(μ+2t+1)(μ2+2tμ+μ+t2t)/t2\mu(\mu+2t+1)(\mu^{2}+2t\mu+\mu+t^{2}-t)/t^{2} with Kt,μ(μ2+2tμ+μ+t2)/tK_{t,\mu\left(\mu^{2}+2t\mu+\mu+t^{2}\right)/t} as a star complement for μ\mu.\quad\square

In [19], Rowlinson gave a lemma to determine whether a connected rr-regular graph with K1,sK_{1,s} as a star complement is a strongly regular graph. Now we extend it to Kt,sK_{t,s}.

Lemma 3.9.

Let GG be a connected rr-regular graph with μ(r)\mu(\neq r) as an eigenvalue of multiplicity kk. Suppose that |V(G)|=k+t+s\left|V(G)\right|=k+t+s. If k+t+s1>rk+t+s-1>r, then

(k+t+s)rr2kμ2(kμ+r)2s+t10,(k+t+s)r-r^{2}-k\mu^{2}-\frac{(k\mu+r)^{2}}{s+t-1}\geq 0,

with equality if and only if GG is strongly regular.

Proof. Since k+t+s1>rk+t+s-1>r, neither GG nor G¯\overline{G} is complete. Let θ1,,θs+t1\theta_{1},\dots,\theta_{s+t-1} be the eigenvalue of GG other than μ\mu and rr. We have

i=1s+t1θi+kμ+r=0 and i=1s+t1θi2+kμ2+r2=(k+t+s)r.\sum_{i=1}^{s+t-1}\theta_{i}+k\mu+r=0\mbox{ and }\sum_{i=1}^{s+t-1}\theta_{i}^{2}+k\mu^{2}+r^{2}=(k+t+s)r.

It follows that if θ¯=1s+t1i=1s+t1θi\overline{\theta}=\frac{1}{s+t-1}\sum\limits_{i=1}^{s+t-1}\theta_{i}, then

i=1s+t1(θiθ¯)2=i=1s+t1θi2(s+t1)θ¯2=(k+t+s)rr2kμ2(kμ+r)2s+t10.\sum\limits_{i=1}^{s+t-1}(\theta_{i}-\overline{\theta})^{2}=\sum\limits_{i=1}^{s+t-1}\theta_{i}^{2}-(s+t-1)\overline{\theta}^{2}=(k+t+s)r-r^{2}-k\mu^{2}-\frac{(k\mu+r)^{2}}{s+t-1}\geq 0.

Equality holds if and only if θi=θ¯(i[s+t1])\theta_{i}=\overline{\theta}\ (i\in[s+t-1]), equivalently GG has just three distinct eigenvalue. By [9, Theorem 1.2.20], a non-complete connected regular graph is strongly regular if and only if it has exactly three distinct eigenvalues. The proof is completed.\quad\square

Remark 3.10.

From Lemma 3.9, we know that the rr-regular graph GG in (3) of Theorem 3.8 is strongly regular when t=1t=1, and the graph GG isn’t strongly regular when t2t\geq 2.

For the cases 1at11\leq a\leq t-1, we cannot give a characterization. Thus we propose a question for further research.

Question 3.11.

Let st1s\geq t\geq 1, μ{1,t,r}\mu\notin\{-1,-t,r\} and 1at11\leq a\leq t-1. Can we give a characterization of the rr-regular graphs with Kt,sK_{t,s} as a star complement?

4 Regular graphs with K3,sK_{3,s} as a star complement

In this section, we completely solve Question 3.11 when t=3t=3. Since the cases when μ{1,3,r}\mu\in\left\{-1,-3,r\right\} have been solved in Section 3, we consider the cases μ{1,3,r}\mu\notin\left\{-1,-3,r\right\} in the following.

Let GG be an rr-regular graph with H=K3,s(s3)H=K_{3,s}\ (s\geq 3) as a star complement for μ{1,3,r}\mu\notin\left\{-1,-3,r\right\}. Then μ0\mu\neq 0 and μ23s\mu^{2}\neq 3s for μ\mu is not an eigenvalue of K3,sK_{3,s}. By Lemma 3.7, the possible types of vertices in XX are shown in Table 2:

Type (a,b)(a,b) s
I (0,μ2+3μ)(0,\mu^{2}+3\mu) μ(μ2+7μ+9)/3\mu(\mu^{2}+7\mu+9)/3
II (1,μ2+2μ2)(1,\mu^{2}+2\mu-2) (μ+2)(μ2+4μ3)/2(\mu+2)(\mu^{2}+4\mu-3)/2
III (2,μ3+3μ22μ+2)(2,\frac{\mu^{3}+3\mu^{2}-2}{\mu+2}) μ4+7μ3+13μ2+2μ6μ+2\frac{\mu^{4}+7\mu^{3}+13\mu^{2}+2\mu-6}{\mu+2}
Table 2: The possible types of vertices in XX
Lemma 4.1.

Let GG be a graph with H=K3,s(s3)H=K_{3,s}\ (s\geq 3) as a star complement for μ\mu. If μ\mu is a non-main eigenvalue of GG and μ{1,3}\mu\notin\left\{-1,-3\right\}, then all the vertices in the star set XX are of the same type, say, Type I, Type II or Type III.

Proof. Now we show it is impossible that there are two or three types of vertices in XX.

Case 1. There exist vertices of Type I and Type III in XX.

From Table 2, we have

s=13μ(μ2+7μ+9)=μ4+7μ3+13μ2+2μ6μ+2s=\frac{1}{3}\mu(\mu^{2}+7\mu+9)=\frac{\mu^{4}+7\mu^{3}+13\mu^{2}+2\mu-6}{\mu+2}

with solution μ=3\mu=-3, 1-1 or 11. Since μ{1,3}\mu\notin\{-1,-3\}, we have μ=1\mu=1, and thus s=17/3s=17/3, a contraction.

Case 2. There exist vertices of Type II and Type III in XX.

From Table 2, we have

s=12(μ+2)(μ2+4μ3)=μ4+7μ3+13μ2+2μ6μ+2s=\frac{1}{2}(\mu+2)(\mu^{2}+4\mu-3)=\frac{\mu^{4}+7\mu^{3}+13\mu^{2}+2\mu-6}{\mu+2}

with solution μ=3\mu=-3 or 0, a contraction.

Case 3. There exist vertices of Type I and Type II in XX.

By Case 1 and Case 2, we know there does not exist vertices of Type III in XX.

By Table 2, we have

s=13μ(μ2+7μ+9)=12(μ+2)(μ2+4μ3)s=\frac{1}{3}\mu(\mu^{2}+7\mu+9)=\frac{1}{2}(\mu+2)(\mu^{2}+4\mu-3)

with solution μ=3\mu=-3 or 22. Since μ3\mu\neq-3, we have μ=2\mu=2, and thus s=18s=18. The vertices in XX are of type (0,10)(0,10), (1,6)(1,6). From (2.8), Table 3 is obtained.

(a,b)(a,b) (c,d)(c,d) auva_{uv} ρuv\rho_{uv}
(0,10)(0,10) (1,6)(1,6) 0 44
(0,10)(0,10) (1,6)(1,6) 11 22
(0,10)(0,10) (0,10)(0,10) 0 22
(0,10)(0,10) (0,10)(0,10) 11 0
(1,6)(1,6) (1,6)(1,6) 0 11/5-11/5
(1,6)(1,6) (1,6)(1,6) 11 21/5-21/5
Table 3: The possible adjacency of vertices in XX

For any two vertices of type (1,6)(1,6), (1,6)(1,6) in XX, ρuv\rho_{uv}\notin\mathbb{Z}, so there is at most one vertex of type (1,6)(1,6) in XX. For any two vertices of type (0,10)(0,10), (0,10)(0,10) in XX, ρuv=0\rho_{uv}=0 or 22. Since s=18s=18, there are at most two vertices of type (0,10)(0,10) in XX, and ρuv=2\rho_{uv}=2 if there are two vertices of type (0,10)(0,10). Therefore, XX has at most three vertices.

Let V(K3,18)=VWV(K_{3,18})=V\cup W with |V|=3\left|V\right|=3, |W|=18\left|W\right|=18. For any viVv_{i}\in V and wiWw_{i}\in W, we have dG(vi)18d_{G}(v_{i})\geq 18 and dG(wi)3+3=6d_{G}(w_{i})\leq 3+3=6, which contradicts with the regularity of GG.

The proof is completed.\quad\square

Now we define three special graphs G1,G2G_{1},G_{2} and G3G_{3} (see Figure 1) as follows. Clearly, the graph G1G_{1} is a 4-regular graph of order 9, its spectrum is [3,22,02,13,4-3,-2^{2},0^{2},1^{3},4]; G2G_{2} is a 5-regular graph of order 12, its spectrum is [33,12,16,5-3^{3},-1^{2},1^{6},5]; G3G_{3} is a 6-regular graph of order 15, its spectrum is [35,19,6-3^{5},1^{9},6]. By Lemma 3.9, we know that G1G_{1} and G2G_{2} isn’t strongly regular while G3G_{3} is strongly regular with parameters (15,6,1,3)(15,6,1,3).

Refer to caption
(a) G1G_{1}
Refer to caption
(b) G2G_{2}
Refer to caption
(c) G3G_{3}
Figure 1: Regular graphs G1G_{1}, G2G_{2}, G3G_{3} of Theorem 4.2
Theorem 4.2.

Let s3s\geq 3, GG be an rr-regular graph with K3,sK_{3,s} as a star complement for an eigenvalue μ\mu. Then one of the following holds:
(1) μ=1\mu=-1, r1(mod3s1(s1,2))r\equiv-1\ ({\rm mod}\ \frac{3s-1}{(s-1,2)}) and GG(r)G\cong G(r), where G(r)G(r) is defined in Theorem 3.5;
(2) μ=±4\mu=\pm 4, r=s=4r=s=4 and GK4,4G\cong K_{4,4};
(3) μ+\mu\in\mathbb{N}_{+}, r=sr=s and GG is an rr-regular graph of order μ(μ+7)(μ+6)(μ+1)/9\mu(\mu+7)(\mu+6)(\mu+1)/9, where r=μ(μ2+7μ+9)/3r=\mu(\mu^{2}+7\mu+9)/3;
(4) μ=1\mu=1, s=3s=3, and GG1(r=4)G\cong G_{1}\ (r=4), GG2(r=5)G\cong G_{2}\ (r=5) or GG3(r=6)G\cong G_{3}\ (r=6) (see Figure 1).

Proof. By Theorem 3.5 and μ=1\mu=-1, we have (1) holds.

By Theorem 3.4 and μ=r\mu=r, we have s=r=4s=r=4 and GK4,4G\cong K_{4,4}.

If μ{1,r}\mu\notin\{-1,r\}, then μ\mu is non-main. From Proposition 3.3, we know μ3\mu\neq-3. Since μ\mu is not an eigenvalue of HK3,sH\cong K_{3,s}, we have μ0\mu\neq 0 and μ23s\mu^{2}\neq 3s. By Lemma 2.4, V(K3,s)V(K_{3,s}) is a location-dominating set, and so GG is connected. Let V(K3,s)=VWV(K_{3,s})=V\cup W with |V|=3\left|V\right|=3, |W|=s\left|W\right|=s. Denote the vertices in VV and WW by v1,v2,v3v_{1},v_{2},v_{3} and w1,w2,,wsw_{1},w_{2},\dots,w_{s}, respectively. In the following, we suppose that μ{1,r,3,0}\mu\neq\left\{-1,r,-3,0\right\} and μ23s\mu^{2}\neq 3s. From Table 2, there are only three possible types of vertex in XX. By Lemma 4.1, we know all the vertices in XX are of the same type, and we consider the following three cases:

Case 1. The vertices in XX are of Type I.

Then (1) or (3) of Theorem 3.8 holds by s3s\geq 3, say, either μ=4\mu=-4, r=s=4r=s=4 and GK4,4G\cong K_{4,4}, or μ+\mu\in\mathbb{N}_{+} and GG is μ(μ2+7μ+9)/3\mu(\mu^{2}+7\mu+9)/3-regular of order μ(μ+7)(μ+6)(μ+1)/9\mu(\mu+7)(\mu+6)(\mu+1)/9 with K3,(μ3+7μ2+9μ)/3K_{3,(\mu^{3}+7\mu^{2}+9\mu)/3} as a star complement for μ\mu.

Combining the above case of μ=r\mu=r, result (2) or (3) holds.

Case 2. The vertices in XX are of Type II.

Then H=K3,(μ+2)(μ2+4μ3)/2H=K_{3,(\mu+2)(\mu^{2}+4\mu-3)/2} and all vertices in XX are of type (1,μ2+2μ2)(1,\mu^{2}+2\mu-2). Applying the compatibility condition (2.8) to vertices u,vu,v of XX, since μ3\mu\neq-3, we find that

ρuv={μ1,uv,2μ1,uv.\rho_{uv}=\left\{\begin{matrix}\mu-1,&u\sim v,\\ 2\mu-1,&u\nsim v.\end{matrix}\right. (4.1)

By regularity of GG, we have dG(v1)=dG(v2)=dG(v3)d_{G}(v_{1})=d_{G}(v_{2})=d_{G}(v_{3}). This implies the vertices in XX are equally divided into three parts, and each vertex in VV is adjacent to every vertex of one part, so

r=s+|X|/3.r=s+\left|X\right|/3. (4.2)

Now we compute the edges between XX and V(H)V(H) by two ways, and we have

|X|(μ2+2μ1)=3(rs)+s(r3).\left|X\right|(\mu^{2}+2\mu-1)=3(r-s)+s(r-3). (4.3)

By (4.2), (4.3) and s=(μ+2)(μ2+4μ3)/2s=(\mu+2)(\mu^{2}+4\mu-3)/2, we have (μ+3)(μ1)(μ2)|X|=32(μ+2)(μ2+4μ3)(μ+3)(μ1)(μ+4).(\mu+3)(\mu-1)(\mu-2)\left|X\right|=-\frac{3}{2}(\mu+2)(\mu^{2}+4\mu-3)(\mu+3)(\mu-1)(\mu+4). Since μ3\mu\neq-3, we consider the following three subcases.

Subcase 2.1. μ1,2\mu\neq 1,2.

Then

|X|=32(μ+2)(μ2+4μ3)μ+4μ2=3s(1+6μ2).\left|X\right|=-\frac{3}{2}(\mu+2)(\mu^{2}+4\mu-3)\frac{\mu+4}{\mu-2}=-3s(1+\frac{6}{\mu-2}).

Since |X|\left|X\right|\in\mathbb{Z} and ss\in\mathbb{Z}, we have μ\mu\in\mathbb{Q}. Notice that μ\mu is an algebraic integer, then μ\mu\in\mathbb{Z}. From (4.1), we know that μ+{1,2}\mu\in\mathbb{N}_{+}\setminus\left\{1,2\right\}. Thus |X|<0\left|X\right|<0, a contradiction.

Subcase 2.2. μ=2\mu=2.

Then s=(μ+2)(μ2+4μ3)2=18s=\frac{(\mu+2)(\mu^{2}+4\mu-3)}{2}=18, H=K3,18H=K_{3,18} and all vertices in XX are of type (1,6)(1,6). By (4.2) and (4.3), we obtain 0=2700=270, it is a contradiction.

Subcase 2.3. μ=1\mu=1.

Then s=3s=3, H=K3,3H=K_{3,3} with V={v1,v2,v3}V=\left\{v_{1},v_{2},v_{3}\right\}, W={w1,w2,w3}W=\left\{w_{1},w_{2},w_{3}\right\} and all vertices in XX are of type (1,1)(1,1). Thus |X|(31)(31)=9\left|X\right|\leq\binom{3}{1}\binom{3}{1}=9. From (4.1), we have

ρuv={0,uv,1,uv.\rho_{uv}=\left\{\begin{matrix}0,&u\sim v,\\ 1,&u\nsim v.\end{matrix}\right. (4.4)

Since GG and HH are regular, XX induces a rr^{\prime}-regular graph, denoted by G[X]G[X].

Claim 1. The graph G[X]G[X] cannot contain S1S_{1}, S2S_{2} or S3S_{3} as an induced subgraph (see Figure 3).

Proof. If G[X]G[X] contains S1S_{1} as an induced subgraph, without loss of generality, we suppose that u1v1,u1w1u_{1}\sim v_{1},u_{1}\sim w_{1}. Then by (4.4), ρu1u2=ρu1u3=ρu1u4=0\rho_{u_{1}u_{2}}=\rho_{u_{1}u_{3}}=\rho_{u_{1}u_{4}}=0, and thus vertices u2u_{2}, u3u_{3} and u4u_{4} are adjacent to one vertex in V{v1}V\setminus\left\{v_{1}\right\} and one vertex in W{w1}W\setminus\left\{w_{1}\right\} such that ρu2u3=ρu2u4=ρu3u4=1\rho_{u_{2}u_{3}}=\rho_{u_{2}u_{4}}=\rho_{u_{3}u_{4}}=1, it is impossible.

If G[X]G[X] contains S2S_{2} as an induced subgraph, since the vertices u5u_{5}, u6u_{6} and u7u_{7} are adjacent in pairs, by (4.4), without loss of generality, we can suppose that u5v1,u5w1,u6v2,u6w2,u7v3,u7w3u_{5}\sim v_{1},u_{5}\sim w_{1},u_{6}\sim v_{2},u_{6}\sim w_{2},u_{7}\sim v_{3},u_{7}\sim w_{3}. Since u8u6,u8u7u_{8}\sim u_{6},u_{8}\sim u_{7}, by (4.4), vertex u8u_{8} is adjacent to vertices in V(K3,3){v2,w2,v3,w3}V(K_{3,3})\setminus\left\{v_{2},w_{2},v_{3},w_{3}\right\}. Thus the HH-neighbourhood of u5u_{5} and u8u_{8} is the same, a contradiction.

Similar to the proof of S2S_{2}, it’s obvious that G[X]G[X] cannot contain S3S_{3} as an induced subgraph.\qquad\square

By G[X]G[X] is rr^{\prime}-regular and (4.2), for any uXu\in X, dG(u)=2+r=r=3+|X|/3d_{G}(u)=2+r^{\prime}=r=3+\left|X\right|/3. Then 2r=1+|X|/342\leq r^{\prime}=1+\left|X\right|/3\leq 4 by |X|9|X|\leq 9.

If r=2r^{\prime}=2, then |X|=3\left|X\right|=3, G[X]C3G[X]\cong C_{3}, GG1G\cong G_{1} by (4.4);

If r=3r^{\prime}=3, then |X|=6\left|X\right|=6 and G[X]G[X] is connected since the minimum degree of G[X]G[X] is equal to |X|2\frac{|X|}{2}. By [13], there are two non-isomorphic connected 33-regular graphs with 66 vertices (see Figure 3). Since A2A_{2} contains S1S_{1} as an induced subgraph, by Claim 1, G[X]A2G[X]\ncong A_{2}. Then G[X]A1G[X]\cong A_{1}, and the graph G2G_{2} in Figure 1 is the only non-isomorphic graph satisfying (4.4).

Refer to caption
(a) S1S_{1}
Refer to caption
(b) S2S_{2}
Refer to caption
(c) S3S_{3}
Figure 2: induced subgraph
Refer to caption
(a) A1A_{1}
Refer to caption
(b) A2A_{2}
Figure 3: 33-regular graphs on 66 vertices.

If r=4r^{\prime}=4, then |X|=9\left|X\right|=9 and G[X]G[X] is connected since the minimum degree of G[X]G[X] is equal to |X|2\left\lfloor\frac{|X|}{2}\right\rfloor. By [13], there are 1616 non-isomorphic connected 44-regular graphs with 99 vertices (see Figure 4). Since Bi(i{2,,12})B_{i}(i\in\{2,\dots,12\}) contain S1S_{1} as an induced subgraph, graph B13B_{13} contain S3S_{3} as an induced subgraph, graph Bj(j{14,15,16})B_{j}(j\in\{14,15,16\}) contain S2S_{2} as an induced subgraph, by Claim 1, we have G[X]Bi(i{2,,16})G[X]\ncong B_{i}(i\in\{2,\dots,16\}). So G[X]B1G[X]\cong B_{1}, and the graph G3G_{3} in Figure 1 is the only non-isomorphic graph satisfying (4.4).

Combining the proof of Subcase 2.3, (4) holds.

Refer to caption
(a) B1B_{1}
Refer to caption
(b) B2B_{2}
Refer to caption
(c) B3B_{3}
Refer to caption
(d) B4B_{4}
Refer to caption
(e) B5B_{5}
Refer to caption
(f) B6B_{6}
Refer to caption
(g) B7B_{7}
Refer to caption
(h) B8B_{8}
Refer to caption
(i) B9B_{9}
Refer to caption
(j) B10B_{10}
Refer to caption
(k) B11B_{11}
Refer to caption
(l) B12B_{12}
Refer to caption
(m) B13B_{13}
Refer to caption
(n) B14B_{14}
Refer to caption
(o) B15B_{15}
Refer to caption
(p) B16B_{16}
Figure 4: 44-regular graphs on 99 vertices

Case 3. The vertices in XX are of Type III.

Then H=K3,(μ4+7μ3+13μ2+2μ6)/(μ+2)H=K_{3,(\mu^{4}+7\mu^{3}+13\mu^{2}+2\mu-6)/(\mu+2)} and all vertices in XX are of type (2,(μ3+3μ22)/(μ+2))(2,(\mu^{3}+3\mu^{2}-2)/(\mu+2)). Since μ3\mu\neq-3, by (2.8), we have

ρuv={2μ+2,uv,μ2+2μ+2μ+2,uv.\rho_{uv}=\left\{\begin{matrix}\frac{2}{\mu+2},&u\sim v,\\ \frac{\mu^{2}+2\mu+2}{\mu+2},&u\nsim v.\end{matrix}\right.

Then (μ+2)2(\mu+2)\mid 2 by ρuv\rho_{uv}\in\mathbb{Z}. Noting that ρuv0\rho_{uv}\geq 0 and μ{1,3,0}\mu\notin\left\{-1,-3,0\right\}, it is impossible. Therefore, there is not an rr-regular graph with K3,sK_{3,s} as a star complement in this case.

Combining the above argument, we complete the proof.\quad\square

5 Regular graphs with Ks,sK_{s,s} as a star complement

By (4) of Theorem 4.2, we know that there are three regular graphs with K3,3K_{3,3} as a star complement for μ=1\mu=1. In the following, we will study some properties of regular graphs with Ks,sK_{s,s} as a star complement for an eigenvalue μ\mu, and then give a sharp upper bound for the multiplicity kk of μ\mu.

Since μ\mu is not an eigenvalue of Ks,sK_{s,s}, we have μ0\mu\neq 0 and μ±s\mu\neq\pm s. When μ=1\mu=-1, we have r1(mod(s+1))r\equiv-1({\rm mod}\ (s+1)) and GG(r)G\cong G(r) by Theorem 3.5. So we discuss the case when μ{1,0}\mu\notin\{-1,0\} in the following.

Proposition 5.1.

Let s2s\geq 2 and GG be an rr-regular graph with Ks,sK_{s,s} as a star complement for an eigenvalue μ\mu, where μ{1,0}\mu\notin\{-1,0\}. Then

(1) μ\mu\in\mathbb{Z} and |μ|<s|\mu|<s.

(2) If μ=1\mu=1, then s=3s=3 and GG1G\cong G_{1}, G2G_{2} or G3G_{3}.

(3) If all vertices in the star set XX are of the same type, then GG is an r-regular graph of order (2μ+1)(rμ1)(2\mu+1)(\frac{r}{\mu}-1).

Proof. Clearly, there is no regular graph GG with Ks,sK_{s,s} as a star complement for μ=r\mu=r by Theorem 3.4. Thus μ\mu is a non-main eigenvalue. Let t=st=s in (2.6), we have 0<a+b=sμ2s0<a+b=s-\mu\leq 2s, thus μ\mu\in\mathbb{Z} and sμ<s-s\leq\mu<s. Since μ±s\mu\neq\pm s, we have |μ|<s|\mu|<s, (1) holds.

Since t=st=s, by (2.6) and (2.7), we have a=x1a=x_{1}, b=x2b=x_{2} or a=x2a=x_{2}, b=x1b=x_{1} where

x1=sμ+(s+μ)(2μ2+μs)2,x2=sμ(s+μ)(2μ2+μs)2.x_{1}=\frac{s-\mu+\sqrt{-(s+\mu)(2\mu^{2}+\mu-s)}}{2},\ x_{2}=\frac{s-\mu-\sqrt{-(s+\mu)(2\mu^{2}+\mu-s)}}{2}. (5.1)

Since a,ba,b\in\mathbb{Z}, (s+μ)(2μ2+μs)=(sμ2)2(μ2+μ)2-(s+\mu)(2\mu^{2}+\mu-s)=(s-\mu^{2})^{2}-(\mu^{2}+\mu)^{2} must be a perfect square. Thus, when μ=1\mu=1, (s1)24(s-1)^{2}-4 must be a perfect square, so s=3s=3, and thus the graphs G1G_{1}, G2G_{2} and G3G_{3} (see Figure 1) are the regular graphs with K3,3K_{3,3} as a star complement for μ=1\mu=1 by (4) of Theorem 4.2, (2) holds.

Let uXu\in X be a vertex of type (a,b)(a,b). Since GG is rr-regular with Ks,sK_{s,s} as a star complement and all vertices in XX are of the same type, we have a=ba=b. By (2.6) and (2.7), we have a=b=s=μa=b=s=-\mu or a=b=μ2,s=2μ2+μa=b=\mu^{2},s=2\mu^{2}+\mu.

If a=b=s=μa=b=s=-\mu, then |X|=1|X|=1, r=2s=1+sr=2s=1+s. Thus s=1s=1 and μ=1\mu=-1, a contradiction.

If a=b=μ2,s=2μ2+μa=b=\mu^{2},s=2\mu^{2}+\mu, counting the edges between XX and V(H)V(H) in two ways, we have (a+b)|X|=2s(rs)(a+b)\cdot|X|=2s(r-s). Thus |V(G)|=|X|+2s=(2μ+1)(rμ1)|V(G)|=|X|+2s=(2\mu+1)(\frac{r}{\mu}-1), (3) holds.\quad\square

Remark 5.2.

In fact, there are a lot of regular graphs with H=Ks,sH=K_{s,s} as a star complement. For example, when μ=2\mu=-2, it follows from (5.1) that (s4)24(s-4)^{2}-4 must be a perfect square, thus s=2s=2 or 66.

When s=2s=2, we have a=b=2a=b=2 by (5.1). Thus |X|=1|X|=1 by Lemma 2.4. In this situation, GG is not regular.

But when s=6s=6, we have a=b=4a=b=4 by (5.1). From (2.8), we have ρuv={4,uv,6,uv.\rho_{uv}=\left\{\begin{matrix}4,&u\nsim v,\\ 6,&u\sim v.\end{matrix}\right. Since |X|8=12(r6)|X|\cdot 8=12(r-6), we have |X|=3(r6)2|X|=\frac{3(r-6)}{2}\in\mathbb{Z}, and then rr is even.

Let (V,W)(V,W) be the bipartition of H=K6,6H=K_{6,6} with V={v1,v2,,v6}V=\left\{v_{1},v_{2},\cdots,v_{6}\right\}, W={w1,w2,,w6}W=\left\{w_{1},w_{2},\cdots,w_{6}\right\}.

If r=8r=8, then |X|=3|X|=3. The graph G4G_{4} defined as follows is a regular graph with K6,6K_{6,6} as a star complement for μ=2\mu=-2: X={u1,u2,u3}X=\{u_{1},u_{2},u_{3}\}, V(G4)=V(H)XV(G_{4})=V(H)\cup X, G4[X]=3K1G_{4}[X]=3K_{1}, and NH(u1)={v1,v2,v3,v4,w1,w2,w3,w4}N_{H}(u_{1})=\left\{v_{1},v_{2},v_{3},v_{4},w_{1},w_{2},w_{3},w_{4}\right\}, NH(u2)={v3,v4,v5,v6,w3,w4,w5,w6}N_{H}(u_{2})=\left\{v_{3},v_{4},v_{5},v_{6},w_{3},w_{4},w_{5},w_{6}\right\}, NH(u3)={v1,v2,v5,v6,w1,w2,w5,w6}N_{H}(u_{3})=\left\{v_{1},v_{2},v_{5},v_{6},w_{1},w_{2},w_{5},w_{6}\right\}. The spectrum of G4G_{4} is [6,23,08,22,8][-6,-2^{3},0^{8},2^{2},8].

If r=10r=10, then |X|=6|X|=6. The graph G5G_{5} defined as follows is a regular graph with K6,6K_{6,6} as a star complement for μ=2\mu=-2: X={u1,u2,,u6}X=\{u_{1},u_{2},\dots,u_{6}\}, V(G5)=V(H)XV(G_{5})=V(H)\cup X, G5[X]=C6G_{5}[X]=C_{6} with the edge set {u1u2,u2u3,u3u4,u4u5,u5u6,u6u1}\{u_{1}u_{2},u_{2}u_{3},u_{3}u_{4},u_{4}u_{5},u_{5}u_{6},u_{6}u_{1}\}, and NH(u1)={v1,v2,v3,v4,w1,w2,w3,w4}N_{H}(u_{1})=\left\{v_{1},v_{2},v_{3},v_{4},w_{1},w_{2},w_{3},w_{4}\right\}, NH(u2)={v2,v3,v4,v5,w2,w3,w4,w5}N_{H}(u_{2})=\left\{v_{2},v_{3},v_{4},v_{5},w_{2},w_{3},w_{4},w_{5}\right\}, NH(u3)={v3,v4,v5,v6,w3,w4,w5,w6}N_{H}(u_{3})=\left\{v_{3},v_{4},v_{5},v_{6},w_{3},w_{4},w_{5},w_{6}\right\}, NH(u4)={v1,v4,v5,v6,w1,w4,w5,w6}N_{H}(u_{4})=\{v_{1},v_{4},v_{5},v_{6},w_{1},\\ w_{4},w_{5},w_{6}\}, NH(u5)={v1,v2,v5,v6,w1,w2,w5,w6}N_{H}(u_{5})=\left\{v_{1},v_{2},v_{5},v_{6},w_{1},w_{2},w_{5},w_{6}\right\}, NH(u6)={v1,v2,v3,v6,w1,w2,w3,w6}N_{H}(u_{6})=\left\{v_{1},v_{2},v_{3},v_{6},w_{1},w_{2},w_{3},w_{6}\right\}. The spectrum of G5G_{5} is [6,26,06,12,32,10][-6,-2^{6},0^{6},1^{2},3^{2},10].

Since |X|12(q+1)(q2)=65|X|\leq\frac{1}{2}(q+1)(q-2)=65, where q=|V(H)|q=|V(H)| ([4]), there are various choices of rr. It can be predicted that there are a lot of graphs that satisfy the conditions. Then the commonalities of the regular graphs with Ks,sK_{s,s} as a star complement seems to be an interesting question worth studying.

It is shown in [17] that if GG is a connected rr-regular graph of order nn with μ{1,0}\mu\notin\left\{-1,0\right\} as an eigenvalue of multiplicity kk and r>2r>2, q=nkq=n-k, then k12(r1)qk\leq\frac{1}{2}(r-1)q. In the following, we will show that when GG has Ks,s(q=2s2)K_{s,s}\ (q=2s\geq 2) as a star complement for μ\mu, then ks(rs)=12q(rq2)12(r1)qk\leq s(r-s)=\frac{1}{2}q(r-\frac{q}{2})\leq\frac{1}{2}(r-1)q.

For subsets U,VU^{\prime},V^{\prime} of V(G)V(G), we write E(V,U)E(V^{\prime},U^{\prime}) for the set of edges between UU^{\prime} and VV^{\prime}. The authors of [5] have determined all the graphs with a star set XX for which E(X,X¯)E(X,\overline{X}) is a perfect matching. The result is as follows.

Theorem 5.3.

([5]) Let GG be a graph with XX as a star set for an eigenvalue μ\mu. If E(X,X¯)E(X,\overline{X}) is a perfect matching, then one of the following holds:
(1) G=K2G=K_{2} and μ=±1\mu=\pm 1; (2) G=C4G=C_{4} and μ=0\mu=0; (3) G is the Petersen graph and μ=1\mu=1.

Theorem 5.4.

Let GG be an rr-regular graph of order nn with Ks,sK_{s,s} as a star complement for the eigenvalue μ{1,0}\mu\notin\{-1,0\} of multiplicity kk. Then ks(rs)k\leq s(r-s), equivalently ns(rs+2)n\leq s(r-s+2), with equality if and only if μ=1\mu=1, GG1,G2G\cong G_{1},G_{2} or G3G_{3} (see Figure 1).

Proof. Let H=Ks,sH=K_{s,s} and V(H)=VWV(H)=V\cup W with |V|=|W|=s|V|=|W|=s. By Lemma 2.4, V(Ks,s)V(K_{s,s}) is a location-dominating set, and so GG is connected. By Lemma 2.4, we have |NH(u)|1|N_{H}(u)|\geq 1. Thus k=|X|uX|NH(u)|=|E(X,X¯)|=2s(rs)k=|X|\leq\sum_{u\in X}|N_{H}(u)|=|E(X,\overline{X})|=2s(r-s).

When the equality holds, we have |NH(u)|=1|N_{H}(u)|=1 for all uXu\in X. Since the neighbourhoods NH(u)(uX)N_{H}(u)\ (u\in X) are distinct, any vertex in HH has at most one adjacent vertex in XX, thus 2s=|X¯||X|=2s(rs)2s=|\overline{X}|\geq|X|=2s(r-s) which means rs+1r\leq s+1. On the other hand, we have rs+1r\geq s+1 by GG is rr-regular and connected. Thus r=s+1r=s+1 and then |X|=2s|X|=2s. Therefore E(X,X¯)E(X,\overline{X}) is a perfect matching, but there is no such graph GG by Theorem 5.3.

Let t=st=s in (2.6), we find that |NH(u)|=a+b=sμ|N_{H}(u)|=a+b=s-\mu is a constant, which means |NH(u1)|=|NH(u2)||N_{H}(u_{1})|=|N_{H}(u_{2})| for any u1,u2Xu_{1},u_{2}\in X. Therefore, we have |NH(u)|2|N_{H}(u)|\geq 2 for any uXu\in X and 2kuX|NH(u)|=|E(X,X¯)|=2s(rs)2k\leq\sum\limits_{u\in X}|N_{H}(u)|=|E(X,\overline{X})|=2s(r-s), equivalently ns(rs+2)n\leq s(r-s+2), with equality if and only if |NH(u)|=2|N_{H}(u)|=2 for any uXu\in X.

If n=s(rs+2)n=s(r-s+2), we have μ=s2\mu=s-2 and the possible types for the vertices in XX are (1,1),(0,2),(2,0)(1,1),(0,2),(2,0).

If there are two vertices of type (0,2)(0,2) (or (2,0)(2,0)) in XX, then it follows from (2.8) that ρuv={ss1,uv,s24s+21s,uv.\rho_{uv}=\left\{\begin{matrix}\frac{s}{s-1},&u\nsim v,\\ \frac{s^{2}-4s+2}{1-s},&u\sim v.\end{matrix}\right. By (2) of Lemma 2.4, we have ρuv2\rho_{uv}\neq 2, then ρuv=0\rho_{uv}=0 or 11, it is a contradiction with s+s\in\mathbb{Z}_{+}. Therefore, there is at most one vertex of type (0,2)(0,2) (or (2,0)(2,0)).

If there is one vertex of type (2,0)(2,0) and one vertex of type (0,2)(0,2) in XX, then it follows from (2.8) that ρuv={s2s1,uv,s24s+41s,uv.\rho_{uv}=\left\{\begin{matrix}\frac{s-2}{s-1},&u\nsim v,\\ \frac{s^{2}-4s+4}{1-s},&u\sim v.\end{matrix}\right. Clearly, ρuv=0\rho_{uv}=0, s=2s=2, and μ=0\mu=0, it is a contradiction. Therefore, the vertex of type (2,0)(2,0) and the vertex of type (0,2)(0,2) cannot exist at the same time in XX.

Suppose that XX contains a vertex of type (0,2)(0,2) (or (2,0)(2,0)), then |E(X,V)||E(X,W)||E(X,V)|\neq|E(X,W)|, a contradiction. Thus all vertices in XX are of type (1,1)(1,1). From (2.8), we have ρuv={1,uv,3s,uv.\rho_{uv}=\left\{\begin{matrix}1,&u\nsim v,\\ 3-s,&u\sim v.\end{matrix}\right.

If for any u,vXu,v\in X, uvu\nsim v, then r=2r=2 and HK1,1H\cong K_{1,1}, GC3G\cong C_{3} and thus s=1s=1 and μ=s2=1\mu=s-2=-1, a contradiction.

If there exists u,vXu,v\in X such that uvu\sim v, then 3s=03-s=0 or 11 by (2) of Lemma 2.4. If 3s=13-s=1, then s=2s=2 and μ=0\mu=0, a contradiction. If 3s=03-s=0, then s=3s=3, μ=1\mu=1 and thus G=G1,G2G=G_{1},G_{2} or G3G_{3} (see Figure 1) by (4) of Theorem 4.2.

The proof is completed. \quad\square

Competing interests

The authors declare that they have no competing interests.

References

  • [1] L. Asgharsharghi, D. Kiani, On regular graphs with complete tripartite star complements, Ars Combin. 122 (2015) 431–437.
  • [2] F.K. Bell, Characterizing line graphs by star complements, Linear Algebra Appl. 296 (1999) 15-25.
  • [3] F.K. Bell, Line graphs of bipartite graphs with hamiltonian paths, J. Graph Theory. 43 (2003) 137-149.
  • [4] F.K. Bell, P. Rowlinson, On the multiplicities of graph eigenvalues, Bull. Lond. Math. Soc. 35 (2003) 401–408.
  • [5] N.E. Clarke, W.D. Garraway, C.A. Hickman, R.J. Nowakowski, Graphs where star sets are matched to their complements, J. Combin. Math. Combin. Comput. 37 (2001) 177–185.
  • [6] D. Cvetković, M. Doob, H. Sachs, Spectra of Graphs: Theory and Application, 3rd edition, Jonah Ambrosius Barth Verlag, Heidelberg–Leipzig, 1995.
  • [7] D. Cvetković, P. Rowlinson, S. Simić, Eigenspaces of Graphs, Cambridge University Press, Cambridge, 1997.
  • [8] D. Cvetković, P. Rowlinson, S. Simić, Some characterization of graphs by star complements, Linear Algebra Appl. 301 (1999) 81–97.
  • [9] D. Cvetković, P. Rowlinson, S. Simić, Spectral Generalizations of Line Graphs, Cambridge University Press, Cambridge, 2004.
  • [10] D. Cvetković, P. Rowlinson, S. Simić, An Introduction to the Theory of Graph Spectra, Cambridge University Press, Cambridge, 2010.
  • [11] M. Ellingham, Basic subgraphs and graph spectra, Australas. J. Combin. 8 (1993) 247–265.
  • [12] X. Fang, L. You, Y. Huang, Maximal graphs with a prescribed complete bipartite graph as a star complement, AIMS Math. 6 (2021) 7153–7169.
  • [13] M. Meringer, Fast generation of regular graphs and construction of cages, J Graph Theory. 30 (1999) 137-146.
  • [14] F. Ramezani, B. Tayfeh-Rezaie, Graphs with prescribed star complement for the eigenvalue 1, Ars Combin. 116 (2014) 129–145.
  • [15] P. Rowlinson, On bipartite graphs with complete bipartite star complements, Linear Algebra Appl. 458 (2014) 149–160.
  • [16] P. Rowlinson, An extension of the star complement technique for regular graphs, Linear Algebra Appl. 557 (2018) 496-507.
  • [17] P. Rowlinson, Eigenvalue multiplicity in regular graphs, Discrete Appl. Math. 269 (2019) 11–17.
  • [18] P. Rowlinson, P.S. Jackson, On graphs with complete bipartite star complements, Linear Algebra Appl. 298 (1999) 9–20.
  • [19] P. Rowlinson, B. Tayfeh-Rezaie, Star complements in regular graphs: old and new results, Linear Algebra Appl. 432 (2010) 2230–2242.
  • [20] Z. Stanić, On graphs whose second largest eigenvalue equals 1 – the star complement technique, Linear Algebra Appl. 420 (2) (2007) 700–710.
  • [21] Z. Stanić, S.K. Simić, On graphs with unicyclic star complement for 1 as the second largest eigenvalue, Faculty of Mathematics, University of Belgrade, 351 (2005) 475-484.
  • [22] J. Wang, X. Yuan, L. Liu, Regular graphs with a prescribed complete multipartite graph as a star complement, Linear Algebra Appl. 579 (2019) 302–319.
  • [23] Y. Yang, Q. Huang, J. Wang, On a conjecture for regular graphs with complete multipartite star complement, arXiv:1912.07594v2, 2021.
  • [24] X. Yuan, H. Chen, L. Liu, On the characterization of graphs by star complements, Linear Algebra Appl. 533 (2017) 491-506.
  • [25] X. Yuan, Q. Zhao, L. Liu, H. Chen. On graphs with prescribed star complements, Linear Algebra Appl. 559 (2018) 80-94.