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

A complete characterization of graphs for which mG(1)=nd1m_{G}(-1)=n-d-1

Songnian Xu School of Mathematics, China University of Mining and Tecnology, Xuzhou, China. Wenhao Zhen School of Mathematics, China University of Mining and Tecnology, Xuzhou, China. Dein Wong Corresponding author. E-mail address: wongdein@163.com. School of Mathematics, China University of Mining and Tecnology, Xuzhou, China. Songnian Xu
Department of Mathematics, China University of Mining and Technology
Xuzhou, 221116, P.R. China
E-mail: xsn1318191@cumt.edu.cn

Wang
Department of Mathematics, China University of Mining and Technology
Xuzhou 221116, P.R. China
E-mail:
Corresponding author

A complete characterization of graphs for which mG(1)=nd1m_{G}(-1)=n-d-1

Songnian Xu School of Mathematics, China University of Mining and Tecnology, Xuzhou, China. Wenhao Zhen School of Mathematics, China University of Mining and Tecnology, Xuzhou, China. Dein Wong Corresponding author. E-mail address: wongdein@163.com. School of Mathematics, China University of Mining and Tecnology, Xuzhou, China. Songnian Xu
Department of Mathematics, China University of Mining and Technology
Xuzhou, 221116, P.R. China
E-mail: xsn1318191@cumt.edu.cn

Wang
Department of Mathematics, China University of Mining and Technology
Xuzhou 221116, P.R. China
E-mail:
Corresponding author
Abstract

Let GG be a simple connected graph of order nn with diameter dd. Let mG(1)m_{G}(-1) denote the multiplicity of the eigenvalue 1-1 of the adjacency matrix of GG, and let P=Pd+1P=P_{d+1} be the diameter path of GG. If 1-1 is not an eigenvalue of PP, then by the interlacing theorem, we have mG(1)nd1m_{G}(-1)\leq n-d-1. In this article, we characterize the extremal graphs where equality holds. Moreover, for the completeness of the results, we also characterize the graphs GG that achieve mG(1)=nd1m_{G}(-1)=n-d-1 when 1-1 is an eigenvalue of PP. Thus, we provide a complete characterization of the graphs GG for which mG(1)=nd1m_{G}(-1)=n-d-1.

footnotetext: Keywords: eigenvalue multiplicity; diameter; extremal graphs

1 Introduction

In this paper, we consider only simple, connected and finite graphs. A simple undireted graph GG is denoted by G=(V(G),E(G))G=(V(G),E(G)), where V(G)V(G) is the vertex set and E(G)E(G) is the edge set. A graph HH is called a subgraph of a graph GG if V(H)V(G)V(H)\subseteq V(G) and E(H)E(G)E(H)\subseteq E(G). Further, if any two vertices of V(H)V(H) are adjacent in HH if and only if they are adjacent in GG, we say that HH is an induced subgraph of GG and denote this relation as HGH\leq G. The order of GG is the number of vertices of GG, denoted by |G||G|. For V1V(G)V_{1}\subseteq V(G) and HGH\leq G, NH(V1)=vV1{uV(H)|uvE(G)}N_{H}(V_{1})=\cup_{v\in V_{1}}\{u\in V(H)|uv\in E(G)\}. Let KV(G)K\subseteq V(G), traditionally, the subgraph of GG induced by KK, written as G[K]G[K]. And we sometimes write GKG-K or GG[K]G-G[K] to denote G[V(G)\V(K)]G[V(G)\backslash V(K)]. A vertex vv of GG is said to be pendant if d(v)=1d(v)=1, where d(v)d(v) denotes the number of adjacent vertices of vv in V(G)V(G). For x,yV(G)x,y\in V(G), distance d(x,y)d(x,y) represents the length of the shortest path between xx and yy, and d(x,H)=min{d(x,y)|yV(H)}d(x,H)=min\{d(x,y)|y\in V(H)\} for HGH\leq G. We denote by Km,nK_{m,n} the complete bipartite graph. Let A(G)A(G) denote the adjacency matrix of graph GG, which is a square matrix and aij=1a_{ij}=1 if and only if vivjv_{i}\sim v_{j}, otherwise aij=0a_{ij}=0.

Let mG(μ)m_{G}(\mu) denote the multiplicity of the eigenvalue μ\mu in the adjacency matrix A(G)A(G). In 2013, Wong [10] characterize graphs with maximum diameter among all connected graph with rank nn. In 2022, Wang [11] characterized the graphs for which mG(0)=ndm_{G}(0)=n-d using the method of star complements. In the same year, Chang and Li [2] characterized the graphs for which mG(0)=ng1m_{G}(0)=n-g-1. In 2023, Du and Fonseca [4] extend the definition of minimal graphs on adjacency matrices to real symmetric matrices. They characterize all the trees for which there is a real symmetric matrix with nullity ndn-d and nd1n-d-1. In 2024, Zhang and Wong [6] provided a complete characterization of the graphs for which mG(1)=ndm_{G}(-1)=n-d. Therefor, we investigate the situation mG(1)=nd1m_{G}(-1)=n-d-1.

Let GG be a graph and u,vV(G)u,v\in V(G). We introduce (see [5]) a binary relation ρ\rho in V(G)V(G) in the following way: uρvu\rho v if and only if NG[u]=NG[v]N_{G}[u]=N_{G}[v]. The relation ρ\rho is symmetric and transitive. By this relation the vertex set V(G)V(G) is divided to kk disjoint subsets C1C_{1}, C2C_{2}, \ldots, CkC_{k}, (1kn1\leq k\leq n, n=|V(G)|n=|V(G)|), such that every graph induced by the set CiC_{i} (i=1,,ki=1,\ldots,k) is a complete graph. The graph GcG^{c}, obtained from GG by identification of all vertices from the same subset CiC_{i} (i=1,,ki=1,\ldots,k) is called the “CC-canonical graph” of the graph GG. If V(G)V(G) does not have any pair of vertices which lie in relation ρ\rho, then Gc=GG^{c}=G and we say that GG is a CC-canonical graph. For instance, all connected bipartite graphs distinct of P2P_{2} are CC-canonical graphs..

2 Preliminaries

Lemma 2.1.

[1] Let vv be a vertex of GG, then mG(μ)1mG(μ)mG(μ)+1m_{G}(\mu)-1\leq m_{G}(\mu)\leq m_{G}(\mu)+1

The eigenvalues of PnP_{n} are {2cosiπn+1|i=1,2,,n}\{2cos\frac{i\pi}{n+1}|i=1,2,...,n\}, we have the following lemma.

Lemma 2.2.

mPn+1(1)1m_{P_{n+1}}(-1)\leq 1, further, mPn+1(1)=1m_{P_{n+1}}(-1)=1 if and only if n1(mod 3)n\equiv 1\ (\text{mod}\ 3)

Lemma 2.3.

([7], Lemma 2.3) Let XX be a star set for μ\mu in GG and let UU be a proper subset of XX. Then XUX-U is a star set for μ\mu in GUG-U.

Lemma 2.4.

([3], Proposition 5.1.4) Let XX be a star set for μ\mu in GG, and let X=V(G)\XX=V(G)\backslash X. If μ=0\mu=0, then XX is a dominating set for GG.

Lemma 2.5.

([6, 3.1]) Let HH be an induced subgraph of a connected graph GG for which rank(A(H)+I)rank(A(G)+I)1rank(A(H)+I)\geq rank(A(G)+I)-1, and vV(G)\V(H)v\in V(G)\backslash V(H). If vhv\sim h and NH(v)=NH[h]N_{H}(v)=N_{H}[h] for a vertex hV(H)h\in V(H), then NG[v]=NG[h]N_{G}[v]=N_{G}[h].

Lemma 2.6.

([6, 3.2]) Let HH be an induced subgraph of a connected graph GG for which rank(A(H)+I)rank(A(G)+I)1rank(A(H)+I)\geq rank(A(G)+I)-1, and u,vV(G)\V(H)u,v\in V(G)\backslash V(H). If uvu\sim v and NH(v)=NH(h)N_{H}(v)=N_{H}(h), then NG[v]=NG[h]N_{G}[v]=N_{G}[h].

Lemma 2.7.

([5, Theorem 5]) Let GG be a graph. If two vertices uu, vv satisfy uρvu\rho v, then mG(1)=mGu(1)+1=mGv(1)+1m_{G}(-1)=m_{G-u}(-1)+1=m_{G-v}(-1)+1.

Lemma 2.8.

([6, 3.4]) Let GcG^{c} be the CC-canonical graph of a graph GG. If GG is not a complete graph, then d(Gc)1d(G^{c})\geq 1 and d(Gc)=d(G)d(G^{c})=d(G).

Lemma 2.9.

Let GcG^{c} be the CC-canonical graph of a graph GG. If GG is not a complete graph, then mG(1)=n(G)d(G)1m_{G}(-1)=n(G)-d(G)-1 if and only if mGc(1)=n(Gc)d(Gc)1m_{G^{c}}(-1)=n(G^{c})-d(G^{c})-1.

Proof.

If GG is not a complete graph, by Lemma 2.8, d(Gc)=d(G)d(G^{c})=d(G). Meanwhile, by Lemma 2.7, mG(1)=mGc(1)+n(G)n(Gc)m_{G}(-1)=m_{G^{c}}(-1)+n(G)-n(G^{c}). So, mG(1)=n(G)d(G)1m_{G}(-1)=n(G)-d(G)-1 if and only if mGc(1)+(n(G)n(Gc)=n(G)d(G)1=n(G)d(Gc)1)m_{G^{c}}(-1)+(n(G)-n(G^{c})=n(G)-d(G)-1=n(G)-d(G^{c})-1). Then we have mG(1)=n(G)d(G)1m_{G}(-1)=n(G)-d(G)-1 if and only if mGc(1)=n(Gc)d(Gc)1m_{G^{c}}(-1)=n(G^{c})-d(G^{c})-1. ∎

By Lemmas 2.8 and 2.9, we know that for the eigenvalue 1-1, GG and GcG^{c} possess identical properties; therefore, in this article, we will only consider the case where G=GcG=G^{c}.

Lemma 2.10.

[9, 2.2] Let GG be a graph obtained from a given graph HH and P3P_{3}, disjoint with HH, by identifying one pendant vertex of P3P_{3} with a vertex ww of HH. Then mG(1)=mGw(1)m_{G}(-1)=m_{G-w}(-1)

Lemma 2.11.

If mG(1)=nd1m_{G}(-1)=n-d-1 and Pd+1P_{d+1} is the diameter path of GG with Pd+1HGP_{d+1}\leq H\leq G, then it follows that rank(A(H)+I)rank(A(G)+I)1rank(A(H)+I)\geq rank(A(G)+I)-1.

Proof.

If -1 is not an eigenvalue of Pd+1P_{d+1}, then by the interlacing theorem, we have rank(H+I)=rank(Pd+1)rank(H+I)=rank(P_{d+1}). If -1 is an eigenvalue of Pd+1P_{d+1}, then rank(H+I)=rank(Pd+1+I)rank(H+I)=rank(P_{d+1}+I) or rank(Pd+1+I)+1rank(P_{d+1}+I)+1. This completes the proof. ∎

For xRnx\in R^{n}, we denote by xux_{u} the component of xx corresponding to uV(G)u\in V(G). With respect to a subset UU of V(G)V(G), we define

(U)={xRn:xu=0,uU}\mathbb{Z}(U)=\{x\in R^{n}:x_{u}=0,\forall u\in U\}.

Lemma 2.12.

[9, Lemma 2.1] Let GG be a graph of order nn and UU be a subset of GG. If V(G,μ))Z(U)=0V(G,\mu))\cap Z(U)=0, then mG(μ)|U|m_{G}(\mu)\leq|U|.

Lemma 2.13.

[8] Let GG be a graph, V(G)={1,2,,n}V(G)=\{1,2,...,n\}, XV(G)X\subseteq V(G). Let FF be the matrix which represents the orthogonal projection of RnR^{n} onto the eigenspace A(G)(μ)\mathcal{E}_{A(G)}(\mu) with respect to the standard orthonormal basis {e1,e2,,en}\{e_{1},e_{2},\ldots,e_{n}\} of RnR^{n}. Then

(i) XX is a star set for μ\mu in GG if and only if the vectors FeiFe_{i}, iXi\in X, form a basis for A(G)(μ)\mathcal{E}_{A(G)}(\mu).

(ii)The matrix FF is a polynomial in AA and μFev=ΣuvFeu\mu Fe_{v}=\Sigma_{u\sim v}Fe_{u}.

3 mP(1)=0m_{P}(-1)=0 and mG(1)=nd1m_{G}(-1)=n-d-1

[Uncaptioned image]

After performing computations using a computer, it can be concluded that mHi(1)=0m_{H_{i}}(-1)=0 for 1i61\leq i\leq 6.

[Uncaptioned image]
[Uncaptioned image]
Theorem 3.1.

Let GG be a connected graph and P=Pd+1=v1v2vd+1P=P_{d+1}=v_{1}\sim v_{2}\sim\cdots\sim v_{d+1} be the diameter path of GG. Given that -1 is not an eigenvalue of PP, we have mG(1)nd1m_{G}(-1)\leq n-d-1. The equality holds if and only if PGGiP\leq G\leq G_{i} for 1i31\leq i\leq 3.

Proof.

Since -1 is not an eigenvalue of PP, we can deduce from Lemma 2.1 that mG(1)nd1m_{G}(-1)\leq n-d-1, with equality holding if and only if X=V(GP)X=V(G-P) forms a star set. From Lemma 2.4, we know that PP is a dominating set for GG, which implies NG(P)=V(GP)N_{G}(P)=V(G-P). Let NG(P)=i1NiN_{G}(P)=\cup_{i\geq 1}N_{i} and Ni={x:|NP(x)|=i}N_{i}=\{x:|N_{P}(x)|=i\}.

Claim 1: NG(P)=N1N2N_{G}(P)=N_{1}\cup N_{2}

If i>3i>3 and xNix\in N_{i}, where xx is adjacent to vj1v_{j_{1}}, vj2v_{j_{2}}, vj3v_{j_{3}}, vj4v_{j_{4}} with j1<j2<j3<j4j_{1}<j_{2}<j_{3}<j_{4}, then the path v1v2vj1xvj4vd+1v_{1}v_{2}\ldots v_{j_{1}}xv_{j_{4}}\ldots v_{d+1} forms a path shorter than PP, which contradicts the fact that PP is a diameter path. If i=3i=3 and xN3x\in N_{3}, with xx being adjacent to vj1v_{j_{1}}, vj2v_{j_{2}}, vj3v_{j_{3}}, it follows that j3=j1+2=j2+1j_{3}=j_{1}+2=j_{2}+1. If this condition is not satisfied, then v1v2vj1xvj3vd+1v_{1}v_{2}\ldots v_{j_{1}}xv_{j_{3}}\ldots v_{d+1} would again represent a path shorter than PP, leading to a contradiction. When j3=j1+2=j2+1j_{3}=j_{1}+2=j_{2}+1, we find that NP(x)=NP[vj2]N_{P}(x)=N_{P}[v_{j_{2}}]. According to Lemma 2.5, this implies NG[x]=NG[vj2]N_{G}[x]=N_{G}[v_{j_{2}}], which contradicts the assumption that G=GcG=G^{c}. Therefore, we conclude that NG(P)=N1N2N_{G}(P)=N_{1}\cup N_{2}.

Claim 2: If x,yV(GP)x,y\in V(G-P), then NP(x)NP(y)N_{P}(x)\neq N_{P}(y).

If xyx\sim y, then NP[x]=NP[y]N_{P}[x]=N_{P}[y]. According to Lemmas 2.6 and 2.11, it follows that NG[x]=NG[y]N_{G}[x]=N_{G}[y], which contradicts the condition G=GcG=G^{c}. If xyx\nsim y, consider P+x+yP+x+y. By Lemma 2.13, we have Fex=FeyFe_{x}=Fe_{y}, which contradicts the fact that FexFe_{x} and FeyFe_{y} are linearly independent. Therefore, we conclude that NP(x)NP(y)N_{P}(x)\neq N_{P}(y).

Since mP(1)=0m_{P}(-1)=0, we can infer from Lemma 2.2 that d0d\equiv 0 or 2(mod 3)2\ (\text{mod}\ 3) . Therefore, we will discuss two cases in the following analysis.

Case 1: d2(mod 3)d\equiv 2\ (\text{mod}\ 3)

Claim 1.1: N1=N_{1}=\emptyset.

First, by the interlacing theorem, we know that mP+u(1)=1m_{P+u}(-1)=1.

If N1N_{1}\neq\emptyset, let xN1x\in N_{1} such that xvmx\sim v_{m}. If m0m\equiv 0 or 1mod31\mod 3, then by Lemma 2.10, we have mP+x(1)=0m_{P+x}(-1)=0, leading to a contradiction. If m2mod3m\equiv 2\mod 3, it follows that mP+x(1)=mK1,3=0m_{P+x}(-1)=m_{K_{1,3}}=0, which also leads to a contradiction. Therefore, we conclude that N1=N_{1}=\emptyset.

Claim 1.2: If uN2u\in N_{2} and uu is adjacent to viv_{i} and vjv_{j} with i<ji<j, then it follows that j=i+1j=i+1.

If ji>2j-i>2, then the path v1v2viuvjvd+1v_{1}v_{2}\ldots v_{i}uv_{j}\ldots v_{d+1} is shorter than PP, leading to a contradiction.

When j=i+2j=i+2 and i0(mod 3)i\equiv 0\ (\text{mod}\ 3), then mP+u(1)=0m_{P+u}(-1)=0, resulting in another contradiction.

When j=i+2j=i+2 and i1(mod 3)i\equiv 1\ (\text{mod}\ 3), then mP+u(1)=mH4(1)=0m_{P+u}(-1)=m_{H_{4}}(-1)=0, resulting in yet another contradiction.

The case for j=i+2j=i+2 and i2(mod 3)i\equiv 2\ (\text{mod}\ 3) can be similarly analyzed, leading to a contradiction. In summary, we conclude that j=i+1j=i+1.

Claim 1.3: If uN2u\in N_{2} and uu is adjacent to vmv_{m} and vm+1v_{m+1}, then m0m\equiv 0 or 1(mod 3)1\ (\text{mod}\ 3).

When m0(mod 3)m\equiv 0\ (\text{mod}\ 3), then by using Lemma 2.10, we obtain mP+u(1)=0m_{P+u}(-1)=0, which leads to a contradiction.

When m1(mod 3)m\equiv 1\ (\text{mod}\ 3), then mP+u(1)=1m_{P+u}(-1)=1, satisfying the condition.

When m2(mod 3)m\equiv 2\ (\text{mod}\ 3), then mP+u(1)=1m_{P+u}(-1)=1, satisfying the condition. Thus, the proof is established.

Claim 1.4: If u,vN2u,v\in N_{2} with uvm,vm+1u\sim v_{m},v_{m+1} and vvh,vh+1v\sim v_{h},v_{h+1}, we may assume m<hm<h. Then, u≁vu\not\sim v and PGG1P\leq G\leq G_{1}.

First, let us assume that uvu\sim v. When m1(mod 3)m\equiv 1\ (\text{mod}\ 3), since PP is a diameter path, we only need to consider the case where h=m+1h=m+1. In this case, mP+u+v(1)=1m_{P+u+v}(-1)=1, which leads to a contradiction. Similarly, when m2(mod 3)m\equiv 2\ (\text{mod}\ 3), we only need to consider the case where h=m+2h=m+2. In this situation, mP+u+v(1)=mH5(1)=0m_{P+u+v}(-1)=m_{H_{5}}(-1)=0, which also results in a contradiction. Thus, it follows that u≁vu\not\sim v.

If m1(mod 3)m\equiv 1\ (\text{mod}\ 3) and h2(mod 3)h\equiv 2\ (\text{mod}\ 3), then mP+u+v(1)=2m_{P+u+v}(-1)=2, which satisfies the conditions. If m2(mod 3)m\equiv 2\ (\text{mod}\ 3) and h1(mod 3)h\equiv 1\ (\text{mod}\ 3), then mP+u+v(1)=mH2(1)=0m_{P+u+v}(-1)=m_{H_{2}}(-1)=0, leading to a contradiction. If both m1(mod 3)m\equiv 1\ (\text{mod}\ 3) and h1(mod 3)h\equiv 1\ (\text{mod}\ 3) (or if m2(mod 3)m\equiv 2\ (\text{mod}\ 3) and h2(mod 3))h\equiv 2\ (\text{mod}\ 3)), we can apply Lemmas 2.7 and 2.10, which yield mP+u+v(1)=2m_{P+u+v}(-1)=2, satisfying the conditions. Therefore, based on the above discussion, G1G_{1} is a maximal graph that satisfies the given conditions. Furthermore, by reapplying Lemmas 2.7 and 2.10, we obtain mG1(1)=t=nd1m_{G_{1}}(-1)=t=n-d-1, which concludes the proof.

We will now begin to consider the case where d0(mod 3)d\equiv 0\ (\text{mod}\ 3), while omitting certain proofs that are similar to those in Case 1.

Caes2: d0(mod 3)d\equiv 0\ (\text{mod}\ 3)

Claim 2.1: If xN1x\in N_{1} and xsinvmx\sin v_{m}, then m1(mod 3)m\equiv 1\ (\text{mod}\ 3).

If m0m\equiv 0 or 22 (mod 3)\ (\text{mod}\ 3), then mP+u(1)=0m_{P+u}(-1)=0, leading to a contradiction.

If m1(mod 3)m\equiv 1\ (\text{mod}\ 3), then mP+u(1)=1m_{P+u}(-1)=1, satisfying the condition.

Claim 2.2: If x,yN1x,y\in N_{1}, then xyx\sim y. Therefore, |N1|2|N_{1}|\leq 2.

Let xvmx\sim v_{m} and yvhy\sim v_{h}. According to Claim 2.1, we know that m=3t+1m=3t+1 and h=3l+1h=3l+1. If x≁yx\not\sim y, then mP+x+y(1)=mP3(1)=0m_{P+x+y}(-1)=m_{P_{3}}(-1)=0, which leads to a contradiction. If xyx\sim y, we may assume m<hm<h, thus h=m+3h=m+3. Otherwise, the sequence v1v2vmxyvhvd+1v_{1}v_{2}\ldots v_{m}xyv_{h}\ldots v_{d+1} would constitute a path shorter than PP. In this case, we have mP+x+y(1)=mC6(1)=2m_{P+x+y}(-1)=m_{C_{6}}(-1)=2, which satisfies the conditions.

Claim 2.3: If uN2u\in N_{2} and uu is adjacent to viv_{i} and vjv_{j} with i<ji<j, then it follows that j=i+1j=i+1.

Claim 2.4: If uN2u\in N_{2} and uu is adjacent to vmv_{m} and vm+1v_{m+1}, m0m\equiv 0 or 1(mod 3)1\ (\text{mod}\ 3).

When m0(mod 3)m\equiv 0\ (\text{mod}\ 3), then by using Lemma 2.10, we obtain mP+u(1)=mP2(1)=1m_{P+u}(-1)=m_{P_{2}}(-1)=1, satisfying the condition.

When m1(mod 3)m\equiv 1\ (\text{mod}\ 3), then mP+u(1)=mP2(1)=1m_{P+u}(-1)=m_{P_{2}}(-1)=1, satisfying the condition.

When m2(mod 3)m\equiv 2\ (\text{mod}\ 3), then mP+u(1)=mH1=0m_{P+u}(-1)=m_{H_{1}}=0, leading to a contradiction;

Claim 2.5: Let u,vN2u,v\in N_{2} such that uv3a,v3a+1u\sim v_{3a},v_{3a+1}, and vv3b+1,v3b+2v\sim v_{3b+1},v_{3b+2}. If bab\geq a, then it follows that uvu\sim v and b=ab=a. Conversely, if b<ab<a, then uvu\nsim v.

If bab\geq a and u≁vu\not\sim v, then according to Lemma 2.10, we have mP+u+v(1)=0<2m_{P+u+v}(-1)=0<2, which leads to a contradiction. Furthermore, since PP is a diameter path, we can conclude that a=ba=b.

If b<ab<a and uvu\sim v, considering that PP is a diameter path, we have b=a1b=a-1. Thus, mP+u+v(1)=mH6(1)=0<2m_{P+u+v}(-1)=m_{H_{6}}(-1)=0<2, which leads to a contradiction. Therefore, it follows that uvu\nsim v.

Claim 2.6: If xN1x\in N_{1}, uN2u\in N_{2}, xvix\sim v_{i}, uvj,vj+1u\sim v_{j},v_{j+1}, and u≁xu\not\sim x, then if j>ij>i, it follows that j0(mod 3)j\equiv 0\ (\text{mod}\ 3). Conversely, by symmetry, if j<ij<i, then j1(mod 3)j\equiv 1\ (\text{mod}\ 3).

From Claim 2.4, we know that j0j\equiv 0 or 1(mod 3)1\ (\text{mod}\ 3). If j>ij>i and j1(mod 3)j\equiv 1\ (\text{mod}\ 3), then mP+x+u(1)=mP3(1)=0m_{P+x+u}(-1)=m_{P_{3}}(-1)=0, which leads to a contradiction. On the other hand, if j>ij>i and j0(mod 3)j\equiv 0\ (\text{mod}\ 3), then mP+x+u(1)=mC3(1)=2m_{P+x+u}(-1)=m_{C_{3}}(-1)=2, satisfying the conditions. By symmetry, the proof is completed.

Claim 2.7: Let xN1x\in N_{1} and uN2u\in N_{2}, with xvix\sim v_{i}, uvju\sim v_{j}, vj+1v_{j+1}. If xux\sim u, then it follows that j=ij=i or j=i1j=i-1.

By symmetry, it is sufficient to prove that when jij\geq i, then j=ij=i. If j1(mod 3)j\equiv 1\ (\text{mod}\ 3), then i=ji=j; otherwise, v1v2vixuvj+1vd+1v_{1}v_{2}\ldots v_{i}xuv_{j+1}\ldots v_{d+1} would form a path shorter than PP ; when i=ji=j, we have mP+u+v(1)=mC3(1)=2m_{P+u+v}(-1)=m_{C_{3}}(-1)=2, which satisfies the conditions. If j0(mod 3)j\equiv 0\ (\text{mod}\ 3), then j=i+2j=i+2; otherwise, a path shorter than PP would emerge. When j=i+2j=i+2, we find mP+u+v(1)=mH3(1)=0m_{P+u+v}(-1)=m_{H_{3}}(-1)=0, leading to a contradiction. Therefore, the conclusion is established.

Claim 2.8: Let N1={x,y}N_{1}=\{x,y\}. If there are two vertices u,vN2u,v\in N_{2} that are adjacent to vertices in N1N_{1}, then P+u+x+yH7P+u+x+y\cong H_{7}.

According to claims 2.6 and 2.7, when there are two vertices u,vN2u,v\in N_{2} that are adjacent to vertices in N1N_{1}, it follows that P+u+v+x+yH7P+u+v+x+y\cong H_{7}. Based on Lemmas 2.7 and 2.10, we have mH7(1)=2+mC6(1)=4m_{H_{7}}(-1)=2+m_{C_{6}}(-1)=4, which satisfies the conditions and thereby concludes the proof.

[Uncaptioned image]

Claim 2.9: When d0(mod 3)d\equiv 0\ (\text{mod}\ 3), PGG2P\leq G\leq G_{2} or G3G_{3}.

Based on the above discussion, it can be concluded that G2G_{2} or G3G_{3} are maximal graphs that satisfy the given conditions. Furthermore, according to Lemmas 2.7 and 2.10, we obtain mG2(1)=mG3(1)=i+k+4=nd1m_{G_{2}}(-1)=m_{G_{3}}(-1)=i+k+4=n-d-1, which satisfies the conditions. The proof is thereby established by the alternating theorem. ∎

4 mP(1)=1m_{P}(-1)=1 and mG(1)=nd1m_{G}(-1)=n-d-1

Theorem 4.1.

Let GG be a connected graph and P=Pd+1=v1v2vd+1P=P_{d+1}=v_{1}\sim v_{2}\sim\cdots\sim v_{d+1} be the diameter path of GG. If mP(1)=1m_{P}(-1)=1 and mG(1)=nd1m_{G}(-1)=n-d-1, then P<GG4P<G\leq G_{4} and N1N_{1}\neq\emptyset.

[Uncaptioned image]
Proof.

Initially, since mP(1)=1m_{P}(-1)=1, we conclude from Lemma 2.2 that d=3k+1d=3k+1. Furthermore, there cannot exist a point zz that is at a distance of 3 from PP. To illustrate this, suppose ( z ) is a point at a distance of 3 from PP, yy is a vertex at a distance of 2 from PP, and xx is adjacent to PP. According to Lemma 2.10, we then have mP+x+y+z(1)=mP(1)=1<2m_{P+x+y+z}(-1)=m_{P}(-1)=1<2, which leads to a contradiction.

Claim 1: If xN1x\in N_{1} and xvix\sim v_{i}, then mP+x(1)=1m_{P+x}(-1)=1 and i0mod3i\equiv 0\mod{3}.

According to Lemma 2.11, we have mP+x(1)1+dim(V(G,1)Z(v1)m_{P+x}(-1)\leq 1+\dim(V(G,-1)\cap Z(v_{1}). Since dim(V(G,1)Z(v1))=0\dim(V(G,-1)\cap Z(v_{1}))=0, it follows that mP+x(1)1m_{P+x}(-1)\leq 1. If mP+u(1)=0m_{P+u}(-1)=0, then it would imply that mG(1)nd2m_{G}(-1)\leq n-d-2, which leads to a contradiction. Therefore, we conclude that mP+u(1)=1m_{P+u}(-1)=1. This also indicates that V(GPx)V(G-P-x) is a star set.

If i0(mod 3)i\equiv 0\ (\text{mod}\ 3), then by utilizing Lemma 2.10, we obtain mP+x(1)=mP2(1)=1m_{P+x}(-1)=m_{P_{2}}(-1)=1, which satisfies the condition. If m1m\equiv 1 or 2(mod 3)2\ (\text{mod}\ 3), then mP+u(1)=0m_{P+u}(-1)=0, leading to a contradiction.

Claim 2: If x,yN1x,y\in N_{1} and xv3ax\sim v_{3a}, yv3by\sim v_{3b}, then aba\neq b and uvu\sim v. Hence, it follows that |N1|2|N_{1}|\leq 2.

First, we demonstrate that aba\neq b. If a=ba=b and xyx\sim y, then NP(x)=NP(y)N_{P}(x)=N_{P}(y). By Lemma 2.6, this implies NG[x]=NG[y]N_{G}[x]=N_{G}[y], which leads to a contradiction. Alternatively, if a=ba=b and x≁yx\not\sim y, then mP+x+y(1)=1<2m_{P+x+y}(-1)=1<2, resulting in another contradiction.

If x≁yx\not\sim y, then by Claim 1 and Lemma 2.10, we have mP+x+y=mP2(1)=1<2m_{P+x+y}=m_{P_{2}}(-1)=1<2, which leads to a contradiction. Hence, it follows that xyx\sim y. Furthermore, considering that PP is a diameter path, we conclude that |N1|2|N_{1}|\leq 2.

Claim 3: Let uN2u\in N_{2} and uvm,vm+1u\sim v_{m},v_{m+1}. Then m1(mod 3)m\equiv 1\ (\text{mod}\ 3) and mP+u(1)=2m_{P+u}(-1)=2.

First, we know that 1mP+u(1)21\leq m_{P+u}(-1)\leq 2. If m0(mod 3)m\equiv 0\ (\text{mod}\ 3) or m2(mod 3)m\equiv 2\ (\text{mod}\ 3) , then mP+u(1)=0m_{P+u}(-1)=0, which leads to a contradiction. If m1(mod 3)m\equiv 1\ (\text{mod}\ 3), then mP+u(1)=mC3(1)=2m_{P+u}(-1)=m_{C_{3}}(-1)=2, thus proving the claim. Similar to Claim 2, for u,vN2u,v\in N_{2}, we have NP(u)NP(v)N_{P}(u)\neq N_{P}(v).

Claim 4: Let xN1x\in N_{1} and u,vN2u,v\in N_{2}. Then x≁N2x\not\sim N_{2} and u≁vu\not\sim v.

Assume uvu\sim v. According to Claim 3, we may assume uv3a+1u\sim v_{3a+1}, v3a+2v_{3a+2}, and vv3b+1v\sim v_{3b+1}, v3b+2v_{3b+2}. Considering that PP is a diameter path, we have |ba|=1|b-a|=1. Thus, mP+u+v(1)=mH8(1)=1<2m_{P+u+v}(-1)=m_{H_{8}}(-1)=1<2, which leads to a contradiction. If xN2x\sim N_{2}, we may assume xv3cx\sim v_{3c} and uu. Given that PP is a diameter path, we have |bc|=1|b-c|=1. Therefore, mP+x+u(1)=1<2m_{P+x+u}(-1)=1<2, which results in another contradiction.

Next, we begin to consider the vertices that are at a distance of 2 from PP.

Claim 5: If d(z,P)=2d(z,P)=2, then zN1z\sim N_{1} and z≁N2z\not\sim N_{2}.

If xN1x\in N_{1} and zxz\sim x, then mP+x+z(1)=2m_{P+x+z}(-1)=2, satisfying the condition. On the other hand, if uN2u\in N_{2} and zuz\sim u, then mP+u+z(1)=1<2m_{P+u+z}(-1)=1<2, which leads to a contradiction.

Claim 6: Let N2={zd(z,P)=2}N^{2}=\{z\mid d(z,P)=2\}. Then it follows that |N2|1|N^{2}|\leq 1.

If z,wNz,w\in N, by Claim 5, we know there exist x,yN1x,y\in N_{1} such that xzx\sim z and ywy\sim w. If x=yx=y and zwz\sim w, let H=P+xH=P+x; then NH(z)=NH(w)N_{H}(z)=N_{H}(w). By Lemma 2.5, we have NG[z]=NG[w]N_{G}[z]=N_{G}[w], which leads to a contradiction. If x=yx=y and z≁wz\not\sim w, then mP+x+z+w(1)=1<3m_{P+x+z+w}(-1)=1<3, which is also a contradiction. Thus, we conclude that xyx\neq y. At this point, based on Claim 1 and Claim 2, we can assume without loss of generality that xv3ax\sim v_{3a} and yv3a+3y\sim v_{3a+3}. If z≁yz\not\sim y, then mP+x+y+z(1)=1<2m_{P+x+y+z}(-1)=1<2, which is again a contradiction. If zyz\sim y, then mPx+y+z(1)=1+mC3(1)=3m_{P{x+y+z}}(-1)=1+m_{C_{3}}(-1)=3, satisfying the condition, which implies zx,yz\sim x,y. Similarly, we can conclude that wx,yw\sim x,y. If zwz\sim w, let H=P+x+y)H=P+x+y); then NH(z)=NH(w)N_{H}(z)=N_{H}(w). By Lemma 2.6, we know NG[z]=NG[w]N_{G}[z]=N_{G}[w], leading to a contradiction. If z≁wz\not\sim w, then mP+x+y+z+w(1)=2<4m_{P+x+y+z+w}(-1)=2<4, which is also a contradiction. In summary, we have |N2|1|N^{2}|\leq 1. If zN2z\in N^{2} and x,yN1x,y\in N_{1}, then it follows that zx,yz\sim x,y.

Claim 7: P<GG4P<G\leq G_{4} and N1N_{1}\neq\emptyset.

First, based on the discussion above, we know that G4G_{4} is a maximal graph that satisfies the conditions. Furthermore, by applying Lemmas 2.7 and 2.10, we can conclude that mG4(1)=m+3=nd1m_{G_{4}}(-1)=m+3=n-d-1, which satisfies the required conditions. If N1=N_{1}=\emptyset, then V(G)=V(P)N2V(G)=V(P)\cup N_{2}, leading to mG(1)=ndm_{G}(-1)=n-d, which presents a contradiction. Therefore, we conclude that N1N_{1}\neq\emptyset, and the theorem is proven.

References

  • [1] A. Brouwer, W. Haemers, Spectra of Graphs, Springer, New York, 2011.
  • [2] S. Chang, J. Li, Graphs GG with nullity n(G)g(G)1n(G)-g(G)-1, Linear Algebra Appl. 642 (202) 251-263.
  • [3] D. Cvetkovic´\acute{c}, P. Rowlinson, S. Simic´\acute{c}, An Introduction to the Theory of Graph Spectra, Cambridge University Press, Cambridge, 2010.
  • [4] Z. Du, C. Fonseca, The characterization of the minimal weighted acyclic graphs, Linear Algebra Appl. 672 (2023) 75–92.
  • [5] M. Petrovic´\acute{c}, On graphs with exactly one eigenvalue less than 1, J. Comb. Theory, Ser. B 52 (1991) 102–112.
  • [6] Y. Zhang, W. Zhen, D. Wang, Minimal graphs with eigenvalue multiplicity of ndn-d, Linear Algebra Appl. 699 (2024) 47–58.
  • [7] P. Rowlinson, An extension of the star complement technique for regular graphs, Linear Algebra Appl. 557 (2018) 496–507.
  • [8] P. Rowlinson, On multiple eigenvalues of trees, Linear Algebra Appl. 432 (2010) 3007–3011.
  • [9] X. Wang, D. Wong, L. Wei, F. Tian, On the multiplicity of -1 as an eigenvalue of a tree with given number of pendant vertices, Linear and Multilinear Algebra, 70:17, 3345-3353.
  • [10] D. Wong, M. Zhu, W. Lv, A characterization of long graphs of arbitrary rank, Linear Algebra Appl. 438 (2013) 1347–1355.
  • [11] Z. Wang, The multiplicities of eigenvalues of a graph, Appl. Math. Comput. 421 (2022) 126957.