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

Two spectral extremal results for graphs with given order and rank

Xiuqing Li,  Xian’an Jin111Corresponding author,  Chao Shi,  Ruiling Zheng
School of Mathematical Sciences, Xiamen University,
Xiamen, Fujian 361005, P. R. China
E-mails: [email protected]; [email protected];
   [email protected]; [email protected]
Abstract

The spectral radius and rank of a graph are defined to be the spectral radius and rank of its adjacency matrix, respectively. It is an important problem in spectral extremal graph theory to determine the extremal graph that has the maximum or minimum spectral radius over certain families of graphs. Monsalve and Rada [Extremal spectral radius of graphs with rank 4, Linear Algebra Appl. 609 (2021) 1–11] obtained the extremal graphs with maximum and minimum spectral radii among all graphs with order nn and rank 44. In this paper, we first determine the extremal graph which attains the maximum spectral radius among all graphs with any given order nn and rank rr, and further determine the extremal graph which attains the minimum spectral radius among all graphs with order nn and rank 55.

Keywords: Rank of graphs; Extremal graphs; Maximum spectral radius; Minimum spectral radius

journal: Linear Algebra and its Applications

1 Introduction

Graphs considered in the paper are all simple, connected and undirected. Let G=(V(G),E(G))G=(V(G),E(G)) be a graph. For vV(G)v\in V(G), the degree d(v)d(v) is the cardinality of the neighborhood NG(v)N_{G}(v) (or N(v)N(v) for short) of vv in GG. Let A(G)A(G) be the adjacency matrix of GG. The characteristic polynomial of a graph GG is the determinantal expansion of xIA(G)xI-A(G), denoted by ϕ(G,x)\phi(G,x). According to the famous Perron-Frobenius theorem, the largest eigenvalue ρ(G)\rho(G) of A(G)A(G) is exactly the spectral radius of GG and there is a unique positive unit eigenvector corresponding to ρ(G)\rho(G), called the principal eigenvector of GG.

Let GG be a graph with vertex set V(G)={v1,v2,,vk}V(G)=\{v_{1},v_{2},\dots,v_{k}\} and m=(n1,n2,,nk)\textbf{m}=(n_{1},n_{2},\dots,n_{k}) be a vector of positive integers. Denote by GmG\circ\textbf{m}, the graph obtained from GG by replacing each vertex viv_{i} with an independent set ViV_{i} with nin_{i} vertices vi1,vi2,,viniv_{i}^{1},v_{i}^{2},\dots,v_{i}^{n_{i}} and joining each vertex in ViV_{i} with each vertex in VjV_{j} if and only if vivjE(G)v_{i}v_{j}\in E(G). The resulting graph GmG\circ\textbf{m} is said to be obtained from GG by multiplication of vertices by Chang, Huang and Yeh in [1]. Further, let GG be a graph of order kk, we define Mn(G)M_{n}(G) to be the set of all graphs G(n1,n2,,nk)G\circ(n_{1},n_{2},\dots,n_{k}) with i=1kni=n\sum_{i=1}^{k}n_{i}=n. Moreover, for a given set of graphs {H1,,Hl}\{H_{1},\dots,H_{l}\}, we denote the set i=1lMn(Hi)\bigcup_{i=1}^{l}M_{n}(H_{i}) by Mn(H1,,Hl)M_{n}(H_{1},\dots,H_{l}).

Let GG be a connected graph of order nn and R(G)R(G) be its rank. Sciriha [4] proved that R(G)=iR(G)=i if and only if GMn(Ki)G\in M_{n}(K_{i}) for i=2,3i=2,3, where KiK_{i} is the complete graph of order ii. Chang, Huang and Yeh [1, 5] characterized the set of all connected graphs with rank 4 and 5, respectively. They obtained the set of connected graphs of order nn and rank 5 is

Mn(G1,G2,,G24),M_{n}(G_{1},G_{2},\dots,G_{24}),

where the graphs G1,G2,,G24G_{1},G_{2},\dots,G_{24} are shown in Figure 1.

Refer to caption
Figure 1: Reduced graphs of rank 5.

For a given class of graphs 𝒢\mathscr{G}, there are many results on characterizing the extramal graphs with maximum and minimum spectral radius among Mn(𝒢)M_{n}(\mathscr{G}). For example, in [6], Stevanović, Gutman and Rehman determined the extremal graphs with the maximum and minimum spectral radii in Mn(Kp)M_{n}(K_{p}). Monsalve and Rada [7] obtained the extremal graphs with maximum and minimum spectral radii among all connected graphs of order nn and rank 44. In the same article, they conjectured that in Mn(Pk)M_{n}(P_{k}), Pk(1,,1,nk+22,nk+22,1,,1)P_{k}\circ(1,\dots,1,\lfloor\frac{n-k+2}{2}\rfloor,\lceil\frac{n-k+2}{2}\rceil,1,\dots,1) and Pk(nk+22,1,,1,P_{k}\circ(\lfloor\frac{n-k+2}{2}\rfloor,1,\dots,1, nk+22)\lceil\frac{n-k+2}{2}\rceil) attain the maximum and minimum spectral radius, respectively, and Ck(nk+22,nk+22,1,,1)C_{k}\circ(\lfloor\frac{n-k+2}{2}\rfloor,\lceil\frac{n-k+2}{2}\rceil,1,\dots,1) attains the maximum spectral radius in Mn(Ck)M_{n}(C_{k}). Recently, Lou, Zhai [2] and Sun, Das [3] independently proved the above conjectures on the extremal graphs with the maximum spectral radius in Mn(Pk)M_{n}(P_{k}) and Mn(Ck)M_{n}(C_{k}) by using different techniques, and they independently constructed a class of graphs disproving the conjecture on the minimum spectral radius in Mn(Pk)M_{n}(P_{k}).

The Turán graph T(n,r)T(n,r) is the complete rr-partite graph on nn vertices where its part sizes are as equal as possible. In this paper, we first determine the extremal graph that attains the maximum spectral radius with any given order and rank, and obtain:

Theorem 1.1.

T(n,r)T(n,r) is the unique extremal graph that attains the maximum spectral radius among all graphs of order nn and rank rr.

However, it seems that it is a difficult task to find the extremal graph that attains the minimum spectral radius with given order and rank. In this paper, we focus on graphs with order nn and rank 55, and obtain:

Theorem 1.2.

The extremal graph that attains the minimum spectral radius among all connected graphs of order nn and rank 5 is:

  • 1.

    G7=C5G_{7}=C_{5}, for n=5n=5;

  • 2.

    G1(1,1,1,1,n4)G_{1}\circ(1,1,1,1,n-4), for 6n106\leq n\leq 10;

  • 3.

    G10(1,1,1,1,1,n5)G_{10}\circ(1,1,1,1,1,n-5), for n=11n=11;

  • 4.

    G10(1,1,1,1,k,nk4)G_{10}\circ(1,1,1,1,k,n-k-4), where k=6n3724n+118k=\lfloor\frac{6n-37-\sqrt{24n+1}}{18}\rfloor or 6n3724n+118\lceil\frac{6n-37-\sqrt{24n+1}}{18}\rceil, for n12n\geq 12.

2 The proof of Theorem 1.1

We will use the following results to prove Theorem 1.1.

Theorem 2.1.

[1] Suppose that GG and HH are two graphs. If HMn(G)H\in M_{n}(G), then R(H)=R(G)R(H)=R(G).

Theorem 2.2.

[8] Let T(n,r)T(n,r) be the rr-partite Turán graph of order n. If GG is a Kr+1K_{r+1}-free graph of order nn, then ρ(G)<ρ(T(n,r))\rho(G)<\rho(T(n,r)) unless G=T(n,r)G=T(n,r).

Proof of Theorem 1.1.

Let GG be a graph of order nn and rank rr. We claim that GG is a Kr+1K_{r+1}-free graph. Otherwise, since Kr+1K_{r+1} is a subgraph of GG, selecting the rows and columns corresponding to the vertices in Kr+1K_{r+1} can obtain a nonzero minor of order r+1r+1 of A(G)A(G), i.e.,

det(011101110)(r+1)×(r+1)=(1)rr0.\displaystyle\det\left(\begin{array}[]{cccc}0&1&\cdots&1\\ 1&0&\cdots&1\\ \vdots&\vdots&\ddots&\vdots\\ 1&1&\cdots&0\\ \end{array}\right)_{(r+1)\times(r+1)}=(-1)^{r}\cdot r\neq 0. (5)

Therefore, we have R(G)r+1R(G)\geq r+1, a contradiction. Since T(n,r)=Kr(nr,,nr,nr,,nr)Mn(Kr)T(n,r)=K_{r}\circ(\lceil\frac{n}{r}\rceil,\dots,\lceil\frac{n}{r}\rceil,\lfloor\frac{n}{r}\rfloor,\dots,\lfloor\frac{n}{r}\rfloor)\in M_{n}(K_{r}), by Theorem 2.1, we have R(T(n,r))=R(Kr)=rR(T(n,r))=R(K_{r})=r. By Theorem 2.2, we obtain ρ(G)<ρ(T(n,r))\rho(G)<\rho(T(n,r)) unless G=T(n,r)G=T(n,r). ∎

3 The proof of Theorem 1.2

In this section, we focus on the extremal graph that has the minimum spectral radius among all connected graphs of order nn and rank 5. We firstly outline our proof for Theorem 1.2.

Step 1. We first apply a result of Monsalve and Rada in [7] to prove that the extremal graph with minimum spectral radius belongs to Mn(G1,G7M_{n}(G_{1},G_{7}, G10)G_{10}).

Step 2. Then, using the method of comparing characteristic polynomials, we characterize the extremal graph with minimum spectral radius in Mn(G1)M_{n}(G_{1}), Mn(G7)M_{n}(G_{7}) and Mn(G10)M_{n}(G_{10}), respectively.

Step 3. Next, for n12n\geq 12, we compare the spectral radii of these three types of extremal graphs by some well-known results and obtain that the extremal graph of order nn and rank 55 with minimum spectral radius is G10(1,1,1,1,k,n4k)G_{10}\circ(1,1,1,1,k,n-4-k) for some integer kk. Further, we determine k{6n3724n+118,6n3724n+118}k\in\{\lfloor\frac{6n-37-\sqrt{24n+1}}{18}\rfloor,\lceil\frac{6n-37-\sqrt{24n+1}}{18}\rceil\}.

Step 4. Finally, for 5n115\leq n\leq 11, we obtain the extremal graphs by calculating directly the spectral radii of the extremal graphs in Mn(G1)M_{n}(G_{1}), Mn(G7)M_{n}(G_{7}) and Mn(G10)M_{n}(G_{10}), respectively.

3.1 Step 1

We begin with recalling a well-known result.

Theorem 3.1.

[9] If HH is a proper subgraph of a connected graph GG, then ρ(H)<ρ(G)\rho(H)<\rho(G).

In [7], Theorem 3.1 is used to prove the following results.

Theorem 3.2.

[7] Let GG be a connected graph with kk vertices and m=(n1,n2,,nk)\textbf{m}=(n_{1},n_{2},\dots,n_{k}) a vector of positive integers. If v1v2E(G)v_{1}v_{2}\in E(G), then

ρ((Gv1v2)m)<ρ(Gm).\rho((G-v_{1}v_{2})\circ\textbf{m})<\rho(G\circ\textbf{m}).
Theorem 3.3.

[7] Let GG be a connected graph with kk vertices and m=(n1,n2,,nk)\textbf{m}=(n_{1},n_{2},\dots,n_{k}) a vector of positive integers. If vivjE(G)v_{i}v_{j}\notin E(G) and N(vi)N(vj)N(v_{i})\subsetneq N(v_{j}), then

ρ(G(n1,,ni,,nj,,nk))<ρ(G(n1,,ni1,,nj+1,,nk)).\rho(G\circ(n_{1},\dots,n_{i},\dots,n_{j},\dots,n_{k}))<\rho(G\circ(n_{1},\dots,n_{i}-1,\dots,n_{j}+1,\dots,n_{k})).

By Theorem 3.2, we obtain the following proposition.

Proposition 3.4.

Let GG be the extremal graph with minimum spectral radius among all connected graphs of order nn and rank 5. Then GMn(G1,G7,G10)G\in M_{n}(G_{1},G_{7},G_{10}).

Proof.

Let m1=(n1,n2,n3,n4,n5),\textbf{m}_{1}=(n_{1},n_{2},n_{3},n_{4},n_{5}), m2=(n1,n2,n3,n4,n5,n6),\textbf{m}_{2}=(n_{1},n_{2},n_{3},n_{4},n_{5},n_{6}), m3=(n1,n2,\textbf{m}_{3}=(n_{1},n_{2}, n3,n4,n5,n6,n7)n_{3},n_{4},n_{5},n_{6},n_{7}) and m4=(n1,n2,n3,\textbf{m}_{4}=(n_{1},n_{2},n_{3}, n4,n5,n6,n7,n8)n_{4},n_{5},n_{6},n_{7},n_{8}) be arbitrary vectors of positive integers with i=1ni=n\sum_{i=1}n_{i}=n. As a consequence of Theorem 3.2, we have

ρ(G1m1)<ρ(Gim1),i=2,3,4,5,6,8,\displaystyle\rho(G_{1}\circ\textbf{m}_{1})<\rho(G_{i}\circ\textbf{m}_{1}),i=2,3,4,5,6,8,
ρ(G10m2)<ρ(Gjm2),j=11,12,13,14,15.\displaystyle\rho(G_{10}\circ\textbf{m}_{2})<\rho(G_{j}\circ\textbf{m}_{2}),j=11,12,13,14,15.

Thus,

GMn(G1,G7,G9,G10,G16,G17,G18,G19,G20,G21,G22,G23,G24).G\in M_{n}(G_{1},G_{7},G_{9},G_{10},G_{16},G_{17},G_{18},G_{19},G_{20},G_{21},G_{22},G_{23},G_{24}).

Let H1=G1(1,1,1,1,2)H_{1}=G_{1}\circ(1,1,1,1,2), H2=G10(1,1,1,1,1,2)H_{2}=G_{10}\circ(1,1,1,1,1,2), H3=G10(1,1,1,1,2,1)H_{3}=G_{10}\circ(1,1,1,1,2,1) and H4=G10(1,1,1,1,1,3)H_{4}=G_{10}\circ(1,1,1,1,1,3), as shown in Figure 2.

Obiviously,

  • 1.

    H1H_{1} is the spanning proper subgraph of G9G_{9};

  • 2.

    H2H_{2} is the spanning proper subgraph of Gi,i{16,17,18,19,21,22}G_{i},i\in\{16,17,18,19,21,22\};

  • 3.

    H3H_{3} is the spanning proper subgraph of G20G_{20};

  • 4.

    H4H_{4} is the spanning proper subgraph of Gj,j{23,24}G_{j},j\in\{23,24\}.

Therefore, it follows from Theorem 3.2 that

ρ(G1m2)=ρ(H1m2)<ρ(G9m2),\displaystyle\rho(G_{1}\circ\textbf{m}_{2}^{\prime})=\rho(H_{1}\circ\textbf{m}_{2})<\rho(G_{9}\circ\textbf{m}_{2}),
ρ(G10m3)=ρ(H2m3)<ρ(Gim3),i=16,17,18,19,21,22,\displaystyle\rho(G_{10}\circ\textbf{m}_{3}^{\prime})=\rho(H_{2}\circ\textbf{m}_{3})<\rho(G_{i}\circ\textbf{m}_{3}),i=16,17,18,19,21,22,
ρ(G10m3′′)=ρ(H3m3)<ρ(G20m3),\displaystyle\rho(G_{10}\circ\textbf{m}_{3}^{\prime\prime})=\rho(H_{3}\circ\textbf{m}_{3})<\rho(G_{20}\circ\textbf{m}_{3}),
ρ(G10m4)=ρ(H4m4)<ρ(Gjm4),j=23,24,\displaystyle\rho(G_{10}\circ\textbf{m}_{4}^{\prime})=\rho(H_{4}\circ\textbf{m}_{4})<\rho(G_{j}\circ\textbf{m}_{4}),j=23,24,

where m2=(n1,n2,n3,n4,n5+n6),\textbf{m}_{2}^{\prime}=(n_{1},n_{2},n_{3},n_{4},n_{5}+n_{6}), m3=(n1,n2,n3,n4,n5,n6+n7),\textbf{m}_{3}^{\prime}=(n_{1},n_{2},n_{3},n_{4},n_{5},n_{6}+n_{7}), m3′′=(n1,n2,n3,n4,n5+n7,n6)\textbf{m}_{3}^{\prime\prime}=(n_{1},n_{2},n_{3},n_{4},n_{5}+n_{7},n_{6}) and m4=(n1,n2,n3,n4,n5,n6+n7+n8).\textbf{m}_{4}^{\prime}=(n_{1},n_{2},n_{3},n_{4},n_{5},n_{6}+n_{7}+n_{8}).

Hence, GMn(G1,G7,G10)G\in M_{n}(G_{1},G_{7},G_{10}).

Refer to caption
Figure 2: The graphs Hi,i=1,2,3,4H_{i},i=1,2,3,4.

3.2 Step 2

In this subsection we characterize the extremal graphs with minimum spectral radii in Mn(G1),Mn(G7)M_{n}(G_{1}),M_{n}(G_{7}) and Mn(G10)M_{n}(G_{10}), respectively. To accomplish this, let’s introduce some classic results in spectral graph theory.

Definition 3.5.

[10] Let AA be an n×nn\times n real matrix whose rows and columns are indexed by X={1,2,,n}X=\{1,2,\dots,n\}. We partition XX into X1,X2,,Xk{X_{1},X_{2},\dots,X_{k}} in order and rewrite AA according to the partition X1,X2,,Xk{X_{1},X_{2},\dots,X_{k}} as follows:

A=(A1,1A1,kAk,1Ak,k),\displaystyle A=\left(\begin{array}[]{ccc}A_{1,1}&\cdots&A_{1,k}\\ \vdots&\ddots&\vdots\\ A_{k,1}&\cdots&A_{k,k}\\ \end{array}\right), (9)

where Ai,jA_{i,j} is the block of AA formed by rows in XiX_{i} and the columns in XjX_{j}. Let bi,jb_{i,j} denote the average row sum of Ai,jA_{i,j}. Then the matrix B=[bi,j]B=[b_{i,j}] will be called the quotient matrix of the partition of AA. In particular, when the row sum of each block Ai,jA_{i,j} is constant, the partition is called an equitable partition.

Theorem 3.6.

[10] Let A0A\geq 0 be an irreducible square matrix, BB be the quotient matrix of an equitable partition of AA. Then the spectrum of AA contains the spectrum of BB and ρ(A)=ρ(B).\rho(A)=\rho(B).

Theorem 3.7.

[11] Let GG and HH be two connected graphs such that ϕ(H,x)>ϕ(G,x)\phi(H,x)>\phi(G,x) for xρ(G)x\geq\rho(G). Then ρ(H)<ρ(G)\rho(H)<\rho(G).

Theorem 3.8.

[9] Let Kn1,n2,,nkK_{n_{1},n_{2},\dots,n_{k}} be the complete multipartite graph of order nn. Then

ϕ(Kn1,n2,,nk,x)=xnk(1i=1knix+ni)i=1k(x+ni).\phi(K_{n_{1},n_{2},\dots,n_{k}},x)=x^{n-k}(1-\sum_{i=1}^{k}\frac{n_{i}}{x+n_{i}})\prod_{i=1}^{k}(x+n_{i}).

The following Propositions 3.9, 3.10 and 3.11 give the extremal graph which attains the minimum spectral radius in Mn(G1)M_{n}(G_{1}), Mn(G10)M_{n}(G_{10}) and Mn(G7)M_{n}(G_{7}), respectively.

Proposition 3.9.

The extremal graph in Mn(G1)M_{n}(G_{1}) which attains minimum spectral radius is of the form

G1(1,1,1,k,nk3),G_{1}\circ(1,1,1,k,n-k-3),

where 1kn32.1\leq k\leq\frac{n-3}{2}.

Proof.

Since N(v5)={v3}N(v1)N(v_{5})=\{v_{3}\}\subsetneq N(v_{1}) and v1v5E(G1)v_{1}v_{5}\notin E(G_{1}), then by Theorem 3.3 we have

ρ(G1(1,n2,n3,n4,n5+n11))ρ(G1(n1,n2,n3,n4,n5)),\rho(G_{1}\circ(1,n_{2},n_{3},n_{4},n_{5}+n_{1}-1))\leq\rho(G_{1}\circ(n_{1},n_{2},n_{3},n_{4},n_{5})),

with equality if and only if n1=1n_{1}=1. It follows that the extremal graph in Mn(G1)M_{n}(G_{1}) which attains minimum spectral radius is of the form F=G1(1,n2,n3,n4,n5)F=G_{1}\circ(1,n_{2},n_{3},n_{4},n_{5}).

Then V(F)V(F) can be naturally partitioned into 55 parts:

{V1,V2,V3,V4,V5},\{V_{1},V_{2},V_{3},V_{4},V_{5}\},

where Vi={vi1,,vini},i=1,2,3,4,5V_{i}=\{v_{i}^{1},\dots,v_{i}^{n_{i}}\},i=1,2,3,4,5. Obviously, this partition of A(F)A(F) is equitable and the corresponding quotient matrix B is

B=(0n2n3n40100n401000n51n200000n300).\displaystyle B=\left(\begin{array}[]{ccccc}0&n_{2}&n_{3}&n_{4}&0\\ 1&0&0&n_{4}&0\\ 1&0&0&0&n_{5}\\ 1&n_{2}&0&0&0\\ 0&0&n_{3}&0&0\\ \end{array}\right). (15)

Then the characteristic polynomial of the quotient matrix BB is:

ϕ(B,x)=x5(n2+n3+n4+n2n4+n3n5)x32n2n4x2+(n2n3n4+n2n3n5+n3n4n5+n2n3n4n5)x+2n2n3n4n5.\begin{split}\phi(B,x)&=x^{5}-(n_{2}+n_{3}+n_{4}+n_{2}n_{4}+n_{3}n_{5})x^{3}-2n_{2}n_{4}x^{2}+\\ &(n_{2}n_{3}n_{4}+n_{2}n_{3}n_{5}+n_{3}n_{4}n_{5}+n_{2}n_{3}n_{4}n_{5})x+2n_{2}n_{3}n_{4}n_{5}.\end{split}

Since R(A(F))=5R(A(F))=5, by Theorem 3.6 we have ϕ(F,x)=xn5ϕ(B,x)\phi(F,x)=x^{n-5}\phi(B,x) and ρ(F)=ρ(A(F))=ρ(B)\rho(F)=\rho(A(F))=\rho(B).

Note that G1(1,n2,n3,n4,n5)G1(1,n4,n3,n2,n5)G_{1}\circ(1,n_{2},n_{3},n_{4},n_{5})\cong G_{1}\circ(1,n_{4},n_{3},n_{2},n_{5}). Therefore, without loss of generality, we suppose that n4n2.n_{4}\geq n_{2}.

Claim 1. n2=1.n_{2}=1.

Assume n22n_{2}\geq 2, let F1=G1(1,n21,n3,n4+1,n5)F_{1}=G_{1}\circ(1,n_{2}-1,n_{3},n_{4}+1,n_{5}) then

r(x)=ϕ(F1,x)ϕ(F,x)=xn5(n4n2+1)(x3+2x2(n3+n3n5)x2n3n5)=xn5(n4n2+1)(x(x2n3(n5+1))+2(x2n3n5)).\begin{split}r(x)&=\phi(F_{1},x)-\phi(F,x)\\ &=x^{n-5}(n_{4}-n_{2}+1)(x^{3}+2x^{2}-(n_{3}+n_{3}n_{5})x-2n_{3}n_{5})\\ &=x^{n-5}(n_{4}-n_{2}+1)\left(x(x^{2}-n_{3}(n_{5}+1))+2(x^{2}-n_{3}n_{5})\right).\end{split}

Since n4n2n_{4}\geq n_{2}, we have n4n2+1>0n_{4}-n_{2}+1>0. It is clear that Kn3,n5+1K_{n_{3},n_{5}+1} is a proper subgraph of FF, we obtain ρ(F)>ρ(Kn3,n5+1)=n3(n5+1)\rho(F)>\rho(K_{n_{3},n_{5}+1})=\sqrt{n_{3}(n_{5}+1)}, then r(x)>0r(x)>0 for xρ(F)x\geq\rho(F).

Thus, by Theorem 3.7, we have ρ(F1)<ρ(F)\rho(F_{1})<\rho(F) which contradicts to the extremality of FF.

Claim 2. n3=1.n_{3}=1.

Now F=G1(1,1,n3,n4,n5)F=G_{1}\circ(1,1,n_{3},n_{4},n_{5}), we claim that n5n3n_{5}\geq n_{3}. If not, let F2=G1(1,1,n5,n4,n3)F_{2}=G_{1}\circ(1,1,n_{5},n_{4},n_{3}), then

r(x)=ϕ(F2,x)ϕ(F,x)=xn4(x2n4)(n3n5).\begin{split}r(x)&=\phi(F_{2},x)-\phi(F,x)=x^{n-4}(x^{2}-n_{4})(n_{3}-n_{5}).\end{split}

Since n3>n5n_{3}>n_{5}, we have n3n5>0n_{3}-n_{5}>0. It can be seen that Kn4,2K_{n_{4},2} is a proper subgraph of FF, we obtain ρ(F)>ρ(Kn4,2)=2n4\rho(F)>\rho(K_{n_{4},2})=\sqrt{2n_{4}}, then r(x)>0r(x)>0 for xρ(F)x\geq\rho(F).

Thus, by Theorem 3.7, we have ρ(F2)<ρ(F)\rho(F_{2})<\rho(F), a contradiction. Therefore n5n3n_{5}\geq n_{3}.

Next, we assume n32n_{3}\geq 2, let F3=G1(1,1,n31,n4,n5+1)F_{3}=G_{1}\circ(1,1,n_{3}-1,n_{4},n_{5}+1) then

r(x)=ϕ(F3,x)ϕ(F,x)=xn5((n5n3+1)(x3(2n4+1)x2n4)+x(x2n4)).\begin{split}r(x)&=\phi(F_{3},x)-\phi(F,x)\\ &=x^{n-5}\left((n_{5}-n_{3}+1)(x^{3}-(2n_{4}+1)x-2n_{4})+x(x^{2}-n_{4})\right).\end{split}

Since n5n3n_{5}\geq n_{3}, we have n5n3+1>0n_{5}-n_{3}+1>0. It is clear that Kn4,1,1K_{n_{4},1,1} is a proper subgraph of FF, by Theorem 3.8, we obtain ρ(F)>ρ(Kn4,1,1)=(8n4+1+1)/2\rho(F)>\rho(K_{n_{4},1,1})=(\sqrt{8n_{4}+1}+1)/2, then r(x)>0r(x)>0 for xρ(F)x\geq\rho(F).

Thus, by Theorem 3.7, we have ρ(F3)<ρ(F)\rho(F_{3})<\rho(F), which contradicts to the extremality of FF.

Claim 3. n5n4.n_{5}\geq n_{4}.

Now F=G1(1,1,1,n4,n5)F=G_{1}\circ(1,1,1,n_{4},n_{5}). Otherwise, let F4=G1(1,1,1,n5,n4)F_{4}=G_{1}\circ(1,1,1,n_{5},n_{4}) then

r(x)=ϕ(F4,x)ϕ(F,x)=xn3(x+2)(n4n5).\begin{split}r(x)&=\phi(F_{4},x)-\phi(F,x)=x^{n-3}(x+2)(n_{4}-n_{5}).\end{split}

Since n4>n5n_{4}>n_{5} and ρ(F)>0\rho(F)>0, then r(x)>0r(x)>0 for xρ(F)x\geq\rho(F). By Theorem 3.7, we have ρ(F4)<ρ(F)\rho(F_{4})<\rho(F) which contradicts to the extremality of FF, thus n5n4n_{5}\geq n_{4}.

From above three claims, we conclude that the extremal graph with minimum spectral radius in Mn(G1)M_{n}(G_{1}) is of the form G1(1,1,1,k,nk3)G_{1}\circ(1,1,1,k,n-k-3), where 1k(n3)/21\leq k\leq(n-3)/2. ∎

Similarly, we characterize the extremal graph with minimum spectral radius in Mn(G10)M_{n}(G_{10}).

Proposition 3.10.

The extremal graph in Mn(G10)M_{n}(G_{10}) which attains minimum spectral radius is of the form

G10(1,1,1,1,k,nk4),G_{10}\circ(1,1,1,1,k,n-k-4),

where 1kn42.1\leq k\leq\frac{n-4}{2}.

Proof.

By Theorem 3.3, we have

ρ(G10(1,n2,n3,n4,n5,n6+n11))ρ(G10(n1,n2,n3,n4,n5,n6)),\rho(G_{10}\circ(1,n_{2},n_{3},n_{4},n_{5},n_{6}+n_{1}-1))\leq\rho(G_{10}\circ(n_{1},n_{2},n_{3},n_{4},n_{5},n_{6})),

with equality if and only if n1=1n_{1}=1. Thus, we may suppose that the extremal graph in Mn(G10)M_{n}(G_{10}) which attains minimum spectral radius is of the form F=G10(1,n2,n3,n4,n5,n6)F=G_{10}\circ(1,n_{2},n_{3},n_{4},n_{5},n_{6}).

Similarly, we obtain

B=(0n2n3n40010n30n501n200n5010000n60n2n3000000n400),\displaystyle B=\left(\begin{array}[]{cccccc}0&n_{2}&n_{3}&n_{4}&0&0\\ 1&0&n_{3}&0&n_{5}&0\\ 1&n_{2}&0&0&n_{5}&0\\ 1&0&0&0&0&n_{6}\\ 0&n_{2}&n_{3}&0&0&0\\ 0&0&0&n_{4}&0&0\\ \end{array}\right), (22)

is the quotient matrix of an equitable partition of A(F)A(F). The characteristic polynomial of the quotient matrix BB is:

ϕ(B,x)=x(x5(n2+n3+n4+n2n3+n2n5+n3n5+n4n6)x3(2n2n3+2n2n3n5)x2+(n2n3n4+n2n4n5+n2n4n6+n3n4n5+n3n4n6+n2n3n4n6+n2n4n5n6+n3n4n5n6)x+2n2n3n4n5+2n2n3n4n6+2n2n3n4n5n6).\begin{split}\phi(B,x)&=x(x^{5}-(n_{2}+n_{3}+n_{4}+n_{2}n_{3}+n_{2}n_{5}+n_{3}n_{5}+n_{4}n_{6})x^{3}\\ &-(2n_{2}n_{3}+2n_{2}n_{3}n_{5})x^{2}+(n_{2}n_{3}n_{4}+n_{2}n_{4}n_{5}+n_{2}n_{4}n_{6}\\ &+n_{3}n_{4}n_{5}+n_{3}n_{4}n_{6}+n_{2}n_{3}n_{4}n_{6}+n_{2}n_{4}n_{5}n_{6}+n_{3}n_{4}n_{5}n_{6})x\\ &+2n_{2}n_{3}n_{4}n_{5}+2n_{2}n_{3}n_{4}n_{6}+2n_{2}n_{3}n_{4}n_{5}n_{6}).\end{split}

Since R(A(F))=5R(A(F))=5, by Theorem 3.6 we have ϕ(F,x)=xn6ϕ(B,x)\phi(F,x)=x^{n-6}\phi(B,x) and ρ(F)=ρ(A(F))=ρ(B)\rho(F)=\rho(A(F))=\rho(B).

Note that G10(1,n2,n3,n4,n5,n6)G10(1,n3,n2,n4,n5,n6)G_{10}\circ(1,n_{2},n_{3},n_{4},n_{5},n_{6})\cong G_{10}\circ(1,n_{3},n_{2},n_{4},n_{5},n_{6}). Therefore, without loss of generality, we suppose that n3n2.n_{3}\geq n_{2}.

Claim 1. n2=1.n_{2}=1.

Assume n22n_{2}\geq 2, let F1=G10(1,n21,n3+1,n4,n5,n6)F_{1}=G_{10}\circ(1,n_{2}-1,n_{3}+1,n_{4},n_{5},n_{6}) then

r(x)=ϕ(F1,x)ϕ(F,x)=xn5(n3n2+1)(x3+2(1+n5)x2(n4+n4n6)x2n4n52n4n62n4n5n6)=xn5(n3n2+1)(x(x2n4(n6+1))+2n5(x2n4(n6+1))+2(x2n4n6)).\begin{split}r(x)&=\phi(F_{1},x)-\phi(F,x)\\ &=x^{n-5}(n_{3}-n_{2}+1)(x^{3}+2(1+n_{5})x^{2}-(n_{4}+n_{4}n_{6})x-2n_{4}n_{5}\\ &-2n_{4}n_{6}-2n_{4}n_{5}n_{6})\\ &=x^{n-5}(n_{3}-n_{2}+1)(x(x^{2}-n_{4}(n_{6}+1))+2n_{5}(x^{2}-n_{4}(n_{6}+1))\\ &+2(x^{2}-n_{4}n_{6})).\end{split}

Since n3n2n_{3}\geq n_{2}, we have n3n2+1>0n_{3}-n_{2}+1>0. It is clear that Kn4,n6+1K_{n_{4},n_{6}+1} is a proper subgraph of FF, we obtain ρ(F)>ρ(Kn4,n6+1)=n4(n6+1)\rho(F)>\rho(K_{n_{4},n_{6}+1})=\sqrt{n_{4}(n_{6}+1)}, then r(x)>0r(x)>0 for xρ(F)x\geq\rho(F).

Thus, by Theorem 3.7, we have ρ(F1)<ρ(F)\rho(F_{1})<\rho(F) which contradicts to the extremality of FF.

Claim 2. n3=1.n_{3}=1.

Now F=G10(1,1,n3,n4,n5,n6)F=G_{10}\circ(1,1,n_{3},n_{4},n_{5},n_{6}), we claim that n5n3n_{5}\geq n_{3}. If not, let F2=G10(1,1,n5,n4,n3,n6)F_{2}=G_{10}\circ(1,1,n_{5},n_{4},n_{3},n_{6}), then

r(x)=ϕ(F2,x)ϕ(F,x)=xn5(n3n5)(x2n4n6)(x+2).\begin{split}r(x)&=\phi(F_{2},x)-\phi(F,x)=x^{n-5}(n_{3}-n_{5})(x^{2}-n_{4}n_{6})(x+2).\end{split}

Since n3>n5n_{3}>n_{5}, we have n3n5>0n_{3}-n_{5}>0. It can be seen that Kn4,n6K_{n_{4},n_{6}} is a proper subgraph of FF, we obtain ρ(F)>ρ(Kn4,n6)=n4n6\rho(F)>\rho(K_{n_{4},n_{6}})=\sqrt{n_{4}n_{6}}, then r(x)>0r(x)>0 for xρ(F)x\geq\rho(F).

Thus, by Theorem 3.7, we have ρ(F2)<ρ(F)\rho(F_{2})<\rho(F), a contradiction. Therefore n5n3n_{5}\geq n_{3}.

Next, we assume n32n_{3}\geq 2, let F3=G10(1,1,n31,n4,n5+1,n6)F_{3}=G_{10}\circ(1,1,n_{3}-1,n_{4},n_{5}+1,n_{6}) then

r(x)=ϕ(F3,x)ϕ(F,x)=xn5(x+2)((n5n3+2)(x2n4(n6+1))+n4).\begin{split}r(x)&=\phi(F_{3},x)-\phi(F,x)\\ &=x^{n-5}(x+2)\left((n_{5}-n_{3}+2)(x^{2}-n_{4}(n_{6}+1))+n_{4}\right).\end{split}

Since n5n3n_{5}\geq n_{3}, we have n5n3+2>0n_{5}-n_{3}+2>0. It is clear that Kn4,n6+1K_{n_{4},n_{6}+1} is a proper subgraph of FF, we obtain ρ(F)>ρ(Kn4,n6+1)=n4(n6+1)\rho(F)>\rho(K_{n_{4},n_{6}+1})=\sqrt{n_{4}(n_{6}+1)}, then r(x)>0r(x)>0 for xρ(F)x\geq\rho(F).

Thus, by Theorem 3.7, we have ρ(F3)<ρ(F)\rho(F_{3})<\rho(F) which contradicts to the extremality of FF.

Claim 3. n4=1.n_{4}=1.

Now F=G10(1,1,1,n4,n5,n6)F=G_{10}\circ(1,1,1,n_{4},n_{5},n_{6}), we claim that n6n4n_{6}\geq n_{4}. If not, let F4=G10(1,1,1,n6,n5,n4)F_{4}=G_{10}\circ(1,1,1,n_{6},n_{5},n_{4}), then

r(x)=ϕ(F4,x)ϕ(F,x)=xn5(n4n6)(x2x2n5)(x+1).\begin{split}r(x)&=\phi(F_{4},x)-\phi(F,x)=x^{n-5}(n_{4}-n_{6})(x^{2}-x-2n_{5})(x+1).\end{split}

Since n4>n6n_{4}>n_{6}, we have n4n6>0n_{4}-n_{6}>0. It can be seen that Kn5,1,1K_{n_{5},1,1} is a proper subgraph of FF, we obtain ρ(F)>ρ(Kn5,1,1)=(8n5+1+1)/2\rho(F)>\rho(K_{n_{5},1,1})=(\sqrt{8n_{5}+1}+1)/2, then r(x)>0r(x)>0 for xρ(F)x\geq\rho(F).

Thus, by Theorem 3.7, we have ρ(F4)<ρ(F)\rho(F_{4})<\rho(F), a contradiction. Therefore n6n4n_{6}\geq n_{4}.

Next, we assume n42n_{4}\geq 2, let F5=G10(1,1,1,n41,n5,n6+1)F_{5}=G_{10}\circ(1,1,1,n_{4}-1,n_{5},n_{6}+1) then

r(x)=ϕ(F5,x)ϕ(F,x)=xn5(x+1)((n6n4+2)(x2x2n52)+2).\begin{split}r(x)&=\phi(F_{5},x)-\phi(F,x)\\ &=x^{n-5}(x+1)\left((n_{6}-n_{4}+2)(x^{2}-x-2n_{5}-2)+2\right).\end{split}

Since n6n4n_{6}\geq n_{4}, we have n6n4+2>0n_{6}-n_{4}+2>0. It is clear that H(n5,1,1,1)H\circ(n_{5},1,1,1) is a proper subgraph of FF, where HH is shown in Figure 3, we obtain ρ(F)>ρ(H(n5,1,1,1))=(8n5+9+1)/2\rho(F)>\rho(H\circ(n_{5},1,1,1))=(\sqrt{8n_{5}+9}+1)/2, then r(x)>0r(x)>0 for xρ(F)x\geq\rho(F).

Thus, by Theorem 3.7, we have ρ(F5)<ρ(F)\rho(F_{5})<\rho(F), which contradicts to the extremality of FF.

Refer to caption
Figure 3: The graph HH.

Claim 4. n6n5.n_{6}\geq n_{5}.

Now F=G10(1,1,1,1,n5,n6)F=G_{10}\circ(1,1,1,1,n_{5},n_{6}). Otherwise, let F6=G10(1,1,1,1,n6,n5)F_{6}=G_{10}\circ(1,1,1,1,n_{6},n_{5}) then

r(x)=ϕ(F6,x)ϕ(F,x)=xn4(x+1)2(n5n6).\begin{split}r(x)&=\phi(F_{6},x)-\phi(F,x)=x^{n-4}(x+1)^{2}(n_{5}-n_{6}).\end{split}

Since n5>n6n_{5}>n_{6} and ρ(F)>0\rho(F)>0, then r(x)>0r(x)>0 for xρ(F)x\geq\rho(F), by Theorem 3.7, we have ρ(F6)<ρ(F)\rho(F_{6})<\rho(F) which contradicts to the extremality of FF, thus n6n5n_{6}\geq n_{5}.

It follows from above four claims that the extremal graph with minimum spectral radius in Mn(G10)M_{n}(G_{10}) is of the form G10(1,1,1,1,k,nk4),G_{10}\circ(1,1,1,1,k,n-k-4), where 1k(n4)/21\leq k\leq(n-4)/2. ∎

Next we determine the extremal graph with minimum spectral radius in Mn(G7)M_{n}(G_{7}).

Proposition 3.11.

The extremal graph in Mn(G7)M_{n}(G_{7}) which attains minimum spectral radius is

G7(n32,1,n32,1,1).G_{7}\circ(\lceil\frac{n-3}{2}\rceil,1,\lfloor\frac{n-3}{2}\rfloor,1,1).
Proof.

Suppose that the extremal graph in Mn(G7)M_{n}(G_{7}) which attains minimum spectral radius is of the form F=G7(n1,n2,n3,n4,n5)F=G_{7}\circ(n_{1},n_{2},n_{3},n_{4},n_{5}). Similarlly, we obtain

B=(0n200n5n10n3000n20n4000n30n5n100n40),\displaystyle B=\left(\begin{array}[]{ccccc}0&n_{2}&0&0&n_{5}\\ n_{1}&0&n_{3}&0&0\\ 0&n_{2}&0&n_{4}&0\\ 0&0&n_{3}&0&n_{5}\\ n_{1}&0&0&n_{4}&0\\ \end{array}\right), (28)

is the quotient matrix of an equitable partition of A(F)A(F) and the characteristic polynomial of BB is:

ϕ(B,x)\displaystyle\phi(B,x) =x5(n1n2+n2n3+n1n5+n3n4+n4n5)x3+(n1n2n3n4+\displaystyle=x^{5}-(n_{1}n_{2}+n_{2}n_{3}+n_{1}n_{5}+n_{3}n_{4}+n_{4}n_{5})x^{3}+(n_{1}n_{2}n_{3}n_{4}+
n1n2n3n5+n1n2n4n5+n1n3n4n5+n2n3n4n5)x2n1n2n3n4n5.\displaystyle n_{1}n_{2}n_{3}n_{5}+n_{1}n_{2}n_{4}n_{5}+n_{1}n_{3}n_{4}n_{5}+n_{2}n_{3}n_{4}n_{5})x-2n_{1}n_{2}n_{3}n_{4}n_{5}.

Since R(A(F))=5R(A(F))=5, by Theorem 3.6 we have ϕ(F,x)=xn5ϕ(B,x)\phi(F,x)=x^{n-5}\phi(B,x) and ρ(F)=ρ(A(F))=ρ(B)\rho(F)=\rho(A(F))=\rho(B).

Without loss of generality, we may suppose that n1=max{ni,i=1,2,3,4,5}n_{1}=\text{max}\{n_{i},i=1,2,3,4,5\} and n2n5n_{2}\leq n_{5}, then we have the following claims.

Claim 1. n2n3n_{2}\leq n_{3} and n5n4n_{5}\leq n_{4}.

Suppose that n2>n3n_{2}>n_{3}. Let F1=G7(n1,n3,n2,n4,n5)F_{1}=G_{7}\circ(n_{1},n_{3},n_{2},n_{4},n_{5}), then

r(x)=ϕ(F1,x)ϕ(F,x)=xn2(n1n4)(n2n3).r(x)=\phi(F_{1},x)-\phi(F,x)=x^{n-2}(n_{1}-n_{4})(n_{2}-n_{3}).

Since n1=max{ni,i=1,2,3,4,5}n_{1}=\text{max}\{n_{i},i=1,2,3,4,5\}, we have n1n4n_{1}\geq n_{4}. And if n1=n4n_{1}=n_{4}, then F1FF_{1}\cong F. Thus, without loss of generality, we may suppose that n1>n4n_{1}>n_{4}.

Since n2>n3n_{2}>n_{3}, n1>n4n_{1}>n_{4} and ρ(F)>0\rho(F)>0, then r(x)>0r(x)>0 for xρ(F)x\geq\rho(F). By Theorem 3.7, we have ρ(F1)<ρ(F)\rho(F_{1})<\rho(F) which contradicts to the extremality of FF.

Similarly, we obtain n5n4n_{5}\leq n_{4}.

Claim 2. n4n3n_{4}\leq n_{3}.

Suppose to the contrary that n4>n3n_{4}>n_{3}. Let F2=G7(n1,n2,n4,n3,n5)F_{2}=G_{7}\circ(n_{1},n_{2},n_{4},n_{3},n_{5}), then

r(x)=ϕ(F2,x)ϕ(F,x)=xn2(n2n5)(n3n4).r(x)=\phi(F_{2},x)-\phi(F,x)=x^{n-2}(n_{2}-n_{5})(n_{3}-n_{4}).

Since n5n2n_{5}\geq n_{2} and if n5=n2n_{5}=n_{2}, then F2FF_{2}\cong F. Thus without loss of generality we may suppose that n5>n2n_{5}>n_{2}.

Since n4>n3n_{4}>n_{3}, n5>n2n_{5}>n_{2} and ρ(F)>0\rho(F)>0. Then r(x)>0r(x)>0 for xρ(F)x\geq\rho(F). By Theorem 3.7, we have ρ(F2)<ρ(F)\rho(F_{2})<\rho(F) which contradicts to the extremality of FF.

From above two claims, we have n1n3n4n5n2n_{1}\geq n_{3}\geq n_{4}\geq n_{5}\geq n_{2}. Next, we will prove n2=n4=n5=1n_{2}=n_{4}=n_{5}=1 and n1n31n_{1}-n_{3}\leq 1.

Claim 3. n2=n5n_{2}=n_{5}

Assume n2<n5n_{2}<n_{5}, let F3=G7(n1+n5n2,n2,n3,n4,n2)F_{3}=G_{7}\circ(n_{1}+n_{5}-n_{2},n_{2},n_{3},n_{4},n_{2}) then

r(x)\displaystyle r(x) =ϕ(F3,x)ϕ(F,x)\displaystyle=\phi(F_{3},x)-\phi(F,x)
=xn5(n5n2)((n12n2+n4)x3(n1n2)((n3n4+n2n3+n2n4)x\displaystyle=x^{n-5}(n_{5}-n_{2})((n_{1}-2n_{2}+n_{4})x^{3}-(n_{1}-n_{2})((n_{3}n_{4}+n_{2}n_{3}+n_{2}n_{4})x
2n2n3n4))\displaystyle-2n_{2}n_{3}n_{4}))
xn5(n5n2)(n1n2)(x3(n3n4+n2n3+n2n4)x+2n2n3n4)\displaystyle\geq x^{n-5}(n_{5}-n_{2})(n_{1}-n_{2})\left(x^{3}-(n_{3}n_{4}+n_{2}n_{3}+n_{2}n_{4})x+2n_{2}n_{3}n_{4}\right)
=xn5(n5n2)(n1n2)g(x).\displaystyle=x^{n-5}(n_{5}-n_{2})(n_{1}-n_{2})g(x).

Since n1n3n4n5>n21n_{1}\geq n_{3}\geq n_{4}\geq n_{5}>n_{2}\geq 1, we have n1n2>0n_{1}-n_{2}>0 and n5n2>0n_{5}-n_{2}>0. It is clear that Kn3,n2+n4K_{n_{3},n_{2}+n_{4}} is a proper subgraph of FF, we obtain ρ(F)>ρ(Kn3,n2+n4)=n3(n2+n4)\rho(F)>\rho(K_{n_{3},n_{2}+n_{4}})=\sqrt{n_{3}(n_{2}+n_{4})}.

Since g(n3(n2+n4))>0g(\sqrt{n_{3}(n_{2}+n_{4})})>0 and n3(n2+n4)>(n3(n2+n4)+n2n4)/3\sqrt{n_{3}(n_{2}+n_{4})}>\sqrt{(n_{3}(n_{2}+n_{4})+n_{2}n_{4})/3}, where (n3(n2+n4)+n2n4)/3\sqrt{\left(n_{3}(n_{2}+n_{4})+n_{2}n_{4}\right)/3} is the largest root of g(x)g^{\prime}(x), we have r(x)>0r(x)>0 for xρ(F)x\geq\rho(F).

Thus, by Theorem 3.7, we have ρ(F3)<ρ(F)\rho(F_{3})<\rho(F) which contradicts to the extremality of FF.

Note that G7(n1,n2,n3,n4,n5)G7(n1,n2,n4,n3,n5)G_{7}\circ(n_{1},n_{2},n_{3},n_{4},n_{5})\cong G_{7}\circ(n_{1},n_{2},n_{4},n_{3},n_{5}) when n2=n5n_{2}=n_{5}, therefore without loss of generality we may suppose that n3n4.n_{3}\geq n_{4}.

Claim 4. n4=1.n_{4}=1.

Assume n42n_{4}\geq 2, let F4=G7(n1,n2,n3+1,n41,n5)F_{4}=G_{7}\circ(n_{1},n_{2},n_{3}+1,n_{4}-1,n_{5}) then

r(x)=ϕ(F4,x)ϕ(F,x)=xn5(xn5)(n3n4+1)(x2+n5x2n1n5).\begin{split}r(x)&=\phi(F_{4},x)-\phi(F,x)\\ &=x^{n-5}(x-n_{5})(n_{3}-n_{4}+1)(x^{2}+n_{5}x-2n_{1}n_{5}).\end{split}

Since n1n3n4n5=n2n_{1}\geq n_{3}\geq n_{4}\geq n_{5}=n_{2}, we have n3n4+1>0n_{3}-n_{4}+1>0. It can be seen that Kn1,2n5K_{n_{1},2n_{5}} is a proper subgraph of FF, we obtain ρ(F)>ρ(Kn1,2n5)=2n1n5>n5\rho(F)>\rho(K_{n_{1},2n_{5}})=\sqrt{2n_{1}n_{5}}>n_{5}, then r(x)>0r(x)>0 for xρ(F)x\geq\rho(F).

Thus, by Theorem 3.7, we have ρ(F4)<ρ(F)\rho(F_{4})<\rho(F) which contradicts to the extremality of FF, therefore n4=1n_{4}=1 and hence n2=n5=1n_{2}=n_{5}=1.

Claim 5. n1n31.n_{1}-n_{3}\leq 1.

Now F=G7(n1,1,n3,1,1)F=G_{7}\circ(n_{1},1,n_{3},1,1). Assume n1n3+2n_{1}\geq n_{3}+2, let F5=G7(n11,1,n3+1,1,1)F_{5}=G_{7}\circ(n_{1}-1,1,n_{3}+1,1,1) then

r(x)=ϕ(F5,x)ϕ(F,x)=xn5(3x2)(n1n31).\begin{split}r(x)&=\phi(F_{5},x)-\phi(F,x)=x^{n-5}(3x-2)(n_{1}-n_{3}-1).\end{split}

Since n1n3+2n_{1}\geq n_{3}+2, we have n1n31>0n_{1}-n_{3}-1>0. It is clear that Kn1,2K_{n_{1},2} is a proper subgraph of FF, we obtain ρ(F)>ρ(Kn1,2)=2n1>1\rho(F)>\rho(K_{n_{1},2})=\sqrt{2n_{1}}>1, then r(x)>0r(x)>0 for xρ(F)x\geq\rho(F).

Thus, by Theorem 3.7, we have ρ(F5)<ρ(F)\rho(F_{5})<\rho(F) which contradicts to the extremality of FF, therefore n1n31n_{1}-n_{3}\leq 1 and hence n1=(n3)/2,n3=(n3)/2.n_{1}=\lceil(n-3)/2\rceil,n_{3}=\lfloor(n-3)/2\rfloor.

From above five claims, we obtain G7(n32,1,n32,1,1)G_{7}\circ(\lceil\frac{n-3}{2}\rceil,1,\lfloor\frac{n-3}{2}\rfloor,1,1) attains the minimum spectral radius in Mn(G7)M_{n}(G_{7}). ∎

3.3 Step 3

We first prove that the extremal graph with minimum spectral radius in Mn(G1,G7)M_{n}(G_{1},G_{7}) must be in Mn(G1)M_{n}(G_{1}) by the following lemma.

Lemma 3.12.

For n8n\geq 8, we have ρ(G1(1,1,1,n32,n32))<ρ(G7(n32,1,n32,1,1))\rho(G_{1}\circ(1,1,1,\lfloor\frac{n-3}{2}\rfloor,\lceil\frac{n-3}{2}\rceil))<\rho(G_{7}\circ(\lceil\frac{n-3}{2}\rceil,1,\lfloor\frac{n-3}{2}\rfloor,1,1)).

Proof.

Let F1=G1(1,1,1,n32,n32)F_{1}=G_{1}\circ(1,1,1,\lfloor\frac{n-3}{2}\rfloor,\lceil\frac{n-3}{2}\rceil) and F2=G7(n32,1,n32,1,1)F_{2}=G_{7}\circ(\lceil\frac{n-3}{2}\rceil,1,\lfloor\frac{n-3}{2}\rfloor,1,1).

For 8n128\leq n\leq 12, we use the MATLAB software to calculate the spectral radii of FiF_{i} for i=1,2i=1,2, as shown in the Table 1.

Table 1: ρ(Fi)\rho(F_{i}).
          nn           ρ(F1)\rho(F_{1})           ρ(F2)\rho(F_{2})
8 2.7676 2.9764
9 3.1474 3.2176
10 3.1713 3.4630
11 3.5047 3.6737
12 3.5223 3.8879

So let us assume that n13n\geq 13.

Case 1. n3=2kn-3=2k is even.

In this case, F1=G1(1,1,1,k,k)F_{1}=G_{1}\circ(1,1,1,k,k) and F2=G7(k,1,k,1,1)F_{2}=G_{7}\circ(k,1,k,1,1), then

r(x)=ϕ(F1,x)ϕ(F2,x)=xn5((k1)x32kx2k2x+4k2).\begin{split}r(x)&=\phi(F_{1},x)-\phi(F_{2},x)=x^{n-5}\left((k-1)x^{3}-2kx^{2}-k^{2}x+4k^{2}\right).\end{split}

It can be seen that K2k,1K_{2k,1} is a proper subgraph of F2F_{2}, we obtain ρ(F2)>ρ(K2k,1)=2k\rho(F_{2})>\rho(K_{2k,1})=\sqrt{2k}. Since n13n\geq 13, we have r(2k)>0r(\sqrt{2k})>0 and 2k>(2k+k3k+1)/3(k1)\sqrt{2k}>(2k+k\sqrt{3k+1})/3(k-1). Since (2k+k3k+1)/3(k1)(2k+k\sqrt{3k+1})/3(k-1) is the largest root of r(x)r^{\prime}(x), we obtain r(x)>0r(x)>0 for xρ(F2)x\geq\rho(F_{2}).

Thus by Theorem 3.7, we have ρ(F1)<ρ(F2)\rho(F_{1})<\rho(F_{2}).

Case 2. n3=2k+1n-3=2k+1 is odd.

In this case, F1=G1(1,1,1,k,k+1)F_{1}=G_{1}\circ(1,1,1,k,k+1) and F2=G7(k+1,1,k,1,1)F_{2}=G_{7}\circ(k+1,1,k,1,1), then

r(x)=ϕ(F1,x)ϕ(F2,x)=xn5k(x32x2(k+1)x+4k+).\begin{split}r(x)&=\phi(F_{1},x)-\phi(F_{2},x)=x^{n-5}k\left(x^{3}-2x^{2}-(k+1)x+4k+\right).\end{split}

It is clear that K2k+1,1K_{2k+1,1} is a proper subgraph of F2F_{2}, we obtain ρ(F2)>ρ(K2k+1,1)=2k+1\rho(F_{2})>\rho(K_{2k+1,1})=\sqrt{2k+1}. Since n13n\geq 13, we have r(2k+1)>0r(\sqrt{2k+1})>0 and 2k+1>(2+3k+7)/3\sqrt{2k+1}>(2+\sqrt{3k+7})/3. Since (2+3k+7)/3(2+\sqrt{3k+7})/3 is the largest root of r(x)r^{\prime}(x), we obtain r(x)>0r(x)>0 for xρ(F2)x\geq\rho(F_{2}).

Thus by Theorem 3.7, we have ρ(F1)<ρ(F2)\rho(F_{1})<\rho(F_{2}).

Next we prove the extremal graph with minimum spectral radius in Mn(G1,G10)M_{n}(G_{1},G_{10}) must be in Mn(G10)M_{n}(G_{10}). We need the following theorem.

Theorem 3.13.

[12] Let GG be a graph with m edges and n vertices. Then ρ(G)2mn+1\rho(G)\leq\sqrt{2m-n+1}, with equality if and only if GG is isomorphic to the star K1,n1K_{1,n-1} or the complete graph KnK_{n}.

Lemma 3.14.

Let G1(1,1,1,k,nk3)G_{1}\circ(1,1,1,k,n-k-3) be the extremal graph with minimum spectral radius in Mn(G1)M_{n}(G_{1}) for n12n\geq 12. Then 2kn322\leq k\leq\frac{n-3}{2}.

Proof.

We denote Fk=G1(1,1,1,k,nk3)F_{k}=G_{1}\circ(1,1,1,k,n-k-3) for convenience. By Proposition 3.9, we have 1k(n3)/21\leq k\leq(n-3)/2.

For 12n1812\leq n\leq 18, we use the MATLAB software to calculate the spectral radii of FkF_{k}, as shown in the Table 2, where the minimum spectral radius is bolded.

Table 2: ρ(Fk)\rho(F_{k}).
nn kk ρ(Fk)\rho(F_{k}) 1 2 3 4 5 6 7
12 3.0751 3.0649 3.2427 3.5223 \ \ \
13 3.2229 3.1791 3.2951 3.5443 3.8231 \ \
14 3.3668 3.3013 3.3616 3.5722 3.8368 \ \
15 3.5064 3.4274 3.4422 3.6076 3.8535 4.1131 \
16 3.6418 3.5544 3.5353 3.6526 3.8742 4.1243 \
17 3.7731 3.6807 3.6377 3.7088 3.8998 4.1376 4.3813
18 3.9006 3.8053 3.7459 3.7767 3.9318 4.1536 4.3906

For n19n\geq 19, note that F1=G1(1,1,1,1,n4),F2=G1(1,1,1,2,n5)F_{1}=G_{1}\circ(1,1,1,1,n-4),F_{2}=G_{1}\circ(1,1,1,2,n-5), let xx be the principal eigenvector of F2F_{2} and xix_{i} correspond to vertices in ViV_{i} for i=1,2,3,4,5.i=1,2,3,4,5. By ρ(F2)𝒙=A(F2)𝒙\rho(F_{2})\bm{x}=A(F_{2})\bm{x}, we have

ρ(F2)x1\displaystyle\rho(F_{2})x_{1} =x2+x3+2x4,\displaystyle=x_{2}+x_{3}+2x_{4}, (29)
ρ(F2)x2\displaystyle\rho(F_{2})x_{2} =x1+2x4,\displaystyle=x_{1}+2x_{4}, (30)
ρ(F2)x3\displaystyle\rho(F_{2})x_{3} =x1+(n5)x5,\displaystyle=x_{1}+(n-5)x_{5}, (31)
ρ(F2)x4\displaystyle\rho(F_{2})x_{4} =x1+x2,\displaystyle=x_{1}+x_{2}, (32)
ρ(F2)x5\displaystyle\rho(F_{2})x_{5} =x3,\displaystyle=x_{3}, (33)

From (29)(\ref{eq3.1})-(31)(\ref{eq3.3}), we have

ρ(F2)(x3x1x2)=x1+(n5)x5x2x32x4x12x4,\rho(F_{2})(x_{3}-x_{1}-x_{2})=x_{1}+(n-5)x_{5}-x_{2}-x_{3}-2x_{4}-x_{1}-2x_{4},

multiplying ρ(F2)\rho(F_{2}) on both sides, by (32) and (33), yields

ρ(F2)2(x3x1x2)=(n5)x3ρ(F2)x3ρ(F2)x24(x1+x2),\rho(F_{2})^{2}(x_{3}-x_{1}-x_{2})=(n-5)x_{3}-\rho(F_{2})x_{3}-\rho(F_{2})x_{2}-4(x_{1}+x_{2}),

then

(ρ(F2)2ρ(F2)4)(x3x1x2)=(n92ρ(F2))x3+ρ(F2)x1.(\rho(F_{2})^{2}-\rho(F_{2})-4)(x_{3}-x_{1}-x_{2})=(n-9-2\rho(F_{2}))x_{3}+\rho(F_{2})x_{1}. (34)

Since n19n\geq 19 and Kn5,1K_{n-5,1} is a proper subgraph of F2F_{2}, we have ρ(F2)>ρ(Kn5,1)=n5>3\rho(F_{2})>\rho(K_{n-5,1})=\sqrt{n-5}>3, thus ρ(F2)2ρ(F2)4>0\rho(F_{2})^{2}-\rho(F_{2})-4>0. By Theorem 3.13 and n19n\geq 19, we obtain ρ(F2)<2m(F2)n+1=2(n+1)n+1=n+3<(n9)/2\rho(F_{2})<\sqrt{2m(F_{2})-n+1}=\sqrt{2(n+1)-n+1}=\sqrt{n+3}<(n-9)/2, therefore n92ρ(F2)>0n-9-2\rho(F_{2})>0. Since xx is the principal eigenvector of F2F_{2}, we have xi>0x_{i}>0.

Thus, it follows from (34) that x3x1x2>0x_{3}-x_{1}-x_{2}>0.

Now we have

ρ(F1)ρ(F2)𝒙TA(F2)𝒙𝒙TA(F1)𝒙=2x4x32x4(x1+x2)=2x4(x3x1x2)>0.\begin{split}\rho(F_{1})-\rho(F_{2})&\geq\bm{x}^{T}A(F_{2})\bm{x}-\bm{x}^{T}A(F_{1})\bm{x}\\ &=2x_{4}x_{3}-2x_{4}(x_{1}+x_{2})\\ &=2x_{4}(x_{3}-x_{1}-x_{2})>0.\end{split}

Therefore, ρ(F1)>ρ(F2)\rho(F_{1})>\rho(F_{2}), which means k2k\geq 2.

Now we prove that ρ(G10(1,1,1,1,k1,nk3))<ρ(G1(1,1,1,k,nk3))\rho(G_{10}\circ(1,1,1,1,k-1,n-k-3))<\rho(G_{1}\circ(1,1,1,k,n-k-3)) for k2k\geq 2 and n12n\geq 12 by using a well-known operation.

Theorem 3.15.

[8] Let v1,v2v_{1},v_{2} be two vertices of a connected graph GG and let {u1,u2,,ut}N(v1)N(v2)\{u_{1},u_{2},\dots,u_{t}\}\subseteq N(v_{1})\setminus N(v_{2}). Let GG^{\prime} be the graph obtained from GG by rotating the edge v1uiv_{1}u_{i} to v2uiv_{2}u_{i} for i=1,2,,ti=1,2,\dots,t. If xv1xv2x_{v_{1}}\leq x_{v_{2}}, where x is the principal eigenvector of GG, then ρ(G)>ρ(G)\rho(G^{\prime})>\rho(G).

Lemma 3.16.

For k2k\geq 2 and n12n\geq 12, we have ρ(G10(1,1,1,1,k1,nk3))<ρ(G1(1,1,1,k,nk3))\rho(G_{10}\circ(1,1,1,1,k-1,n-k-3))<\rho(G_{1}\circ(1,1,1,k,n-k-3)).

Proof.

Let F1=G1(1,1,1,k,nk3)F_{1}=G_{1}\circ(1,1,1,k,n-k-3) and F2=G10(1,1,1,1,k1,nk3)F_{2}=G_{10}\circ(1,1,1,1,k-1,n-k-3). Let xx be the principal eigenvector of F2F_{2} and xix_{i} correspond to vertices in ViV_{i} for i=1,2,3,4,5,6i=1,2,3,4,5,6.

Let us first suppose that x3x1x_{3}\geq x_{1}, then by Theorem 3.15 we have ρ(F2)<ρ(F)\rho(F_{2})<\rho(F^{\prime}), where FF^{\prime} is obtained from F2F_{2} by rotating the edge v1v4v_{1}v_{4} to v3v4v_{3}v_{4}. Since FF1F^{\prime}\cong F_{1}, we obtain ρ(F2)<ρ(F1)\rho(F_{2})<\rho(F_{1}).

Now, suppose that x3<x1x_{3}<x_{1}. Since F′′F1F^{\prime\prime}\cong F_{1}, where F′′F^{\prime\prime} is obtained from F2F_{2} by rotating the edge v3v5iv_{3}v_{5}^{i} to v1v5iv_{1}v_{5}^{i} for i=1,2,,k1i=1,2,\dots,k-1, we have ρ(F2)<ρ(F′′)=ρ(F1)\rho(F_{2})<\rho(F^{\prime\prime})=\rho(F_{1}).

Thus, we complete the proof of the Lemma.

Now we know that the extremal graph of order nn and rank 55 with minimum spectral radius is G10(1,1,1,1,k,n4k)G_{10}\circ(1,1,1,1,k,n-4-k) for some integer kk with 1kn421\leq k\leq\frac{n-4}{2} when n12n\geq 12.

For convenience, we set Fn(i)=G10(1,1,1,1,i,n4i)F_{n}(i)=G_{10}\circ(1,1,1,1,i,n-4-i) and ={Fn(i):1in42}\mathcal{F}=\{F_{n}(i):1\leq i\leq\frac{n-4}{2}\}. It is only remained to find the extremal graph with minimum spectral radius in \mathcal{F}.

Theorem 3.17.

[10] Let AA be an n×nn\times n nonnegative matrix. Then the largest eigenvalue ρ(A)𝐱TA𝐱\rho(A)\geq\bm{x}^{T}A\bm{x} for any unit vector 𝐱\bm{x}, with equality if and only if A𝐱=ρ(A)𝐱A\bm{x}=\rho(A)\bm{x}.

Lemma 3.18.

Let α=6n3724n+118\alpha=\frac{6n-37-\sqrt{24n+1}}{18} and n12n\geq 12. Then for 1in421\leq i\leq\frac{n-4}{2}, we have

ρ(Fn(i))>min{ρ(Fn(α)),ρ(Fn(α))}\rho(F_{n}(i))>\min\{\rho(F_{n}(\lfloor\alpha\rfloor)),\rho(F_{n}(\lceil\alpha\rceil))\}

unless i=αi=\lfloor\alpha\rfloor or α\lceil\alpha\rceil.

Proof.

Let ρi=ρ(Fn(i))\rho_{i}=\rho(F_{n}(i)). Our aim is to prove that ρi<ρi1\rho_{i}<\rho_{i-1} if 2iα2\leq i\leq\lfloor\alpha\rfloor and ρi<ρi+1\rho_{i}<\rho_{i+1} if αin62\lceil\alpha\rceil\leq i\leq\frac{n-6}{2}.

Let 𝒙i\bm{x}_{i} be the principal eigenvector of Fn(i)F_{n}(i) and xjix_{j}^{i} correspond to vertices in VjV_{j} for j=1,2,3,4,5,6j=1,2,3,4,5,6. Then by ρi𝒙i=A(Fn(i))𝒙i\rho_{i}\bm{x}_{i}=A(F_{n}(i))\bm{x}_{i} we have

ρix1i\displaystyle\rho_{i}x_{1}^{i} =x2i+x3i+x4i,\displaystyle=x_{2}^{i}+x_{3}^{i}+x_{4}^{i}, (35)
ρix2i\displaystyle\rho_{i}x_{2}^{i} =x1i+x3i+ix5i,\displaystyle=x_{1}^{i}+x_{3}^{i}+ix_{5}^{i}, (36)
ρix3i\displaystyle\rho_{i}x_{3}^{i} =x1i+x2i+ix5i,\displaystyle=x_{1}^{i}+x_{2}^{i}+ix_{5}^{i}, (37)
ρix4i\displaystyle\rho_{i}x_{4}^{i} =x1i+(ni4)x6i,\displaystyle=x_{1}^{i}+(n-i-4)x_{6}^{i}, (38)
ρix5i\displaystyle\rho_{i}x_{5}^{i} =x2i+x3i,\displaystyle=x_{2}^{i}+x_{3}^{i}, (39)
ρix6i\displaystyle\rho_{i}x_{6}^{i} =x4i,\displaystyle=x_{4}^{i}, (40)

By (36)(\ref{eq3.8}) and (37)(\ref{eq3.9}), we have

ρi(x2ix3i)=x3ix2i, i.e.,\displaystyle\rho_{i}(x_{2}^{i}-x_{3}^{i})=x_{3}^{i}-x_{2}^{i}\text{, i.e.,}
(ρi+1)(x2ix3i)=0,\displaystyle(\rho_{i}+1)(x_{2}^{i}-x_{3}^{i})=0,

which implies that

x2i=x3i.\displaystyle x_{2}^{i}=x_{3}^{i}. (41)

Therefore, by (35) and (38)-(41), we have

x5i=2x2iρi\displaystyle x_{5}^{i}=\frac{2x_{2}^{i}}{\rho_{i}} =x1ix4iρi\displaystyle=x_{1}^{i}-\frac{x_{4}^{i}}{\rho_{i}}
=x1ix6i\displaystyle=x_{1}^{i}-x_{6}^{i}
=ρix4i(ni4)x6ix6i\displaystyle=\rho_{i}x_{4}^{i}-(n-i-4)x_{6}^{i}-x_{6}^{i}
=ρi2x6i(ni4)x6ix6i\displaystyle=\rho_{i}^{2}x_{6}^{i}-(n-i-4)x_{6}^{i}-x_{6}^{i}
=(ρi2n+i+3)x6i,\displaystyle=(\rho_{i}^{2}-n+i+3)x_{6}^{i},

and from (35)-(36) and (39)-(41), we have

x6i=x4iρi\displaystyle x_{6}^{i}=\frac{x_{4}^{i}}{\rho_{i}} =x1i2x2iρi\displaystyle=x_{1}^{i}-\frac{2x_{2}^{i}}{\rho_{i}}
=x1ix5i\displaystyle=x_{1}^{i}-x_{5}^{i}
=(ρi1)x2iix5ix5i\displaystyle=(\rho_{i}-1)x_{2}^{i}-ix_{5}^{i}-x_{5}^{i}
=ρi(ρi1)2x5iix5ix5i\displaystyle=\frac{\rho_{i}(\rho_{i}-1)}{2}x_{5}^{i}-ix_{5}^{i}-x_{5}^{i}
=12(ρi2ρi2i2)x5i.\displaystyle=\frac{1}{2}(\rho_{i}^{2}-\rho_{i}-2i-2)x_{5}^{i}.

Hence, we obtain that

{(ρi2n+i+3)(ρi2ρi2i2)=2,ρi2n+i+3>0,ρi2ρi2i2>0.\displaystyle\begin{cases}(\rho_{i}^{2}-n+i+3)(\rho_{i}^{2}-\rho_{i}-2i-2)=2,\\ \rho_{i}^{2}-n+i+3>0,\\ \rho_{i}^{2}-\rho_{i}-2i-2>0.\\ \end{cases} (42)

Note that, if we let

{ρi2n+i+3=1,ρi2ρi2i2=2,\displaystyle\begin{cases}\rho_{i}^{2}-n+i+3=1,\\ \rho_{i}^{2}-\rho_{i}-2i-2=2,\end{cases}

then we have

{ρi=ni2,ρi=1+8i+172.\displaystyle\begin{cases}\rho_{i}=\sqrt{n-i-2},\\ \rho_{i}=\frac{1+\sqrt{8i+17}}{2}.\end{cases}

By calculation, we can find that i=α=(6n3724n+1)/18i=\alpha=(6n-37-\sqrt{24n+1})/18 is the only solution of ni2=(1+8i+17)/2\sqrt{n-i-2}=(1+\sqrt{8i+17})/2. Since ii\in\mathbb{N}, we will complete the proof by classifying the value of ii.

Case 1. If 2iα2\leq i\leq\lfloor\alpha\rfloor.

We have ni2(1+8i+17)/2\sqrt{n-i-2}\geq(1+\sqrt{8i+17})/2. We claim that (1+8i+17)/2ρini2(1+\sqrt{8i+17})/2\leq\rho_{i}\leq\sqrt{n-i-2}. Suopose that ρi<(1+8i+17)/2\rho_{i}<(1+\sqrt{8i+17})/2. By (42), we have 0<ρi2n+i+3<10<\rho_{i}^{2}-n+i+3<1 and 0<ρi2ρi2i2<20<\rho_{i}^{2}-\rho_{i}-2i-2<2. Then (ρi2n+i+3)(ρi2ρi2i2)<2(\rho_{i}^{2}-n+i+3)(\rho_{i}^{2}-\rho_{i}-2i-2)<2, a contradiction. Suopose that ρi>ni2\rho_{i}>\sqrt{n-i-2}. By (42), we obtain that ρi2n+i+3>1\rho_{i}^{2}-n+i+3>1 and ρi2ρi2i2>2\rho_{i}^{2}-\rho_{i}-2i-2>2. Then (ρi2n+i+3)(ρi2ρi2i2)>2(\rho_{i}^{2}-n+i+3)(\rho_{i}^{2}-\rho_{i}-2i-2)>2, a contradiction.

Thus we have (1+8i+17)/2ρini2(1+\sqrt{8i+17})/2\leq\rho_{i}\leq\sqrt{n-i-2}. This induces that ρi2n+i+31\rho_{i}^{2}-n+i+3\leq 1 and ρi2ρi2i22\rho_{i}^{2}-\rho_{i}-2i-2\geq 2, which lead to x6ix5ix_{6}^{i}\geq x_{5}^{i}. Therefore

ρi1ρi𝒙𝒊TA(Fn(i1))𝒙𝒊𝒙𝒊TA(Fn(i))𝒙𝒊=2x5i(x4ix2ix3i)=2ρix5i(x6ix5i)0.\begin{split}&\rho_{i-1}-\rho_{i}\\ \geq&\bm{x_{i}}^{T}A(F_{n}(i-1))\bm{x_{i}}-\bm{x_{i}}^{T}A(F_{n}(i))\bm{x_{i}}\\ =&2x_{5}^{i}(x_{4}^{i}-x_{2}^{i}-x_{3}^{i})\\ =&2\rho_{i}x_{5}^{i}(x_{6}^{i}-x_{5}^{i})\\ \geq&0.\end{split} (43)

Now we only need to prove ρi1ρi\rho_{i-1}\neq\rho_{i}. Suppose that ρi1=ρi\rho_{i-1}=\rho_{i}, then ρi1=𝒙𝒊TA(Fn(i1))𝒙𝒊\rho_{i-1}=\bm{x_{i}}^{T}A(F_{n}(i-1))\bm{x_{i}}. By Theorem 3.17, we have

ρi1x4i=x1i+(ni4)x6i+x5i,\displaystyle\rho_{i-1}x_{4}^{i}=x_{1}^{i}+(n-i-4)x_{6}^{i}+x_{5}^{i},

and since

ρix4i=x1i+(ni4)x6i,\displaystyle\rho_{i}x_{4}^{i}=x_{1}^{i}+(n-i-4)x_{6}^{i},

we obtain 0=(ρi1ρi)x4i=x5i0=(\rho_{i-1}-\rho_{i})x_{4}^{i}=x_{5}^{i}, which contradicts to the definition of the principal eigenvector.

Therefore, from (43) we have ρi1>ρi\rho_{i-1}>\rho_{i} for 2iα2\leq i\leq\lfloor\alpha\rfloor.

Case 2. If αin62\lceil\alpha\rceil\leq i\leq\frac{n-6}{2}.

We have ni2(1+8i+17)/2\sqrt{n-i-2}\leq(1+\sqrt{8i+17})/2. Similarly, by (42), we conclude that ni2ρi(1+8i+17)/2\sqrt{n-i-2}\leq\rho_{i}\leq(1+\sqrt{8i+17})/2. This induces that ρi2n+i+31\rho_{i}^{2}-n+i+3\geq 1 and ρi2ρi2i22\rho_{i}^{2}-\rho_{i}-2i-2\leq 2, which lead to x5ix6ix_{5}^{i}\geq x_{6}^{i}, therefore

ρi+1ρi𝒙𝒊TA(Fn(i+1))𝒙𝒊𝒙𝒊TA(Fn(i))𝒙𝒊=2x6i(x2i+x3ix4i)=2ρix6i(x5ix6i)0.\begin{split}&\rho_{i+1}-\rho_{i}\\ \geq&\bm{x_{i}}^{T}A(F_{n}(i+1))\bm{x_{i}}-\bm{x_{i}}^{T}A(F_{n}(i))\bm{x_{i}}\\ =&2x_{6}^{i}(x_{2}^{i}+x_{3}^{i}-x_{4}^{i})\\ =&2\rho_{i}x_{6}^{i}(x_{5}^{i}-x_{6}^{i})\\ \geq&0.\end{split} (44)

Similarly, we have ρi+1ρi\rho_{i+1}\neq\rho_{i}. Using this, from (44), we obtain ρi+1>ρi\rho_{i+1}>\rho_{i} for αin62\lceil\alpha\rceil\leq i\leq\frac{n-6}{2}.

Therefore, the proof of Lemma is completed.

3.4 Step 4

It only remains for the case that 5n115\leq n\leq 11. Applying Proposition 3.9, 3.10 and 3.11, we obtain the extremal graphs with minimum spectral radius in Mn(G1)M_{n}(G_{1}), Mn(G7)M_{n}(G_{7}) and Mn(G10)M_{n}(G_{10}), respectively. And then calculate their spectral radii by using MATLAB, as shown in Table 3, where the extremal graphs and the minimum spectral radii are bolded.

Table 3: The extremal graph with minimum spectral radius in Mn(G1),Mn(G7),Mn(G10).M_{n}(G_{1}),M_{n}(G_{7}),M_{n}(G_{10}).
n Mn(G1)M_{n}(G_{1}) Mn(G7)M_{n}(G_{7}) Mn(G10)M_{n}(G_{10})
Extremal graph Spectral radius Extremal graph Spectral radius Extremal graph Spectral radius
5 G1(1,1,1,1,1)G_{1}\circ(1,1,1,1,1) 2.2143 G7(1,1,1,1,1)G_{7}\circ(1,1,1,1,1) 2.0000 \\backslash \\backslash
6 G1(1,1,1,1,2)G_{1}\circ(1,1,1,1,2) 2.2784 G7(2,1,1,1,1)G_{7}\circ(2,1,1,1,1) 2.3912 G10(1,1,1,1,1,1)G_{10}\circ(1,1,1,1,1,1) 2.6544
7 G1(1,1,1,1,3)G_{1}\circ(1,1,1,1,3) 2.3686 G7(2,1,2,1,1)G_{7}\circ(2,1,2,1,1) 2.6813 G10(1,1,1,1,1,2)G_{10}\circ(1,1,1,1,1,2) 2.6751
8 G1(1,1,1,1,4)G_{1}\circ(1,1,1,1,4) 2.4860 G7(3,1,2,1,1)G_{7}\circ(3,1,2,1,1) 2.9764 G10(1,1,1,1,1,3)G_{10}\circ(1,1,1,1,1,3) 2.7033
9 G1(1,1,1,1,5)G_{1}\circ(1,1,1,1,5) 2.6239 G7(3,1,3,1,1)G_{7}\circ(3,1,3,1,1) 3.2176 G10(1,1,1,1,1,4)G_{10}\circ(1,1,1,1,1,4) 2.7448
10 G1(1,1,1,1,6)G_{1}\circ(1,1,1,1,6) 2.7724 G7(4,1,3,1,1)G_{7}\circ(4,1,3,1,1) 3.4630 G10(1,1,1,1,1,5)G_{10}\circ(1,1,1,1,1,5) 2.8060
11 G1(1,1,1,1,7)G_{1}\circ(1,1,1,1,7) 2.9243 G7(4,1,4,1,1)G_{7}\circ(4,1,4,1,1) 3.6737 G10(1,1,1,1,1,6)G_{10}\circ(1,1,1,1,1,6) 2.8915

By Table 4, we obtain that when 5n115\leq n\leq 11, the extremal graph with minimum spectral radius of rank 55 is:

  • 1.

    G7=C5G_{7}=C_{5}, for n=5n=5;

  • 2.

    G1(1,1,1,1,n4)G_{1}\circ(1,1,1,1,n-4), for 6n106\leq n\leq 10;

  • 3.

    G10(1,1,1,1,1,n5)G_{10}\circ(1,1,1,1,1,n-5), for n=11n=11.

4 Concluding remarks

In the last case of Theorem 1.2, we obtain that k{6n3724n+118k\in\{\lfloor\frac{6n-37-\sqrt{24n+1}}{18}\rfloor ,6n3724n+118},\lceil\frac{6n-37-\sqrt{24n+1}}{18}\rceil\}. When 12n2312\leq n\leq 23, we use the MATLAB software to calculate the spectral radii of the graphs in ={Fn(i):1in42}\mathcal{F}=\{F_{n}(i):1\leq i\leq\frac{n-4}{2}\}, as shown in the Table 4, where the minimum spectral radius is bolded. It demonstrates that k=6n3724n+118k=\lfloor\frac{6n-37-\sqrt{24n+1}}{18}\rfloor or 6n3724n+118\lceil\frac{6n-37-\sqrt{24n+1}}{18}\rceil depends on nn.

Table 4: ρ(Fn(i))\rho(F_{n}(i)).
nn ii ρ(Fn(i))\rho(F_{n}(i)) 1 2 3 4 5 6 7 8 9 6n3724n+118\frac{6n-37-\sqrt{24n+1}}{18}
12 3 3.1370 3.4319 3.7362 \ \ \ \ \ 1
13 3.1239 3.1818 3.4431 3.7404 \ \ \ \ \ 1.2949
14 3.255 3.2470 3.4588 3.7457 4.0278 \ \ \ \ 1.5912
15 3.3894 3.3347 3.4817 3.7525 4.0308 \ \ \ \ 1.8889
16 3.5227 3.4402 3.5160 3.7616 4.0344 4.2979 \ \ \ 2.1877
17 3.6539 3.5563 3.5674 3.7743 4.0389 4.3001 \ \ \ 2.4876
18 3.7824 3.6770 3.6394 3.7926 4.0446 4.3027 4.5506 \ \ 2.7884
19 3.9079 3.7889 3.7303 3.8199 4.0523 4.3058 4.5522 \ \ 3.0901
20 4.0303 3.9201 3.8338 3.8612 4.0628 4.3097 4.5542 4.7888 \ 3.3927
21 4.1498 4.0396 3.9439 3.9211 4.0779 4.3147 4.5565 4.7900 \ 3.6960
22 4.2663 4.1570 4.0564 4 4.1002 4.3213 4.5593 4.7915 5.0146 4
23 4.3801 4.2721 4.1694 4.0929 4.1341 4.3303 4.5627 4.7933 5.0157 4.3047

It is a natural problem to determine the extramal spectral radii of the graphs of order nn and rank rr. By Theorem 1.1, we know that the maximum spectral radius of all connected graphs of order nn and rank rr is ρ(T(n,r))\rho(T(n,r)). Feng et al. gave the spectral radius of T(n,r)T(n,r) in [13].

Theorem 4.1.

[13] Let T(n,r)T(n,r) be a Turán graph. Then

ρ(T(n,r))=12(n2nr1+(n+1)24(nrnr)nr)nnr\rho(T(n,r))=\frac{1}{2}\left(n-2\lfloor\frac{n}{r}\rfloor-1+\sqrt{(n+1)^{2}-4(n-r\lfloor\frac{n}{r}\rfloor)\lceil\frac{n}{r}\rceil}\right)\leq n-\lfloor\frac{n}{r}\rfloor

with the last equality if and only if T(n,r)T(n,r) is regular.

Further, we obtain a sharp upper and lower bound for the spectral radius of the extremal graph GG which attains the minimum spectral radius among all connected graphs of order n12n\geq 12 and rank 55. By Theorem 1.2, we know that

ρ(G)=min{ρ(Fn(α)),ρ(Fn(α))},\displaystyle\rho(G)=\ \text{min}\ \{\rho(F_{n}(\lfloor\alpha\rfloor)),\rho(F_{n}(\lceil\alpha\rceil))\},

where α=6n3724n+118\alpha=\frac{6n-37-\sqrt{24n+1}}{18}.

From the proof of Lemma 3.18, we have

1+8α+172ρ(Fn(α)nα2,\displaystyle\frac{1+\sqrt{8\lfloor\alpha\rfloor+17}}{2}\leq\rho(F_{n}(\lfloor\alpha\rfloor)\leq\sqrt{n-\lfloor\alpha\rfloor-2},
nα2ρ(Fn(α)1+8α+172.\displaystyle\sqrt{n-\lceil\alpha\rceil-2}\leq\rho(F_{n}(\lceil\alpha\rceil)\leq\frac{1+\sqrt{8\lceil\alpha\rceil+17}}{2}.

Therefore, we obtain that

ρ(G)min{1+8α+172,nα2},\displaystyle\rho(G)\geq\text{min}\{\frac{1+\sqrt{8\lfloor\alpha\rfloor+17}}{2},\sqrt{n-\lceil\alpha\rceil-2}\},

and

ρ(G)min{nα2,1+8α+172}.\displaystyle\rho(G)\leq\text{min}\{\sqrt{n-\lfloor\alpha\rfloor-2},\frac{1+\sqrt{8\lceil\alpha\rceil+17}}{2}\}.

In general, the problem of determining the minimum spectral radius of all connected graphs with order nn and rank rr deserves further study.

Declaration of compting interest

There is no competing interest.

Acknowledgement

This research is supported by the National Natural Science Foundation of China [Grant number, 12171402].

References

  • [1] G. Chang, L. Huang, H.G. Yeh, A characterization of graphs with rank 4, Linear Algebra Appl. 434 (2011) 1793–1798.
  • [2] Z.Z. Lou, M.Q. Zhai, Proof of a conjecture on extremal spectral radii of blow-up graphs, Linear Algebra Appl. 617 (2021) 168–178.
  • [3] S.W. Sun, K.C. Das, Proof and disproof of conjectures on spectral radii of coclique extension of cycles and paths, Linear Algebra Appl. 618 (2021) 1–11
  • [4] I. Sciriha, On the rank of graphs, in: Y. Alavi, D.R. Lick, A. Schwenk (Eds.), Combinatorics, Graph Theory, and Algorithms, vol. II, New Issue Press, Western Michigan University, Kalamazoo, Michigan, 1999, pp. 769–778.
  • [5] G.J. Chang, L.H. Huang, H.G. Yeh, A characterization of graphs with rank 5, Linear Algebra Appl. 436 (2012) 4241–4250.
  • [6] D. Stevanović, I. Gutman, M.U. Rehman, On spectral radius and energy of complete multipartite graphs, Ars Math. Contemp. 9 (2015) 109–113.
  • [7] J. Monsalve, J. Rada, Extremal spectral radius of graphs with rank 4, Linear Algebra Appl. 609 (2021) 1–11.
  • [8] V. Nikiforov, Bounds on graph eigenvalues II, Linear Algebra Appl. 427 (2007) 183–189.
  • [9] D. Stevanović, Spectral Radius of Graphs, Academic Press, Amsterdam, 2015.
  • [10] A. Brouwer, W. Haemers, Spectra of Graphs, Springer, New York, 2012.
  • [11] Q. Li, K.Q. Feng, On the largest eigenvalue of graphs, Acta Math. Appl. Sinica. 2 (1979) 167–175 (in Chinese).
  • [12] Y. Hong, A bound on the spectral radius of graphs, Linear Algebra Appl. 108 (1988) 135–139.
  • [13] L.H. Feng, Q. Li, X.D. Zhang, Spectral radii of graphs with given chromatic number, Appl. Math. Lett. 20 (2007) 158–162.