1 Introduction

Let V(G) be the vertex set and E(G) be the edge set of a graph G. The numbers of vertices and edges of G are denoted by n(G) and m(G), respectively. For a vertex vV(G), let N G (v) be the set of vertices adjacent to v and d G (v)=| N G (v)| be the degree of v. Particularly, denote by Δ(G) the maximum degree of G. The diameter of a connected graph G, denoted by d(G), is the maximum distance among all pairs of vertices in G. Let A(G) be the adjacency matrix of G and D(G) be the diagonal matrix of vertex degrees. The matrix D(G)A(G) is called the Laplacian matrix of G and its eigenvalues are called the Laplacian eigenvalues of G. Let μ 1 μ 2 μ n be the Laplacian eigenvalues of a graph G with n vertices. It is well known that μ n =0 and i = 1 n 1 μ i =2m(G). In particular, μ n 1 is called the algebraic connectivity of G and it is denoted by α(G).

The Laplacian matrix is an important topic in the theory of graph spectra. Particularly, much literature has paid attention to μ 1 , μ 2 , μ n 1 or μ 1 μ n 1 for trees (see, for example, [16]). Let S n be the star of order n, S a , b k be the tree obtained from two stars S a + 1 , S b + 1 by joining a path of length k between their central vertices (see Figure 1). As is well known, among all trees of order n, S n has the largest value of μ 1 (see [7]) and S n 3 , 1 1 has the second largest value of μ 1 (see [6]). On the other hand, Guo [4] proved that these two trees also attain the first two smallest values of μ 2 , respectively. This implies that μ 1 , μ 2 cannot attain simultaneously the maximal (or minimal) value and even the relation between them seems like a seesaw. Therefore, it is interesting to investigate the value of μ 1 + μ 2 . Moreover, Zhang [6] showed that the S k 1 , k 1 1 , S k 1 , k 2 2 , S k 2 , k 2 3 attain simultaneously the largest value of μ 2 among all trees with 2k vertices. Then Shao et al.[5] showed that S k 1 , k 1 2 attains the largest value of μ 2 among all trees with 2k+1 vertices.

Figure 1
figure 1

S a , b k : a tree of order a+b+k+1 .

Another motivation to study the value of μ 1 + μ 2 came from a result of Haemers et al.[8], who showed that μ 1 + μ 2 m(G)+3 for any graph G. This result implies that Brouwer’s conjecture [9],

μ 1 + μ 2 ++ μ k m(G)+ ( k + 1 2 ) ,

is true for k=2. Considering a tree T, we have μ 1 + μ 2 n(T)+2. Recently, Fritscher et al.[10] improved this bound by giving μ 1 + μ 2 <n(T)+2 2 n ( T ) . This paper determines the extremal tree that attains the bound of μ 1 + μ 2 . Moreover, for general connected graphs, we also give a conjecture on the extremal graphs for μ 1 + μ 2 .

2 A sharp upper bound of μ 1 + μ 2

Let S k (G) be the sum of the largest k Laplacian eigenvalues of a graph G. When k=2, we shall write S(G) instead of S k (G) for simplicity. For graphs G and H, we denote by GH the graph with vertex set V(G)V(H) and edge set E(G)E(H). The following lemmas come from an important result as regards a real symmetric matrix.

Lemma 2.1 ([8])

Let G 1 , G 2 ,, G r be some edge-disjoint graphs. Then S k ( i = 1 r G i ) i = 1 r S k ( G i )for any k.

Lemma 2.2 ([8])

For any graph G, S(G)m(G)+3.

Lemma 2.3 Let G be a connected graph, d i = d G ( v i )and m i = Σ v j N G ( v i ) d j / d i . Then

  1. (i)

    [11] μ 1 (G)Δ(G)+1, with equality if and only ifΔ(G)=n(G)1.

  2. (ii)

    [7] μ 1 (G)n(G), with equality if and only if the complement of G is disconnected.

  3. (iii)

    [12] μ 1 (G)max{ d i + m i | v i V(G)}.

Lemma 2.4 ([6])

Let T be a tree of order n. IfT S n , then μ 1 (T) μ 1 ( S n 3 , 1 1 ), with equality if and only ifT S n 3 , 1 1 .

Corollary 2.5 Let T be a tree with n vertices and diameterd3. Then μ 1 (T)<n0.5.

Proof Note that any tree T has diameter d3 if T S n . According to Lemma 2.4, μ 1 (T) μ 1 ( S n 3 , 1 1 ). Further, by Lemma 2.3,

μ 1 ( S n 3 , 1 1 ) max{ d i + m i }=n2+ n 1 n 2 =n1+ 1 n 2 <n0.5

for n5. For n=4, a straightforward calculation shows that μ 1 ( S 1 , 1 1 )=2+ 2 <3.5. □

Lemma 2.6 ([2])

Let T be a tree of order n and diameterd3. Thenα(T)α( S n d + 1 2 , n d + 1 2 d 2 ), with equality if and only ifT S n d + 1 2 , n d + 1 2 d 2 .

Lemma 2.7 ([11])

Let G be a graph with a vertex u of degree one. Thenα(G)α(Gu).

Lemma 2.7 implies that the algebraic connectivity of a tree is not greater than that of its subtree.

Lemma 2.8 ([4])

Let T n k (n2k+1) be a tree obtained from a star S n k by replacing its k edges with k paths of length two, respectively. Ifk2, then μ 2 ( T n k )= 3 + 5 2 .

The following lemma can be found in [13] and is known as the Interlacing Theorem of Laplacian eigenvalues.

Lemma 2.9 Let G be a graph of order n and H be a graph obtained from G by deleting an edge. Then

μ 1 (G) μ 1 (H) μ n (G) μ n (H)=0.

Next we give the main theorem of this section. Its proof is divided into several sequent claims.

Theorem 2.10 For any tree T with ordern4, S(T)S( S n 2 2 , n 2 2 1 ). The equality holds if and only ifT S n 2 2 , n 2 2 1 .

Claim 2.11 For any tree T with ordern4and diameterd3, S(T)<S( S n 2 2 , n 2 2 1 )except thatT S n 2 2 , n 2 2 1 .

Proof If d(T)=3, then T S a , b 1 for some positive integers a, b with a+b=n2. It is well known that the Laplacian characteristic polynomial of S a , b 1 is μ ( μ 1 ) n 4 f a , b (μ), where

f a , b (μ)= μ 3 (n+2) μ 2 +(ab+2n+1)μn.
(1)

Note that S a , b 1 contains S 1 , 1 1 as a subtree. By Lemma 2.9, μ 2 ( S a , b 1 ) μ 2 ( S 1 , 1 1 )=2. Moreover, we know that for any tree T, α(T)1, with equality if and only if T is a star. These imply that μ 1 ( S a , b 1 ), μ 2 ( S a , b 1 ), and α( S a , b 1 ) consist of the three roots of f a , b (μ). As follows from (1), we have

μ 1 ( S a , b 1 ) + μ 2 ( S a , b 1 ) +α ( S a , b 1 ) =n+2.
(2)

By virtue of Lemma 2.6, we have α( S a , b 1 )>α( S n 2 2 , n 2 2 1 ) except that (a,b)=( n 2 2 , n 2 2 ). Equivalently, S( S a , b 1 )<S( S n 2 2 , n 2 2 1 ) except that (a,b)=( n 2 2 , n 2 2 ).

If d(T)=2, then T S n . We first give a lower bound of S( S n 2 2 , n 2 2 1 ) for n6:

S ( S n 2 2 , n 2 2 1 ) >n+1.5.
(3)

Indeed, by (2) it suffices to show α( S n 2 2 , n 2 2 1 )<0.5. Note that for n6, S n 2 2 , n 2 2 1 contains S 2 , 2 1 as a subtree. By Lemma 2.7, α( S n 2 2 , n 2 2 1 )α( S 2 , 2 1 )= 5 17 2 <0.5.

Note that S( S n )=n+1 for n3. According to (3), we have S( S n )<S( S n 2 2 , n 2 2 1 ) for n6. As for n{4,5}, a straightforward calculation shows that

S ( S 1 , 1 1 ) 5.4142,S ( S 2 , 1 1 ) 6.4811.
(4)

Also we have S( S n )<S( S n 2 2 , n 2 2 1 ). □

Claim 2.12 For any tree T with order n and diameterd5, S(T)<S( S n 2 2 , n 2 2 1 ).

Proof Since d(T)5, then n6 and there is a path of length 5 in T. By inequality (3), it suffices to show S(T)n+1.5. First suppose that there is a path v 0 v 1 v 5 in T such that either max{ d T ( v 0 ), d T ( v 5 )}2 or max{ d T ( v 2 ), d T ( v 3 )}3. Let T 1 , T 2 be the two components of T v 2 v 3 . Clearly, both T 1 and T 2 have at least two edges.

If μ 1 , μ 2 of T 1 T 2 attain at the same component, say T 1 , then by Lemma 2.2,

S( T 1 T 2 )=S( T 1 )m( T 1 )+3m( T 1 T 2 )+1.
(5)

Note that S( v 2 v 3 )=S( S 2 )=2. By Lemma 2.1,

S(T)S( T 1 T 2 )+S( v 2 v 3 )m( T 1 T 2 )+3=m(T)+2=n+1.
(6)

Otherwise, S( T 1 T 2 )= μ 1 ( T 1 )+ μ 1 ( T 2 ). Whether max{ d T ( v 0 ), d T ( v 5 )}2 or max{ d T ( v 2 ), d T ( v 3 )}3, we can observe that max{d( T 1 ),d( T 2 )}3. Say d( T 2 )3, then by Corollary 2.5, μ 1 ( T 2 )<n( T 2 )0.5. By Lemma 2.3(ii), μ 1 ( T 1 )n( T 1 ). Hence,

S(T)S( T 1 T 2 )+S( v 2 v 3 )=n( T 1 )+n( T 2 )0.5+2=n+1.5.

Next, we may assume that each path v 0 v 1 v 5 of length 5 in T has d T ( v 0 )= d T ( v 5 )=1 and d T ( v 2 )= d T ( v 3 )=2. This implies that d(T)=5 and T S a , b 3 for some integers a, b with a+b=n4. If a=b=1, then T is isomorphic to a path of order 6 and a straightforward calculation shows that S(T)=5+ 3 <n+1.5, as claimed. Otherwise, assume without loss of generality that a2. Then d T ( v 1 )3. Let T 3 , T 4 be the two components of T v 1 v 2 with v 0 v 1 E( T 3 ). Then both T 3 and T 4 have at least two edges. If μ 1 , μ 2 of T 3 T 4 attain at the same component, say T 3 , then by Lemmas 2.1 and 2.2,

S(T)S( T 3 T 4 )+S( v 1 v 2 )=S( T 3 )+2m( T 3 )+5m(T)+2=n+1.

Otherwise, S( T 3 T 4 )= μ 1 ( T 3 )+ μ 1 ( T 4 ). Note that μ 1 ( T 3 )n( T 3 ). Since d( T 4 )=3, by Corollary 2.5, μ 1 ( T 4 )<n( T 4 )0.5. So

S(T)S( T 3 T 4 )+S( v 1 v 2 )n( T 3 )+n( T 4 )0.5+2=n+1.5.

 □

Claim 2.13 For any tree T with order n and diameter 4, S(T)<S( S n 2 2 , n 2 2 1 ).

Proof First suppose that T contains a path v 0 v 1 v 4 such that max{ d T ( v 1 ), d T ( v 3 )}3. Now n6 and it suffices to show S(T)n+1.5. Without loss of generality assume that d T ( v 1 )3. Let T 1 , T 2 be the two components of T v 1 v 2 with v 0 v 1 E( T 1 ). Then both T 1 and T 2 have at least two edges.

If μ 1 , μ 2 of T 1 T 2 attain at the same component, say T 1 , then similarly to inequalities (5) and (6), we can observe that S(T)n+1.

Now let S( T 1 T 2 )= μ 1 ( T 1 )+ μ 1 ( T 2 ). If d T ( v 2 )3, then d( T 2 )3 and hence μ 1 ( T 2 )<n( T 2 )0.5. So

S(T)S( T 1 T 2 )+S( v 1 v 2 )<n( T 1 )+n( T 2 )0.5+2=n+1.5.

If d T ( v 2 )=2, then T S a , b 2 for some positive integers a, b with 3a+b=n3. Moreover, since d T ( v 1 )3, then a2. If (a,b){(2,1),(3,1)}, a straightforward calculations show that S( S a , b 2 )<n+1.5. Otherwise, S a , b 2 contains either S 4 , 1 2 or S 2 , 2 2 as a subtree. Since

μ 3 ( S 4 , 1 2 ) 1.5068, μ 3 ( S 2 , 2 2 ) 1.5858,

it follows from Lemma 2.9 that μ 3 ( S a , b 2 )>1.5. Since S a , b 2 is not a star, μ n 1 ( S a , b 2 )<1. On the other hand, note that the matrix 1 I n [D( S a , b 2 )A( S a , b 2 )] has a identical rows and b different identical rows, so the multiplicity of eigenvalue 1 is at least a+b2 and else five eigenvalues are μ 1 , μ 2 , μ 3 , μ n 1 and μ n =0. Since i = 1 n μ i ( S a , b 2 )=2(n1), we have

i = 1 3 μ i ( S a , b 2 ) + μ n 1 ( S a , b 2 ) + μ n ( S a , b 2 ) =2(n1)(a+b2)=n+3.

This implies that S( S a , b 2 )<n+3 μ 3 ( S a , b 2 )<n+1.5.

Next, it suffices to consider the case that each path v 0 v 1 v 4 of T has d T ( v 1 )= d T ( v 3 )=2. This implies that T T n k for some k2 and n2k+1, since d(T)=4. According to Lemma 2.8, μ 2 ( T n k )= 3 + 5 2 . Moreover, by Lemma 2.3,

μ 1 ( T n k ) max{ d i + m i }=nk1+ n 1 n k 1 n2+ 2 n 3 .

Thus for n6,

S ( T n k ) n2+ 2 n 3 + 3 + 5 2 <n+1.5<S ( S n 2 2 , n 2 2 1 ) .

When n=5, T n k is a path. Comparing with (4), S( T n k )=4+ 5 <S( S 2 , 1 1 ). This completes the proof. □

Following from Claims 2.11-2.13, Theorem 2.10 holds and the unique tree with maximal S(T) is S n 2 2 , n 2 2 1 . According to (2),

μ ( S n 2 2 , n 2 2 1 ) <n+2=m ( S n 2 2 , n 2 2 1 ) +3.

Theorem 2.14 Let m, n be two positive integers withnm2n3and G m , n be a graph of order n and size m obtained from a given edge uv by joiningmn+1independent vertices with u and v, respectively, and another2nm3independent vertices with u. ThenS( G m , n )=m+3.

Proof Let H s , t be a graph obtained by joining a vertex to s vertices of a given complete graph of order s+t and H s , t c be its complement graph. Then H s , t c is isomorphic to the union of S t + 1 and s isolated vertices. Clearly, the Laplacian eigenvalues of H s , t c consist of t+1, 1 with multiplicity t1 and 0 with multiplicity s+1. Recall that for any graph G with n vertices, μ i (G)=n μ n i ( G c ) for 1in1 and μ n (G)=0. So the Laplacian eigenvalues of H s , t consist of s+t+1 with multiplicity s, s+t with multiplicity t1, s and 0.

Now G m , n c is isomorphic to the union of H 2 n m 3 , m n + 1 and an isolated vertex. So the Laplacian eigenvalues of G m , n c consist of n1 with multiplicity 2nm3, n2 with multiplicity mn, 2nm3, and 0 with multiplicity 2. Therefore, the Laplacian eigenvalues of G m , n consist of n, mn+3, 2 with multiplicity mn, 1 with multiplicity 2nm3 and 0. So S( G m , n )=n+(mn+3)=m+3. □

Recall that μ 1 (G)n(G) for any graph G. When m(G)>2n(G)3, Haemers’ bound is clearly not attainable. Theorem 2.14 implies that if m(G)2n(G)3, Haemers’ bound is always sharp for connected graphs other than trees. Ending the paper, we present a conjecture on the uniqueness of the extremal graph.

Conjecture 2.15 Among all connected graphs with n vertices andnm2n3edges, G m , n is the unique graph with maximal value of μ 1 + μ 2 .