1 Introduction

Let G be a simple graph with n vertices. The adjacency matrix A=A(G) is the symmetric matrix [ a i j ] n × n , where a i j = a j i =1 if v i v j is an edge of G, otherwise, a i j = a j i =0. We call det(λIA) the characteristic polynomial of G, denoted by ϕ(G;λ) (or abbreviated to ϕ(G)). Since A is symmetric, its eigenvalues λ 1 (G), λ 2 (G),, λ n (G) are real, and we assume that λ 1 (G) λ 2 (G) λ n (G). We call ρ(G)= λ 1 (G) the adjacency spectral radius of G.

The class of all graphs G whose largest (adjacency) spectral radius is bounded by 2 has been completely determined by Smith; see, for example, [1, 2]. Later, Hoffman [3], Cvetković et al. [4] gave a nearly complete description of all graphs G with 2<ρ(G)< 2 + 5 (≈2.0582). Their description was completed by Brouwer and Neumaier [5]. Then Woo and Neumaier [6] investigated the structure of graphs G with 2 + 5 < λ max (G)< 3 2 2 (≈2.1312), Wang et al. [7] investigated the structure of graphs whose largest eigenvalue is close to 3 2 2 .

Another interesting problem that arises in the context of graph eigenvalues is to order graphs in some class with respect to the spectral radius or least eigenvalue. In 2003, Guo [8] gave the first six unicyclic graphs of order n with larger spectral radius. Belardo et al. [9] ordered graphs with spectral radius in the interval (2, 2 + 5 ). In the paper [10], the first five unicyclic graphs on order n in terms of their smaller least eigenvalues were determined.

The graph obtained from a simple undirected graph by assigning an orientation to each of its edges is referred as the oriented graph. Let G σ be an oriented graph with vertex set { v 1 , v 2 ,, v n } and edge set E( G σ ). The skew-adjacency matrix S=S( G σ )= [ s i j ] n × n related to G σ is defined as

where (note that the definition is slightly different from the one of the normal skew-adjacency matrix given by Adiga et al. [11]). Since S( G σ ) is an Hermitian matrix, the eigenvalues λ 1 ( G σ ), λ 2 ( G σ ),, λ n ( G σ ) of S( G σ ) are all real numbers and, thus, can be arranged non-increase as

λ 1 ( G σ ) λ 2 ( G σ ) λ n ( G σ ) .

The skew-spectral radius and the skew-characteristic polynomial of G σ are defined respectively as

ρ s ( G σ ) =max { | λ 1 ( G σ ) | , | λ 2 ( G σ ) | , , | λ n ( G σ ) | }

and

ϕ ( G σ ; λ ) =det ( λ I n S ( G σ ) ) .

Recently, much attention has been devoted to the skew-adjacency matrix of an oriented graph. In 2009, Shader and So [12] investigated the spectra of the skew-adjacency matrix of an oriented graph. In 2010, Adiga et al. [11] discussed the properties of the skew-energy of an oriented graph. In papers [13, 14], all the coefficients of the skew-characteristic polynomial of G σ in terms of G were interpreted. Cavers et al. [15] discussed the graphs whose skew-adjacency matrices are all cospectral, and the relations between the matchings polynomial of a graph and the characteristic polynomials of its adjacency and skew-adjacency matrices. In [16], the author established a relation between ρ s ( G σ ) and ρ(G). Also, the author gave some results on the skew-spectral radii of G σ and its oriented subgraphs.

A connected graph in which the number of edges equals the number of vertices is called a unicyclic graph. In this paper, we will investigate the skew-spectral radius of an oriented unicyclic graph. The rest of this paper is organized as follows: In Section 2, we introduce some notations and preliminary results. In Section 3, all the oriented unicyclic graphs whose skew-spectral radius does not exceed 2 are determined. The result tells us that there is a big difference between the (adjacency) spectral radius of an undirected graph and the skew-spectral radius of its corresponding oriented graph. Furthermore, we order all the oriented unicyclic graphs with n vertices whose skew-spectral radius is bounded by 2 in Section 4.

2 Preliminaries

Let G=(V,E) be a simple graph with vertex set V=V(G)={ v 1 , v 2 ,, v n } and eE(G). Denote by Ge the graph obtained from G by deleting the edge e and by Gv the graph obtained from G by removing the vertex v together with all edges incident to it. For a nonempty subset W of V(G), the subgraph with vertex set W and edge set consisting of those pairs of vertices that are edges in G is called an induced subgraph of G. Denote by C n , K 1 , n 1 and P n the cycle, the star and the path on n vertices, respectively. Certainly, each subgraph of an oriented graph is also referred as an oriented graph and preserves the orientation of each edge.

Recall that the skew-adjacency matrix S( G σ ) of any oriented graph G σ is Hermitian, then the well-known interlacing theorem for Hermitian matrices applies equally well to oriented graphs; see, for example, Theorem 4.3.8 of [17].

Lemma 2.1 Let G σ be an arbitrary oriented graph on n vertices and V V(G). Suppose that | V |=k. Then

λ i ( G σ ) λ i ( G σ V ) λ i + k ( G σ ) for i=1,2,,nk.

Let G σ be an oriented graph, and let C= v 1 v 2 v k v 1 (k3) be a cycle of G, where v j adjacent to v j + 1 for j=1,2,,k1 and v 1 adjacent to v k . Let also S( G σ )= [ s i j ] n × n be the skew-adjacency matrix of G σ whose first k rows and columns correspond to the vertices v 1 , v 2 ,, v k . The sign of the cycle C σ , denoted by sgn( C σ ), is defined by

sgn ( C σ ) = s 1 , 2 s 2 , 3 s k 1 , k s k , 1 .

Let C ¯ = v 1 v k v 2 v 1 be the cycle by inverting the order of the vertices along the cycle C. Then one can verify that

sgn ( C ¯ σ ) = { sgn ( C σ ) , if  k  is odd ; sgn ( C σ ) , if  k  is even .

Moreover, sgn( C σ ) is either 1 or −1 if the length of C is even; and sgn( C σ ) is either or if the length of C is odd. For an even cycle, we simply refer it as a positive cycle or a negative cycle according to its sign. A positive even cycle is also named as oriented uniformly by Hou et al. [14].

On the skew-spectral radius of an oriented graph, we have obtained the following results. They will be useful in the proofs of the main results of this paper.

Lemma 2.2 ([[16], Theorem 2.1])

Let G σ be an arbitrary connected oriented graph. Denote by ρ(G) the (adjacency) spectral radius of G. Then

ρ s ( G σ ) ρ(G)

with equality if and only if G is bipartite and each cycle of G is a positive even cycle.

Lemma 2.3 ([[16], Theorem 3.2])

Let G σ be a connected oriented graph. Suppose that each even cycle of G is positive. Then

  1. (a)

    ρ s ( G σ )> ρ s ( G σ u) for any uG;

  2. (b)

    ρ s ( G σ )> ρ s ( G σ e) for any eG.

Lemma 2.4 ([[14], Theorem 2.4], [[16], Theorem 3.1])

Let G σ be an oriented graph, and let ϕ( G σ ,λ) be its skew-characteristic polynomial. Then

(a)ϕ ( G σ , λ ) =λϕ ( G σ u , λ ) v N ( u ) ϕ ( G σ u v , λ ) 2 u C sgn(C)ϕ ( G σ C , λ ) ,

where the first summation is over all the vertices in N(u), and the second summation is over all even cycles of G containing the vertex u,

(b)ϕ ( G σ , λ ) =ϕ ( G σ e , λ ) ϕ ( G σ u v , λ ) 2 ( u , v ) C sgn(C)ϕ ( G σ C , λ ) ,

where e=(u,v) and the summation is over all even cycles of G containing the edge e, and sgn(C) denotes the sign of the even cycle C.

Lemma 2.5 ([[13], A part of Theorem 2.5])

Let G σ be an oriented graph, and let ϕ( G σ ,λ) be its skew-characteristic polynomial. Then

d d λ ϕ ( G σ , λ ) = v V ( G ) ϕ ( G σ v , λ ) ,

where d d λ ϕ( G σ ,λ) denotes the derivative of ϕ( G σ ,λ).

Finally, we introduce a class of undirected graphs that will be often mentioned in this manuscript.

Denote by P l 1 , l 2 , , l k a pathlike graph, which is defined as follows: we first draw k (≥2) paths P l 1 , P l 2 ,, P l k of orders l 1 , l 2 ,, l k respectively along a line and put two isolated vertices between each pair of those paths, then add edges between the two isolated vertices and the nearest end vertices of such a pair of paths such that the four newly added edges form a cycle C 4 , where l 1 , l k 0 and l i 1 for i=2,3,,k1. Then P l 1 , l 2 , , l k contains i = 1 k l i +2k2 vertices. Notice that if l i =1 (i=2,3,,k1), the two end vertices of the path P l i are referred as overlap; if l 1 =0 ( l k =0), the left (right) of the graph P l 1 , l 2 , , l k has only two pendent vertices. Obviously, P 1 , 0 = K 1 , 2 , the star of order 3, and P 1 , 1 = C 4 . In general, P l 1 , l 2 , P 0 , l 1 , l 2 , P 0 , l 1 , l 2 , 0 are all unicyclic graphs containing C 4 , where l 1 , l 2 1.

3 The oriented unicyclic graphs whose skew-spectral radius does not exceed 2

In this section, we determine all the oriented unicyclic graphs whose skew-spectral radius does not exceed 2.

First, we introduce more notations. Denote by T l 1 , l 2 , l 3 the starlike tree with exactly one vertex v of degree 3, and T l 1 , l 2 , l 3 v= P l 1 P l 2 P l 3 , where P l i is the path of order l i (i=1,2,3).

Due to Smith, all undirected graphs whose (adjacency) spectral radius is bounded by 2 are completely determined as follows.

Lemma 3.1 ([2] or [[1], Chapter 2.7.12])

All undirected graphs whose spectral radius does not exceed 2 are C m , P 0 , n 4 , 0 , T 2 , 2 , 2 , T 1 , 3 , 3 , T 1 , 2 , 5 and their subgraphs, where m3 and n5.

By Lemma 2.4, to study the skew-spectrum properties of an oriented graph, we need only consider the sign of those even cycles. Moreover, Shader and So showed that S( G σ ) has the same spectrum as that of its underlying tree for any oriented tree G σ ; see Theorem 2.5 of [12]. Consequently, combining with Lemma 2.2, the skew-spectral radius of each oriented graph whose underlying graph is as described in Lemma 3.1, regardless of the orientation of the oriented cycle C n σ , does not exceed 2.

For convenience, we write:

U={G|G is a unicyclic graph}.

U(m)={G|G is a unicyclic graph in U containing the cycle  C m }.

U (m)={G|G is a unicyclic graph in U(m) which is not the cycle  C m }.

U n ={G|G is a unicyclic graph on order n}.

Moreover, let C m = v 1 v 2 v m v 1 be a cycle on m vertices, and let P l 1 , P l 2 ,, P l m be m paths with lengths l 1 , l 2 ,, l m (perhaps some of them are empty), respectively. Denote by C m l 1 , l 2 , , l m the unicyclic undirected graph obtained from C m by joining v i to a pendent vertex of P l i for i=1,2,,m. Suppose, without loss of generality, that l 1 =max{ l i :i=1,2,,m}, l 2 l m , and write C m l 1 , l 2 , , l j instead of the standard C m l 1 , l 2 , , l j , 0 , , 0 if l j + 1 = l j + 2 == l m =0.

By Lemmas 2.2 and 2.4 or papers [11, 12], for a given unicyclic graph GU(m), we know that the skew-spectral radius of G σ is independent of its orientation if m is odd. Therefore, we will briefly write G instead of the normal notation G σ if each cycle of G is odd. If m is even, then essentially, there exist two orientations σ 1 (the sign of the even cycle is positive) and σ 2 (the sign of the even cycle is negative) such that ρ s ( G σ 1 )=ρ(G) and ρ s ( G σ 2 )<ρ(G). Henceforth, we will briefly write G (or G + ) instead of G σ if the sign of each even cycle is negative (or positive). In particular, G will also denote the oriented graph if G is a tree since ρ s ( G σ )=ρ(G) in this case.

3.1 The C 4 -free oriented unicyclic graphs whose skew-spectral radius does not exceed 2

Let G σ be an oriented graph with the property

ρ s ( G σ ) 2.
(3.1)

The property (3.1) is hereditary, because, as a direct consequence of Lemma 2.1, for any induced subgraph HG, H σ also satisfies (3.1). The inheritance (hereditary) of property (3.1) implies that there are minimal connected graphs that do not obey (3.1); such graphs are called forbidden subgraphs. It is easy to verify the following.

Lemma 3.2 Let GUU(4) with ρ s ( G σ )2. Then C 3 3 , C 3 1 , 1 , C 3 (2), C 3 1 (2), C 5 1 , C 7 1 are forbidden, where C 3 (2) (or C 3 1 (2)) denotes the oriented graph obtained by adding two pendent vertices to a vertex (or the pendent vertex) of C 3 (or C 3 1 ).

Combining with Lemma 3.2 and the fact that ρ s (T)>2 if the oriented tree T contains an arbitrary tree described as Lemma 3.1 as a proper subgraph, we have the following result.

Theorem 3.1 Let GUU(4) and G C m . Let also ρ s ( G σ )2. Then G σ is one of C 3 2 , ( C 6 2 , 0 , 0 , 2 ) , ( C 6 1 , 0 , 1 , 0 , 1 ) , ( C 8 1 , 0 , 0 , 0 , 1 ) and their induced oriented unicyclic subgraphs.

Proof Denote by gir(G) the girth of G. Let gir(G)=m and C m be the cycle of G with vertex set { v 1 , v 2 ,, v m } such that v i adjacent to v i + 1 for i=1,2,,m1 and v m adjacent to v 1 . (We should point out once again that in C m l 1 , l 2 , , l j (jm), we always refer v i adjacent to one pendent vertex of P l i , a path with length l i , for i=1,2,,j.) We divide our proof into the following four claims.

Claim 1 If gir(G)=3, then G σ { C 3 1 , C 3 2 }.

The result follows from Lemma 3.2 that C 3 3 , C 3 1 , 1 and C 3 (2), C 3 1 (2) are forbidden.

Claim 2 If gir(G)3, then gir(G){6,8}. Moreover, each induced even cycle of G σ is negative.

Let gir(G)=m. Notice that G is C 4 -free, then m5 if m3, and, thus, G contains the induced subgraph C m 1 as G C n . From Lemma 3.2, both C 5 1 and C 7 1 are forbidden, thus, m5,7. Moreover, the graph obtained from C m 1 by deleting the vertex v 5 is the tree T 1 , 3 , m 5 for m6. Thus, there is an induced subgraph T 1 , 3 , 4 if gir(G)9, which is a contradiction to Lemma 3.1. Hence, the former follows.

Assume to the contrary that there exists a positive even cycle C m + , then by Lemma 2.3, ρ s ( G σ ) ρ s ( ( C m 1 ) + )> ρ s ( C m + )=2, a contradiction. Thus, the latter follows.

Claim 3 If gir(G)=6, then G σ is one of ( C 6 1 , 0 , 1 , 0 , 1 ) , ( C 6 2 , 0 , 0 , 2 ) or their induced subgraphs.

By Claim 2, we always suppose that each cycle C ˆ 6 is negative.

We first claim that G is of C 6 l 1 , l 2 , l 3 , l 4 , l 5 , l 6 , that is, each pendent tree adjacent to v i of C 6 is a path for i=1,2,,6. Otherwise, assume that the pendent tree adjacent to v 1 is not a path, then the resultant graph by deleting vertex v 3 of G is a tree and contains the tree P 0 , l , 0 as a proper induced subgraph, and, thus, ρ s ( G σ )> ρ s ( P 0 , l , 0 )=2 combining with Lemmas 2.3 and 3.2, a contradiction. Moreover, we have l 1 2. Otherwise, G v 4 contains T 2 , 2 , 3 as an induced subgraph. Notice that both C 6 1 , 1 v 4 and C 6 2 , 0 , 1 v 5 are trees and contain P 0 , 2 , 0 as a proper induced subgraph, then G may be C 6 1 , 0 , 1 , 0 , 1 and C 6 2 , 0 , 0 , 2 . By calculation, we have ρ s ( ( C 6 1 , 0 , 1 , 0 , 1 ) )=2 and ρ s ( ( C 6 2 , 0 , 0 , 2 ) )=2. Thus, the result follows.

Claim 4 If gir(G)=8, then G σ is one of ( C 8 1 , 0 , 0 , 0 , 1 ) or its induced subgraphs.

By Claim 2, the cycle C 8 σ of G σ is negative. Notice that C 8 2 v 5 = T 2 , 3 , 3 , C 8 1 , 1 v 5 = T 2 , 2 , 3 , C 8 1 , 0 , 1 v 5 = T 2 , 2 , 3 , C 8 1 , 0 , 0 , 1 v 5 = T 2 , 2 , 3 , each of them has skew-spectral radius greater than 2. Then G σ may be ( C 8 1 , 0 , 0 , 0 , 1 ) . By calculation, we have ρ s ( ( C 8 1 , 0 , 0 , 0 , 1 ) )=2. Thus, the result follows.  □

3.2 The oriented unicyclic graphs in U(4) whose skew-spectral radius does not exceed 2

Now, we consider the oriented unicyclic graphs in U(4). First, we have the following.

Lemma 3.3 Let l 1 , l 2 1. Then

  1. (a)

    ρ s ( P l 1 , l 2 )<2;

  2. (b)

    ρ s ( P 0 , l 1 , l 2 )= ρ s ( P 0 , l 1 , l 2 , 0 )=2.

Proof (a) Let n= l 1 + l 2 +2. We first show by induction on n that

ϕ ( P l 1 , l 2 , 2 ) =4.
(3.2)

Let l 1 l 2 . Then there is exactly one pathlike graph if n=4, namely, P 1 , 1 = C 4 . By calculation, we have

ϕ ( P 1 , 1 , 2 ) =4.

Suppose now that n5, and the result is true for the order no more than n1. Applying Lemma 2.4 to the left pendent vertex of P l 1 , l 2 , we have

ϕ ( P l 1 , l 2 , λ ) =λϕ ( P l 1 1 , l 2 , λ ) ϕ ( P l 1 2 , l 2 , λ ) .

Then ϕ( P l 1 , l 2 ,2)=4 by induction hypothesis, and, thus, the result follows.

Let now v be a vertex with degree 2 in C 4 of P l 1 , l 2 . Then P l 1 , l 2 v= P n 1 , a path of order n1. Let λ 1 λ 2 λ n and λ ¯ 1 λ ¯ 2 λ ¯ n 1 be all eigenvalues of P l 1 , l 2 and P n 1 , respectively. By Lemma 2.1 and the fact that λ ¯ 1 <2, we have λ 2 λ ¯ 1 <2. On the other hand, we have

ϕ ( P l 1 , l 2 , λ ) = i = 1 n (λ λ i ).

Consequently, λ 1 <2, and, thus, ρ s ( P l 1 , l 2 )<2 by Eq. (3.2). Thus, the result (a) holds.

  1. (b)

    We first show that 2 is an eigenvalue of P 0 , l 1 , l 2 .

    ϕ ( P 0 , l 1 , l 2 , λ ) =λϕ ( P l 1 + 1 , l 2 , λ ) λϕ ( P l 1 1 , l 2 , λ ) .

By the proof of the result (a), we know that

ϕ ( P l 1 + 1 , l 2 , 2 ) =ϕ ( P l 1 1 , l 2 , 2 ) =4.

It tells us that

ϕ ( P 0 , l 1 + 1 , l 2 , 2 ) =0.

Note that λ 2 ( P 0 , l 1 + 1 , l 2 )<2. We know that ρ s ( P 0 , l 1 , l 2 )=2.

Now, we show that 2 is also an eigenvalue of P 0 , l 1 , l 2 , 0 . Applying Lemma 2.5, we have

d d λ ϕ ( P 0 , l 1 , l 2 , 0 , λ ) = v ϕ ( P 0 , l 1 , l 2 , 0 v , λ ) .

It is easy to see that 2 is an eigenvalue of each oriented graph P 0 , l 1 , l 2 , 0 v. Thus, 2 is an eigenvalue of P 0 , l 1 , l 2 , 0 with multiplicity 2. □

By calculation, we have the following.

Lemma 3.4 Let GU(4) with ρ s ( G σ )2. Then G i (i=1,2,,7) are forbidden, where G 1 = C 4 5 , 1 , G 2 = C 4 3 , 2 , G 3 = C 4 4 , 1 , 1 , G 4 = C 4 3 , 1 , 2 , G 5 = C 4 2 , 2 , 1 , G 6 = C 4 3 , 1 , 0 , 1 and G 7 = P 0 , l 1 , l 2 , 0 1 , which denotes the graph obtained by adding a pendent vertex to a vertex of P 0 , l 1 , l 2 , 0 .

Combining with Lemma 3.4 and the fact that ρ s (T)>2 if the oriented tree T contains an arbitrary tree described as Lemma 3.1 as a proper subgraph, we have the following result.

Theorem 3.2 Let G U (4) and ρ s ( G σ )2. Then G σ is one of ( C 4 4 , 1 ) , ( C 4 3 , 1 , 1 ) , ( C 4 2 , 1 , 2 , 1 ) , ( C 4 2 , 2 ) and P 0 , l 1 , l 2 , 0 or their induced oriented unicyclic subgraphs.

Proof Note that the induced cycle C 4 σ of G σ must be negative. By Lemma 3.3, we can assume that G P 0 , l 1 , l 2 , 0 .

Case 1. G C 4 l 1 , l 2 , l 3 , l 4 .

Then G contains an induced tree T such that T has a proper induced subgraph P 0 , l , 0 . It means that ρ s ( G σ )>ρ( P 0 , l , 0 )=2, a contradiction.

Case 2. G= C 4 l 1 , l 2 , l 3 , l 4 .

Then, by Lemma 3.4, we know that l 1 4 and l 2 1. Thus, it is not difficult to see that the possible oriented graphs are ( C 4 4 , 1 ) , ( C 4 3 , 1 , 1 ) , ( C 4 2 , 1 , 2 , 1 ) , ( C 4 2 , 2 ) or their induced oriented unicyclic subgraphs by Lemma 3.4. Moreover, taking some computations, we know the skew-spectral radius of each above oriented graph does not exceed 2.

Combining with Lemma 3.3, the result follows. □

3.3 The oriented unicyclic graphs whose skew-spectral radius does not exceed 2

Putting Lemma 3.1 together with Theorem 3.1 and Theorem 3.2, we have the following.

Theorem 3.3 Let GU and ρ s ( G σ )2. Then G σ is one of C m σ , C 3 2 , ( C 4 4 , 1 ) , ( C 4 3 , 1 , 1 ) , ( C 4 2 , 1 , 2 , 1 ) , ( C 4 2 , 2 ) , ( C 6 2 , 0 , 0 , 2 ) , ( C 6 1 , 0 , 1 , 0 , 1 ) , ( C 8 1 , 0 , 0 , 0 , 1 ) and P 0 , l 1 , l 2 , 0 or their induced oriented unicyclic subgraphs, where the orientation of C m σ is arbitrary.

Moreover, by calculation, we have the following two corollaries from Theorem 3.3.

Corollary 3.1 Let GU and ρ s ( G σ )=2. Then G σ is one of the following oriented graphs.

  1. (a)

    C m + , where m is even;

  2. (b)

    P 0 , l 1 , l 2 , P 0 , l 1 , l 2 , 0 , where l 1 , l 2 1;

  3. (c)

    C 3 2 , ( C 4 4 , 1 ) , ( C 4 3 , 1 , 1 ) , ( C 4 2 , 1 , 2 , 1 ) , ( C 4 2 , 1 , 2 ) , ( C 4 2 , 1 , 1 , 1 ) , ( C 4 2 , 1 , 0 , 1 ) , ( C 4 2 , 2 ) , ( C 6 2 , 0 , 0 , 2 ) , ( C 6 1 , 0 , 1 , 0 , 1 ) , ( C 8 1 , 0 , 0 , 0 , 1 ) and ( C 8 1 ) .

Corollary 3.2 Let GU and ρ s ( G σ )<2. Then G σ is one of the following oriented graphs or their induced oriented unicyclic subgraphs.

  1. (a)

    C m σ , where m is odd, or m is even, and the sign of C m σ is negative;

  2. (b)

    P l 1 , l 2 , where l 1 , l 2 1;

  3. (c)

    C 3 1 , ( C 4 3 , 1 ) , ( C 4 2 , 1 , 1 ) , ( C 4 1 , 1 , 1 , 1 ) , ( C 6 2 , 0 , 0 , 1 ) , ( C 6 1 , 0 , 1 ) .

4 Ordering the oriented unicyclic graphs whose skew-spectral radius is bounded by 2

In this section, we discuss the skew-spectral radii of oriented unicyclic graphs in U n . Let G U n and ρ s ( G σ )<2. By Corollary 3.2, we know that G σ is C n σ (where n is odd, or n is even, and the sign is negative) or P l , n l (where l1) if n10. This makes it possible to order the oriented unicyclic graphs whose skew-spectral radius is bounded by 2.

Lemma 4.1 Let l 2 l 1 2. Then ρ s ( P l 1 , l 2 )< ρ s ( P l 1 1 , l 2 + 1 ).

Proof By Lemma 2.4, we have

ϕ ( P l 1 , l 2 ) = λ ϕ ( P l 1 + l 2 + 1 ) ϕ ( P l 1 1 ) ϕ ( P l 2 + 1 ) ϕ ( P l 1 + 1 ) ϕ ( P l 2 1 ) + 2 ϕ ( P l 1 1 ) ϕ ( P l 2 1 ) ; ϕ ( P l 1 1 , l 2 + 1 ) = λ ϕ ( P l 1 + l 2 + 1 ) ϕ ( P l 1 2 ) ϕ ( P l 2 + 2 ) ϕ ( P l 1 ) ϕ ( P l 2 ) + 2 ϕ ( P l 1 2 ) ϕ ( P l 2 ) .

Thus,

ϕ ( P l 1 , l 2 ) ϕ ( P l 1 1 , l 2 + 1 ) = [ ϕ ( P l 1 2 ) ϕ ( P l 2 + 2 ) ϕ ( P l 1 1 ) ϕ ( P l 2 + 1 ) ] + [ ϕ ( P l 1 ) ϕ ( P l 2 ) ϕ ( P l 1 + 1 ) ϕ ( P l 2 1 ) ] + 2 [ ϕ ( P l 1 1 ) ϕ ( P l 2 1 ) ϕ ( P l 1 2 ) ϕ ( P l 2 ) ] .

Moreover, we have

ϕ ( P l 1 2 ) ϕ ( P l 2 + 2 ) ϕ ( P l 1 1 ) ϕ ( P l 2 + 1 ) = ϕ ( P l 1 2 ) [ λ ϕ ( P l 2 + 1 ) ϕ ( P l 2 ) ] ϕ ( P l 2 + 1 ) [ λ ϕ ( P l 1 2 ) ϕ ( P l 1 3 ) ] = ϕ ( P l 1 3 ) ϕ ( P l 2 + 1 ) ϕ ( P l 1 2 ) ϕ ( P l 2 ) = ϕ ( P 0 ) ϕ ( P l 2 l 1 + 4 ) ϕ ( P 1 ) ϕ ( P l 2 l 1 + 3 ) = ϕ ( P l 2 l 1 + 4 ) λ ϕ ( P l 2 l 1 + 3 ) = ϕ ( P l 2 l 1 + 2 ) , ϕ ( P l 1 ) ϕ ( P l 2 ) ϕ ( P l 1 + 1 ) ϕ ( P l 2 1 ) = ϕ ( P l 1 ) [ λ ϕ ( P l 2 1 ) ϕ ( P l 2 2 ) ] ϕ ( P l 2 1 ) [ λ ϕ ( P l 1 ) ϕ ( P l 1 1 ) ] = ϕ ( P l 1 1 ) ϕ ( P l 2 1 ) ϕ ( P l 1 ) ϕ ( P l 2 2 ) = ϕ ( P 0 ) ϕ ( P l 2 l 1 ) ϕ ( P 1 ) ϕ ( P l 2 l 1 1 ) = ϕ ( P l 2 l 1 ) λ ϕ ( P l 2 l 1 1 ) = ϕ ( P l 2 l 1 2 ) ,

where l 2 l 1 2. It is easy to know that

ϕ( P l 1 )ϕ( P l 2 )ϕ( P l 1 + 1 )ϕ( P l 2 1 )=1if  l 2 l 1 =0

and

ϕ( P l 1 )ϕ( P l 2 )ϕ( P l 1 + 1 )ϕ( P l 2 1 )=0if  l 2 l 1 =1.

Similarly, we have

ϕ ( P l 1 1 ) ϕ ( P l 2 1 ) ϕ ( P l 1 2 ) ϕ ( P l 2 ) = ϕ ( P l 2 1 ) [ λ ϕ ( P l 1 2 ) ϕ ( P l 1 3 ) ] ϕ ( P l 1 2 ) [ λ ϕ ( P l 2 1 ) ϕ ( P l 2 2 ) ] = ϕ ( P l 1 2 ) ϕ ( P l 2 2 ) ϕ ( P l 1 3 ) ϕ ( P l 2 1 ) = ϕ ( P 1 ) ϕ ( P l 2 l 1 + 1 ) ϕ ( P 0 ) ϕ ( P l 2 l 1 + 2 ) = λ ϕ ( P l 2 l 1 ) ϕ ( P l 2 l 1 + 1 ) = ϕ ( P l 2 l 1 ) .

Hence,

ϕ ( P l 1 , l 2 ) ϕ ( P l 1 1 , l 2 + 1 ) =ϕ( P l 2 l 1 + 2 )ϕ( P l 2 l 1 2 )+2ϕ( P l 2 l 1 ).

Let l 2 l 1 =k. Then for k2, we have

ϕ ( P l 1 , l 2 ) ϕ ( P l 1 1 , l 2 + 1 ) = ϕ ( P k + 2 ) ϕ ( P k 2 ) + 2 ϕ ( P k ) = [ ϕ ( P 2 ) ϕ ( P k ) ϕ ( P 1 ) ϕ ( P k 1 ) ] ϕ ( P k 2 ) + 2 ϕ ( P k ) = ( 3 λ 2 ) ϕ ( P k ) + ϕ ( P 1 ) ϕ ( P k 1 ) ϕ ( P k 2 ) = ( 4 λ 2 ) ϕ ( P k ) .

Obviously, the above equality also holds for k=0,1. It means that ϕ( P l 1 , l 2 , ρ s ( P l 1 1 , l 2 + 1 ))>0, since ρ s ( P l 1 1 , l 2 + 1 )<2. Thus, ρ s ( P l 1 , l 2 )< ρ s ( P l 1 1 , l 2 + 1 ). □

By Lemma 4.1, we know that

ρ s ( P n 2 2 , n 2 2 ) << ρ s ( P 2 , n 4 ) < ρ s ( P 1 , n 3 ) .

Now, we need only to compare the skew-spectral radii of P l 1 , l 2 and C n σ . In fact, we have the following.

Lemma 4.2 Let n4. Then we have

  1. (a)

    ρ s ( P 1 , n 3 )< ρ s ( C n ) if n is odd;

  2. (b)

    ρ s ( P n 2 2 , n 2 2 )= ρ s ( C n ) if n is even.

Proof Note that by paper [11]

ρ s ( C n σ ) = { 2 cos π 2 n , if  n  is odd ; 2 cos π n , if  n  is even and the sign of the cycle is negative .

Moreover, we have ρ s ( P 0 , n 2 )=2cos π 2 n 2 . Thus, ρ s ( C n )> ρ s ( P 0 , n 2 ) if n is odd.

On the other hand, we have

ϕ ( P 1 , n 3 ) = λ ϕ ( P n 1 ) ϕ ( P n 2 ) ϕ ( P 2 ) ϕ ( P n 4 ) + 2 ϕ ( P n 4 ) ; ϕ ( P 0 , n 2 ) = λ ϕ ( P n 1 ) λ ϕ ( P n 3 ) .

Thus,

ϕ ( P 1 , n 3 ) ϕ ( P 0 , n 2 ) = ϕ ( P 2 ) ϕ ( P n 4 ) + 2 ϕ ( P n 4 ) = ( 4 λ 2 ) ϕ ( P n 4 ) .

It means that ρ s ( P 1 , n 3 )< ρ s ( P 0 , n 2 ). Then the result (a) follows.

If n is even, then let l= n 2 2 . We have

ϕ ( P l , l ) = λ ϕ ( P n 1 ) 2 ϕ ( P l 1 ) ϕ ( P l + 1 ) + 2 ϕ ( P l 1 ) ϕ ( P l 1 ) ; ϕ ( C n ) = λ ϕ ( P n 1 ) 2 ϕ ( P n 2 ) + 2 ϕ ( C n ) = λ ϕ ( P n 1 ) 2 ϕ ( P l 1 ) ϕ ( P l + 1 ) + 2 ϕ ( P l 2 ) ϕ ( P l ) + 2 .

Thus,

ϕ ( P l , l ) ϕ ( C n ) = 2 ϕ ( P l 2 ) ϕ ( P l ) + 2 ϕ ( P l 1 ) ϕ ( P l 1 ) 2 = 2 [ ϕ ( P l 2 ) ϕ ( P l 2 ) ϕ ( P l 3 ) ϕ ( P l 1 ) ] 2 = 2 [ ϕ ( P 1 ) ϕ ( P 1 ) ϕ ( P 0 ) ϕ ( P 2 ) ] 2 = 0 .

Then the result (b) holds. □

By Lemmas 4.1 and 4.2, we obtain the following interesting result.

Theorem 4.1 Let G σ be an oriented unicyclic graph on order n (n10). G σ P l 1 , l 2 , C n σ , where n= l 1 + l 2 +2 and C n σ = C n if n is even. Then

  1. (a)

    ρ s ( P n 3 2 , n 1 2 )<< ρ s ( P 1 , n 3 )< ρ s ( C n )<2 ρ s ( G σ ) if n is odd;

  2. (b)

    ρ s ( C n )= ρ s ( P n 2 2 , n 2 2 )<< ρ s ( P 1 , n 3 )<2 ρ s ( G σ ) if n is even.

Combining with Corollary 3.1, we have ordered all the oriented unicyclic graphs with n vertices whose skew-spectral radius is bounded by 2.