# Some inequalities on the skew-spectral radii of oriented graphs

- Guang-Hui Xu
^{1}Email author

**2012**:211

https://doi.org/10.1186/1029-242X-2012-211

© Xu; licensee Springer 2012

**Received: **11 February 2012

**Accepted: **18 September 2012

**Published: **2 October 2012

## Abstract

Let *G* be a simple graph and ${G}^{\sigma}$ be an oriented graph obtained from *G* by assigning a direction to each edge of *G*. The adjacency matrix of *G* is $A(G)$ and the skew-adjacency matrix of ${G}^{\sigma}$ is $S({G}^{\sigma})$. The adjacency spectral radius $\rho (G)$ of *G* and the skew-spectral radius $\varrho ({G}^{\sigma})$ of ${G}^{\sigma}$ are defined as the spectral radius of $A(G)$ and $S({G}^{\sigma})$ respectively.

In this paper, we firstly establish a relation between $\varrho ({G}^{\sigma})$ and $\rho (G)$. Also, we give some results on the skew-spectral radii of ${G}^{\sigma}$ and its oriented subgraphs. As an application of these results, we obtain a sharp upper bound of the skew-spectral radius of an oriented unicyclic graph.

**MSC:**05C50, 15A18.

## Keywords

## 1 Introduction

In this paper, *i* will always denote an imaginary unit. The identity matrix is denoted by *I* and the transpose of the matrix *A* by ${A}^{T}$. Let *G* be a simple graph with *n* vertices. The *adjacency matrix* $A=A(G)$ is the symmetric matrix ${[{a}_{jk}]}_{n\times n}$, where ${a}_{jk}={a}_{kj}=1$ if ${v}_{j}{v}_{k}$ is an edge of *G*, otherwise ${a}_{jk}={a}_{kj}=0$. We call $det(\lambda I-A)$ the *characteristic polynomial* of *G*, denoted by $P(G;\lambda )$. The *adjacency spectrum* ${S}_{p}(G)$ of *G* is defined as the spectrum of $A(G)$. Since *A* is symmetric, its eigenvalues ${\lambda}_{1}(G),{\lambda}_{2}(G),\dots ,{\lambda}_{n}(G)$ are real, and we assume that ${\lambda}_{1}(G)\ge {\lambda}_{2}(G)\ge \cdots \ge {\lambda}_{n}(G)$. We call $\rho (G)={\lambda}_{1}(G)$ the *adjacency spectral radius* of *G*.

Let ${G}^{\sigma}$ be a simple graph with an orientation *σ*, which assigns to each edge of *G* a direction so that ${G}^{\sigma}$ becomes a directed graph. The *skew-adjacency matrix* $S=S({G}^{\sigma})$ is the real skew-symmetric matrix ${[{s}_{jk}]}_{n\times n}$, where ${s}_{jk}=1$ and ${s}_{kj}=-1$ if $({v}_{j},{v}_{k})$ is an arc of ${G}^{\sigma}$, otherwise ${s}_{jk}={s}_{kj}=0$. We call $det(\lambda I-S)$ the *skew-characteristic polynomial* of ${G}^{\sigma}$, denoted by $\varphi ({G}^{\sigma};\lambda )$. The *skew spectrum* ${S}_{p}({G}^{\sigma})$ of ${G}^{\sigma}$ is defined as the spectrum of $S({G}^{\sigma})$. Since *S* is skew-symmetric, its eigenvalues are purely imaginary numbers ${\lambda}_{1}({G}^{\sigma})i,{\lambda}_{2}({G}^{\sigma})i,\dots ,{\lambda}_{n}({G}^{\sigma})i$. Also, we assume that ${\lambda}_{1}({G}^{\sigma})\ge {\lambda}_{2}({G}^{\sigma})\ge \cdots \ge {\lambda}_{n}({G}^{\sigma})$ and call $\varrho ({G}^{\sigma})={\lambda}_{1}({G}^{\sigma})$ the *skew-spectral radius* of ${G}^{\sigma}$.

In this paper, we will denote by $\mathfrak{D}(G)$ the set of all the oriented graphs obtained from *G* by giving an arbitrary orientation to each edge. Also, we refer to [1–3] for more terminologies and notations not defined here.

Unlike the adjacency matrix of a graph, there is little research on the skew-adjacency matrix $S({G}^{\sigma})$, except that into enumeration of perfect matchings of a graph. As early as in 1947, Tutte [4] derived his famous characterization of the graphs with no perfect matchings. Tutte’s result motivates a lot of work on the matchings polynomial and enumerating perfect matchings of graphs in terms of its skew-adjacency matrix; see, for example, [5–7] and references therein.

Recently, many researchers have paid a great deal of attention to the spectral properties of skew-symmetric matrices in terms of oriented graphs. IMA-ISU research group on minimum rank [8] studied the minimum rank of skew-symmetric matrices. In 2009, Shader and So [9] investigated the spectra of the skew-adjacency matrix of an oriented graph. And in 2010, Adiga *et al.* [10] discussed the properties of the skew energy of an oriented graph. In the papers [11] and [12], all the coefficients of the skew-characteristic polynomial of ${G}^{\sigma}$ in terms of *G* were interpreted.

The motivation of this paper is to study more carefully the skew spectra of oriented graphs. In Section 2, we firstly establish a relation between $\varrho ({G}^{\sigma})$ and $\rho (G)$. In Section 3, we give some results on the skew-spectral radii of ${G}^{\sigma}$ and its oriented subgraphs. Finally, in Section 4, we give an application of the previous results - to obtain a sharp upper bound of the skew-spectral radius of an oriented unicyclic graph.

## 2 A relation between $\rho (G)$ and $\varrho ({G}^{\sigma})$

It is interesting to discuss the relations between $\rho (G)$ and $\varrho ({G}^{\sigma})$. Let *G* be a simple graph. We denoted by $N(v)$ the *neighborhood* of the vertex *v* in *G*, by $G-e$ the subgraph obtained from *G* by deleting the edge *e* and by $G-v$ the subgraph obtained from *G* by removing the vertex *v* together with all edges incident to it. A *walk* *W* of length *k* from *u* to *v* in *G* is a sequence of $k+1$ vertices starting at *u* and ending at *v* such that consecutive vertices are adjacent. If all vertices in a walk are distinct, then such a walk is called a *path* of *G*, denoted by *P*. Let $P={v}_{1}{v}_{2}\cdots {v}_{k}$ be a path with $k\ge 3$. Then *P* together with the edge ${v}_{k}{v}_{1}$ is called a *cycle* of *G*, denoted by *C*.

*G*(we often call it a $(k+1)$-walk). The sign of

*W*in ${G}^{\sigma}$, denoted by $sgn({W}^{\sigma})$, is defined by

*W*. Then one can find that

Obviously, for an even closed walk (that is to say, ${u}_{k+1}={u}_{1}$), we can simply refer to it as a positive (or negative) even closed walk according to its sign, regardless of the order of its vertices. Similarly, we can define a positive (or negative) even cycle. Moreover, an even cycle ${C}_{2k}$ with 2*k* vertices is said to be *oriented uniformly* if its sign in ${G}^{\sigma}$ is $sgn({C}_{2k}^{\sigma})={(-1)}^{k}$.

Note that in terms of defining walks, paths, cycles *etc.*, we focus only on the underlying undirected graph. And their signs are based on the oriented graph.

For ${G}^{\sigma}\in \mathfrak{D}(G)$, one reverses the orientations of all the arcs incident to a particular vertex of ${G}^{\sigma}$. We call such an operation a *reversal* of ${G}^{\sigma}$. Let ${G}^{\tau}$ be the digraph obtained from ${G}^{\sigma}$ by doing some reversals. Then ${G}^{\sigma}$ and ${G}^{\tau}$ are said to be *quasi-isomorphic* and denoted by ${G}^{\sigma}\simeq {G}^{\tau}$. For example, if *T* is a tree and ${T}^{\sigma},{T}^{\tau}\in \mathfrak{D}(T)$, then we always have ${T}^{\sigma}\simeq {T}^{\tau}$.

On quasi-isomorphic oriented graphs, Adiga *et al.* [10] obtained the following result.

**Lemma 2.1** ([10])

*Let*${G}^{\sigma}\in \mathfrak{D}(G)$

*and*${G}^{\tau}\simeq {G}^{\sigma}$.

*Then the skew*-

*adjacency matrices*$S({G}^{\sigma})$

*and*$S({G}^{\tau})$

*are orthogonally similar*.

*And therefore*,

For oriented bipartite graphs, we have

**Lemma 2.2**

*Let*$G=(U,V;E)$

*be a bipartite graph and*$|U|=m$, $|V|=n$.

*The adjacency matrix of*

*G*

*is*

*Let*${G}^{\sigma}\in \mathfrak{D}(G)$

*and each even cycle of*

*G*

*be oriented uniformly*.

*Then*${G}^{\sigma}\simeq D$,

*where*$D\in \mathfrak{D}(G)$

*and its skew*-

*adjacency matrix is*

*Proof*Firstly, we will prove that each even closed walk of

*G*is oriented uniformly in ${G}^{\sigma}$ too. That is to say, for an arbitrary closed 2

*k*-walk of

*G*, its sign in ${G}^{\sigma}$ is ${(-1)}^{k}$. Let $W={u}_{1}{u}_{2}\cdots {u}_{2k}{u}_{1}$ be a closed 2

*k*-walk of

*G*. We may assume the closed 2

*k*-walk is exactly constituted by the cycles (or closed 2-walks) ${C}_{j}$ with $2{k}_{j}$ vertices ($j=1,2,\dots ,r$). Hence,

Now, to prove the lemma, we can consider the following two cases.

Case 1. $G\cong {K}_{mn}$;

Let ${u}_{1}\in U$, ${v}_{1}\in V$. Obviously, we can take some reversals to obtain a new oriented graph ${G}^{\tau}$ such that all the arcs $({u}_{1},{v}_{j}),({u}_{l},{v}_{1})\in {G}^{\tau}$ ($j=1,2,\dots ,n$, $l=1,2,\dots ,m$). Now, if ${G}^{\tau}\ncong D$, then there exists an arc $({v}_{p},{u}_{q})\in {G}^{\tau}$, where ${u}_{q}\in U$ and ${v}_{p}\in V$. Thus, for the cycle ${C}_{4}={u}_{1}{v}_{1}{u}_{q}{v}_{p}{u}_{1}$, we have $sgn({C}_{4}^{\sigma})=sgn({C}_{4}^{\tau})=-1\ne {(-1)}^{2}$. Contradicting the condition of this lemma that each even cycle of *G* is oriented uniformly in ${G}^{\sigma}$, ${G}^{\tau}\cong D$, and then ${G}^{\sigma}\simeq D$.

Case 2. $G\ncong {K}_{mn}$;

*G*is connected. Then there must exist two vertices $u\in U$, $v\in V$ not adjacent in

*G*. Let ${W}_{1}=u{u}_{1}\cdots {u}_{2{k}_{1}}v$ be a walk of

*G*. Now, we join the vertices

*u*and

*v*and suppose the resulting graph is ${G}_{1}$. Obviously, we can give the edge

*uv*a direction (for instance, from

*u*to

*v*) such that the closed $2({k}_{1}+1)$-walk $W={W}_{1}+uv$ satisfies $sgn({W}^{\varsigma})={(-1)}^{{k}_{1}+1}$ in ${G}_{1}^{\varsigma}={G}^{\sigma}+(u,v)$. Of course, it means that

*G*, then

*G*. Thus,

Hence, the closed $2({k}_{2}+1)$-walk $\overline{W}=uv+{W}_{2}$ of ${G}_{1}$ satisfies $sgn({\overline{W}}^{\varsigma})={(-1)}^{{k}_{2}+1}$ in ${G}_{1}^{\varsigma}$. Thus, each even cycle of ${G}_{1}$ is oriented uniformly in ${G}_{1}^{\varsigma}$. It tells us that the graph ${G}_{1}$ satisfies the conditions of this lemma too. Furthermore, if the result holds for ${G}_{1}$, we can get immediately the same result for the graph *G*. By taking some similar operations, we can obtain the graph ${K}_{mn}$. And then, by the Case 1 and the above discussion, we know the result holds. □

Moreover, the following result tells us that ${S}_{p}(D)=i{S}_{p}(G)$ for a bipartite graph *G* and its oriented graph *D* defined as in Lemma 2.2.

**Lemma 2.3** ([9])

*Let*

*and*

*be two real matrices*. ${S}_{p}(A)$ *and* ${S}_{p}(B)$ *are denoted by the spectrum of* *A* *and* *B* *respectively*. *Then* ${S}_{p}(B)=i{S}_{p}(A)$.

To obtain the main result of this section, we also need the following result.

**Lemma 2.4** ([2])

*Let* *A* *be an irreducible nonnegative* $n\times n$ *matrix and* *B* *be a complex* $n\times n$ *matrix such that* $|B|\le A$ (*entry*-*wise*). *Then* $|\lambda (B)|\le \rho (A)$ *for each eigenvalue* $\lambda (B)$ *of* *B*; *and* $\lambda (B)=\rho (A){e}^{i\theta}$ *iff* $B={e}^{i\theta}LA{L}^{-1}$, *where* ${e}^{i\theta}=cos\theta +isin\theta $ *and* $|L|={I}_{n}$, *the identity matrix*.

Now, by the above lemmas, we can give the relation between $\rho (G)$ and $\varrho ({G}^{\sigma})$ as follows.

**Theorem 2.1** *Let* *G* *be a connected graph*, ${G}^{\sigma}\in \mathfrak{D}(G)$. *Then* $\varrho ({G}^{\sigma})\le \rho (G)$ *with equality if and only if* *G* *is a bipartite graph and each even cycle of* *G* *is oriented uniformly in* ${G}^{\sigma}$.

*Proof* By Lemma 2.4, we know that $\varrho ({G}^{\sigma})\le \rho (G)$.

*r*-cycle of

*G*. Then $sgn({C}^{\sigma})={s}_{12}{s}_{23}\cdots {s}_{r-1,r}{s}_{r1}$. Thus,

Since $sgn({C}^{\sigma})=1$ or −1, *r* must be an even number. Hence, *G* is a bipartite graph, and each even cycle of *G* is oriented uniformly in ${G}^{\sigma}$.

*G*be a bipartite graph and each even cycle of

*G*be oriented uniformly in ${G}^{\sigma}$. Assume the adjacency matrix of

*G*is

*D*is

Thus, by Lemma 2.1 and Lemma 2.3, we have $\varrho ({G}^{\sigma})=\varrho (D)=\rho (G)$. □

**Remark 2.1** By this theorem and those known results on the adjacency spectral radii of bipartite graphs, we can obtain similar results on the skew-spectral radii of oriented bipartite graphs.

For example, we can give the following sharp upper bound on the skew-spectral radii of oriented bipartite graphs.

**Theorem 2.2**

*Let*$G=(U,V;E)$

*be a bipartite graph with*$|U|={n}_{1}$

*and*$|V|={n}_{2}$.

*Let*${G}^{\sigma}\in \mathfrak{D}(G)$.

*Then*

*with equality if and only if*${G}^{\sigma}\simeq D$,

*where*$D\in \mathfrak{D}({K}_{{n}_{1}{n}_{2}})$,

*and its skew*-

*adjacency matrix is*

*where* ${J}_{{n}_{1}{n}_{2}}$ *is the* ${n}_{1}\times {n}_{2}$ *matrix with all entries equal to* 1.

*Proof* It is well known that $\rho (G)\le \sqrt{{n}_{1}{n}_{2}}$ with equality if and only if $G\cong {K}_{{n}_{1}{n}_{2}}$. Thus, by Lemma 2.2 and Theorem 2.1, we know the result holds. □

## 3 Some relations on the skew-spectral radii of an oriented graph and its subgraphs

In this section, we will discuss the relations on the skew-spectral radii of an oriented graph and its subgraphs due to vertex (or edge) deletion. Certainly, in this paper, each subgraph of an oriented graph is also referred to as an oriented graph and preserves the direction of each arc, even if we do not indicate specially.

Let *G* be a simple graph and ${G}^{\sigma}\in \mathfrak{D}(G)$. A subgraph *H* of *G* is called a *basic subgraph* if each component of *H* is an edge or an even cycle. Of course, $|V(H)|$ is an even number.

In the papers [11] and [12], the following result was obtained.

*Let*${G}^{\sigma}\in \mathfrak{D}(G)$

*and the skew*-

*characteristic polynomial is*

*Then the coefficient of*${\lambda}^{n-j}$

*satisfies*

*where the summation is over all the basic subgraphs of* *G* *having* 2*r* *vertices* ($0\le r\le \lfloor \frac{n}{2}\rfloor $), ${c}^{+}$ *and* *c* *are respectively the number of positive even cycles and the number of even cycles contained in* ${H}^{\sigma}$.

As an application of this lemma, we can obtain the following results (the result (b) of the following theorem was also obtained by Hou and Lei [12]) which can be used to find recursions for the skew-characteristic polynomial of some oriented graphs.

**Theorem 3.1**

*Let*$G=(V,E)$

*be a simple graph*, $u\in V(G)$, $e\in E(G)$.

*Then the skew*-

*characteristic polynomial of an oriented graph*${G}^{\sigma}$

*satisfies the following identities*:

*where the first summation is over all the vertices in*$N(u)$

*and the second summation is over all the even cycles of*

*G*

*containing the vertex*

*u*.

*Moreover*,

*where* $e=uv$ *and the summation is over all the even cycles of* *G* *containing the edge* *e*.

*Proof*(a) Obviously, all basic subgraphs of

*G*having $j=2r$ vertices can be divided into three parts: those that do not contain the vertex

*u*, those that contain

*u*together with a neighbor

*v*as an edge of it, and those that contain

*u*together with the vertices of an even cycle containing

*u*as the component of it. Then ${a}_{j}={a}_{j}^{(1)}+{a}_{j}^{(2)}+{a}_{j}^{(3)}$, where

the summation is over all the even cycles of *G* containing the vertex *u*. And ${a}_{j-|V(Z)|}={\sum}_{{H}_{3}}{(-1)}^{{c}^{+}}{2}^{c}$, where the summation is over all the basic subgraphs of $G-V(Z)$ having $j-|V(Z)|$ vertices.

When *j* is an odd number, we write ${a}_{j}^{(t)}={a}_{j}=0$ ($t=1,2,3$).

Thus, the result holds.

(b) The proof is similar to that of (a). □

Now, we consider the relations on the skew-spectral radii of an oriented graph and its subgraphs due to vertex (or edge) deletion. Firstly, according to a classical result in the matrix theory, we have

**Lemma 3.2** (Interlacing of eigenvalues)

*Then we have*

*Proof* Let $S({G}^{\sigma})$ be the skew-adjacency matrix of ${G}^{\sigma}$. Since $iS({G}^{\sigma})$ is a Hermitian matrix, by the well-known Cauchy-Poincare theorem, we know the result holds. □

Now, we can get the following strict inequalities on the skew-spectral radii of an oriented graph and its subgraphs.

**Theorem 3.2**

*Let*

*G*

*be a connected simple graph on*

*n*

*vertices*, ${G}^{\sigma}\in \mathfrak{D}(G)$.

*If each even cycle of*

*G*

*is oriented uniformly in*${G}^{\sigma}$,

*then*

- (a)
$\varrho ({G}^{\sigma})>\varrho ({G}^{\sigma}-u)$,

*where**u**is an arbitrary vertex of**G*. - (b)
$\varrho ({G}^{\sigma})>\varrho ({G}^{\sigma}-e)$,

*where**e**is an arbitrary edge of**G*.*Moreover*,*for*$\varrho \ge \varrho ({G}^{\sigma}-e)$,*we have*${i}^{-n}\varphi ({G}^{\sigma}-e,\varrho i)>{i}^{-n}\varphi ({G}^{\sigma},\varrho i)$.

*Proof* (a) We prove the statement by induction on *n*. Obviously, the result holds for $n=2,3$.

*n*. We consider the case that the number of vertices equals

*n*. Firstly, we have

Let $G-u={\bigcup}_{l=1}^{r}{G}_{l}$ and $\varrho ({G}^{\sigma}-u)=\varrho ({G}_{1}^{\sigma})={\lambda}_{0}$.

Case 1. $|V({G}_{1})|\le n-2$;

*G*induced by the vertex set $V({G}_{1})\cup u$. By induction assumption, we have

Case 2. $|V({G}_{1})|=n-1$;

*v*in $N(u)$. Thus,

*k*-cycle

*Z*(if there exists) containing

*u*, we have

where $q>0$. Thus the result holds.

*Z*(if it exists) containing

*e*, we have

Furthermore, we know $\varrho ({G}^{\sigma})>\varrho ({G}^{\sigma}-e)$. □

Moreover, by this theorem, we have the following

**Corollary 3.1**

*Let*

*G*

*be a connected simple graph of order*

*n*, ${G}^{\sigma}\in \mathfrak{D}(G)$, ${V}^{\prime}\subset V(G)$, ${E}^{\prime}\subset E(G)$.

*If each even cycle of*

*G*

*is oriented uniformly in*${G}^{\sigma}$,

*then*

- (a)
$\varrho ({G}^{\sigma})>\varrho ({G}^{\sigma}-{V}^{\prime})$.

- (b)
${i}^{-n}\varphi ({G}^{\sigma}-{E}^{\prime},\varrho i)>{i}^{-n}\varphi ({G}^{\sigma},\varrho i)$

*for*$\varrho \ge \varrho ({G}^{\sigma}-{E}^{\prime})$.*In particular*,*we have*$\varrho ({G}^{\sigma})>\varrho ({G}^{\sigma}-{E}^{\prime})$.

**Remark 3.1** Note that the above results will help us to compare the skew-spectral radii of two oriented graphs and then to clarify the skew-spectral properties of some oriented graphic classes. In the next section, we will just give such an example.

## 4 An application of the previous results

A so-called unicyclic graph is a connected simple graph in which the number of edges equals the number of vertices. In this section, we will give an application of the previous results - to obtain a sharp upper bound of the skew-spectral radius of an oriented unicyclic graph.

Also, we denoted by ${C}_{k}^{n-k}$ the unicyclic graph obtained from ${C}_{k}$ by joining a vertex of ${C}_{k}$ with $(n-k){K}_{1}$.

Regarding the adjacency spectral radii of unicyclic graphs, Hong [13] obtained the following results.

**Lemma 4.1** ([13])

*Let*$G\in {\mathcal{U}}_{n}(k)$.

*Then*

*with equality if and only if* $G\cong {C}_{k}^{n-k}$.

**Lemma 4.2** ([13])

$\rho ({C}_{k}^{n-k})>\rho ({C}_{k+1}^{n-k-1})$, $3\le k\le n-1$.

Now, we begin to discuss the skew-spectral radii of unicyclic graphs. Firstly, we consider the bipartite unicyclic graphs. For this case, we have

**Lemma 4.3**

*Let*$G\in {\mathcal{U}}_{n}(2k)$

*and*${G}^{\sigma}\in \mathfrak{D}(G)$.

*Then*

*The equality holds if and only if* $G\cong {C}_{4}^{n-4}$ *and the sign of the cycle in* ${G}^{\sigma}$ *is positive*.

*Proof* By Lemma 2.1, we know $\varrho ({G}^{\sigma})\le \rho (G)$ and the equality holds if and only if the sign of the 2*k*-cycle in ${G}^{\sigma}$ is ${(-1)}^{k}$. Moreover, by Lemmas 4.1 and 4.2, $\rho (G)\le \rho ({C}_{4}^{n-4})$ and the equality holds if and only if $G\cong {C}_{4}^{n-4}$.

Thus, $\varrho ({G}^{\sigma})\le \rho ({C}_{4}^{n-4})$ with equality if and only if $G\cong {C}_{4}^{n-4}$ and the sign of the cycle in ${G}^{\sigma}$ is ${(-1)}^{2}=1$. That is to say the sign of the cycle in ${G}^{\sigma}$ is positive.

This completes the proof. □

Now, we begin to consider the non-bipartite unicyclic graphs. Let $G\in {\mathcal{U}}_{n}(2k+1)$ and ${G}^{\sigma},{G}^{\tau}\in \mathfrak{D}(G)$. We can easily know that ${G}^{\sigma}\simeq {G}^{\tau}$ and ${S}_{p}({G}^{\sigma})={S}_{p}({G}^{\tau})$. Moreover, we have

**Lemma 4.4**

*Let*$G\in {\mathcal{U}}_{n}(2k+1)$

*and*${G}^{\sigma}\in \mathfrak{D}(G)$.

*Then*${i}^{-n}\varphi ({G}^{\sigma},\varrho i)\ge {i}^{-n}\varphi ({D}_{2k+1}^{n-2k-1},\varrho i)$

*for*$\varrho \ge \varrho ({G}^{\sigma})$.

*And thus*

*where* ${D}_{2k+1}^{n-2k-1}\in \mathfrak{D}({C}_{2k+1}^{n-2k-1})$. *The above two equalities hold if and only if* $G\cong {C}_{2k+1}^{n-2k-1}$.

*Proof* We prove the statement by induction on $n-2k-1$. Obviously, the result holds for $n-2k-1=0,1$.

Hence, the result follows. □

Furthermore, we have

**Lemma 4.5**

*Let*${D}_{2k+1}^{n-2k-1}\in \mathfrak{D}({C}_{2k+1}^{n-2k-1})$

*and*${D}_{2k-1}^{n-2k+1}\in \mathfrak{D}({C}_{2k-1}^{n-2k+1})$ ($n>5,k\ge 2$).

*Then for*$\varrho \ge \varrho ({D}_{2k+1}^{n-2k-1})$,

*we have*

*In particular*, $\varrho ({D}_{2k+1}^{n-2k-1})<\varrho ({D}_{2k-1}^{n-2k+1})$.

*Proof* We prove the result by induction on *n*. Obviously, the result holds for $n=6$.

*n*. We consider the case that the number of vertices is

*n*. By Theorem 3.1, we know

Thus, similar to the proof of Lemma 4.4, we know the result holds. □

Hence, for the non-bipartite unicyclic graphs, we can give

**Lemma 4.6**

*Let*$G\in {\mathcal{U}}_{n}(2k+1)$

*and*${G}^{\sigma}\in \mathfrak{D}(G)$, ${D}_{3}^{n-3}\in \mathfrak{D}({C}_{3}^{n-3})$.

*Then*

*The equality holds if and only if* $G\cong {C}_{3}^{n-3}$.

*Proof* By Lemmas 4.4 and 4.5, we know $\varrho ({G}^{\sigma})\le \varrho ({D}_{3}^{n-3})$, and the equality holds if and only if $G\cong {C}_{3}^{n-3}$.

Thus the result holds. □

Now, it is sufficient to compare the skew-spectral radius of ${D}_{3}^{n-3}$ and the adjacency spectral radius of ${C}_{4}^{n-4}$.

**Lemma 4.7** *If* $n>5$, *then* $\rho ({C}_{4}^{n-4})<\varrho ({D}_{3}^{n-3})$.

*Proof*By Lemmas 4.3 and 4.6, we have

It is easy to know that $\rho ({C}_{4}^{n-4})<\varrho ({D}_{3}^{n-3})$ when $n>5$. □

Finally, by the above results, we can get the main result of this section.

**Theorem 4.1**

*Let*

*G*

*be a unicyclic graph of order*

*n*, ${G}^{\sigma}\in \mathfrak{D}(G)$.

*If*$n>5$,

*then*

*The equality holds if and only if* $G\cong {C}_{3}^{n-3}$.

*Proof* By Lemmas 4.3, 4.6 and 4.7, we know the result holds. □

**Remark 4.1** For the adjacency spectral radius of a unicyclic graph, it is well known that $\rho (G)<\rho ({C}_{3}^{n-3})$ for $G\ncong {C}_{3}^{n-3}$. It is interesting that for the skew-spectral radius of an oriented unicyclic graph, we also have $\varrho ({G}^{\sigma})<\varrho ({D}_{3}^{n-3})$ ($n>5$) for $G\ncong {C}_{3}^{n-3}$. But for $n=4,5$, the inequality does not hold. In fact, by making a simple comparison, we know that $\varrho ({D}_{3}^{1})<\rho ({C}_{4})$ and $\varrho ({D}_{3}^{2})=\rho ({C}_{4}^{1})$.

## Declarations

### Acknowledgements

This work was supported by the National Natural Science Foundation of China (No. 11171373).

## Authors’ Affiliations

## References

- Cvetković D, Doob M, Sachs H:
*Spectra of Graphs*. Academic Press, New York; 1980.MATHGoogle Scholar - Godsil CD:
*Algebraic Combinatorics*. Chapmam & Hall, London; 1993.MATHGoogle Scholar - Horn R, Johnson C:
*Matrix Analysis*. Cambridge University Press, Cambridge; 1989.Google Scholar - Tutte WT: The factorization of linear graphs.
*J. Lond. Math. Soc.*1947, 22: 107–111. 10.1112/jlms/s1-22.2.107MathSciNetView ArticleMATHGoogle Scholar - Lovász L, Plummer M Ann. of Discrete Math. 29. In
*Matching Theory*. North-Holland, New York; 1988.Google Scholar - Robertson N, Seymour PD, Thomas R: Permanents, Pfaffian orientations and even directed circuits.
*Ann. Math.*1999, 150: 929–975. 10.2307/121059MathSciNetView ArticleMATHGoogle Scholar - Yan WG, Zhang FJ: Enumeration of perfect matchings of a type of Cartesian products of graphs.
*Discrete Appl. Math.*2006, 154: 145–157. 10.1016/j.dam.2005.07.001MathSciNetView ArticleMATHGoogle Scholar - IMA-ISU research group on minimum rank: Minimum rank of skew-symmetric matrices described by a graph.
*Linear Algebra Appl.*2010, 432: 2457–2472. 10.1016/j.laa.2009.10.001MathSciNetView ArticleGoogle Scholar - Shader B, So WS: Skew spectra of oriented graphs.
*Electron. J. Comb.*2009., 16: Article ID N32Google Scholar - Adiga C, Balakrishnan R, So WS: The skew energy of a graph.
*Linear Algebra Appl.*2010, 432: 1825–1835. 10.1016/j.laa.2009.11.034MathSciNetView ArticleMATHGoogle Scholar - Gong SC, Xu GH: The characteristic polynomial and the matchings polynomial of a weighted oriented graph.
*Linear Algebra Appl.*2012, 436: 3597–3607. 10.1016/j.laa.2011.12.033MathSciNetView ArticleMATHGoogle Scholar - Hou YP, Lei TG: Characteristic polynomials of skew-adjacency matrices of oriented graphs.
*Electron. J. Comb.*2011., 18: Article ID P156Google Scholar - Hong Y: Bounds on the spectra of unicyclic graphs.
*J. East China Norm. Univ.*1986, 1: 31–34. in ChineseMathSciNetMATHGoogle Scholar

## Copyright

This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.