Let \(G\) be a simple graph with vertex set \(V=\{v_1,v_2,\ldots,v_n \}\), and let \(d_i\) be the degree of the vertex \(v_i\). The Sombor matrix of \(G\) is the square matrix \(\mathbf A_{SO}\) of order \(n\), whose \((i,j)\)-element is \(\sqrt{d_i^2+d_j^2}\) if \(v_i\) and \(v_j\) are adjacent, and zero otherwise. We study the characteristic polynomial, spectrum, and energy of \(\mathbf A_{SO}\). A few results for the coefficients of the characteristic polynomial, and bounds for the energy of \(\mathbf A_{SO}\) are established.
The Sombor index \(SO\) is a recently introduced vertex-degree-based topological index [1]. It promptly attracted much attention and its mathematical properties and chemical applications became a topic of a remarkably large number of studies, e.g., [2,3,4,5,6,7,8,9]. Also promptly, the concept of Sombor index was extended to linear algebra, by defining the Sombor matrix, which then led to the investigation of its spectrum and various spectrum-based properties [10,11,12,13,14]. In particular, the energy of the Sombor matrix was much examined [11,12,13,14]. In the present paper we report a few additional results on this matter, with emphasis on the characteristic polynomial and energy.
In this paper, we considered simple, finite, undirected, and connected graphs. Let \(G\) be such a graph, with vertex set \(\mathbf V(G)\) and edge set \(\mathbf E(G)\). If two vertices have a common edge then they are said to be adjacent. If the vertices \(u\) and \(v\) are adjacent, then the edge connecting them is denoted by \(uv\). The number of edges incident to a vertex \(v\) is called the degree of that vertex \(v\), and is denoted by \(d_v\).
In the mathematical and chemical literature, a great number of vertex-degree-based graph invariants of the form
The adjacency matrix \(\mathbf A(G)=(a_{ij})_{n \times n}\) of the graph \(G\) with vertex set \(\mathbf V(G)=\{v_1,v_2,\dots,v_n\}\), is the symmetric matrix of order \(n\), whose elements are defined as [18]:
The energy of the graph \(G\) is defined as [19]:
\[ En(G) = \sum_{i=1}^n |\lambda_i|\,. \] The theory of graph spectra, including the theory of graph energy, is nowadays a well elaborated part of discrete mathematics. In parallel with the above specified graph-spectral concepts, we now introduce their Sombor-index-related counterparts. The following definition is an application to the Sombor index of the general spectral theory of matrices associated with vertex-degree-based topological indices of the form (1) [20,21,22].Definition 1.
Remark 1. Comparing Equations (2) and (3), we see that the Sombor matrix can be viewed as the ordinary adjacency matrix of a graph with weighted edges, such that the weight of the edge \(v_iv_j\) is \(\sqrt{d_{v_i}^2+d_{v_j}^2}\). This observation allows us to apply to the Sombor matrix and its spectrum the standard methods of graph spectral theory [18], in particular the Sachs coefficient theorem [23].
Lemma 1. Let \(G\) be a graph with Sombor eigenvalues \(\sigma_1, \sigma_2,\ldots,\sigma_n\). Then \begin{align*} \sum_{i=1}^n \sigma_i = & 0,\;\;\;\;\;\;\;\;\;\;\;\;\;\;\; \nonumber \end{align*}
Lemma 2. Let \(p\) be the size of smallest odd cycle contained in the graph \(G\), and let \(\sum_{C_p}\) indicate summation over all cycles of size \(p\) contained in \(G\). Then for \(q=1,3,\ldots,p-2\),
Proof. Take into account Remark 1, and use the analogous result for ordinary graphs [18].
Lemma 3. [24,25] Suppose that \(a_i\) and \(b_i\) are non negative real numbers for \(1\leq i \leq n\). Then, \[ \left( \sum_{i=1}^n a_i^2\right) \left( \sum_{i=1}^n b_i^2\right) \leq \dfrac{1}{4}\left( \sqrt{\dfrac{M_1\,M_2}{m_1\,m_2}} + \sqrt{\dfrac{m_1\,m_2}{M_1\,M_2}}\right)^2 \left(\sum_{i=1}^n a_i\,b_i \right)^2 \] where \(M_1=\max\limits_{1\leq i \leq n} a_i\)  ,\(M_2=\max\limits_{1\leq i \leq n} b_i\),   \(m_1=\min\limits_{1\leq i \leq n}a_i\) ,  and \(m_2=\min\limits_{1\leq i \leq n}b_i\) .
Lemma 4. [24,25] Using the same notation as in Lemma 3, \[ \left( \sum_{i=1}^n a_i^2\right) \left( \sum_{i=1}^n b_i^2\right) – \left(\sum_{i=1}^n a_i\,b_i \right)^2 \leq \dfrac{n^2}{4}\left( M_1\,M_2-m_1\,m_2\right)^2\,. \]
We first recall a result by Lin and Miao [13], that can be stated in terms of traces of the Sombor matrix. It should be compared with the below Theorem 2. The upper bound was obtained also in [12]. Note that \(tr(\mathbf A_{SO}(G)^2)=2F(G)\) follows from Equation (5).
Theorem 1. [13] Denote the trace of a square matrix \(\mathbf M\) by \(tr(\mathbf M)\). Let \(G\) be a graph on \(n\) vertices. Then \[ \sqrt{tr(\mathbf A_{SO}(G)^2)} \leq En_{SO}(G) \leq \sqrt{n\,tr(\mathbf A_{SO}(G)^2)} \] i.e., \[ \sqrt{2F(G)} \leq En_{SO}(G) \leq \sqrt{2n\,F(G)}\,. \]
Theorem 2. Let \(G\) be a non-trivial graph. Then \[ En_{SO}(G) \geq \sqrt{\dfrac{\big[tr(\mathbf A_{SO}(G)^2)\big]^3}{tr(\mathbf A_{SO}(G)^4)}}\,. \]
Proof. By the Hölder inequality, \[ \sum_{i=1}^n a_i\,b_i \leq \left( \sum_{i=1}^n a_i^p \right)^{1/p} \left( \sum_{i=1}^n b_i^q \right)^{1/q} \] where, \(a_i,b_i \in \mathbf{R}^+\), \((i=1,2,3\dots,n)\). Setting \(a_i=|\sigma_i|^{2/3}\), \(b_i=|\sigma_i|^{4/3}\), \(p=3/2\), and \(q=3\), we get \[ \sum_{i=1}^n |\sigma_i|^2 \leq \left( \sum_{i=1}^n |\sigma_i| \right)^{2/3} \left(\sum_{i=1}^n |\sigma_i|^4 \right)^{1/3} \] which by Equation (4), and bearing in mind that since \(G\) is not an empty graph and thus \(\sum_{i=1}^n|\sigma_i|^4\neq0\), yields Theorem 2.
Theorem 3. If \(\sigma_1\) is the greatest Sombor eigenvalue, then \(En_{SO}(G) \leq 2\sigma_1\). For connected graphs, equality holds if and only if \(G\) is a complete bipartite graph.
Proof. Bearing in mind Equation (4), \[ En_{SO}(G) = |\sigma_1|+\sum_{i=2}^n |\sigma_i| \geq |\sigma_1|+ \left| \sum_{i=2}^n \sigma_i\right|\,. \] On the other hand, \[ \sum_{i=1}^n \sigma_i=0 \ \ \ \ \implies \ \ \ \ \sigma_1=-\sum_{i=2}^n \sigma_i \ \ \ \ \text{and so} \ \ \ \ |\sigma_1|=\left| \sum_{i=2}^n \sigma_i\right|\,. \] Equality holds if \(\sigma_1\) and \(\sigma_n\) are the only non-zero eigenvalues. In view of Remark 1, this happens only if \(G\) is a complete bipartite graph.
Theorem 4. Let \(G\) be a graph with \(n\) vertices. If no Sombor eigenvalue of \(G\) is equal to zero, then \[ En_{SO}(G) \geq \sqrt{\dfrac{8n\,F(G)\,\sigma_\ell\,\sigma_s}{|\sigma_\ell|+|\sigma_s|}} \] where, \(|\sigma_\ell|\) and \(|\sigma_s|\) are, respectively, the largest and smallest absolute values of the eigenvalues in the Sombor spectrum of \(G\). Of course, \(|\sigma_\ell|=\sigma_1\).
Proof. Setting in Lemma 3, \(a_i=|\sigma_i|\) and \(b_i=1\) for \(1\leq i \leq n\), we get \[ \left(\sum_{i=1}^n |\sigma_i|^2 \right) \left( \sum_{i=1}^n 1\right) \leq \dfrac{1}{4}\left( \sqrt{\dfrac{|\sigma_\ell|}{|\sigma_s|}} + \sqrt{\dfrac{|\sigma_s|}{|\sigma_\ell|}}\right)^2 \left(\sum_{i=1}^n |\sigma_i| \right)^2, \] where \(|\sigma_\ell| = \max\limits_{1\leq i \leq n}\{|\sigma_i|\}\) and \(|\sigma_s| = \min\limits_{1\leq i \leq n}\{|\sigma_i|\}\). Then \[ 2F(G)\,n \leq \dfrac{1}{4}\left( \sqrt{\dfrac{|\sigma_\ell|}{|\sigma_s|}} + \sqrt{\dfrac{|\sigma_s|}{|\sigma_\ell|}}\right)^2 \left(\sum_{i=1}^n |\sigma_i| \right)^2 \] and thus \[ \sqrt{8n\,F(G)} \leq \left(\dfrac{|\sigma_\ell|+|\sigma_s|}{\sqrt{\sigma_\ell\,\sigma_s}} \right) En_{SO}(G) \] which straightforwardly leads to Theorem 4.
Theorem 5. Let \(G\) be a connected graph with \(n\) vertices, and \(\sigma_\ell\,,\,\sigma_s\) same as in Theorem 4. Then \[ En_{SO}(G)\geq \sqrt{2n\,F(G)-\dfrac{n^2}{4}(|\sigma_\ell|-|\sigma_s|)} \]
Proof. Setting in Lemma 4, \(a_i=|\sigma_i|\) and \(b_i=1\) for \(1\leq i \leq n\), we get \[ \left( \sum_{i=1}^{n} |\sigma_i|^2\right) \left( \sum_{i=1}^{n} 1\right)-\left(\sum_{i=1}^{n} |\sigma_i| \right)^2 \leq \dfrac{n^2}{4}\left( |\sigma_\ell|-|\sigma_s|\right)^2, \] implying \[ 2F(G)\,n -En_{SO}(G)^2 \leq \dfrac{n^2}{4}(|\sigma_\ell|-|\sigma_s|)^2\,. \] Theorem 5 follows.
As explained in Remark 1, the matrix \(\mathbf A_{SO}(G)\) can be viewed as the adjacency matrix of a graph with weighted edges. This, of course, applies also to trees.
According to the Sachs coefficient theorem [18,23], for the Sombor characteristic polynomial of a tree \(T\), an expression analogous to Equation (8) would hold, namely
The weight of a single matching \(M\) is equal to \(\prod\limits_{uv \in M} \big(d_u^2+d_v^2 \big)\) and therefore
We thus see that the coefficients \(m_{SO}(T,k)\) are positive if \(m(T,k)>0\) and are equal to zero if \(m(T,k)=0\). This implies:
Theorem 6. The inertia of the Sombor matrix and of the ordinary adjacency matrix of any tree coincide.
The energy of a tree can be computed from its matching polynomial as [28]:Since, evidently, \(m_{SO}(T,k) > m(T,k)\) holds whenever the tree \(T\) has at least one \(k\)-matching, by comparing Equations (11) and (12), we immediately arrive at:
Theorem 7. For any tree \(T\), \(En_{SO}(T) > En(T)\).