Abstract
In this paper, one of our main purposes is to prove the boundedness of the solution set of tensor complementarity problems such that the specific bounds depend only on the structural properties of such a tensor. To achieve this purpose, firstly, we prove that this class of structured tensors is strictly semi-positive. Subsequently, the strictly lower and upper bounds of operator norms are given for two positively homogeneous operators. Finally, with the help of the above upper bounds, we show that the solution set of tensor complementarity problems has the strictly lower bound. Furthermore, the upper bounds of spectral radius are obtained, which depends only on the principal diagonal entries of tensors.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
As a natural extension of linear complementarity problem, the tensor complementarity problem is a new topic emerged from the tensor community. Meanwhile, such a problem is a special type of nonlinear complementarity problems. So the tensor complementarity problem seems to have similar properties to the linear complementarity problem, and to have its particular properties other than ones of the classical nonlinear complementarity problem, and to have some nice properties that depended on itself special structure. The notion of the tensor complementarity problem was used firstly by Song and Qi [1, 2]. Recently, Huang and Qi [3] formulated a multiplayer non-cooperative game as a tensor complementarity problem and showed that a Nash equilibrium point of the multilinear game is equivalent to a solution of the tensor complementarity problem. By using specially structured properties of tensors, the properties of the tensor complementarity problem have been well studied in the literatures. For example, see Song and Yu [4], Song and Qi [5] for strictly semi-positive tensors, Gowda et al. [6], Luo et al. [7] for Z-tensors, Ding et al. [8] for P-tensors, Wang et al. [9] for exceptionally regular tensors, Bai et al. [10] for strong P-tensors, Che et al. [11] for some special tensors, Huang et al. [12] for Q-tensors. Song and Qi [13], Ling et al. [14, 15], Chen et al. [16] studied the tensor eigenvalue complementarity problems.
In past several decades, numerous mathematical works concerned with error bound analysis for the solution of linear complementarity problem by means of the special structure of the matrices. For more details, see [17,18,19,20,21,22]. Recently, motivated by the study on error bounds for linear complementarity problems, Song and Yu [4] and Song and Qi [23] extended the error bounds results of the linear complementarity problems to the tensor complementarity problems with strictly semi-positive tensors. However, there are relatively few works in the specific upper or lower bounds of the tensor complementarity problems, which is a weak link in this topics.
In this paper, we will give the boundedness of solution set of tensor complementarity problem with B-tensors. Moreover, we will present the specific lower bounds of such a problem, which depend only upon the structural properties of B-tensors.
To achieve the above goal, we need study the structured properties of B-tensors. Nowadays, miscellaneous structured tensors have been widely studied (Qi and Luo [24]), which is one of hot research topics. For more detail, see Zhang et al. [25] and Ding et al. [26] for M-tensors, Song and Qi [1] for P-tensors and B-tensors, Li and Li [27] for double B-tensors, Song and Qi [28, 29] and Mei and Song [30] for Hilbert tensors. Recently, the concept of B-tensors was first used by Song and Qi [1]. They gave many nice structured properties which are similar to ones of B-matrices. For more nice properties and applications of B-matrices, see Peña [31, 32]. It is well known that each B-matrix is a P-matrix. However, the same conclusion only holds for even-order symmetric B-tensors [33]. Qi and Song [34] showed that an even-order symmetric B-tensor is positive definite. Yuan and You [33] proved that a non-symmetric B-tensor is not P-tensor in general. So there are many special properties of B-tensors for further and serious consideration.
In Sect. 3, we prove that each B-tensor is strictly semi-positive. So, the solution set of tensor complementarity problem with B-tensor is bounded. In order to presenting the specific lower bounds of such a problem, in Sect. 4, we give the strictly lower and upper bounds of the norms for two positively homogeneous operators induced by B-tensors. By means of the above upper bounds, we establish the strictly lower bound of solution set of tensor complementarity problem with B-tensors. Furthermore, we achieve our another objective with the help of upper bounds of operator norms. That is, we obtain the upper bounds of spectral radius and E-spectral radius of B-tensor, which depends only on the principal diagonal entries of tensors.
2 Preliminaries and Basic Facts
An m-order n-dimensional tensor (hypermatrix) \(\mathcal {A} = (a_{i_1\cdots i_m})\) is a multi-array of real entries \(a_{i_1\cdots i_m}\in \mathbb {R}\), where \(i_j \in [n]=\{1,2,\ldots ,n\}\) for \(j \in [m]=\{1,2,\ldots ,m\}\). The set of all m-order n-dimensional tensors is denoted by \(T_{m,n}\). If the entries \(a_{i_1\cdots i_m}\) are invariant under any permutation of their indices, then \(\mathcal {A}\) is called a symmetric tensor, denoted by \(S_{m,n}\). Let \(\mathcal {A} = (a_{i_1\cdots i_m})\in T_{m,n}\) and a vector \(x=(x_{1},x_{2},\ldots ,x_{n})^\top \in \mathbb {R}^{n}\). Then \(\mathcal {A}x^{m-1}\) is a vector with its ith component defined by
and \(\mathcal {A}x^{m}\) is a homogeneous polynomial of degree m, defined by
For any \(q\in \mathbb {R}^{n}\), the tensor complementarity problem, denoted by \(TCP(\mathcal {A},q)\), is to find \(x\in \mathbb {R}^{n}\) such that
or to show that no such vector exists.
An n-dimensional B-matrix \(B=(b_{ij})\) is a square real matrix with its entries satisfying that for all \(i\in [n]\)
As a natural extension of B-matrices. Song and Qi [1] gave the definitions of B-tensors and B\(_0\)-tensors.
Definition 2.1
Let \(\mathcal {B}=(b_{i_{1}\cdots i_{m}})\in T_{m,n}\). Then \(\mathcal {B}\) is said to be
-
(i)
a B-tensor iff for all \(i\in [n],\mathop {\sum }\limits _{i_{2},\ldots ,i_{m}=1}^{n}b_{ii_{2}i_{3}\cdots i_{m}}>0\) and
$$\begin{aligned} \frac{1}{n^{m-1}}\left( \sum _{i_{2},\ldots ,i_{m}=1}^{n}b_{ii_{2}i_{3}\cdots i_{m}}\right) >b_{ij_{2}j_{3}\cdots j_{m}}, \hbox { for all } (j_{2},j_{3},\ldots ,j_{m})\ne (i,i,\ldots ,i); \end{aligned}$$ -
(ii)
a B\(_{0}\)-tensor iff for all \(i\in [n],\mathop {\sum }\limits _{i_{2},\ldots ,i_{m}=1}^{n}b_{ii_{2}i_{3}\cdots i_{m}}\ge 0\) and
$$\begin{aligned} \frac{1}{n^{m-1}}\left( \sum _{i_{2},\ldots ,i_{m}=1}^{n}b_{ii_{2}i_{3}\cdots i_{m}}\right) \ge b_{ij_{2}j_{3}\cdots j_{m}}, \hbox { for all } (j_{2},j_{3},\ldots ,j_{m})\ne (i,i,\ldots ,i). \end{aligned}$$
For each \(i\in [n]\), let
and
Lemma 2.1
(Song and Qi [1, Theorems 5.1, 5.2, 5.3]) Let \(\mathcal {B}=(b_{i_{1}\cdots i_{m}})\in T_{m,n}\). If \(\mathcal {B}\) is a B-tensor, then, for each \(i\in [n]\),
-
(i)
\( b_{ii\cdots i}>|b_{ij_{2}j_{3}\cdots j_{m}}| for all (j_{2},j_{3},\ldots ,j_{m})\ne (i,i,\ldots ,i), \ j_{2},j_{3},\ldots ,j_{m}\in [n]\);
-
(ii)
\(\mathop {\sum }\limits _{i_2,\ldots , i_m=1}^{n} b_{ii_2i_3\cdots i_m}>n^{m-1}\beta _i(\mathcal {B})\);
-
(iii)
\(b_{ii\cdots i}>\mathop {\sum }\limits _{b_{ii_2\cdots i_m}<0} |b_{ii_2i_3\cdots i_m}|\).
If \(\mathcal {B}\) is a B\(_0\)-tensor, then, the above three inequalities hold with “>” being replaced by “\(\ge \)”.
The concepts of tensor eigenvalues were introduced by Qi [35, 36] to the higher-order symmetric tensors, and the existence of the eigenvalues and some applications were studied there. Lim [37] independently introduced the concept of real tensor eigenvalues and obtained some existence results using a variational approach.
Definition 2.2
Let \(\mathcal {A}=(a_{i_{1}\cdots i_{m}})\in T_{m,n}\).
-
(i)
A number \(\lambda \in \mathbb {C}\) is called an eigenvalue of \(\mathcal {A}\) iff there is a nonzero vector x such that
$$\begin{aligned} \mathcal {A}x^{m-1}=\lambda x^{[m-1]}. \end{aligned}$$(3)and x is called an eigenvector of \(\mathcal {A}\) associated with \(\lambda \). Such an eigenvalue is called H-eigenvalue if it is real and has a real eigenvector x, and such a real eigenvector x is called H-eigenvector.
-
(ii)
A number \(\mu \in \mathbb {C}\) is said to be an E-eigenvalue of \(\mathcal {A}\) iff there exists a nonzero vector x such that
$$\begin{aligned} \mathcal {A}x^{m-1}=\mu x(x^{T}x)^{\frac{m-2}{2}}, \end{aligned}$$(4)and x is called an E-eigenvector of \(\mathcal {A}\) associated with \(\mu \). It is clear that if x is real, then, \(\mu \) is also real. In this case, \(\mu \) and x are called a Z-eigenvalue of \(\mathcal {A}\) and a Z-eigenvector of \(\mathcal {A}\), respectively.
We now give the definitions of (strictly) semi-positive tensors (Song and Qi [2, 5]).
Definition 2.3
Let \(\mathcal {A} = (a_{i_1\cdots i_m}) \in T_{m, n}\). \(\mathcal {A}\) is said to be
-
(i)
semi-positive iff for each \(x\ge 0\) and \(x\ne 0\), there exists an index \(k\in [n]\) such that
$$\begin{aligned} x_k>0\hbox { and }\left( \mathcal {A} x^{m-1}\right) _k\ge 0; \end{aligned}$$ -
(ii)
strictly semi-positive iff for each \(x\ge 0\) and \(x\ne 0\), there exists an index \(k\in [n]\) such that
$$\begin{aligned} x_k>0\hbox { and }\left( \mathcal {A} x^{m-1}\right) _k>0. \end{aligned}$$
For \(x\in \mathbb {R}^{n}\), it is well known that
are two main norms defined on \(\mathbb {R}^{n}\). Then, for a continuous, positively homogeneous operator \(T:\mathbb {R}^{n}\rightarrow \mathbb {R}^{n}\), it is obvious that
are two operator norms of T. For \(\mathcal {A}\in T_{m,n}\), we may define a continuous, positively homogeneous operator \(T_{\mathcal {A}}:\mathbb {R}^{n}\rightarrow \mathbb {R}^{n}\) by
When m is even, we may define another continuous, positively homogeneous operator \(F_\mathcal {A} : \mathbb {R}^n \rightarrow \mathbb {R}^n\) by for any \(x \in \mathbb {R}^n\),
The following upper bounds and properties of the operator norm were established by Song and Qi [23, 38].
Lemma 2.2
(Song and Qi [38, Theorem 4.3] and [23, Lemma 3, Lemma 4]) Let \(\mathcal {A}=(a_{i_{1}\cdots i_{m}})\in T_{m,n}\). Then,
-
(i)
\(\Vert T_{\mathcal {A}}\Vert _{\infty }\le \mathop {\max }\limits _{i\in [n]}\mathop {\sum }\limits _{i_{2},\ldots ,i_{m}=1}^{n}|a_{ii_{2}\cdots i_{m}}|;\)
-
(ii)
\(\Vert F_{\mathcal {A}}\Vert _{\infty }\le \mathop {\max }\limits _{i\in [n]}\left( \mathop {\sum }\limits _{i_{2},\ldots ,i_{m}=1}^{n}|a_{ii_{2}\cdots i_{m}}|\right) ^{\frac{1}{m-1}}\) if m is even;
-
(iii)
\(\Vert T_{\mathcal {A}}\Vert _{p}\le n^{\frac{m-2}{p}}\left( \mathop {\sum }\limits _{i=1}^{n}\left( \mathop {\sum }\limits _{i_{2},\ldots ,i_{m}=1}^{n}|a_{ii_{2}\cdots i_{m}}|\right) ^{p}\right) ^{\frac{1}{p}};\)
-
(iv)
\(\Vert F_{\mathcal {A}}\Vert _{p} \le \left( \mathop {\sum }\limits _{i=1}^{n} \left( \mathop {\sum }\limits _{i_{2},\ldots ,i_{m}=1}^{n}|a_{ii_{2}\cdots i_{m}}|\right) ^{\frac{p}{m-1}}\right) ^{\frac{1}{p}}\) if m is even.
3 The Strictly Semi-Positivity of B-Tensors
Theorem 3.1
Let \(\mathcal {B}\) be an m-order n-dimensional B-tensor. Then, \(\mathcal {B}\) is strictly semi-positive.
Proof
Suppose that \(\mathcal {B}\) is not strictly semi-positive. Then, there exists \(x\ge 0\) and \(x\ne 0\), for all \(k\in [n]\) with \(x_k>0\) such that
Choose \(x_i>0\) with \(x_i\ge x_k\) for all \(k\in [n]\). Clearly, \(\left( \mathcal {B} x^{m-1}\right) _i\le 0.\) Then, we have
By Lemma 2.1 (iii), we obtain
a contradiction. Consequently, \(\mathcal {B}\) must be strictly semi-positive. \(\square \)
Using the similar proof technique, it is easy to prove the following conclusion.
Theorem 3.2
Let \(\mathcal {B}\) be an m-order n-dimensional B\(_0\)-tensor. Then, \(\mathcal {B}\) is semi-positive.
Clearly, by Theorem 3.2 of Song and Yu [4] and Corollary 3.3 of Song and Qi [2], we easily obtain the following.
Corollary 3.3
Let \(\mathcal {B}\) be an m-order n-dimensional B-tensor. Then, for each \(q \in \mathbb {R}^n\), the tensor complementarity problem \(TCP(\mathcal {B}, q)\) has always a solution. Furthermore, the solution set of the \(TCP(\mathcal {B}, q)\) is bounded for each \(q \in \mathbb {R}^n\).
4 Boundedness About B-Tensors
We now present the upper and lower bounds of the operator norm associated with a B-tensor.
Theorem 4.1
Let \(\mathcal {B}\) be an m-order n-dimensional B tensor. Then,
-
(i)
\( n^{\frac{m}{2}}\max _{i\in [n]}\beta _i(\mathcal {B})<n^{\frac{2-m}{2}}\mathop {\max }\limits _{i\in [n]}\mathop {\sum }\limits _{i_{2},\ldots ,i_{m}=1}^{n}b_{ii_{2}\cdots i_{m}}\le \Vert T_{\mathcal {B}}\Vert _{\infty }< n^{\frac{m}{2}}\max \limits _{i\in [n]}b_{ii\cdots i}\);
-
(ii)
\(n^{\frac{mp-2}{2p}} \left( \mathop {\sum }\limits _{i=1}^{n}\left( \beta _i(\mathcal {B})\right) ^p\right) ^\frac{1}{p}<n^{\frac{2p-pm-2}{2p}}\left( \mathop {\sum }\limits _{i=1}^{n} \left( \mathop {\sum }\limits _{i_{2},\ldots ,i_{m}=1}^{n}b_{ii_{2}\cdots i_{m}}\right) ^p\right) ^\frac{1}{p}\le \Vert T_{\mathcal {B}}\Vert _{p}< n^{\frac{mp-2}{2p}}(\mathop {\sum }\limits _{i=1}^{n}b_{ii\cdots i}^{p})^{\frac{1}{p}}\) if \(p\ge 1,\)
where \(\beta _i(\mathcal {B})\) is defined by (2).
Proof
(i) Choose \(e=(1,1,\ldots ,1)^\top \). Then, \(\Vert e\Vert _\infty =1\) and \(\Vert e\Vert _2=n^{\frac{1}{2}}\), and hence, by Lemma 2.1 (ii), we have
Consequently,
Now we show the right inequality. It follows from Lemma 2.1 (i) together with the fact that \(\Vert x\Vert _{1}\le n\Vert x\Vert _\infty \) and \(\Vert x\Vert _{2}\le \sqrt{n}\Vert x\Vert _{\infty }\) that
(ii) Choose \(y=(n^{-\frac{1}{p}},n^{-\frac{1}{p}},\ldots ,n^{-\frac{1}{p}})^\top \). Then, \(\Vert y\Vert _p=1\) and \(\Vert y\Vert _2=n^{\frac{p-2}{2p}}\), and hence, by Lemma 2.1 (ii), we have
Consequently,
Next we show the right inequality. By Lemma 2.1 (i) together with the fact that \(\Vert x\Vert _{q}\le \Vert x\Vert _{p}\le n^{\frac{1}{p}-\frac{1}{q}}\Vert x\Vert _{q} \hbox { for } q>p \), we obtain
The desired conclusions follow. \(\square \)
Theorem 4.2
Let \(\mathcal {B}\) be an m-order n-dimensional B-tensor. If m is even, then,
-
(i)
\(n\left( \beta _{i}(\mathcal {B})\right) ^{\frac{1}{m-1}}<\mathop {\max }\limits _{i\in [n]}\left( \mathop {\sum }\limits _{i_{2},\ldots ,i_{m}=1}^{n}b_{ii_{2}\cdots i_{m}}\right) ^{\frac{1}{m-1}}\le \Vert F_{\mathcal {B}}\Vert _{\infty }\le \mathop {\max }\limits _{i\in [n]}\left( \mathop {\sum }\limits _{i_{2},\ldots ,i_{m}=1}^{n}|b_{ii_{2}\cdots i_{m}}|\right) ^{\frac{1}{m-1}}<n\mathop {\max }\limits _{i\in [n]}b_{ii\cdots i}^{\frac{1}{m-1}}\);
-
(ii)
\(n^{\frac{p-1}{p}} \left( \mathop {\sum }\limits _{i=1}^{n}\left( \beta _i(\mathcal {B})\right) ^{\frac{p}{m-1}}\right) ^\frac{1}{p}<\frac{1}{\root p \of {n}}\left( \mathop {\sum }\limits _{i=1}^{n} \left( \mathop {\sum }\limits _{i_{2},\ldots ,i_{m}=1}^{n}b_{ii_{2}\cdots i_{m}}\right) ^{\frac{p}{m-1}}\right) ^{\frac{1}{p}}\le \Vert F_{\mathcal {B}}\Vert _{p}<n^{\frac{p-1}{p}} \left( \mathop {\sum }\limits _{i=1}^{n}b_{ii\cdots i}^{\frac{p}{m-1}}\right) ^\frac{1}{p}\) (\(p\ge 1).\)
Proof
-
(i)
Choose \(e=(1,1,\ldots ,1)^\top \). Then, \(\Vert e\Vert _\infty =1\), and hence, by Lemma 2.1 (ii), we have
$$\begin{aligned} \Vert F_{\mathcal {B}}(e)\Vert _{\infty }= & {} \max _{i\in [n]}\left| \sum _{i_{2},\ldots ,i_{m}=1}^{n}b_{ii_{2}\cdots i_{m}}\right| ^{\frac{1}{m-1}}\\= & {} \max _{i\in [n]}\left( \sum _{i_{2},\ldots ,i_{m}=1}^{n}b_{ii_{2}\cdots i_{m}}\right) ^{\frac{1}{m-1}}\\> & {} n \max _{i\in [n]}\left( \beta _i(\mathcal {B})\right) ^{\frac{1}{m-1}}. \end{aligned}$$So, by the definition of the operator norm, we have
$$\begin{aligned} n\max _{i\in [n]}\left( \beta _i(\mathcal {B})\right) ^{\frac{1}{m-1}}<\max _{i\in [n]}\left( \sum _{i_{2},\ldots ,i_{m}=1}^{n}b_{ii_{2}\cdots i_{m}}\right) ^{\frac{1}{m-1}}\le \Vert F_{\mathcal {B}}\Vert _{\infty }. \end{aligned}$$From Lemmas 2.2 (ii) and 2.1 (i), it follows that
$$\begin{aligned} \Vert F_{\mathcal {B}}\Vert _{\infty }\le \max _{i\in [n]}\left( \sum _{i_{2},\ldots ,i_{m}=1}^{n}|b_{ii_{2}\cdots i_{m}}|\right) ^{\frac{1}{m-1}}<\max _{i\in [n]}\left( \sum _{i_{2},\ldots ,i_{m}=1}^{n}b_{ii\cdots i}\right) ^{\frac{1}{m-1}}= n\max _{i\in [n]}b_{ii\cdots i}^{\frac{1}{m-1}}. \end{aligned}$$ -
(ii)
Choose \(y=(n^{-\frac{1}{p}},n^{-\frac{1}{p}},\ldots ,n^{-\frac{1}{p}})^\top \). Then, \(\Vert y\Vert _p=1\), and hence, by Lemma 2.1 (ii), we have
$$\begin{aligned} \Vert F_{\mathcal {B}}(y)\Vert _p^p= & {} \sum _{i=1}^{n}\left| \sum _{i_{2},\ldots ,i_{m}=1}^{n}b_{ii_{2}\cdots i_{m}}(n^{-\frac{1}{p}})^{m-1}\right| ^{\frac{p}{m-1}}\\= & {} \sum _{i=1}^{n} \frac{1}{n}\left( \sum _{i_{2},\ldots ,i_{m}=1}^{n}b_{ii_{2}\cdots i_{m}}\right) ^{\frac{p}{m-1}}\\> & {} \frac{1}{n}\sum _{i=1}^{n} \left( n^{m-1}\beta _i(\mathcal {B})\right) ^{\frac{p}{m-1}}\\= & {} n^{p-1} \sum _{i=1}^{n}\left( \beta _i(\mathcal {B})\right) ^{\frac{p}{m-1}}. \end{aligned}$$Consequently,
$$\begin{aligned} n^{\frac{p-1}{p}} \left( \mathop {\sum }\limits _{i=1}^{n}\left( \beta _i(\mathcal {B})\right) ^{\frac{p}{m-1}}\right) ^\frac{1}{p}<\frac{1}{\root p \of {n}}\left( \sum _{i=1}^{n} \left( \sum _{i_{2},\ldots ,i_{m}=1}^{n}b_{ii_{2}\cdots i_{m}}\right) ^{\frac{p}{m-1}}\right) ^{\frac{1}{p}}\le \Vert F_{\mathcal {B}}\Vert _{p}. \end{aligned}$$Next we show the right inequality. By Lemma 2.1 (i) together with the fact that \( \Vert x\Vert _1\le n^{1-\frac{1}{p}}\Vert x\Vert _p \hbox { for } p>1 \), we obtain
$$\begin{aligned} \Vert F_{\mathcal {B}}\Vert _{p}^{p}= & {} \max _{\Vert x\Vert _{p}=1}\Vert F_{\mathcal {B}}(x)\Vert _{p}^{p}= \max _{\Vert x\Vert _{p}=1}\sum _{i=1}^{n}\left| \sum _{i_{2},\ldots ,i_{m}=1}^{n}b_{ii_{2}\cdots i_{m}}x_{i_{2}}x_{i_{3}}\cdots x_{i_{m}}\right| ^{\frac{p}{m-1}}\\< & {} \max _{\Vert x\Vert _{p}=1}\sum _{i=1}^{n}\left( b_{ii\cdots i}\sum _{i_{2},\ldots ,i_{m}=1}^{n}|x_{i_{2}}||x_{i_{3}}|\cdots |x_{i_{m}}|\right) ^{\frac{p}{m-1}}\\= & {} \max _{\Vert x\Vert _{p}=1}\sum _{i=1}^{n}\left( b_{ii\cdots i}\left( \sum _{k=1}^{n}|x_{k}|\right) ^{m-1}\right) ^{\frac{p}{m-1}}= \max _{\Vert x\Vert _{p}=1}\Vert x\Vert _1^{p}\sum _{i=1}^{n}b_{ii\cdots i}^{\frac{p}{m-1}}\\\le & {} \max _{\Vert x\Vert _{p}=1}\left( n^{1-\frac{1}{p}}\Vert x\Vert _p\right) ^p\sum _{i=1}^{n}b_{ii\cdots i}^{\frac{p}{m-1}}\\= & {} n^{p-1}\sum _{i=1}^{n}b_{ii\cdots i}^{\frac{p}{m-1}}. \end{aligned}$$The desired conclusions follow.
\(\square \)
Theorem 4.3
Let \(\mathcal {B}\) be an m-order n-dimensional B\(_{0}\)-tensor. Then,
-
(i)
\( n^{\frac{m}{2}}\max _{i\in [n]}\beta _i(\mathcal {B})\le \Vert T_{\mathcal {B}}\Vert _{\infty }\le n^{\frac{m}{2}}\mathop {\max }\limits _{i\in [n]}b_{ii\cdots i}\);
-
(ii)
\(n^{\frac{mp-2}{2p}} \left( \mathop {\sum }\limits _{i=1}^{n}\left( \beta _i(\mathcal {B})\right) ^p\right) ^\frac{1}{p}\le \Vert T_{\mathcal {B}}\Vert _{p}\le n^{\frac{mp-2}{2p}}\left( \mathop {\sum }\limits _{i=1}^{n}b_{ii\cdots i}^{p}\right) ^{\frac{1}{p}}\) if \(p\ge 1;\)
-
(iii)
\( n\left( \beta _i(\mathcal {B})\right) ^{\frac{1}{m-1}}\le \Vert F_{\mathcal {B}}\Vert _{\infty }\le n\mathop {\max }\limits _{i\in [n]}b_{ii\cdots i}^{\frac{1}{m-1}}\) if m is even;
-
(iv)
\(n^{\frac{p-1}{p}} \left( \mathop {\sum }\limits _{i=1}^{n}\left( \beta _i(\mathcal {B})\right) ^{\frac{p}{m-1}}\right) ^\frac{1}{p}\le \Vert F_{\mathcal {B}}\Vert _{p}\le n^{\frac{p-1}{p}} \left( \mathop {\sum }\limits _{i=1}^{n}b_{ii\cdots i}^{\frac{p}{m-1}}\right) ^\frac{1}{p}\) if m is even and \(p\ge 1.\)
For a B-tensor \(\mathcal {B}\), it is obvious that the upper bounds of its operator norms are simpler in form than ones of Lemma 2.2, but we cannot compare their size relationship. By constructing following two examples, we show that there exists B-tensors such that the above upper bounds of \(\Vert T_{\mathcal {B}}\Vert \) and \(\Vert F_{\mathcal {B}}\Vert \) are smaller than the ones of Lemma 2.2.
Example 4.1
Let \(\mathcal {B}=(b_{i_{1}i_{2}i_{3}i_{4}})\in S_{4,3}\), where \(b_{1111}=b_{3333}=6\), \(b_{2222}=5\), \(b_{1333}=b_{3133}=b_{3313}=b_{3331}=1\), \(b_{2322}=b_{2232}=b_{2223}=b_{3222}=1.5\) and all other \(b_{i_{1}i_{2}i_{3}i_{4}}=2\). Then, \(\mathcal {B}\) is a B-tensor.
Proof
It is obvious that \(\mathop {\sum }\limits _{i_{2},i_{3},i_{4}=1}^3b_{ii_{2}i_{3}i_{4}}>0\) for \(i=1,2,3.\)
For \(i=1\), we have
and hence, \(\frac{1}{n^{m-1}}\left( \mathop {\sum }\limits _{i_{2},i_{3},i_{4}=1}^3b_{1i_{2}i_{3}i_{4}}\right) >b_{1j_{2}j_{3}j_{4}}\) for all \((j_{2},j_{3},j_{4})\ne (1,1,1)\).
For \(i=2\), we have
and so, \(\frac{1}{n^{m-1}}\left( \mathop {\sum }\limits _{i_{2},i_{3},i_{4}=1}^{n}b_{2i_{2}i_{3}i_{4}}\right) >b_{2j_{2}j_{3}j_{4}}\) for all \((j_{2},j_{3},j_{4})\ne (2,2,2)\).
For \(i=3\), we have
and so, \(\frac{1}{n^{m-1}}\left( \mathop {\sum }\limits _{i_{2},i_{3},i_{4}=1}^{n}b_{3i_{2}i_{3}i_{4}}\right) >b_{3j_{2}j_{3}j_{4}}\) for all \((j_{2},j_{3},j_{4})\ne (3,3,3)\).
Thus, \(\mathcal {B}\) is a B-tensor. \(\square \)
Clearly, for the upper bounds of \(\Vert T_{\mathcal {B}}\Vert _\infty \), we have
and hence, \(n^{\frac{m}{2}}\mathop {\max }\limits _{i\in [n]}b_{iiii}< \max \limits _{i\in [n]}\left( \mathop {\sum }\limits _{i_{2},i_{3},i_{4}=1}^{n}|b_{ii_{2}i_{3}i_{4}}|\right) \).
It is obvious that for the upper bounds of \(\Vert F_{\mathcal {B}}\Vert _p\) and \(p=1\),
Example 4.2
Let \(\mathcal {B}=(b_{i_{1}i_{2}i_{3}i_{m}})\in S_{4,4}\), where \(b_{1111}=b_{2222}=b_{3333}=b_{4444}=3\), \(b_{1444}=b_{4144}=b_{4414}=b_{4441}=0.7\), \(b_{2333}=b_{3233}=b_{3323}=b_{3332}=0.5\) and all other \(b_{i_{1}i_{2}i_{3}i_{4}}=1\). Then, \(\mathcal {B}\) is a B-tensor.
Proof
It is obvious that \(\mathop {\sum }\limits _{i_{2},i_{3},i_{4}=1}^4b_{ii_{2}i_{3}i_{4}}>0\) for \(i=1,2,3,4.\)
For \(i=1\),
So we have \(\frac{1}{n^{m-1}}\left( \mathop {\sum }\limits _{i_{2},i_{3},i_{4}=1}^{n}b_{1i_{2}i_{3}i_{4}}\right) >b_{1j_{2}j_{3}j_{4}}\) for all \((j_{2},j_{3},j_{4})\ne (1,1,1)\).
For \(i=2\),
So we have \(\frac{1}{n^{m-1}}\left( \mathop {\sum }\limits _{i_{2},i_{3},i_{4}=1}^{n}b_{2i_{2}i_{3}i_{4}}\right) >b_{2j_{2}j_{3}j_{4}}\) for all \((j_{2},j_{3},j_{4})\ne (2,2,2)\).
For \(i=3\),
So we have \(\frac{1}{n^{m-1}}\left( \mathop {\sum }\limits _{i_{2},i_{3},i_{4}=1}^{n}b_{3i_{2}i_{3}i_{4}}\right) >b_{3j_{2}j_{3}j_{4}}\) for all \((j_{2},j_{3},j_{4})\ne (3,3,3)\).
For \(i=4\)
So we have \(\frac{1}{n^{m-1}}\left( \mathop {\sum }\limits _{i_{2},i_{3},i_{4}=1}^{n}b_{4i_{2}i_{3}i_{4}}\right) >b_{4j_{2}j_{3}j_{4}}\) for all \((j_{2},j_{3},j_{4})\ne (4,4,4)\).
Therefore, \(\mathcal {B}\) is a B-tensor. \(\square \)
It is obvious that for the upper bounds of \(\Vert T_{\mathcal {B}}\Vert _p\),
and
Since \(4^{\frac{3}{p}}>1\), we have \(n^{\frac{mp-2}{2p}}\left( \mathop {\sum }\limits _{i=1}^{n}b_{iiii}^{p}\right) ^{\frac{1}{p}}< n^{\frac{m-2}{p}}\left( \mathop {\sum }\limits _{i=1}^{n}\left( \mathop {\sum }\limits _{i_{2},i_{3},i_{4}=1}^{n}|b_{ii_{2}i_{3}i_{4}}|\right) ^{p}\right) ^{\frac{1}{p}}\).
Theorem 4.4
Let \(\mathcal {B}\) be a B-tensor. Then,
-
(i)
\(|\lambda | < \left( \mathop {\sum }\limits _{i=1}^{n}b_{ii\cdots i}^{\frac{1}{m-1}}\right) ^{m-1}\) for all eigenvalues \(\lambda \) of \(\mathcal {B}\) if m is even;
-
(ii)
\(|\mu |< n^\frac{m-1}{2}\left( \mathop {\sum }\limits _{i=1}^nb_{ii\cdots i}^2\right) ^\frac{1}{2}\) for all E-eigenvalues \(\mu \) of \(\mathcal {B}\).
Proof
-
(i)
From the definition of the operator \(F_\mathcal {B}\), it follows that \(\lambda \) is an eigenvalue of \(\mathcal {B}\) if and only if \(\lambda ^{\frac{1}{m-1}}\) is an eigenvalue of \(F_\mathcal {B}\). By Theorem 4.2 of Song and Qi [38], we have
$$\begin{aligned} |\lambda |^{\frac{1}{m-1}}\le \Vert F_\mathcal {B}\Vert _1,\hbox { i.e., } |\lambda |\le \Vert F_\mathcal {B}\Vert _1^{m-1}. \end{aligned}$$By Theorem 4.2 (ii), we have
$$\begin{aligned} \Vert F_{\mathcal {B}}\Vert _1< \mathop {\sum }\limits _{i=1}^{n}b_{ii\cdots i}^{\frac{1}{m-1}}. \end{aligned}$$So we obtain
$$\begin{aligned} |\lambda |\le \Vert F_\mathcal {B}\Vert _1^{m-1}<\left( \mathop {\sum }\limits _{i=1}^{n}b_{ii\cdots i}^{\frac{1}{m-1}}\right) ^{m-1}. \end{aligned}$$ -
(ii)
From the definition of the operator \(T_\mathcal {B}\), it follows that \(\mu \) is an E-eigenvalue of \(\mathcal {B}\) if and only if \(\mu \) is an eigenvalue of \(T_\mathcal {B}\). By Theorem 4.2 of Song and Qi [38], we have
$$\begin{aligned} |\mu |\le \Vert T_\mathcal {B}\Vert _\infty \hbox { and }|\mu |\le \Vert T_\mathcal {B}\Vert _2. \end{aligned}$$By Theorem 4.1, we obtain
$$\begin{aligned} \Vert T_\mathcal {B}\Vert _\infty<n^{\frac{m}{2}}\mathop {\max }\limits _{i \in [n]}b_{ii\cdots i}\hbox { and } \Vert T_\mathcal {B}\Vert _2<n^{\frac{m-1}{2}}\left( \mathop {\sum }\limits _{i=1}^nb_{ii\cdots i}^2\right) ^\frac{1}{2}. \end{aligned}$$Then, we have
$$\begin{aligned} |\mu |<n^{\frac{m}{2}}\mathop {\max }\limits _{i \in [n]}b_{ii\cdots i}\hbox { and } |\mu |<n^{\frac{m-1}{2}}\left( \mathop {\sum }\limits _{i=1}^nb_{ii\cdots i}^2\right) ^\frac{1}{2}. \end{aligned}$$So, the desired conclusion follows. \(\square \)
Similarly, we easily show the following.
Theorem 4.5
Let \(\mathcal {B}\) be a B\(_0\)-tensor. Then,
-
(i)
\(|\lambda | \le \left( \mathop {\sum }\limits _{i=1}^{n}b_{ii\cdots i}^{\frac{1}{m-1}}\right) ^{m-1}\) for all eigenvalues \(\lambda \) of \(\mathcal {B}\) if m is even;
-
(ii)
\(|\mu |\le n^{\frac{m-1}{2}}\left( \mathop {\sum }\limits _{i=1}^nb_{ii\cdots i}^2\right) ^\frac{1}{2}\) for all E-eigenvalues \(\mu \) of \(\mathcal {B}\).
For the tensor complementarity problem with a B-tensor \(\mathcal {B}\), denoted by \(TCP(\mathcal {B},q)\), we show the lower bound of the solution set of \(TCP(\mathcal {B},q)\).
Theorem 4.6
Let \(\mathcal {B}\) be a B-tensor. Assume that x is a nonzero solution of \(TCP(\mathcal {B},q)\) and \(y_{+}=(\max \{y_{1},0\},\max \{y_{2},0\},\ldots ,\max \{y_{n},0\})^\top \). Then,
-
(i)
\(\dfrac{\Vert (-q)_+\Vert _\infty }{n^{m-1}\mathop {\max }\limits _{i\in [n]}b_{ii\cdots i}}<\Vert x\Vert _\infty ^{m-1}\);
-
(ii)
\(\dfrac{\Vert (-q)_+\Vert _2}{n^{\frac{m-1}{2}}\left( \mathop {\sum }\limits _{i=1}^{n}b_{ii\cdots i}^{2}\right) ^{\frac{1}{2}}}<\Vert x\Vert _2^{m-1}\);
-
(iii)
\(\dfrac{\Vert (-q)_+\Vert _m}{n^{\frac{(m-1)^2}{m}}\left( \mathop {\sum }\limits _{i=1}^{n}b_{ii\cdots i}^{\frac{m}{m-1}}\right) ^{\frac{m-1}{m}}}<\Vert x\Vert _m^{m-1}\) if m is even.
Proof
Theorem 3.1 implies that each B-tensor is strictly semi-positive. From Theorem 5 of Song and Qi [23], it follows that
By Theorem 4.1, we have
Then, combining the inequalities (9) and (10), the conclusions (i) and (ii) are proved.
Similarly, we may show (iii). In fact, if m is even, Theorem 5 of Song and Qi [23] implies
By Theorem 4.2, we have
So, the desired conclusion follows by combining the inequalities (11) and (12). \(\square \)
5 Conclusions
In this paper, we show that each B-tensor is strictly semi-positive. We obtain the strictly lower bounds of the solution set of tensor complementarity problem with B-tensors. We establish the upper bounds of spectral radius and E-spectral radius of B-tensors. The following topics are worth of further and serious consideration.
-
For two positively homogeneous operators induced by B-tensors, whether or not their operator norms may be found exactly.
-
Do there exist similar bounds of the operator norm associated with some other structured tensors such as Z-tensors, H-tensors, M-tensors and so on?
-
Are the upper bounds and lower bounds best in this paper?
-
How to design an effective algorithm to compute the upper or lower bounds?
References
Song, Y., Qi, L.: Properties of some classes of structured tensors. J. Optim. Theory Appl. 165(3), 854–873 (2015)
Song, Y., Qi, L.: Properties of tensor complementarity problem and some classes of structured tensors. Ann. Appl. Math. 33(3), 308–323 (2017)
Huang, Z., Qi, L.: Formulating an \(n\)-person noncooperative game as a tensor complementarity problem. Comput. Optim. Appl. 66(3), 557–576 (2017)
Song, Y., Yu, G.: Properties of solution set of tensor complementarity problem. J. Optim. Theory Appl. 170, 85–96 (2016)
Song, Y., Qi, L.: Tensor complementarity problem and semi-positive tensors. J. Optim. Theory Appl. 169, 1069–1078 (2016)
Gowda, M. S., Luo, Z., Qi, L., Xiu, N.: Z-tensors and complementarity problems. arXiv:1510.07933v2 (2016)
Luo, Z., Qi, L., Xiu, N.: The sparsest solutions to Z-tensor complementarity problems. Optim. Lett. 11(3), 471–482 (2017)
Ding, W., Luo, Z., Qi, L.: P-Tensors, \(\text{P}_{0}\)-Tensors, and tensor complementarity problem. arXiv:1507.06731v1 (2015)
Wang, Y., Huang, Z., Bai, X.: Exceptionally regular tensors and tensor complementarity problems. Optim. Method Softw. 31, 815–828 (2016)
Bai, X., Huang, Z., Wang, Y.: Global uniqueness and solvability for tensor complementarity problems. J. Optim. Theory Appl. 170, 72–84 (2016)
Che, M., Qi, L., Wei, Y.: Positive definite tensors to nonlinear complementarity problems. J. Optim. Theory Appl. 168, 475–487 (2016)
Huang, Z., Suo, S., Wang, J.: On Q-tensors. Pac. J. Optim. arXiv:1509.03088 (2015)
Song, Y., Qi, L.: Eigenvalue analysis of constrained minimization problem for homogeneous polynomial. J. Global Optim. 64(3), 563–575 (2016)
Ling, C., He, H., Qi, L.: On the cone eigenvalue complementarity problem for higher-order tensors. Comput. Optim. Appl. 63, 143–168 (2016)
Ling, C., He, H., Qi, L.: Higher-degree eigenvalue complementarity problems for tensors. Comput. Optim. Appl. 64(1), 149–176 (2016)
Chen, Z., Yang, Q., Ye, L.: Generalized eigenvalue complementarity problem for tensors. Pac. J. Optim. 13(3), 527–545 (2017)
Chen, T., Li, W., Wu, X., Vong, S.: Error bounds for linear complementarity problems of MB-matrices. Numer. Algorithm 70(2), 341–356 (2015)
Dai, P.: Error bounds for linear complementarity problems of DB-matrices. Linear Algebra Appl. 434, 830–840 (2011)
Dai, P., Li, Y., Lu, C.: Error bounds for linear complementarity problems of SB-matrices. Numer. Algorithm 61(1), 121–139 (2012)
Dai, P., Lu, C., Li, Y.: New error bounds for linear complementarity problems with an SB-matrices. Numer. Algorithm 64(4), 741–757 (2013)
García-Esnaola, M., Peña, J.M.: Error bounds for linear complementarity problems for B-matrices. Appl. Math. Lett. 22(7), 1071–1075 (2009)
Sun, H., Wang, Y.: Further discussion on the error bound for generalized linear complementarity problem over a polyhedral cone. J. Optim. Theory Appl. 159(1), 93–107 (2013)
Song, Y., Qi, L.: Strictly semi-positive tensors and the boundedness of tensor complementarity problems. Optim. Lett. 11, 1407–1426 (2017)
Qi, L., Luo, Z.: Tensor Analysis: Spectral Theory and Special Tensors. Society for Industrial and Applied Mathematics, Philadelphia (2017)
Zhang, L., Qi, L., Zhou, G.: M-tensors and some applications. SIAM. J. Matrix Anal. Appl. 34(2), 437–452 (2014)
Ding, W., Qi, L., Wei, Y.: M-tensors and nonsingular M-tensors. Linear Algebra Appl. 439, 3264–3278 (2013)
Li, C., Li, Y.: Double B-tensors and quasi-double B-tensors. Linear Algebra Appl. 466, 343–356 (2015)
Song, Y., Qi, L.: Infinite and finite dimensional Hilbert tensors. Linear Algebra Appl. 451, 1–14 (2014)
Song, Y., Qi, L.: Infinite dimensional Hilbert tensors on spaces of analytic functions. Commun. Math. Sci. 15(7), 1897–1911 (2017)
Mei, M., Song, Y.: Infinite and finite dimensional generalized Hilbert tensors. Linear Algebra Appl. 532(1), 8–24 (2017)
Peña, J.M.: A class of P-matrices with applications to the localization of the eigenvalues of a real matrix. SIAM. J. Matrix Anal. Appl. 22, 1027–1037 (2001)
Peña, J.M.: On an alternative to Gerschgorin circles and ovals of cassini. Numer. Math. 95, 337–345 (2003)
Yuan, P., You, L.: Some remarks on P,\(\text{ P }_{0}\), B and \(\text{ B }_{0}\) tensors. Linear Algebra Appl. 459(3), 511–521 (2014)
Qi, L., Song, Y.: An even order symmetric B-tensor is positive definite. Linear Algebra Appl. 457, 303–312 (2014)
Qi, L.: Eigenvalues of a real supersymmetric tensor. J. Symb. Comput. 40, 1302–1324 (2005)
Qi, L.: Rank and eigenvalues of a supersymmetric tensor, the multivariate homogeneous polynomial and the algebraic hypersurface it defines. J. Symb. Comput. 41, 1309–1327 (2006)
Lim, L.H.: Singular values and eigenvalues of tensors: a variational approach. In: Proceedings of the 1st IEEE International Workshop on Computational Advances of Multi-Tensor Adaptive Processing, pp. 129-132, Dec 13–15 (2005)
Song, Y., Qi, L.: Spectral properties of positively homogeneous operators induced by higher order tensors. SIAM J. Matrix Anal. Appl. 34, 1581–1595 (2013)
Acknowledgements
The authors would like to thank the anonymous referees/editors for their valuable suggestions which helped us to improve this manuscript. This work was supported by the National Natural Science Foundation of P.R. China (Grant Nos. 11571095, 11601134, 11701154).
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by Liqun Qi.
Rights and permissions
About this article
Cite this article
Song, Y., Mei, W. Structural Properties of Tensors and Complementarity Problems. J Optim Theory Appl 176, 289–305 (2018). https://doi.org/10.1007/s10957-017-1212-2
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10957-017-1212-2