Abstract
In this paper, several new estimates of the minimum H-eigenvalue for weakly irreducible nonsingular \(\mathcal {M}\)-tensors, including the new Brauer-type estimates and the new S-type estimates, are derived. It is proved that the new estimates are tighter than some existing ones and numerical examples are given to verify this fact. The other main result of this paper is to provide a sharper Ky Fan-type theorem which is better than the original Ky Fan theorem for the nonsingular \(\mathcal {M}\)-tensors.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Eigenvalue problems of higher-order tensors have become an important topic of study in a new applied mathematics branch and numerical multilinear algebra, and they have a wide range of practical applications [1,2,3,4,5].
The class of \(\mathcal {M}\)-tensor introduced in [6, 7] is the generalization M-matrices [8]. And some important properties of \(\mathcal {M}\)-tensors and nonsingular \(\mathcal {\mathcal {M}}\)-tensors have been established in [7, 9]. It is noteworthy that some applications of \(\mathcal {M}\)-tensors [6, 7, 9, 10] are related to the eigenvalue problems of \(\mathcal {M}\)-tensors. In [11,12,13,14], some bounds for the minimum H-eigenvalue of nonsingular \(\mathcal {M}\)-tensors have been proposed. The main aim of this paper is to present some new bounds for the minimum H-eigenvalue of weakly irreducible nonsingular \(\mathcal {M}\)-tensors, and these bounds improve some existing ones.
Let \(\mathbb {C}(\mathbb {R})\) denote the set of all complex (real) field and \(N=\{1,2,\ldots ,n\}\). We consider an m-order n-dimensional tensor \(\mathcal {A}=(a_{i_{1}i_{2}\ldots i_{m}})\) consisting of \(n^{m}\) entries, denoted by \(\mathcal {A}\in \mathbb {C}^{[m,n]}(\mathbb {R}^{[m,n]})\), if
where \(i_{j}=1,2,\ldots ,n\) for \(j=1,2,\ldots ,m\) [9, 15]. Obviously, a vector is a tensor of order 1 and a matrix is a tensor of order 2. Moreover, an m-order n-dimensional tensor \(\mathcal {I}=(\delta _{i_{1}i_{2}\ldots i_{m}})\) is called the unit tensor [16], if its entries are \(\delta _{i_{1}\ldots i_{m}}\) for \(i_{1}, \ldots , i_{m}\in N\), where
Let \(\mathcal {A}\in \mathbb {C}^{[m,n]}\), if there exist a number \(\lambda \in \mathbb {C}\) and a nonzero vector \(x=(x_{1},x_{2},\ldots ,x_{n})^\mathrm{T}\in \mathbb {C}^{n}\) that are solutions of the following homogeneous polynomial equations:
then \(\lambda \) is an eigenvalue of \(\mathcal {A}\) and x is the eigenvector of \(\mathcal {A}\) associated with \(\lambda \) [1, 15, 17, 18], where \(\mathcal {A}x^{m-1}\) and \(\lambda x^{[m-1]}\) are vectors, whose ith components are
and
Furthermore, if \(\lambda \) and x are restricted to the real field, then we call \(\lambda \) an H-eigenvalue of \(\mathcal {A}\) and x an H-eigenvector of \(\mathcal {A}\) associated with \(\lambda \) [15].
Let \(\Gamma \) be a digraph with vertex set V and arc set E. If there exist directed paths from i to j and j to i for any \(i,j\in V (i\ne j)\), then \(\Gamma \) is called strongly connected. For each vertex \(i\in V\), if there exists a circuit such that i belong to the circuit, then \(\Gamma \) is called weakly connected. For a tensor \(\mathcal {A}=(a_{i_{1}\ldots i_{m}})\in \mathbb {C}^{[m,n]}\), we associate \(\mathcal {A}\) with a digraph \(\Gamma _{\mathcal {A}}\) as follows. The vertex set of \(\Gamma _{\mathcal {A}}\) is \(V(\mathcal {A})=\{1,\ldots , n\}\), and the arc set of \(\Gamma _{\mathcal {A}}\) is \( E=\{(i,j)|a_{ii_{2}\ldots i_{m}}\ne 0,j\in \{i_{2}.\ldots ,i_{m}\}\ne \{i,\ldots ,i\}\}\). Let \(C(\mathcal {A})\) denote the set of circuits of \(\Gamma _{\mathcal {A}}\). A tensor \(\mathcal {A}\) is called weakly irreducible if \(\Gamma _{\mathcal {A}}\) is strongly connected [19,20,21]. The tensor \(\mathcal {A}\) is called reducible if there exists a nonempty proper index subset \(J\in N\) such that \(a_{i_{1}i_{2}\ldots i_{m}}=0, \forall i_{1}\in J, \forall i_{2},\ldots , i_{m}\notin J\). If \(\mathcal {A}\) is not reducible, then we call \(\mathcal {A}\) irreducible [22].
Let \(\rho (A)=\max \{|\lambda |:\lambda ~\mathrm {is~an~eigenvalue~of}~A\}\), where \(|\lambda |\) denotes the modulus of \(\lambda \). We call \(\rho (A)\) the spectral radius of tensor \(\mathcal {A}\) [23]. An m-order n-dimensional tensor \(\mathcal {A}\) is called nonnegative [1, 2, 16, 23, 24], if each entry is nonnegative. We call a tensor \(\mathcal {A}\) a \(\mathcal {Z}\)-tensor, if all of its off-diagonal entries are nonpositive, which is equivalent to write \(\mathcal {A} = s\mathcal {I}-\mathcal {B}\), where \(s>0\) and \(\mathcal {B}\) is a nonnegative tensor \((\mathcal {B}\ge 0)\), and the set of m-order and n-dimensional \(\mathcal {Z}\)-tensors is denoted by \(\mathbb {Z}\). A \(\mathcal {Z}\)-tensor \(\mathcal {A} = s\mathcal {I}-\mathcal {B}\) is an \(\mathcal {M}\)-tensor if \(s\ge \rho (\mathcal {B})\), and it is a nonsingular (strong) \(\mathcal {M}\)-tensor if \(s>\rho (\mathcal {B})\) [6, 7, 9].
Denote by \(\tau (\mathcal {A})\) the minimum value of the real part of all eigenvalues of the tensor \(\mathcal {A}\). Let \(\mathcal {A}=(a_{i_{1}i_{2}\ldots i_{m}})\in \mathbb {R}^{[m,n]}\). For \(i,j\in N, j\ne i\), we denote
In recent years, much literature has focused on the bounds of the minimum H-eigenvalue of nonsingular \(\mathcal {M}\)-tensors. In [11], He and Huang first proposed the upper and lower bounds for the minimum H-eigenvalue of irreducible nonsingular \(\mathcal {M}\)-tensors as follows.
Lemma 1.1
[11] Let \(\mathcal {A}\in \mathbb {R}^{[m,n]}\) be an irreducible nonsingular \(\mathcal {M}\)-tensor. Then
Lemma 1.2
[11] Let \(\mathcal {A}\in \mathbb {R}^{[m,n]}\) be an irreducible nonsingular \(\mathcal {M}\)-tensor. Then
where
Recently, Zhao and Sang in [13] pointed out that there are some errors in the calculation process of Lemma 1.2, and the correction is as follows:
Lemma 1.3
[13] Let \(\mathcal {A}\in \mathbb {R}^{[m,n]}\) be an irreducible nonsingular \(\mathcal {M}\)-tensor. Then
where
In addition, Wang and Wei presented the upper and lower bounds on \(\tau (\mathcal {A})\) for a weakly irreducible nonsingular \(\mathcal {M}\)-tensor as follows.
Lemma 1.4
[12] Let \(\mathcal {A}\in \mathbb {R}^{[m,n]}\) be a weakly irreducible nonsingular \(\mathcal {M}\)-tensor. Then
where
is a nonnegative matrix and \(\widetilde{\Delta }_{ij}(\mathcal {A})=(a_{i\ldots i}-a_{j\ldots j}-\widetilde{r}_{i}(\mathcal {A}))^{2}+4r_{i}(M(\mathcal {A}))r_{j}(\mathcal {A})\), with \(r_{i}(M(\mathcal {A}))=\sum \nolimits _{j\ne i}(M(\mathcal {A}))_{ij},\widetilde{r}_{i}(\mathcal {A})=r_{i}(\mathcal {A})-r_{i}(M(\mathcal {A}))\).
In this paper, we continue this research on the estimates of the minimum H-eigenvalue for weakly irreducible nonsingular \(\mathcal {M}\)-tensors; inspired by the ideas of [25, 26], we obtain two new estimates of the minimum H-eigenvalue for weakly irreducible nonsingular \(\mathcal {M}\)-tensors. They are proved to be tighter than Lemmas 1.1 and 1.2 in corrected form. Finally, we derive a sharper bound in Ky Fan theorem for nonsingular \(\mathcal {M}\)-tensors.
The remainder of the paper is organized as follows. In Sect. 2, we recollect some useful lemmas on tensors which are utilized in the following proofs, then focus on the estimates of \(\tau (\mathcal {A})\) and establish some new bounds for \(\tau (\mathcal {A})\). In Sect. 3, a sharper bound in Ky Fan theorem is obtained. Finally, some conclusions are given to end this paper in Sect. 4.
2 Several New Estimates of the Minimum H-eigenvalue
In this section, we give several new estimates of the minimum H-eigenvalue for weakly irreducible nonsingular \(\mathcal {M}\)-tensors.
Lemma 2.1
[12] If a tensor \(\mathcal {A}\) is irreducible, then \(\mathcal {A}\) is weakly irreducible.
Lemma 2.2
[11] Let \(\mathcal {A}\) be a nonsingular \(\mathcal {M}\)-tensor and denote by \(\tau (\mathcal {A})\) the minimum value of the real part of all eigenvalues of \(\mathcal {A}\). Then \(\tau (\mathcal {A})\) is an eigenvalue of \(\mathcal {A}\) with a nonnegative eigenvector. Moreover, if \(\mathcal {A}\) is irreducible, then \(\tau (\mathcal {A})\) is the unique eigenvalue with a positive eigenvector.
Zhang et al. [6] obtained some results similar to those of Lemma 2.2 for weakly irreducible nonsingular \(\mathcal {M}\)-tensors in the following lemma.
Lemma 2.3
[6] Let \(\mathcal {A}\) be a nonsingular \(\mathcal {M}\)-tensor and denote by \(\tau (\mathcal {A})\) the minimum value of the real part of all eigenvalues of \(\mathcal {A}\). Then \(\tau (\mathcal {A})\) is an H-eigenvalue of \(\mathcal {A}\) with a nonnegative eigenvector. Moreover, if \(\mathcal {A}\) is a weakly irreducible \(\mathcal {Z}\)-tensor, then \(\tau (\mathcal {A})\) is the unique eigenvalue with a positive eigenvector.
Lemma 2.4
[27] Let \(\mathcal {A}\) be a weakly irreducible nonsingular \(\mathcal {M}\)-tensor. Then \(\tau (\mathcal {A})<\min \limits _{i\in N}\{a_{ii\ldots i}\}\).
For any given diagonal nonsingular matrix \(D=\mathrm {diag}(d_{1},\ldots , d_{n})\), we define a tensor \(\mathcal {A}_{D}\) as follows:
where \(\times _{k}\) is k-mode tensor-matrix multiplication between \(\mathcal {A}\) and D [28]. Here the entries of \(\mathcal {A}_{D}\) are given by [9] as follows:
Lemma 2.5
[23] The tensors \(\mathcal {A}_{D}\) and \(\mathcal {A}\) have the same set of eigenvalues.
Lemma 2.6
Let \(f(x)=a_{1}x^{2}+b_{1}x+c_{1}\), \(g(x)=a_{2}x^{2}+b_{2}x+c_{2}\), where \(a_{1}>0\) and \(a_{2}>0\). Assume that \(x_{1},x_{2}\) and \(\widetilde{x}_{1},\widetilde{x}_{2}\) are roots of \(f(x)=0\) and \(g(x)=0\), respectively. Then the solution of \(f(x)\le 0\) is \([x_{1},x_{2}]\), and that of \(g(x)\le 0\) is \([\widetilde{x}_{1},\widetilde{x}_{2}]\). If \(g(x)\le 0\) under the condition \(f(x)\le 0\), then \([x_{1},x_{2}]\subseteq [\widetilde{x}_{1},\widetilde{x}_{2}]\).
Proof
It is obvious that \([x_{1},x_{2}]\) and \([\widetilde{x}_{1},\widetilde{x}_{2}]\) are the solutions of \(f(x)\le 0\) and \(g(x)\le 0\), respectively. Since \(g(x)\le 0\) under the condition \(f(x)\le 0\), we get \(g(x)\le 0\) for any \(x\in [x_{1},x_{2}]\). And because \([\widetilde{x}_{1},\widetilde{x}_{2}]\) is the solution of \(g(x)\le 0\), we can obtain that \(x\in [\widetilde{x}_{1},\widetilde{x}_{2}]\), i.e., \([x_{1},x_{2}]\subseteq [\widetilde{x}_{1},\widetilde{x}_{2}]\). \(\square \)
Lemma 2.7
([29], Lemmas 2.2 and 2.3) Let \(a,b,c \ge 0\) and \(d >0\).
(I) If \(\frac{a}{b+c+d}\le 1\), then
(II) If \(\frac{a}{b+c+d}\ge 1\), then
2.1 The New Brauer-Type Estimates of Minimum H-eigenvalue
In this subsection, we present the new Brauer-type estimates of minimum H-eigenvalue for weakly irreducible nonsingular \(\mathcal {M}\)-tensors, which are tighter than the results in Lemmas 1.1 and 1.2 in corrected form.
We denote
and let
Then, \(r_{i}(\mathcal {A})=r_{i}^{\Delta _{i}}(\mathcal {A})+r_{i}^{\overline{\Delta } _{i}}(\mathcal {A}).\)
Theorem 2.1
Let \(\mathcal {A}=(a_{i_{1}i_{2}\ldots i_{m}})\in \mathbb {R}^{[m,n]}\) be a weakly irreducible nonsingular \(\mathcal {M}\)-tensor with \(n\ge 2\). Then
where
Proof
Since \(\mathcal {A}\) is weakly irreducible nonsingular \(\mathcal {M}\)-tensor, by Lemma 2.3, there exists \(x=(x_{1},x_{2},\ldots , x_{n})^\mathrm{T}>0\) such that
Now, the proof is proceeded in two steps.
(i) Let \(x_{t}\ge x_{l}\ge \max \{x_{k}:k\in N,k\ne t,k\ne l\}\) (where the last term above is defined to be zero if \(n=2\)). From (2.1), we have
Using the inequality technique gives
Equivalently
If \(a_{tt\ldots t}-\tau (\mathcal {A})-r_{t}^{\Delta _{t}}(\mathcal {A})\le 0\), then
Otherwise, we have \(a_{tt\ldots t}-\tau (\mathcal {A})-r_{t}^{\Delta _{t}}(\mathcal {A})> 0\), which means that
On the other hand, by (2.1) we can get
i.e.,
Multiplying Inequalities (2.3) and (2.4) yields
Note that \(x_{t}^{m-1}x_{l}^{m-1}>0\), thus
which is equivalent to
This gives the following bounds for \(\tau (\mathcal {A})\),
where
Furthermore, by Inequality (2.3), we can get that
consequently,
Combining Inequalities (2.5) and (2.6), we have
The first inequality in Theorem 2.1 follows from Inequalities (2.2) and (2.7).
(ii) Let \(x_{p}\le x_{q}\le \min \{x_{k}:k\in N,k\ne p,k\ne q\}\). By (2.1), we derive that
and
Using the inequality technique gives
and
Combining Inequalities (2.8) and (2.9) and using the same method as the proof in (i), we can deduce the following result:
This completes our proof of Theorem 2.1. \(\square \)
We now give the following comparison theorem for Theorem 2.1 and Lemma 1.2 in corrected form. First, we prove that the lower bound of Theorem 2.1 is better than that of Lemma 1.2 in corrected form.
Theorem 2.2
Let \(\mathcal {A}=(a_{i_{1}i_{2}\ldots i_{m}})\in \mathbb {R}^{[m,n]}\) be a weakly irreducible nonsingular \(\mathcal {M}\)-tensor with \(n\ge 2\). Then
Proof
From proof of Lemma 1.3, we can see that \(\tau (\mathcal {A})\ge \min \nolimits _{\begin{array}{c} i,j\in N,\\ j\ne i \end{array}}L_{ij}(\mathcal {A})\) is obtained by solving the following quadratic inequality
Let \(g^{ij}(\tau (\mathcal {A}))=(a_{ii\ldots i}-\tau (\mathcal {A})-r_{i}^{j}(\mathcal {A}))(a_{jj\ldots j}-\tau (\mathcal {A}))-(-a_{ij\ldots j})r_{j}(\mathcal {A})\), and the left solution of \(g^{ij}(\tau (\mathcal {A}))=0\) is \(L_{ij}(\mathcal {A})\). If \(\Lambda _{\min }=\widetilde{\Lambda }_{\min } =\min \nolimits _{i\in N}\{a_{ii\ldots i}-r_{i}^{\Delta _{i}}(\mathcal {A})\}\), then there exists \(i_{0}\in N\) such that
From Theorem 2.1, we get
which together with Lemma 2.4 results in
By Lemma 2.6, we derive that
If \(\Lambda _{\min }=\overline{\Lambda }_{\min }=\min \limits _{\begin{array}{c} i,j\in N,\\ j\ne i \end{array}}\max \{\frac{1}{2} (a_{ii\ldots i}+a_{j j\ldots j}-r_{i}^{\Delta _{i}}(\mathcal {A})-r_{j}^{\overline{\Delta }_{i}}(\mathcal {A}) -\Omega _{i,j}^{\frac{1}{2}}),R_{i}(\mathcal {A})\}\), then there exist \(i_{1},j_{1}\in N\) such that
which means that
and
By proof of Theorem 2.1, we see that \(K_{i_{1}j_{1}}(\mathcal {A}):=\frac{1}{2} (a_{i_{1}\ldots i_{1}}+a_{j_{1}\ldots j_{1}}-r_{i_{1}}^{\Delta _{i_{1}}}(\mathcal {A})-r_{j_{1}}^{\overline{\Delta }_{i_{1}}}(\mathcal {A}) -\Omega _{i_{1},j_{1}}^{\frac{1}{2}})\) is the left root of the following equation
so, we let
By Lemma 2.6, if \(g^{i_{1}j_{1}}(\tau (\mathcal {A}))\le 0\) under the condition \(f^{i_{1}j_{1}}(\tau (\mathcal {A}))\le 0\), then \(K_{i_{1}j_{1}}(\mathcal {A})\ge L_{i_{1}j_{1}}(\mathcal {A})\ge \min \nolimits _{\begin{array}{c} i,j\in N,\\ j\ne i \end{array}}L_{ij}(\mathcal {A})\). Combining with (2.11), we can derive that \(\Lambda _{\min }\ge \min \limits _{\begin{array}{c} i,j\in N,\\ j\ne i \end{array}}L_{ij}(\mathcal {A})\). Therefore, now we only need to prove that \(g^{i_{1}j_{1}}(\tau (\mathcal {A}))\le 0\) under the condition \(f^{i_{1}j_{1}}(\tau (\mathcal {A}))\le 0\).
When \(a_{i_{1}\ldots i_{1}}-\tau (\mathcal {A})-r_{i_{1}}^{\Delta _{i_{1}}}(\mathcal {A})\le 0\), it is not difficult to get the following form
Otherwise, we have \(a_{i_{1}\ldots i_{1}}-\tau (\mathcal {A})-r_{i_{1}}^{\Delta _{i_{1}}}(\mathcal {A})> 0\). From the condition \(f^{i_{1}j_{1}}(\tau (\mathcal {A}))\le 0\), we have
If \(r_{i_{1}}^{\overline{\Delta }_{i_{1}}}(\mathcal {A})r_{j_{1}}^{\Delta _{i_{1}}}(\mathcal {A})=0\), then
which leads to
In addition, by (2.12) we have
i.e.,
Note that \(\tau (\mathcal {A})<a_{j_{1}\ldots j_{1}}\), then multiplying Inequality (2.15) with Inequality (2.16) gives
which implies that \(g^{i_{1}j_{1}}(\tau (\mathcal {A}))\le 0\).
If \(r_{i_{1}}^{\overline{\Delta }_{i_{1}}}(\mathcal {A})r_{j_{1}}^{\Delta _{i_{1}}}(\mathcal {A})>0\), then by dividing Inequality (2.14) by \(r_{i_{1}}^{\overline{\Delta }_{i_{1}}}(\mathcal {A})r_{j_{1}}^{\Delta _{i_{1}}}(\mathcal {A})\), we get
By (2.12), we have
Then it follows from Inequality (2.17) that
or
When \(-a_{i_{1}j_{1}\ldots j_{1}}>0\), from the part (I) in Lemma 2.7 and Inequality (2.18) we have
Furthermore, if \(\frac{a_{j_{1}\ldots j_{1}} -\tau (\mathcal {A})-r_{j_{1}}^{\overline{\Delta }_{i_{1}}}(\mathcal {A})}{r_{j_{1}}^{\Delta _{i_{1}}}(\mathcal {A})}\ge 1\), it follows from the part (II) in Lemma 2.7 that
Then multiplying Inequality (2.19) with Inequality (2.20), together with (2.17), gives
equivalently,
that is \(g^{i_{1}j_{1}}(\tau (\mathcal {A}))\le 0\). And if \(\frac{a_{j_{1}\ldots j_{1}} -\tau (\mathcal {A})-r_{j_{1}}^{\overline{\Delta }_{i_{1}}}(\mathcal {A})}{r_{j_{1}}^{\Delta _{i_{1}}}(\mathcal {A})}\le 1\), then
Inequality (2.18) implies
The above two inequalities lead to
i.e., \(g^{i_{1}j_{1}}(\tau (\mathcal {A}))\le 0\).
When \(a_{i_{1}j_{1}\ldots j_{1}}=0\), from (2.18), we easily get
Hence,
i.e., \(g^{i_{1}j_{1}}(\tau (\mathcal {A}))\le 0\).\(\square \)
By using the technique in the proof of Theorem 2.2, we can get \(\overline{\Lambda }_{\max }\le \max \nolimits _{\begin{array}{c} i,j\in N,\\ j\ne i \end{array}}L_{ij}(\mathcal {A})\). Combining with Theorem 5 in [13], we can easily obtain the bounds in Theorem 2.1 are shaper than Lemmas 1.1 and 1.2 in corrected form.
Now we take an example to show the efficiency of the bounds established in Theorem 2.1.
Example 2.1
Let \(\mathcal {A}=(a_{ijk})\in \mathbb {R}^{[3,3]}\) be a weakly irreducible \(\mathcal {M}\)-tensor with entries defined as follows:
where
We compare the results derived in Theorem 2.1 with those of Lemmas 1.1, 1.2 in the correct form and Lemma 1.4. By Lemma 1.1, we have
By Lemma 1.2 in the corrected form, we get
By Lemma 1.4, we obtain
By Theorem 2.1 , we have
This example shows that the upper and lower bounds in Theorem 2.1 are better than those in Lemmas 1.1, 1.2 and 1.4.
2.2 The New S-type Estimates of Minimum H-eigenvalue
In this subsection, the new S-type estimates of minimum H-eigenvalue for weakly irreducible nonsingular \(\mathcal {M}\)-tensor are derived, which are better than the ones in Lemmas 1.1 and 1.2 in corrected form.
Given a nonempty proper subset S of N, we denote
This implies that for \(i\in S\), we have
where
Theorem 2.3
Let \(\mathcal {A}=(a_{i_{1}i_{2}\ldots i_{m}})\in \mathbb {R}^{[m,n]}\) be a weakly irreducible nonsingular \(\mathcal {M}\)-tensor with \(n\ge 2\), and S be a nonempty proper subset of N. Then
where
Proof
Since \(\mathcal {A}\) is a weakly irreducible nonsingular \(\mathcal {M}\)-tensor, by Lemma 2.3, there exists \(x=(x_{1},x_{2},\ldots ,x_{n})^\mathrm{T}>0\) such that
(i) Let \(x_{l}=\max \nolimits _{i\in S}x_{i}\) and \(x_{t}=\max \nolimits _{i\in \overline{S}}x_{i}\). Next, we divide into two cases to prove.
Case I\(x_{t}\ge x_{l}\), that is, \(x_{t}=\max \nolimits _{i\in N}x_{i}\). From (2.21), we have
Using the inequality technique, together with \(\tau (\mathcal {A})< a_{tt\ldots t}\), gives
hence,
On the other hand, by (2.21), we also get that
Multiplying (2.22) with (2.23) gives
Solving the above quadratic inequality yields
with
Furthermore, by Inequality (2.22), we can get that
i.e.,
It follows from Inequalities (2.24) and (2.25) that
Case II\(x_{l}\ge x_{t}\), that is, \(x_{l}=\max \limits _{i\in N}x_{i}\). In a similar manner to the proof of Case I, we have
and
Note that \(x_{t}x_{l}>0\). Thus,
and
Then, solve for \(\tau (\mathcal {A})\),
Combining (2.26) and (2.27) yields the first inequality of Theorem 2.3.
(ii) Let \(x_{p}=\min \nolimits _{i\in S}x_{i}\) and \(x_{q}=\min \nolimits _{i\in \overline{S}}x_{i}\). Dividing into two cases to prove: \(x_{p}\ge x_{q}\) and \(x_{q}\ge x_{p}\) and by the analogical proof as (i), we can prove the second inequality of Theorem 2.3.\(\square \)
Next, we show the bounds of Theorem 2.3 are sharper than those of Lemma 1.2 in corrected form. We first proof that the lower bound of Theorem 2.3 is greater than or equal to than that of Lemma 1.2 in corrected form.
Theorem 2.4
Let \(\mathcal {A}=(a_{i_{1}i_{2}\ldots i_{m}})\in \mathbb {R}^{[m,n]}\) be a weakly irreducible nonsingular \(\mathcal {M}\)-tensor with \(n\ge 2\). Then
Proof
By Theorem 2.3, we have \(\Upsilon _{\min }(\mathcal {A})=\overline{\Upsilon }^{S}(\mathcal {A})\) or \(\Upsilon _{\min }(\mathcal {A}) =\overline{\Upsilon }^{\overline{S}}(\mathcal {A})\). Without loss of generality, we suppose that \(\Upsilon _{\min }(\mathcal {A})=\overline{\Upsilon }^{S}(\mathcal {A})\) (we can prove it similarly if \(\Upsilon _{\min }(\mathcal {A})=\overline{\Upsilon }^{\overline{S}}(\mathcal {A})\)). Then there are \(i_{2}\in S,j_{2}\in \overline{S}\) such that
which leads to
and
From proof of Theorem 2.3, Inequality (2.29) is derived by solving the following quadratic inequality
So we let \(h^{i_{2}j_{2}}(\tau (\mathcal {A}))=(a_{j_{2}\ldots j_{2}}-\tau (\mathcal {A})-r_{j_{2}}^{\overline{\Delta ^{S}}}(\mathcal {A}))(a_{i_{2}\ldots i_{2}} -\tau (\mathcal {A}))-r_{j_{2}}^{\Delta ^{S}}(\mathcal {A})r_{i_{2}}(\mathcal {A})\) and \(W_{i_{2}j_{2}}(\mathcal {A}):=\frac{1}{2} (a_{j_{2}\ldots j_{2}}+a_{i_{2}\ldots i_{2}}-r_{j_{2}}^{\overline{\Delta ^{S}}}(\mathcal {A}) -(\Psi ^{S}_{i_{2},j_{2}})^{\frac{1}{2}})\) is the left root of the equation \(h^{i_{2}j_{2}}(\tau (\mathcal {A}))=0\). By Lemma 2.6, if \(g^{j_{2}i_{2}}(\tau (\mathcal {A}))\le 0\) under the condition \(h^{i_{2}j_{2}}(\tau (\mathcal {A}))\le 0\), then \(W_{i_{2}j_{2}}(\mathcal {A}) \ge L_{j_{2}i_{2}}(\mathcal {A})\ge \min \nolimits _{\begin{array}{c} i,j\in N,\\ j\ne i \end{array}}L_{ij}(\mathcal {A})\), that is, \(\Upsilon _{\min }(\mathcal {A})\ge \min \nolimits _{\begin{array}{c} i,j\in N,\\ j\ne i \end{array}}L_{ij}(\mathcal {A})\). We now prove that \(g^{j_{2}i_{2}}(\tau (\mathcal {A}))\le 0\) under the condition \(h^{i_{2}j_{2}}(\tau (\mathcal {A}))\le 0\). From the condition \(h^{i_{2}j_{2}}(\tau (\mathcal {A}))\le 0\), we have
If \(r_{j_{2}}^{\Delta ^{S}}(\mathcal {A})r_{i_{2}}(\mathcal {A})=0\), then \(r_{j_{2}}^{\Delta ^{S}}(\mathcal {A})=0\) or \(r_{i_{2}}(\mathcal {A})=0\). When \(r_{j_{2}}^{\Delta ^{S}}(\mathcal {A})=0\), we get \(-a_{j_{2}i_{2}\ldots i_{2}}=0, r_{j_{2}}^{\overline{\Delta ^{S}}}(\mathcal {A})=r_{j_{2}}^{i_{2}}(\mathcal {A})\). Therefore,
consequently, \(g^{i_{2}j_{2}}(\tau (\mathcal {A}))\le 0\). When \(r_{i_{2}}(\mathcal {A})=0\),
This leads to \(g^{j_{2}i_{2}}(\tau (\mathcal {A}))\le 0\).
If \(r_{j_{2}}^{\Delta ^{S}}(\mathcal {A})r_{i_{2}}(\mathcal {A})>0\), then we can equivalently express Inequality (2.30) as
By (2.28), we have \(\frac{a_{j_{2}\ldots j_{2}}-\tau (\mathcal {A})-r_{j_{2}}^{\overline{\Delta ^{S}}}(\mathcal {A})}{r_{j_{2}}^{\Delta ^{S}}(\mathcal {A})}\le 1\), and when \(a_{j_{2}i_{2}\ldots i_{2}}>0\), from the part (I) in Lemma 2.7 we have
together with Inequality (2.31), we can derive that
i.e., \(g^{j_{2}i_{2}}(\tau (\mathcal {A}))\le 0\). When \(a_{j_{2}i_{2}\ldots i_{2}}=0\), by (2.28) we easily get
Hence,
This also implies \(g^{j_{2}i_{2}}(\tau (\mathcal {A}))\le 0\). This completes our proof of Theorem 2.4. \(\square \)
By using the technique in the proof of Theorem 2.4, we can get \(\Upsilon _{\max }(\mathcal {A})\le \max \nolimits _{\begin{array}{c} i,j\in N,\\ j\ne i \end{array}}L_{ij}(\mathcal {A})\). Together with Theorem 5 in [13], we can easily see the bounds in Theorem 2.3 are better than Lemmas 1.1 and 1.2 in corrected form.
Let us show the advantage of Theorem 2.3 over the results in Lemma 1.1, 1.2 which are corrected, Lemma 1.4 and newly derived by Huang et al. [14] by a simple example as follows.
Example 2.2
Let \(\mathcal {A}=(a_{ijk})\in \mathbb {R}^{[3,4]}\) be a weakly irreducible \(\mathcal {M}\)-tensor with entries defined as follows:
where
We now compute the bounds for \(\tau (\mathcal {A})\). Let \(S=\{1,2\}\), then \(\overline{S}=\{3,4\}\). By Lemma 1.1, we have
By Lemma 1.2 in the corrected form, we get
By Lemma 1.4, we obtain
By Theorem 3.5 in [14], we get
By Theorem 2.3, we have
Obviously, the bounds given in Theorem 2.3 are sharper than the aforementioned existing results.
3 Ky Fan Theorem
In [11], He and Huang gave the Ky Fan theorem for nonsingular \(\mathcal {M}\)-tensors as follows:
Lemma 3.1
[11] Let \(\mathcal {A}\), \(\mathcal {B} \) be of order m dimension n, suppose that \(\mathcal {B}\) is a nonsingular \(\mathcal {M}\)-tensor and \(|b_{i_{1},\ldots ,i_{m}}|\ge |a_{i_{1},\ldots ,i_{m}}|\) for any \(i_{1},\ldots ,i_{m} \in N \) and \(\delta _{i_{1},\ldots , i_{m}}\ne 0\). Then, for any eigenvalue \(\lambda \) of \(\mathcal {A}\), there exists \(i\in N\) such that
In [19], Bu et al. derived the following Brualdi-type eigenvalue inclusion sets of tensors.
Lemma 3.2
[19] Let \(\mathcal {A}=(a_{i_{1},\ldots ,i_{m}})\in \mathbb {C}^{[m,n]}\) be a tensor such that \(\Gamma _{\mathcal {A}}\) is weakly connected. Then,
Based on Lemma 3.2, we derive a new set in Ky Fan theorem, which is sharper than the one in (3.1).
Theorem 3.1
Let \(\mathcal {A}, \mathcal {B}\) be m-order n-dimensional tensors such that \(\Gamma _{\mathcal {A}}\) is weakly connected and \(\mathcal {B}\) be a nonsingular \(\mathcal {M}\)-tensor, and \(|b_{i_{1}\ldots i_{m}}|\ge |a_{i_{1}\ldots i_{m}}|\) for all \(i_{1}\ne \ldots \ne i_{m}\). Then, there exists a circuit \(\gamma \in C(\mathcal {A})\), such that
Proof
We first suppose that \(\mathcal {B}\) is irreducible, by Lemma 2.2, there exists \(x=(x_{1},x_{2},\ldots , x_{n})^\mathrm{T}>0\) such that
Let \(D=\mathrm {diag}(x_{1},\ldots ,x_{n})\), \(\mathcal {A}_{D}=\mathcal {A}D^{1-m}\overbrace{D\ldots D}^{m-1}\), \(y=(y_{1},\ldots ,y_{n})^\mathrm{T}\) be an eigenvector of \(\mathcal {A}_{D}\) corresponding to \(\lambda \). Then
By Lemma 2.5, we have
Equation (3.2) implies that for any i,
which is equivalent to
Since \(\Gamma _{\mathcal {A}}\) is weakly connected, so is \(\Gamma _{\mathcal {A}_{D}}\). From Lemma 3.2 and the above equation, for any eigenvalue \(\lambda \) of \(\mathcal {A}_{D}\), there exists a circuit \(\gamma \in C(\mathcal {A})\), such that
When the tensor \(\mathcal {B}\) is reducible, by replacing the zero entries of \(\mathcal {B}\) with \(-\frac{1}{k}\), where k is a positive integer, we see that the Z-tensor \(\mathcal {B}_{k}\) is irreducible and \(|(\mathcal {B}_{k})_{i_{1}\ldots i_{m}}|\ge |\mathcal {A}_{i_{1}\ldots i_{m}}|\). Then there exists a circuit \(\gamma \in C(\mathcal {A})\) such that
From the proof process of Theorem 3.6 in [14], we have
In Inequality (3.3), letting \(k\rightarrow \infty \) results in
This completes our proof of Theorem 3.1.\(\square \)
Denote
It follows from Lemma 3.1 and Theorem 3.1 that \(\sigma (\mathcal {A})\subseteq G(\mathcal {A})\) and \(\sigma (\mathcal {A})\subseteq S(\mathcal {A})\). Next, we compare the sets \(S(\mathcal {A})\) and \(G(\mathcal {A})\) in the following theorem, showing that Theorem 3.1 is better than the Ky Fan theorem.
Theorem 3.2
Let \(\mathcal {A}, \mathcal {B}\) be m-order n-dimensional tensors such that \(\Gamma _{\mathcal {A}}\) is weakly connected, \(\mathcal {B}\) be a nonsingular \(\mathcal {M}\)-tensor, and \(|b_{i_{1}\ldots i_{m}}|\ge |a_{i_{1}\ldots i_{m}}|\) for all \(i_{1}\ne \ldots \ne i_{m}\). Then
Proof
For any \(z\in S(\mathcal {A})\), if \(z\notin G(\mathcal {A})\), then \(|z-a_{ii\ldots i}|>b_{ii\ldots i}-\tau (\mathcal {B})~(i=1,2,\ldots ,n)\). In this case, \(\prod \nolimits _{i\in \gamma }|z-a_{ii\ldots i}|>\prod \nolimits _{i\in \gamma }(b_{ii\ldots i}-\tau (\mathcal {B}))\) for any \(\gamma \in C(\mathcal {A})\), a contradiction to \(z\in S(\mathcal {A})\). Hence \(z\in G(\mathcal {A})\), i.e., \(S(\mathcal {A})\subseteq G(\mathcal {A})\).\(\square \)
4 Conclusions
In this paper, several new estimates of the minimum H-eigenvalue for weakly irreducible nonsingular \(\mathcal {M}\)-tensors are presented, which are proved to be sharper than those of [11, 12]. On the other hand, we have studied a new Ky Fan-type theorem. It should be noted that the new Ky Fan theorem is based on the condition that \(\Gamma _{\mathcal {A}}\) is weakly connected and \(\mathcal {B}\) is a nonsingular \(\mathcal {M}\)-tensor, and the new Ky Fan-type theorem improves the one in [11].
However, the new S-type estimates for minimum H-eigenvalue depend on the set S. Then an interesting problem is how to pick S to make the bounds exhibited in Theorem 2.3 as tight as possible. But it is very difficult when the dimension of the tensor \(\mathcal {A}\) is large. Therefore, future work will include numerical or theoretical studies for finding the best choice for S.
References
Qi, L.Q.: Eigenvalues and invariants of tensor. J. Math. Anal. Appl. 325, 1363–1377 (2007)
Chang, K.C., Pearson, K., Zhang, T.: On eigenvalue problems of real symmetric tensors. J. Math. Anal. Appl. 350, 416–422 (2009)
Qi, L.Q.: Symmetric nonnegative tensors and copositive tensors. Linear Algebra Appl. 439, 228–238 (2013)
Liu, Y.J., Zhou, G.L., Ibrahim, N.F.: An always convergent algorithm for the largest eigenvalue of an irreducible nonnegative tensor. J. Comput. Appl. Math. 235, 286–292 (2010)
Ng, M., Qi, L.Q., Zhou, G.L.: Finding the largest eigenvalue of a non-negative tensor. SIAM J. Matrix Anal. Appl. 31, 1090–1099 (2009)
Zhang, L.P., Qi, L.Q., Zhou, G.L.: \(\cal{M}\)-tensors and some applications. SIAM J. Matrix Anal. Appl. 32, 437–452 (2014)
Ding, W.Y., Qi, L.Q., Wei, Y.M.: \(\cal{M}\)-tensors and nonsingular \(\cal{M}\)-tensors. Linear Algebra Appl. 439, 3264–3278 (2013)
Berman, A., Plemmons, R.J.: Nonnegative Matrices in the Mathematical Sciences. Academic Press, New York (1994)
Kannan, M.R., Monderer, N.S., Berman, A.: Some properties of strong \(\cal{H}\)-tensors and general \(\cal{H}\)-tensors. Linear Algebra Appl. 476, 42–55 (2015)
Zhang, L.P., Qi, L.Q., Zhou, G.L.: \(\cal{M}\)-tensors and the positive definiteness of a multivariate form. preprint arXiv:1202.6431 (2012)
He, J., Huang, T.Z.: Inequalities for \(\cal{M}\)-tensors. J. Inequal. Appl. 114, 2014 (2014)
Wang, X.Z., Wei, Y.M.: Bounds for eigenvalues of nonsingular \(\cal{H}\)-tensor. Electron. J. Linear Algebra 29, 3–16 (2015)
Zhao, J.X., Sang, C.L.: Two new lower bounds for the minimum eigenvalue of \(M\)-tensors. J. Inequal. Appl. 268, 2016 (2016)
Huang, Z.G., Wang, L.G., Xu, Z., Cui, J.J.: A new \(S\)-type eigenvalue inclusion set for tensors and its applications. J. Inequal. Appl. 254, 2016 (2016)
Qi, L.Q.: Eigenvalues of a real supersymmetric tensor. J. Symb. Comput. 40, 1302–1324 (2005)
Yang, Y.N., Yang, Q.Z.: Further results for Perron-Frobenius theorem for nonnegative tensors. SIAM J. Matrix Anal. Appl. 31, 2517–2530 (2010)
Kolda, T.G., Mayo, J.R.: Shifted power method for computing tensor eigenpairs. SIAM J. Matrix Anal. Appl. 32(4), 1095–1124 (2011)
Lim, L.H.: Singular values and eigenvalues of tensors: a variational approach. In: CAMSAP05: Proceeding of the IEEE International Workshop on Computational Advances in MultiSensor Adaptive Processing, pp. 129–132 (2005)
Bu, C.J., Wei, Y.P., Zhou, J.: Brualdi-type eigenvalue inclusion sets of tensors. Linear Algebra Appl. 480, 168–175 (2015)
Friedland, J.S., Gaubert, S., Han, L.: Perron-Frobenius theorem for nonnegative multilinear forms and extensions. Linear Algebra Appl. 438, 738–749 (2013)
Pearson, K., Zhang, T.: On spectral hypergraph theory of the adjacency tensor. Graphs Comb. 30, 1233–1248 (2014)
Li, W., Liu, D.D., Vong, S.W.: \(Z\)-eigenpair bounds for an irreducible nonnegative tensor. Linear Algebra Appl. 483, 182–199 (2015)
Yang, Y.N., Yang, Q.Z.: Further results for Perron-Frobenius theorem for nonnegative tensors II. SIAM J. Matrix Anal. Appl 32, 1236–1250 (2011)
Hu, S.L., Huang, Z.H., Qi, L.Q.: Strictly nonnegative tensors and nonnegative tensor partition. Sci. China Math. 57, 181–195 (2014)
Li, C.Q., Zhao, J.J., Li, Y.T.: A new Brauer-type eigenvalue localization set for tensors. Linear Multilinear Algebra 64, 727–736 (2015)
Li, C.Q., Jiao, A.Q., Li, Y.T.: An \(S\)-type eigenvalue localization set for tensors. Linear Algebra Appl. 493, 469–483 (2016)
Huang, Z.G., Wang, L.G., Xu, Z., Cui, J.J.: Some new inequalities for the minimum \(H\)-eigenvalue of nonsingular \(M\)-tensors. Linear Algebra Appl. (2016). (submitted for publication)
Li, W., Ng, M.K.: Some bounds for the spectral radius of nonnegative tensors. Numer. Math. 130, 315–335 (2015)
Li, C.Q., Li, Y.T.: An eigenvalue localization set for tensors with applications to determine the positive (semi-)definiteness of tensors. Linear and Multilinear Algebra 64, 587–601 (2016)
Acknowledgements
This work was supported by the National Natural Science Foundations of China (10802068).
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no competing interests.
Additional information
Communicated by Emrah Kilic.
Rights and permissions
About this article
Cite this article
Cui, J., Peng, G., Lu, Q. et al. Several New Estimates of the Minimum H-eigenvalue for Nonsingular \(\mathcal {M}\)-tensors. Bull. Malays. Math. Sci. Soc. 42, 1213–1236 (2019). https://doi.org/10.1007/s40840-017-0544-2
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s40840-017-0544-2