# Permutation transformations of tensors with an application

- Yao-Tang Li†
^{1}Email author, - Zheng-Bo Li†
^{1}, - Qi-Long Liu†
^{1}and - Qiong Liu†
^{1}

**Received: **12 April 2016

**Accepted: **21 November 2016

**Published: **28 November 2016

## Abstract

The permutation transformation of tensors is introduced and its basic properties are discussed. The invariance under permutation transformations is studied for some important structure tensors such as symmetric tensors, positive definite (positive semidefinite) tensors, *Z*-tensors, *M*-tensors, Hankel tensors, *P*-tensors, *B*-tensors and *H*-tensors. Finally, as an application of permutation transformations of tensors, the canonical form theorem of tensors is given. The theorem shows that some problems of higher dimension tensors can be translated into the corresponding problems of lower dimension weakly irreducible tensors so as to handle easily.

## Keywords

## Mathematics Subject Classification

## Background

The study of tensors with their various applications has attracted extensive attention and interest, since the work of Qi (2005) and Lim (2005). Lately, the research topic on structure tensors has also attracted much attention, such as symmetric tensors (Qi 2005), \(P(P_0)\)-tensors (Song and Qi 2014), \(B(B_0)\)-tensors (Song and Qi 2014), *Z*-tensors (Zhang et al. 2014), (strong) *M*-tensors (Zhang et al. 2014), *H*-tensors (Li et al. 2014) and so on. In the researches on tensors with its application, the reducibility and higher dimension of tensors are two important factors to cause difficulties. Therefore, it is interesting that how to translate problems of higher dimension reducible tensors into the corresponding problems of lower dimension irreducible tensors.

As we all know, the permutation transformation of matrices plays a very important role in linear algebra and matrix theory. Some problems of higher dimension reducible matrices can be translated into the corresponding problems of lower dimension irreducible matrices by using the permutation transformation of matrices. Inspired by this, we introduce permutation transformations of tensors, and discuss its basic properties and and their applications in this paper.

In the next section, we will introduce the permutation transformation of tensors and give its expression. In third section, we will discuss basic properties of permutation transformations of tensors. In fourth section, we will discuss the invariance under permutation transformations for some important structure tensors such as symmetric tensors, positive definite tensors, *M*-tensors, Hankel tensors, *P*-tensors, *B*-tensors, *H*-tensors and so on. In fifth section, we will give the canonical form theorem of tensors and a numerical example which shows that some problems of higher dimension tensors can be translated into the corresponding problems of lower dimension weakly irreducible tensors by using permutation transformations. Finally, we draw some conclusions in the last section.

## Permutation transformations of tensors and its expression

###
**Definition 1**

- (i)
\(\mathcal {A}+\mathcal {B}=(a_{i_1\ldots i_m}+b_{i_1\ldots i_m}).\)

- (ii)
\(k\mathcal {A}=(ka_{i_1\ldots i_m}).\)

###
*Remark 1*

Obviously, both \(\mathbb {C}^{[m,n]}\) and \(\mathbb {R}^{[m,n]}\) are linear spaces about the addition and the multiplication in Definition 1.

###
**Definition 2**

(Qi 2005) A tensor \(\mathcal {A}=(a_{i_1\ldots i_m})\in \mathbb {R}^{[m,n]}\) is called a symmetric tensor if its entries \(a_{i_1\ldots i_m}\) are invariant under any permutation of their indices.

Now, we give the definition of permutation transformation of tensors.

###
**Definition 3**

###
*Remark 2*

\(P_\pi\) is called as a permutation transformation on \(\mathbb {R}^{[m,n]}\) if \(\mathbb {C}^{[m,n]}\) is replaced by \(\mathbb {R}^{[m,n]}\) in Definition 3.

###
**Definition 4**

For further discussing property of the permutation transformation of tensors, we introduce the following general product of two \(n\)-dimensional tensors defined in Shao (2013). For the sake of simplicity, we sometime use the following “condensed notation” for the subscripts of the tensor. For example, we will write \(a_{i_1i_2\ldots i_m}\) as \(a_{i_1\alpha }\), where \(\alpha =i_2\ldots i_m\in [n]^{m-1}\)and \([n]^{m-1}\) is \(m-1\) dimensional array whose every element varies from 1 to \(n\).

###
**Definition 5**

###
**Definition 6**

(Shao 2013) Let \(\mathcal {A}\), \(\mathcal {B}\in \mathbb {C}^{[m,n]}\). If there exists a permutation \(\pi\) on the set \([n]\), and the corresponding permutation matrix \(P=(P_{ij})\) (where \(P_{ij}=1 \Longleftrightarrow j=\pi (i)\); \(P_{ij}=0\), otherwise.) such that \(\mathcal {B}=P \mathcal {A}P^T\), then we say that \(\mathcal {A}\) and \(\mathcal {B}\) are permutational similar.

###
*Remark 3*

- (i)
\(b_{i_1\ldots i_m}=a_{\pi (i_1)\ldots \pi (i_m)}\),

- (ii)
\(P \mathcal {I}P^T=\mathcal {I}\).

###
**Definition 7**

(Shao 2013) Let \(\mathcal {A},\mathcal {B}\in \mathbb {C}^{[m,n]}\). Suppose that there exist two matrices \(P\) and \(Q\) of dimension \(n\) with \(P\mathcal {I}Q=\mathcal {I}\) such that \(\mathcal {B}=P\mathcal {A}Q\), then we say that the two tensors \(\mathcal {A}\) and \(\mathcal {B}\) are similar.

Now, we present the relationship between the permutation transformation of tensors and permutational similar.

###
**Theorem 1**

*Let*
\(\mathcal {A},\mathcal {B}\in \mathbb {C}^{[m,n]}\). *Then*
\(\mathcal {A}\)
*and*
\(\mathcal {B}\)
*are permutational similar if and only if there exists a permutation*
\(\pi\)
*on*
\([n]\)
*such that*
\(\mathcal {B}=P_\pi (\mathcal {A})\).

###
*Proof*

By Theorem 1 and Remark 3, we have the following expression theorem of permutational transformation of tensors.

###
**Theorem 2**

*Let*\(\mathcal {A}\in \mathbb {C}^{[m,n]}\),

*and*\(\pi\)

*be a permutation on*\([n]\).

*Then*

*where*\(P_{ij}=1 \Longleftrightarrow j=\pi (i)\); \(P_{ij}=0\),

*otherwise.*

## Basic properties of permutation transformations

*n*-dimensional vector \(x=(x_1,x_2,\ldots ,x_n)\), real or complex, we define the

*n*-dimensional vector:

*n*-dimensional vector:

###
**Definition 8**

The set of all eigenvalues of \(\mathcal {A}\) is denoted by \(\sigma (\mathcal {A})\) and it is called the spectral of \(\mathcal {A}\). Let \(\rho (\mathcal {A})=\max \{|\lambda |:\lambda \in \sigma (\mathcal {A})\}\). It is called the spectral radius of \(\mathcal {A}\).

###
**Definition 9**

###
**Definition 10**

###
**Definition 11**

(Bu et al. 2014) A tensor \(\mathcal {A}\in \mathbb {C}^{n_1\times \cdots \times n_k}\) is said to have rank-one if there exist nonzero \(a_i\in \mathbb {C}^{n_i}(i=1,\ldots ,k)\) such that \(\mathcal {A}= a_1\otimes a_2\otimes \cdots \otimes a_k\), where \(a_1\otimes a_2\otimes \cdots \otimes a_k\) is the segre outer product of \(a_1\in \mathbb {C}^{n_1},\ldots ,a_k\in \mathbb {C}^{n_k}\) with entries \(a_{i_1\ldots i_k}= (a_1)_{i_1}\ldots (a_k)_{i_k}\). The rank of a tensor \(\mathcal {A}\), denoted by \(rank(\mathcal {A})\), is defined to be the smallest \(r\) such that \(\mathcal {A}\) can be written as a sum of \(r\) rank-one tensors. Especially, if \(\mathcal {A}=0\), then \(rank(\mathcal {A})=0\).

###
**Lemma 1**

*Assume that*\(\mathcal {A}\in \mathbb {R}^{[m,n]}\)

*is an even-order symmetric tensor. The following conclusions hold for*\(\mathcal {A}\),

- (i)
\({\mathcal {A}}\)

*always has H-eigenvalues*. \(\mathcal {A}\)*is positive definite (positive semidefinite) if and**only if all of its H-eigenvalues are positive*(*nonnegative*). - (ii)
\(\mathcal {A}\)

*always has Z-eigenvalues*. \(\mathcal {A}\)*is positive definite*(*positive semidefinite*)*if and only if all of its Z-eigenvalues are positive*(*nonnegative*).

Next, we present some basic properties of permutation transformation of tensor as follows.

###
**Theorem 3**

*Let*\(\mathcal {A}\), \(\mathcal {B}\in \mathbb {C}^{[m,n]}~(\mathbb {R}^{[m,n]})\),

*and*\(\pi\)

*be a permutation on*\([n]\).

*Then*

- (i)
\(P_\pi (\mathcal {A}+\mathcal {B})=P_\pi (\mathcal {A})+P_\pi (\mathcal {B})\).

- (ii)
\(P_\pi (k\mathcal {A})=kP_\pi (\mathcal {A})\),

*where*\(k\in \mathbb {C}~(\mathbb {R})\). - (iii)
\(P_\pi (\mathcal {I})=\mathcal {I}\).

- (iv)
\(P_\pi ^{-1}(P_\pi (\mathcal {A}))=\mathcal {A}\).

- (v)
\(P_\pi (P_\pi ^{-1}(\mathcal {A}))=\mathcal {A}\).

- (vi)
\(\sigma (P_\pi (\mathcal {A}))=\sigma (\mathcal {A})\).

- (vii)
\(\rho (P_\pi (\mathcal {A}))=\rho (\mathcal {A})\).

- (viii)
\(rank(P_\pi (\mathcal {A}))=rank(\mathcal {A})\).

- (ix)
*If*\(\mathcal {A}\in \mathbb {S}^{[m,n]}\),*then*\(P_\pi (\mathcal {A})\in \mathbb {S}^{[m,n]}\). - (x)
If \(\mathcal {A}\)

*is a strictly diagonally**dominant tensor, then*\(P_\pi (\mathcal {A})\)*is also a strictly diagonally dominant tensor*. - (xi)
*If m is even and*\(\mathcal {A}\in \mathbb {S}^{[m,n]}\)*is positive definite*(*positive semidefinite*) tensor, then \(P_\pi (\mathcal {A})\in \mathbb {S}^{[m,n]}\) and is*also a positive definite*(*positive semidefinite*)*tensor*.

###
*Proof*

- (i)
\(P_\pi (\mathcal {A}+\mathcal {B})_{i_1\ldots i_m}=a_{\pi (i_1)\ldots \pi (i_m)}+b_{\pi (i_1)\ldots \pi (i_m)}=\big (P_\pi (\mathcal {A})+P_\pi (\mathcal {B})\big )_{i_1\ldots i_m}.\)

- (ii)
\(P_\pi (k\mathcal {A})_{i_1\ldots i_m}=ka_{\pi (i_1)\ldots \pi (i_m)}=kP_\pi (\mathcal {A})_{i_1\ldots i_m}.\)

- (iii)
Since \(P_\pi (\mathcal {I})_{i_1\ldots i_m}=\mathcal {I}_{\pi (i_1) \ldots \pi (i_m)}\) and \(\delta _{\pi (i_1)\ldots \pi (i_m)}=1\) if and only if \(\delta _{i_1\ldots i_m}=1\), then \(P_\pi (\mathcal {I})=\mathcal {I}\).

From the Definition 3, it is easy to obtain (iv) and (v) are hold.

- (vi)By Theorem 1, \(P_\pi (\mathcal {A})\) and \(\mathcal {A}\) are permutational similar. By Theorem 2.1 in Shao (2013),where \(\phi _{\mathcal {A}}(\lambda )\) is the characteristic polynomial of the tensor \(\mathcal {A}\). In Qi (2005), Qi has proved that a number \(\lambda \in \mathbb {C}\) is an eigenvalue of \(\mathcal {A}\) if and only if it is a root of \(\phi _{\mathcal {A}}(\lambda )\). Hence, similar tensors have the same eigenvalues. Then$$\phi _{P_\pi (\mathcal {A})}(\lambda )=\phi _{\mathcal {A}}(\lambda ),$$$$\sigma (\mathcal {A})=\sigma (P_\pi (\mathcal {A})).$$
- (vii)
It is easy to be got from the results of (vi).

- (viii)Let \(rank(\mathcal {A})=r\), and \(P_\pi (\mathcal {A})=(a_{\pi (i_1)\ldots \pi (i_m)})\), where \(\pi\) is a permutation on \([n]\).
**Case 1**. If \(r=1\), then there exists \(a_i\in \mathbb {C}^{n_i}~(i\in [m],a_i\ne 0)\) such thatwhich implies,$$\mathcal {A}=a_1\otimes a_2\otimes \cdots \otimes a_m$$Therefore,$$a_{i_1\ldots i_m}=(a_1)_{i_1}(a_2)_{i_2}\ldots (a_m)_{i_m}.$$Hence, \(P_\pi (\mathcal {A})={P_\pi (a_1)}\otimes {P_\pi (a_2)}\otimes \cdots \otimes {P_\pi (a_m)}\). Since \(P_\pi (a_i)\in \mathbb {C}^{n_i}\) and \(P_\pi (a_i)\ne 0\), then \(rank(P_\pi (\mathcal {A}))=1\).$$\begin{aligned} a_{\pi (i_1)\pi (i_2)\ldots \pi (i_m)}&=(a_1)_{\pi (i_1)}(a_2)_{\pi (i_2)}\ldots (a_m)_{\pi (i_m)}\\&=(P_\pi (a_1))_{i_1}(P_\pi (a_2))_{i_2}\ldots (P_\pi (a_m))_{i_m}.\end{aligned}$$**Case 2**. If \(r>1\), then \(\mathcal {A}\) can be written at least as a sum of \(r\) rank-one tensors. Let \(\mathcal {A}=\sum \nolimits _{i\in [r]}\mathcal {A}_i\), where \(\mathcal {A}_i\in \mathbb {C}^{[m,n]},rank(\mathcal {A}_i)=1,i\in [r]\). Then,By the results of Case 1, we have$$P_\pi (\mathcal {A})=\sum \limits _{i\in [r]}P_\pi (\mathcal {A}_i).$$Hence, \(rank(P_\pi (\mathcal {A}))\le r\). Next, we will prove that \(rank(P_\pi (\mathcal {A}))< r\) is impossible. Suppose that \(rank(P_\pi (\mathcal {A}))=r'<r\). Then \(P_\pi (\mathcal {A})\) can be written at least as a sum of \(r'\) rank-one tensors as follows$$rank(P_\pi (\mathcal {A}_i))=1,i\in [r].$$and \(\pi ^{-1}\) be the inverse transformation of \(\pi\) on \([n]\). Then from \((iv)\),$$P_\pi (\mathcal {A})=\sum \limits _{i\in [r']}\mathcal {D}_i,\quad where\quad \mathcal {D}_i\in \mathbb {C}^{[m,n]},\quad rank(\mathcal {D}_i)=1,i\in [r'],$$From case 1, \(rank(P_\pi ^{-1}(\mathcal {D}_i))=1\), so \(rank(\mathcal {A})\le r'< r\). It’s a contradiction. Therefore, \(rank(P_\pi (\mathcal {A}))=r\).$$\mathcal {A}=P_\pi ^{-1}(P_\pi (\mathcal {A}))= \sum \limits _{i\in [r']}P_{\pi ^{-1}}(\mathcal {D}_i).$$ - (ix)
It is easy to be proved from the definition of symmetric tensors.

- (x)Suppose that \(\mathcal {A}\) is a strictly diagonally dominant tensor, thenThus, \(P_\pi (\mathcal {A})\) is also a strictly diagonally dominant tensor.$$\begin{aligned}\mid P_\pi (\mathcal {A})_{ii\ldots i}\mid&=\mid \mathcal {A}_{ \pi (i)\pi (i)\ldots \pi (i)}\mid \\&> \mathop {\mathop {\sum }\limits _{i_2,\ldots , i_m\in [n],}}\limits _{{\delta _{\pi (i)i_2\ldots i_m=0}}} \mid a_{\pi (i)i_2\ldots i_m}\mid \\&= \mathop {\mathop {\sum }\limits _{i_2,\ldots , i_m\in [n],}}\limits _{\delta _{ii_2\ldots i_m=0}} \mid P_\pi (\mathcal {A})_{ii_2\ldots i_m}\mid,~\forall i\in [n]. \end{aligned}$$
- (xi)
Suppose that \(\mathcal {A}\) is positive definite (semidefinite), then all H-eigenvalues of \(\mathcal {A}\) are positive (nonnegative). By (vi) and (ix) of Theorem 3, \(P_\pi (\mathcal {A})\) is an even-order symmetric tensor, and \(\sigma (P_\pi (\mathcal {A}))=\sigma (\mathcal {A})\). From the results above, we know that all H-eigenvalues of \(P_\pi (\mathcal {A})\) are positive (nonnegative). Then, by Lemma 1, \(P_\pi (\mathcal {A})\) is positive definite (positive semidefinite).

###
*Remark 4*

By the results of (i) and (ii) of Theorem 3, we know that a permutation transformation of tensor is a linear transformation on \(\mathbb {C}^{[m,n]}~(\mathbb {R}^{[m,n]})\).

## Permutation transformation on some structure tensors

It is universally acknowledged that some structure tensors with good properties have been well studied, such as nonnegative tensor, symmetric tensor, positive definite (positive semidefinite) tensor, \(Z\)-tensor, (strong) \(M\)-tensor, Hankel tensor, \(P(P_0)\)-tensor, \(B(B_0)\)-tensor and \(H\)-tensor. In this section, we discuss the invariance under permutation transformations for some important structure tensors.

###
**Definition 12**

(Chang et al. 2008) A tensor \(\mathcal {A}=(a_{i_1\ldots i_m})\in \mathbb {R}^{[m,n]}\) is called a nonnegative tensor, denoted by \(\mathcal {A}\ge 0\), if each entry is nonnegative.

###
**Definition 13**

(Zhang et al. 2014) We call a tensor \(\mathcal {A}\) as an \(Z\)-tensor, if all of its off-diagonal entries are non-positive, which is equivalent to write \(\mathcal {A}=s\mathcal {I}-\mathcal {B}\), where \(s>0\) and \(\mathcal {B}\) is a nonnegative tensor.

From Definition 12, we easily get the following two lemmas.

###
**Lemma 2**

*Let*
\(\mathcal {A}\in \mathbb {R}^{[m,n]}\)
*be a nonnegative tensor,*
\(\pi\)
*be a permutation on*
\([n]\). *Then*
\(P_\pi (\mathcal {A})\)
*is also a nonnegative tensor*.

###
**Lemma 3**

*Let*
\(\mathcal {A}\in \mathbb {R}^{[m,n]}\)
*be an*
\(Z\)-*tensor, and*
\(\pi\)
*be a permutation on*
\([n]\). *Then*
\(P_\pi (\mathcal {A})\)
*is also an*
\(Z\)-*tensor.*

###
**Definition 14**

(Zhang et al. 2014) We call an \(Z\)-tensor \(\mathcal {A}=s\mathcal {I}-\mathcal {B}(\mathcal {B}\ge 0)\) as an \(M\)-tensor if \(s\ge \rho (\mathcal {B})\); we call it as a strong \(M\)-tensor if \(s>\rho (\mathcal {B})\), where \(\rho (\mathcal {B})\) is the spectral radius of \(\mathcal {B}\).

###
**Theorem 4**

*Let*
\(\mathcal {A}\in \mathbb {R}^{[m,n]}\)
*be an* (*strong*) \(M\)-*tensor, and*
\(\pi\)
*be a permutation on*
\([n]\). *Then*
\(P_\pi (\mathcal {A})\)
*is also an (strong)*
\(M\)-*tensor*.

###
*Proof*

###
**Definition 15**

- (i)
a \(P_0\)-tensor if for any nonzero vector \(x\in \mathbb {R}^n\), there exists \(i\in [n]\) such that \(x_i\ne 0\) and \(x_i(\mathcal {A}x^{m-1})_i\ge 0\);

- (ii)
a \(P\)-tensor if for any nonzero vector \(x\in \mathbb {R}^n\), \(\max \nolimits _{i\in [n]}x_i(\mathcal {A}x^{m-1})_i>0\).

###
**Lemma 4**

(Song and Qi 2014) *A symmetric tensor is a*
\(P(P_0)\)-*tensor if and only if it is positive (semi)definite.*

###
**Theorem 5**

*Let*
\(\mathcal {A}\in \mathbb {S}^{[m,n]}\)
*be an even-order*
\(P(P_0)\)-*tensor, and*
\(\pi\)
*be a permutation on*
\([n]\). *Then*
\(P_\pi (\mathcal {A})\)
*is also an even-order symmetric*
\(P(P_0)\)-*tensor*.

###
*Proof*

Suppose that \(\mathcal {A}\) is an even-order symmetric \(P(P_0)\)-tensor. According to Lemma 4, \(\mathcal {A}\) is positive (semi)definite. It follows from (ix) of Theorem 3 that \(P_\pi (\mathcal {A})\) is positive (semi)definite. Hence, by Lemma 4, \(P_\pi (\mathcal {A})\) is a \(P(P_0)\)-tensor. \(\square\)

###
**Definition 16**

###
**Theorem 6**

*Let*
\(\mathcal {A}\in \mathbb {R}^{[m,n]}\)
*be a*
\(B(B_0)\)-*tensor, and*
\(\pi\)
*be a permutation on*
\([n]\). *Then*
\(P_\pi (\mathcal {A})\)
*is also a*
\(B(B_0)\)-*tensor*.

###
*Proof*

###
**Definition 17**

###
**Theorem 7**

*Let*
\(\mathcal {A}\in \mathbb {C}^{[m,n]}\)
*be an*
\(H\)-*tensor, and*
\(\pi\)
*be a permutation on*
\([n]\). *Then*
\(P_\pi (\mathcal {A})\)
*is also an*
\(H\)-*tensor*.

###
*Proof*

## Canonical form of tensors

In this section, as an application of the permutation transformation of tensors, we would introduce some results which show that the canonical form theorem for matrices could be generalized to tensors. For the convenience discussion, we starts with the following definitions and lemmas.

Let \(\mathcal {A}\in \mathbb {R}^{[m,n]}\) be a nonnegative tensor. The directed graph \(G(\mathcal {A})\) (Chang et al. 2013; Friedland et al. 2013) associated to \(\mathcal {A}\) is the directed graph with vertices \(1,2,\ldots , n\) and an edge from \(i\) to \(j\) if and only if \(a_{ii_2\ldots i_m}\ne 0\) for some \(i_l=j,l=2,3,\ldots ,m.\)

###
**Definition 18**

(Chang et al. 2013; Friedland et al. 2013; Hu et al. 2014) A nonnegative tensor \(\mathcal {A}\in \mathbb {R}^{[m,n]}\) is called weakly irreducible if the associate directed graph \(G(\mathcal {A})\) is strongly connected. A tensor \(\mathcal {A}\) is said to be weakly irreducible if \(|\mathcal {A}|\) is weakly irreducible, where \(|\mathcal {A}|\) denote the tensor whose \((i_1,\ldots ,i_m)\)-th entry is defined as \(|a_{i_1\ldots i_m}|\).

###
**Definition 19**

###
**Definition 20**

###
**Definition 21**

###
**Lemma 5**

(Shao et al. 2013) *Let*
\(\mathcal {A}\in \mathbb {R}^{[m,n]}\)
*and*
\(m\ge 2\). *Then there exists positive integers*
\(r\ge 1\)
*and*
\(n_1,\ldots ,n_r\)
*with*
\(n_1+\cdots +n_r=n (r\ge 1)\)
*such that*
\(\mathcal {A}\)
*is permutational similar to some*
\((n_1+\cdots +n_r)\)-*lower triangular block tensor, where all the diagonal blocks*
\(\mathcal {A}_1,\ldots ,\mathcal {A}_r\)
*are weakly irreducible.*

###
**Lemma 6**

*Let*\(r\ge 2\)

*and*\(n_1,\ldots ,n_r\)

*be positive integers with*\(n_1+\cdots +n_r=n\).

*Let*\(\mathcal {A}\)

*be a*\((n_1+\cdots +n_r)\)-

*lower triangular block tensor with the diagonal blocks*\(\mathcal {A}_1,\ldots ,\mathcal {A}_r\).

*Then we have:*

*and thus*

*where*\(\phi _{\mathcal {A}}(\lambda )\)

*is the characteristic polynomial of the tensor*\(\mathcal {A}\).

From Theorem 1, Lemmas 5 and 6, we can easily obtain the following theorems.

###
**Theorem 8**

*Let*
\(\mathcal {A}\in \mathbb {R}^{[m,n]}\), *then there exists a permutation*
\(\pi\)
*on*
\([n]\)
*such that*
\(P_\pi (\mathcal {A})\)
*is a*
\((n_1+\cdots +n_r)\)-*lower triangular block tensor, where all the diagonal blocks*
\(\mathcal {A}_1,\ldots ,\mathcal {A}_r\)
*are weakly irreducible*.

###
*Remark 5*

Theorem 8 could be called the “canonical form” theorem and the \((n_1+\cdots +n_r)\)-lower triangular block tensor is called the “canonical form” of \(\mathcal {A}\).

###
**Theorem 9**

*Let*\(\mathcal {A}\in \mathbb {R}^{[m,n]}\)

*and*\(\overline{\mathcal {A}}\)

*be the “canonical form” of*\(\mathcal {A}\).

*Then*

###
*Proof*

###
*Example 1*

## Conclusions

In this paper, the definition of permutation transformation of tensors is introduced, and its properties are studied. As applications of permutation transformations, we give the canonical form theorem of tensors and a numerical example which show that some problems of higher dimension tensors can be translated into the corresponding problems of lower dimension weakly irreducible tensors by using the permutation transformation. There are many problems unsolved for permutation transformations of tensors and their applications. Hence, this paper is only a starting point for studying permutation transformations of tensors.

## Notes

## Declarations

### Authors' contributions

YTL, ZBL, QLL and QL are contributed equally to this work. All authors read and approved the final manuscript.

### Acknowledgements

The authors are grateful to the anonymous referees for their valuable comments, which helped improve the quality of the paper. This work was supported by the National Nature Science Foundation of China (Grant No. 11361074).

### Competing interests

The authors declare that they have no competing interests.

**Open Access**This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.

## Authors’ Affiliations

## References

- Brachat J, Comon P, Mourrain B, Tsigaridas E (2009) Symmetric tensor decomposition. In: Signal processing conference, 2009 17th European, pp 525–529Google Scholar
- Bu C, Zhang X, Zhou J, Wang W, Wei Y (2014) The inverse, rank and product of tensors. Linear Algebra Appl 446:269–280MathSciNetView ArticleMATHGoogle Scholar
- Chang K, Qi L, Zhang T (2013) A survey on the spectral theory of nonnegative tensors. Numer Linear Algebra Appl 20(6):891–912MathSciNetView ArticleMATHGoogle Scholar
- Chang KC, Pearson K, Zhang T et al (2008) Perron-frobenius theorem for nonnegative tensors. Commun Math Sci 6(2):507–520MathSciNetView ArticleMATHGoogle Scholar
- Comon P (1992) Independent component analysis. In: Lacoume J-L (ed) Higher-order statistics. Elsevier, Amsterdam, pp 29–38Google Scholar
- Comon P, Golub G, Lim LH, Mourrain B (2008) Symmetric tensors and symmetric tensor rank. SIAM J Matrix Anal Appl 30(3):1254–1279MathSciNetView ArticleMATHGoogle Scholar
- De Lathauwer L, De Moor B, Vandewalle J (2000) On the best rank-1 and rank-(\(r_1, r_2, \ldots, r_n\)) approximation of higher-order tensors. SIAM J Matrix Anal Appl 21(4):1324–1342MathSciNetView ArticleMATHGoogle Scholar
- Friedland S, Gaubert S, Han L (2013) Perron-frobenius theorem for nonnegative multilinear forms and extensions. Linear Algebra Appl 438(2):738–749MathSciNetView ArticleMATHGoogle Scholar
- Hu S, Huang Z, Qi L (2014) Strictly nonnegative tensors and nonnegative tensor partition. Sci China Math 57(1):181–195MathSciNetView ArticleMATHGoogle Scholar
- Kolda TG, Bader BW (2009) Tensor decompositions and applications. SIAM Rev 51(3):455–500ADSMathSciNetView ArticleMATHGoogle Scholar
- Li C, Wang F, Zhao J, Zhu Y, Li Y (2014) Criterions for the positive definiteness of real supersymmetric tensors. J Comput Appl Math 255:1–14MathSciNetView ArticleMATHGoogle Scholar
- Lim LH (2005) Singular values and eigenvalues of tensors: a variational approach. In: CAMSAP’05: proceeding of the IEEE international workshop on computational advances in multisensor adaptive pp 129–132Google Scholar
- Liu Y, Zhou G, Ibrahim NF (2010) An always convergent algorithm for the largest eigenvalue of an irreducible nonnegative tensor. J Comput Appl Math 235(1):286–292MathSciNetView ArticleMATHGoogle Scholar
- Ng M, Qi L, Zhou G (2009) Finding the largest eigenvalue of a nonnegative tensor. SIAM J Matrix Anal Appl 31(3):1090–1099MathSciNetView ArticleMATHGoogle Scholar
- Qi L (2005) Eigenvalues of a real supersymmetric tensor. J Symb Comput 40(6):1302–1324MathSciNetView ArticleMATHGoogle Scholar
- Robeva E (2016) Orthogonal decomposition of symmetric tensors. SIAM J Matrix Anal Appl 37(1):86–102MathSciNetView ArticleGoogle Scholar
- Shao JY (2013) A general product of tensors with applications. Linear Algebra Appl 439(8):2350–2366MathSciNetView ArticleMATHGoogle Scholar
- Shao JY, Shan HY, Zhang L (2013) On some properties of the determinants of tensors. Linear Algebra Appl 439(10):3057–3069MathSciNetView ArticleMATHGoogle Scholar
- Song Y, Qi L (2014) An initial study on p, p0, b and b0 tensors. arXiv preprint arXiv:14031118
- Yang Y, Yang Q (2010) Further results for Perron–Frobenius theorem for nonnegative tensors. SIAM J Matrix Anal Appl 31(5):2517–2530MathSciNetView ArticleMATHGoogle Scholar
- Zhang L, Qi L, Zhou G (2014) M-tensors and some applications. SIAM J Matrix Anal Appl 35(2):437–452MathSciNetView ArticleMATHGoogle Scholar
- Zhang T, Golub GH (2001) Rank-one approximation to high order tensors. SIAM J Matrix Anal Appl 23(2):534–550MathSciNetView ArticleMATHGoogle Scholar