Abstract
In this paper, a unified approach for various extended inverses of tensors, the generalized bilateral inverse of tensors via Einstein products, is introduced and we show that a number of known generalized tensor inverses can be regarded as special cases of this idea. Some characterizations of the CMP, DMP, and MPD inverse of tensors by using Einstein products are provided. The notion of generalized bilateral inverses’ dual and self-duality are investigated. In addition, the bilateral inverse solutions for singular linear tensor equations are studied.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Tensors are higher-dimensional generalizations of matrices and can thus be viewed as multidimensional array (Weiyang and Yimin 2016; Wei et al. 2018). Tensors have various applications, such as data mining (Eldén 2007), machine learning (Rabanser et al. 2017), computer vision (Cyganek and Gruszczyński 2014), automation systems (Zhao et al. 2017), neuroscience (Beckmann and Smith 2005) etc.
Let \( {\mathbb {C}}^{I_1\times \cdots \times I_M} \) denotes the set of all tensors of order M and their elements are denoted as \( {\mathcal {A}}=(a_{i_{1},i_{2},\cdots ,i_{M}})_{1\le i_j \le I_j}, j=1,\ldots ,M\). Suppose that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_M \times J_1\times \cdots \times J_N} \). Then \({\mathcal {A}}^{*}\in {\mathbb {C}}^{ J_1\times \cdots \times J_N \times I_1\times \cdots \times I_M} \) is a conjugate transpose of \({\mathcal {A}}\) and is defined as \( ({\mathcal {A}}^{*})_{j_1\cdots j_Ni_1 \cdots i_M}={\bar{a}}_{i_1\cdots i_Mj_1\cdots j_N},\) where the over-line stands for the conjugate of \( a_{i_1\cdots i_Mj_1\cdots j_N}\). If the tensor \( {\mathcal {A}} \) is real, then its transpose is represented by \( {\mathcal {A}}^{T}\).
Consider the Einstein product of two tensors, \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times K_1\times \cdots \times K_N} \) and \( {\mathcal {B}}\in {\mathbb {C}}^{K_1\times \cdots \times K_N \times J_1\times \cdots \times J_M}\). The Einstein product \( {\mathcal {A}}*_{N}{\mathcal {B}}\in {\mathbb {C}}^{K_1\times \cdots \times K_N \times J_1\times \cdots \times J_M} \) was defined as in Einstein (2007), using the operation via \(*_N\)
Suppose that \( {\mathcal {B}}\in {\mathbb {C}}^{K_1\times \cdots \times K_N} \). Thus,
Definition 1
Sun et al. (2016) Let \( {\mathcal {D}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N\times I_1\times \cdots \times I_N}\). Then the tensor \( {\mathcal {D}}\) is diagonal if \(({\mathcal {D}})_{i_1 \cdots i_N \times j_1 \cdots j_N} =0\) for \((i_1, \ldots , i_N ) \ne ( j_1, \ldots , j_N)\).
Suppose that \({\mathcal {I}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N\times I_1\times \cdots \times I_N}\) is the identity tensor. Then the tensor \({\mathcal {X}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N\times I_1\times \cdots \times I_N}\) is considered the inverse of tensor \({\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N\times I_1\times \cdots \times I_N}\) if it satisfies the condition \({\mathcal {X}}*_N{\mathcal {A}}={\mathcal {A}}*_N{\mathcal {X}}={\mathcal {I}}\) and it is represented by \({\mathcal {A}}^{-1}\) (see Brazell et al. 2013).
Suppose that \({\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times J_1\times \cdots \times J_M}\). If \( {\mathcal {X}}\in {\mathbb {C}}^{J_1\times \cdots \times J_M \times I_1\times \cdots \times I_N}\) satisfies \({\mathcal {A}}*_{M}{\mathcal {X}}*_N{\mathcal {A}}={\mathcal {A}}\), then \({\mathcal {X}}\) is referred to as an inner inverse of tensor \({\mathcal {A}}\). Alternatively, if \( {\mathcal {X}}*_N{\mathcal {A}}*_M{\mathcal {X}}={\mathcal {X}} \), then \( {\mathcal {X}}\) is referred to as an outer inverse of tensor \({\mathcal {A}}\). Throughout this paper, the following notations are established.
Furthermore, if \( {\mathcal {X}}\in {\mathcal {G}}_r({\mathcal {A}}):= {\mathcal {G}}_i({\mathcal {A}}) \cap {\mathcal {G}}_o({\mathcal {A}})\), then \({\mathcal {X}}\) is represented as the reflexive inverse of \({\mathcal {A}}\).
Definition 2
Sun et al. (2016, Definition 2.2) Suppose that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times J_1\times \cdots \times J_M}.\) The tensor \({\mathcal {X}}\in {\mathcal {G}}_r({\mathcal {A}}) \) that satisfies the following:
is referred to as the Moore-Penrose inverse of the tensor \({\mathcal {A}}\).
For \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times K_1\times \cdots \times K_N}, \) the null space \( N({\mathcal {A}}) \) and the range \( R({\mathcal {A}}) \) are defined by:
where \( {\mathcal {O}} \) is the zero tensor (see Ji and Wei 2018).
Let \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}.\) Define \( {\mathcal {A}}^{e}:={\mathcal {A}}^{e-1}*_N {\mathcal {A}},\quad for\quad e\ge 2.\)
Note that
In Ji and Wei (2018), the index of a tensor \({\mathcal {A}}\) is represented by \(index({\mathcal {A}})\) is defined as the smallest non-negative integer e such that \(R({\mathcal {A}}^{e+1})=R({\mathcal {A}}^e)\) or \(N({\mathcal {A}}^{e+1})=N({\mathcal {A}}^e)\).
Definition 3
Ji and Wei (2018, Theorem 3.3) The Drazin inverse of \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) with index(\({\mathcal {A}}\))=k, is the tensor \( {\mathcal {X}}\in {\mathcal {G}}_o\), which satisfies:
The Drazin inverse is represented by \({\mathcal {A}}^d\). For more information (see Sahoo et al. 2020; Du et al. 2019; Ma et al. 2019; Wang et al. 2023; Wang and Wei 2022; Sun et al. 2018; Bu et al. 2014).
Theorem 4
Wang et al. (2020, Theorem 1.1) Let \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}. \) Then \( {\mathcal {A}} \) can be represented as the sum of two tensors \( C_{\mathcal {A}} \) and \( N_{\mathcal {A}}, \) such that, \( {\mathcal {A}}=C_{\mathcal {A}} +N_{\mathcal {A}}, \) where \( index(C_{\mathcal {A}})\leqslant 1, N_{\mathcal {A}}\) is nilpotent and \( C_{\mathcal {A}}*_N N_{\mathcal {A}}=N_{\mathcal {A}}*_NC_{\mathcal {A}}={\mathcal {O}}.\)
The tensors \( C_{\mathcal {A}} \) and \( N_{\mathcal {A}} \) are referred to as the core part and the nilpotent part of \({\mathcal {A}}\), respectively. It is readily seen that \( C_{\mathcal {A}}={\mathcal {A}}*_N{\mathcal {A}}^d*_N{\mathcal {A}}.\)
Let \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}.\) If the following conditions hold, the unique matrix \( {\mathcal {X}} \in {\mathcal {G}}_o({\mathcal {A}})\) is referred to as the DMP inverse of \({\mathcal {A}}\) and is represented by \({\mathcal {A}}^{d,\dagger }\) Wang et al. (2020, Theorem 2.2).
Note that \( {\mathcal {A}}^{d,\dagger }={\mathcal {A}}^{d}*_N{\mathcal {A}}*_N{\mathcal {A}}^{\dagger }\).
By employing the same approach as in Wang et al. (2020, Theorem 2.2), the following holds.
Proposition 1
Suppose that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) with index(\( {\mathcal {A}})=k\). Then \( {\mathcal {X}}= {\mathcal {A}}^{\dagger ,d}={\mathcal {A}}^{\dagger }*_N{\mathcal {A}}*_N{\mathcal {A}}^{d}\) is the unique solution of the following:
Definition 5
Let \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) with index(\( {\mathcal {A}})=k\). Then The MPD inverse of \( {\mathcal {A}}\), represented by \( {\mathcal {A}}^{\dagger , d} \), the definition is as follows
Let \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}.\) If the following conditions hold, the unique matrix \( {\mathcal {X}} \in {\mathcal {G}}_o({\mathcal {A}})\) is referred to as the CMP inverse of \({\mathcal {A}}\) and is represented by \({\mathcal {A}}^{c,\dagger }={\mathcal {A}}^{\dagger }*_NC_{{\mathcal {A}}}*_N{\mathcal {A}}^{\dagger }\)Wang et al. (2020).
2 CMP and DMP generalized inverses of tensors
This section introduces novel characterizations of CMP, DMP, and MPD inverses of tensors.
The theorem below demonstrates that one of the conditions in Wang et al. (2020, Theorem 2.7) is unnecessary.
Theorem 6
Suppose that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}. \) Then \( {\mathcal {X}}= {\mathcal {A}}^{c,\dagger }\) is the unique solution of the following:
Proof
It is obvious that the tensor \( {\mathcal {X}}={\mathcal {A}}^{c,\dagger } \) satisfies the system (4). Assume that two tensors \({\mathcal {X}}_{1}\) and \( {\mathcal {X}}_{2} \) satisfy (4), then
\(\square \)
A novel characterization of DMP inverses of tensors, which does not rely on the index of \({\mathcal {A}}\), is presented in the following (see Wang et al. (2020, Theorem 2.2)).
Theorem 7
Suppose that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}.\) Then \( {\mathcal {X}}= {\mathcal {A}}^{d,\dagger }\) is the unique solution of the following:
Proof
It is evident that the tensor \( {\mathcal {X}}={\mathcal {A}}^{d}*_NA*_NA^{\dagger } \) satisfies the system (5). Assume that two tensors \({\mathcal {X}}_{1}\) and \( {\mathcal {X}}_{2} \) satisfy (5), then
\(\square \)
By employing the same approach as in the proof of Theorem 7, the following holds.
Corollary 8
Let \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}.\) Then \( {\mathcal {X}}= {\mathcal {A}}^{\dagger ,d}\) is the unique solution of the following:
In the following theorem we state a new characterization of \(A^{c,\dagger }\).
Theorem 9
Let \({\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}.\) Then \( {\mathcal {X}}={\mathcal {A}}^{c,\dagger }\) is the unique solution satisfies in 6.
Proof
By (3),
where \({\mathcal {U}}=({\mathcal {A}}^{\dagger })^{*}*_N{\mathcal {A}}^{d,\dagger }\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N} \) and \( {\mathcal {V}}={\mathcal {A}}^{\dagger ,d}*_N({\mathcal {A}}^{\dagger })^{*} \in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\). Therefore, by Stanimirović et al. (2020, Lemma 2.2 (a)), we obtain that \(R({\mathcal {X}})\subseteq R({\mathcal {A}}^{*})\) and \( R({\mathcal {X}}^{*})\subseteq R({\mathcal {A}})\) are equivalent to \({\mathcal {A}}^{c,\dagger }={\mathcal {A}}^{*}*_N{\mathcal {U}}\) and \( {\mathcal {A}}^{c,\dagger }={\mathcal {V}}*_N{\mathcal {A}}^{*}\), respectively. By the Eq. (7), it is clear to see that \( {\mathcal {A}}^{c,\dagger } \) satisfies (6). Assume possible, there exist \( {\mathcal {X}}_{1} \) and \({\mathcal {X}}_{2}\) such that \({\mathcal {X}}_1 \ne {\mathcal {X}}_2\), we have that
where \({\mathcal {U}}_1,{\mathcal {U}}_2, {\mathcal {V}}_1, {\mathcal {V}}_2 \in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\). Let
It then follows from (6), (8), (9) and (10),
By Panigrahy et al. (2020, Lemma 3.7), we have
Therefore, \( {\mathcal {X}}*_N{\mathcal {A}}={\mathcal {O}}. \) Meanwhile,
by Panigrahy et al. (2020, Remark 3.8), yields that \( {\mathcal {X}}={\mathcal {O}}\), and hence \( {\mathcal {X}}_1={\mathcal {X}}_2\). Therefore, we conclude that unique tensor \( {\mathcal {X}}={\mathcal {A}}^{c,\dagger } \) satisfying (6). \(\square \)
Corollary 10
Let \({\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\). If there exist \( {\mathcal {X}}\) and \({\mathcal {Z}} \) in \( {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N} \) satisfying
then \({\mathcal {X}}={\mathcal {A}}^{c,\dagger }\).
By employing the same approach as in the proof of Theorem 9, the following holds.
Corollary 11
Let \({\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}.\) Then \( {\mathcal {X}}={\mathcal {O}}\) is the unique solution satisfies in 11.
By using Corollary 11, we characterize \({\mathcal {A}}^{c,\dagger }\) by two relations.
Theorem 12
Let \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\). Then \( {\mathcal {X}}={\mathcal {A}}^{c,\dagger }\) is the unique solution satisfies in 12.
In the following theorem, we characterize \({\mathcal {A}}^{d,\dagger }\) by the relations in 13.
Theorem 13
Let \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\). Then \( {\mathcal {X}}={\mathcal {A}}^{d,\dagger }\) is the unique solution satisfies in 13.
Proof
It is clear that \({\mathcal {A}}^{\dagger }*_N{\mathcal {A}}^{d,\dagger }*_N{\mathcal {A}}={\mathcal {A}}^{\dagger ,d}\), \(R({\mathcal {A}}^{d,\dagger }) = R({\mathcal {A}}*_N {\mathcal {A}}^{d} *_N{\mathcal {A}}^{\dagger }) \subseteq R({\mathcal {A}})\), and \(R(({\mathcal {A}}^{d,\dagger })^*) = R(({\mathcal {A}}^{d} *_N{\mathcal {A}}*_N{\mathcal {A}}^{\dagger })^*) = R(({\mathcal {A}}^{\dagger })^{*} *_N{\mathcal {A}}^{*} *_N({\mathcal {A}}^{d})^*) \subseteq R(({\mathcal {A}}^{\dagger })^*)= R({\mathcal {A}})\). That is, we have proved that \( {\mathcal {A}}^{d,\dagger }\) satisfies (13). By Stanimirović et al. (2020, Lemma 2.2 (a)), from (13), we can assume that \({\mathcal {X}}=({\mathcal {A}}^{\dagger })^{*}*_N{\mathcal {U}}\) and \({\mathcal {X}}={\mathcal {V}}*_N{\mathcal {A}}^{*}\) for some \({\mathcal {U}}, {\mathcal {V}} \in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\).
Assume possible, there exist \( {\mathcal {X}}_{1} \) and \({\mathcal {X}}_{2}\) such that \({\mathcal {X}}_1 \ne {\mathcal {X}}_2\) and
where \({\mathcal {U}}_1, {\mathcal {U}}_2, {\mathcal {V}}_1, {\mathcal {V}}_2 \in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\). Let
It then follows from (13), (14), (15) and (16),
By Panigrahy et al. (2020, Lemma 3.7), we have that
We obtain \( {\mathcal {X}}*_N{\mathcal {A}}={\mathcal {O}}\). Meanwhile, we find
by Panigrahy et al. (2020, Remark 3.8), we obtain \( {\mathcal {X}}={\mathcal {O}}\)., and hence \( {\mathcal {X}}_1={\mathcal {X}}_2\). Therefore, we conclude that unique tensor \({\mathcal {X}}={\mathcal {A}}^{d,\dagger } \) satisfying (13). \(\square \)
Corollary 14
Let \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\). If there exist \( {\mathcal {X}}, {\mathcal {Z}} \in M_{n}({\mathbb {C}}) \) satisfying
then \({\mathcal {X}}={\mathcal {A}}^{d,\dagger }\).
By using Corollary 11, we characterize \({\mathcal {A}}^{d,\dagger }\) by two relations.
Theorem 15
Let \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\). Then \( {\mathcal {X}}={\mathcal {A}}^{d,\dagger }\) is the unique solution satisfies in 18.
By employing the same approach as in the proof of Theorem 13, the following hold.
Theorem 16
Suppose that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\). Then the solution satisfies in 19.
is unique and is given by \( {\mathcal {X}}={\mathcal {A}}^{\dagger ,d}\).
Corollary 17
Let \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\). If there exist \( {\mathcal {X}}, {\mathcal {Z}} \in M_{n}({\mathbb {C}}) \) satisfying
then \({\mathcal {X}}={\mathcal {A}}^{\dagger ,d}\).
By using Corollary 11, we characterize \({\mathcal {A}}^{\dagger ,d}\) by two relations.
Theorem 18
Let \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\). Then the solution satisfies in 21.
is unique and is given by \( {\mathcal {X}}={\mathcal {A}}^{\dagger ,d}\).
First, we obtain the null space and the range of the outer inverse of the tensor \({\mathcal {A}}.\)
Lemma 1
Let \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times J_1\times \cdots \times J_M}\) and \( {\mathcal {X}}\in {\mathcal {G}}_o({\mathcal {A}})\). Then
Proof
Given that \({\mathcal {X}}*_N {\mathcal {A}}\) and \({\mathcal {A}}*_M{\mathcal {X}}\) are projections, we can conclude that:
\(\square \)
Lemma 2
(Panigrahy and Mishra (2022, Lemma 2.3)) If \({\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) is a Hermitian idempotent tensor, then \( {\mathcal {A}}^{\dagger }={\mathcal {A}}\).
Remark 1
Let \({\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) is Hermitian idempotent tensor. Then \( {\mathcal {C}}_{\mathcal {A}}={\mathcal {A}}^{d,\dagger }={\mathcal {A}}^{\dagger ,d}\).
Theorem 19
Let \({\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) with index(\({\mathcal {A}})=k\). The solution to the system of following:
is unique and is given by \( {\mathcal {X}}= C_{\mathcal {A}}\).
Proof
It is evident that the tensor \( {\mathcal {X}}= C_{\mathcal {A}} \) satisfies the system (22). Assume that two tensors \({\mathcal {X}}_{1}\) and \( {\mathcal {X}}_{2} \) satisfy (22), then by Behera et al. (2020, Lemma 3.1), we have
\(\square \)
Next result gives the aforementioned relationships in terms of mainly the core part of the tensor \({\mathcal {A}}\).
Theorem 20
Let \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) with index(\({\mathcal {A}})=k\). Then
-
(i)
\({\mathcal {A}}^{d,\dagger }*_NC_{\mathcal {A}}=C_{\mathcal {A}}*_N{\mathcal {A}}^{d,\dagger } \) if and only if \({\mathcal {A}}^{k+1}*_N{\mathcal {A}}^{\dagger }={\mathcal {A}}^{k} \).
-
(ii)
\({\mathcal {A}}^{\dagger ,d}*_NC_{\mathcal {A}}=C_{\mathcal {A}}*_N{\mathcal {A}}^{\dagger ,d}\) if and only if \({\mathcal {A}}^{\dagger }*_N{\mathcal {A}}^{k+1}={\mathcal {A}}^{k} \).
-
(iii)
\(C_{\mathcal {A}}={\mathcal {A}}^{d, \dagger }*_N{\mathcal {A}} \) if and only if \({\mathcal {A}}^k={\mathcal {A}}^{k+1}\).
-
(iv)
\(C_{\mathcal {A}}={\mathcal {A}}^{\dagger ,d}*_N{\mathcal {A}} \) if and only if \({\mathcal {A}}^\dagger *_N{\mathcal {A}}^{k} = {\mathcal {A}}^{k}\).
Proof
(i) By Ji and Wei (2018, Theorem 3.4 (1)) and Lemma 1, we have
(ii) and (iii) are similar to part (i).
(iv)
By Ji and Wei (2018, Theorem 3.4 (1)), we can conclude that
Therefore, \( R(C_{\mathcal {A}})=R({\mathcal {A}}^{k}) \). We obtain \( R({\mathcal {A}}^{k})\subseteq N({\mathcal {I}}-{\mathcal {A}}^{\dagger }) \Leftrightarrow {\mathcal {A}}^\dagger *_N{\mathcal {A}}^{k} = {\mathcal {A}}^{k}.\) \(\square \)
Hartwig and Spindelböck decomposition of tensor \( {\mathcal {A}} \) arrived at the following lemma.
Lemma 3
Wang et al. (2020, Lemma 1.3) Let \({\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}.\) Then there exist unitary \({\mathcal {U}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) such that
where \( \Sigma \in {\mathbb {C}}^{R_1\times \cdots \times R_N \times R_1\times \cdots \times R_N}\) is a diagonal tensor of singular values of tensor \({\mathcal {A}}\), and the tensors \( {\mathcal {K}}\in {\mathbb {C}}^{R_1\times \cdots \times R_N \times R_1\times \cdots \times R_N},\quad {\mathcal {L}} \in {\mathbb {C}}^{R_1\times \cdots \times R_N \times (I_1- R_1)\times \cdots \times (I_N-R_N)}\) satisfy:
Using the same approach as described in the proof of Wang et al. (2020, Theorem 2.3), the following holds.
Corollary 21
Suppose that \({\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) is in the form of (23). Then
where \({\tilde{\Sigma }}={\mathcal {K}}*_N(\Sigma *_N{\mathcal {K}})^{d}\).
We extend the recently obtained properties by using CMP inverse to the tensor (see Mehdipour and Salemi (2018, p. 4 (9))).
Theorem 22
Let \({\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) be of the form (23). Then
Proof
Suppose that \({\mathcal {A}} \) is expressed as shown in (23) and
By Wang et al. (2020, p. 7(2.6)) and (24), we have that
According to Definition 2, it is straightforward to calculate the first equation, which states that.
Moreover, the second equation
The third equation follows from
The fourth equation follows from
The tensor \( {\mathcal {X}} \) fulfills four equations. Assume that both \( {\mathcal {W}} \) and \( {\mathcal {Z}} \) also satisfy four equations each. In order to demonstrate the uniqueness, we need to show that
\(\square \)
By using Theorem 22, we conclude the following.
Theorem 23
Suppose that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N} \) be as in (23). Then \( {\mathcal {A}}^{\dagger ,d}*_N({\mathcal {A}}^{c,\dagger })^{\dagger }=({\mathcal {A}}^{c,\dagger })^{\dagger }*_N{\mathcal {A}}^{d,\dagger }\) if and only if the following conditions hold.
-
1.
\({\mathcal {K}}^{*}*_N{\tilde{\Delta }} *_N{\mathcal {K}}=({\tilde{\Sigma }})^{\dagger }*_N{\tilde{\Sigma }}\),
-
2.
\({\mathcal {L}}^{*}*_N{\tilde{\Delta }} =0\),
where \( {\tilde{\Delta }} ={\tilde{\Sigma }}*_N({\tilde{\Sigma }})^{\dagger }\).
Proof
By Wang et al. (2020, Theorem 2.3), (24), (25) and (26), we have
Then \( {\mathcal {A}}^{\dagger ,d}*_N({\mathcal {A}}^{c,\dagger })^{\dagger }=({\mathcal {A}}^{c,\dagger })^{\dagger }*_N{\mathcal {A}}^{d,\dagger }\) if and only if the following conditions hold.
Note that the Eq. (27) and the Part 1 of Theorem 23 are equivalent. Since using (24), by left-multiplying the Eqs. (28) and (29) by \( {\mathcal {K}} \) and \( {\mathcal {L}}\), respectively, we obtain \({\tilde{\Sigma }}*_N({\tilde{\Sigma }})^{\dagger }*_N {\mathcal {L}}= {\mathcal {O}}\), equivalent to \({\mathcal {L}}^{*}*_N{\tilde{\Sigma }}*_N({\tilde{\Sigma }})^{\dagger }={\mathcal {O}}\), that is the Part 2 of Theorem 23. \(\square \)
3 Generalized bilateral inverse of tensor via Einstein product
In this section, we expand upon the recently introduced concept of a generalized bilateral inverse for a tensor \({\mathcal {A}}\) using the Einstein product. Furthermore, we demonstrate that certain well-known generalized inverses can be viewed as specific instances of the generalized bilateral inverses for tensors (see Kheirandish and Salemi 2023).
Definition 24
Let \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times J_1\times \cdots \times J_M}\) and let \( {\mathcal {X}}_1, {\mathcal {X}}_2 \in {\mathcal {G}}_i({\mathcal {A}}) \cup {\mathcal {G}}_o({\mathcal {A}}).\) Then \({\mathcal {X}}_1*_N{\mathcal {A}}*_M{\mathcal {X}}_2\) is referred to as generalized bilateral inverse of tensor \({\mathcal {A}}\).
We will now present a theorem that characterizes the generalized bilateral inverses of tensors.
Theorem 25
Suppose that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times J_1\times \cdots \times J_M}\) and suppose that \( {\mathcal {X}}_1\in {\mathcal {G}}_o({\mathcal {A}}) \) and \({\mathcal {X}}_2 \in {\mathcal {G}}_i({\mathcal {A}}).\) The unique solution to the system of following:
is given by \({\mathcal {X}}={\mathcal {X}}_1*_N{\mathcal {A}}*_M {\mathcal {X}}_2.\)
Proof
Assume that \({\mathcal {X}}={\mathcal {X}}_1*_N {\mathcal {A}}*_M {\mathcal {X}}_2 \) is a solution. Then
Suppose that two tensors \({\mathcal {W}}\) and \( {\mathcal {Z}} \) satisfy (30), then
\(\square \)
Using the same approach as described in the proof of Theorem 25, the following holds.
Corollary 26
Suppose that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times J_1\times \cdots \times J_M}\) and suppose that \( {\mathcal {X}}_1\in {\mathcal {G}}_o({\mathcal {A}}) \) and \({\mathcal {X}}_2 \in {\mathcal {G}}_i({\mathcal {A}}).\) The unique solution to the system of following:
is given by \({\mathcal {X}}={\mathcal {X}}_2*_N{\mathcal {A}}*_M {\mathcal {X}}_1.\)
The following proposition demonstrates that certain well-known generalized inverses of tensors can be regarded as generalized bilateral inverses of tensors.
Proposition 2
Suppose that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\). Then
Next, will define the dual of the generalized bilateral inverse for tensors in the following manner:
Definition 27
Suppose that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) and suppose that \( {\mathcal {X}}_1, {\mathcal {X}}_2 \in {\mathcal {G}}_i({\mathcal {A}}) \cup {\mathcal {G}}_o({\mathcal {A}}).\) Then the dual of generalized bilateral inverse of tensor \({\mathcal {X}}_1 *_N{\mathcal {A}}*_N {\mathcal {X}}_2\) is denoted by
and \({\mathcal {X}}_1*_N {\mathcal {A}}*_N {\mathcal {X}}_2\) is called self dual, if \({\mathcal {X}}_1*_N {\mathcal {A}}*_N {\mathcal {X}}_2={\mathcal {X}}_2 *_N{\mathcal {A}}*_N {\mathcal {X}}_1.\)
Now, we extend the recently obtained properties in Kheirandish and Salemi (2023) for tensors. Let \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}, {\mathcal {X}}_1\in {\mathcal {G}}_o({\mathcal {A}}) \) and \({\mathcal {X}}_2 \in {\mathcal {G}}_i({\mathcal {A}}).\) The following theorem presents the necessary and sufficient conditions for a generalized bilateral inverse of tensors \({\mathcal {X}}_1*_N{\mathcal {A}}*_N{\mathcal {X}}_2\) to be self-dual.
Theorem 28
Suppose that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}, {\mathcal {X}}_1\in {\mathcal {G}}_o({\mathcal {A}}) \) and \({\mathcal {X}}_2 \in {\mathcal {G}}_i({\mathcal {A}}).\) Then, the following statements are equivalent.
-
(i)
\({\mathcal {X}}_1*_N{\mathcal {A}}*_N{\mathcal {X}}_2\) is self dual,
-
(ii)
\({\mathcal {X}}_1={\mathcal {X}}_1*_N{\mathcal {A}}*_N{\mathcal {X}}_2={\mathcal {X}}_2*_N{\mathcal {A}}*_N{\mathcal {X}}_1,\)
-
(iii)
\( N({\mathcal {A}}*_N{\mathcal {X}}_{2})\subseteq N({\mathcal {X}}_{1}) \) and \( R({\mathcal {X}}_{1})\subseteq R({\mathcal {X}}_{2}*_N{\mathcal {A}}). \)
Proof
\(((i)\rightarrow (ii))\) Assume \({\mathcal {X}}_1*_N{\mathcal {A}}*_N{\mathcal {X}}_2={\mathcal {X}}_2*_N{\mathcal {A}}*_N{\mathcal {X}}_1.\) Since \({\mathcal {A}}*_N{\mathcal {X}}_2*_N{\mathcal {A}}={\mathcal {A}}\) and \({\mathcal {X}}_1*_N{\mathcal {A}}*_N{\mathcal {X}}_1={\mathcal {X}}_1,\) we obtain that
Then \({\mathcal {X}}_1={\mathcal {X}}_1*_N{\mathcal {A}}*_N{\mathcal {X}}_2={\mathcal {X}}_2*_N{\mathcal {A}}*_N{\mathcal {X}}_1.\)
\(((ii)\rightarrow (iii))\) Since \({\mathcal {X}}_1={\mathcal {X}}_1*_N{\mathcal {A}}*_N{\mathcal {X}}_2\), we obtain that \( N({\mathcal {A}}*_N{\mathcal {X}}_{2})\subseteq N({\mathcal {X}}_{1}*_N{\mathcal {A}}*_N{\mathcal {X}}_2)=N({\mathcal {X}}_{1}). \) Also, since \({\mathcal {X}}_1={\mathcal {X}}_2*_N{\mathcal {A}}*_N{\mathcal {X}}_1,\) we obtain that \( R({\mathcal {X}}_{1})=R({\mathcal {X}}_{2}*_N{\mathcal {A}}*_N{\mathcal {X}}_1)\subseteq R({\mathcal {X}}_{2}*_N{\mathcal {A}}). \)
\(((iii)\rightarrow (i))\) Using Lemma 1, we can see that \(R(I-{\mathcal {A}}*_N{\mathcal {X}}_{2})=N({\mathcal {A}}*_N{\mathcal {X}}_{2})\) and \(N({\mathcal {A}}*_N{\mathcal {X}}_{2})\subseteq N({\mathcal {X}}_{1})\). Therefore, \(R(I-{\mathcal {A}}*_N{\mathcal {X}}_{2})\subseteq N({\mathcal {X}}_{1})\) which implies that \({\mathcal {X}}_1*_N(I-{\mathcal {A}}*_N{\mathcal {X}}_2)=0.\) Hence, we have \({\mathcal {X}}_1={\mathcal {X}}_1*_N{\mathcal {A}}*_N{\mathcal {X}}_2\). Similarly, using Lemma 1, we arrive \(R({\mathcal {X}}_{2}*_N{\mathcal {A}})=N(I-{\mathcal {X}}_{2}*_N{\mathcal {A}})\) and \( R({\mathcal {X}}_{1})\subseteq R({\mathcal {X}}_{2}*_N{\mathcal {A}})=N(I-{\mathcal {X}}_{2}*_N{\mathcal {A}})\). This implies that \((I-{\mathcal {X}}_2*_N{\mathcal {A}})*_N{\mathcal {X}}_1=0.\) Therefore we have \({\mathcal {X}}_1={\mathcal {X}}_2*_N{\mathcal {A}}*_N{\mathcal {X}}_1\), which completes the proof. \(\square \)
Let \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N},~ {\mathcal {X}}_1={\mathcal {A}}^D,~ {\mathcal {X}}_2={\mathcal {A}}^{\dagger }\). By Theorem 28, Lemma 1, Ji and Wei (2018, Theorem 3.4 (1)) and Sahoo et al. (2020, Theorem 3.7 a(i)), we deduce the following.
Proposition 3
Suppose that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) with index(\({\mathcal {A}})=k\). Then, the following statements are equivalent.
-
(i)
\({\mathcal {A}}^{d}*_N{\mathcal {A}}*_N{\mathcal {A}}^{\dagger }={\mathcal {A}}^{\dagger }*_N{\mathcal {A}}*_N{\mathcal {A}}^{d},\)
-
(ii)
\( {\mathcal {A}}^{d}={\mathcal {A}}^{d,\dagger }={\mathcal {A}}^{\dagger ,d}\),
-
(iii)
\(N({\mathcal {A}}^{*}) \subseteq N({\mathcal {A}}^{k}) \) & \(R({\mathcal {A}}^{k}) \subseteq R({\mathcal {A}}^{*}).\)
The following theorem states the necessary and sufficient conditions for the generalized bilateral inverse of tensors to be self-dual.
Theorem 29
Suppose that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}, {\mathcal {X}}_2\in {\mathcal {G}}_o({\mathcal {A}})\) and \({\mathcal {X}}_1 \in {\mathcal {G}}_i({\mathcal {A}}) \cup {\mathcal {G}}_o({\mathcal {A}}).\) Then \({\mathcal {X}}_1*_N{\mathcal {A}}*_N{\mathcal {X}}_2\) is self dual, \({\mathcal {X}}_1*_N{\mathcal {A}}*_N{\mathcal {X}}_2=({\mathcal {X}}_1*_N{\mathcal {A}}*_N{\mathcal {X}}_2)'={\mathcal {X}}_2*_N{\mathcal {A}}*_N{\mathcal {X}}_1, \) if and only if \( N({\mathcal {X}}_{2})\subseteq N({\mathcal {X}}_{2}*_N{\mathcal {A}}*_N{\mathcal {X}}_{1}) \) and \( R({\mathcal {X}}_{1}*_N{\mathcal {A}}*_N{\mathcal {X}}_{2})\subseteq R({\mathcal {X}}_{2}). \)
Proof
Assume \({\mathcal {X}}_1*_N{\mathcal {A}}*_N{\mathcal {X}}_2={\mathcal {X}}_2*_N{\mathcal {A}}*_N{\mathcal {X}}_1.\) Since \({\mathcal {X}}_2*_N{\mathcal {A}}*_N{\mathcal {X}}_2={\mathcal {X}}_2\) and by Lemma 1, we obtain the following relations:
Therefore,
Conversely, we know that Eqs. (31), (32), (33), and (34) are equivalent. Therefore, \({\mathcal {X}}_2*_N{\mathcal {A}}*_N{\mathcal {X}}_{1}={\mathcal {X}}_2*_N{\mathcal {A}}*_N{\mathcal {X}}_{1}*_N{\mathcal {A}}*_N{\mathcal {X}}_{2}\) and \( {\mathcal {X}}_{1}*_N{\mathcal {A}}*_N{\mathcal {X}}_2={\mathcal {X}}_{2}*_N{\mathcal {A}}*_N{\mathcal {X}}_{1}*_N{\mathcal {A}}*_N{\mathcal {X}}_2.\) Therefore, \( {\mathcal {X}}_{1}*_N{\mathcal {A}}*_N{\mathcal {X}}_2={\mathcal {X}}_{2}*_N{\mathcal {A}}*_N{\mathcal {X}}_{1}. \) \(\square \)
Theorem 30
Suppose that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\). Then
-
(i)
\({\mathcal {A}}^{d}={\mathcal {A}}^{d}*_N{\mathcal {A}}*_N{\mathcal {A}}^{\dagger } \) if and only if \(({\mathcal {A}}^{d}*_N{\mathcal {A}}*_N{\mathcal {A}}^{\dagger })' ={\mathcal {A}}^{c,\dagger } \).
-
(ii)
\({\mathcal {A}}^{d}={\mathcal {A}}^{\dagger }*_N{\mathcal {A}}*_N{\mathcal {A}}^{d} \) if and only if \(({\mathcal {A}}^{\dagger }*_N{\mathcal {A}}*_N{\mathcal {A}}^{d})'={\mathcal {A}}^{c, \dagger } \).
Proof
(i)
(ii)
\(\square \)
The remark below demonstrates that the dual of a generalized bilateral inverse \({\mathcal {X}}_1*_N{\mathcal {A}}*_N{\mathcal {X}}_2\) is closely linked to \({\mathcal {X}}_1\) and \({\mathcal {X}}_2\).
Remark 2
Let \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}.\) Using Definition 27 and Proposition 2(iii)-(iv), it follows that \( {\mathcal {A}}^{c\dagger }={\mathcal {A}}^{\dagger }*_N{\mathcal {A}}*_N{\mathcal {A}}^{d,\dagger }={\mathcal {A}}^{\dagger ,d}*_N{\mathcal {A}}*_N{\mathcal {A}}^{\dagger }.\) But
The theorem below presents the necessary and sufficient conditions for the generalized bilateral inverses of tensor \({\mathcal {A}}^{\dagger }*_N{\mathcal {A}}*_N{\mathcal {A}}^{d,\dagger }\) to be self dual.
Theorem 31
Suppose that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) with index(\({\mathcal {A}})=k\). The following statements are equivalent:
-
(i)
\( {\mathcal {A}}^{\dagger }*_N{\mathcal {A}}*_N{\mathcal {A}}^{d,\dagger }={\mathcal {A}}^{d,\dagger }*_N{\mathcal {A}}*_N{\mathcal {A}}^{\dagger }\).
-
(ii)
\( R({\mathcal {A}}^{k})\subseteq R({\mathcal {A}}^{*})\) & \( N({\mathcal {A}}^{*}) \subseteq N({\mathcal {A}}^{k}*_NA^{\dagger })\).
Proof
From Theorem 28, we can conclude that \( {\mathcal {A}}^{\dagger }*_N{\mathcal {A}}*_N{\mathcal {A}}^{d,\dagger }={\mathcal {A}}^{d,\dagger }*_N{\mathcal {A}}*_N{\mathcal {A}}^{\dagger }\) if and only if \( R({\mathcal {A}}^{d,\dagger })\subseteq R({\mathcal {A}}^{\dagger }*_N{\mathcal {A}})\) and \(N({\mathcal {A}}*_N{\mathcal {A}}^{\dagger })\subseteq N({\mathcal {A}}^{d,\dagger })\). By applying Lemma 1, \(R({\mathcal {A}}^{\dagger }*_N{\mathcal {A}}) =R({\mathcal {A}}^{\dagger })\) and \(N({\mathcal {A}}^{\dagger })=N({\mathcal {A}}*_N{\mathcal {A}}^{\dagger })\). Furthermore, according to Sahoo et al. (2020, the first part Theorem 3.7), we can conclude that \( R({\mathcal {A}}^{\dagger })=R({\mathcal {A}}^{*})\) and \(N({\mathcal {A}}^{\dagger })=N({\mathcal {A}}^{*}) \). Moreover, by Ji and Wei (2018, Theorem 3.4 (1)) and Behera et al. (2020, Lemma 3.1), we have that
Then, we have \( R({\mathcal {A}}^{d,\dagger })=R({\mathcal {A}}^k) \) and \( N({\mathcal {A}}^{d,\dagger })=N({\mathcal {A}}^k*_N{\mathcal {A}}^{\dagger })\). Therefore, \( R({\mathcal {A}}^{k})\subseteq R({\mathcal {A}}^{*})\) and \( N({\mathcal {A}}^{*})\subseteq N({\mathcal {A}}^{k}*_N{\mathcal {A}}^{\dagger })\). \(\square \)
4 Bilateral inverse solutions of singular tensor equations
Let \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N} \) be a tensor with \(index( {\mathcal {A}}) \ge 1\) and \({\mathcal {B}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N}\). As an application of the DMP, MPD and CMP inverses of tensor, we consider the following equation
First, we state the following theorem.
Theorem 32
Suppose that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) with index\(({\mathcal {A}})= k\). Then
-
(i)
The Eq. (35) has a solution \({\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}}\) if and only if \({\mathcal {B}} \in R({\mathcal {A}}^k) \).
-
(ii)
The Eq. (35) has a solution \({\mathcal {A}}^{\dagger ,d}*_N{\mathcal {B}}\) if and only if \({\mathcal {B}} \in R({\mathcal {A}}^k) \).
-
(iii)
The Eq. (35) has a solution \({\mathcal {A}}^{c,\dagger }*_N{\mathcal {B}}\) if and only if \({\mathcal {B}} \in R({\mathcal {A}}^k) \).
Proof
(i) Let \( {\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}}\) is a solution of (35). By Ji and Wei (2018, Theorem 3.4 (1)), we have
Suppose that \( {\mathcal {B}}\in R({\mathcal {A}}^{k})\), by Stanimirović et al. (2020, Lemma 2.2 (a)), we can conclude that is a tensor \({\mathcal {U}} \in {\mathbb {C}}^{I_1\times \cdots \times I_N}\) such that \({\mathcal {B}}={\mathcal {A}}^{k}*_N{\mathcal {U}}\). Set \( {\mathcal {X}}_1={\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}}\). Thus,
implying that \({\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}}\) is a solution of (35).
(ii) and (iii) have similar proofs to that of (i). \(\square \)
Using the same approach as described in the proof of Theorem 32, the following holds.
Remark 3
Let \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) with index\(({\mathcal {A}})= k\). Then
-
(i)
\( {\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}}={\mathcal {A}}^{d}*_N{\mathcal {B}},\) if \({\mathcal {B}}\in R({\mathcal {A}}).\)
-
(ii)
\( {\mathcal {A}}^{\dagger , d}*_N{\mathcal {B}}={\mathcal {A}}^{\dagger }*_N{\mathcal {B}},\) if \({\mathcal {B}}\in R({\mathcal {A}}^k).\)
-
(iii)
\( {\mathcal {A}}^{c,\dagger }*_N{\mathcal {B}}={\mathcal {A}}^{\dagger }*_N{\mathcal {B}},\) if \({\mathcal {B}}\in R({\mathcal {A}}^k).\)
Theorem 33
Assume that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) with index\(({\mathcal {A}})= k\) and assume that \( {\mathcal {B}}\in R({\mathcal {A}}^k)\). Then
-
(i)
The general solution of (35) takes of the form
$$\begin{aligned} {\mathcal {X}}={\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}}+({\mathcal {I}}-{\mathcal {A}}^{\dagger }*_N{\mathcal {A}})*{\mathcal {Y}} \end{aligned}$$(36)for any tensor \( {\mathcal {Y}} \in {\mathbb {C}}^{I_1\times \cdots \times I_N}\).
-
(ii)
The Eq. (35) has the unique solution \( {\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}} \in R({\mathcal {A}}^k)\).
Proof
(i) By Theorem 32(i), \( {\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}} \) is a solution (35). Assume that \({\mathcal {X}}={\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}}+({\mathcal {I}}-{\mathcal {A}}^{\dagger }*_N{\mathcal {A}})*({\mathcal {Y}}_1+{\mathcal {Y}}_2)\), where \( {\mathcal {Y}}_1 \in {\mathbb {C}}^{I_1\times \cdots \times I_N} \) and \({\mathcal {A}}*_N{\mathcal {Y}}_2={\mathcal {O}}. \) Then
that is \( {\mathcal {X}} \) is a solution (35). Assume that \( {\mathcal {W}} \) is any arbitrary solution of (35). It is clear that \( R({\mathcal {I}}-{\mathcal {A}}^{\dagger }*_N{\mathcal {A}}) \subseteq N({\mathcal {A}})\) and \( {\mathcal {W}}-{\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}} \in N({\mathcal {A}})\). Because
we have that \({\mathcal {W}}-{\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}}=({\mathcal {I}}-{\mathcal {A}}^{\dagger }*_N{\mathcal {A}})*_N{\mathcal {W}}_1+{\mathcal {W}}_2,\) where \( {\mathcal {W}}_2\in N({\mathcal {A}})\cap R({\mathcal {I}}-{\mathcal {A}}^{\dagger }*_N{\mathcal {A}})^{\perp }\). Because \( {\mathcal {W}}_2\in N({\mathcal {A}}) \), we obtain \( {\mathcal {A}}*_N{\mathcal {W}}_2={\mathcal {O}}\). Moreover,
Therefore, \( {\mathcal {W}}-{\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}}=({\mathcal {I}}-{\mathcal {A}}^{\dagger }*_N{\mathcal {A}})*({\mathcal {W}}_1+{\mathcal {W}}_2)\), where \( {\mathcal {W}}_1 \in {\mathbb {C}}^{I_1\times \cdots \times I_N}\) and \( {\mathcal {W}}_2 \in N({\mathcal {A}})\).
(ii) Let \( {\mathcal {X}} \) be a solution in \(R({\mathcal {A}}^k)\). By Theorem 32(i), \( {\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}} \) is a solution in \( R(A^k)\). By the proof of Theorem 31, we have \(R({\mathcal {A}}^{d,\dagger })=R({\mathcal {A}}^{k})\). We have that \( {\mathcal {X}}-{\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}} \in R(A^k)\). Moreover, as stated in Part (i) of this theorem, we have that \( {\mathcal {X}}-{\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}}= ({\mathcal {I}}-{\mathcal {A}}^{\dagger }*_N{\mathcal {A}})*{\mathcal {Y}}\) for some \( {\mathcal {Y}}\). Now \( {\mathcal {A}}^{k}*_N({\mathcal {X}}-{\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}})= ({\mathcal {A}}^{k}-{\mathcal {A}}^{k}*_N{\mathcal {A}}^{\dagger }*_N{\mathcal {A}})*{\mathcal {Y}}={\mathcal {O}}.\) Hence \( {\mathcal {X}}-{\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}} \in N({\mathcal {A}}^k).\) Thus, \( {\mathcal {X}}-{\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}} \in R({\mathcal {A}}^k)\cap N({\mathcal {A}}^k)=\{{\mathcal {O}}\}\), that is \( {\mathcal {X}}={\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}}\). \(\square \)
Theorem 34
Let \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) with index\(({\mathcal {A}})= k\) and assume that \( {\mathcal {B}}\in R({\mathcal {A}}^k)\). Then
-
(i)
The general solution of (35) is of the form
$$\begin{aligned} {\mathcal {X}}={\mathcal {A}}^{\dagger ,d}*_N{\mathcal {B}}+({\mathcal {I}}-{\mathcal {A}}^{\dagger }*_N{\mathcal {A}})*{\mathcal {Y}} \end{aligned}$$(37)for any tensor \( {\mathcal {Y}} \in {\mathbb {C}}^{I_1\times \cdots \times I_N}\).
-
(ii)
The Eq. (35) has the unique solution \( {\mathcal {A}}^{\dagger ,d}*_N{\mathcal {B}} \in R({\mathcal {A}}^{\dagger }*_N{\mathcal {A}}^k)\).
Proof
(i) By using a method similar to the one employed in the proof of Theorem 33(i).
(ii) Suppose that \( {\mathcal {X}} \) is a solution in \(R({\mathcal {A}}^{\dagger }*_N{\mathcal {A}}^k)\). Using a similar method as in the proof of Theorem 31, we obtain \(R({\mathcal {A}}^{\dagger ,d})=R({\mathcal {A}}^{\dagger }*_N{\mathcal {A}}^k)\). This implies \( {\mathcal {X}}-{\mathcal {A}}^{\dagger ,d}*_N{\mathcal {B}} \in R({\mathcal {A}}^{\dagger }*_N{\mathcal {A}}^k)\). Moreover, as stated in Part (i) of this theorem, we have that \( {\mathcal {X}}-{\mathcal {A}}^{\dagger ,d}*_N{\mathcal {B}}= ({\mathcal {I}}-{\mathcal {A}}^{\dagger }*_N{\mathcal {A}})*{\mathcal {Y}}\) for some \( {\mathcal {Y}}\). Now \( {\mathcal {A}}^{k}*_N({\mathcal {X}}-{\mathcal {A}}^{\dagger ,d}*_N{\mathcal {B}})= ({\mathcal {A}}^{k}-{\mathcal {A}}^{k}*_N{\mathcal {A}}^{\dagger }*_N{\mathcal {A}})*{\mathcal {Y}}={\mathcal {O}}.\) Hence \( {\mathcal {X}}-{\mathcal {A}}^{\dagger ,d}*_N{\mathcal {B}} \in N({\mathcal {A}}^k).\) Therefore,
that is \( {\mathcal {X}}={\mathcal {A}}^{\dagger ,d}*_N{\mathcal {B}}\). \(\square \)
Using the same approach as described in the proof of Theorem 34, the following holds.
Corollary 35
Let \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) with index\(({\mathcal {A}})= k\) and assume that \( {\mathcal {B}}\in R({\mathcal {A}}^k)\). Then
-
(i)
The general solution of (35) takes of the form
$$\begin{aligned} {\mathcal {X}}={\mathcal {A}}^{c,\dagger }*_N{\mathcal {B}}+({\mathcal {I}}-{\mathcal {A}}^{\dagger }*_N{\mathcal {A}})*{\mathcal {Y}} \end{aligned}$$(38)for any tensor \( {\mathcal {Y}} \in {\mathbb {C}}^{I_1\times \cdots \times I_N}\).
-
(ii)
The Eq. (35) has the unique solution \( {\mathcal {A}}^{c,\dagger }*_N{\mathcal {B}} \in R({\mathcal {A}}^{\dagger }*_N{\mathcal {A}}^k)\).
As an application of the DMP, MPD and CMP inverses of tensor, we consider the following equation
where \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N} \) and \( {\mathcal {X}}, {\mathcal {B}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N}\). Using the same approach as described in the proof of Theorem 32, the following holds.
Corollary 36
Suppose that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) with index\(({\mathcal {A}})= k\). Then
-
(i)
The Eq. (39) has a solution \({\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}}\) if and only if \({\mathcal {B}} \in R({\mathcal {A}}^k) \).
-
(ii)
The Eq. (39) has a solution \({\mathcal {A}}^{\dagger ,d}*_N{\mathcal {B}}\) if and only if \({\mathcal {B}} \in R({\mathcal {A}}^k) \).
-
(iii)
The Eq. (39) has a solution \({\mathcal {A}}^{c,\dagger }*_N{\mathcal {B}}\) if and only if \({\mathcal {B}} \in R({\mathcal {A}}^k) \).
Using the same approach as described in the proof of Theorems 33 and 34, the following hold.
Corollary 37
Assume that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) with index\(({\mathcal {A}})= k\) and assume that \( {\mathcal {B}}\in R({\mathcal {A}}^k)\). Then
-
(i)
The general solution of (39) takes of the form
$$\begin{aligned} {\mathcal {X}}={\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}}+({\mathcal {I}}-{\mathcal {A}}^{d}*_N{\mathcal {A}})*_N {\mathcal {Y}}, \end{aligned}$$for any tensor \( {\mathcal {Y}} \in {\mathbb {C}}^{I_1\times \cdots \times I_N}\).
-
(ii)
The Eq. (39) has the unique solution \( {\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}} \in R({\mathcal {A}}^k)\).
Corollary 38
Assume that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) with index\(({\mathcal {A}})= k\) and assume that \( {\mathcal {B}}\in R({\mathcal {A}}^k)\). Then
-
(i)
The general solution of (39) is of the form
$$\begin{aligned} {\mathcal {X}}= {\mathcal {A}}^{\dagger ,d}*_N {\mathcal {B}}+( {\mathcal {I}}- {\mathcal {A}}^{d}*_N {\mathcal {A}})*_N {\mathcal {Y}}, \end{aligned}$$for any tensor \( {\mathcal {Y}} \in {\mathbb {C}}^{I_1\times \cdots \times I_N}\).
-
(ii)
The Eq. (39) has the unique solution \( {\mathcal {A}}^{\dagger ,d}*_N{\mathcal {B}} \in R({\mathcal {A}}^{\dagger }*_N{\mathcal {A}}^k)\).
Corollary 39
Assume that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) with index\(({\mathcal {A}})= k\) and assume that \( {\mathcal {B}}\in R({\mathcal {A}}^k)\). Then
-
(i)
The general solution of (39) takes of the form
$$\begin{aligned} {\mathcal {X}}= {\mathcal {A}}^{c,\dagger }*_N {\mathcal {B}}+( {\mathcal {I}}- {\mathcal {A}}^{d}*_N {\mathcal {A}})*_N {\mathcal {Y}}, \end{aligned}$$for any tensor \( {\mathcal {Y}} \in {\mathbb {C}}^{I_1\times \cdots \times I_N}\).
-
(ii)
The Eq. (39) has the unique solution \( {\mathcal {A}}^{c,\dagger }*_N{\mathcal {B}} \in R({\mathcal {A}}^{\dagger }*_N{\mathcal {A}}^k)\).
Suppose that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N} \) and \( {\mathcal {X}}, {\mathcal {B}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N}\). As an application of the core-part of \({\mathcal {A}}\), we consider the following equation:
Theorem 40
Let \({\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) and \({\mathcal {B}}\in R({\mathcal {A}}^{\dagger }*_N{\mathcal {A}}^{d})\). Then the general solution of (40) takes of the form
for any tensor \( {\mathcal {Y}} \in {\mathbb {C}}^{I_1\times \cdots \times I_N}\).
Proof
Using a similar method as in the proof of Theorem 32, we obtain that \({\mathcal {X}}_0=C_{\mathcal {A}}*_N{\mathcal {B}}\) is a solution of Eq. (40) if and only if \({\mathcal {B}} \in R({\mathcal {A}}^{\dagger }*_N{\mathcal {A}}^{d}) \). Also, by Lemma 1, it is clear that \(R({\mathcal {I}}-{\mathcal {A}}*_N{\mathcal {A}}^{\dagger })=N({\mathcal {A}}*_N{\mathcal {A}}^{\dagger })=N({\mathcal {A}}^{\dagger })\). Thus,
\(\square \)
As an application of the DMP, MPD and CMP inverses of tensor, we consider the following equation
where \( index({\mathcal {A}})=k \) and \({\mathcal {B}}\in R({\mathcal {A}}^k)\). If \({\mathcal {B}}\in R({\mathcal {A}}^k)\) and \(index({\mathcal {A}})=k\), then each member of the set \( \{{\mathcal {A}}^{d}*_N{\mathcal {B}}, {\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}}, {\mathcal {A}}^{\dagger ,d}*_N{\mathcal {B}}, {\mathcal {A}}^{c,\dagger }*_N{\mathcal {B}}\}\) is a solution of Eqs. (35) and (41) (see Behera et al. (2020, P. 21)).
Theorem 41
Let \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) and \({\mathcal {B}} \in R({\mathcal {A}}^{k}) \) with index\(({\mathcal {A}})=k\). Then, the set of all solutions of (41) can be represented as
Furthermore, the Eq. (41) has the unique solution \({\mathcal {X}}={\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}}\in R({\mathcal {A}}^k)\).
Proof
Assume \( {\mathcal {B}}\in R({\mathcal {A}}^{k})\). By Stanimirović et al. (2020, Lemma 2.2 (a)), we can conclude that there is a tensor \({\mathcal {U}} \in {\mathbb {C}}^{I_1\times \cdots \times I_N}\) such that \({\mathcal {B}}={\mathcal {A}}^{k}*_N{\mathcal {U}}\).
From Ji and Wei (2018, Theorem 3.2), we have that \( {\mathcal {X}}-{\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}} \in N({\mathcal {A}}^{k+1})=N({\mathcal {A}}^{k})\). Therefore, \( {\mathcal {X}}={\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}}+N({\mathcal {A}}^{k})\). Let \( {\mathcal {X}} \) be a solution in \( R({\mathcal {A}}^k)\). Moreover, by Theorem 32(i) and the proof of Theorem 31, we arrive \({\mathcal {X}}-{\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}} \in R({\mathcal {A}}^{k})\). For the uniqueness in \( R({\mathcal {A}}^{k})\), let \( {\mathcal {V}}\in R({\mathcal {A}}^k) \) be any solution of (41). Now \( {\mathcal {V}}-{\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}} \in R({\mathcal {A}}^k)\), we have \( {\mathcal {A}}^{k+1}*_N{\mathcal {V}}-{\mathcal {A}}^{k+1}*_N{\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}}={\mathcal {O}}\). So, \( {\mathcal {V}}-{\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}} \in N({\mathcal {A}}^k)\). Hence, \( {\mathcal {V}}-{\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}} \in R({\mathcal {A}}^k)\cap N({\mathcal {A}}^k)=\{{\mathcal {O}}\}\). i.e., \( {\mathcal {V}}={\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}}\). \(\square \)
Using the same approach as described in the proof of Theorems 34(ii) and 41, the following hold.
Corollary 42
Suppose that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) and \({\mathcal {B}} \in R({\mathcal {A}}^{k}) \) with index\(({\mathcal {A}})=k\). Then, the set of all solutions of (41) can be represented as
Furthermore, the Eq. (41) has the unique solution \({\mathcal {X}}={\mathcal {A}}^{\dagger ,d}*_N{\mathcal {B}} \in R({\mathcal {A}}^{\dagger }*_N{\mathcal {A}}^k)\).
Corollary 43
Suppose that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) and \( {\mathcal {B}} \in R({\mathcal {A}}^{k}) \) with index\(({\mathcal {A}})=k\). Then, the set of all solutions of (41) can be represented as
Furthermore, the Eq. (41) has the unique solution \({\mathcal {X}}={\mathcal {A}}^{c,\dagger }*_N{\mathcal {B}} \in R({\mathcal {A}}^{\dagger }*_N{\mathcal {A}}^k)\).
As an application of the DMP, MPD and CMP inverses of tensor, we consider the following equation
where \( index({\mathcal {A}})=k \) and \({\mathcal {B}}\in R({\mathcal {A}}^k)\). If \({\mathcal {B}}\in R({\mathcal {A}}^k)\) and \(index({\mathcal {A}})=k\), then each member of the set \( \{{\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}}, {\mathcal {A}}^{\dagger ,d}*_N{\mathcal {B}}, {\mathcal {A}}^{c,\dagger }*_N{\mathcal {B}}\}\) is a solution of Eqs. (41) and (42).
Theorem 44
Suppose that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) and \( {\mathcal {X}},{\mathcal {B}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N} \) with \( index({\mathcal {A}})=k\). Then \({\mathcal {X}}={\mathcal {A}}^{c,\dagger }*_N{\mathcal {B}}\) is a solution of Eq. (42). Moreover, \({\mathcal {X}}={\mathcal {A}}^{c,\dagger }*_N{\mathcal {B}}+({\mathcal {I}}-{\mathcal {A}}^d*_N{\mathcal {A}})*_N{\mathcal {Y}}\) is the general solution of Eq. (42), where \( {\mathcal {Y}} \in {\mathbb {C}}^{I_1\times \cdots \times I_N}\) is an arbitrary tensor.
Proof
Set \( {\mathcal {X}}={\mathcal {A}}^{c,\dagger }*_N{\mathcal {B}}\). Then,
By Ji and Wei (2018, Theorem 3.4), we have \( R({\mathcal {I}}-{\mathcal {A}}^d*_N{\mathcal {A}}) =N({\mathcal {A}}^d*_N{\mathcal {A}})=N({\mathcal {A}}^d)=N({\mathcal {A}}^k). \)
Thus, \({\mathcal {A}}^k*_N[{\mathcal {A}}^{c,\dagger }*_N{\mathcal {B}}+({\mathcal {I}}-{\mathcal {A}}^d*_N{\mathcal {A}})*_N{\mathcal {Y}}]={\mathcal {A}}^k*_N{\mathcal {A}}^{c,\dagger }*_N{\mathcal {B}} + {\mathcal {O}}= {\mathcal {A}}^k*_N{\mathcal {A}}^{\dagger }*_N{\mathcal {B}}.\) \(\square \)
Using the same approach as described in the proof of Theorem 44, the following holds.
Corollary 45
Suppose that \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) and \( {\mathcal {X}},{\mathcal {B}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N} \) with \( index({\mathcal {A}})=k\). Then \({\mathcal {X}}={\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}}\) is a solution of the Eq. (42). Moreover, \({\mathcal {X}}={\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}}+({\mathcal {I}}-{\mathcal {A}}*_N{\mathcal {A}}^d)*_N{\mathcal {Y}}\) is the general solution of Eq. (42), where \( {\mathcal {Y}} \in {\mathbb {C}}^{I_1\times \cdots \times I_N}\) is an arbitrary tensor.
Using the same approach as described in the proof of Theorems 34(ii) and 41, the following holds.
Remark 4
Let \( {\mathcal {A}}\in {\mathbb {C}}^{I_1\times \cdots \times I_N \times I_1\times \cdots \times I_N}\) and \( {\mathcal {B}} \in R({\mathcal {A}}^{k}) \) with \( index({\mathcal {A}})=k\). Then
-
(i)
The Eq. (42) has a unique solution \({\mathcal {X}}={\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}} \in R({\mathcal {A}}^k)\) and its general solution \({\mathcal {X}}={\mathcal {A}}^{d,\dagger }*_N{\mathcal {B}}+N({\mathcal {A}}^k).\)
-
(ii)
The Eq. (42) has a unique solution \({\mathcal {X}}={\mathcal {A}}^{\dagger ,d}*_N{\mathcal {B}} \in R({\mathcal {A}}^{\dagger }*_N{\mathcal {A}}^k)\) and its general solution \({\mathcal {X}}={\mathcal {A}}^{\dagger ,d}*_N{\mathcal {B}}+N({\mathcal {A}}^k).\)
-
(iii)
The Eq. (42) has a unique solution \({\mathcal {X}}={\mathcal {A}}^{c,\dagger }*_N{\mathcal {B}} \in R({\mathcal {A}}^{\dagger }*_N{\mathcal {A}}^k)\) and its general solution \({\mathcal {X}}={\mathcal {A}}^{c,\dagger }*_N{\mathcal {B}}+N({\mathcal {A}}^k).\)
References
Beckmann CF, Smith SM (2005) Tensorial extensions of independent component analysis for multisubject FMRI analysis. Neuroimage 25(1):294–311
Behera R, Nandi AK, Sahoo JK (2020) Further results on the Drazin inverse of even-order tensors. Numer Linear Algebra Appl 27(5):2317
Brazell M, Li N, Navasca C, Tamon C (2013) Solving multilinear systems via tensor inversion. SIAM J Matrix Anal Appl 34(2):542–570
Bu C, Zhang X, Zhou J, Wang W, Wei Y (2014) The inverse, rank and product of tensors. Linear Algebra Appl 446:269–280
Cyganek B, Gruszczyński S (2014) Hybrid computer vision system for drivers’ eye recognition and fatigue monitoring. Neurocomputing 126:78–94
Du H-M, Wang B-X, Ma H-F (2019) Perturbation theory for core and core-ep inverses of tensor via Einstein product. Filomat 33(16):5207–5217
Einstein A (2007) The foundation of the general theory of relativity. Ann Phys 49(7):769–822
Eldén L (2007) Matrix methods in data mining and pattern recognition. SIAM
Ji J, Wei Y (2018) The Drazin inverse of an even-order tensor and its application to singular tensor equations. Comput Math Appl 75(9):3402–3413
Kheirandish E, Salemi A (2023) Generalized bilateral inverses. J Comput Appl Math 428:115137
Ma H, Li N, Stanimirović PS, Katsikis VN (2019) Perturbation theory for Moore-Penrose inverse of tensor via Einstein product. Comput Appl Math 38:1–24
Mehdipour M, Salemi A (2018) On a new generalized inverse of matrices. Linear Multilinear Algebra 66(5):1046–1053
Panigrahy K, Mishra D (2022) Extension of Moore-Penrose inverse of tensor via Einstein product. Linear Multilinear Algebra 70(4):750–773
Panigrahy K, Behera R, Mishra D (2020) Reverse-order law for the Moore-Penrose inverses of tensors. Linear Multilinear Algebra 68(2):246–264
Rabanser S, Shchur O, Günnemann S (2017) Introduction to tensor decompositions and their applications in machine learning. arXiv preprint arXiv:1711.10781
Sahoo JK, Behera R, Stanimirović PS, Katsikis VN, Ma H (2020) Core and core-ep inverses of tensors. Comput Appl Math 39(1):9
Sahoo JK, Behera R, Stanimirović PS, Katsikis VN (2020) Computation of outer inverses of tensors using the QR decomposition. Comput Appl Math 39(3):1–20
Stanimirović PS, Ćirić M, Katsikis VN, Li C, Ma H (2020) Outer and (b, c) inverses of tensors. Linear Multilinear Algebra 68(5):940–971
Sun L, Zheng B, Bu C, Wei Y (2016) Moore-Penrose inverse of tensors via Einstein product. Linear Multilinear Algebra 64(4):686–698
Sun L, Zheng B, Wei Y, Bu C (2018) Generalized inverses of tensors via a general product of tensors. Front Math China 13:893–911
Wang Y, Wei Y (2022) Generalized eigenvalue for even order tensors via Einstein product and its applications in multilinear control systems. Comput Appl Math 41(8):419
Wang B, Du H, Ma H (2020) Perturbation bounds for DMP and CMP inverses of tensors via Einstein product. Comput Appl Math 39(1):1–17
Wang X, Che M, Mo C, Wei Y (2023) Solving the system of nonsingular tensor equations via randomized Kaczmarz-like method. J Comput Appl Math 421:114856
Wei Y, Stanimirovic P, Petkovic M (2018) Numerical and symbolic computations of generalized inverses. World Scientific
Weiyang D, Yimin W (2016) Theory and computation of tensors. Elsevier, Academic Press, Amsterdam, Boston, London, New York, Oxford, Paris, San Diego, San Francisco, Singapore, Sydney, Tokyo
Zhao Y, Yang LT, Zhang R (2017) A tensor-based multiple clustering approach with its applications in automation systems. IEEE Trans Ind Inform 14(1):283–291
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
There is no conflict of interest in the manuscript.
Additional information
Communicated by Yimin Wei.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Kheirandish, E., Salemi, A. Generalized bilateral inverses of tensors via Einstein product with applications to singular tensor equations. Comp. Appl. Math. 42, 343 (2023). https://doi.org/10.1007/s40314-023-02483-8
Received:
Revised:
Accepted:
Published:
DOI: https://doi.org/10.1007/s40314-023-02483-8