Quaternion MPCEP, CEPMP, and MPCEPMP Generalized Inverses

*Ivan I. Kyrchei*

## **Abstract**

A generalized inverse of a matrix is an inverse in some sense for a wider class of matrices than invertible matrices. Generalized inverses exist for an arbitrary matrix and coincide with a regular inverse for invertible matrices. The most famous generalized inverses are the Moore–Penrose inverse and the Drazin inverse. Recently, new generalized inverses were introduced, namely the core inverse and its generalizations. Among them, there are compositions of the Moore–Penrose and core inverses, MPCEP (or MP–Core–EP) and EPCMP (or EP–Core–MP) inverses. In this chapter, the notions of the MPCEP inverse and CEPMP inverse are expanded to quaternion matrices and introduced new generalized inverses, the right and left MPCEPMP inverses. Direct method of their calculations, that is, their determinantal representations are obtained within the framework of theory of quaternion row-column determinants previously developed by the author. In consequence, these determinantal representations are derived in the case of complex matrices.

**Keywords:** Moore–Penrose inverse, Drazin inverse, generalized inverse, core-EP inverse, quaternion matrix, noncommutative determinant

## **1. Introduction**

The field of complex (or real) numbers is designated by (). The set of all *m* � *n* matrices over the quaternion skew field

$$\mathbb{H} = \left\{ h\_0 + h\_1 \mathbf{i} + h\_2 \mathbf{j} + h\_3 \mathbf{k} | \mathbf{i}^2 = \mathbf{j}^2 = \mathbf{k}^2 = \mathbf{i} \mathbf{j} \mathbf{k} = -\mathbf{1}, h\_0, h\_1, h\_2, h\_3 \in \mathbb{R} \right\},$$

is represented by *<sup>m</sup>*�*<sup>n</sup>*, while *<sup>m</sup>*�*<sup>n</sup> <sup>r</sup>* is reserved for the subset of *<sup>m</sup>*�*<sup>n</sup>* with matrices of rank *r*. If *h* ¼ *h*<sup>0</sup> þ *h*1**i** þ *h*2**j** þ *h*3**k** ∈ , its conjugate is *h* ¼ *h*<sup>0</sup> � *h*1**i** � *h*2**j** � *h*3**k**, and its norm ∥*h*∥ ¼ ffiffiffiffiffi *hh* p ¼ ffiffiffiffiffi *hh* <sup>p</sup> ¼ ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi *h*2 <sup>0</sup> <sup>þ</sup> *<sup>h</sup>*<sup>2</sup> <sup>1</sup> <sup>þ</sup> *<sup>h</sup>*<sup>2</sup> <sup>2</sup> <sup>þ</sup> *<sup>h</sup>*<sup>2</sup> 3 q . For **A** ∈ *<sup>m</sup>*�*<sup>n</sup>*, its rank and conjugate transpose are given by rankð Þ **<sup>A</sup>** and **<sup>A</sup>**<sup>∗</sup> , respectively. A matrix **<sup>A</sup>** <sup>∈</sup> *<sup>n</sup>*�*<sup>n</sup>* is said to be Hermitian if **<sup>A</sup>**<sup>∗</sup> <sup>¼</sup> **<sup>A</sup>**. Also,

• <sup>C</sup>*r*ð Þ¼ **<sup>A</sup> <sup>c</sup>** <sup>∈</sup> *<sup>m</sup>*�<sup>1</sup> : **<sup>c</sup>** <sup>¼</sup> **Ad**, **<sup>d</sup>**<sup>∈</sup> *<sup>n</sup>*�<sup>1</sup> � � is the right column space of **<sup>A</sup>**;

$$\bullet \,\, \mathcal{R}\_l(\mathbf{A}) = \left\{ \mathbf{c} \in \mathbb{H}^{1 \times n} : \mathbf{c} = \mathbf{d} \mathbf{A}, \mathbf{d} \in \mathbb{H}^{1 \times m} \right\} \text{ is the left row space of } \mathbf{A};$$


Let us recall the definitions of some well-known generalized inverses that can be extend to quaternion matrices as follows.

**Definition 1.1.** *The Moore–Penrose inverse of* **<sup>A</sup>** <sup>∈</sup> *<sup>n</sup>*�*<sup>m</sup> is the unique matrix* **<sup>A</sup>**† <sup>¼</sup> **<sup>X</sup>** *determined by equations*

$$(1)\;\textbf{AXA}=\textbf{A};\\(2)\;\textbf{XAX}=\textbf{X};\\(3)\;(\textbf{AX})^{\*}=\textbf{AX};\\(4)\;(\textbf{XA})^{\*}=\textbf{XA}.\tag{1}$$

**Definition 1.2.** *The Drazin inverse of* **<sup>A</sup>** <sup>∈</sup> *<sup>n</sup>*�*<sup>n</sup> is the unique* **<sup>A</sup>***<sup>d</sup>* <sup>¼</sup> **<sup>X</sup>** *that satisfying Eq.(2) from* (1) *and the following equations,*

$$(\mathbf{S})\,\mathbf{A}^k = \mathbf{X}\mathbf{A}^{k+1}, \quad (\mathbf{6})\,\mathbf{X}\mathbf{A} = \mathbf{A}\mathbf{X},$$

where *k* ¼ Indð Þ **A** is the index of **A**, i.e. the smallest positive number such that rank **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> <sup>¼</sup> rank **<sup>A</sup>***<sup>k</sup>* . If Indð Þ **<sup>A</sup>** <sup>≤</sup>1, then **<sup>A</sup>***<sup>d</sup>* <sup>¼</sup> **<sup>A</sup>**# is the group inverse of **<sup>A</sup>**. If Indð Þ¼ **<sup>A</sup>** 0, then **<sup>A</sup>**# <sup>¼</sup> **<sup>A</sup>**† <sup>¼</sup> **<sup>A</sup>**�<sup>1</sup> .

A matrix **A** satisfying the conditions ð Þ*i* ,ð Þ*j* , … is called an f g *i*, *j*, … -inverse of **A**, and is denoted by **A**ð Þ *<sup>i</sup>*,*j*, … . In particular, **A**ð Þ<sup>1</sup> is called the *inner inverse*, **A**ð Þ<sup>2</sup> is called the *outer inverse*, and **A**ð Þ 1,2 is called the *reflexive inverse*, and **A**ð Þ 1,2,3,4 is the Moore–Penrose inverse, etc.

Note that the Moore–Penrose inverse inducts the orthogonal projectors **<sup>P</sup>***<sup>A</sup>* <sup>¼</sup> **AA**† and **<sup>Q</sup>***<sup>A</sup>* <sup>¼</sup> **<sup>A</sup>**† **A** onto the right column spaces of **A** and **A**<sup>∗</sup> , respectively.

In [1], the core-EP inverse over the quaternion skew field was presented similarly as in [2].

**Definition 1.3.** *The core-EP inverse of* **<sup>A</sup>** <sup>∈</sup> *<sup>n</sup>*�*<sup>n</sup> is the unique matrix* **<sup>A</sup>**† <sup>¼</sup> **<sup>X</sup>** *which satisfies*

$$\mathbf{X} = \mathbf{X} \mathbf{A} \mathbf{X}, \quad \mathcal{C}\_r(\mathbf{X}) = \mathcal{C}\_r \left( \mathbf{A}^d \right) = \mathcal{C}\_r(\mathbf{X}^\*).$$

According to [3], (Theorem 2.3), for *<sup>m</sup>* <sup>≥</sup> Indð Þ **<sup>A</sup>** , we have that **<sup>A</sup>**◯† <sup>¼</sup> **<sup>A</sup>***<sup>d</sup>* **<sup>A</sup>***<sup>m</sup>* **<sup>A</sup>***<sup>m</sup>* ð Þ† *:* In a special case that Indð Þ **<sup>A</sup>** <sup>≤</sup>1, **<sup>A</sup>**◯† <sup>¼</sup> **<sup>A</sup>**◯# is the core inverse of **<sup>A</sup>** [4].

**Definition 1.4.** *The dual core-EP inverse of* **<sup>A</sup>** <sup>∈</sup> *<sup>n</sup>*�*<sup>n</sup> is the unique matrix* **<sup>A</sup>**◯† <sup>¼</sup> **<sup>X</sup>** *for which*

$$\mathbf{X} = \mathbf{X} \mathbf{A} \mathbf{X}, \quad \mathcal{R}\_l(\mathbf{X}) = \mathcal{R}\_l \left( \mathbf{A}^d \right) = \mathcal{R}\_l(\mathbf{X}^\*).$$

Recall that, **<sup>A</sup>**◯† <sup>¼</sup> **<sup>A</sup>***<sup>m</sup>* ð Þ† **<sup>A</sup>***<sup>m</sup>***A***<sup>d</sup>* for *<sup>m</sup>* <sup>≥</sup>Indð Þ **<sup>A</sup>** .

Since the quaternion core-EP inverse **<sup>A</sup>**† is related to the right space <sup>C</sup>*r*ð Þ **<sup>A</sup>** of **<sup>A</sup>** <sup>∈</sup> *<sup>n</sup>*�*<sup>n</sup>* and the quaternion dual core-EP inverse **<sup>A</sup>**† is related to its left space <sup>R</sup>*l*ð Þ **<sup>A</sup>** . So, in [1], they are also named the right and left core-EP inverses, respectively.

Various representations of core-EP inverse can be found in [1, 5–7]. In [8], continuity of core-EP inverse was investigated. Bordering and iterative methods to find the core-EP inverse were proved in [9, 10], and its determinantal representation for complex matrices was derived in [2]. New determinantal representations of the

### *Quaternion MPCEP, CEPMP, and MPCEPMP Generalized Inverses DOI: http://dx.doi.org/10.5772/intechopen.103087*

complex core-EP inverse and its various generalizations were obtained in [11]. The core-EP inverse was generalized to rectangular matrices [12], Hilbert space operators [13], Banach algebra elements [14], tensors [15], and elements of rings [3]. Combining the core-EP inverse or the dual core-EP inverse with the Moore–Penrose inverse, the MPCEP inverse and CEPMP inverse were introduced in [16] for bounded linear Hilbert space operators.

In the last years, interest in quaternion matrix equations is growing significantly based on the increasing their applications in various fields, among them, robotic manipulation [17], fluid mechanics [18, 19], quantum mechanics [20–22], signal processing [23, 24], color image processing [25–27], and so on.

The main goals of this chapter are investigations of the MPCEP and CEPMP inverses, introductions and representations of new right and left MPCEPMP inverses over the quaternion skew field, and obtaining of their determinantal representations as a direct method of their constructions. The chapter develops and continues the topic raised in a number of other works [28–33], where determinantal representations of various generalized inverses were obtained.

The remainder of our chapter is directed as follows. In Section 2, we introduce of the quaternion MPCEP and CEPMP inverses and give characterizations of new generalized inverses, namely left and right MPCEPMP-inverses. In Section 3, we commence with introducing determinantal representations of the projection matrices inducted by the Moore–Penrose inverse and of core-EP inverse previously obtained within the framework of theory of quaternion row-column determinants and, based of them, determinantal representations of the MPCEP, CEPMP, and left and right MPCEPMP inverses are derived. Finally, the conclusion is drawn in Section 4.

## **2. Characterizations of the quaternion MPCEP, CEPMP, and MPCEPMP inverses**

Analogously as in [16], the MPCEP inverse and CEPMP inverse can be defined for quaternion matrices.

**Definition 2.1.** *Let* **A** ∈ *<sup>n</sup>*�*<sup>n</sup> . The MPCEP (or MP-Core-EP) inverse of* **A** *is the unique solution* **<sup>A</sup>**†,◯† <sup>¼</sup> **<sup>X</sup>** *to the system*

$$\mathbf{X} = \mathbf{X}\mathbf{A}\mathbf{X}, \qquad \mathbf{X}\mathbf{A} = \mathbf{A}^{\dagger}\mathbf{A}\mathbf{A}^{\dagger}\mathbf{\widehat{A}}, \qquad \mathbf{A}\mathbf{X} = \mathbf{A}\mathbf{A}^{\dagger}\mathbf{\widehat{A}}$$

The CEPMP (or Core-EP-MP) inverse of **<sup>A</sup>** is the unique solution **<sup>A</sup>**◯† ,† <sup>¼</sup> **<sup>X</sup>** to the system

$$\mathbf{X} = \mathbf{X}\mathbf{A}\mathbf{X}, \qquad \mathbf{A}\mathbf{X} = \mathbf{A}\mathbf{A}\_{\bigotimes}\mathbf{A}\mathbf{A}^{\dagger}, \qquad \mathbf{X}\mathbf{A} = \mathbf{A}\_{\bigotimes}\mathbf{A}.$$

We can represent the MPCEP inverse and CEPMP inverse, by [16], as

$$\mathbf{A}^{\dagger,\{\mathbb{G}\}} = \mathbf{A}^{\dagger} \mathbf{A} \mathbf{A}^{\mathbb{G}},\tag{2}$$

$$\mathbf{A}^{\widehat{\odot}\downarrow\uparrow} = \mathbf{A}\_{\widehat{\odot}} \mathbf{A} \mathbf{A}^{\uparrow}. \tag{3}$$

According to our concepts, we can define the left and right MPCPMP inverses. **Definition 2.2.** *Suppose* **A** ∈ *<sup>n</sup>*�*n. The right MPCEPMP inverse of* **A** *is defined as*

$$\mathbf{A}^{\dagger,\bigoplus,\dagger,r} = \mathbf{A}^{\dagger}\mathbf{A}\mathbf{A}^{\bigoplus}\mathbf{A}\mathbf{A}^{\dagger}.$$

The left MPCEPMP inverse of **A** is defined as

$$\mathbf{A}^{\dagger,\widehat{\odot},\dagger,l} = \mathbf{A}^{\dagger}\mathbf{A}\mathbf{A}\_{\widehat{\odot}}\mathbf{A}\mathbf{A}^{\dagger}.$$

The following gives the characteristic equations of these generalized inverses. **Theorem 2.3.** *Let* **A**, **X** ∈ *<sup>n</sup>*�*n. The following statements are equivalent:*

i. **X** *is the right MPCEPMP inverse of* **A***.*

ii.

$$\mathbf{X} = \mathbf{A}^{\dagger, \{\mathbf{\}\}} \mathbf{P}\_A. \tag{4}$$

iii. **X** *is the unique solution to the three equations:*

$$\mathbf{1.X} = \mathbf{X} \mathbf{A} \mathbf{X}, \qquad \mathbf{2.X} = \mathbf{A}^{\dagger, \{\overline{\mathbb{G}}\}} \mathbf{A}, \qquad \mathbf{3.A} \mathbf{X} = \mathbf{A} \mathbf{A}^{\overline{\mathbb{G}}} \mathbf{P}\_{\mathbf{A}}.\tag{5}$$

*Proof.* ½ � ð Þi ↦ ½ � ð Þii . By Eq. (2) and the denotation of **P***A*, it is evident that

$$\mathbf{A}^{\dagger,\widehat{\mathbb{G}}\dagger,r} = \mathbf{A}^{\dagger}\mathbf{A}\mathbf{A}^{\widehat{\mathbb{G}}}\mathbf{A}\mathbf{A}^{\dagger} = \mathbf{A}^{\dagger,\widehat{\mathbb{G}}\dagger}\mathbf{P}\_A.$$

½ � ð Þ<sup>i</sup> <sup>↦</sup> ½ � ð Þ iii . Now, we verify the condition (5). Let **<sup>X</sup>** <sup>¼</sup> **<sup>A</sup>**†,◯† ,†,*<sup>r</sup>* <sup>¼</sup> **<sup>A</sup>**† **AA**◯† **AA**† . Then, from the Definition 1.1 and the representation (2), we have

$$\mathbf{AX}\mathbf{AX} = \mathbf{A}^{\dagger}\mathbf{A}\mathbf{A}^{\odot}\mathbf{A} (\mathbf{A}^{\dagger}\mathbf{A}\mathbf{A}^{\dagger})\mathbf{A}\mathbf{A}^{\odot}\mathbf{A}\mathbf{A}^{\dagger} = \mathbf{A}^{\dagger}\mathbf{A}\mathbf{A}^{\odot}(\mathbf{A}\mathbf{A}^{\dagger}\mathbf{A})\mathbf{A}^{\odot}\mathbf{A}\mathbf{A}^{\dagger} = \mathbf{A}^{\dagger}\mathbf{A}\mathbf{A}^{\dagger}\mathbf{A}\mathbf{A}^{\dagger}\mathbf{A}^{\dagger}$$

$$= \mathbf{A}^{\dagger}\mathbf{A}\left(\mathbf{A}^{\odot}\mathbf{A}\mathbf{A}^{\odot}\right)\mathbf{A}\mathbf{A}^{\dagger} = \mathbf{A}^{\dagger}\mathbf{A}\mathbf{A}^{\odot}\mathbf{A}\mathbf{A}^{\dagger} = \mathbf{X},$$

$$\mathbf{XA} = \mathbf{A}^{\dagger}\mathbf{A}\mathbf{A}^{\odot}(\mathbf{A}\mathbf{A}^{\dagger}\mathbf{A}) = \mathbf{A}^{\dagger, \odot}\mathbf{A},$$

$$\mathbf{AX} = (\mathbf{A}\mathbf{A}^{\dagger}\mathbf{A})\mathbf{A}^{\odot}\mathbf{A}\mathbf{A}^{\dagger} = \mathbf{A}\mathbf{A}^{\odot}\mathbf{P}\_{\mathbf{A}}.$$

To prove that the system (5) has unique solution, suppose that **X** and **X**<sup>1</sup> are two solutions of this system. Then **XA** <sup>¼</sup> **<sup>A</sup>**†,◯† **<sup>A</sup>** <sup>¼</sup> **<sup>X</sup>**1**<sup>A</sup>** and **AX** <sup>¼</sup> **AA**◯† **<sup>P</sup>***<sup>A</sup>* <sup>¼</sup> **AX**1, which give **X AX** ð Þ¼ ð Þ **XA X**<sup>1</sup> ¼ **X**1**AX**<sup>1</sup> ¼ **X**1. Therefore, **X** is the unique solution to the system. □

The next theorem can be proved in the same way. **Theorem 2.4.** *Let* **A**, **X** ∈ *<sup>n</sup>*�*n. The following statements are equivalent:*

i. **X** *is the left MPCEPMP inverse of* **A**.

ii.

$$\mathbf{X} = \mathbf{Q}\_A \mathbf{A}^{\oplus \uparrow}.\tag{6}$$

iii. **X** *is the unique solution to the system:*

$$\mathbf{1.X} = \mathbf{X} \mathbf{A} \mathbf{X}, \qquad \mathbf{2.A} \mathbf{X} = \mathbf{A} \mathbf{A}^{\bigotimes \uparrow}, \qquad \mathbf{3.X} \mathbf{A} = \mathbf{A}\_{\bigotimes} \mathbf{A}.$$

#### **3. Determinantal representations of the quaternion MPCEP and \* CEPMP inverses**

It is well known that the determinantal representation of the regular inverse is given by the cofactor matrix. The construction of determinantal representations of generalized inverses is not so evident and unambiguous even for matrices with complex or real entries. Taking into account the noncommutativity of quaternions, this task is more complicated due to a problem of defining the determinant of a matrix with noncommutative elements (see survey articles [34–36] for detail). Only now, the solving this problem begins to be decided thanks to the theory of noncommutative column-row determinants introduced in [37, 38].

For arbitrary quaternion matrix **A** ∈ *<sup>n</sup>*�*<sup>n</sup>*, there exists an exact technique to generate *n* row determinants (ℜ-determinants) and *n* column determinants (ℭ-determinants) by stating a certain order of factors in each term.

**Definition 3.1.** *Let* **<sup>A</sup>** <sup>¼</sup> *aij* � �<sup>∈</sup> *<sup>n</sup>*�*n.*

• *For an arbitrary row index i*∈ *In, the ith* ℜ*-determinant of* **A** *is defined as*

$$\mathbf{rdet}\_i \mathbf{A} := \sum\_{\sigma \in \mathcal{S}\_n} (-1)^{n-r} \left( a\_{i i\_{k\_1}} a\_{i\_{k\_1} i\_{k\_1+1}} \dots a\_{i\_{k\_1+l\_1} i} \right) \dots \left( a\_{i\_{k\_r} i\_{k\_r+1}} \dots a\_{i\_{k\_r+l\_r} i\_{k\_r}} \right),$$

in which *Sn* denotes the symmetric group on *In* ¼ f g 1, … , *n* , while the permutation *σ* is defined as a product of mutually disjunct subsets ordered from the left to right by the rules

$$\sigma = (i i\_{k\_1} i\_{k\_1 + 1} \dots i\_{k\_1 + l\_1})(i\_{k\_2} i\_{k\_2 + 1} \dots i\_{k\_2 + l\_2}) \dots (i\_{k\_r} i\_{k\_r + 1} \dots i\_{k\_r + l\_r}),$$

$$i\_{k\_t} < i\_{k\_t + s}, \ i\_{k\_2} < i\_{k\_3} < \dots < i\_{k\_r}, \ \forall \ t = 2, \ \dots, r, \ s = 1, \dots, l\_t.$$

• *For an arbitrary column index j*∈*In, the jth* ℭ*-determinant of* **A** *is defined as the sum*

$$\mathbf{cdet}\_{j}\mathbf{A} = \sum\_{\mathbf{r}\in\mathcal{S}\_{\mathbf{n}}} (-\mathbf{1})^{n-r} \left( a\_{\ \ j\_{k\_{r}}j\_{k\_{r}+l\_{r}}} \cdots a\_{\ j\_{k\_{r}+1}j\_{k\_{r}}} \right) \cdots \left( a\_{\vec{\jmath}\_{k\_{1}+l\_{1}}} \cdots a\_{\ j\_{k\_{1}+1}j\_{k\_{1}}} a\_{\ j\_{k\_{l}}j} \right),$$

in which a *permutation τ* is ordered from the right to left in the following way:

$$\pi = \left(j\_{k\_r+l\_r}\cdots j\_{k\_r+1}j\_{k\_r}\right)\cdots\left(j\_{k\_2+l\_2}\cdots j\_{k\_2+1}j\_{k\_2}\right)\left(j\_{k\_1+l\_1}\cdots j\_{k\_1+1}j\_{k\_2}j\right), j\_{k\_i} < j\_{k\_i+l}, \quad j\_{k\_2} < j\_{k\_3} < \cdots < j\_{k\_r}.$$

It is known that all ℜ- and ℭ-determinants are different in general. However, in [37], the following equalities are verified for a Hermitian matrix **A** that introduce a determinant of a Hermitian matrix: rdet1**A** ¼ ⋯ ¼ rdet*n***A** ¼ cdet1**A** ¼ ⋯ ¼ cdet*n***A** ≔ det**A** ∈ *:*

D-Representations of various generalized inverses were developed by means of the theory of ℜ- and ℭ-determinants (see e.g. [28–31]).

The following notations are used for determinantal representations of generalized inverses.

Let *α* ≔ f g *α*1, … , *α<sup>k</sup>* ⊆f g 1, … , *m* and *β* ≔ *β*<sup>1</sup> f g , … , *β<sup>k</sup>* ⊆f g 1, … , *n* be subsets with <sup>1</sup>≤*k*<sup>≤</sup> min f g *<sup>m</sup>*, *<sup>n</sup>* . Suppose that **<sup>A</sup>***<sup>α</sup> <sup>β</sup>* is a submatrix of **A** ∈ *<sup>m</sup>*�*<sup>n</sup>* whose rows and columns are indexed by *α* and *β*, respectively. Then, **A***<sup>α</sup> <sup>α</sup>* is a principal submatrix of **A**

whose rows and columns are indexed by *<sup>α</sup>*. If **<sup>A</sup>** is Hermitian, then j j **<sup>A</sup>** *<sup>α</sup> <sup>α</sup>* stands for a principal minor of det**A**. The collection of strictly increasing sequences of 1≤*k*≤*n* integers chosen from 1, f g … , *n* is denoted by

*Lk*,*<sup>n</sup>* ≔ f g *α* : *α* ¼ ð Þ *α*1, … , *α<sup>k</sup>* , 1≤*α*<sup>1</sup> < ⋯ <*α<sup>k</sup>* ≤*n :* For fixed *i* ∈*α* and *j*∈ *β*, put *Ir*,*m*f g*i* ≔ f g *α* : *α* ∈*Lr*,*m*, *i* ∈*α* , *Jr*,*n*f g*j* ≔ f g *β* : *β* ∈*Lr*,*n*, *j*∈ *β* .

Let **a***:<sup>j</sup>* and **a** <sup>∗</sup> *:<sup>j</sup>* be the *j*th columns, **a***<sup>i</sup>:* and **a** <sup>∗</sup> *<sup>i</sup>:* be the *i*th rows of **A** and **A**<sup>∗</sup> , respectively. Suppose that **A***<sup>i</sup>:*ð Þ **b** and **A***:j*ð Þ**c** stand for the matrices obtained from **A** by replacing its *i*th row with the row vector **b**∈ <sup>1</sup>�*<sup>n</sup>* and its *j*th column with the column vector **c** ∈ *<sup>m</sup>*, respectively.

Based on determinantal representations of the Moore–Penrose inverse obtained in [28], we have determinantal representations of the projections.

**Lemma 3.2.** [28] *If* **A** ∈ *<sup>m</sup>*�*<sup>n</sup> <sup>r</sup> , then the determinantal representations of the projection matrices* **A**† **<sup>A</sup>**≕**Q***<sup>A</sup>* <sup>¼</sup> *qA ij* � � *n*�*n and* **AA**† <sup>≕</sup>**P***<sup>A</sup>* <sup>¼</sup> *<sup>p</sup><sup>A</sup> ij* � � *<sup>m</sup>*�*<sup>m</sup> can be expressed as follows*

$$q^{A}\_{\vec{\mu}} = \frac{\sum\_{\beta \in I\_{rs} \{i\}} \mathsf{cdet}\_{i} \left( (\mathbf{A}^{\*} \mathbf{A})\_{.i} (\dot{\mathbf{a}}\_{j}) \right)^{\beta}\_{\beta}}{\sum\_{\beta \in I\_{rs}} |\mathbf{A}^{\*} \mathbf{A}|\_{\beta}^{\beta}} = \frac{\sum\_{a \in I\_{rs} \{j\}} \mathsf{rdet}\_{j} \left( (\mathbf{A}^{\*} \mathbf{A})\_{j} (\dot{\mathbf{a}}\_{i}) \right)^{a}\_{a}}{\sum\_{a \in I\_{rs}} |\mathbf{A}^{\*} \mathbf{A}|\_{a}^{a}},\tag{7}$$

$$p\_{\vec{\mu}}^{A} = \frac{\sum\_{a \in I\_{r,n}\{\cdot\}} \text{rdet}\_{\vec{\}} \Big( (\mathbf{A} \mathbf{A}^{\*})\_{\cdot \vec{\cdot}} (\ddot{\mathbf{a}}\_{i.}) \Big)\_{a}^{a}}{\sum\_{a \in I\_{r,n}} |\mathbf{A} \mathbf{A}^{\*}|\_{a}^{a}} = \frac{\sum\_{\beta \in I\_{r,n}\{\cdot\}} \text{cdet}\_{i} \Big( (\mathbf{A} \mathbf{A}^{\*})\_{\cdot \vec{\cdot}} (\ddot{\mathbf{a}}\_{\cdot \vec{\cdot}}) \Big)\_{\beta}^{\beta}}{\sum\_{\beta \in I\_{r,n}} |\mathbf{A} \mathbf{A}^{\*}|\_{\beta}^{\beta}},\tag{8}$$

where **a**\_*:<sup>i</sup>* and **a**\_*:<sup>j</sup>*, €**a***<sup>i</sup>:* and **a**€*:<sup>j</sup>* are the *i*th rows and the *j*th columns of **A**<sup>∗</sup> **A** ∈ *<sup>n</sup>*�*<sup>n</sup>* and **AA**<sup>∗</sup> ∈ *<sup>m</sup>*�*<sup>m</sup>*, respectively.

Recently, D-representations of the quaternion core-EP inverses were obtained in [1] as well.

**Lemma 3.3.** [1] *Suppose that* **<sup>A</sup>** <sup>∈</sup> *<sup>n</sup>*�*n,* Indð Þ¼ **<sup>A</sup>** *k and* rank **<sup>A</sup>***<sup>k</sup>* � � <sup>¼</sup> *s. Then* **<sup>A</sup>**† <sup>¼</sup> *a*†,*<sup>r</sup> ij* � � *and* **<sup>A</sup>**† <sup>¼</sup> *<sup>a</sup>*†,*<sup>l</sup> ij* � � *possess the determinantal representations, respectively,*

$$a\_{\vec{\eta}}^{\uparrow r} = \frac{\sum\_{a \in I, \mathfrak{s} \{\cdot\}} \text{rdet}\_{\vec{\jmath}} \left( \left( \mathbf{A}^{k+1} \left( \mathbf{A}^{k+1} \right)^{\*} \right)\_{\vec{\jmath}.} \left( \hat{\mathbf{a}}\_{\cdot} \right) \right)\_{a}^{a}}{\sum\_{a \in I, \mathfrak{s}} \left| \mathbf{A}^{k+1} \left( \mathbf{A}^{k+1} \right)^{\*} \right|\_{a}^{a}},\tag{9}$$

$$a\_{ij}^{\star,l} = \frac{\sum\_{\beta \in I\_{\iota \pi} \{i\}} \mathbf{cdet}\_i \left( \left( \left( \mathbf{A}^{k+1} \right)^\* \mathbf{A}^{k+1} \right)\_{\ i} \left( \bar{\mathbf{a}}\_j \right) \right)\_{\beta}^{\beta}}{\sum\_{\beta \in I\_{\iota \pi}} \left| \left( \mathbf{A}^{k+1} \right)^\* \mathbf{A}^{k+1} \right|\_{\beta}^{\beta}},\tag{10}$$

where **<sup>a</sup>**^*<sup>i</sup>:* is the *<sup>i</sup>*th row of **<sup>A</sup>**^ <sup>¼</sup> **<sup>A</sup>***<sup>k</sup>* **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> � �<sup>∗</sup> and **a** *^ :<sup>j</sup>* is the *j*th column of **A** *^* ¼ **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> � �<sup>∗</sup> **A***k* .

**Theorem 3.4.** *Let* **<sup>A</sup>** <sup>∈</sup> *<sup>n</sup>*�*<sup>n</sup> <sup>s</sup> ,* Indð Þ¼ **<sup>A</sup>** *k and* rank **<sup>A</sup>***<sup>k</sup>* � � <sup>¼</sup> *<sup>s</sup>*1*. Then its MPCEP inverse* **<sup>A</sup>**†,† <sup>¼</sup> *<sup>a</sup>*†,† *ij* � � *is expressed by componentwise*

*Quaternion MPCEP, CEPMP, and MPCEPMP Generalized Inverses DOI: http://dx.doi.org/10.5772/intechopen.103087*

$$a\_{ij}^{\dagger,\dagger} = \frac{\sum\_{a \in I\_{1,\pi}\{\cdot\}} \text{rdet}\_{\cdot j} \left( \left( \mathbf{A}^{k+1} \left( \mathbf{A}^{k+1} \right)^{\*} \right)\_{j\cdot} \left( \mathbf{v}\_{i\cdot}^{(1)} \right) \right)\_{a}^{a}}{\sum\_{\beta \in I\_{\imath,\imath}} |\mathbf{A}^{\*}\mathbf{A}|\_{\beta}^{\beta} \sum\_{a \in I\_{\imath,\imath}} \left| \mathbf{A}^{k+1} \left( \mathbf{A}^{k+1} \right)^{\*} \right|\_{a}^{a}} \tag{11}$$

$$\begin{split} \mathbf{1} = \frac{\sum\_{\boldsymbol{\beta} \in I\_{\boldsymbol{\mu}, \boldsymbol{\mu}} \{\boldsymbol{i}\}} \mathbf{c} \mathbf{det}\_{i} \Big( \left( \mathbf{A}^{\*} \mathbf{A} \right)\_{i} \Big( \mathbf{u}\_{j}^{(1)} \Big) \Big)\_{\boldsymbol{\beta}}^{\boldsymbol{\beta}}}{\sum\_{\boldsymbol{\beta} \in I\_{\boldsymbol{\mu}, \boldsymbol{\mu}}} |\mathbf{A}^{\*} \mathbf{A}|\_{\boldsymbol{\beta}}^{\boldsymbol{\beta}} \sum\_{a \in I\_{\boldsymbol{\mu}, \boldsymbol{\mu}}} \Big| \mathbf{A}^{k+1} \Big( \mathbf{A}^{k+1} \Big)^{\*} \Big|\_{a}^{a}}, \end{split} \tag{12}$$

where

$$\mathbf{v}\_{i.}^{(1)} = \left[\sum\_{\boldsymbol{\beta} \in I\_{\boldsymbol{s}\boldsymbol{u}}\{\boldsymbol{i}\}} \text{cdet}\_{i} \big( (\mathbf{A}^{\ast}\mathbf{A})\_{\boldsymbol{i}}(\tilde{\mathbf{a}}\_{\boldsymbol{I}}) \big)\_{\boldsymbol{\beta}}^{\boldsymbol{\beta}} \right] \in \mathbb{H}^{1 \times n}, \ l = \mathbf{1}, \ldots, n,\tag{13}$$
 
$$\mathbf{u}\_{j}^{(1)} = \left[\sum\_{\boldsymbol{a} \in I\_{\boldsymbol{u}\boldsymbol{s}}\{\boldsymbol{j}\}} \text{rdet}\_{\boldsymbol{j}} \big( \left( \mathbf{A}^{k+1} \left( \mathbf{A}^{k+1} \right)^{\ast} \right)\_{\boldsymbol{j}} \big( \tilde{\mathbf{a}}\_{\boldsymbol{f}} \big) \big)\_{\boldsymbol{a}}^{\boldsymbol{a}} \right] \in \mathbb{H}^{n \times 1}, \ f = \mathbf{1}, \ldots, n,$$

$$
\mathbf{u}\_j^{(1)} = \left[ \sum\_{a \in l\_{1^k}(f)} \mathbf{r} \text{det}\_j \left( \left\langle \mathbf{A}^{k+1} \left( \mathbf{A}^{k+1} \right) \right\rangle\_{f.} \left( \tilde{\mathbf{a}}\_{f.} \right) \right)\_a \right] \in \mathbb{H}^{n \times 1}, f = \mathbf{1}, \dots, n,
$$

and  $\tilde{\mathbf{a}}\_l$  and  $\tilde{\mathbf{a}}\_{f.}$  are the  $l$ th column and the  $f$ th row of  $\tilde{\mathbf{A}} = \mathbf{A}^\* \mathbf{A}^{k+1} \left( \mathbf{A}^{k+1} \right)^\*.
$ 

*Proof.* By (2), we have

$$a\_{ij}^{\dagger,\dagger} = \sum\_{l=1}^{n} q\_{il} a\_{lj}^{\dagger,r} \,. \tag{14}$$

Using (7) and (9) for the determinantal representations of **<sup>Q</sup>***<sup>A</sup>* <sup>¼</sup> **<sup>A</sup>**† **<sup>A</sup>** <sup>¼</sup> *qij* � � and **A**† , respectively, from (14) it follows

*a*†,† *ij* <sup>¼</sup> <sup>X</sup>*<sup>n</sup> l*¼1 X*<sup>n</sup> f*¼1 P *<sup>β</sup>* <sup>∈</sup>*Js*,*n*f g*<sup>i</sup>* cdet*<sup>i</sup>* **<sup>A</sup>**<sup>∗</sup> ð Þ **<sup>A</sup>** *:<sup>i</sup>* **<sup>a</sup>**\_*:<sup>f</sup>* � � � � *<sup>β</sup> β* P *<sup>β</sup>* <sup>∈</sup>*Js*,*<sup>n</sup>* **<sup>A</sup>**<sup>∗</sup> j j **<sup>A</sup>** *<sup>β</sup> β* � P *α*∈*Is* <sup>1</sup>,*n*f g*<sup>j</sup>* rdet*<sup>j</sup>* **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> � � � �<sup>∗</sup> *j:* ð Þ **a**^*<sup>l</sup>:* � �*<sup>α</sup> α* P *α*∈*Is* <sup>1</sup>,*<sup>n</sup>* **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> � �<sup>∗</sup> � � � � � � *α α* ¼ <sup>¼</sup> <sup>X</sup>*<sup>n</sup> l*¼1 X*<sup>n</sup> f*¼1 P *<sup>β</sup>* <sup>∈</sup>*Js*,*n*f g*<sup>i</sup>* cdet*<sup>i</sup>* **<sup>A</sup>**<sup>∗</sup> ð Þ **<sup>A</sup>** *:<sup>i</sup>* **<sup>e</sup>***:<sup>f</sup>* � � � � *<sup>β</sup> β* P *<sup>β</sup>* <sup>∈</sup>*Js*,*<sup>n</sup>* **<sup>A</sup>**<sup>∗</sup> j j **<sup>A</sup>** *<sup>β</sup> β a*~*fl* P *<sup>α</sup>*∈*Is*1,*n*f g*<sup>j</sup>* rdet*<sup>j</sup>* **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> � � � �<sup>∗</sup> *j:* ð Þ **e***<sup>l</sup>:* � �*<sup>α</sup> α* P *α* ∈*Is* <sup>1</sup>,*<sup>n</sup>* **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> � �<sup>∗</sup> � � � � � � *α α* ,

where **e***:<sup>f</sup>* and **e***<sup>l</sup>:* are the *f*th column and the *l*th row of the unit matrix **I***n*, **a**^*<sup>l</sup>:* is the *<sup>l</sup>*th row of **<sup>A</sup>**^ <sup>¼</sup> **<sup>A</sup>***<sup>k</sup>* **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> � �<sup>∗</sup> , and *<sup>a</sup>*~*fl* is the (*fl*)th element of **<sup>A</sup>**<sup>~</sup> <sup>¼</sup> **<sup>A</sup>**<sup>∗</sup> **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> � �<sup>∗</sup> . If we denote by

$$\mathbf{b}\_{il}^{(1)} \coloneqq \sum\_{f=1}^{n} \sum\_{\boldsymbol{\beta} \in I\_{l,n}\{i\}} \mathbf{c} \mathbf{det}\_{i} \big( (\mathbf{A}^{\*} \mathbf{A})\_{,i} (\mathbf{e}\_{f}) \big)\_{\boldsymbol{\beta}}^{\boldsymbol{\beta}} \tilde{\mathbf{a}}\_{\boldsymbol{\beta}} = \sum\_{\boldsymbol{\beta} \in I\_{l,n}\{i\}} \mathbf{c} \mathbf{det}\_{i} \big( (\mathbf{A}^{\*} \mathbf{A})\_{,i} (\tilde{\mathbf{a}}\_{l}) \big)\_{\boldsymbol{\beta}}^{\boldsymbol{\beta}}$$

the *l*th component of a row-vector **v**ð Þ<sup>1</sup> *<sup>i</sup>:* ¼ *v* ð Þ1 *<sup>i</sup>*<sup>1</sup> , … , *v* ð Þ1 *in* h i, then

$$\begin{split} & \sum\_{l=1}^{n} \boldsymbol{\nu}\_{il}^{(1)} \sum\_{a \in I\_{1,n}\{\,\,\} } \mathbf{rdet}\_{j} \Big( \left( \mathbf{A}^{k+1} \left( \mathbf{A}^{k+1} \right)^{\*} \right)\_{j.} \left( \mathbf{e}\_{l} \right) \Big)\_{a}^{a} \\ & \quad = \sum\_{a \in I\_{\boldsymbol{\nu}\_{l},n}\{\,\,\boldsymbol{j}\} } \mathbf{rdet}\_{j} \Big( \left( \mathbf{A}^{k+1} \left( \mathbf{A}^{k+1} \right)^{\*} \right)\_{\boldsymbol{j}.} \left( \mathbf{v}\_{i.}^{(1)} \right) \Big)\_{a}^{a} . \end{split}$$

**129**

So, we have (11). By putting

$$\begin{split} u\_{\hat{f}}^{(1)} &:= \sum\_{l=1}^{n} \tilde{a}\_{\hat{f}l} \sum\_{a \in I\_{1,\mathbf{r}}\{\,\,\}} \mathsf{rdet}\_{j} \Big( \left( \mathbf{A}^{k+1} \Big( \mathbf{A}^{k+1} \Big)^{\*} \right)\_{\stackrel{\cdot}{.}} \Big( \mathbf{e}\_{l.} \Big) \Big)\_{a}^{a} \\ &= \sum\_{a \in I\_{1,\mathbf{r}}\{\,\,\}} \mathsf{rdet}\_{j} \Big( \left( \mathbf{A}^{k+1} \Big( \mathbf{A}^{k+1} \Big)^{\*} \right)\_{\stackrel{\cdot}{.}} \Big( \tilde{\mathbf{a}}\_{\stackrel{\cdot}{.}} \Big) \Big)\_{a}^{a} \end{split}$$

as the *f*th component of a column-vector **u**ð Þ<sup>1</sup> *:<sup>j</sup>* <sup>¼</sup> *<sup>u</sup>*ð Þ<sup>1</sup> <sup>1</sup>*<sup>j</sup>* , … , *<sup>u</sup>*ð Þ<sup>1</sup> *nj* h i*<sup>T</sup>* , it follows

$$\sum\_{j=1}^{n} \sum\_{\boldsymbol{\beta} \in I\_{\boldsymbol{\mu}, \boldsymbol{\upbeta}}\{\boldsymbol{i}\}} \mathsf{cdet}\_{i} \big( (\mathbf{A}^{\*}\mathbf{A})\_{\boldsymbol{i}} \big( \mathbf{e}\_{\boldsymbol{\beta}} \big) \big)\_{\boldsymbol{\beta}}^{\boldsymbol{\beta}} u\_{\boldsymbol{j}\boldsymbol{\upbeta}}^{(1)} = \sum\_{\boldsymbol{\beta} \in I\_{\boldsymbol{\upbeta}}\{\boldsymbol{i}\}} \mathsf{cdet}\_{i} \big( (\mathbf{A}^{\*}\mathbf{A})\_{\boldsymbol{i}} \big( \mathbf{u}\_{\boldsymbol{j}}^{(1)} \big) \big)\_{\boldsymbol{\beta}}^{\boldsymbol{\beta}}.$$

Hence, we obtain (12). □ Determinantal representations of a complex MPCEP inverse are obtained by substituting row-column determinants for usual determinants in (11)–(12).

**Corollary 3.5.** *Let* **A** ∈ *<sup>n</sup>*�*<sup>n</sup> <sup>s</sup> ,* Indð Þ¼ **<sup>A</sup>** *k and* rank **<sup>A</sup>***<sup>k</sup>* � � <sup>¼</sup> *<sup>s</sup>*1*. Then its MPCEP inverse* **<sup>A</sup>**†,† <sup>¼</sup> *<sup>a</sup>*†,† *ij* � � *has the following determinantal representations*

$$\begin{split} a\_{ij}^{\dagger,\dagger} &= \frac{\sum\_{\alpha \in I\_{\imath,\imath} \{\,\,\}} \Big| \left( \mathbf{A}^{k+1} \left( \mathbf{A}^{k+1} \right)^{\ast} \right)\_{j,\cdot} \left( \mathbf{v}\_{i}^{(1)} \right) \Big|\_{\alpha}^{\alpha}}{\sum\_{\beta \in I\_{\imath,\imath} \,\vert} |\mathbf{A}^{\ast}\,\mathbf{A}|\_{\beta}^{\beta} \sum\_{\alpha \in I\_{\imath,\imath}} \Big| \mathbf{A}^{k+1} \left( \mathbf{A}^{k+1} \right)^{\ast} \Big|\_{\alpha}^{\alpha}} \\ &= \frac{\sum\_{\beta \in I\_{\imath,\imath} \,\vert i} \Big| \left( \mathbf{A}^{\ast} \,\mathbf{A} \right)\_{i} \left( \mathbf{u}\_{j}^{(1)} \right) \Big|\_{\beta}^{\beta}}{\sum\_{\beta \in I\_{\imath,\imath} \,\vert} |\mathbf{A}^{\ast}\,\mathbf{A}|\_{\beta}^{\beta} \sum\_{\alpha \in I\_{\imath\_{1},\imath}} \Big| \mathbf{A}^{k+1} \Big( \mathbf{A}^{k+1} \right)^{\ast} \Big|\_{\alpha}^{\alpha}}, \end{split}$$

where

$$\begin{aligned} \mathbf{v}\_{i.}^{(1)} &= \left[ \sum\_{\boldsymbol{\beta} \in I\_{\boldsymbol{\beta}, \boldsymbol{\pi}} \{\boldsymbol{\beta}\}} \left| (\mathbf{A}^{\*} \mathbf{A})\_{\boldsymbol{\beta}} (\widetilde{\mathbf{a}}\_{\boldsymbol{\beta}}) \right|\_{\boldsymbol{\beta}}^{\boldsymbol{\beta}} \right] \in \mathbb{C}^{1 \times n}, \; l = 1, \ldots, n, \\\ \mathbf{u}\_{j}^{(1)} &= \left[ \sum\_{\boldsymbol{a} \in I\_{\boldsymbol{1}, \boldsymbol{\pi}} \{\boldsymbol{\beta}\}} \left| \left( \mathbf{A}^{k+1} \left( \mathbf{A}^{k+1} \right)^{\ast} \right)\_{j.} \right|\_{\boldsymbol{\beta}} (\widetilde{\mathbf{a}}\_{\boldsymbol{f}, \boldsymbol{\epsilon}}) \right|\_{\boldsymbol{a}}^{\boldsymbol{a}} \in \mathbb{C}^{n \times 1}, \; f = \mathbf{1}, \ldots, n, \end{aligned} \tag{15}$$

*and* **<sup>a</sup>**~*:<sup>l</sup> and* **<sup>a</sup>**<sup>~</sup> *<sup>f</sup>: are the lth column and the fth row of* **<sup>A</sup>**<sup>~</sup> <sup>¼</sup> **<sup>A</sup>**<sup>∗</sup> **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> � �<sup>∗</sup> . **Theorem 3.6.** *Let* **<sup>A</sup>** <sup>∈</sup> *<sup>n</sup>*�*<sup>n</sup> <sup>s</sup> ,* Indð Þ¼ **<sup>A</sup>** *k and* rank **<sup>A</sup>***<sup>k</sup>* � � <sup>¼</sup> *<sup>s</sup>*1*. Then its CEPMP inverse* **<sup>A</sup>**†,† <sup>¼</sup> *<sup>a</sup>*†,† *ij* � � *has the following determinantal representations*

*Quaternion MPCEP, CEPMP, and MPCEPMP Generalized Inverses DOI: http://dx.doi.org/10.5772/intechopen.103087*

$$a\_{ij}^{\dagger,\dagger} = \frac{\sum\_{a \in I\_{\imath s} \{\cdot\} \text{ } \text{vert}\_j} \text{rdet}\_j \left( \left( \mathbf{A} \mathbf{A}^\* \right)\_{j.} \left( \mathbf{v}\_{i.}^{(2)} \right) \right)\_a^a}{\sum\_{a \in I\_{\imath s} \: \vert} |\mathbf{A} \mathbf{A}^\* \; \vert\_a^a \sum\_{\beta \in I\_{\imath s} \: \vert} \left| \left( \mathbf{A}^{k+1} \right)^\* \mathbf{A}^{k+1} \right|\_{\beta}^{\beta}} \tag{16}$$

$$=\frac{\sum\_{\boldsymbol{\beta}\in J\_{\boldsymbol{\gamma},\boldsymbol{\pi}}\{\boldsymbol{i}\}}\mathtt{cdet}\_{i}\left(\left(\left(\mathbf{A}^{k+1}\right)^{\*}\mathbf{A}^{k+1}\right)\_{\boldsymbol{i}}\left(\mathbf{u}\_{j}^{(2)}\right)\right)\_{\boldsymbol{\beta}}^{\boldsymbol{\beta}}}{\sum\_{\boldsymbol{a}\in I\_{\boldsymbol{a}}\boldsymbol{\pi}}|\mathbf{A}\mathbf{A}^{\*}|\_{\boldsymbol{a}}^{\boldsymbol{a}}\sum\_{\boldsymbol{\beta}\in J\_{\boldsymbol{\gamma},\boldsymbol{\pi}}}\left|\left(\mathbf{A}^{k+1}\right)^{\*}\mathbf{A}^{k+1}\right|\_{\boldsymbol{\beta}}^{\boldsymbol{\beta}}},\tag{17}$$

where

$$\begin{aligned} \mathbf{v}\_{i}^{(2)} &= \left[ \sum\_{\boldsymbol{\beta} \in I\_{1s}(i)} \text{cdet}\_{i} \big( \left( \left( \mathbf{A}^{k+1} \right)^{\*} \mathbf{A}^{k+1} \right)\_{i} (\hat{\mathbf{a}}\_{\boldsymbol{l}}) \big)\_{\boldsymbol{\beta}}^{\boldsymbol{\beta}} \right] \in \mathbb{H}^{1 \times n}, \; l = \mathbf{1}, \ldots, n, \\\ \mathbf{u}\_{j}^{(2)} &= \left[ \sum\_{\boldsymbol{a} \in I\_{1s}(\boldsymbol{j})} \text{rdet}\_{j} \big( \left( \mathbf{A} \mathbf{A}^{\*} \right)\_{\boldsymbol{j}} (\hat{\mathbf{a}}\_{\boldsymbol{f}}) \big)\_{\boldsymbol{a}}^{\boldsymbol{a}} \right] \in \mathbb{H}^{n \times 1}, \; f = 1, \ldots, n. \end{aligned} \tag{18}$$

Here **<sup>a</sup>**^*:<sup>l</sup>* and **<sup>a</sup>**^ *<sup>f</sup>:* are the *<sup>l</sup>*th column and the *<sup>f</sup>*th row of **<sup>A</sup>**^ <sup>¼</sup> **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> � �<sup>∗</sup> **A***<sup>k</sup>*þ<sup>1</sup> **A**<sup>∗</sup> .

*Proof.* The proof is similar to the proof of Theorem 3.4 by using the representation (3) for the CEPMP inverse.

**Corollary 3.7.** *Let* **A** ∈ *<sup>n</sup>*�*<sup>n</sup> <sup>s</sup> ,* Indð Þ¼ **<sup>A</sup>** *k and* rank **<sup>A</sup>***<sup>k</sup>* � � <sup>¼</sup> *<sup>s</sup>*1*. Then its CEPMP inverse* **<sup>A</sup>**†,† <sup>¼</sup> *<sup>a</sup>*†,† *ij* � � *has the following determinantal representations*

$$\begin{split} a\_{ij}^{\dagger,\dagger} &= \frac{\sum\_{\boldsymbol{\alpha} \in I\_{\boldsymbol{\alpha}}\left(\boldsymbol{\beta}\right)} \left| \left(\mathbf{A}\mathbf{A}^{\ast}\right)\_{\boldsymbol{\beta}} \left(\mathbf{v}\_{i.}^{(2)}\right) \right|\_{\boldsymbol{\alpha}}^{\boldsymbol{\alpha}}}{\sum\_{\boldsymbol{\alpha} \in I\_{\boldsymbol{\alpha}}\left(\boldsymbol{\mathbf{A}}\mathbf{A}^{\ast}\right)^{\boldsymbol{\alpha}}} \left| \left(\mathbf{A}^{k+1}\right)^{\ast} \mathbf{A}^{k+1} \right|\_{\boldsymbol{\beta}}^{\boldsymbol{\beta}}} \\ &= \frac{\sum\_{\boldsymbol{\beta} \in I\_{\boldsymbol{\alpha},\boldsymbol{\alpha}}\left(\boldsymbol{i}\right)} \mathbf{c} \mathbf{d} \mathbf{e}\_{i} \left( \left(\left(\mathbf{A}^{k+1}\right)^{\ast} \mathbf{A}^{k+1}\right)\_{\boldsymbol{\beta}} \left(\mathbf{u}\_{j}^{(2)}\right) \right)\_{\boldsymbol{\beta}}^{\boldsymbol{\beta}}}{\sum\_{\boldsymbol{\alpha} \in I\_{\boldsymbol{\alpha}}\left(\mathbf{A} \mathbf{A}^{\ast}\right)\_{\boldsymbol{\alpha}} \left| \left(\mathbf{A}^{k+1}\right)^{\ast} \mathbf{A}^{k+1} \right|\_{\boldsymbol{\beta}}^{\boldsymbol{\beta}}} \right)^{\boldsymbol{\beta}}}, \end{split}$$

where

$$\begin{aligned} \mathbf{v}\_{i.}^{(2)} &= \left[ \sum\_{\boldsymbol{\beta} \in I\_{\boldsymbol{\eta}, \boldsymbol{u}}\{i\}} \Big| \left( \left( \mathbf{A}^{k+1} \right)^{\*} \mathbf{A}^{k+1} \right)\_{i} (\hat{\mathbf{a}}\_{\boldsymbol{d}}) \Big|\_{\boldsymbol{\beta}}^{\boldsymbol{\beta}} \right] \in \mathbb{C}^{1 \times n}, \; l = \mathbf{1}, \ldots, n, \\\ \mathbf{u}\_{j}^{(2)} &= \left[ \sum\_{\boldsymbol{a} \in I\_{\boldsymbol{\eta}, \boldsymbol{u}}\{j\}} \Big| \left( \mathbf{A} \mathbf{A}^{\*} \right)\_{j} (\hat{\mathbf{a}}\_{\boldsymbol{f}}) \Big|\_{\boldsymbol{\alpha}}^{\boldsymbol{a}} \right] \in \mathbb{C}^{n \times 1}, \; f = \mathbf{1}, \ldots, n. \end{aligned} \tag{19}$$

Here **<sup>a</sup>**^*:<sup>l</sup>* and **<sup>a</sup>**^ *<sup>f</sup>:* are the *<sup>l</sup>*th column and the *<sup>f</sup>*th row of **<sup>A</sup>**^ <sup>¼</sup> **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> � �<sup>∗</sup> **A***<sup>k</sup>*þ<sup>1</sup> **A**<sup>∗</sup> . **Theorem 3.8.** *Let* **<sup>A</sup>** <sup>∈</sup> *<sup>n</sup>*�*<sup>n</sup> <sup>s</sup> ,* Indð Þ¼ **<sup>A</sup>** *k and* rank **<sup>A</sup>***<sup>k</sup>* � � <sup>¼</sup> *<sup>s</sup>*1*. Then its right MPCEPMP inverse* **<sup>A</sup>**†,†,†,*<sup>r</sup>* <sup>¼</sup> *<sup>a</sup>*†,†,†,*<sup>r</sup> ij* � � *has the following determinantal representations*

$$a\_{ij}^{\star,\uparrow,\uparrow,r} = \frac{\sum\_{a \in I\_{\iota,\mathbf{z}}\{\cdot\}\text{ rdet}\_{j}\big(\left(\mathbf{A}\mathbf{A}^{\star}\right)\_{j}\big(\boldsymbol{\phi}\_{i.}^{(1)}\big)\big)\_{a}^{a}}{\left(\sum\_{a \in I\_{\iota,\mathbf{z}}\big|} \left|\mathbf{A}\mathbf{A}^{\star}\right|\_{a}^{a}\right)^{2}\sum\_{\beta \in J\_{\iota,\mathbf{z}}\big|}{\sum\_{\beta \in J\_{\iota,\mathbf{z}}\big|} \left(\left(\mathbf{A}^{k+1}\right)^{\ast}\mathbf{A}^{k+1}\right)\_{\cdot,i}^{\beta}\mathbf{A}^{k+1}\big|\_{\beta}^{\beta}}} = \tag{20}$$
 
$$= \frac{\sum\_{\beta \in J\_{\iota,\mathbf{z}}\{i\}} \text{cdet}\_{i}\big(\left(\left(\mathbf{A}^{k+1}\right)^{\ast}\mathbf{A}^{k+1}\right)\_{\cdot,i}\big(\boldsymbol{\nu}\_{j}^{(1)}\big)\big)\_{\beta}^{\beta}}{\left(\sum\_{\beta \in J\_{\iota,\mathbf{z}}\big|} \left|\mathbf{A}\mathbf{A}^{\star}\right|\_{\beta}^{\beta}\right)^{2}\sum\_{a \in J\_{\iota,\mathbf{z}}\big|}\left(\left(\mathbf{A}^{k+1}\right)^{\ast}\mathbf{A}^{k+1}\right)\_{a}^{a}}},\tag{21}$$

where

$$\begin{aligned} \boldsymbol{\phi}\_{i.}^{(1)} &= \left[ \sum\_{\boldsymbol{\beta} \in I\_{\boldsymbol{\eta}\_{1}\boldsymbol{n}} \{i\}} \mathsf{cdet}\_{i} \Big( \left( \left( \mathbf{A}^{k+1} \right)^{\*} \mathbf{A}^{k+1} \right)\_{\boldsymbol{a}} (\hat{\mathbf{u}}\_{\boldsymbol{l}}) \Big)\_{\boldsymbol{\beta}}^{\boldsymbol{\beta}} \right] \in \mathbb{H}^{1 \times n}, \; l = 1, \ldots, n, \\\boldsymbol{\nu}\_{j}^{(1)} &= \left[ \sum\_{\boldsymbol{a} \in I\_{\boldsymbol{n}} \{\boldsymbol{j} \}} \mathsf{rdet}\_{j} \Big( \left( \mathbf{A} \mathbf{A}^{\*} \right)\_{\boldsymbol{j}.} (\hat{\mathbf{u}}\_{\boldsymbol{f}}) \Big)\_{\boldsymbol{a}}^{\boldsymbol{a}} \right] \in \mathbb{H}^{n \times 1}, \; f = 1, \ldots, n. \end{aligned}$$

*Here* **<sup>u</sup>**^*:<sup>l</sup> and* **<sup>u</sup>**^ *<sup>f</sup>: are the lth column and the fth row of* **<sup>U</sup>**^ <sup>¼</sup> **<sup>U</sup>**2**AA**<sup>∗</sup> *, and the matrix* **<sup>U</sup>**<sup>2</sup> *is constructed from the columns* (18).

*Proof.* Owing to (4), we have

$$a\_{ij}^{\dagger,\dagger,\dagger,r} = \sum\_{t=1}^{n} a\_{it}^{\dagger,\dagger} p\_{tj}.\tag{22}$$

Applying (8) for the determinantal representation of **<sup>P</sup>***<sup>A</sup>* <sup>¼</sup> **AA**† <sup>¼</sup> *pij* � � and (17) for the determinantal representation of **A**†,† in (22), we obtain

*a*†,†,†,*<sup>r</sup> ij* <sup>¼</sup> <sup>X</sup>*<sup>n</sup> t*¼1 P *β* ∈ *Js* <sup>1</sup>,*n*f g*<sup>i</sup>* cdet*<sup>i</sup>* **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> � �<sup>∗</sup> **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> � � *:<sup>i</sup>* **<sup>u</sup>**ð Þ<sup>2</sup> *:t* � � � � *<sup>β</sup> β* P *<sup>α</sup>*∈*Is*,*<sup>n</sup>* **AA**<sup>∗</sup> j j*<sup>α</sup> α* P *β* ∈*Is* <sup>1</sup>,*<sup>n</sup>* **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> � �<sup>∗</sup> **A***<sup>k</sup>*þ<sup>1</sup> � � � � � � *β β* � P *<sup>α</sup>*∈*Is*,*n*f g*<sup>j</sup>* rdet*<sup>j</sup>* **AA**<sup>∗</sup> ð Þ *<sup>j</sup>:* ð Þ **a**€*t:* � �*<sup>α</sup>* <sup>P</sup> *<sup>α</sup> <sup>α</sup>*∈*Is*,*<sup>n</sup>* **AA**<sup>∗</sup> j j*<sup>α</sup> α* ¼ <sup>¼</sup> <sup>X</sup>*<sup>n</sup> l*¼1 X*<sup>n</sup> f*¼1 P *β* ∈*Js* <sup>1</sup>,*n*f g*<sup>i</sup>* cdet*<sup>i</sup>* **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> � �<sup>∗</sup> **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> � � *:<sup>i</sup>* **<sup>e</sup>***:<sup>f</sup>* � � � � *<sup>β</sup> β* P *β* ∈*Js* <sup>1</sup>,*<sup>n</sup>* **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> � �<sup>∗</sup> **A***<sup>k</sup>*þ<sup>1</sup> � � � � � � *β β u*^*fl* � P *<sup>α</sup>*∈*Is*,*n*f g*<sup>j</sup>* rdet*<sup>j</sup>* **AA**<sup>∗</sup> ð Þ *<sup>j</sup>:* ð Þ **e***<sup>l</sup>:* � �*<sup>α</sup> α* P *<sup>α</sup>*<sup>∈</sup> *Is*,*<sup>n</sup>* **AA**<sup>∗</sup> j j*<sup>α</sup> α* � �<sup>2</sup> ,

where **e***:<sup>f</sup>* and **e***<sup>l</sup>:* are the *f*th column and the *l*th row of the unit matrix **I***n*, and *u*^*fl* is the (*fl*)th element of **<sup>U</sup>**^ <sup>¼</sup> **<sup>U</sup>**2**AA**<sup>∗</sup> . The matrix **<sup>U</sup>**<sup>2</sup> <sup>¼</sup> **<sup>u</sup>**ð Þ<sup>2</sup> *:*<sup>1</sup> , … , **<sup>u</sup>**ð Þ<sup>2</sup> *:n* h i is constructed from the columns (18). If we denote by

$$\begin{split} \boldsymbol{\phi}\_{il}^{(1)} &:= \sum\_{f=1}^{n} \sum\_{\beta \in J\_{\mathbf{i}\_{1},\mathbf{a}}\{i\}} \mathsf{cdet}\_{i} \Big( \Big( \Big( \mathbf{A}^{k+1} \Big)^{\*} \mathbf{A}^{k+1} \Big)\_{:i} (\mathbf{e}\_{f}) \Big)\_{\beta}^{\beta} \boldsymbol{\hat{u}}\_{\beta} \\ &= \sum\_{\beta \in J\_{\mathbf{i}\_{1},\mathbf{a}}\{i\}} \mathsf{cdet}\_{i} \Big( \Big( \Big( \Big( \mathbf{A}^{k+1} \Big)^{\*} \mathbf{A}^{k+1} \Big)\_{:i} (\mathbf{\hat{u}}\_{\cdot l}) \Big)\_{\beta}^{\beta} \Big) \end{split}$$

*Quaternion MPCEP, CEPMP, and MPCEPMP Generalized Inverses DOI: http://dx.doi.org/10.5772/intechopen.103087*

the *l*th component of a row-vector *ϕ*ð Þ<sup>1</sup> *<sup>i</sup>:* <sup>¼</sup> *<sup>ϕ</sup>*ð Þ<sup>1</sup> *<sup>i</sup>*<sup>1</sup> , … , *<sup>ϕ</sup>*ð Þ<sup>1</sup> *in* h i, then

$$\sum\_{l=1}^{n} \phi\_{il}^{(1)} \sum\_{a \in I\_{l,\mathbf{x}}\{\,\,\}} \mathbf{rdet}\_{j} \Big( (\mathbf{A} \mathbf{A}^{\*} \,)\_{j,\cdot} (\mathbf{e}\_{l,\cdot}) \Big)\_{a}^{a} = \sum\_{a \in I\_{l,\mathbf{x}}\{\,\,\,j\}} \mathbf{rdet}\_{j} \Big( (\mathbf{A} \mathbf{A}^{\*} \,)\_{j,\cdot} \Big( \phi\_{i.}^{(1)} \Big) \Big)\_{a}^{a}.$$

Therefore, (20) holds. By putting

$$\mathbb{V}\boldsymbol{\mu}\_{\widehat{f}}^{(1)} \coloneqq \sum\_{l=1}^{n} \hat{\mathbf{a}}\_{\widehat{f}} \sum\_{a \in I\_{\iota,n}\{\,\,j\}} \mathsf{rdet}\_{j} \Big( (\mathbf{A}\mathbf{A}^{\*} \,)\_{\,j.} (\mathbf{e}\_{l.}) \Big)\_{a}^{a} = \sum\_{a \in I\_{\iota,n}\{\,\,j\}} \mathsf{rdet}\_{j} \Big( (\mathbf{A}\mathbf{A}^{\*} \,)\_{\,j.} (\hat{\mathbf{u}}\_{f.})\_{a}^{a} \Big)$$

as the *f*th component of a column-vector *ψ*ð Þ<sup>1</sup> *:<sup>j</sup>* <sup>¼</sup> *<sup>ψ</sup>*ð Þ<sup>1</sup> <sup>1</sup>*<sup>j</sup>* , … , *<sup>ψ</sup>*ð Þ<sup>1</sup> *nj* h i*<sup>T</sup>* , it follows

$$\sum\_{f=1}^{n} \sum\_{\substack{\boldsymbol{\rho} \in I\_{\boldsymbol{\rho}\_{1}} \{\boldsymbol{i}\}}} \mathsf{cdet}\_{i} \Big( \Big( \left( \mathbf{A}^{k+1} \right)^{\*} \mathbf{A}^{k+1} \Big)\_{\boldsymbol{i}} \Big( \mathbf{e}\_{f} \Big) \Big)\_{\boldsymbol{\rho}}^{\boldsymbol{\rho}} \boldsymbol{\mu}\_{\widehat{f}}^{(1)} = \sum\_{\boldsymbol{\rho} \in I\_{\boldsymbol{\rho}\_{1}} \{\boldsymbol{i}\}} \mathsf{cdet}\_{i} \Big( \Big( \Big( \mathbf{A}^{k+1} \Big)^{\*} \mathbf{A}^{k+1} \Big)\_{\boldsymbol{i}} \Big( \boldsymbol{\mu}\_{\widehat{j}}^{(1)} \Big) \Big)\_{\boldsymbol{\rho}}^{\boldsymbol{\rho}}$$

Thus, Eq. (21) holds.

**Corollary 3.9.** *Let* **A** ∈ *<sup>n</sup>*�*<sup>n</sup> <sup>s</sup> ,* Indð Þ¼ **<sup>A</sup>** *k and* rank **<sup>A</sup>***<sup>k</sup>* � � <sup>¼</sup> *<sup>s</sup>*1*. Then its right MPCEPMP inverse* **<sup>A</sup>**†,†,†,*<sup>r</sup>* <sup>¼</sup> *<sup>a</sup>*†,†,†,*<sup>r</sup> ij* � � *has the following determinantal representations*

$$\begin{split} \boldsymbol{a}\_{ij}^{\dagger,\dagger} &= \frac{\sum\_{\boldsymbol{a} \in I\_{\boldsymbol{a}} \left\{ \boldsymbol{j} \right\}} \Big| \left( \mathbf{A} \mathbf{A}^{\ast} \right)\_{j} \left( \boldsymbol{\phi}\_{i.}^{(1)} \right) \Big|\_{\boldsymbol{a}}^{\boldsymbol{a}}}{\left( \sum\_{\boldsymbol{a} \in I\_{\boldsymbol{a},\boldsymbol{a}} \left\{ \mathbf{A} \mathbf{A}^{\ast} \right\}\_{\boldsymbol{a}}^{\boldsymbol{a}} \right)^{2} \sum\_{\boldsymbol{\beta} \in I\_{\boldsymbol{\gamma}\_{1},\boldsymbol{a}} \left\{ \left( \mathbf{A}^{k+1} \right)^{\ast} \right\}\_{\boldsymbol{\beta}}^{\boldsymbol{a}} \mathbf{A}^{k+1} \Big|\_{\boldsymbol{\beta}}^{\boldsymbol{\beta}}}} \\\\ &= \frac{\sum\_{\boldsymbol{\beta} \in I\_{\boldsymbol{\gamma}\_{1},\boldsymbol{a}} \left\{ \boldsymbol{i} \right\}} \Big| \left( \left( \mathbf{A}^{k+1} \right)^{\ast} \mathbf{A}^{k+1} \right)\_{\boldsymbol{i}} \left( \boldsymbol{\nu}\_{\boldsymbol{j}}^{(1)} \right) \Big|\_{\boldsymbol{\beta}}^{\boldsymbol{\beta}}}{\left( \sum\_{\boldsymbol{\beta} \in I\_{\boldsymbol{a},\boldsymbol{a}}} \Big| \mathbf{A} \mathbf{A}^{\ast} \big|\_{\boldsymbol{\beta}}^{\boldsymbol{\beta}} \right)^{2} \sum\_{\boldsymbol{a} \in I\_{\boldsymbol{a},\boldsymbol{a}} \left\{ \left( \mathbf{A}^{k+1} \right)^{\ast} \right\}\_{\boldsymbol{a}}^{\boldsymbol{a}} \mathbf{A}^{k+1} \Big|\_{\boldsymbol{a}}^{\boldsymbol{a}}}}, \end{split}$$

where

$$\phi\_{i}^{(1)} = \left[ \sum\_{\boldsymbol{\beta} \in I\_{1,\boldsymbol{\pi}}\{i\}} \Big| \left( \left( \mathbf{A}^{k+1} \right)^{\*} \mathbf{A}^{k+1} \right)\_{\boldsymbol{\beta}} (\hat{\mathbf{u}}\_{J}) \Big|\_{\boldsymbol{\beta}}^{\boldsymbol{\beta}} \right] \in \mathbb{C}^{1 \times n}, \ l = \mathbf{1}, \ldots, n,$$

$$\boldsymbol{\Psi}\_{j}^{(1)} = \left[ \sum\_{a \in I\_{\boldsymbol{\pi}}\{j\}} \Big| \left( \mathbf{A} \mathbf{A}^{\*} \right)\_{j\cdot} (\hat{\mathbf{u}}\_{J}) \Big|\_{a}^{a} \right] \in \mathbb{C}^{n \times 1}, \ f = \mathbf{1}, \ldots, n.$$

*Here* **<sup>u</sup>**^*:<sup>l</sup> and* **<sup>u</sup>**^ *<sup>f</sup>: are the lth column and the fth row of* **<sup>U</sup>**^ <sup>¼</sup> **<sup>U</sup>**2**AA**<sup>∗</sup> *, and the matrix* **<sup>U</sup>**<sup>2</sup> *is constructed from the columns* (19).

**Theorem 3.10.** *Let* **<sup>A</sup>** <sup>∈</sup> *<sup>n</sup>*�*<sup>n</sup> <sup>s</sup> ,* Indð Þ¼ **<sup>A</sup>** *k and* rank **<sup>A</sup>***<sup>k</sup>* � � <sup>¼</sup> *<sup>s</sup>*1*. Then its left MPCEPMP inverse* **<sup>A</sup>**†,†,†,*<sup>l</sup>* <sup>¼</sup> *<sup>a</sup>*†,†,†,*<sup>l</sup> ij* � � *has the following determinantal representations*

$$a\_{ij}^{\star,\uparrow,\uparrow,\downarrow} = \frac{\sum\_{a \in I\_{\iota,\pi} \{\cdot\}} \text{rdet}\_j \left( \left( \mathbf{A}^{k+1} \left( \mathbf{A}^{k+1} \right)^\* \right)\_{j} \left( \phi\_{i.}^{(2)} \right) \right)\_a^a}{\left( \sum\_{\beta \in I\_{\iota,\pi}} |\mathbf{A}^\* \cdot \mathbf{A}|\_{\beta}^{\beta} \right)^2 \sum\_{a \in I\_{\iota,\pi}} \left| \mathbf{A}^{k+1} \left( \mathbf{A}^{k+1} \right)^\* \right|\_a^a} \tag{23}$$

$$=\frac{\sum\_{\boldsymbol{\beta}\in I\_{\boldsymbol{s},\boldsymbol{u}}\{\boldsymbol{i}\}}\mathsf{cdet}\_{i}\left(\left(\mathbf{A}^{\ast}\mathbf{A}\right)\_{i}\left(\boldsymbol{\nu}\_{\boldsymbol{j}}^{(2)}\right)\right)\_{\boldsymbol{\beta}}^{\boldsymbol{\beta}}}{\left(\sum\_{\boldsymbol{\beta}\in I\_{\boldsymbol{s},\boldsymbol{u}}}\left|\mathbf{A}^{\ast}\mathbf{A}\right|\_{\boldsymbol{\beta}}^{\boldsymbol{\beta}}\right)^{2}\sum\_{a\in I\_{\boldsymbol{s},\boldsymbol{u}}}\left|\mathbf{A}^{\boldsymbol{k}+1}\left(\mathbf{A}^{\boldsymbol{k}+1}\right)^{\ast}\right|\_{a}^{a}},\tag{24}$$

where

$$\begin{aligned} \boldsymbol{\phi}\_{i}^{(2)} &= \left[ \sum\_{\boldsymbol{\beta} \in I\_{\boldsymbol{u}}\{i\}} \mathsf{cdet}\_{i} \big( (\mathbf{A}^{\*}\mathbf{A})\_{i} (\bar{\mathbf{v}}\_{\boldsymbol{I}}) \big)\_{\boldsymbol{\beta}}^{\boldsymbol{\beta}} \right] \in \mathbb{H}^{1 \times n}, \quad t = 1, \ldots, n, \\\boldsymbol{\nu}\_{j}^{(2)} &= \left[ \sum\_{a \in I\_{\boldsymbol{u},\boldsymbol{\epsilon}}\{\boldsymbol{j}\}} \mathsf{rdet}\_{j} \Big( \left( \mathbf{A}^{k+1} \left( \mathbf{A}^{k+1} \right)^{\*} \right)\_{\boldsymbol{j}} (\bar{\mathbf{v}}\_{\boldsymbol{f}}.) \Big)\_{a}^{a} \right] \in \mathbb{H}^{n \times 1}, \quad f = 1, \ldots, n, \end{aligned}$$

*and* **<sup>v</sup>**~*:<sup>l</sup> and* **<sup>v</sup>**<sup>~</sup> *<sup>f</sup>: are the lth column and the fth row of* **<sup>V</sup>**<sup>~</sup> <sup>¼</sup> **<sup>A</sup>**<sup>∗</sup> **AV**1*, where the matrix* **V**<sup>1</sup> *is determined from the rows* (13).

*Proof.* Due to (6),

$$a\_{ij}^{\dagger,\dagger,\dagger,l} = \sum\_{t=1}^{n} q\_{it} a\_{tj}^{\dagger,\dagger}.\tag{25}$$

Using (7) for the determinantal representation of **<sup>Q</sup>***<sup>A</sup>* <sup>¼</sup> **<sup>A</sup>**† **<sup>A</sup>** <sup>¼</sup> *qij* � � and (9) for the determinantal representation of **A**†,† in (14), we obtain

*a*†,†,†,*<sup>l</sup> ij* <sup>¼</sup> <sup>X</sup>*<sup>n</sup> t*¼1 P *<sup>β</sup>* <sup>∈</sup> *Js*,*n*f g*<sup>i</sup>* cdet*<sup>i</sup>* **<sup>A</sup>**<sup>∗</sup> ð Þ **<sup>A</sup>** *:<sup>i</sup>* **<sup>a</sup>**\_ ð Þ*:<sup>t</sup>* � �*<sup>β</sup> β* P *<sup>β</sup>* <sup>∈</sup> *Js*,*<sup>n</sup>* **<sup>A</sup>**<sup>∗</sup> j j **<sup>A</sup>** *<sup>β</sup> β* � P *<sup>α</sup>*<sup>∈</sup> *Is*1,*n*f g*<sup>j</sup>* rdet*<sup>j</sup>* **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> � � � �<sup>∗</sup> *<sup>j</sup>:* **<sup>v</sup>**ð Þ<sup>1</sup> *t:* � � � � *<sup>α</sup> α* P *<sup>β</sup>* <sup>∈</sup>*Js*,*<sup>n</sup>* **<sup>A</sup>**<sup>∗</sup> j j **<sup>A</sup>** *<sup>β</sup> β* P *α* ∈*Is* <sup>1</sup>,*<sup>n</sup>* **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> � �<sup>∗</sup> � � � � � � *α α* <sup>¼</sup> <sup>X</sup>*<sup>n</sup> t*¼1 X*<sup>n</sup> f*¼1 P *<sup>β</sup>* <sup>∈</sup> *Js*,*n*f g*<sup>i</sup>* cdet*<sup>i</sup>* **<sup>A</sup>**<sup>∗</sup> ð Þ **<sup>A</sup>** *:<sup>i</sup>* **<sup>e</sup>***:<sup>f</sup>* � � � � *<sup>β</sup> β* P *<sup>β</sup>* <sup>∈</sup> *Js*,*<sup>n</sup>* **<sup>A</sup>**<sup>∗</sup> j j **<sup>A</sup>** *<sup>β</sup> β* � �<sup>2</sup> <sup>~</sup>*vft* � P *α*∈ *Is* <sup>1</sup>,*n*f g*<sup>j</sup>* rdet*<sup>j</sup>* **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> � � � �<sup>∗</sup> *j:* ð Þ **e***<sup>t</sup>:* � �*<sup>α</sup> α* P *α*∈ *Is* <sup>1</sup>,*<sup>n</sup>* **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> **<sup>A</sup>***<sup>k</sup>*þ<sup>1</sup> � �<sup>∗</sup> � � � � � � *α α* ,

where <sup>~</sup>*vft* is the (*ft*)th element of **<sup>V</sup>**<sup>~</sup> <sup>¼</sup> **<sup>A</sup>**<sup>∗</sup> **AV**<sup>1</sup> and the matrix **<sup>V</sup>**<sup>1</sup> is constructed from the rows (13). If we put

$$\begin{split} \phi\_{it}^{(2)} &:= \sum\_{f=1}^{n} \sum\_{\beta \in J\_{\iota,n}\{i\}} \mathsf{cdet}\_{i} \big( (\mathbf{A}^\* \mathbf{A})\_{.i} (\mathbf{e}\_f) \big)\_{\beta}^{\beta} \ddot{v}\_{\beta} \\ &= \sum\_{\beta \in J\_{\iota,n}\{i\}} \mathsf{cdet}\_{i} \big( (\mathbf{A}^\* \mathbf{A})\_{.i} (\ddot{\mathbf{v}}\_{\iota}) \big)\_{\beta}^{\beta} \end{split}$$

$$\begin{split} & \sum\_{t=1}^{n} \phi\_{\text{if}}^{(2)} \sum\_{a \in I\_{\iota\_{1},\iota} \{\,\} } \text{rdet}\_{j} \Big( \left( \mathbf{A}^{k+1} \Big( \mathbf{A}^{k+1} \Big)^{\*} \right)\_{j.} \big)\_{a}^{a} \\ & = \sum\_{a \in I\_{\iota\_{1},\iota} \{\,\,\} } \text{rdet}\_{j} \Big( \left( \mathbf{A}^{k+1} \Big( \mathbf{A}^{k+1} \Big)^{\*} \right)\_{j.} \left( \phi\_{i.}^{(2)} \right) \Big)\_{a}^{a}, \end{split}$$

$$\begin{split} \mathbb{W}\_{\mathcal{\tilde{f}}}^{(2)} &:= \sum\_{t=1}^{n} \tilde{u}\_{\mathcal{f}t} \sum\_{a \in I\_{\mathbb{I}\_{1},n}\{\,\,\}} \mathsf{rdet}\_{j} \Big( \left( \mathbf{A}^{k+1} \Big( \mathbf{A}^{k+1} \Big)^{\*} \right)\_{j} \Big)\_{\cdot}^{a} \big( \mathbf{e}\_{\mathsf{f},} \Big)\_{a}^{a} = \\ &= \sum\_{a \in I\_{\mathbb{I}\_{1},n}\{\,\,\}} \mathsf{rdet}\_{j} \Big( \left( \mathbf{A}^{k+1} \Big( \mathbf{A}^{k+1} \Big)^{\*} \right)\_{\cdot} \Big( \tilde{\mathbf{u}}\_{\cdot f,} \Big)\_{a}^{a} \end{split}$$

$$\sum\_{j=1}^{n} \sum\_{\beta \in I\_{\iota,\mathbf{v}}\{i\}} \mathsf{cdet}\_{i} \Big( (\mathbf{A}^\* \mathbf{A})\_{\,i} (\mathbf{e}\_{\mathcal{J}}) \Big)\_{\beta}^{\beta} \boldsymbol{\upmu}\_{\widehat{\mathcal{J}}}^{(2)} = \sum\_{\beta \in I\_{\iota,\mathbf{v}}\{i\}} \mathsf{cdet}\_{i} \Big( (\mathbf{A}^\* \mathbf{A})\_{\,i} \Big( \boldsymbol{\upmu}\_{\boldsymbol{j}}^{(2)} \Big) \Big)\_{\beta}^{\beta}$$

$$\begin{split} \boldsymbol{a}\_{ij}^{\dagger,\dagger,\dagger,l} &= \frac{\sum\_{\alpha \in I\_{\boldsymbol{\alpha},\boldsymbol{\alpha}} \{\boldsymbol{j}\}} \left| \left( \mathbf{A}^{k+1} \left( \mathbf{A}^{k+1} \right)^{\ast} \right)\_{\boldsymbol{j},\boldsymbol{\alpha}} \left( \boldsymbol{\phi}\_{\boldsymbol{i}}^{(2)} \right) \right|\_{a}^{a}}{\left( \sum\_{\beta \in I\_{\boldsymbol{\beta},\boldsymbol{\alpha}}} |\mathbf{A}^{\ast} \mathbf{A}|\_{\boldsymbol{\beta}}^{\beta} \right)^{2} \sum\_{\alpha \in I\_{\boldsymbol{\alpha},\boldsymbol{\alpha}}} \left| \mathbf{A}^{k+1} \left( \mathbf{A}^{k+1} \right)^{\ast} \right|\_{a}^{a}} \\ &= \frac{\sum\_{\beta \in I\_{\boldsymbol{\alpha},\boldsymbol{\alpha}} \{\boldsymbol{i}\}} \left| \left( \mathbf{A}^{\ast} \mathbf{A} \right)\_{\boldsymbol{i}} \left( \boldsymbol{\nu}\_{\boldsymbol{j}}^{(2)} \right) \right|\_{\boldsymbol{\beta}}^{\beta}}{\left( \sum\_{\beta \in I\_{\boldsymbol{\beta},\boldsymbol{\alpha}}} |\mathbf{A}^{\ast} \mathbf{A}|\_{\boldsymbol{\beta}}^{\beta} \right)^{2} \sum\_{\alpha \in I\_{\boldsymbol{1},\boldsymbol{\alpha}}} \left| \mathbf{A}^{k+1} \left( \mathbf{A}^{k+1} \right)^{\ast} \right|\_{a}^{a}}, \end{split}$$

$$\begin{aligned} \boldsymbol{\phi}\_{i.}^{(2)} &= \left[ \sum\_{\boldsymbol{\beta} \in I\_{\boldsymbol{\beta}, \boldsymbol{\pi}} \{i\}} \left| (\mathbf{A}^\* \mathbf{A})\_{\boldsymbol{\beta}} (\bar{\mathbf{v}}\_{\boldsymbol{\beta}}) \right|\_{\boldsymbol{\beta}}^{\boldsymbol{\beta}} \right] \in \mathbb{C}^{1 \times n}, \quad t = \mathbf{1}, \ldots, n, \\\boldsymbol{\nu}\_{j}^{(2)} &= \left[ \sum\_{a \in I\_{\boldsymbol{\pi}, \boldsymbol{\pi}} \{j\}} \left| \left( \mathbf{A}^{k+1} \left( \mathbf{A}^{k+1} \right)^{\*} \right)\_{j.} (\bar{\mathbf{v}}\_{\boldsymbol{f}, \boldsymbol{\cdot}}) \right|\_{a}^{a} \right] \in \mathbb{C}^{n \times 1}, \quad f = \mathbf{1}, \ldots, n, \end{aligned}$$

## **4. Conclusions**

In this chapter, notions of the MPCEP and CEPMP inverses are extended to quaternion matrices, and the new right and left MPCEPMP inverses are introduced and their characterizations are explored. Their determinantal representations are obtained within the framework of the theory of noncommutative column-row determinants previously introduced by the author. Also, determinantal representations of these generalized inverses for complex matrices are derived by using regular determinants. The obtained determinantal representations give new direct methods of calculations of these generalized inverses.

## **Acknowledgements**

The author thanks the Erwin Schrödinger Institute for Mathematics and Physics (ESI) at the University of Vienna for the support given by the Special Research Fellowship Programme for Ukrainian Scientists.

## **Author details**

Ivan I. Kyrchei Pidstryhach Institute for Applied Problems of Mechanics and Mathematics, NAS of Ukraine, Lviv, Ukraine

\*Address all correspondence to: ivankyrchei26@gmail.com

© 2022 The Author(s). Licensee IntechOpen. This chapter is distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

## **References**

[1] Kyrchei II. Determinantal representations of the quaternion core inverse and its generalizations. Advances in Applied Clifford Algebras. 2019;**29**(5): 104. DOI: 10.1007/s00006-019-1024-6

[2] Prasad KM, Mohana KS. Core-EP inverse. Linear and Multilinear Algebra. 2014;**62**(6):792-802. DOI: 10.1080/ 03081087.2013.791690

[3] Gao Y, Chen J. Pseudo core inverses in rings with involution. Communications in Algebra. 2018;**46**(1):38-50. DOI: 10.1080/ 00927872.2016.1260729

[4] Baksalary OM, Trenkler G. Core inverse of matrices. Linear and Multilinear Algebra. 2010;**58**(6):681-697. DOI: 10.1080/03081080902778222

[5] Ma H, Stanimirović PS. Characterizations, approximation and perturbations of the core-EP inverse. Applied Mathematics and Computation. 2019;**359**:404-417. DOI: 10.1080/ 03081087.2013.791690

[6] Wang H. Core-EP decomposition and its applications. Linear Algebra and its Applications. 2016;**508**:289-300. DOI: 10.1016/j.laa.2016.08.008

[7] Zhou MM, Chen JL, Li TT, Wan DG. Three limit representations of the core-EP inverse. Univerzitet u Nišu. 2018;**32**: 5887-5894. DOI: 10.2298/FIL1817887Z

[8] Gao Y, Chen J, Patricio P. Continuity of the core-EP inverse and its applications. Linear and Multilinear Algebra. 2021;**69**(3):557-571. DOI: 10.1080/03081087.2019.1608899

[9] Prasad KM, Raj MD. Bordering method to compute core-EP inverse. Special Matrices. 2018;**6**:193-200. DOI: 10.1515/spma-2018-0016

[10] Prasad KM, Raj MD, Vinay M. Iterative method to find core-EP inverse. Bulletin of Kerala Mathematics Association. 2018;**16**(1):139-152

[11] Kyrchei II. Determinantal representations of the core inverse and its generalizations with applications. Journal of Mathematics. 2019;**1631979**:13. DOI: 10.1155/2019/ 1631979

[12] Ferreyra DE, Levis FE, Thome N. Revisiting the core EP inverse and its extension to rectangular matrices. Quaestiones Mathematicae. 2018;**41**(2): 265-281. DOI: 10.2989/16073606.2017. 1377779

[13] Mosić D, Djordjević DS. The gDMP inverse of Hilbert space operators. Journal of Spectral Theory. 2018;**8**(2): 555-573. DOI: 10.4171/JST/207

[14] Mosić D. Core-EP inverses in Banach algebras. Linear and Multilinear Algebra. 2021;**69**(16):2976-2989. DOI: 10.1080/ 03081087.2019.1701976

[15] Sahoo JK, Behera R, Stanimirović PS, Katsikis VN, Ma H. Core and core-EP inverses of tensors. Computational and Applied Mathematics. 2020;**39**:9. DOI: 10.1007/s40314-019-0983-5

[16] Chen JL, Mosić D, Xu SZ. On a new generalized inverse for Hilbert space operators. Quaestiones Mathematicae. 2020;**43**(9):1331-1348. DOI: 10.2989/ 16073606.2019.1619104

[17] Udwadia F, Schttle A. An alternative derivation of the quaternion equations of motion for rigid-body rotational dynamics. Journal of Applied Mechanics. 2010;**77**:044505.1-044505.4. DOI: 10.1115/1.4000917

[18] Gibbon JD. A quaternionic structure in the three-dimensional Euler and ideal magneto-hydrodynamics equation. Physica D: Nonlinear Phenomena. 2002; **166**:17-28. DOI: 10.1016/S0167-2789(02) 00434-7

[19] Gibbon JD, Holm DD, Kerr RM, Roulstone I. Quaternions and particle dynamics in the Euler fluid equations. Nonlinearity. 2006;**19**:1969-1983. DOI: 10.1088/0951-7715/19/8/011

[20] Adler SL. Quaternionic Quantum Mechanics and Quantum Fields. New York: Oxford University Press; 1995

[21] Jiang T, Chen L. Algebraic algorithms for least squares problem in quaternionic quantum theory. Computer Physics Communications. 2007;**176**: 481-485. DOI: 10.1016/j.cpc.2006.12.005

[22] Leo SD, Ducati G. Delay time in quaternionic quantum mechanics. Journal of Mathematical Physics. 2012;**53**:022102.8. DOI: 10.1063/ 1.3684747

[23] Took CC, Mandic DP. A quaternion widely linear adaptive filter. IEEE Transactions on Signal Processing. 2010; **58**:4427-4431. DOI: 10.1109/ TSP.2010.2048323

[24] Took CC, Mandic DP. Augmented second-order statistics of quaternion random signals. Signal Processing. 2011; **91**:214-224. DOI: 10.1016/j. sigpro.2010.06.024

[25] Le Bihan N, Sangwine SJ. Quaternion principal component analysis of color images. Proceedings ICIP. 2003;I-809. DOI: 10.1109/ICIP.2003.1247085

[26] Jia Z, Ng MK, Song GJ. Robust quaternion matrix completion with applications to image inpainting. Numerical Linear Algebra with

Applications. 2019;**26**(4):e2245. DOI: 10.1002/nla.2245

[27] Jia Z, Ng MK, Song GJ. Lanczos method for large-scale quaternion singular value decomposition. Numerical Algorithms. 2019;**82**:699-717. DOI: 10.1007/s11075-018-0621-0

[28] Kyrchei II. Determinantal representations of the Moore–Penrose inverse over the quaternion skew field and corresponding Cramer's rules. Linear and Multilinear Algebra. 2011;**59**: 413-431. DOI: 10.1080/030810810035 86860

[29] Kyrchei II. Determinantal representations of the Drazin inverse over the quaternion skew field with applications to some matrix equations. Applied Mathematics and Computation. 2014;**238**:193-207. DOI: 10.1016/j. amc.2014.03.125

[30] Kyrchei II. Determinantal representations of the Drazin and Wweighted Drazin inverses over the quaternion skew field with applications. In: Griffin S, editor. Quaternions: Theory and Applications. New York: Nova Science Publishers; 2017. pp. 201-275

[31] Kyrchei II. Determinantal representations of the quaternion weighted Moore–Penrose inverse and its applications. In: Baswell AR, editor. Advances in Mathematics Research: Vol. 23. New York: Nova Science Publishers; 2017. pp. 35-96

[32] Kyrchei II. Determinantal representations of the weighted core-EP, DMP, MPD, and CMP inverses. J. Math. 2020;**9816038**: 12 p. DOI: 10.1155/2020/ 9816038

[33] Kyrchei II. Weighted quaternion core-EP, DMP, MPD, and CMP inverses and their determinantal representations. *Quaternion MPCEP, CEPMP, and MPCEPMP Generalized Inverses DOI: http://dx.doi.org/10.5772/intechopen.103087*

Revista de La Real Academia Ciencias Exactas, Físicas y Naturales. Serie A. Matemáticas RACSAM. 2020;**114**:198. DOI: 10.1007/s13398-020-00930-3

[34] Aslaksen H. Quaternionic determinants. Mathematical Intelligence. 1996;**18**(3):57-65. DOI: 10.1007/ BF03024312

[35] Cohen N, De Leo S. The quaternionic determinant. Electronic Journal of Linear Algebra. 2000;**7**:100-111. DOI: 10.13001/ 1081-3810.1050

[36] Zhang FZ. Quaternions and matrices of quaternions. Linear Algebra and its Applications. 1997;**251**:21-57. DOI: 10.1016/0024-3795(95)00543-9

[37] Kyrchei II. Cramer's rule for quaternionic systems of linear equations. Journal of Mathematical Sciences. 2008; **155**(6):839-858. DOI: 10.1007/ s10958-008-9245-6

[38] Kyrchei II. The theory of the column and row determinants in a quaternion linear algebra. In: Baswell AR, editor. Advances in Mathematics Research: Vol. 15. New York: Nova Science Publishers; 2012. pp. 301-359

Section 2 Applications

## **Chapter 7**
