Professional Documents
Culture Documents
QUATERNIONS
THEORY AND APPLICATIONS
No part of this digital document may be reproduced, stored in a retrieval system or transmitted in any form or
by any means. The publisher has taken reasonable care in the preparation of this digital document, but makes no
expressed or implied warranty of any kind and assumes no responsibility for any errors or omissions. No
liability is assumed for incidental or consequential damages in connection with or arising out of information
contained herein. This digital document is sold with the clear understanding that the publisher is not engaged in
rendering legal, medical or any other professional services.
MATHEMATICS RESEARCH DEVELOPMENTS
QUATERNIONS
THEORY AND APPLICATIONS
SANDRA GRIFFIN
EDITOR
New York
Copyright 2017 by Nova Science Publishers, Inc.
All rights reserved. No part of this book may be reproduced, stored in a retrieval system or transmitted in any
form or by any means: electronic, electrostatic, magnetic, tape, mechanical photocopying, recording or
otherwise without the written permission of the Publisher.
We have partnered with Copyright Clearance Center to make it easy for you to obtain permissions to reuse
content from this publication. Simply navigate to this publications page on Novas website and locate the
Get Permission button below the title description. This button is linked directly to the titles permission
page on copyright.com. Alternatively, you can visit copyright.com and search by title, ISBN, or ISSN.
For further questions about using the service on copyright.com, please contact:
Copyright Clearance Center
Phone: +1-(978) 750-8400 Fax: +1-(978) 750-4470 E-mail: info@copyright.com.
Independent verification should be sought for any data, advice or recommendations contained in this book. In
addition, no responsibility is assumed by the publisher for any injury and/or damage to persons or property
arising from any methods, products, instructions, ideas or otherwise contained in this publication.
This publication is designed to provide accurate and authoritative information with regard to the subject
matter covered herein. It is sold with the clear understanding that the Publisher is not engaged in rendering
legal or any other professional services. If legal or any other expert assistance is required, the services of a
competent person should be sought. FROM A DECLARATION OF PARTICIPANTS JOINTLY ADOPTED
BY A COMMITTEE OF THE AMERICAN BAR ASSOCIATION AND A COMMITTEE OF
PUBLISHERS.
Additional color graphics may be available in the e-book version of this book.
Index 277
PREFACE
functions. Three series of energy levels are found, of which two substantially
differ from those for spin 1 particles without anomalous magnetic moment.
For assigning to them physical sense for all the values of the main quantum
number n 0,1, 2, ... , one must impose special restrictions on a parameter
related to the anomalous moment. Otherwise, only some part of the energy
levels corresponds to bound states. The neutral spin 1 particle is considered as
well. In this case no bound states exist in the system, and the main qualitative
manifestation of the anomalous magnetic moment consists in the occurrence of
a space scaling of the arguments of the wave functions, compared to a particle
without such a moment. Traditionally, the automotive industry has been the
largest employer of robots, but their control is inline and programmed to
follow planning trajectories. As shown in Chapter 4, in this case, in the
department motors test of Volkswagen Mexico a semi-autonomous robot is
developed. To date, some critical technical problems must be solved in a
number of areas, including in dynamics control. Generally, the attitude
estimation and the measurement of the angular velocity are a requirements for
the attitude control. As a result, the computational cost and the complexity of
the control loop is relatively high.
Chapter 4 deals with the implementation of a cheap Micro AHRS
(Attitude and Heading Reference System) using low-cost inertial sensors. In
Chapter 4, the technique proposed is designed with attitude estimation and the
prediction movement via the kinematic of a 4GDL robot. With this approach,
only the measurements of at least two non-collinear directional sensors are
needed. Since the control laws are highly simple and a model-based observer
for angular velocity reconstruction is not needed, the proposed new strategy is
very suitable for embedded implementations. The global convergence of the
estimation and prediction techniques is proved. Simulation with some
robustness tests is performed.
Chapter 5 - The quaternions, reduced biquaternions (RBs) and their
respective Fourier transformations, i.e., discrete quaternion Fourier transform
(DQFT) and discrete reduced biquaternion Fourier transform (DRBQFT), are
very useful for multi-dimensional signal processing and analysis. In Chapter 5,
the basic concepts of quaternion and RB algebra are reviewed, and the author
introduce the two dimensional Hermite-Gaussian functions (2D-HGF) as the
eigenfunction of DQFT/DRBQFT, and the eigenvalues of 2D-HGF for three
types of DQFT and two types of DRBQFT. After that, the relation between
2D-HGF and Gauss-Laguerre circular harmonic function (GLCHF) is given.
From the aforementioned relation and some derivations, the GLCHF can be
proved as the eigenfunction of DQFT/DRBQFT and its eigenvalues are
x Sandra Griffin
summarized. These GLCHFs can be used as the basis to perform color image
expansion. The expansion coefficients can be used to reconstruct the original
color image and as a rotation invariant feature. The GLCHFs can also be
applied to color matching applications.
Chapter 6 - The unit quaternion system was invented in 1843 by William
Rowan Hamilton as an extension to the complex number to find an answer to
the question (how to multiply triplets?). Yet, quaternions are extensively used
to represent the attitude of a rigid body such as quadrotors, which is able to
alleviate the singularity problem caused by the Euler angles representation.
The singularity is in general a point at which a given mathematical object is
not defined and it outcome of the so called gimbal lock. The singularity is
occur when the pitch angles rotation is 90 . In Chapter 6, a leader-
follower formation control problem of quadrotors is investigated. The
quadrotor dynamic model is represented by unit quaternion with the
consideration of external disturbance. Three different control techniques are
proposed for both the leader and the follower robots. First, a nonlinear H
design approach is derived by solving a Hamilton-Jacobi inequality following
from a result for general nonlinear affine systems. Second, integral
backstepping (IBS) controllers are also addressed for the leader-follower
formation control problem. Then, an iterative Linear Quadratic Regulator
(iLQR) is derived to solve the problem of leader-follower formation. The
simulation results from all types of controllers are compared and robustness
performance of the H controllers, fast convergence and small tracking errors
of iLQR controllers over the IBS controllers are demonstrated.
Chapter 7 - A generalized inverse of a given quaternion matrix (similarly,
as for complex matrices) exists for a larger class of matrices than the invertible
matrices. It has some of the properties of the usual inverse, and agrees with the
inverse when a given matrix happens to be invertible. There exist many
different generalized inverses. The authors consider determinantal
representations of the Drazin and weighted Drazin inverses over the
quaternion skew field. Due to the theory of column-row determinants recently
introduced by the author, the authors derive determinantal representations of
the Drazin inverse for both Hermitian and arbitrary matrices over the
quaternion skew field. Using obtained determinantal representations of the
Drazin inverse we get explicit representation formulas (analogs of Cramer's
rule) for the Drazin inverse solutions of the quaternionic matrix equations
AXB = D and, consequently, AX = D , XB = D in both cases when A and
B are Hermitian and arbitrary, where A , B can be noninvertible matrices of
Preface xi
Chapter 1
Jingjing Ma
Department of Mathematics and Statistics,
University of Houston Clear Lake, Houston, TX, US
Abstract
This paper collects some old problems on lattice orders and directed
partial orders on complex numbers and quaternions, and summarizes re-
cent development in answering those questions.
1. Introduction
We will introduce some definitions and terminologies in this section. The reader
is referred to [2, 3, 5] for more information on partially ordered rings and lattice-
ordered rings (-rings).
Let R be a partially ordered ring. The positive cone of R is defined as
R+ = {r R | r 0}. The positive cone R+ is closed under the addition
E-mail address: ma@uhcl.edu.
2 Jingjing Ma
CT = {a + bi | a, b T, i2 = 1},
HT = {a0 + a1 i + a2 j + a3 k | a0 , a1 , a2 , a3 T, i2 = j 2 = k 2 = 1}.
(a0 + a1 i + a2 j + a3 k)(b0 + b1 i + b2 j + b3 k) =
(a0 b0 a1 b1 a2 b2 a3 b3 ) + (a0 b1 + a1 b0 + a2 b3 a3 b2 )i +
(a0 b2 + a2 b0 + a3 b1 a1 b3 )j + (a0 b3 + a3 b0 + a1 b2 a2 b1 )k.
The following questions have been left unanswered for some time now, how-
ever they have greatly motivated research activities in the area.
Problem 1. (G. Birkhoff and R. Pierce, 1956) Can the field of complex
numbers be made into a lattice-ordered ring?
Problem 2. (L. Fuchs, 1963) Describe the directed orders of the fields of
complex numbers and quaternions.
Directed Partial Orders on Quaternions - A Brief Summary 3
Problem 3. (G. Birkhoff, 1967) In how many ways can the quaternions
be made into an -ring? an -algebra? a directed algebra?
We will summarize below recent developments in finding lattice orders and
directed partial orders on quaternion algebras. Since this activity is closely re-
lated to and motivated by the same research for complex numbers, results for
complex numbers are also included.
In [14], Y. Yang showed that for some totally ordered field Q, CQ admits
directed partial orders to make it into a directed algebra with 1 > 0, and hence
i is an element with negative square, that is, i2 = 1 < 0. Then in [13], N.
Schwartz and Y. Yang proved that C can be made into a directed algebra over R,
and in [12], W. Rump and Y. Yang constructed directed partial orders on K(i),
where K could be any non-archimedean totally ordered field and i2 = 1.
Their method has used multiplicative segment that is a convex additive subgroup
of F containing identity element 1.
Motivated by the above work, in [9], L. Wu, Y. Zhang and the present author
have introduced a more general method to produce directed partial orders on
CF . Take an additive semigroup S F + with 0, 1 S, and take x, y F +
with 0 < x, y 1. Define the positive cone Px,y (S) as follows.
Then (CF , Px,y ) is a partially ordered algebra over F and it is a directed algebra
if there exists z F + such that s z for all s S [9, Theorem 2.2].
For a non-archimedean totally ordered field F , take S = Z+ and x = y = 1,
then P1,1 (Z+ ) is a directed partial order on CF that makes CF into a directed
algebra over F . We also observe that P1,1 (Z+ ) is the largest directed partial
order on CF over a non-archimedean totally ordered field. Therefore P1,1 (Z+ )
is division closed in the sense that for any a, b CF , if a, ab P1,1 (Z+ ), then
b P1,1 (Z+ ).
We notice that the partial orders defined above have positive identity ele-
ment, that is, 1 Px,y (S). This begs the question whether we can construct
directed partial orders on CF such that 1 is not positive?
Let S be an additive semigroup of F + containing 0, 1. Suppose that there
exists w F + such that s w for all s S. Define
Theorem 4. P (S)> and P (S)< are directed partial order on CF with 1 6> 0.
Directed Partial Orders on Quaternions - A Brief Summary 5
Proof. Lets just consider P (S)> . We leave it to the reader to check that P (S)>
is a partial order on CF . Take w F + such that s w for all s S. For any
a + bi CF ,
and
P (S)> + F + = {a + bi P1,1 | b 0}.
Then P is a directed partial order on H that makes it into a directed algebra over
R with R P = R+ [6, Theorem 1].
In [9], the authors proved a more general method to produce directed partial
orders on HF over a non-archimedean totally ordered field F . Take an additive
semigroup S F + with 0, 1 S, and take x F with 0 < x 1. Define the
positive cone Px (S) as follows.
Px (S) = {a0 + a1 i + a2 j + a3 k HF | a0 0, |a1 | S xa0 , |a2 | S xa0 , |a3 | S xa0 },
where |a1 | S xa0 means xa0 sa1 xa0 for all s S. Similarly for
|a2 | S xa0 and |a3 | S xa0 . Then Px is a partial order on HF to make it into
a partially ordered algebra over F , and if there exists an element z F + such
that s z for all s S, then Px is a directed partial order and HF is a directed
algebra [9, Theorem 3.2].
For instance, for a non-archimedean totally ordered field F , take S = Z+
and x = 1, then Px is the positive cone P introduced in the previous paragraph,
and P1 (Z+ ) is the largest directed partial order on HF .
The directed partial orders constructed for complex numbers and quater-
nions over non-archimedean totally ordered fields have been generalized to
complex numbers and quaternions over non-archimedean partially ordered
fields that contain a totally ordered subfield [7, Theorems 1 and 2].
Acknowledgment
The author thanks Professor Warren McGovern, Florida Atlantic University, for
kindly reading the manuscript and providing helpful feedback.
8 Jingjing Ma
References
[1] G. Birkhoff, Lattice Theory, vol. 25, 3rd ed. Colloquium Publications
AMS, New York (1967).
[4] J. Ma, Finite dimensional simple algebras that do not admit a lattice order,
Comm. Alg. 32 (2004) 1615-1617.
[7] J. Ma, Directed partial orders on complex number and real quaternions II,
Positivity (2015), DOI: 10.1007/s11117-015-0388-7.
[9] J. Ma, L. Wu, Y. Zhang, Directed partial orders on complex numbers and
quaternions over non-archimedean linearly ordered fields, Order (First
Online: 05 March 2016, doi:10.1007/s11083-016-9387-y).
[10] J. Ma, L. Wu, Y. Zhang, Describing directed partial orders on F (i) with
1 > 0, (preprint).
[13] N. Schwartz, Y. Yang, Fields with directed partial orders, J. Algebra 336
(2011) 342-348.
[14] Y. Yang, On the existence of directed rings and algebras with negative
squares, J. Algebra 295 (2006) 452-457.
In: Quaternions: Theory and Applications ISBN: 978-1-53610-768-5
Editor: Sandra Griffin
c 2017 Nova Science Publishers, Inc.
Chapter 2
Abstract
Within the matrix 10-dimensional Duffin-Kemmer-Petiau formalism
applied to the Shamaly-Capri field, we study the behavior of a vector
particle with anomalous magnetic moment in the presence of an external
uniform electric field. The separation of variables in the wave equation
is performed by using projective operator techniques and the theory of
DKP-algebras. The whole wave function is decomposed into the sum of
three components 0 , + , + . It is enough to solve the equation for the
E-mail address: e.ovsiyuk@mail.ru.
E-mail address: voinyuschka@mail.ru.
E-mail address: vasiliy-bspu@mail.ru.
E-mail address: vladimir.balan@upb.ro.
E-mail address: redkov@dragon.bas-net.by.
12 E. M. Ovsiyuk, Ya. A. Voynova, V. V. Kisel et al.
1. Intoduction
Commonly, we shall use only the simplest wave equations for fundamental par-
ticles of spin 0, 1/2, 1. Meanwhile, it is known that other more complicated
equations can be assigned to particles with such spins, which are based on
the application of extended sets of Lorentz group representations (see [1]-[16]).
Such generalized wave equations allow to describe more complicated objects,
which have besides mass, spin, and electric charge other electromagnetic
characteristics, like polarizability or anomalous magnetic moment. These addi-
tional characteristics manifest themselves explicitly in the presence of external
electromagnetic fields.
In particular, within this approach, Petras [3] proposed a 20-component the-
ory for spin 1/2 particle, which after excluding 16 subsidiary components
turns to be equivalent to the Dirac particle theory modified by the presence of
Pauli interaction term. In other words, this theory describes a spin 1/2 particle
with anomalous magnetic moment.
A similar equation was proposed by ShamalyCapri [6, 7] for spin 1 par-
ticles (also see [16, 17]). In the following, we investigate and solve this wave
equation in the presence of the external uniform electric field.
The wave equation for spin 1 particle with anomalous magnetic moment
Spin 1 Particle with Anomalous Magnetic Moment ... 13
0 = P0 , + = P+ , = P , = 0 + + + .
Acting on (2) by the operator P0 , and taking into account the algebraic identities
we get
(1 1 + 2 2 + M ) 0 + P0 3 3 + (4 eEx3 ) P0 4 = 0 . (3)
Let us consider the operator P0 3 (we shall further use the computation
rules within the DKP-algebra):
P0 3 = (1 Y 2 )3 = (1 + 23 3 4 4 3 3 4 4 )3 =
Spin 1 Particle with Anomalous Magnetic Moment ... 15
= 3 + 23 3 4 4 3 3 4 4 3 =
= 23 3 (3 3 4 4 ) (3 3 4 4 ) = 3 3 4 4 .
Considering the identities
3 (1 P0 ) = 3 Y 2 = 3 [3 3 + 4 4 23 3 4 4 ] =
= 3 + 3 4 4 23 4 4 = 3 3 4 4 ,
the previous can be written in the form
P0 3 = 3 (1 P0 ) = 3 (P+ + P ) . (4)
P0 4 = 4 (1 P0 ) = 4 (P+ + P ) . (5)
Taking into account the relations (4)(5), (3) reduces to the form
(1 1 + 2 2 + M ) 0 +
+ [3 3 + 4 (4 eEx3 )] + + [3 3 + 4 (4 eEx3 )] = 0 . (6)
1
3 P+ = (3 + i3 3 4 3 4 4 ) .
2
As well, for 43 = 4 and 4 3 4 = 0, we infer
1
4 P+ = 4 [i(3 4 4 3 ) 23 3 4 4 + 3 3 + 4 4 ] =
2
1
= [i4 4 3 24 3 3 4 4 + 4 3 3 + 4 ] .
2
16 E. M. Ovsiyuk, Ya. A. Voynova, V. V. Kisel et al.
4 4 3 = 3 3 4 4 , 4 3 3 = 4 3 3 4 ,
we get
1
4 P+ = [i(3 3 4 4 ) 2(4 3 3 4 )4 4 + (4 3 3 4 ) + 4 ] =
2
i
= [3 3 4 4 + i3 3 4 ] .
2
Hence, we obtain the algebraic relation
3 P+ = i4 P+ = (3 i4 )P+ = 0 . (7)
(1 1 + 2 2 + M ) 0 +
1 P0 = Y 2 = 3 3 + 4 4 23 3 4 4 ,
we get
(1 P0 )3 = 3 + (3 3 4 4 ) 23 3 (3 3 4 4 ) = +3 4 4 .
Similarly, we derive
3 P0 = 3 (1 Y 2 ) = 3 (1 3 3 4 4 + 23 3 4 4 ) = +3 4 4 .
(1 P0 )3 = 3 P0 .
4 3 3 4 = (1 P0 )4 , 4 3 3 4 = 4 P0 , (1 P0 )4 = 4 P0 ,
By acting on (11) by the operator 12 (1 + Y ) and with the help of the easy to
check identities
1 1 1 1
(1 + Y )P+ = (1 + Y ) Y (1 + Y ) = Y (1 + Y ) = P+ ,
2 2 2 2
1 1
(1 + Y )P = (Y + Y 2 )(Y 1) = 0 ,
2 4
we derive
1 1 + 2 2 i0 P Y + M + +
x x
1 1
+ (1 + Y )3 3 0 + (1 + Y )4 ( 4 eEx3 )0 = 0 , (12)
2 x 2 x
We need three auxiliary relations. From the known formula
J[] J[] =
it follows
3 Y Y 3 = +i4 = Y 3 = 3 Y i4 ,
4 Y Y 4 = i3 = Y 4 = 4 Y + i4 .
Therefore, (12) can be written as
1 x 1 + 2 x2 i0 P Y + M + +
+ 21 (3 + 3 Y i4 ) x 1
3 0 + 2 (4 + 4 Y + i4 )( x4 eEx3 )0 = 0 ,
From this, taking into account Y P0 0, we obtain the more simple form
1 x 1 + 2 x2 i0 P Y + M + +
+ 21 (3 i4 ) x
1
3 0 + 2 (4 + i3 )( x4 eEx3 )0 = 0 ,
or
1
1 1 + 2 2 i0 P Y + M + + (3 i4 ) 0 + i eEx 3 0 = 0.
x x 2 x3 x4
(1 1 + 2 2 + M )0 +
+ 12 + [(3 + ieEx3 ) i4 ]+ + 1 [(3
2
ieEx3 ) + i4 ] = 0 ,
(1 1 + 2 2 i0 P + M )+ + 1 [(3 ieEx3 ) + i4 ]0 = 0 ,
2
(1 1 + 2 2 + i0 P + M ) + 1 + [(3 + ieEx3 ) i4 ]0 = 0 ,
2
where
1 1
+ = (3 + i4 ), = (3 i4 ) .
2 2
We look for solutions of the form:
0 = eip4 x4 eip1 x1 eip2 x2 f0 (x3 ) ,
+ = eip4 x4 eip1 x1 eip2 x2 f+ (x3 ) ,
= eip4 x4 eip1 x1 eip2 x2 f (x3 ) .
(ip1 1 + ip2 2 + M )0 +
+ 12 + [( dxd 3 + ieEx3 ) + p4 ]+ + 1 [( d
2 dx3 ieEx3) p4 ] = 0 ,
i0 = , p1 1 + p2 2 = p ;
(i + b = 0 ,
p + M )0 + + a (13)
p P + M )+ b0 = 0 ,
(i (14)
(i
p + P + M ) + + a0 = 0 . (15)
M P
,
M
we infer
M P M P M P
i
p+ (M P ) + b0 = 0 .
M M M
With the help of the identities
M P 1
(M P ) = (M 2 M P M P ) = M ,
M M
it reads
M P M P
i
p+M + b0 = 0 .
M M
By using the notations
M P M P 0
i
p = A, = ,
M M
the previous equation shortens to
0
(A + M )+ b0 = 0 .
M + P
,
M +
Spin 1 Particle with Anomalous Magnetic Moment ... 21
we get
M + P M + P M + P
i
p+ (M + P ) + + a
0 = 0 .
M + M M +
M + P 1
(M + P ) = (M 2 + M P + M P ) = M ;
M + M +
we derive
M + P M + P
i
p + M + + a0 = 0 .
M + M +
= P + P = P + P , P = P , P = P ,
P P = P P = 0, P = P , P = P ,
P + P = 1, P P = P P = 0,
we get
1 M p2
A2 = (M 2 2
p
+ M
p 2
) = .
(M )2 M
22 E. M. Ovsiyuk, Ya. A. Voynova, V. V. Kisel et al.
We calculate A3 :
M )(i M p2 (M P )
A3 = (M P p)
p 2
= (i
p) ,
(M )2 (M ) M
so, the minimal polynomial of A (or the Cayley-Hamilton identity for A) has
the form
M p2
A3 + A = 0.
M
Similar results are valid for the operator C:
M p2
C3 + C = 0.
M +
The Cayley-Hamilton identity for i
p) has the form
i p)2 + p2 ] = 0 .
p [(i
p + M ) f0 + + af+ bf = 0 ,
(i
0
(A + M )f+ bf0 = 0 ,
0 a
(C + M )f + + f0 = 0 .
To proceed with these equations, we introduce the matrices5 with the properties
(i p + M ) = p2 + M 2 ,
p + M )(i
(A + M )(A + M ) = p2 + M 2 ,
(C + M )(C + M ) = p2 + M 2 . (16)
The first equation in (17), with the help of the other two ones, transforms
into an equation in the component f0 (r):
0
p + M )(p2 + M 2 )2 f0 + + a(A + M )
(i bf0 + b(C + M )+
0
a
m f0 = 0 , (18)
In fact, the equations (19) mean that it suffices to solve (18) with respect to f0 ; the two
other components f+ and f can be calculated by means of the equations (19).
To proceed further, we need to know the explicit form of the inverse operators (16).
To solve this task, we first establish the minimal polynomials for the relevant matrices.
Therefore, the needed inverse operators must be quadratic with respect to the rele-
vant matrices. They are given by the formulas:
1
p)2 M (i p) + (p2 + M 2 ) ,
(M + ip) = M (i
2 2
h i
(A + M ) = p +M M 1 2
M
2
p +M M A + 2
M
2
M (p +M M ) A2
,
2 2
h i
(C + M ) = p +M M
M +
1 p2 +M M +
2 +M C + M (p2 +M 2 +M ) C
2
.
Taking into account the explicit form of the inverse operators, we get
(p2 + M 2 )f0 +
1
p)2 M (i
p) + (p2 + M 2 ) +
2
(i
+
M
M M
1 2 A + A 2
0
bf0 +
a
p + M2 M M (p2 + M 2 M )
1
p)2 M (i
p) + (p2 + M 2 )
+ 2 (i
M
M + M + 2 0
1 2 C+ C + baf0 = 0 .
p + M2 + M M (p2 + M 2 + M )
Now, by considering the formulas
M P M p2
A= i
p, A2 = ,
M M
M + P M p2
C= i
p, C2 = ,
M + M +
24 E. M. Ovsiyuk, Ya. A. Voynova, V. V. Kisel et al.
M P 0 0 M + P
= , + = + ,
M M +
we transform the above equation into
(p2 + M 2 )f0 +
1
p)2 M (i
p) + (p2 + M 2 ) +
(i +
M2
M P p)2 M P
(i
1 2 p+ 2
i abf0 +
p + M2 M p + M2 M M
1
p)2 M (i
p) + (p2 + M 2 )
+ 2 (i
M
M + P p)2 M + P
(i
1 2 i
p + + b
af0 = 0 .
p + M2 + M p2 + M 2 + M M +
After some manipulation, this becomes
1 1
{ (p2 + M 2) + a bm
M 2 (p2 + M 2 M ) M
p) + (p2 + M 2 ) + (p2 + M 2 M ) (M P ) i
p)2 M (i p)2 (M P ) +
(i p + (i
1 1
+
b
a
M 2 (p2 + M 2 + M ) M +
p) + (p2 + M 2 ) (p2 + M 2 + M ) (M P ) i
p)2 M (i p)2
(i p + (i
(M + P )+ } f0 = 0
or
1 1
b
{ (p2 + M 2 ) + a
M 2 (p2 2
+ M M ) M
p)2 M (i
p) + (p2 + M 2 ) + (p2 + M 2 M ) i p)2 (M P ) +
(i p(M P ) + (i
1 1
+ba
M 2 (p2 + M 2 + M ) M +
h i h i
2 M (ip)
(ip) + (p2 + M 2 ) (p2 + M 2 + M ) ip(M
2 (M + P
+ P ) + (ip) )+ } f0 = 0 .
+
b
a 1 1 [ (p2 + M 2 + M )(ip)
2 M (p2 + M 2 + M )ip
+
M 2 (p2 +M 2 +M ) M +
+(p2 + M 2 ) (ip)
2 ] (M + P
)+ }f0 = 0
Spin 1 Particle with Anomalous Magnetic Moment ... 25
Then we obtain
0 f1 1 8 0 (M )f 1
1
f >
B 02 C B (M )f 2C
>
>
0
>
>
B 0 C
> B C
B 0 C a b < 0
(p2 +M 2 ) B 2 2 2 2
B C
0
B f 0 C+ M 2 (M )(p2 + M 2 M ) (p + M )(p + M M )B
C C
B 0 C
B [12] C 0
>
> B C
0
@ 0 A >
> @ A
>
0 >
: 0
f[34] M f[34]
0 0
1 0 p2 (M )(p2 f1 p1 f2 ) 1
0 p1 (M )(p2 f1 p1 f2 )
0
B
0
C B 0 C
0
B C B C
2 2 0 2 2
iM (p + M M ) B C (p + M M ) B 0 C+
B C B C
B (M )(p20f1 p1 f2 ) C B 0 C
0
0
@ A @ A
0
0 0
0 0
26 E. M. Ovsiyuk, Ya. A. Voynova, V. V. Kisel et al.
0 p1 f[34] 1 0p
1 (M )(p1 f1 +p2 f2 )
9
1
p2 (M )(p1 f1 +p2 f2 ) >
>
p1 f[34]
B C B 0 C>>
B 0 C B 0
C >
>
0 C=
+(p2 + M 2 )M (M (p2 + 2 B 0
B C B
) B
B 0 C M )B 0 C> +
C
0 C
0
B 0 C B
0
C>
A>
@ 0 A @
0
>
>
0
>
M p2 f[34]
;
(p1 f1 +p2 f2 )
8 0 (M +)f 1
> 1
B (M +)f 2C
>
>
0
>
>
b
a
>
< B 0
C
(p2 + M 2 )(p2 + M 2 + M ) B
B C
+ 2 0 C
2 2 0
M (M + )(p + M + M ) >
>
B
B 0
C
C
0
>
> @ A
>
>
: 0
M f[34]
0 0
1 0 p2 (M +)(p2 f1 p1 f2 ) 1
0 p (M +)(p2 f1 p1 f2 )
B 0 C B 1 0 C
B 0 C B 0
C
0
iM (p2 + M 2 + M ) B (p2 + M 2 + M ) B 0 C+
B C B C
0 C
0
B (M +)(p2 f1 p1 f2 ) C B C
0
0
@ A @ A
0
0 0
0 0
0 p1 f[34] 1 0p
1 (M +)(p1 f1 +p2 f2 )
1 9
p2 (M +)(p1 f1 +p2 f2 ) >
>
p1 f[34]
B C B 0 C>>
B 0 C B 0
C >
>
0 C=
+(p2 + M 2 )M (M + (p2 + 2 B 0
B C B
) B
B 0 C M )B 0
C = 0.
0 C
0
C>
B 0 C B
0
C>
A>
@ 0 A @
0
>
>
0
>
2
;
(p1 f1 +p2 f2 ) M p f[34]
b
a
(p2 + M 2 )f1 +
M 2 (p2 + M2 M
b
a
(p2 + M 2 )f2 +
M 2 (p2 + M2 M
{(p2 + M 2 )(p2 + M 2 M )f2 + p1 (p2 + M 2 M )(p2 f1 p1 f2 )+
b
a
+M (p2 + M 2 )p2 f[34] p2 (p2 + M 2 )(p1 f1 + p2 f2 )} +
M 2 (p2 + M2 + M
Spin 1 Particle with Anomalous Magnetic Moment ... 27
b
a b
a
(p2 + M 2 )f[12] + {i(p2 f1 p1 f2 )} + {i(p2 f1 p1 f2 )} = 0 , (22)
M M
a b
f[34] + {(p2 + M 2 M )f[34] (M )(p1f1 + p2 f2 )) p2f[34] }+
M (M )(p2 + M 2 M )
b
a
+ {(p2 + M 2 + M )f[34] + (M + )(p1 f1 + p2 f2 )) p2 f[34] } = 0 .
M (M + )(p2 + M 2 + M )
(23)
ab
[EQ.I] (p2 + M 2 )f1 + M 2(p2 +M 2 M
a b
[EQ.II] (p2 + M 2 )f2 + M 2 (p2 +M 2 M
28 E. M. Ovsiyuk, Ya. A. Voynova, V. V. Kisel et al.
{(p2 + M 2 )(p2 + M 2 M )f2 + p1 (p2 + M 2 M )(p2f1 p1 f2 )+
+M (p2 + M 2)p2 f[34] p2 (p2 + M 2 )(p1 f1 + p2 f2 )}+
b
a 2 2 2 2 2 2
+ M 2 (p2 +M 2 +M {(p + M )(p + M + M )f2 + p1 (p + M + M )(p2 f1 p1 f2 )
we derive
b
a
(p1 f1 + p2 f2 ) +
M 2 (p2 + M 2 M )
{(p2 + M 2 M )(p1 f1 + p2 f2 ) + M p2 f[34] p2 (p1 f1 + p2 f2 )}+
ba
+
M 2 (p2 + M 2 + M )
{(p2 + M 2 + M )(p1 f1 + p2 f2 ) M p2 f[34] p2 (p1 f1 + p2 f2 )} = 0 ,
(p2 + M 2 )(p2 f1 p1 f2 )+
b
a
+ {(p2 + M 2 )(p2 f1 p1 f2 ) p2 (p2 f1 p1 f2 )}+
M2
ba
+ 2 {(p2 + M 2 )(p2 f1 p1 f2 ) p2 (p2 f1 p1 f2 )} = 0 .
M
After elementary manipulations, they read
(p1 f1 + p2 f2 ) +
b
a
+ {(M )(p1 f1 + p2 f2 ) + p2 f[34]} +
M (p2 + M 2 M )
ba
+ {(M + )(p1 f1 + p2 f2 ) p2 f[34]} = 0 , (24)
M (p2 + M 2 + M )
ab + b
(p2 + M 2 )(p2 f1 p1 f2 ) + ( a)(p2 f1 p1 f2 ) = 0 . (25)
Let us write down here the remaining two equations (see (22)(23)) as well:
i
(p2 + M 2 )f[12] (
ab + b
a)(p2 f1 p1 f2 ) = 0 . (26)
M
Spin 1 Particle with Anomalous Magnetic Moment ... 29
b
a
f[34] + {M f[34] (p1 f1 + p2 f2 )} +
M (p2
+ M 2 M )
b
a
+ {M f[34] + (p1 f1 + p2 f2 )} = 0 (27)
M (p2 + M 2 + M )
ab + b
[( a) + (p2 + M 2 )](p2 f1 p1 f2 ) = 0 ,
1
f[12] = iM (p2 f1 p1 f2 ) .
F = f[34] , G = p1 f1 + p2 f2 ,
b
a b
a
F+ (M F G) + (M F + G) = 0|,
M (p2 + M 2 M ) M (p2 + M 2 + M )
(28)
b
a
G+ [(M )G + p2 F ] +
M (p2
+ M 2 M )
b
a
+ [(M + )G p2 F ] = 0 . (29)
M (p + M 2 + M )
2
F+ 1
M (p2+M 2 M )(p2 +M 2 +M )
{(p2 + M 2 + M ) ab (M F G))+
+(p2 + M 2 M ) b
a (M F + G)} = 0 ,
{M (p2 + M 2 ) ab F + M 2 a
b F (p2 + M 2 ) ab G M a bG+
+M (p2 + M 2 ) b
a M F M 2 b
a F + (p2 + M 2 ) b
a G M ba G} = 0,
30 E. M. Ovsiyuk, Ya. A. Voynova, V. V. Kisel et al.
or,
[EQ.I] F+ 1
M [(p2 +M 2 )2 M 2 2 ]
ab + b
{M (p2 + M 2 ) ( ab b
a) F + M 2 ( a) F
a b + b
M ( ab b
a)G (p2 + M 2 ) ( a) G} = 0 .
or,
1
G+ M [(p2 +M 2 )2 M 2 2 )
b G + (M 2 M 2 ) a
{ (p2 + M 2 )(M ) a b G + p2 (p2 + M 2 ) a
b F + p2 M a
b F +
+(M + )(p2 + M 2 ) b
a G (M 2 + M 2 ) b
a G p2 (p2 + M 2 ) b
a F + p2 M b
aF} = 0,
so we infer
[EQ.II] G+ 1
M [(p2 +M 2 )2 M 2 2 )
a
{ M (p2 + M 2 )(b+ a
a) G (p2 + M 2 )(
b b
b
a)G
b +
M 2 ( a a
a) G + M 2 (
b b a
a)G + M p2 (
b b + b a
a) F + p2 (p2 + M 2 ) (b
b
a) F } = 0 ,
ab + b
[( a) + p2 + M 2 ]G = p2 ( ieE
M )F ,
ab + ba) + p2 + M 2 ]F = ieE ieE
[( M G+ M F.
Spin 1 Particle with Anomalous Magnetic Moment ... 31
This sub-system is solved by diagonalizing the mixing matrix. To this aim, let
us introduce the new functions
1 = G 1 F , 2 = G 2 F ,
where p p
+2 4p2 2 4p2
1 = , 2 = .
2 2
So we get two independent equations:
h i
ab + b
a) + p2 + M 2 ] 01 1 = 0 , 01 = 1 E
( M ;
0
h i
ab + b
a) + p2 + M 2 ] 02 2 = 0 , 02 = 2 E
( M .
0
For the second order operator, we have the explicit for, e.g., x3 = z:
d2
ab + b
( a) = 2 e2 E 2 z 2 2eEz 2 .
dz
Thus, we get the equation6:
2
d 2 2 2 2 2 2 2 0
+ (e E z + 2eEz + ) p M + 1,2 1,2 (z) = 0 ,
dz 2
where 2 = p2 M 2 + 01,2.
This task coincides with that which arises for the scalar KleinFockGordon
particle in external uniform electric field modified by an anomalous magnetic
moment.
6
We consider both variants.
32 E. M. Ovsiyuk, Ya. A. Voynova, V. V. Kisel et al.
2(M 2 + p2 )
q
> 0 20 + 4p2 ,
(y 0 )
which is valid without any additional restrictions.
We first address the variant 0 < 0, (+) upper sign:
2(M 2 + p2 )
q
> 0 + 20 + 4p2 .
(y 0 )
7
In our considerations, 0 is real-valued
8
For definiteness, we assume that eE > 0
Spin 1 Particle with Anomalous Magnetic Moment ... 33
This yields
2(M 2 + p2 )
q
0 > + 20 + 4p2 ;
(y 0 )
which, after squaring, takes the form
4(M 2 + p2 )2 2(M 2 + p2 )
20 4p2 > 0 ,
(y 0 )2 (y 0 )
or
4(M 2 + p2 )2 40 (M 2 + p2 )(y 0 ) 4p2 (y 0 )2 > 0 .
It is convenient to use the variable x:
y 0 = x > 0 ,
which leads to
(M 2 + p2 )2 (y x)x(M 2 + p2 ) p2 x2 > 0 ,
equivalent to
(M 2 + p2 )y (M 2 + p2 )2
x2 2x + >0.
2M 2 M2
The roots of this quadratic equations are
(M 2 + p2 ) (M 2 + p2 )y p 2
x1,2 = y 4M 2 ,
2M 2 2M 2
and the whole parabola lays above the horizontal axes only if the discriminant
is negative, and this yields
eE
y 2 4M 2 < 0 = < 2M .
M
Thus, we get the essential restriction on the magnitude of the electric field 0 <
0.
We further address the variant 0 > 0, () lower sign:
2(M 2 + p2 )
q
> 0 20 + 4p2 ;
(y 0 )
which is evidently valid.
34 E. M. Ovsiyuk, Ya. A. Voynova, V. V. Kisel et al.
2(M 2 + p2 )
q
0 > 20 + 4p2 .
(y 0 )
Now, we can repeat the previous analysis. With the help of the variable y0 =
x, we get
(M 2 + p2 )2 (y x)x(M 2 + p2 ) p2 x2 > 0
or
(M 2 + p2 )y (M 2 + p2 )2
x2 2x + >0;
2M 2 M2
the corresponding roots are
(M 2 + p2 ) (M 2 + p2 )y p 2
x1,2 = y 4M 2 .
2M 2 2M 2
The whole parabola lays above the horizontal axis only if the discriminant is
negative, which yields
eE
y 2 4M 2 < 0 = < 2M .
M
eE
Considering the previous assumption 0 < 0 , we derive 0 < 0 < M.
All in all, we conclude that 2 is positive, if the following double inequality
is valid
eE
0 < < 2M .
M
We remark that this equation, after its transforming to a new variable x is from
mathematical point of view very similar to that arising for the non-relativistic
quantum harmonic oscillator:
2 2
+ eEz d 2 2 2 d 2
x= , + (eE) x = 0 , + E kx f = 0 .
eE dx2 dx2
Let us use in (30) a new variable Z
( + eEz)2 2
Z=i (let it be = , eE > 0) .
eE 4eE
Then we obtain an equation of the form
2
d 1/2 d 1 i
+ + (Z) = 0 ,
dZ 2 Z dZ 4 Z
which has two singular points. The point Z = 0 is regular, and the behavior of
the solutions in its neighborhood may be as follows
A 1
Z 0, (Z) = Z , A 0, .
2
The point Z = is an irregular singularity of rank 2. Indeed, in terms of the
variable y = Z 1 , the above equation reads
2
d 3/2 d 1 i
+ + =0.
dy 2 y dy 4y 4 y 3
When y 0, the corresponding asymptotic structure is given by
whence it follows
1 1 1 1
D1 = + , C1 = + i ; D2 = , C2 = i .
2 4 2 4
Thus, at infinity, two asymptotics are possible
( C D Z
C DZ
Z 1 e 1 = Z 1/4i e+Z/2
Z , =Z e =
Z C2 eD2 Z = Z 1/4+i eZ/2 ,
where9
(+eEz)2
Z=i eE = iZ0 , Z0 > 0 , eZ/2 = eiZ0 /2 ,
1/4i 1/4i
Z 1/4i = eln iZ0 = eln Z0 +i/2 .
We shall further construct a solution in the whole region of Z. We start with
the substitution
(Z) = Z A eBZ f (Z) .
This leads to
d2
1 d 2 1 A (2A 1) B
Z + (2A + + 2BZ) + (B ) Z + + 2AB + + i f (Z) = 0 .
dZ 2 2 dZ 4 2Z 2
11
We use (again) the main branch of the logarithmic function
38 E. M. Ovsiyuk, Ya. A. Voynova, V. V. Kisel et al.
1/4i
Z ac = (iZ0 )1/4i = eln Z0 +i/2 = e+(1/4i)i/2 e(1/4i) ln Z0 ,
3/4i
Z a1 = (iZ0 )3/4i = eln Z0 +i/2 = e+(3/4i)i/2 e(3/4i) ln Z0 ,
(2 c) (3/2) (2 c) (3/2)
= , = ,
(1 a) (3/4 + i) (a c + 1) (3/4 i)
we derive the asymptotic formula
n
(3/2)
F (a c + 1, 2 c, Z) = eiZ0 /2 (3/4+i)
e(3/4+i)i/2 e(3/4+i) ln Z0 eiZ0 /2 +
o
(3/2)
+ (3/4i) e+(3/4i)i/2 e(3/4i) ln Z0 e+iZ0 /2 .
Spin 1 Particle with Anomalous Magnetic Moment ... 39
(3/2)
e(3/4+i)i/2 e(1/4+i) ln Z0 eiZ0 /2 +
(3/4 + i)
(3/2) +(3/4i)i/2 (1/4i) ln Z0 +iZ0 /2
+ e e e . (34)
(3/4 i)
The two pairs {Y5 , Y7 } and {Y1 , Y2 } relate to each other by the Kummer for-
mulas
(1 c) (c 1) (1 c) (c 1) ic
Y5 = Y1 + Y2 , Y7 = Y1 e Y2 .
(a c + 1) (a) (1 a) (c a)
These formulas after passing to the functions (Z) take the form
1/4+i
5 = eZ/2 Y5 = eiZ0 /2 eln Z0 +i/2 ,
i/2 1/4i
7 = eZ/2 Y7 (Z) = e+iZ0 /2 eln Z0
.
We see that these functions are conjugate to each other; only these ones enter
the superpositions (33) and (34).
12
We recall that Z = e(1/2)(ln Z0 +i/2) .
40 E. M. Ovsiyuk, Ya. A. Voynova, V. V. Kisel et al.
2E 2eE 2E
e 0, , = ,
M M M
where is a dimensionless parameter; the new has the electric charge dimen-
sion. We consider below only the main relations:
= 12 + dxd3 + i , b = 12 dxd3 + i ,
a
d2
ab + b
( a) = dz 2
2 , ab ba) = 0 ;
(
2
1 d 2 p2 M 2 (p f p f ) = 0 ;
f[12] = iM (p2 f1 p1 f2 ) , dz2
+ 2 1 1 2
F = f[34] , G = p1 f1 + p2 f2 ,
h i
1 ( ab + b
a ) + p2 + M 2 G = p2 F ,
h i
1 (ab + b
a) + p2 + M 2 F = G + F ;
1 = G 1 F , 2 = G 2 F ,
1 = 21 ( + 2 4p2 ) , 2 = 12 ( 2 4p2 ) ;
p p
2
d 2 p2 M 2 +
dz 2 + 1,2 1,2 (z) = 0 .
where
i0 0
p q q
1,2 = ( 2 4p2) = i0 20 4p2 = 0 20 + 4p2 .
2 2 2
The solutions will have the form of plane waves 1,2 (z) = eip3 z , only if
1
q
2 2 2
pz = 0 0 0 + 4p > 0 .
2
Spin 1 Particle with Anomalous Magnetic Moment ... 41
20 < 2 ,
2
2 > ,
+ p2
and thus conclude by the restriction
1. 0 < 0 < p , 0 > 0 . = 2 p2 M 2 > 0 .
+ p2
2
20 < , 0 < 0 .
+ p2
By summing, we conclude that the parameter 0 must lay within the fol-
lowing bounds:
0 < + p , = 2 p2 M 2 > 0 .
+ p2
Conclusion
Within the matrix 10-dimensional Duffin-Kemmer-Petiau formalism applied to
the Shamaly-Capri field, we study the behavior of a vector particle with anoma-
lous magnetic moment in the presence of an external uniform electric field. The
separation of variables in the wave equation is performed by using projective
operator techniques and the theory of DKP-algebras. The whole wave function
is decomposed into the sum of three components 0 , + , + . It is enough to
solve the equation for the main component 0 , the two remaining ones being
uniquely determined by it. Consequently, the problem reduces to three inde-
pendent differential equations for three functions, which are of the type of one-
dimensional KleinFockGordon equation in the presence of a uniform electric
field modified by the non-vanishing anomalous magnetic moment of the parti-
cle. The solutions are constructed in terms of confluent hypergeometric func-
tions. For assigning physical sense for these solutions, one must impose special
restrictions on a certain parameter related to the anomalous moment of the par-
ticle. The neutral spin 1 particle is considered as well. In this case, the main
manifestation of the anomalous magnetic moment consists in the modification
of the ordinary plane wave solution along the electric field direction. Again,
one must impose special restrictions on a parameter related to the anomalous
moment of the particle.
Spin 1 Particle with Anomalous Magnetic Moment ... 43
References
[1] E. S. Fradkin. To the theory of particles with higher spins. JETP. 1950.
Vol. 20. P. 2738.
[2] V. Ya. Feinberg. To interaction of the particlesof higher spins with electro-
magnetic fields. Trudy FIAN. USSR. 1955. Vol. 6. P. 269332.
[3] M. Petras. A note to Bhabhas equation for a particle with maximum spin
3/2 M. Petras // Czech. J. Phys.-1955. Vol. 5. Vol. 3. P. 418419.
[4] I. Ulehla. Anomalous equations for spin 1/2 particles. JETP. 1957. Vol.
33. P. 473477.
[6] A. Z. Capri. Nonuniqueness of the spin 1/2 equation. Phys. Rev. 1969. Vol.
178. Vol. 5. P. 1811815; First order wave equations for half-odd-integral
spin. Phys. Rev. 1969. Vol. 178. P. 24272433; Electromagnetic properties
of a new spin-1/2 field. Progr. Theor. Phys. 1972. Vol. 48. P. 13641374.
Chapter 3
Abstract
Within the matrix 10-dimensional DuffinKemmer-Petiau formalism
applied to the ShamalyCapri field, we study the behavior of a vector
E-mail address: vasiliy-bspu@mail.ru.
E-mail address: voinyuschka@mail.ru.
E-mail address: e.ovsiyuk@mail.ru.
E-mail address: vladimir.balan@upb.ro.
E-mail address: redkov@dragon.bas-net.by.
48 V. V. Kisel, Ya. A. Voynova, E. M. Ovsiyuk et al.
1. Introduction
Commonly, we shall use only the simplest wave equations for fundamental par-
ticles of spin 0, 1/2, 1. Meanwhile, it is known that other more complicated
equations can be proposed for particles with such spins, which are based on
the application of extended sets of Lorentz group representations (see [1]-[16]).
Such generalized wave equations allow to describe more complicated objects,
which have besides mass, spin, and electric charge, other electromagnetic char-
acteristics, like polarizability or anomalous magnetic moment. These additional
characteristics manifest themselves explicitly in presence of external electro-
magnetic fields.
In particular, within this approach Petras [3] proposed a 20-component the-
ory for spin 1/2 particle, which-after excluding 16 subsidiary components - turns
to be equivalent to the Dirac particle theory modified by presence of the Pauli
interaction term. In other words, this theory describes a spin 1/2 particle with
anomalous magnetic moment.
Techniques of Projective Operators Used to Construct Solutions ... 49
+ = + , [ , J] ] = ;
0 0 0 0 0 0 1 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0
0 0 0 0 0 0 0 0 1 0
2 = 0 0 1 0 0 0 0 0 0 0 ,
0 0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 1 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 1 0 0 0 0
0 0 0 0 1 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 1
3 = 0 1 0 0 0 0 0 0 0 0 ,
1 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 1 0 0 0 0 0 0
Techniques of Projective Operators Used to Construct Solutions ... 51
0 0 0 0 0 0 0 1 0 0
0 0 0 0 0 0 0 0 1 0
0 0 0 0 0 0 0 0 0 1
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
4 = .
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0 0
0 = P0 , + = P+ , = P , = + 0 + + .
52 V. V. Kisel, Ya. A. Voynova, E. M. Ovsiyuk et al.
eB B0
= B0 , 4 3 3 = . (4)
2 M
We further act on (3) by P0 ; by applying the identities
P0 3 = 3 P0 , P0 4 = 4 P0 , PY = Y P ,
P0 1 = 1 (1 P0 ) = 1 (P+ + P ) , P0 2 = 2 (1 P0 ) = 2 (P+ + P ) ,
we get
[3 3 + 4 4 + M ] 0 +
sin cos
+ 1 (cos ) + 2 (sin + ) + iB0 r sin 1 iB0 r cos 2 +
r r r r
sin cos
+ 1 (cos ) + 2 (sin + ) + iB0 r sin 1 iB0 r cos 2 = 0,
r r r r
Y P0 0 = Y 0 = (Y P0 ) = 0.
[3 3 + 4 4 + M ] 0 +
1 +i i i i
+ e ( + + B0 r) + e + ( B0 r) + +
2 r r r r
1 i i
+ e+i ( + + B0 r) + ei + ( B0 r) = 0 .
2 r r r r
By making use of the projective operators
1
P = [1 1 21 1 2 2 i(1 2 2 1 )],
2
and the commutation relations for DKP-matrices, we prove the identities
P+ = + P = 0; so the above equation is written simpler
[3 3 + 4 4 + M ] 0 +
+ 12 ei + ( r
i
r B0 r)+ + 1 e+i (
2 r + i
r + B0 r) = 0 .
(5)
Now, we act on (3) by 1 P0 = P+ + P ; this gives
(1 P0 )1 cos r sinr
+ iB0 r sin +
+(1 P0 )2 sin r + cosr
iB0 r cos +
+ (3 3 + 4 4 + P Y + M ) (+ + ) = 0 .
[3 3 + 4 4 + M ] 0 +
1 i 1 i
+ ei + ( B0 r)+ + e+i ( + + B0 r) = 0 ,
2 r r 2 r r
1 i
(3 3 + 4 4 + P + M ) + + e+i ( + + B0 r)0 = 0 ,
2 r r
1 i
(3 3 + 4 4 P + M ) + ei + ( B0 r)0 = 0 . (9)
2 r r
To separate the variables, we search for three components of the wave func-
tion in the form
The resulting from (9) radial equations are written in symbolic form as
(ip3 3 + ip4 4 + M ) f0 +
Techniques of Projective Operators Used to Construct Solutions ... 55
1 d m+1 1 d m1
+ + ( + B0 r)f+ + ( + B0 r)f = 0 ,
2 dr r 2 dr r
1 d m
(ip3 3 + ip4 4 + P + M ) f+ + ( + B0 r)f0 = 0 ,
2 dr r
1 d m
(ip3 3 + ip4 4 P + M ) f + + ( + B0 r)f0 = 0 . (10)
2 dr r
(i m+1 f+ bm1 f = 0 ,
p + M ) f0 + + a (11)
p + P + M ) f+ bm f0 = 0 ,
(i (12)
(i
p P + M ) f + + a
m f0 = 0 .
To proceed with these equations, we introduce the matrices3 with the prop-
erties
(i p + M ) = p2 + M 2 ,
p + M )(i
(A + M )(A + M ) = p2 + M 2 ,
(C + M )(C + M ) = p2 + M 2 . (13)
3
We take in the account that p2 = p23 + p24
Techniques of Projective Operators Used to Construct Solutions ... 57
In fact these formulas determine the inverse matrices to within numerical factors
(p2 + M 2 )1 . Then the system of radial equations can be rewritten alternatively
p + M )(p2 + M 2 )f0 +
(i
0
++ a
m+1 (A + M ) bm f0 + bm1 (C + M )+ 0
m f0 = 0 ;
a (15)
In fact, the equations (16) mean that it suffices to solve (15) with respect to f0 ;
the two other components f+ and f can be calculated by means of equations
(16).
To proceed further, we need to know the explicit form of the inverse opera-
tors (13). To solve this task, we first establish the minimal polynomials for the
relevant matrices. The minimal polynomial for (i p) is [3]
i p)2 + p2 ] = 0 .
p [(i (17)
= P + P = P + P , P = P , P = P ,
58 V. V. Kisel, Ya. A. Voynova, E. M. Ovsiyuk et al.
P P = P P = 0, P = P , P = P ,
P + P = 1, P P = P P = 0,
we find
1 M p2
A2 = (M 2 2
p
M
p 2
) = .
(M + )2 M +
Thus, we get the minimal polynomial for A
M )(i M p2 (M + P ) M p2
A3 = (M +P p)
p 2
= (i
p)= A.
(M + )2 (M + ) M + M +
Similarly, we find
M p2
C3 = C.
M
Therefore, the needed inverse operators must be quadratic with respect to
the relevant matrices. They are given by the formulas:
1
p)2 M (i p) + (p2 + M 2 ) ,
(M + ip) = M (i
h i
p2 +M 2 M + M + 2
(A + M ) = M 1 p2 +M 2 +M A + M (p2+M 2 +M ) A ,
2 2
h i
(C + M ) = p +M M 1 2
M
2
p +M M
C + 2
M
2
M (p +M M )
C 2 .
0 0 0 0 0 p3 0 p4 0 0
0 0 0 0 p3 0 0 0 p4 0
0 0 0 0 0 0 0 0 0 p4
0 0 0 0 0 0 0 0 0 p3
0 p 3 0 0 0 0 0 0 0 0
i
p = i ,
p 3 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
p 0 0 0 0 0 0 0 0 0
4
0 p4 0 0 0 0 0 0 0 0
0 0 p4 p3 0 0 0 0 0 0
Techniques of Projective Operators Used to Construct Solutions ... 59
p2 0 0 0 0 0 0 0 0 0
0 p2 0 0 0 0 0 0 0 0
0 0 p24 p3 p4 0 0 0 0 0 0
0 0 p3 p4 p23 0 0 0 0 0 0
2
0 0 0 0 p23 0 0 0 p3 p4 0
(i
p) = ,
0 0 0 0 0 p23 0 p3 p4 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 p3 p4 0 p24 0 0
0 0 0 0 p3 p4 0 0 0 p24 0
0 0 0 0 0 0 0 0 0 p2
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 p23 p3 p4 0 0 0 0 0 0
0 0 p3 p4 p24 0 0 0 0 0 0
0 0 0 0 p24 0 0 0 p3 p4 0
p)2 + p2 =
(i ;
0 0 0 0 0 p24 0 p3 p4 0 0
0 0 0 0 0 0 p2 0 0 0
0 0 0 0 0 p3 p4 0 p23 0 0
0 0 0 0 p3 p4 0 0 0 p23 0
0 0 0 0 0 0 0 0 0 0
we may prove the validness of the identity
i p)2 + p2 ] =
p[(i
0 0 0 0 0 p3 0 p4 0 0
0 0 0 0 p3 0 0 0 p4 0
0 0 0 0 0 0 0 0 0 p4
0 0 0 0 0 0 0 0 0 p3
0 p3 0 0 0 0 0 0 0 0
i ,
p 3 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
p 0 0 0 0 0 0 0 0 0
4
0 p 0 0 0 0 0 0 0 0
4
0 0 p4 p3 0 0 0 0 0 0
60 V. V. Kisel, Ya. A. Voynova, E. M. Ovsiyuk et al.
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 p23 p3 p4
0 0 0 0 0 0 0
0 p3 p4 p24
0 0 0 0 0 0 0
2
0 0 0 0 p4 0 0 0 p3 p4 0
0.
0 0 0 0 0 p24 0 p3 p4 0 0
0 0 0 0 0 0 p 2 0 0 0
0 0 0 0 0 p3 p4 0 p23 0 0
0 0 0 0 p3 p4 0 0 0 p23 0
0 0 0 0 0 0 0 0 0 0
p) bm1 (C + M ) +
+(M + i 0
am f0 = 0 .
Taking into account the explicit form for inverse operators, we get
(p2 + M 2 )f0 + M1 2 (i p)2 M (ip) + (p2 + M 2 ) + a
m+1
h i
1 p2 +MM + M + 2 0
2 +M A + M (p2 +M 2 +M ) A bm f0 +
+ M1 2 (i
p)2 M (i p) + (p2 + M 2 ) bm1
h i
M M 2 0
1 p2 +M 2 M C + M (p2+M 2 M )
C + am f0 = 0 .
p)2
h i
M +P (i M +P
1 p2 +M 2 +M i
p + p2 +M 2 +M M + a m+1 bm f0 +
p)2
h i
M P (i M P
1 p2 +M 2 M ip + p2 +M 2 M M + bm1 am f0 = 0 .
Techniques of Projective Operators Used to Construct Solutions ... 61
m+1 bm M 2 (p2 +M
{ (p2 + M 2 ) + a 1 1
2 +M ) M +
[ (p2 + M 2 + M )(i
p)2 + M (p2 + M 2 + M )i
p + +
+(p2 + M 2 )(p2 + M 2 + M )+ (p2 + M 2 )+ i
p(M + P )+
p)2 ] (M + P ) + bm1 am M 2 (p2+M
+(p2 + M 2 )+ (i 1 1
2 M ) M
[ (p2 + M 2 M )(i
p)2 M (p2 + M 2 M )i
p +
+(p2 + M 2 )(p2 + M 2 M ) (p2 + M 2 ) i
p(M P )+
p)2 ] (M P )+ }f0 = 0.
+(p2 + M 2 ) (i
p, and matrices + , , P .
Now we take into account the explicit form of f0 , i
Then we obtain
0
0
f3
f4
2 0 1 1
2
+ am+1bm 2 2
(p + M ) 2
0 M (p + M + M ) M +
f12
0
0
f
34
0 0
8
>
0 0
>
>
>
>
(M + )f3 0
>
>
>
>
(M + )f4 0
>
>
>
>
0 0
<
2 2 2 2
(p + M )(p + M + M ) iM (M + )(p2 + M 2 + M )
>
> 0 0
M f12 0
>
>
>
>
0 0
>
>
>
>
0 0
>
>
>
:
0 (p f p f )
4 3 3 4
62 V. V. Kisel, Ya. A. Voynova, E. M. Ovsiyuk et al.
0 0
0 0
p4 (p4f3 p3 f4 ) p3 (M + )(p3 f3 + p4 f4 )
p3 (p4 f3 p3 f4 ) p4 (M + )(p3 f3 + p4 f4 )
0 0
2 2 2 2
(M + )(p + M + M ) (p + M ) +
0 0
0
2
p M f12
0 0
0
0
0 0
0
9
>
0
>
>>
>
p3 f12
>
>
>
>
p4 f12
>
>>
> 1 1
0
=
+M (M + )(p2 + M 2 ) + m 2 2
bm1 a
0 >> M (p + M 2 M ) M
(p3 f3 + p4 f4 ) >
>>
0 >>
>>
>
0
>
>>
;
0
0 0
8
>
0 0
>
>
>
>
(M )f 0
>
3
>
>
>
(M )f4 0
>
>
>
>
0 0
<
2 2 2 2 2 2
(p + M )(p + M M ) iM (M )(p + M M )
>
> 0 0
M f12 0
>
>
>
>
0 0
>
>
>
>
>
>
>
:
0
0
0 (p f p f )
4 3 3 4
0 0
0 0
p4 (p4f3 p3 f4 ) p3 (M )(p3 f3 + p4 f4 )
p3 (p4 f3 p3 f4 ) p4 (M )(p3 f3 + p4 f4 )
0 0
(M )(p2 + M 2 M ) (p2 + M 2 )
0 0
0
2
p M f12
0
0
0 0
0 0
9
0 >
>
0 >
>
>
>
p3 f12 >
>
>
>
p4 f12 >
>
>
>
2 2
0 =
M (M )(p + M ) = 0.
0 >
>
(p3 f3 + p4 f4 ) >
>
>
>
0 >
>
>
>
0 >
>
>
>
0 ;
Techniques of Projective Operators Used to Construct Solutions ... 63
(18)
1 1
m+1 bm
(p2 + M 2 )f4 + a
M 2 (p2 2
+ M + M ) M +
{ (p2 + M 2 )(p2 + M 2 + M )(M + )f4 +
+p3 (M + )(p2 + M 2 + M )(p4 f3 p3 f4 )
p4 (p2 + M 2 )(M + )(p3 f3 + p4 f4 ) + p4 M (M + )(p2 + M 2 )f12 }+
1 1
+bm1
am
M 2 (p2 2
+ M M ) M
{(p2 + M 2 )(p2 + M 2 M )(M )f4 +
+p3 (M )(p2 + M 2 M )(p4 f3 p3 f4 )
p4 (p2 + M 2 )(M )(p3 f3 + p4 f4 )
p4 M (M )(p2 + M 2 )f12 } = 0 ,
(19)
1 1
m+1 bm
(p2 + M 2 )f12 + a
M 2 (p2 2
+ M + M ) M +
64 V. V. Kisel, Ya. A. Voynova, E. M. Ovsiyuk et al.
(20)
(p2 + M 2 )f34 +
1 1
am+1 bm
+ {iM (p2 + M 2 + M )(M + )(p4 f3 p3 f4 )}+
M 2 (p2 + M 2 + M ) M +
1 1
+bm1
am {iM (p2 +M 2 M )(M )(p4 f3 p3 f4 )} = 0 ,
M 2 (p2 + M 2 M ) M
i
(p2 + M 2 )f34 am+1 bm + bm1 am )(p4 f3 p3 f4 ) = 0 .
( (21)
M
The equations (18) and (19) may be simplified to
m+1 bm
a
(p2 + M 2 )f3 + {(p2 + M 2 )f3 p4 (p4 f3 p3 f4 )
M2
(p2 + M 2 ) M (p2 + M 2 )
p (p f
3 3 3 + p f
4 4 ) + p3 f12 }+
p2 + M 2 + M p2 + M 2 + M
bm1 a
m
+ 2
{ (p2 + M 2 )f3 p4 (p4 f3 p3 f4 )
M
(p2 + M 2 ) M (p2 + M 2 )
2 p 3 (p 3 f3 + p 4 f4 ) p3 f12 } = 0 ,
p + M2 M p2 + M 2 M
m+1 bm
a
(p2 + M 2 )f4 + { (p2 + M 2 )f4 + p3 (p4 f3 p3 f4 )
M2
(p2 + M 2 ) M (p2 + M 2 )
2 p 4 (p 3 f3 + p 4 f4 ) + p4 f12 }+
p + M2 + M p2 + M 2 + M
bm1 am
+ {(p2 + M 2 )f4 + p3 (p4 f3 p3 f4 )
M2
Techniques of Projective Operators Used to Construct Solutions ... 65
(p2 + M 2 ) M (p2 + M 2 )
p (p f
4 3 3 + p f
4 4 ) p4 f12 } = 0 .
p2 + M 2 M p2 + M 2 M
By multiplying the first equation by p4 , and the second one by p3 , and then
summing the two results, we find
(p2 + M 2 )(p4 f3 p3 f4 )+
m+1 bm 2
a
+ [(p + M 2 )(p4 f3 p3 f4 ) p2 (p4 f3 p3 f4 )]+
M2
bm1 a
m 2
+ 2
[(p + M 2 )(p4 f3 p3 f4 ) p2 (p4 f3 p3 f4 )] = 0
M
or h i
am+1 bm + bm1 a
m + p2 + M 2 (p4 f3 p3 f4 ) = 0 .
am+1bm + bm1 a
( m )(p4 f3 p3 f4 ) = iM (p2 + M 2 )f34 .
we obtain
i
f34 = (p4 f3 p3 f4 )
M
We further consider (20), which simplifies to te form
(p2 + M 2 )
m+1 bm
(p2 + M 2 )f12 + a {M f12 (p3 f3 + p4 f4 )}+
M (p2 + M 2 + M )
(p2 + M 2 )
+bm1 a
m {M f12 + (p3 f3 + p4 f4 )} = 0 ,
M (p2 + M 2 M )
or
am+1
bm
f12 + M (p2+M 2 +M ) {M f12 (p3 f3 + p4 f4 )}+
b
am
+ M (p2m1
+M 2 M )
{M f12 + (p3 f3 + p4 f4 )} = 0 ,
whence after elementary transformation we get
h i
(p2 +M 2 )2 M 2 2
(am+1 bm + bm1 am ) + p2 +M 2
+ 2M B0
p2 +M 2
f12 +
h i
+ p2 +M
am+1 bm + bm1 am ) + 2B
2 ( M
0
(p3 f3 + p4 f4 ) = 0 ;
66 V. V. Kisel, Ya. A. Voynova, E. M. Ovsiyuk et al.
m+1 bm bm1 a
a m = 2B0 .
m+1 bm
a
(p2 + M 2 )f3 + {(p2 + M 2 )f3 p4 (p4 f3 p3 f4 )
M2
(p2 + M 2 ) M (p2 + M 2 )
p 3 (p 3 f3 + p 4 f4 ) + p3 f12 }+
p2 + M 2 + M p2 + M 2 + M
bm1 a
m
+ 2
{ (p2 + M 2 )f3 p4 (p4 f3 p3 f4 )
M
(p2 + M 2 ) M (p2 + M 2 )
2 p 3 (p 3 f3 + p 4 f4 ) p3 f12 } = 0 ,
p + M2 M p2 + M 2 M
m+1 bm
a
(p2 + M 2 )f4 + { (p2 + M 2 )f4 + p3 (p4 f3 p3 f4 )
M2
(p2 + M 2 ) M (p2 + M 2 )
2 2
p4 (p3 f3 + p4 f4 ) + 2 p4 f12 }+
p + M + M p + M2 + M
bm1 am
+ {(p2 + M 2 )f4 + p3 (p4 f3 p3 f4 )
M2
(p2 + M 2 ) M (p2 + M 2 )
2 2
p4 (p3 f3 + p4 f4 ) 2 p4 f12 } = 0 ,
p + M M p + M2 M
By multiplying the first relation by p3 , and the second one by p4 , and summing
the results, we find
1
(p3 f3 + p4 f4 ) + am+1bm [(M + )(p3 f3 + p4 f4 ) + p2 f12 ]+
M (p2 +2 +M )
1
+bm1 a
m [(M )(p3 f3 + p4 f4 )] p2 f12 ] = 0 .
M (p2 + M 2 M )
Thus, we have found two equations for (p3 f3 + p4 f4 ) and f12 :
(p2 +M 2 )2 M 2 2
h i
am+1
( bm + m ) +
bm1 a p2 +M 2 i
+ 2M B0
p2 +M 2
f12 +
h
2B0
+ p2 +M 2 (
am+1 bm + bm1 am ) + M (p3 f3 + p4 f4 ) = 0 ;
am+1
{(p2 + M 2 )2 M 2 2 }(p3 f3 + p4 f4 ) + (p2 + M 2 2 )( bm + m )(p3 f3 + p4 f4 )
bm1 a
2
h 2 2 i
2BM0 p (p f + p f ) p2 (
3 3 4 4 am+1bm + bm1 am ) + 2B0 (pM+M ) f12 = 0.
Techniques of Projective Operators Used to Construct Solutions ... 67
These equations may be reduced to such a form, that the 2-nd order operator
am+1bm + bm1 a
( m ) acts on a single function:
h i
m+1 bm + bm1
` 2B
am + p2 + M 2 2 2BM0 f12 = 0 ,
M
0
(p3 f3 + p4 f4 ) + a
h i 2
am+1 bm + bm1 a
m + p2 + M 2 (p3 f3 + p4 f4 ) + p2 2BM
0p
f12 = 0 .
Thus, the final form of the equations for the four functions f3 , f4 , f12 , f34
has the following relatively simple structure:
i
f34 = (p4 f3 p3 f4 ) , (22)
M
h i
am+1 bm + bm1 a
m + p2 + M 2 (p4 f3 p3 f4 ) = 0 , (23)
h i 2B0
am+1 bm + bm1 am + p2 + M 2 (p3 f3 + p4 f4 ) = p2 ( )f12 , (24)
M
h i
am+1 bm + bm1 am + p2 + M 2 f12 =
(25)
2B0 2B0
= M f12 + M (p3 f3 + p4 f4 ) .
The analysis of (22) and (23) can be now clearly done. The second sub-
system (24)(25) is solved through diagonalizing the mixing matrix.
To this aim, let us introduce the new functions
where 01 = ( 2B 0 2B0
M )1 , and 2 = ( M )2 .
0
68 V. V. Kisel, Ya. A. Voynova, E. M. Ovsiyuk et al.
d2 2 M 2 p23 01
d |m| + m + 1 1 = 0 .
x + (|m| + 1 x)
dx2 dx 2 4|B0 |
|m| + m + 1 2 M 2 p23 01
= n ;
2 4|B0 |
z 2 x2 2
z 2N + 2M 2 + x, E 2 E(z + x2 ) + = 0;
4
and the roots are
z + x2 1 p
E1 = + (z + x2 )2 (z 2 x2 2 ),
2 2
z + x2 1 p
E2 = (z + x2 )2 (z 2 x2 2 ). (27)
2 2
To have both E1 and E2 real-valued and positive (such that these refer to phys-
ical energy levels), we require
z + x2 = (2N + 2M 2 ) + x + x2 > 0;
is valid due to (28). The third inequality 2zx2 + x4 + x2 2 > 0 is valid due to
z = 2N + 2M 2 + x, x > 0.
4
We remind that B0 = |B0| < 0.
70 V. V. Kisel, Ya. A. Voynova, E. M. Ovsiyuk et al.
We note that the possibility of positive values > 0, > 2|B0 |/M is ignored,
because in this case the admissible region for does not contain the close to
zero values. The two remaining inequalities are valid as well:
z + x2 = (2N + 2M 2 ) + x + x2 > 0;
5. Conclusion to Sections 24
Let us summarize the main results of the Sections 24. Three series of the
energy levels are found; two of them substantially differ from those for spin 1
particles without anomalous magnetic moment.
The formula (27) and its restriction (29) provide us with two series for en-
ergy levels6 in both cases B0 = |B0 |, and B0 = +|B0 |:
2 1
E1 = z+x
p
2 + 2 (z + x2 )2 (z 2 x2 2 ),
2 1
E2 = z+x
p
2 2 (z + x2 )2 (z 2 x2 2 ).
To assign to the energies E1 and E2 a physical sense for all the values of the
main quantum number n = 0, 1, 2, ..., one must impose special restrictions
which are explicitly formulated on the values of the anomalous magnetic mo-
ment. Without these restrictions, only some part of the energy levels correspond
to bound states.
The third series of the energy levels (see (23)) has the form:
in these states the anomalous magnetic moment does not manifest itself at all.
6
We remember the formal change m = m, when inverting the orientation of the magnetic
field
72 V. V. Kisel, Ya. A. Voynova, E. M. Ovsiyuk et al.
Solving (30) and (31) is a trivial task. The system (31)(33) can be solved
trough the diagonalization of the mixing matrix. Let us introduce the notation
1h i
= m+1 bm + bm1 am + p2 + M 2 , (p3 f3 +p4 f4 ) = 1 , f12 = 2 ;
a
then (32)(33) reads in matrix form as follows
p2 1 p2 1 1
1 0 1
=S 0
= = S S S
2 1 2 2 1 2
Requiring
0 p2 1 1 0
s s
S = 11 12
S
S = , ;
1 0 2 s21 s22
we derive
0 p2 1 0
s11 s12 s11 s12
1 = 0 2
,
s21 s22 s21 s22
d2 m2
1 d
2 M 2 p23 + 1,2 D , + + D 1,2 = 0 .
dr 2 r dr r2
Let us search for solutions of the form = r A eBr f (r); for f (r), we derive
d2 f
2
A m2 2AB + B
2A + 1 df 2
+ + 2B + + +B +D f =0.
dr 2 r dr r2 r
d2 f df
r 2
+ (2A + 1 + 2Br) + (2AB + B) f = 0.
dr dr
If we take the positive case A = + | m |, then the solutions are vanishing near
the point r = 0. Moreover, from physical considerations, we must require the
parameter D be positive, in order to agree with the correspondence principle:
=0 = D D0 = 2 M 2 p23 > 0 .
Without loss of generality, assume that B = +i D. In the new variable, the
above equation will read as a confluent hypergonetric equation
d2 f
df 1
2Br = x, x 2 + (2A + 1 x) A+ f =0
dx dx 2
that is
where q
x = 2Br = 2i M 2 p23 + 1,2 .
74 V. V. Kisel, Ya. A. Voynova, E. M. Ovsiyuk et al.
Thus, for a neutral particle, no bound states exist, and the qualitative manifes-
tation of the anomalous magnetic moment is mainly revealed by appearing of
space scaling of the arguments of the wave functions, in comparison with the
case of particles without the magnetic moment. Formally, we have two sorts of
states depending on the sign of :
r
1 p
1,2 , x = 2Br = 2i M 2 p23 + ( 2 4p2 ) ,
2
There exists a third type of states in which the parameter does not manifest
itself in any way (see (31)):
2
m2
d 1 d 2 2 2
+ + M p 3 (p4 f3 p3 f4 ) = 0 ;
dr 2 r dr r2
for these states, the solutions depend on the ordinary (non-modified) argument
x: q
x = 2Br = 2i M 2 p23 .
( i aa m) (x) = 0 , (34)
where the 20-component wave function includes the tensors , a, [ab], (ab),
and transforms by the following representation of the Lorentz group SO(3, 1)
a = i ( 1 e4,a 1 ea,4 +
+2 gkn en,[ka] 2 gkn e[ka],n 3 gkn en,(ka) 3 gkn e(ka),n ; (35)
Techniques of Projective Operators Used to Construct Solutions ... 75
(ab) 1 ab 1
(cd) = ca db + da cb g gcd , g (ab),(cd) = g ac g bd + g ad g bc g ab g cd ;
2 2
and also the generators J ab for the Lorentz group representation
Let us transform now (34) to its tensor form with respect to , a, [ab], (ab):
1 a a = m ,
1 b + 2 a[ba] 3 a (ba) = m b ,
2 ( ba ab ) = m [ba] ,
1
3 ( ba + a b gab c c ) = m (ab) . (37)
2
From the first and fourth equations in (37), by considering the relations (36), we
obtain
1
1 b 3 a (ba) = 3 3 a(a b b a)
m
or
3
1 b 3 a(ba) = 3 a[ab] .
2
76 V. V. Kisel, Ya. A. Voynova, E. M. Ovsiyuk et al.
So, the equations (34) and (40) are equivalent from physical standpoint, because
their solutions must be unambiguously mutually related.
The generalization of (34) to the case of arbitrary curved space-time with
the metric g (x) and any relevant e(a)(x), may be performed in accordance
with the tetrad method of TetrodeWeylFockIvanenko [19]. Such an equation
has the form
[ i ( + B ) m ] = 0 , (41)
or
i a cd
( ia(a) + J cda m ) = 0 . (42)
2
We use the notation
1 ab
= a e(a) , B = J e(a) e(b) ,
2
(a) = e(a) , abc = ( e(a))e(b)e(c) ;
here represents the covariant derivative, while abc stand for the Ricci rota-
tion coefficients.
Let us show that any two equations of the type (41)
0
[ i( + B ) m ] = 0 , [ i ( + B0 ) m ] 0 = 0 , (43)
Techniques of Projective Operators Used to Construct Solutions ... 77
which is a known relation, which ensures the Lorentz invariance of the wave
equation (34) in Special Relativity. To prove (45), in its form
1
SBS 1 = SJ ab S 1 e(a)
e(b)
2
we express the tetrad e(a) in terms of primed tetrad. This leads to
1 0
SB S 1 = (SJ abS 1 ) (L1 )ak [ ((L1 )bl gkl + (L1 )bl e(k) e0(l) ] .
2
By using the explicit form of S (see (44)) and of J ab , we prove
SJ ab S 1 = J mn Lma Lnb .
Then we obtain
1 mn b
SB S 1 = J Ln (Lmb ) + B 0 .
2
We can infer the conclusion provided that
1
SB S 1 = J mn Lnb (Lmb ) .
2
78 V. V. Kisel, Ya. A. Voynova, E. M. Ovsiyuk et al.
But the last relationship can by proved by using the known pseudo-orthogonality
condition for Lorentz transformations and the explicit form of S and J ab . Thus,
(43) are mutually related to each other by a local transformation of the type (44).
The general covariant matrix wave equation (42) may be translated to the
tetrad tensor form
1 ( (a) + bab ) a = m ,
1 (r) + 2 ( (a)[ra] + rbc [bc] + bcc [br] )
3 ( (a)(ra) + rdc (dc) + dcc (dr) ) = m r ,
2 ( (r)s (s) r + drs d dsr d ) = m [rs] ,
3 [ ( (r)s + (s) r + rd s d + sd r d )
1
grs ( (a)a + ada d ) ] = m (rs) . (46)
2
Let us eliminate the components, and obtain the equation for the main compo-
nents a and [cd] . To this end, from the first and the fourth equation in (46) we
express and (ad) and substitute the results into the second one. Due to the
conditions (36) and the third equation in (46), we get
3 3 (a)
= [ [ra] + rbc [bc] + bcc[br] ]
2
23 3 b ,(a)
[ ar b + bab,(r) a
m
abr (a)b r ab (a)b + r ab dba d abb car c ] .
Hence, we obtain
1
( (a)[ra] + rbc [bc] + bcc [br] )
2
23 3 b ,(a)
[ ar b + bab,(r) a abr (a)b r ab (a) b +
m
+ r ab dba d abb car c ] = mr . (47)
Techniques of Projective Operators Used to Construct Solutions ... 79
It remains to translate the equations (46) and (47) to the variables a and ab
according to (38). In the end, we derive the tetrad generalized Proca system:
Rabcd = abc,(d) + abd,(c) + akc kbd + abn ncd akd kbc abn ndc ,
1 D = m ,
1 D + 2 D [] 3 D () = m ,
2 (D D ) = m [] ,
1
3 [ D + D g ] = m () . (50)
2
The relations between the tetrad and the tensor components are:
(b) (b)
= e(a)
a , [] = e(a)
e [ab] , () = e(a)
e (ab) .
7
We consider the notation described by the equality D = ieAa (x).
80 V. V. Kisel, Ya. A. Voynova, E. M. Ovsiyuk et al.
As for (46), the system (50) can be reduced to the minimal form
1 233
2 D [] + m [D, D ] = m ,
2 (D D ) = m [] .
or, alternatively, to
23 3
D [] + [D, D ] = m , (51)
m
D D = m [] .
Taking into account that
[D, D ] = (ieF + R ) ,
we conclude that the parameter 3m 3 in (51) determines both the anomalous
magnetic moment of the spin 1 particle and the additional interaction term with
non-Euclidean space-time background through the Ricci tensor R .
Conclusion
By applying the matrix 10-dimensional DuffinKemmer-Petiau formalism to
the ShamalyCapri field, the behavior of a vector particle with anomalous mag-
netic moment is studied in the presence of external uniform magnetic field. The
separation of variables in the wave equation is performed by using projective
operator techniques and the DKP-algebra theory. The problem is reduced to
a system of 2-nd order differential equations for three independent functions,
which are solved in terms of confluent hypergeometric functions. Three series
of the energy levels are found; two of them substantially differ from those for
spin 1 particle without anomalous magnetic moment. To assign to them physical
sense for all the values of the main quantum number n = 0, 1, 2, ... one must
impose special restrictions on a parameter related to the anomalous moment.
Otherwise, the energy levels corresponds only partially of to bound states. The
neutral spin 1 particle is considered as well. In this case no bound states exist
in the system, and the main qualitative manifestation of the anomalous mag-
netic moment consists in the occurrence of space scaling of the arguments of
the wave functions, in comparison with a particle which has no such moment.
Some features of theory of the ShamalyCapri particle within General Relativ-
ity are given.
Techniques of Projective Operators Used to Construct Solutions ... 81
References
[1] E. S. Fradkin. To the theory of particles with higher spins. JETP. 1950.
Vol. 20. P. 2738.
[2] V. Ya. Feinberg. To interaction of the particlesof higher spins with electro-
magnetic fields. Trudy FIAN. USSR. 1955. Vol. 6. P. 269332.
[3] M. Petras. A note to Bhabhas equation for a particle with maximum spin
3/2 M. Petras // Czech. J. Phys.-1955. Vol. 5. Vol. 3. P. 418419.
[4] I. Ulehla. Anomalous equations for spin 1/2 particles. JETP. 1957. Vol.
33. P. 473477.
[6] A. Z. Capri. Nonuniqueness of the spin 1/2 equation. Phys. Rev. 1969. Vol.
178. Vol. 5. P. 1811815; First order wave equations for half-odd-integral
spin. Phys. Rev. 1969. Vol. 178. P. 24272433; Electromagnetic properties
of a new spin-1/2 field. Progr. Theor. Phys. 1972. Vol. 48. P. 13641374.
Chapter 4
A QUATERNION APPROACH IN
THE ESTIMATION OF THE ATTITUDE OF
AN INDUSTRIAL ROBOT
Abstract
Traditionally, the automotive industry has been the largest employer of
robots, but their control is inline and programmed to follow planning trajectories.
In this case, in the department motors test of Volkswagen Mexico a semi-
autonomous robot is developed. To date, some critical technical problems must
be solved in a number of areas, including in dynamics control. Generally, the
attitude estimation and the measurement of the angular velocity are a
86 B. B. Salmern-Quiroz, G. Villegas-Medina et al.
requirements for the attitude control. As a result, the computational cost and the
complexity of the control loop is relatively hight. This chapter deals with the
implementation of a cheap Micro AHRS (Attitude and Heading Reference
System) using low-cost inertial sensors. In the present chapter, the technique
proposed is designed with attitude estimation and the prediction movement via
the kinematic of a 4GDL robot. With this approach, only the measurements of at
least two non-collinear directional sensors are needed. Since the control laws are
highly simple and a model-based observer for angular velocity reconstruction is
not needed, the proposed new strategy is very suitable for embedded
implementations. The global convergence of the estimation and prediction
techniques is proved. Simulation with some robustness tests is performed.
1. Introduction
Robots have considerable potential for application in Volkswagen plants. Looking
at the four major sectors of a vehicle assembly operation, as follow:
In the motor test area, the assemblies of all of the instrumentation and wiring
systems, and the test per se, autonomous robotic and telerobotic systems have
been suggested. Industrial Robot has been considered for the different test.
To date, some critical technical problem must be solved in a number of areas,
including in dynamics and control. A prerequisite is state estimation where the
states typically are position, velocity and orientation. State estimation is especially
important where attitude control is needed. With attitude we refer to the robots
orientation relative to the gravity vector, usually described by pitch and roll
movements. Attitude estimation is usually performed by combining measurements
from three kinds of sensors: rate gyros, inclinometers and accelerometers.
A Quaternion Approach in the Estimation of the Attitude 87
2. Mathematical Background
As mentioned in the introduction, the attitude of a rigid body can be represented
by a unit quaternion, consisting of a unit vector known as the Euler axis, and a
rotation angle about this axis. The quaternion is then defined as follows:
2 0
=(
) = ( ) (1)
2
where
{ 0 2 + = 1
= (2)
= [0 ] , 0 , 3}
= [1 2 3 ] and 0 are known as the vector and scalar partis of the quaternion
recpectively. In attitude control aplications, the unit quaternion represents the
rotation from an inertial coordinate system ( , , ) located at some point in
the space (for instance, the earth NED frame), to the body coordinate system
( , , ) located on the center of mass of a rigid body.
If is a vector expressed in , then its coordinates in are expressed by:
= (3)
where = [0 ] and = [0 ] are the quaternions associated to vectors
and respectively. denotes the quaternion multiplication and is the
conjugate quaternion multiplication of q, defined as:
= [0 ] (4)
() = (0 2 )3 + 2( 0 [ ]) (5)
where 3 is the identity matrix and [ ]is a skew symmetric tensor associated with
the axis vector :
A Quaternion Approach in the Estimation of the Attitude 89
1 0 3 2
[ ] = (2 ) = ( 3 0 1 ) (6)
3 2 1 0
= () (7)
The quaternion attitude error used to quantify the mismatch between two
attitudes 1 and 2 is computed by:
= 1 2 (8)
Note
3. Problem Statement
Good models of industrial robots are necessary in a variety of applications, such
as mechanical design, performance simulation, control, diagnosis, supervision and
offline programming. This motivates the need for good modelling tools. In the
first part of this book the foreword kinematic modelling of serial industrial robots
is studied, the focus is on modelling the foreword kinematics. The main interest is
the principal structure, and issues regarding efficiently implementation have not
been considered. The work is based on homogeneous transformations using the
Denavit-Hartenberg (D-H) representation, which gives coordinate frames adapted
to the robot structure and a quaternion approach in the attitude estimation
modelling is presented.
Attitude is a term used to describe the orientation of a vehicle in three-
dimensional space. The attitude of the articulated arm can be described using
several different parameters (e.g., quaternions, Euler angles, direction cosine
matrices, etc.) that have been described in some detail in earlier chapters of this
book. Robotic attitude determination systems provide a means for measuring or
estimating these parameters that describe the end effector and the robots
orientation. The outputs of these systems are used for vehicle guidance,
navigation, and in this case, control and attitude determination. The focus of this
chapter is attitude determination systems for small robots, which, for the purposes
here, are defined to be robot with a total mass between 150 kg and 300 kg and fit
in a volume envelope roughly 50 cm 50 cm 60 cm in dimension. Designing
attitude determination systems for these small articulated robots represents a
challenging and specialized task; sensors and their processing must also be made
to fit within the limited size, weight, and mass specifications while still
performing to high accuracies for many applications. This chapter highlights these
challenges (particularly for a robot like showed in Figure 3) and discusses current
and future developments aimed at addressing them.
In the case of the attitude estimation, one seeks to estimate the attitude and
accelerations of a rigid body. From now on, it is assumed that the system (AHRS)
is equipped with a triaxis accelerometer, three magnetometery and three rate gyros
mounted orthogonally.
In this chapter, we describe the bodys kinematic of the model. In order to
estimate the arms robot position with respect to an inertial frame, a module
A Quaternion Approach in the Estimation of the Attitude 91
The equation describing the relation between the quaternion and the bodys
kinematic is given in introducing the angular variation
= [ ] from
this, it follows.
1 1
= 2
)() =
( 2
()
() (9)
where (
) y () are defined as:
[
]
(
) = [ ] (10)
0
0 3 3 + []
() = [ . ] (11)
0 3 2
[ ] = [ 3 0 1 ] (12)
2 1 0
= + 0 2 = 1 (13)
()() = 33 (14)
() () = 44
()() = 031
() = (0 2 )33 + 2 20 [ ] (15)
A Quaternion Approach in the Estimation of the Attitude 93
4. Robot Configuration
The robot links form a kinematic chain. When the kinematic chain is open, every
link is connected to every other link by one and only one chain. If, on the other
hand, a sequence of the links forms one or more loops, the robot contains closed
kinematic chains. In Figure 2 example of configurations of robots is showed.
The robot has a closed kinematic chain due to the so-called parallelogram-
linkage structure, represented by a mechanical coupling between motor, placed on
the foot of the robot, and the actual link 3. It can also be seen from the figure
(from the ABB internet web site) that has an open kinematic-chain structure.
Robots having an open kinematic chain can be divided into the following types,
based on geometry. The Cartesian robot has three prismatic joints and the links
are mutually orthogonal, which gives that each degree of mobility corresponds to
a degree of freedom in the Cartesian space. Changing the first prismatic joint to a
revolute joint gives a cylindrical geometry, where each degree of mobility
corresponds to a degree of freedom in cylindrical coordinates. Replacing the first
two prismatic joints by revolute joints gives a spherical robot, where the degrees
of freedom are in spherical coordinates, similar to the cases above. SCARA stands
94 B. B. Salmern-Quiroz, G. Villegas-Medina et al.
for selective compliance assembly robot arm, and a SCARA robot has a
mechanical structure with high stiffness to vertical load and compliance to
horizontal load. The anthropomorphic robot has three revolute joints, and has
similarities to a human arm. The workspace is approximately a portion of a sphere
and the robot structure can be seen in many industrial applications. The types are
explained in more detail in Sciavicco and Siciliano (2000) and Spong et al.
(2006), among others. The work in this thesis is limited to serial robots, that is,
robots with an open kinematic structure, and the class of parallel robots that can
be rewritten to this structure using a bilinear transformation. Especially serial
robots having only revolute joints, so-called anthropomorphic robots, are studied.
In this chapter we obtain the attitude from the robot show in Figure 3 with an
open kinematic structure.
Using coordinate frames attached to each joint, shown in Figure 3, the
position and orientation of the robot tool can be defined in the Cartesian
coordinates Ci, with respect to the base frame R0 of the robot by successive
coordinate transformations. This results in the relation.
, = 1: 3, Distance between the center of rotation and the center of attitude
sensor module.
, = 1: 3, Distance between the position of the attitude sensor module and
the end of the segment in consideration.
, = 1: 3, is the attitude sensor module.
The point 0 is supposed to be fixe.
A Quaternion Approach in the Estimation of the Attitude 95
link
1 1 0 1
2
2 2 1 0 0
3 3 2 0
2
Quaternion
1
= [cos 1 0 0 sin ] 1
2 2
2 2
= [cos sin 0 0] 2
2 2
= [cos 3 sin 3 0 0] 3
2 2
= 2 ( )
(16)
According to the robot, the model for foreword kinematics based on the
convention Denavit-Hartenberg is defined for each one of the coordinate systems
in Table 1.
This model can be compared with the quaternion-based model, which is
Simplified in Table 2, so it is possible to observe that the quaternion-based model
is significantly more compact, which reduces the number of operations
significantly.
1) Rate Gyros: The angular velocity 123 is measured by the rate gyros,
T
which are supposed to be orthogonally mounted. The output signal of a rate gyro
is influenced by various factors, such as bias drift and noise. In the absence of
rotation, the output signal can be modeled as the sum of a white Gaussian noise
and of a slowly varying function. Since an integration step is required in order to
obtain the current attitude quaternion (9), even the smallest variation of the rate
96 B. B. Salmern-Quiroz, G. Villegas-Medina et al.
gyro measurement will produce a wrong estimation of the attitude. The bias is
denoted by , belonging to space R 3 . The rate gyro measurements are modeled
by (Brown y Hwang, 1997):
G G (17)
T 1 (18)
bA C (q) a g A (19)
where g 0 0 g and a R
T 3
are the gravity vector and the inertial
accelerations of the body respectively. Both are expressed in frame N. g = 9:81
m/sec2 denotes the gravitational constant and A R is the vector of noises that
3
bM C (q)hM M (20)
1
q (q ) G K1 (21)
2
T 1 K2 . (22)
where T has been defined in (18) and Ki; i = 1; 2 are positive constant parameters
q is the prediction of the attitude at time t. It this obtained via the integration of
the kinematics equation (14) using the measured angular velocity G , the bias
estimate and
qe which is the vector part of the quaternion error qe.
Remember that qe measures the discrepancy between q and the pseudo measured
attitude qps (17). In this chapter, qps is obtained thanks to an appropriate treatment
of the accelerometer and magnetometer measurements and it will be explained in
the next section.
Combining (17), (18), (21) and (22) the error model is expressed as:
1 0 T qe 0
qe (23)
2 2 qe
T 1 K2 (24)
qe 0 1 0
qe0 1 2 generally 2n
98 B. B. Salmern-Quiroz, G. Villegas-Medina et al.
that is, there is only one equilibrium point in the physical 3D space.
We can observe that the global asymptotically convergence of the error to
zero ( qe0 1 , qe 0 , 0 ) and consequently the convergence of q to the real q is
given by:
G 0 q ps q (25)
where q is the true attitude quaternion of the rigid body. Thus, the convergence
is guaranteed if and only if:
qe0 1, qe 0, 0 (26)
Theorem 1. Consider the equilibrium states of the system (21)-(22) and let G be
the measured angular velocity. Then, the equilibrium point ( qe0 1 , qe 0 , 0 ) is
globally asymptotically stable.
1
V K 2 1 qe 0 q Te qe T
2
2
(27)
The derivative of (27), together with (23) and (24), is given by (28):
V 2 K 2 qe 0 T
K 2 T qe T T 1 K 2 (28)
K 2 T K1 T qe T T 1 K 2 T
V K2 K1qTe qe T T 1 0 (29)
A Quaternion Approach in the Estimation of the Attitude 99
1 2
() = 2
[ (=1( ()) ) + 22 ] (30)
with 1 = 0.
The process of Estimation and Prediction needs the determination of his
gradient; this one is obtained by equation (24)
= [ ( )]
2
= [2 ] (31)
Similarly, the gradient of the state for the case of acceleration is obtained.
Finally, the total Gradient is obtained by the fusion between the calculation
show for the quaternion case and the gradient omitted for the acceleration case.
100 B. B. Salmern-Quiroz, G. Villegas-Medina et al.
() = [ ] (32)
For the predictions process of , several tehcnique have been validated, for
purpose of simplicity, the prediction via spline is chosen. Cubic spline is a spline
constructed of piecewise third-order polynomials which pass through a set of n
control points.
Suppose we are n+1 data points ( , ) such that.
= 0 <. . < , Then the coefficients of the vector exists cubic
polynomials with coefficients , 0 3 such that the following hold.
() = () = 3=0[ ] [ +1 ] 0 1
( ) = 0 1
(+1 ) = +1 (+1 ) 0 2
(+1 ) = +1
(+1 ) 0 2
(+1 ) = +1
(+1 ) 0 2
So we see that the cubic spline not only interpolates the data ( , ) but
matches the first and second derivatives at the knots. Notice, from the above
definition, one is free to specify constrains on the endpoints. The end point
constrain () = 0 () = 0 is chosen.
7. Results
In this section, some simulation and results of the articulated arm (Figure 4) are
represented in order to show the performance of the proposed control laws via
quaternion. A rigid body with low momentum of inertia is taken as the
experimental system. In fact, the low momentum of inertia makes the system
vulnerable to high angular accelerations which prove the importance to apply the
control.
A Quaternion Approach in the Estimation of the Attitude 101
Figure 5. Estimation and prediction of the quaternion for the end effector.
102 B. B. Salmern-Quiroz, G. Villegas-Medina et al.
also provides the Euler angles. The methodology of estimation and prediction are
implemented in real-time using the LabView environment.
On the other hand, this system doesnt provide the acceleration of the body so
for validation we have done slowly movement and abrupt movement to appreciate
the effect of the acceleration in our method.
Conclusion
The first key to keep in mind from this chapter is that hybrid methodogyes of
attitude estimation via queternion, constitutes a viable alternative for improving
the overall performance and robustness of embedded attitude estimation systems
dealing with faulty sensor measurements.
By modeling the sensor fusion problem via queternion within the hybrid
systems framework, we are able to exploit the redundancy of information
emerging from the different sensors in order to perform real-time diagnosis of
their modes of operation, therefore allowing the attitude estimation system to
compensate for both methodogies and unmodeled faulty behavior.
Figure 9. Attitude measured from the first articulation.
Figure 10. Attitude measured from the end effect.
A Quaternion Approach in the Estimation of the Attitude 107
This chapter presents a new strategy for attitude estimation of possibly non-
symmetric rigid bodies. Two globally methods of calculation of the bodys
attitude are proposed, namely one methodology fussing data information with a
three-axis accelerometer, three magnetometer and three rate gyros mounted
orthogonally jointly, with prediction of the movement via cubic splines are
studied and simulated. Furthermore, the attitude estimation is independent of the
bodys inertia. The numerical simulations have shown the effectiveness of the
proposed methodologies and their robustness with respect to sensors noise and far
initial points. Moreover, their simplicity makes them suitable for embedded
implementation. This control estimation is tested in real application, consisting in
a set of ABB 6 Degrees of Freedom robot mounted in the laboratorys Motor.
This later is located in the laboratory: Departamento de Pruebas de Motores
Volkswagen Mxico.
Acknowledgments
The authors would like to thank to the Instituto Politcnico Nacional (I.P.N.),
SEPI ESIME Azcapotzalco, the Secretaria de Investigacin y Posgrado for the
projects SIP-IPN and Volkswagen Mxico for their support given in the process
of validation of the methodology of estimation and for the use of their laboratory
in VW-Mxico, Puebla.
References
[1] G. Whaba, A least squares estimate of spacecraft attitude, SIAM Review,
Vol. 7, No. 3, July 1965, p. 409.
[2] John L. Crassidis and F. Landis Markley. Minimum Model Error Approach
for Attitude Estimation, Journal of Guidance, Control, and Dynamics, Vol.
20, No. 6 (1997), pp. 1241-1247.
[3] Lefferts, E. J., F. L. Markley, and M. D. Shuster, Kalman Filtering for
Spacecraft Attitude Estimation, Journal of Guidance, Control, and
Dynamics, Vol. 5, No. 5, Sept.-Oct. 1982, pp. 417-429.
[4] M. D. Shuster, A survey of attitude representations, Journal of the
Astronautically Sciences (1993) 439-5177.
[5] Markley, F. L., New Quaternion Attitude Estimation Method, Journal of
Guidance, Control, and Dynamics, Vol. 17, No. 2, March-April 1994, pp.
407-409.
108 B. B. Salmern-Quiroz, G. Villegas-Medina et al.
[6] Pires, JN, and S da Costa JMG, A Real Time System for Position/Force
Control of Mechanical Manipulators, Proceedings of the 7th International
Machine Design Conference, Ankara, Turkey, 1996.
[7] Siciliano B., Villani L., Robot Force Control, Kluwer Academic Publishers
International Series in Engineering and Computer Science, Boston, MA,
1999.
[8] Thienel, J.; Sanner, R.M.; A coupled nonlinear spacecraft attitude
controller and observer with an unknown constant gyro bias and gyro
noise, Automatic Control, IEEE Transactions on, vol.48, no. 11, pp. 2011-
2015, Nov. 2003.
[9] R. L. Bellaire, E. W. Kamen and S. M. Zabin A new nonlinear iterated
filter with applications to target tracking, Proc. AeroSense: 8th Int. Symp.
Aerospace/Defense Sensing, Simulation and Controls, vol. 2561, pp. 240
1995.
[10] H. Rehbinder, X.Hu, Drif-free attitude estimation for accelerated rigid
bodies, Automatica (2004) 653-659.
[11] B.B Salmern-Quiroz, C.F. Mendez-Barrios, J.F. Guerrero-Castellanos, G.
Villegas-Medina, S.A. Rodrguez-Paredes. Towards Human Capture
Movement: Estimation of Anatomical Movements of the Shoulder,
Discrete Dynamics in Nature and Society, Vol. 2013, Article ID 586480, 13
pages, 2013. doi:10.1155/2013/586480., ISSN: 1026-0226.
[12] S. Lesecq, A. Barraud, Capture de mouvement: Report, October 2004-July
2005, GIPSA-lab - CEA-Leti, 2005, Grenoble, France.
[13] K. Xianwen and C. M. Gosselin, Type synthesis of 3T1R 4-DOF parallel
manipulators based on screw theory, IEEE Trans. Robot. Autom, vol. 20,
no. 2, pp. 181-190, Apr. 2004.
In: Quaternions: Theory and Applications ISBN: 978-1-53610-768-5
Editor: Sandra Griffin 2017 Nova Science Publishers, Inc.
Chapter 5
2D HERMITE-GAUSSIAN AND
GAUSS-LAGUERRE CIRCULAR HARMONIC
EIGENFUNCTION OF THE QUATERNION
AND REDUCED BIQUATERNION
FOURIER TRANSFORM
Abstract
The quaternions, reduced biquaternions (RBs) and their respective Fourier
transformations, i.e., discrete quaternion Fourier transform (DQFT) and discrete
reduced biquaternion Fourier transform (DRBQFT), are very useful for multi-
dimensional signal processing and analysis. In this paper, the basic concepts of
quaternion and RB algebra are reviewed, and we introduce the two dimensional
Hermite-Gaussian functions (2D-HGF) as the eigenfunction of DQFT/DRBQFT,
and the eigenvalues of 2D-HGF for three types of DQFT and two types of
DRBQFT. After that, the relation between 2D-HGF and Gauss-Laguerre circular
harmonic function (GLCHF) is given. From the aforementioned relation and
some derivations, the GLCHF can be proved as the eigenfunction of
1. Introduction
Properties of eigenvalues and eigenfunctions for Fourier transform and its variants
are widely surveyed in the literatures. In 1982, Dickinson and Steiglitz [1]
proposed a matrix which commutes with discrete Fourier transform (DFT) matrix
and used this commuting matrix to compute orthonormal eigenfunctions for DFT.
In the same year as [1], Grunbaum [2] discussed the Hermite function as
eigenfunction of DFT. Later in 2006, Pei et al. [3] proposed a nearly tridiagonal
commuting matrix to obtain orthonormal eigenfunctions for DFT with smaller
error than [1]. In 2008, Santhanam et al. [4] inspired by the ideas from quantum
mechanics in finite dimensions and presented an approach which computes
commuting matrix whose eigenvalue spectrum is closely approximated to that of
the Hermite-Gaussian differential operator. Recently in 2010, Pei et al. [5] derived
the eigenvalues of discrete 2D-HGF for two-side DQFT [6] and two-side
DRBQFT [7].
The quaternions have been applied to many research fields such as computer
science, mathematics, signal processing and image processing, since Hamilton
introduced the concept in 1843 [8]. The fundamental theorems are well developed,
and mathematical operations like Fourier transform, wavelet transform, convolution
of this four-dimensional, non-commutative algebra have been constructed maturely
[9-16]. The usefulness and effectiveness of quaternions in dealing with multi-
dimensional computations are demonstrated, especially those involving operations
of computer graphics and image processing, like three-dimensional rotations and
many other geometrical transformations. However, there are many other interesting
variants of quaternions like the biquaternions [17], the reduced biquaternions (RBs)
[18], and the quad-quaternions [19], which are very useful and possess special
properties and abilities that quaternions dont have.
These variants have also found several applications. Among these variants, in
particular, we concentrate on discussing the quaternion, RBs and derive the
eigenvalues and eigenfunctions for their DQFT/DRBFT. The multiplication rule of
quaternions is non-commutative, on the other hand, the multiplication rule of RBs is
commutative. The commutative property is one of the advantages of RBs over
quaternions. Due to the commutativity of multiplications, many operations, such as
Fourier transform, correlation, convolution, singular value decomposition (SVD), of
2D Hermite-Gaussian and Gauss-Laguerre Circular Harmonic 111
RB algebra are much simpler and more convenient to the users than the ones of
quaternion algebra. In [7], the efficient implementation of DRBQFT, convolution,
correlation, phase-only correlation and RB linear-time-invariant and symmetric
multichannel complex system are developed and they are much simpler than the
existing implementation of quaternions. The commutative property is important and
useful because the commutative DRBQFT is much simpler than non-commutative
DQFT. In [20], the SVD of RB matrices are introduced. Compared with the
quaternion matrix SVD, the complexity of the RB matrix SVD is reduced to a small
factor of one-fourth. From the above discoveries, we can tackle with many
arithmetic problems more efficiently in signal and image processing by using RBs.
The RBs also have their limitations. The algebra of RBs is not a division algebra,
and their geometric meaning is unfamiliar to most engineers. However, these
problems have almost no influence on signal and image processing applications. We
will briefly summarize the comparison of quaternions and RBs in the following
section. In this work, we derive the eigenvalues of 2D-HGF for three types of
DQFT and two types of DRBQFT, i.e., right-side, left-side and two-side for DQFT
and type1, type2 for DRBQFT. By applying the relation between 2D-HGF and
GLCHF mentioned in [21], we extended the conventional GLCHF in quaternion
and RB spaces. We found that the GLCHF is the eigenfunction of left-side and
right-side DQFT (type 1 and type2 DRBQFT), and the modified GLCHF is
eigenfunction of two-side DQFT (type 1 and type2 DRBQFT). Over all, the major
contribution of this paper are:
1) Three types of DQFT (right-side, left-side, and two-side) and two types
of DRBQFT (type1 and type2) are introduced.
2) 2D Hermite-Gaussian and Gauss-Laguerre circular harmonic
eigenfunctions and eigenvalues are derived for the above quaternion and
reduced biquaternion Fourier transforms.
3) Discrete 2D Gauss-Laguerre circular harmonic eigenfunctions can be
efficiently computed using the linear combination coefficients and 2D
discrete Hermite-Gaussian eigenfunctions. Both 2D eigenfunctions forms
a complete and orthonormal basis in the 2D plane.
4) DQFT and DRBQFT can be efficiently implemented using the
conventional 2D FFT.
5) Discrete 2D Gaussian-Laguerre circular harmonic functions are suitable
for circular pattern analysis and expansions of the color images.
6) Color image expansion, reconstruction, rotation invariant features, and
color shape matching are proposed and demonstrated using GLCHFs for
color image processing applications.
112 Soo-Chang Pei and Yu-Zhe Hsiao
2. Preliminaries
2.1. Quaternions
q qr qii q j j qk k (1)
i 2 j 2 k 2 1 , ij ji k , jk kj i , ki ik j (2)
q* qr qii q j j qk k (3)
2D Hermite-Gaussian and Gauss-Laguerre Circular Harmonic 113
The idea of reduced biquaternions (RBs) was suggested by Schtte and Wenzel [18]
in 1990. Other similar ideas can be found in the articles about bicomplex algebra
[22-24]. They [18] proposed to apply the RBs to the implementation of digital filter,
and demonstrated that fourth order real filter can be realized by means of a first
order RB filter. The RBs are another types of four-dimensional hypercomplex
numbers, and are also represented as the form of (1), but {1, i , j , k } obeys
different multiplication rules from those of (2). The rules are given by:
As (5) shows, the major difference between RBs and quaternions is the
multiplication rules. If we define the norm of the RBs as
q (qr2 qi2 q 2j qk2 )1/2 , then q1q2 q1 q2 , where q1 and q2 are two
arbitrary RB, and if we define the conjugate of the RBs as
q qr qii q j j qk k , then qq is still a RB. There are three different
* *
definitions of conjugate for the RBs in [2, 3], but the product of a RB and one of
these three conjugates are still not real, therefore, we define the modulus and
conjugate of RB as follows:
114 Soo-Chang Pei and Yu-Zhe Hsiao
qr qi qj qk qr qi qj qk
M q can be written as q qr qk q j , and qi qr qk qj =
i Mq
q j qk qr qi qj qk qr qi
qk qj qi qr qk qj qi qr
q* q 1 q / q
2
(7)
qq* q and (q1 q2 )* q1*q2* .If 0 , then the inverse of q (RB) and q do
2
not exist.
2D Hermite-Gaussian and Gauss-Laguerre Circular Harmonic 115
In [25], Davenport had found that there exists two special nonzero numbers e1
and e2 in HCA4 such that e1e2 0 , e1 e1
n 1
n
... e1 , and e2n e2n1 ... e2
. Therefore, e1 and e2 are both idempotent elements and divisors of zero. For
complex numbers and quaternions, the idempotent elements are only 0 and 1, and
the divisor of zero is only the number 0. For RBs, e1 (1 j ) / 2 , e2 (1 j ) / 2 .
Any RB can be represented by the linear combination of e1 and e2 :
about RBs, such as multiplication and Fourier transform, can be reduced by the
use of e1 and e2 , and the analysis about RBs become easier. For example, the
multiplication of two RBs q1 and q2 can be computed by following equation:
q1q2 q2 q1 (q1a q1b )(q2a q2b )e1 (q1a q1b )(q2a q2b )e2 (10)
where a=0,1,2, and () are a-th order Hermite polynomials [26]. (m,n) is
spatial location in Cartesian coordinate. The GLCHFs can be linearly combinated
by using 2D-HGFs, that is:
s
Ls t ,t (m, n) t(,sg) H s g , g (m, n) (12)
g 0
( s g )! g ! min( g ,mt ) k s t t
t(,sg) j g (1)
(13)
2s ( s t )!t ! k max(0, g t ) s t k t g k
2D Hermite-Gaussian and Gauss-Laguerre Circular Harmonic 117
L1=L50 =Re{L50 }+i Im{L50 }= 0.1768 H50 -0.3953i H41 -0.5590 H32
(14)
+ 0.5590i H23 +0.3953 H14 -0.1768i H05
We define three types of DQFT used to derive their eigenvalues as follows (only
foreword transforms are given):
Left-side DQFT (denote as L-DQFT):
M 1 N 1 2 ( um vn )
1
FQL (u, v) e
a
M N
f (m, n) (15)
MN m0 n 0
M 1 N 1 2a (
um vn
)
1
FQR (u, v)
MN
f (m, n)e
m0 n 0
M N (16)
M 1 N 1 2 um 2b
vn
1
FQT (u, v) e
a
M
f (m, n)e N (17)
MN m0 n 0
Figure 1. Fifth order 2D-HGFs in top row can be transformed to the corresponding
GLCHFs in rightmost columns (L1 to L6, as shown in real and imaginary parts) by using
linear combination with the coefficients in matrix form.
1 M 1 N 1 2 ( um vn
) 1 M 1 2 um 1 N 1 2 vnN
MN m0 n 0
e M N
H ab (m, n)
M m 0
e M H a ( m) e H b ( n)
N n 0
(18)
( )a H a (m)( )b H b (n) ( ) a b H ab (m, n) H ab (m, n)( ) a b
30 (, ) = 30 (, ) + 21 (, ) + 12 (, ) +
+ 03 (, ) = + ,
where the transformation axis are unit pure quaternion, (A,B,C,D) are the
coefficients of linear combination obtained from (13), the original imaginary axis
i in (13) is now extended to unit pure quaternion , R is real part, I is imaginary
part (R and I are also linear combination of 2D-HGFs). Therefore,
M 1 N 1 2 ( um vn ) M 1 N 1 2 ( um vn )
1 1
MN
e
m 0 n 0
M N
Lab (m, n)
MN
e
m 0 n 0
M N
(R I )
M 1 N 1 2 ( um vn ) M 1 N 1 2 ( um vn )
1 1
MN
e
m 0 n 0
M N
R
MN
e
m 0 n 0
M N
I
1 M 1 N 1 2 ( um vn
) 1 M 1 N 1 2 ( um vn )
(19)
MN m 0 n 0
e M N
R(
MN
e
m 0 n 0
M N
I )
( ) R ( ) I
a b a b
1 M 1 N 1 um vn
2 ( ) 1 M 1 2
um
1 N 1 2
vn
MN m0 n 0
H ab (m, n)e M N
M m 0
H a (m)e M
N n 0
H b ( n )e N (20)
H a (m)( ) a H b (n)( )b H ab (m, n)( ) a b ( ) a b H ab (m, n)
M 1 N 1 2 (
um vn
) M 1 N 1 2 (
um vn
)
1 1
MN
Lab (m, n)e
m 0 n 0
M N
MN
( R I )e
m 0 n 0
M N
M 1 N 1 2 (
um vn
) M 1 N 1 2 (
um vn
)
1 1
MN
Re
m 0 n 0
M N
MN
I e
m 0 n 0
M N
1 M 1 N 1 um vn
2 ( ) 1 M 1 N 1 2 (
um vn
) (21)
MN m 0 n 0
R e M N
(
MN
I e
m0 n 0
M N
)
R( ) I ( )
a b a b
From (21), we can see that the eigenvalue of GLCHF (, ) for right-side
QFTis ()+ .
For 2D-HGF (, ) and two transformation axes 1 and 2 are unit pure
quaternions:
1 M 1 N 1 21 um 22
vn
1 M 1 21 um 1 N 1 22
vn
MN m0 n 0
e M
H ab (m, n)e N
M m 0
e M
H a (m)
N n 0
H b ( n )e N
(22)
( 1 )a H a (m) H b (n)( 2 )b ( 1 ) a ( 2 )b H ab (m, n) H ab ( m, n)( 1 ) a ( 2 ) b
simple theorem is introduced first and it may be helpful for us to derive the
eigenvalues.
M 1 N 1 2 um 22
vn M 1 N 1 2 um 22
vn
1 1
e Lab (m, n)( 1 2 )e e ( R 1I )( 1 2 )e
1 1
M N M N
MN m0 n 0 MN m0 n 0
M 1 N 1 2 um 22
vn M 1 N 1 2 um 22
vn
1 1
e R1 e e R 2 e
1 1
M N M N
MN m 0 n 0 MN m 0 n 0
1 M 1 N 1 21 um 22
vn
1 M 1 N 1 21 um 22
vn
MN m0 n 0
e M
11 I e N
MN m0 n 0
e M
12 I e N
1 ( 1 ) a ( 2 )b R ( 1 ) a ( 2 )b 2 R 1 ( 1 ) a ( 2* )b 1 I 1 ( 1 ) a ( 2 ) b 2 I
R1 ( 1 ) a ( 2 )b 1 I 1 ( 1 ) a ( 2 )b R( 1 ) a ( 2 )b 2 1 I ( 1 ) a ( 2 )b 2
( R 1 I )( 1 ( 1 ) a ( 2 )b ( 1 ) a ( 2 )b 2 )
(25)
( R 1 I )( 1 ) a ( 1 2 )( 2 )b ( R 1 I )( 1 2 )( 2 ) a ( 2 )b
( ( 1 ) a ( 1 2 ) ( 1 2 )( 2 ) a (Thm 1))
( R 1 I )( 1 2 )( 2 ) a b
Lab (m, n)( 1 2 )( 2 ) a b
M 1 N 1 21
um
22
vn M 1 N 1 21
um
22
vn
1 1
MN
e
m0 n 0
M
( 1 2 ) Lab (m, n)e N
MN
e
m0 n 0
M
( 1 2 )( R 2 I )e N
M 1 N 1 21
um
22
vn M 1 N 1 21
um
22
vn
1 1
MN
e
m 0 n 0
M
1 R e N
MN
e
m 0 n 0
M
2 R e N
1 M 1 N 1 21 um 22
vn
1 M 1 N 1 21 um 22
vn
MN m0 n 0
e M
12 I e N
MN m0 n 0
e M
2 2 I e N
1 ( 1 ) a ( 2 )b R ( 1 ) a ( 2 )b 2 R 1 ( 1 ) a ( 2 )b 2 I 2 ( 1* ) a ( 2 ) b 2 I
1 ( 1 ) a ( 2 )b R 1 ( 1 ) a ( 2 )b 2 I ( 1 ) a ( 2 )b 2 R ( 1 ) a ( 2 )b 2 2 I
( 1 ( 1 ) a ( 2 )b ( 1 ) a ( 2 )b 2 )( R 2 I )
(26)
( 1 ) a ( 1 2 )( 2 )b ( R 2 I ) ( 1 ) a ( 1 )b ( 1 2 )( R 2 I )
( ( 1 ) n ( 1 2 ) ( 1 2 )( 2 ) n (Thm 1))
( 1 ) a b ( 1 2 )( R 2 I )
( 1 ) a b ( 1 2 ) Lab (m, n)
DRBFT of type 2 (denote as DRBFT II, one imaginary unit axis,12 = 1):
M 1 N 1 2 u1 (
pm sn
)
1
FRB 2 ( p, s)
MN
f (m, n)e
m 0 n 0
M N (28)
2D Hermite-Gaussian and Gauss-Laguerre Circular Harmonic 123
M 1 N 1 2 ( u1
pm sn
u2 )
1
MN
H
m 0 n 0
ab (m, n)e M N
1 M 1 2 u1
pm
1 N 1 2 u2
sn
(29)
M m 0
H a (m)e M
N n 0
H b ( n )e N
M 1 N 1 2 u1 (
pm sn
)
1
MN
H ab (m, n)e
m 0 n 0
M N
1 M 1 2 u1
pm
1 N 1 2 u1
sn
M m0
aH ( m )e M
b
N n0
H ( n ) e N (30)
where {, , }.
124 Soo-Chang Pei and Yu-Zhe Hsiao
M 1 N 1 2 ( u1
pm sn
u2 )
1
LRB1 ( p, s )
MN
L
m0 n0
ab (m, n)e M N
1 M 1 N 1 2 ( u1
pm sn
u2 ) 1 M 1 N 1 2 ( u1
pm sn
u2 )
1
MN m 0 n 0
H ( m, n )e M N
u 2
MN m 0 n 0
H ( m, n )e M N
(31)
(u1 ) a (u2 )b H1 (m, n) u (u1 ) a (u2 )b H 2 (m, n)
(u1 ) a (u2 )b ( H1 (m, n) u H 2 (m, n))
(u1 ) a (u2 )b ( H abR (m, n) u H abI (m, n))
(u1 ) a (u2 )b Lab (m, n)
(u1 )a (u2 )b .
M 1 N 1 2 u1 (
pm sn
) M 1 N 1 2 u1 (
pm sn
)
1 1
MN m0 n 0
H1 (m, n)e M N
u
MN
H
m0 n 0
2 (m, n)e M N
(u1 )a b .
Assume:
1 j 1 j
u1 i, u2 k , u j, e 1 ,e 2
2 2
pm sn
2 ( u1 u2 ) sn 2 u1 pm sn 2 u1 pm
e M N
cos(2 )e M
u (u1 ) sin(2 )e M
N N (33)
pm sn pm sn
2 u1 ( ) 2 u1 ( )
e M N
e1 e M N
e2
where
M 1 N 1 2 ( u1
pm sn
u2 )
1
LRB1 ( p, s)
MN
L
m0 n 0
ab (m, n)e M N
(36)
1 M 1 N 1 2 u1 (
pm sn
) 1 M 1 N 1 2 u1 (
pm sn
)
(
MN m0 n 0
H1 2 (m, n)e M N
) e1 (
MN
H12 (m, n)e
m 0 n 0
M N
) e2
H1 2 ( p, s) e1 H12 ( p, s) e2
H1 2 ( p, s) DFTu1 ( H1 2 (m, n)), H12 ( p, s ) DFTu1 ( H12 (m, n))
Then,
M 1 N 1 2 ( u1
pm sn
u3 )
1
LRB1 ( p, s )
MN
Lab (m, n) (u1 u3 )e
m0 n0
M N
M 1 N 1
1 2 ( u1
pm sn
u3 ) 1 M 1 N 1 2 ( u1
pm sn
u3 )
MN m 0 n 0
H f (m, n)e M N
u2
MN m 0 n 0
H g (m, n)e M N
(38)
(u1 ) (u3 ) H f (m, n) u2 (u1 ) (u3 ) H g (m, n)
a b a b
(u1 )a (u3 )b .
M 1 N 1 2 u (
pm sn
)
1
LRB 2 ( p, s )
MN
L
m 0 n 0
ab (m, n) (u1 u3 )e M N
M 1 N 1 2 u (
pm sn
) M 1 N 1 2 u (
pm sn
)
1 1
MN m 0 n 0
H f ( m, n )e M N
u2
MN
H g (m, n)e
m0 n 0
M N
(39)
(u ) a b H f (m, n) u2 ( ) a b H g (m, n)
(u ) a b ( H f (m, n) u2 H g (m, n))
(u ) a b ( H1 (m, n) u1 H 2 (m, n))(u1 u3 )
(u ) a b Lab (m, n) (u1 u3 )
where
4. Experimental Results
4.1. Verification of Derived Eigenvalue
First, we use Matlab, QTFM toolbox [27] developed by Sangwine et al. and our
reduced biquaternion toolbox to perform DQFT and DRBQFT for 2D-HGF,
GLCHF, MGLCHF I and MGLCHF II. The transformed spectrums of functions
mentioned above and the ratio of spectrum to original function multiplied by
derived eigenvalue, i.e., 1, will be demonstrated to verify the results summarized
in Table 3. For QFT, we verify all cases in Table 3. Because of the page limit, we
only verify 2D-HGF for DRBQFT II, and MGLCHF I&II for DRBQFT I. The
parameters and methods used in these experiments are summarized in Table 4 and
the experimental results are shown in Figure 4~13.
From the above experiments and discussions, we know that GLCHFs are the
eigenfunctions of DQFT and DRBQFT. Therefore, GLCHFs possess some useful
2D Hermite-Gaussian and Gauss-Laguerre Circular Harmonic 133
where (m, n) is the red color channel, (m, n) is the green color channel, and
(m, n) is the blue color channel of the color image, respectively. {, , } is the
quaternion basis introduced in (2) and { , , } is the RB basis mentioned in
(5). , is the Gauss-Laguerre circular harmonic (GL-CH) expansion
coefficient of order (s-t,t) for quaternion encoded color image and , (m,n) is
quaternion encoded 2D GLCHF basis of order (s-t,t). On the other hand, , is
the GL-CH expansion coefficient of order (s-t,t) for RB encoded color image and
Method 1.
For a fixed K, we increase L from zero to a predefined number (the increment
step d can be verified). The sum of order (a+b) satisfy the following constraint,
(a + b) =L2N+2 (41)
134 Soo-Chang Pei and Yu-Zhe Hsiao
| | <K (42)
From Figure 15, the partially reconstructed color images by using the sifted
expansion coefficients , , with method 1reveal the details of image
circularly and gradually. We cannot recognize the content of original image well
when the order (a + b = L) is low, but when the order (a + b = L) is high, e.g., L
= 100 ~ 130, we only use finite, small number of coefficients and GL-CH basis to
approximate original images. It can be seen that the approximated images are
visually pleasing with high fidelity.
Method 2.
For a fixed L, increasing K from zero to a predefined number (41) and (42)
are also satisfied. From Figure 16, the partially reconstructed color images by
using the sifted expansion coefficients , , with method 2 approximate
the support of the original image. We can recognize the content and support size
of original image even when the order (K) is low. As for K is high, the
approximations are slightly inferior to those of L is high, especially in the border
of the reconstructed images. This is because the main purpose of method 2 is to
obtain the support and roughly sketch of original images.
LQ1 I Q hQ (44)
1
L I
RB RB hRB
Figure 13. Eigenvalue verification of Figure 4~12. The transformed spectrums of functions
and the ratio of spectrum to original function multiplied by derived eigenvalue is 1.
Figure 14. The illustration of partial GL-CH coefficients used to reconstruct color image.
b
Figure 15. (a) Partially reconstructed quaternion encoded color image (Lena) by using ,
((a,b) satisfies (41) and (42)) and , (m,n) with method 1. (b) Partially reconstructed RB
encoded color image (Baboon) by using , and
, (m,n) with method 1. d is the
increment for L=a+b=0~130,K=65>N.
b
Figure 16. (a) Partially reconstructed quaternion encoded color image (Lena) by using ,
, (m,n)
((a,b) satisfies (41) and (42)) and with method 2. (b) Partially reconstructed RB
encoded color image (Peppers) by using , and
, (m,n) with method 2. |
| <K=0~23, L=2N+2=130.
Figure 22. Color patches that can be used to do color shape matching (Clipped From jersey
in Figure 18-21).
2. Use GLCHFs to approximate these patches and use those GLCHFs with
circular shape (the fourth row, fourth column one, marked by black
square) to perform the color matching task of human, below are examples
of decomposition of patches:
f (ms , ns )
3. Transform the interested color image and GLCHFs
approximated color patch h(ms , ns ) into I-H-S color space by:
2D Hermite-Gaussian and Gauss-Laguerre Circular Harmonic 143
1 1 1
3 f (m , n )
f I (ms , ns ) 3 3
1 1 2 R s s (45)
fV1 (ms , ns ) f ( m , n )
6
G s s
6 6
f B (ms , ns )
fV2 (ms , ns ) 1 1
2 0
2
1
fV2 (ms , ns )
f H (ms , ns ) tan ( ) f S (ms , ns ) fV21 (ms , ns ) fV22 (ms , ns )
fV1 (ms , ns )
We can use RB polar form to represent the color image and patch as follows:
M 1 N 1 M 1 N 1
Eh ( h(ms , ns ) )2
ms 0 ns 0
A (m , n )
ms 0 ns 0
2
h s s (47)
After normalizing by Eh , if input image matches the color patch, then the
correlation values at the matching positions are nearly equal to 1 (1 + 0i + 0j + 0k).
Figure 23. Decomposition of color patches by using GLCHFs and RBs, we use the one
marked by black square to perform color shape matching (the one with circular shape).
2D Hermite-Gaussian and Gauss-Laguerre Circular Harmonic 145
g (ms , ns ) IDRBFT F( RB ) ( p, s ) H c ,n ( RB ) ( p, s )
F( RB ) ( p, s) H c ,n ( RB ) ( p, s )
(50)
g (ms , ns ) IDRBFT
F( RB ) ( p, s) H c ,n ( RB ) ( p, s )
g (ms , ns ) f (ms , ns ) RB , PO h(ms , ns )
F ( p, s ) H c ( p, s ) (51)
IDRBFT ( )
F ( p, s ) H c ( p, s )
g (ms , ns ) f (ms , ns ) RB , PO h(ms , ns )
F ( p, s ) H c ( p, s ) (54)
IDRBFT ( )
F ( p, s ) H c ( p, s )
Using the result of the phase-only correlation we can find the positions of
human object that have similar shape as color patch. Because the shape of human
is nearly circular or ellipsoidal, therefore we can use GLCHFs circularly
146 Soo-Chang Pei and Yu-Zhe Hsiao
approximated color patch to do correlation and thus find out the locations of
human or children in our test target images.
g r (ms , ns )
1
g r (ms , ns ) gi (ms , ns ) g j (ms , ns ) g k (ms , ns )
g r (ms , ns ) g k (ms , ns )
2
g r (ms , ns ) gi (ms , ns ) g j (ms , ns ) g k (ms , ns )
g r (ms , ns ) gi (ms , ns )
3
g r (ms , ns ) gi (ms , ns ) g j (ms , ns ) g k (ms , ns )
where (r,i,j,k) are real part (r),i imaginary part (i), j imaginary part (j), and k
imaginary part (k) respectively.
2D Hermite-Gaussian and Gauss-Laguerre Circular Harmonic 147
Figure 24. Color shape matching results of two test input images. (marked by red square)
(a)-(e) color patches. (f)-(j) results of different color matching. (k)-(o) are original input
color images. As can be seen from these results, the location of human can be found and
we can separate human with different color clothes by performing the whole color shape
matching task.
148 Soo-Chang Pei and Yu-Zhe Hsiao
a b
c d
Figure 25. Color shape matching results. (marked by red square)(a) the matching result by using
white color patch. (b) the matching result by using black patch (c)-(d) original test color image.
a b
c d
Figure 26. Color matching results. (marked by red square)(a) the matching result by using blue
color patch. (b) the matching result by using white patch (c)-(d) original test color image.
2D Hermite-Gaussian and Gauss-Laguerre Circular Harmonic 149
8. Finally, we can find the location of human object and match the colors or
their clothes by eq.(55).
Bellow (see Figure 32-34) are results of the color matching by applying above
methods. We can see that the locations of human can be found and after
thresholding, we can successfully accomplish the color matching mask of their
clothes by using patches with different colors
Conclusion
In this work, we derive the eigenvalues of 2D-HGFs and GLCHFs for DQFT and
DRBQFT. The experimental results verify our derivations. Two partial
reconstruction methods of color image are proposed based on GLCHFs and the
reconstructed results demonstrate the efficacy and usefulness of our methods. We
also show that the expansion coefficients can be used as a rotation invariant
feature and also propose a novel method to perform color shape matching of
human jersey effectively.
References
[1] B.W. Dickinson and K. Steiglitz, Eigenvectors and functions of the
discrete Fourier transform, IEEE Trans. Acoust, Speech, Signal Process.,
vol. 30, pp. 25-31, 1982.
[2] F. A. Grunbaum, The eigenvectors of the discrete Fourier transform: A
version of the Hermite functions, J. Math. Anal. Appl., vol. 88, pp.355-363,
1982.
[3] S. C. Pei, W. L. Hsue, and J. J. Ding, Discrete fractional Fourier transform
based on new nearly tridiagonal commuting matrices, IEEETrans. Signal
Process., vol. 54, no.10, pp.3815-3828, 2006.
[4] B. Santhanam and T. S. Santhanam, On discrete Gauss Hermite functions
and eigenvectors of thediscrete Fourier transform, J. Signal. Process., vol.
88, pp.2738-2746, Nov. 2008.
[5] S. C. Pei, J. J. Ding, K. W. Chang, Eigenfunctions, eigenvalues, and
fractionalization of the quaternion and biquternion Fourier transforms,
EUSIPCO 2010, pp.1874-1878, Aug. 2010.
[6] S. Sangwine, The discrete quaternion Fourier transform, IEE Conf., vol.
2, pp. 790-793, 1997.
150 Soo-Chang Pei and Yu-Zhe Hsiao
Chapter 6
Abstract
The unit quaternion system was invented in 1843 by William Rowan
Hamilton as an extension to the complex number to find an answer to
the question (how to multiply triplets?). Yet, quaternions are extensively
used to represent the attitude of a rigid body such as quadrotors, which is
able to alleviate the singularity problem caused by the Euler angles rep-
resentation. The singularity is in general a point at which a given mathe-
matical object is not defined and it outcome of the so called gimbal lock.
The singularity is occur when the pitch angles rotation is = 90o .
In this chapter, a leader-follower formation control problem of quadro-
tors is investigated. The quadrotor dynamic model is represented by unit
quaternion with the consideration of external disturbance. Three differ-
ent control techniques are proposed for both the leader and the follower
robots. First, a nonlinear H design approach is derived by solving a
Hamilton-Jacobi inequality following from a result for general nonlinear
E-mail address: wmj r@yahoo.com.
E-mail address: dgu@essex.ac.uk
154 Wesam Jasim and Dongbing Gu
1. Introduction
In the last decade, the focus on control single unit quadrotors has expanded
to controlling a team of quadrotors for these to be able to achieve their tasks
in variable weather and complicated environments. Team formation flight also
provides advantages over the use of an individual quadrotor in both civil and
military applications, such as inspection of an inaccessible area, disaster man-
agement, and search and rescue in risky circumstances, etc. Most of these
applications demand more than one quadrotor to accomplish the desired ob-
jective [1, 2]. The leader-follower approach is one of the main approaches of
formation control design.
Distributed and decentralised control techniques were used in the literature
to solve the leader-follower control problem. The distributed control technique
assumes that not all followers receive the leaders information and there is a kind
of cooperation among them [314], while the decentralised control technique
proposes that all followers are able to receive the leaders information [1524].
Different controllers have been implemented with both distributed and decen-
tralised control techniques.
tion controllers with its stability analysis. In section 6, presents the derivation
of iLQR controller for the leader and the follower. Section 7 shows the per-
formance of the presented controllers while the conclusions of this study are
indicated in section 8.
2. Mathematical Model
To control the motion and rotation of the quadrotor UAV, the mathematical dy-
namic model should be achieved. The quadrotor UAV system has a nonlinear
dynamic system and complicated structure; therefore, it is difficult to represent
its motion and rotation in a simple model. The dynamic model of the quadrotor
UAV depends on some assumptions [30]:
The structure of the quadrotor is rigid and symmetrical;
The propellers are rigid;
The centre of mass and body fixed frame are coincides;
Thrust and drugs are proportional to the square of the propellers; and
The difference of gravity by altitude or the spin of the earth is minor.
According to these assumptions, the mathematical model can be derived to
perform the quadrotor UAV fuselage dynamics in space, where it will be easy to
add to it the effects of aerodynamic forces generated by the rotation of the pro-
pellers. The coordinate reference system of the quadrotor includes two frames
of reference, the inertial (earth fixed) frame mentioned I(xI , yI , zI ) and the
rigid (body fixed) frame mentioned B(xB , yB , zB ). Several techniques can be
used to perform the rigid body rotation in space such as Euler angles, Quater-
nions and Tait-Bryan angles [8]. The method used to describe the position and
orientation of the quadrotor is the quaternion method. It is a hyper complex
number of 4-tuple (q0 , q1 , q2 , q3 ) R4 which can be written in many ways as
Q = q0 + q1 i + q2 j + q3 k and Q = [q0 , qT ]T [31] [32] [33].
The north east down (NED) coordinate system is used to parametrise the
dynamic model of the quadrotor with an angle of one-axis rotation around the
Euler axis of unit vector k R3 which has a direct physical connection and can
be written as:
cos 2
Q= (1)
k sin 2
The Quaternions with an Application of Quadrotors Team Formation 159
q
where k = kqk and = 2 arccos q0 . Moreover, as any complex number the
norm, complex conjugate and inverse of the quaternion can be defined as:
q
kQk = q02 + q12 + q22 + q32 (2)
q0
= q1
Q q2 (3)
q3
Q
Q1 = . (4)
kQk
The unit quaternion can be used to represent the coordinate transforma-
tion between the inertial frame I and the body frame B by defining the mul-
tiplication and the inverse quaternion. The multiplication of two quaternions
Q = [q0 , qT ]T and Q = [q 0 , q T ]T is defined as:
qT
q0 q0
Q Q =
q q0 I + S(q) q
q0 q 0 q T q
= .
q 0 q + q0 q + S(q)q
qT
q0 q0
QQ =
q q0 I + S(q) q
q0 q 0 q T q
= (5)
q 0 q + q0 q + S(q)q
160 Wesam Jasim and Dongbing Gu
where k R3 is a vector to be rotated from the fixed frame to the body frame
and
2
q0 + q12 q22 q32
2(q1 q2 q0 q3 ) 2(q1 q3 + q0 q2 )
Rq = 2(q1 q2 + q0 q3 ) q02 q12 + q22 q32 2(q2 q3 q0 q1 ) ;
2(q1 q3 q0 q2 ) 2(q2 q3 + q0 q1 ) q0 q12 q22 + q32
2
(12)
that is k = Rq k and k = Rq k .
T
Q(t + t) Q(t) 1
Q = lim = S()Q (17)
t0 t 2
where
0 z y x
z 0 x y
SS () =
y x
. (18)
0 z
x y z 0
Then the time derivative of the quaternion kinematics can be written in the
following two forms:
1 0 1 0
Q= Q = Q. (19)
2 2
dv
m = f. (20)
dtI
Applying the Coriolis equation to (20) we have
dv dv
m = m( + B/I v) = f. (21)
dtI dtI
Applying Equation (21) in body coordinates with vB = (u, , w)T and B/I
B =
T
(x , y , z ) it will be:
u x u fx
m = m(0 + y ) = fy (22)
w z w fz
The Quaternions with an Application of Quadrotors Team Formation 163
or
u z y w fx
= x w z u + 1 fy . (23)
m
w y u x fz
dh dh
= + B/I h = m. (25)
dtI dtB
From the body coordinate we have hB = JB/I B , then Equation (25) can be
resolved in the body coordinate frame. The equations of motion of the quadrotor
UAVs depend on the two frames which can be written as in [36].
Jx 0 0 x x Jx 0 0 x
0 Jy 0 y = 0+ y 0 Jy 0 y =
0 0 Jz z z 0 0 Jz z
(26)
or
1
0 0
x Jx y z (Jy Jz )
1
y = 0 Jy 0 ( x z (Jx Jz ) + ). (27)
z 0 0 J1 z
x y (Jx Jy )
or
y z (Jy Jz )
x Jx Jx
y = x z (Jx Jz )
+ . (28)
Jy Jy
z x y (Jx Jy )
Jz
Jz
The relationship between position and velocities is given by
x u
d
y = RT . (29)
dt
z w
164 Wesam Jasim and Dongbing Gu
Then the full model for the quadrotor kinematics and dynamics can be sum-
marised as follows:
x 0 2(q1 q3 + q0 q2 )
y = 0 + 2(q2 q3 q0 q1 ) f (35)
m
z g q02 q12 q22 + q32
The Quaternions with an Application of Quadrotors Team Formation 165
q0 q0 q1 q2 q3 0
q1 1 q1 q0 q3 q2 x
q2 = 2 q2 q3 (36)
q0 q1 y
q3 q3 q2 q1 q0 z
y z (Jy Jz ) q1
x Jx J
y = x z (Jx Jz ) qx2
Jy +
Jy . (37)
q3
z (J
x y x J y )
J
Jz z
p i = vi
fi
v
= ge + m Re
i i i
qTi i
qi0 1 (39)
=2
q i (qi0 I + S(qi )) i
Ji i = S(i )Ji i G(i ) + i
166 Wesam Jasim and Dongbing Gu
where i is L for the leader and F for the follower, mi is the quadrotor mass,
i = [ix , iy , iz ]T is the angular velocity in the body frame, Ji is the 3 3
diagonal matrix representing three inertial moments in the body frame, G(i )
represents the gyroscopic effect, i is the torque vector applied on the quadrotor,
the unit quaternion [qi0 , qi1 , qi2 , qi3 ]T = [qi0 , qTi ]T where qi = [qi1 , qi2 , qi3 ]T is
the vector part and qi0 is the scalar part of the quaternion, vi = [vix , viy , viz ]T
is the linear velocity, pi = [xi , yi , zi ]T is the position vector, the vector e =
[0, 0, 1]T , and I is the 3 3 unit matrix. The rotation matrix Ri is related to the
unit quaternion through the Rodrigues formula:
Now, the formation control problem for the follower is to satisfy the follow-
ing conditions:
limt (pF d pF ) = 0
limt (qL0 qF 0 ) = 0 (40)
limt (qL3 qF 3 ) = 0
The leader just tracks a desired trajectory represented by (pLd , qL0d , qL3d ).
So, the formation control problem for the leader is to satisfy the following con-
ditions:
limt (pLd pL ) = 0
limt (qL0d qL0 ) = 0 (41)
limt (qL3d qL3 ) = 0
4. Formation H Controllers
The controller design for the leader and the follower is based on H suboptimal
control. The follower H controller is designed by following the introduction
of an error state model, and the introduction of a H control theorem for general
affine systems. Then the leader H controller is briefly presented later.
168 Wesam Jasim and Dongbing Gu
F = pF d pF
p
v F = vF d vF
qF 0 qF 0d qF 0
=
F
q qF d qF
F = F d F
for all T 0 and all d L2 (0, T ) with initial condition x(0), where 0
K(x) < and K(x ) = 0.
For the nonlinear system (44) and > 0, define the Hamiltonian
H (x, V (x)) as below:
T
V (x) 1 V (x) 1 T T V (x)
H (x, V (x)) = f (x) + k(x)k (x) g(x)g (x)
x 2 x 2 x
1
+ hT (x)h(x) (47)
2
T V (x)
u = g T (x) (49)
x
has L2 -gain less than or equal to , and K(x) = 2V (x).
The nonlinear system (44) is called zero-state observable if for any trajec-
tory x(t) such that y(t) = 0, u(t) = 0, d(t) = 0 implies x(t) = x .
V (xF )
F + KF p v
= [CF p p F 2CF q JF KF q F
xF
F + KF v v
KF p p F + JF KF
F JF KF q q F ]
T V (xF )
uF = g T (xF )
xF
F + KF v v
KF p p F
= (51)
F + KF
KF q q F
172 Wesam Jasim and Dongbing Gu
The following weighting matrices are chosen with diagonal matrices WF 1 > 0,
WF 2 > 0, WF 3 > 0 and WF 4 > 0.
TF
h(xF ) = [ WF 1 p TF
WF 2 q FT
WF 3 v FT ]T
WF 4
which satisfies h(xF ) = 0, where the equilibrium point xF = [013 , 1, 013 ,
013 , 013 ]T . And we know
V (xF ) = 0 (52)
Now the team formation problem of the quadrotors under the disturbance dF is
defined below.
Problem 2. Given the equilibrium point xF , find the parameters KF p , KF q ,
KF v , KF , CF p , CF q in order to enable the closed-loop system (43) with the
above controller uF (51) to have L2 -gain less than or equal to F .
Next we want to show our main result in the following theorem.
Theorem 2. If the following conditions are satisfied, the closed-loop system
(43) with the above controller uF (51) has L2 -gain less than or equal to F .
And the closed loop system (43), (51) with dF = 0 is asymptotically locally
stable for the equilibrium point xF .
CF p CF q 0
CF p KF v KF2 p
CF p CF q KF v KF CF p JF KF2 q KF v JF KF2 q KF2 p + CF q KF2 p KF
1
C F p = KF p KF v 1 2
F
1
C F q = KF q KF 1
F2
F2 kWF 1 k
kKF p k2 (53)
F2 1
F2 kWF 2 k
kKF q k2 (54)
F2 1
F2 (kWF 3 k + 2kKF p k)
kKF v k2 (55)
F2 1
2 F2 (kWF 4 k 3kJF kkKF q k)
kKF k (56)
F2 1
kWF 1 k > 0; kWF 2 k > 0; kWF 3 k > 0; kWF 4 k > 0
The Quaternions with an Application of Quadrotors Team Formation 173
Proof. With the given conditions, we need to show (1) V (xF ) 0 and (2) the
Hamiltonian HF (xF , V (xF )) 0. Then the first part of the theorem can be
proved by using Theorem 1.
(1) Since
TF q
2(1 qF 0 ) = (1 qF 0 )2 + q F
TF q
q F
then
CF p I 033 KF p 033 F
p
1 T 033 CF q I 033 JF K F q F
q
V (xF ) pF TF
q v
FT
FT
2 KF p 033 KF v 033 v F
033 JF K F q 033 JF K F F
CF p CF q 0
CF p KF v KF2 p
CF p CF q KF v KF CF p JF KF2 q KF v JF KF2 q KF2 p + CF q KF2 p KF
(2)
1 T
TF CF p v
HF (xF , V (xF )) = p TF CF q
F q F + v FT KF p v
F JF KF q ( qF 0 I
2 F
+ S( qF ))
F + q TF KF q S(
F )JF F + FT KF S(F )JF
F
1 1 1 1
+ 2 1 kKF p p F + KF v v F k2 + 1 kKF q q F
2 F 2 F2
1 1
+ KF F k2 + kWF 1 kk pF k2 + kWF 2 kk qF k2
2 2
1 1
+ kWF 3 kk vF k2 + kWF 4 kk F k2
2 2
By choosing
1
CF p = KF p KF v 1 2
F
1
CF q = KF q KF 1
F2
174 Wesam Jasim and Dongbing Gu
then
1 T
FT KF p v
HF (xF , V (xF )) = v F + q TF KF q S(F )JF F JF KF q (
qF 0 I
2 F
1 1
+ S( qF ))
F + FT KF S( F )JF F + 1
2 F2
2 2 2 2 1 1
(kKF p k k pF k + kKF v k k vF k ) + 1
2 F2
1
(kKF q k2 kqF k2 + kKF k2 k F k2 ) + kWF 1 kkpF k2
2
1 1 1
+ kWF 2 kk qF k2 + kWF 3 kk vF k2 + kWF 4 kkF k2
2 2 2
By using kS( F ))k = k F k, |vF T KF p v vF k2 , k(
F | kKF p kk qF 0 I +
qF ))k
S(
3, | FT JF KF q (
qF 0 I + S( qF ))F | kKF q kkJF kk F k2 k(
qF 0 I +
qF ))k,
S(
TF KF q S(
q F )JF F = 0 and FT KF S(F )JF F = 0 we have
3 1 1
HF (xF , V (xF )) kKF q kkJF kkF k2 + kKF p kk vF k 2 + 1
2 2 F2
1 1
(kKF p k2 k
pF k2 + kKF v k2 k
vF k 2 ) + 1 (kKF q k2 k
qF k 2
2 F2
1 1
+ kKF k2 k
F k2 ) + kWF 1 kk pF k2 + kWF 2 kk qF k 2
2 2
1 1
+ kWF 3 kk vF k2 + kWF 4 kk F k 2
2 2
1 1 1
1 kKF p k2 + kWF 1 k 0
2 F2 2
1 1 1
1 kKF q k2 + kWF 2 k 0
2 F2 2
1 1 1
kKF p k + 2 1 kKF v k2 + kWF 3 k 0
2 F 2
3 1 1 1
kJF kkKF q k + 2 1 kKF k2 + kWF 4 k 0
2 2 F 2
The Quaternions with an Application of Quadrotors Team Formation 175
i.e.
F2 kWF 1 k
kKF p k2
F2 1
F2 kWF 2 k
kKF q k2
F2 1
F2 (kWF 3 k + 2kKF p k)
kKF v k2
F2 1
2 F2 (kWF 4 k 3kJF kkKF p k)
kKF k
F2 1
Then the total force and the torque vector applied to the follower, fF and F
are obtained,
F + KF
F =KF q q F + G(
F )
176 Wesam Jasim and Dongbing Gu
where
R11 R12 R13
T = R21 R22 R23
R L
R31 R32 R33
L = pLd pL
p
L = vLd vL
v
qL0 qL0d qL0
=
L
q qLd qL
L = Ld L
where qL0d , qLd , vLd , Ld are assumed to be constant for the desired tracking
trajectory. Then equation (39) can be rewritten in an error form as:
L = v
p L
fL
L = ge m
v R e
L L
qL0 TL
1 q L (57)
= 2 ( + ))
q L q L0 I S(
q L
L
L = JL1 S(
L + JL1 G(
L )JL L ) JL1 L
Let
L
p
qL0
xL =
L
q
L
v
L
" #
fL
ge m RL e
uL = L
L ) L
G(
The Quaternions with an Application of Quadrotors Team Formation 177
The nonlinear dynamic system (57) with the disturbance vector dL can be
written into an affine nonlinear form:
where
L
v
1 T
1 L
2q L
f (xL ) = 2 (
qL0 I + S(qL ))
L
031
1
JL S( L )JL
L
033 033
013 013
g(xL ) = k(xL ) = 033 033
I 033
033 JL1
from our previous work [39] and its main result is just presented in this section.
They will be used in the simulation later for evaluating the robustness of H
controllers.
Then the position tracking error between the leader and the follower can be
calculated as
cos cos
T
F = p F d p F = p L RL
p d cos sin pF (60)
sin
F = p F d p F = p F d vF
p (61)
where vF is a virtual control, and its desirable value can be described as:
vFd = p F d + bF p
F + kF p
F (62)
The Quaternions with an Application of Quadrotors Team Formation 179
R
where bF and kF are two positive matrices, p F = p F dt is the integral of the
follower position error and added to minimize the steady-state error.
Now, consider the linear velocity error between the leader and the follower
as:
F = vFd p F
v (63)
F = p F d + bF p
v F + kF p
F p F (64)
F = p
v F + kF p
F d + bF p F p
F (65)
Then from (62) and (63) we can rewrite (61) in terms of linear velocity error as:
F = v
p F bF p
F kF p
F (66)
By substituting (59) and (66) into (65), the time derivative of linear velocity
error can be rewritten as:
F =
v p F d + bF v F bF kF p
F b2F p F + kF p
F f (pF ) g(pF )fF (67)
The desirable time derivative of the linear velocity error is supposed to be:
F = cF v
v F p
F (68)
where cF is a positive diagonal matrix. Now, the total thrust fF , the longitudinal
uF x and lateral uF y motion control can be found by subtracting (67) from (68)
as follows:
fF =(g + v Lz + (1 b2F z + kF z )
zF + (bF z + cF z )vF z bF z kF z zF d(R31 cos cos
mF
+ R32 cos sin + R33 sin )) (69)
(qF 0 2 qF 1 2 qF 2 2 + qF 3 2 )
uF x =(v Lx + (1 b2F x + kF x )
xF + (bF x + cF x )
vF x bF x kF x x
F d(R11 cos cos
mF
+ R12 cos sin + R13 sin )) (70)
fF
uF y =(v Ly + (1 b2F y + kF y )
yF + (bF y + cF y )
vF y bF y kF y yF d(R21 cos cos
mF
+ R22 cos sin + R23 sin )) (71)
fF
180 Wesam Jasim and Dongbing Gu
F + KF
F = KF q q F + G(
F )
The attitude stability for the follower was demonstrated in [39] . Next, we show
the stability of followers translational part.
V = p F + v
TF p F + kF p
FT v TF p
F (73)
F = p
By substituting p F , and equations (66) and (68) into (73), equation (73)
becomes:
V = bF p
TF p FT v
F cF v F 0 (74)
Finally, (74) is less than zero provided bF and cF are positive diagonal ma-
trices, i.e. V < 0, ( F ) 6= 0 and V (0) = 0. It can be concluded from the
pF , v
positive definition of V and applying LaSalle theorem that a global asymptotic
stability is guaranteed. This leads to the conclude that limt p F = 0 and
limt vF = 0, which meets the position condition of (40).
xk+1 = Ak xk + Bk uk (80)
The following Hamiltonian function is a first step to proceed towards the optimal
control
k = Sk xk + k (87)
for some unknown sequences Sk and k . The boundary conditions for Sk and
k are
SN = QN
(88)
N = QN (xN x )
and from the boundary condition (85), uk is obtained as:
where
K = (BkT Sk+1 Bk + R)1 BkT Sk+1 Ak (91)
K = (BkT Sk+1 Bk + R) 1
BkT (92)
Ku = (BkT Sk+1 Bk + R)1 R. (93)
Backward recursion equations are used to solve the entire sequences Sk and k
as:
Sk = ATk Sk+1 (Ak Bk K) + Q (94)
184 Wesam Jasim and Dongbing Gu
where
zLkd zLk
zLk =
vLzkd vLzk
The Quaternions with an Application of Quadrotors Team Formation 185
qLkd qLk
qLk =
Lkd Lk
and the follower control law set is
fF k = KF z zF k KF z zF k+1 Kf F fF k
F qk = KF q qF k KF q qF k+1 K qF F qk
f = (g + v d(R cos cos + R cos sin
Fk Lz 31 32
mF
+R 33 sin )) 2 2 2 2
qF 0k qF 1k qF 2k +qF 3k
=f
f F k + fF k
F k =
F qk F qk + F qk
where
zF kd zF k
zF k =
vF zkd vF zk
qLk qF k
qF k =
F kd F k
7. Simulations
The proposed H , IBS and iLQR controllers were tested in a MATLAB simu-
lator of two quadrotors, one leader and one follower. The quadrotor parameters
used in the simulation are described in Table 1. Same path was presented in
the simulation to show the performance of using the proposed three controllers.
The desired path to be tracked by the leader was
xLd = 2 cos(t/80) ; yLd = 2 sin(t/80)
zLd = 1 + 0.1t ; qL3d = 0
with the initial conditions pL = [2, 0, 0]T metres and [qL0 , qTL ]T =
[1, 0, 0, 0]T . The follower tried to maintain the desired distance with the leader
d = 2 metres, the desired incidence angle = 0 and the desired bearing an-
gle = /12. The initial condition of the follower was pF = [0.1, 0.5, 0]T
metres and [qF 0 , qTF ]T = [1, 0, 0, 0]T .
7.1. H Controller
To test the robustness of the proposed H controller, the model parameter un-
certainties (mass and inertia) were increased and decreased by 20% and a
186 Wesam Jasim and Dongbing Gu
Table 1. Quadrotor Parameters
5 Nominal
Disturbanced
4
+20% Mass
z (m)
-20% Mass
3
0
3
2 3
2
1 1
0 0
-1
y (m) -1 -2 x (m)
The obtained results are shown in Figures 3 - 6 with the conditions (1) no
disturbance, (2) force disturbance dviz = 2Nm at 10 t 10.25 seconds,
dvix = 2Nm at 20 t 20.25 seconds, dviy = 2Nm at 30 t 30.25
seconds and the attitude part for the leader and the follower is disturbed using
(97), (3) +20% model parameter uncertainty, and (4) 20% model parameter
uncertainty. The above conditions were applied for the leader and the follower
at the same time.
Figure 3 shows the formation trajectories of two quadrotors obtained using
the H controllers when they tracked the desired path. From this figure we can
see that the H controllers produced good formation performances with small
acceptable errors, fast rejection of the external disturbances, and quick recovery
of the model parameter uncertainties. The quaternions of the leader and the
follower are shown in Figures 4 and 5 with small oscillations. The distances
between the leader and the follower for are shown in Figure 6. Again, less
oscillation in disturbance rejection was observed from the result.
Figure 7 shows the performance of the two quadrotors when only the leader
188 Wesam Jasim and Dongbing Gu
-0.996 0.1
Nominal
Disturbanced
-0.997
+20% Mass 0.05
-20% Mass
q0
q1
-0.998
0
-0.999
-1 -0.05
0 10 20 30 40 0 10 20 30 40
seconds seconds
0.05 0.05
0
q2
q3
0
-0.05
-0.1 -0.05
0 10 20 30 40 0 10 20 30 40
seconds seconds
-0.9985 0.04
Nominal
Disturbanced
-0.999 +20% Mass 0.02
-20% Mass
q0
q1
-0.9995 0
-1 -0.02
0 10 20 30 40 0 10 20 30 40
seconds seconds
0.02 0.03
0 0.02
q2
q3
-0.02 0.01
-0.04 0
-0.06 -0.01
0 10 20 30 40 0 10 20 30 40
seconds seconds
2.2
Nominal
Disturbanced
2.1 +20% Mass
-20% Mass
Distance (m) 2
1.9
1.8
1.7
1.6
1.5
0 5 10 15 20 25 30 35 40
Time (sec)
Figure 6. The Distance Between the Leader and the Follower under H Con-
troller.
5 Nominal
4 Disturbanced
z (m)
0
3
2 3
2
1 1
0 0
1
y (m) 1 2
x (m)
5 Nominal
Disturbanced
4
+20% Mass
z (m)
3 20% Mass
0
3
2 3
2
1 1
0 0
1
y (m) 1 2
x (m)
0.992 0.1
Nominal
0.994 Disturbanced 0.05
+20% Mass
20% Mass
q0
q1
0.996 0
0.998 0.05
1 0.1
0 10 20 30 40 0 10 20 30 40
seconds seconds
0.1 0.15
0.05 0.1
q2
q3
0 0.05
0.05 0
0.1 0.05
0 10 20 30 40 0 10 20 30 40
seconds seconds
1 1
Nominal
0.5 Disturbanced 0.5
+20% Mass
q0
q1
0 20% Mass 0
0.5 0.5
1 1
0 10 20 30 40 0 10 20 30 40
seconds seconds
1 1
0.5 0.5
q2
q3
0 0
0.5 0.5
1 1
0 10 20 30 40 0 10 20 30 40
seconds seconds
Figure 10. Follower Quaternions in First Path under IBS Controller Based on
Quaternion Representation.
2.2
Nominal
Disturbanced
2.15
+20% Mass
-20% Mass
2.1
2.05
Distance (m)
1.95
1.9
1.85
1.8
0 5 10 15 20 25 30 35 40
Time (sec)
Figure 11. The Distance between the Leader and the Follower under IBS Con-
troller.
The Quaternions with an Application of Quadrotors Team Formation 193
5
Nominal
4 Disturbanced
z (m)
0
3
2 3
2
1 1
0 0
1
y (m) 1 2
x (m)
Figure 12. Leader-Follower Formation in First Path under IBS Controller Based
on Quaternion Representation with Leader Disturbance Only.
5
Desired
4 iLQR (1 iteration)
iLQR (5 iterations)
z (m)
3 LQR
0
3
2 3
2
1 1
0 0
1
y (m) 1 2
x (m)
Figure 13. Leader-Follower Formation in First Path under iLQR and LQR Con-
trollers.
194 Wesam Jasim and Dongbing Gu
0.01
iLQR (1 iteration)
0.9998 iLQR (5 iterations)
LQR 0
q0
q1
0.9999
0.01
0.9999
1 0.02
0 10 20 30 40 0 10 20 30 40
sec sec
3
x 10
0.01 2
0.005
0
q2
q3
0
2
0.005
0.01 4
0 10 20 30 40 0 10 20 30 40
sec sec
Figure 14. Leader Quaternions in First Path under iLQR and LQR Controllers.
0.01
iLQR (1 iteration)
0.9998 iLQR (5 iterations)
LQR 0
q0
q1
0.9999
0.01
0.9999
1 0.02
0 10 20 30 40 0 10 20 30 40
sec sec
3 3
x 10 x 10
10 2
0
5
q2
q3
2
0
4
5 6
0 10 20 30 40 0 10 20 30 40
sec sec
Figure 15. Follower Quaternions in First Path under iLQR and LQR Controllers.
the follower in tracking the desired path are shown in Figures 14 and 15, re-
spectively. Figure 16 shows the distances between the leader and the follower.
The Quaternions with an Application of Quadrotors Team Formation 195
2.02
iLQR (1 iteration)
iLQR (5 iterations)
2.01
LQR
1.99
Distance (m)
1.98
1.97
1.96
1.95
1.94
1.93
0 5 10 15 20 25 30 35 40
Time (sec)
Figure 16. The Distance between the Leader and the Follower under iLQR and
LQR Controllers.
The error in using the iLQR controller was smaller than that in using LQR.
However, when the iLQR controller ran for five iterations, the response was
slightly improved.
In conclusion, it is obvious that the proposed iLQR controller maintained
the distance between the leader and the follower faster than LQR controller.
Conclusion
This chapter has presented the performance of applying the H , IBS and iLQR
controllers to the leader-follower formation control problem of quadrotors when
its dynamic model was derived based on unit quaternion. The H controller
was developed to reject the external disturbances and recover the model param-
eter uncertainties change. Then, its stability and robustness were analysed and
a set of corresponding conditions were given.
The IBS controller was developed based on BS control theory with adding
an integral action to minimise the steady state error which appeared when the BS
controller was used for leader-follower formation problem. The main drawback
of the IBS controller is that its stability is guaranteed but the performance is not,
196 Wesam Jasim and Dongbing Gu
References
[1] S. Islam, P. Liu, and A. El Saddik, Robust control of four-rotor unmanned
aerial vehicle with disturbance uncertainty, Industrial Electronics, IEEE
Transactions on, vol. 62, pp. 15631571, March 2015.
[2] J. Ferruz, V. M. Vega, A. Ollero, and V. Blanco, Reconfigurable control
architecture for distributed systems in the hero autonomous helicopter,
Industrial Electronics, IEEE Transactions on, vol. 58, no. 12, pp. 5311
5318, December 2011.
[3] T. Dierks and S. Jagannathan, Neural network control of quadrotor UAV
formations, in American Control Conference ACC 09, (Hyatt Regency
Riverfront, St. Louis, MO, USA), pp. 29902996, 2009.
The Quaternions with an Application of Quadrotors Team Formation 197
[4] G. Damien, Unmanned aerial vehicle formation flight using sliding mode
disturbance observers aerial vehicles, in Thanh Mung Lam (Ed.),ISBN:
978-953-7619-41-1, InTech, 2009.
[6] U. Pilz, A. Popov, and H. Werner, Robust controller design for formation
flight of quad-rotor helicopters, in Joint 48th IEEE Conference on De-
cision and Control and 28th Chinese Control Conference, (Shanghai, P.R.
China), pp. 83228327, 2009.
[11] M. Saska, Z. Kasl, and L. Preucil, Motion planing and control of forma-
tions of micro aerial vehicles, in Preprints of the 19th World Congress,
The International Federation of Automatic Control IFAC, (Cape Town,
South Africa), pp. 12281233, 2014.
[19] M. Turpin, N. Michael, and V. Kumar, Trajectory design and control for
aggressive formation flight with quadrotors, Autonomous Robots, vol. 33,
no. 1-2, pp. 143156, 2012.
[24] Y. X. Min, K.-C. Cao, H. H. Sheng, and Z. Tao, Formation tracking con-
trol of multiple quadrotors based on backstepping, in Control Conference
(CCC), 2015 34th Chinese, pp. 44244430, July 2015.
[26] R. Tjahjana and P. Sasongko, Multi flying vehicle control for saving fuel,
Applied Mathematical Sciences, vol. 8, no. 15, pp. 733742, 2014.
[27] B. Yu, X. Dong, Z. Shi, and Y. Zhong, Formation control for quadrotor
swarm systems: Algorithms and experiments, in 32nd Chines Control
Conference CCC, (Xian, China), pp. 70997104, 2013.
[29] A. van der Schaft, L2 -gain analysis of nonlinear systems and nonlinear
state feedback H control, IEEE Transactions on Automatic Control,
vol. 37, no. 6, pp. 770784, 1992.
[32] J. Diebel, Representing attitude: Euler angles, unit quaternions, and rota-
tion vectors, Stanford University Report, 2006.
[38] W. Jasim and D. Gu, H path tracking control for quadrotors based on
quaternion representation, in 15th Towards Autonomous Robotic Sys-
tems (TAROS 2014), (Birmingham, UK), pp. 7284, 2014.
[39] W. Jasim and D. Gu, Integral backstepping controller for quadrotor path
tracking, in Advanced Robotics, The 17th International Conference on,
pp. 5055, July 2015.
In: Quaternions: Theory and Applications ISBN: 978-1-53610-768-5
Editor: Sandra Griffin
c 2017 Nova Science Publishers, Inc.
Chapter 7
D ETERMINANTAL R EPRESENTATIONS OF
THE D RAZIN AND W-W EIGHTED D RAZIN
I NVERSES OVER THE Q UATERNION S KEW
F IELD WITH A PPLICATIONS
Ivan Kyrchei
Pidstryhach Institute for Applied Problems of Mechanics
and Mathematics, National Academy of Sciences of Ukraine,
Lviv, Ukraine
Abstract
A generalized inverse of a given quaternion matrix (similarly, as for
complex matrices) exists for a larger class of matrices than the invertible
matrices. It has some of the properties of the usual inverse, and agrees
with the inverse when a given matrix happens to be invertible. There
E-mail address: kyrchei@online.ua.
202 Ivan Kyrchei
1. Introduction
Let R and C be the real and complex number fields, respectively. Throughout
the paper, we denote the set of all m n matrices over the quaternion algebra
H = {a0 + a1 i + a2 j + a3 k | i2 = j 2 = k2 = 1, a0 , a1 , a2 , a3 R}
by Hmn , and by Hmnr the set of all m n matrices over H with a rank r.
Let M (n, H) be the ring of n n quaternion matrices and I be the identity
matrix with the appropriate size. For A Hnm , we denote by A , rank A the
conjugate transpose (Hermitian adjoint) matrix and the rank of A. The matrix
A = (aij ) Hnn is Hermitian if A = A.
The definitions of the generalized inverse matrices can be extended to
quaternion matrices as follows.
Drazin and W-Weighted Drazin Inverses Over the Quaternion ... 203
AXA = A; (1.1)
XAX = X; (1.2)
(AX) = AX; (1.3)
(XA) = XA. (1.4)
Definition 1.2. For A Hnn with k = Ind A the smallest positive number
such that rank Ak+1 = rank Ak , the Drazin inverse of A is defined to be the
unique matrix X that satisfying (1..2) and the following equations,
AX = XA; (1.5)
k+1 k
A X=A . (1.6)
rdeti A =
X
(1)nr (ai ik1 aik1 ik1 +1 . . .aik1 +l1 i ) . . . (aikr ikr +1 . . . aikr +lr ikr ),
Sn
= (i ik1 ik1 +1 . . . ik1 +l1 ) (ik2 ik2 +1 . . . ik2 +l2 ) . . . (ikr ikr +1 . . . ikr +lr ) ,
with conditions ik2 < ik3 < . . . < ikr and ikt < ikt +s for t = 2, r and s = 1, lt.
Definition 2.2. The j-th column determinant of A = (aij ) M (n, H) is de-
fined for all j = 1, n by putting
cdetj A =
X
(1)nr (ajkr jkr +lr . . . ajkr +1 ikr ) . . . (aj jk1 +l1 . . . ajk1 +1 jk1 ajk1 j ),
Sn
= (jkr +lr . . . jkr +1 jkr ) . . . (jk2 +l2 . . . jk2 +1 jk2 ) (jk1 +l1 . . . jk1 +1 jk1 j) ,
with conditions, jk2 < jk3 < . . . < jkr and jkt < jkt +s for t = 2, r and
s = 1, lt.
Suppose Ai j denotes the submatrix of A obtained by deleting both the i-th
row and the j-th column. Let a.j be the j-th column and ai. be the i-th row of
A. Suppose A.j (b) denotes the matrix obtained from A by replacing its j-th
column with the column-vector b, and Ai. (b) denotes the matrix obtained from
A by replacing its i-th row with the row-vector b.
We note some properties of column and row determinants of a quaternion
matrix A = (aij ), where i In , j Jn and In = Jn = {1, . . ., n}. These
properties completely have been proved in [2, 3].
Proposition 2.1. If b H, then rdeti Ai. (b ai. ) = b rdeti A for all i = 1, n.
Proposition 2.2. If b H, then cdetj A.j (a.j b) = cdetj Ab for all j = 1, n.
Proposition 2.3. If for A M (n, H) there exists t In such that atj = bj +cj
for all j = 1, n, then
where Ai. ji (a. i ) is obtained from A by replacing the j-th column with the i-th
column, and then by deleting both the i-th row and column.
Lemma 2.4. Let Li j be the ij-th left cofactor of A M (n, H), that is,
P
n
cdetj A = Li j ai j for all j = 1, n. Then
i=1
jj i, if j > i;
cdetk Ai . (aj. ) , i 6= j, k =
Li j = i 1, if j < i; (2.2)
cdetk Aii , i = j, k = min {Jn \ j},
where Ajj
i . (aj . ) is obtained from A by replacing the i-th row with the j-th row,
and then by deleting both the j-th row and column.
The following theorem has a key value in the theory of column-row deter-
minants.
Theorem 2.5. If A M (n, H) is Hermitian, then rdet1 A = = rdetn A =
cdet1 A = = cdetn A R.
208 Ivan Kyrchei
Remark 2.6. Since all column and row determinants of a Hermitian matrix
over H are equal, we can define the determinant of Hermitian A M (n, H) by
putting for all i = 1, n,
The matrix A 1 is obtained from A by replacing all entries of the i-th row and
the j-th column with zeroes except for 1 in the (i, j)-entry. Since elementary
212 Ivan Kyrchei
(k)
transformations of a matrix do not change its rank, then rank Ak+1
i. aj.
n o
min rank Ak , rank A . By rank A 1 rank Ak , the proof is completed.
(ij) (k)
where cn = cdeti Ak+1 .i
a. j and
X
c(ij)
s = cdeti Ak+1 ak.j
.i
Js, n {i}
for all s = 1, n 1, i, j = 1, n.
Proof. Denote by b. i the i-th column of Ak+1 =: (bij )nn . Consider the
Hermitian matrix I + Ak+1 . i (b. i ) Hnn . It differs from I + Ak+1
in an entry bii. Taking into account Theorem 2.13, we obtain
det I + Ak+1 (b. i ) = d1 n1 + d2 n2 + . . . + dn , (3.3)
.i
Drazin and W-Weighted Drazin Inverses Over the Quaternion ... 213
P
where ds = | Ak+1 | is the sum of all principal minors of order
Js, n {i}
s that contain the i-th column for all s = 1, n 1 and dn = det Ak+1 .
P (k)
a1l ali
l
P (k)
a a P (k)
l 2l li (k)
Consequently, we have b. i = = a. l ali , where a. l is
. . l
.
P (k)
anl ali
l
the l-th column of Ak for all l = 1, n. Due to Theorem 2.5, Lemma 2.4 and
Proposition 2.2, we obtain on the one hand
det I + Ak+1 (b. i ) = cdeti I + Ak+1 (b. i ) =
.i
X .i
k+1 (k)
cdeti I + A a. l al i =
.l
l
X
(k)
cdeti I + Ak+1 a. l ali . (3.4)
.i
l
On the other hand having changed the order of summation, for all s = 1, n 1
we have
X X
ds = det Ak+1 = cdeti Ak+1 =
Js, n {i} Js, n {i}
X X
(k)
cdeti Ak+1 a. l al i =
.i
Js, n {i} l
X X
(k)
cdeti Ak+1 a. l al i. (3.5)
.i
l Js, n {i}
or
P (k)
rdetj (Ak+1 )j . (ai. )
Ir,n {j}
aD
ij = P . (3.7)
|(Ak+1 ) |
Ir, n
P
(ij) (k)
for all i, j = 1, n, where cs = cdeti (Ak+1 ). i a.j for all
Js, n {i}
(ij) (k) (ij)
s = 1, n 1 and cn = cdeti Ak+1 .i a.j . We shall prove that ck = 0,
when k r + 1 for all i, j = 1, n.
(k)
Since by Lemma 3.2, Ak+1 . i a.j r, then the matrix
(k)
Ak+1 . i a.j has no more r right-linearly independent columns. Consider
(k)
(Ak+1 ) . i a.j , when Js,n {i}. This is a principal submatrix of
(k)
Ak+1 . i a.j of order s r + 1. Deleting both its i-th row and column, we
obtain a principal submatrix of order s 1 of Ak+1 . We denote it by M. The
following cases are possible.
(ij) (k)
and cn = cdeti Ak+1
.i
a.j = 0 for all i, j = 1, n.
(k) (ij) (ij)
Hence, cdeti I + Ak+1 . i a. j = c1 n1 + c2 n2 + . . . +
(ij)
cr nr for all i, j = 1, n. By substituting these values in the matrix from
(3.8), we obtain
or P
rdetj (A2)j . (ai. )
Ir,n {j}
agij = P .
|(A2 ) |
Ir, n
and
P
(k+1)
rdetj (Ak+1 )j . (ai. )
Ir,n {j}
AAD =
P
. (3.10)
|(Ak+1 ) |
Ir, n
nn
218 Ivan Kyrchei
Proof. At first we prove (3.9). Let AD A = (vij )nn . Using (3.6) for all
i, j = 1, n, we have
P ` (k)
cdeti Ak+1 . i a.j
X Jr, n {i}
vi j = P as j =
k+1
s (A )
Jr, n
P P ` (k) P `
(k+1)
cdeti Ak+1 . i a.j as j cdeti Ak+1 .i
a.j
Jr, n {i} s Jr, n {i}
P = P .
k+1 k+1
(A ) (A )
Jr, n Jr, n
P
cdeti (A A) . i a.j
Jr, n {i}
a+
ij = P , (3.11)
(A A)
Jr, n
or P
rdetj ((AA)j . (ai. ))
Ir,m {j}
a+
ij = P , (3.12)
|(AA ) |
Ir, m
Drazin and W-Weighted Drazin Inverses Over the Quaternion ... 219
for all i = 1, n, j = 1, m.
for all i, j = 1, n.
Denote by a t. the t-th row of Ak (A2k+1 ) =: A = ( aij ) Hnn for all
(k) P (2k+1)
t = 1, n. It follows from ait at . i . and (3.11) that
=a
t
220 Ivan Kyrchei
P ` (2k+1)
rdets A2k+1 A2k+1 a.t
n
X + Xn .s
(k) (2k+1) (k) Ir, n {s}
ait ats = ait P =
t=1 t=1
|(A2k+1 (A2k+1 ) )
|
Ir, n
P `
rdets A2k+1 A2k+1 (
ai . )
Ir, n {s} .s
P
|(A2k+1 (A2k+1 ) )
|
Ir, n
for all i, j = 1, n.
and
P
n
(k+1) P
ait cdett A2k+1 A2k+1 (
a. j )
t=1 .t
Jr, n {t}
AAD = P 2k+1 2k+1 ,
(A ) (A )
Jr, n
= (
where Ak (A2k+1 ) = A = (
aij ) and (A2k+1 ) Ak = A aij ).
or
P
rdetj Bk2 +1 j.
dA
i.
Ir2 ,m {j}
xij = P P , (4.3)
k1 +1 (Bk2 +1 )
(A )
Jr1 ,n Ir2 ,m
where
X
dB Bk2 +1 l. Hn1 , l = 1, n, (4.4)
.j = rdetj d
j.
Ir2 ,m {j}
X
dA cdeti Ak1 +1 .t H1m , t = 1, m, (4.5)
i. = d
.i
Jr1 ,n {i}
i. and d
are the column vector and the row vector, respectively. d .j are the i-th
e
row and the j-th column of D for all i = 1, n, j = 1, m.
Drazin and W-Weighted Drazin Inverses Over the Quaternion ... 223
P
(k )
rdetj (Bk2 +1 )j . (bs. 2 )
Ir2 ,m {j}
bD
sj = P . (4.8)
|(Bk2 +1 ) |
Ir2 ,m
P P
n ` (k ) P `
cdeti Ak1 +1 . i a.t 1
dts cdeti A k1 +1 . i d.s
Jr1 , n {i} t=1 Jr1 , n {i}
P = P
k1 +1 k1 +1
(A ) (A )
Jr1 , n J r1 , n
Suppose es. and e. s are respectively the unit row-vector and the unit column-
vector whose components are 0, except the s-th components, which are 1. Sub-
stituting (4.7) and (4.8) in (4.6), we obtain
P ` P
Ak1 +1 . i d. s (k )
m
cdeti rdetj (Bk2 +1 )j . (bs. 2 )
X Jr1 , n {i} Ir2 ,m {j}
xij = P P .
k1 +1 |(Bk2 +1 )
|
s=1 (A ) Ir2 ,m
Jr1 , n
Since
n
X m
X m
X
. s = (k ) (k )
d e. t dts , bs.
(k2 )
= bsl 2 el. , dts bsl 2 = detl ,
t=1 l=1 s=1
224 Ivan Kyrchei
then we have
xij =
P
m P
m P
n P `` k +1 (k ) P `
cdeti A 1 .i
(e. t ) dts bsl 2 rdetj (Bk2 +1 )j . (el. )
s=1 l=1 t=1 Jr1 , n {i} Ir2 ,m {j}
P P =
k1 +1
(A ) |(Bk2 +1 )
|
Jr1 , n Ir2 ,m
P
n P
m P P
cdeti Ak1 +1 .i
(e. l ) detl rdetj (Bk2 +1 )j . (et.)
t=1 l=1 Jr1 , n {i} Ir2 ,m {j}
P .
k1 +1 P
(A ) |(Bk2 +1 ) |
Jr1 , n Ir2 ,m
(4.9)
Denote by
dAil :=
X X n X
cdeti Ak1 +1 .l
d = cdeti Ak1 +1 (e. t) detl
.i .i
Jr1 , n {i} t=1 Jr , n {i}
1
P
m
Since dA A
il el. = di . , then it follows (4.3).
l=1
If we denote by
m
X X X
dB
tj := detl rdetj (Bk2 +1 )j . (el. ) = e t. ) ,
rdetj (Bk2 +1 )j . (d
l=1 Ir2 ,m {j} Ir2 ,m {j}
AX = D, (4.10)
XB = D, (4.12)
P
n P B
(k )
ait 1 cdett A2k1 +1 A2k1 +1 .t
d.j
t=1 1 Jr , n {t}
xij = P
2k1+1 2k1 +1 P B2k2 +1 (B2k2 +1 )
(A ) (A )
Jr1 , n Ir2 , m
(4.14)
or
!
m
P P (k )
rdets B2k2 +1 B2k2 +1 dA
i. asj 2
s=1 Ir2 , m {s} .s
xij = P ,
2k1 +1 2k1 +1 P B2k2 +1 (B2k2 +1 )
(A ) (A )
Jr1 , n Ir2 , m
(4.15)
where
0 0 1 1
Xm X
e q. A (k2 ) A
dB
.j =@ @ rdets B2k2 +1 B2k2 +1 d asj Hn1 ,
.s
s=1 Ir2 , m {s}
(4.16)
and
n
X X
(k ) e .p H1m ,
dA
i. =
ait 1 cdett A2k1 +1 A2k1 +1 d
.t
t=1 Jr1 , n {t}
(4.17)
p. , d
and d .q are the p-th row and the q-th column of D,
e respectively, for all
q = 1, n, p = 1, m.
Corollary 4.3. If rank Ak+1 = rank Ak = r n for A Hnn , then, for the
Drazin and W-Weighted Drazin Inverses Over the Quaternion ... 227
4.3. Examples
In this section, we give examples to illustrate our results.
1. Let us consider the matrix equation
AXB = D, (4.20)
where
1 k i 1 i
1 i
A = k 2 j , B = , D = k 1 .
i 1
i j 1 1 j
3 4k 3i
Since A2 = 4k 6 4j , det A = det A2 = 0, and
3i 4j 3
1 k 3 4k
det = 1, det = 2, then, by Theorem 2.11, Ind A = 1
k 2 4k 6
2 2 2i
and r1 = rank A = 2. Similarly, since B = , then Ind B = 1
2i 2
and r2 = rank B = 1.
228 Ivan Kyrchei
Since
1i 1+i
e = ADB = i + j 1 k ,
D
1 + i 1 + i
then by (4.4)
X
dB l.
B2 1. d Hn1 , l = 1, 2, 3; j = 1, 2.
.j = rdetj
I1,2 {j}
Thus, we have
1i 1+i
dB
.1 =
i + j , dB
.2 =
1 k .
1+i 1 + i
So
1i 4k 3i
2
A .1
d.1B
= i + j 6 4j ,
1 + i 4j 3
and finally we obtain
P
cdet A2 .1
dB
.1
J2, 3 {1}
xd11 = P 2 P =
(A ) |(B2 ) |
J2,3 I1,2
1 1 i 4k 1 i 3i 3 i + 2j
cdet1 + cdet1 = .
16 i + j 6 1+i 3 8
Drazin and W-Weighted Drazin Inverses Over the Quaternion ... 229
Similarly,
1 1 + i 4k 1 + i 3i 1 + 3i 2k
xd12 = cdet1 + cdet1 = ,
16 1k 6 1+i 3 8
d 1 3 1i i + j 4j 3i j + 4k
x21 = cdet2 + cdet1 = ,
16 4k i + j 1+i 3 8
d 1 3 1+i 1 k 4j 3 + 4j + k
x22 = cdet2 + cdet1 = ,
16 4k 1 k 1 + i 3 8
d 1 3 1i 6 i + j 1 + 3i + 2k
x31 = cdet2 + cdet2 = ,
16 3i 1 + i 4j 1 + i 8
1 3 1+i 6 1k 3 + i + 2j
xd32 = cdet2 + cdet2 = .
16 3i 1 + i 4j 1 + i 8
So,
3 i + 2j 1 + 3i 2k
1
Xd = 3i j + 4k 3 + 4j + k
8
1 + 3i + 2k 3 + i + 2j
is the Drazin inverse solution of (4.20).
2. Let us consider the matrix equation
AX = D, (4.21)
where
i j k 1 i
A = 1 k j , D = k 1 .
1 0 i 1 j
1 + j + k i + k 1
Since A2 = i + j k 1 + j i , A A =
2i j 1 + k
3 2k i + 2j
2k 2 2i ,
i + 2j 2i 3
10 2 + 2j 6k 2 + 2i + 4j + 2k
(A2 ) A2 = 2 2j + 6k 5 3i + j + 2k ,
2 2i 4j 2k 3 j 2k 4
230 Ivan Kyrchei
3 2k
2 2
det A A = det(A ) A = 0, det = 2,
2k 2
10 2 + 2j 6k
det = 6,
2 2j + 6k 5
then
X
3 3 23 2 + 3i + 5j 17k
(A ) A = det
2 3i 5j + 17k 15
J2, 3
15 3 13i + 2j + 5k
+ det
3 + 13i 2j 5k 15
23 8 + 4i + 15j + 2k
+ det = 72.
8 4i 15j 2k 15
Further,
0 1
11 9i 6j + 2k 9 6i j
= (A3 ) AD = @5 + 5i 4j 10k
D 1 2i 7j + 6kA ,
10 4i + 7j 3k 3 4i 7j 4k
and
0 1
` 11 9i 6j + 2k 2 + 3i + 5j 17k 8 + 4i + 15j + 2k
3 3 .1
(A ) A .1
d = @5 + 5i 4j 10k 15 3 13i + 2j + 5kA .
10 4i + 7j 3k 3 + 13i 2j 5k 15
Drazin and W-Weighted Drazin Inverses Over the Quaternion ... 231
Consider a matrix valued function A(t) = (aij (t)) Hnn R, where aij (t)
are quaternion-valued functions with the real variable t for all i, j = 1, n. Then
Z b Z b
dA(t) daij (t)
= , A(t)dt = aij (t)dt .
dt dt nn a a nn
and
q0 (t) = q(t)a(t), (5.6)
with the initial condition q(t0 ) = q0 have been considered and the following
proposition has been derived.
Proposition 5.4. Let q(t) = l (t)q0 and q(t) = q0 r (t) be solutions of (5.5)
and (5.6), respectively. If
Z t Z t
a(t) a( )d = a( )d a(t), (5.7)
t0 t0
then Z
t
l (t) = r (t) = exp a( )dt .
t0
234 Ivan Kyrchei
Rt
If a is constant, then, evidently, t0 a( )dt = a (t t0 ), and l (t) =
r (t) = exp(a (t t0 )).
The similar result has been obtained in [43] as well. In [43], the following
nonhomogeneous differential equation corresponding to (5.5) has been consid-
ered,
q0 (t) = a(t)q(t) + f (t), (5.8)
where f : [0, T ] H and a : [0, T ] H. It has been shown, if condition (5.7)
is satisfied, then the solutions of (5.8) are given by
Z t Z t Z s
q(t) = exp a( )d q(0) + exp (a( )) d f (s)ds , (t [0, T ]).
0 0 0
(5.9)
In the special case when a is constant and q(0) = 1, then the solutions of (5.8)
are given by
Z t
q(t) = exp (at) exp (as) f (s)ds , (t [0, T ]). (5.10)
0
X0 + AX = B, (5.11)
If A is invertible, then
Z
exp (At) dt = A1 exp (At) + G, (5.12)
Proof. Differentiate the right-hand side of (5.13), and use the series expansion
for exp (At).
Using (5.13) and the series expansion for exp (At), we get an explicit
form for a general solution of (5.11),
ff
A A2 2 Ak1 k1
X(t) = AD + (I AAD )t I t + t ...(1)k1 t + G B.
2 3! k!
0 P 1
cdeti Ak+1 b (k+2)
b
.i .j
1Bbb(1) J r, n {i} C 2
B
P k+1 C t + ...
2@ ij
(A )
A
Jr, n
0 P 1
cdeti Ak+1 b (2k)
b
k B . i .j
(1) Bb(k1) Jr, n {i} C k
b Ct (5.15)
k! @ ij P A
(Ak+1 )
Jr, n
(l)
b (l) = (bb ) Hnn for all l = k, 2k;
where Al B =: B ij
236 Ivan Kyrchei
2. when A is arbitrary,
P
n P (0)
(k) b
ait cdets A2k+1 A2k+1 .s
d .j
s=1 Jr, n {s}
xij = P 2k+1 2k+1
(A ) (A )
Jr, n
P
n
(k) P
ais cdets A2k+1 A2k+1 b (1)
d
s=1
.s .j
Jr, n {s}
+ bij
P 2k+1 2k+1 t
(A ) (A )
Jr, n
n (2)
P (k) P b 2k+1 2k+1
ait cdets A A d
1
.s .j
(1) t=1 Jr, n {s} 2
bb
P 2k+1 2k+1 t +...
2 ij (A ) (A )
Jr, n
(k)
P
n
(k) P b 2k+1 2k+1
ais cdets A A d .j
(1)k s=1
.s
b(k1) Jr, n {s}
k
bij P 2k+1 2k+1 t
k! (A ) (A )
Jr, n
(5.16)
2. The proof of (5.16) is similar to the proof of (5.15) by using the determi-
nantal representation of AD by (3.14).
X(t) =
D D A A2 2 k1 A
k1
k1
B A + (I AA )t I t + t + ...(1) t +G .
2 3! k!
If we put G = 0, then we obtain the following partial solution of (5.17),
1 (1)k1
X(t) = BAD +(BBAAD )t (BABA2 AD )t2 +... (BAk1 BAk AD )tk .
2 k!
(5.18)
!
P
n P ` 2k+1
rdets A2k+1
A (0)
(d )
(k)
asj
s=1 Ir, n {s} .s
xij = P
|(A2k+1 (A2k+1 ) )
|
Ir, n
0 ! 1
P
n P `
2k+1 2k+1 (1)) (k)
B rdets A A (d asj C
B s=1 Ir, n {s} .s C
+B
Bbij P Ct
C
@ |(A2k+1 (A2k+1 ) )
| A
Ir, n
0 ! 1
P
n P `
2k+1 2k+1 (2) ) (k)
B rdets A A (d asj C
1 B (1) s=1 Ir, n {s} .s C 2
Bb P C t + ...
2 B ij |(A2k+1 (A2k+1 ) ) C
@ | A
Ir, n
0 ! 1
P
n P ` 2k+1
B rdets 2k+1
A A (k))
(d
(k)
asj C
(1)k B (k1) s=1 Ir, n {s} .s C k
Bb P Ct ,
B ij |(A2k+1 (A2k+1 ) ) C
k! @ | A
Ir, n
Proof. The proof is similar to the proof of Theorem 5.2 by using the determi-
nantal representation of the Drazin inverse (3.6) and (3.14), respectively.
5.3. An Example
Let us consider the matrix equation
X0 + AX = B, (5.19)
where
1 k i i j k
A = k 2 j , B = 1 k j .
i j 1 1 0 i
Drazin and W-Weighted Drazin Inverses Over the Quaternion ... 239
3 4k 3i
Since A 2 = 4k 6 4j , det A = det A2 = 0, and
3i 4j 3
1 k 3 4k
det = 1, det = 2, then, by Theorem 2.11, Ind A = 1
k 2 4k 6
and r1 = rank A = 2. Since A is Hermitian and Ind A = 1, then we shall find
the solutions (xij ) H33 by (5.15),
P P
cdeti A2. i bb (1)
cdeti A2. i bb (2)
.j .j
J2, 3 {i} J2, 3 {i}
xij = P 2 +
bij
P 2 t
(A ) (A )
J2, 3 J2, 3
P 2
for all i, j = 1, 2, 3. We have, A = 4,
J2, 3
k 1+j 1i+k
b (1) = AB = 2 i 2k 1 + 2j k ,
B
j i + k 1+ij
4k 4 + 3j 3 4i + 3k
Bb (2) = A2 B = 6 4i 6k 4 + 6j 4k .
4j 4i + 3k 4 + 3i 3j
Therefore,
1 k 4k k 3i 1 4k 4k
x11 = cdet1 + cdet1 + i cdet1 +
4 2 6 j 3 4 6 6
4k 3i 1 1
cdet1 t = (2k) + i [0] t = 0.5k + (i)t;
4j 3 4 4
1 1+j 4k 1+j 3i 1 4 + 3j 4k
x12 = cdet1 + cdet1 + j cdet1 +
4 i 2k 6 i+k 3 4 4i 6k 6
4 + 3j 3i 1 1
cdet1 t = (2 + 2j) + j [2j] t = 0.5 + 0.5j + (0.5j) t;
4i + 3k 3 4 4
1 1 i + k 4k 1 i + k 3i
x13 = cdet1 + cdet1 +
4 1 + 2j k 6 1 + ij 3
1 3 4i + 3j 4k 3 4i + 4k 3i
k cdet1 + cdet1 t=
4 4 + 6j 4k 6 4 + 3i 3j 3
1 1
(2 + 2i + 2k) + k [2 + 11k] t = 0.5 + 0.5i + 0.5k + (0.5 4.5k) t;
4 4
240 Ivan Kyrchei
1 3 k 2 4j 1 3 4k
x21 = cdet2 + cdet1 + 1 cdet2 +
4 4k 2 j 3 4 4k 6
6 4j 1 1
cdet1 t= (4) + 1 [4k] t = 1 + (1 + k)t;
4j 3 4 4
1 3 1+j i 2k 4j
x22 = cdet2 + cdet1 +
4 4k i 2k i+k 3
1 3 4 + 3j 4i 6k 4j
k cdet2 + cdet1 t=
4 4k 4i 6k 4i + 3k 3
1 1
(2i 4k) + k [4k] t = 0.5i k;
4 4
1 3 1i+k 1 + 2j k 4j
x23 = cdet2 + cdet1 +
4 4k 1 + 2j k 1+ij 3
1 3 3 4i + 3k 4 + 6j 4k 4j
j cdet2 + cdet1 t=
4 4k 4 + 6j 4k 4 + 3i 3j 3
1 1
(2 + 4j + 2k) + j [4j] t = 0.5 + j + 0.5k;
4 4
1 3 k 6 2 1 3 4k
x31 = cdet2 + cdet2 + 1 cdet2 +
4 3i j 4j j 4 3i 4j
6 6 1 1
cdet2 t = (2j) + i [0] t = 0.5j + t;
4j 4j 4 4
1 3 1+j 6 i 2k
x32 = cdet2 + cdet2 +
4 3i i + k 4j i + k
1 3 4 + 3j 6 4i 6k
0 cdet2 + cdet2 t=
4 3i 4i + 3k 4j 4i + 3k
1 1
(2i + 2k) + [2k] t = 0.5i + 0.5k + (0.5k)t;
4 4
1 3 1i+k 6 1 + 2j k
x33 = cdet2 + cdet2 +
4 3i 1 + i j 4j 1+ij
1 3 3 4i + 3jk 6 4 + 6j 4k
i cdet2 + cdet2 t=
4 3i 4 + 3i 3j 4j 4 + 3i 3j
1 1
(2 + 2i 2j) + i [2i 2j] t = 0.5 + 0.5i 0.5j + (0.5i 0.5j)t.
4 4
Drazin and W-Weighted Drazin Inverses Over the Quaternion ... 241
Denote
WAW B
M= .
C 0
Then the W-weighted Drazin inverse Ad,W = (a)ij Hnm has the following
determinantal representations:
Pm+ns
k=1 Lki mkj
aij = , i = 1, m, j = 1, n, (6.3)
det M M
or Pm+ns
mik Rjk
aij = k=1 , i = 1, m, j = 1, n, (6.4)
det MM
where Lij are the left (ij)-th cofactor of M M and Rij are the right (ij)-th
cofactor of MM , respectively, for all i, j = 1, m + n s.
242 Ivan Kyrchei
As can be seen, the auxiliary matrices B and C have been used in the de-
terminantal representations (6.3) and (6.4). In this chapter we escape it. We
shall derive determinantal representations of the W-weighted Drazin inverse of
an arbitrary matrix A Hmn with respect to W Hnm by using the de-
terminantal representations of the Drazin inverse, of the Moore-Penrose inverse,
and the limit representation of the W-weighted Drazin inverse in some particular
case.
where
n
X
(2) D,f
(uD
qj ) = uD,l
qp upj (6.8)
p=1
Drazin and W-Weighted Drazin Inverses Over the Quaternion ... 243
or
!
P
m P
(k)
rdets V 2k+1 V 2k+1 (
vi . )
vsj
s=1 Ir, m {s} .s
D,2
vij = P 2k+1 , (6.12)
V (V 2k+1 )
Ir, m
for all i = 1, m, j = 1, n.
Denote by w t. the t-th row of V k W =: W = (w ij ) Hmn for all
P (k)
t = 1, m. It follows from vtl wl . = w t . and (3.12) that
l
P
m m
rdetj (WW )j. (wl. )
X (k) +
X (k) Ir1 , n {j}
vtl wlj = vtl P =
|(WW )
|
l=1 l=1
Ir1 , n
P
rdetj (WW )j . (w
t. )
Ir1 , n {j}
P , (6.16)
|(WW )
|
Ir1 , n
246 Ivan Kyrchei
(6.18)
Thus, we have proved the following theorem.
Theorem 6.4. Let A Hmn and W Hrnm 1
with k =
max{Ind(AW), Ind(WA)} and r = rank(AW)k+1 = rank(AW)k . Then
the W-weighted Drazin inverse of A with respect to W possesses the deter-
minantal representation (6.18), where V = AW, V = V k (V 2k+1 ) , and
W=V W .k
(6.19)
Drazin and W-Weighted Drazin Inverses Over the Quaternion ... 247
= (U2k+1 ) Uk , and W
where U = WA, U = W Uk .
d,W
Proof. Using (6.14), an entry aij of the W-weighted Drazin inverse Ad,W
can be obtained as follows
X n Xn Xn
d,W + (k) (2k+1) + (k)
aij = wis ust utl ulj (6.20)
s=1 t=1 l=1
and 1
Ad,W = lim A(WA)k In + (WA)k+2 (6.24)
0
k := A(WA)k Hmn .
where U
250 Ivan Kyrchei
(ij) (k)
where cn = cdeti V k+2 .i
.j
v and
X
(k)
c(ij)
s = cdeti V k+2 v .j
.i
Js, n {i}
for all s = 1, n 1, i, j = 1, n.
(k+2)
Proof. Consider the Hermitian matrix tI + V k+2 . i (v. i ) Hnn . Taking
into account Theorem 2.13, we obtain
(k+2)
det I + V k+2 v. i = d1 n1 + d2 n2 + . . . + dn , (6.27)
.i
P
where ds = | V k+2
| is the sum of all principal minors of order
Js, n {i}
s that contain the i-th column for all s = 1, n 1 and dn = det V k+2 .
P (k)
v1l w
li
l
P (k)
li P
v2l w
(k+2) (k) (k)
Consequently, we have v. i = l = . l w
v li , where v
. l is
.. l
.
P (k)
vnl w
li
l
k = (AW)k A and WAW = W
the l-th column of V = (w
li) for all l = 1, n.
By Theorem 2.5, we obtain on the one hand
(k+2) (k+2)
det I + V k+2 v. i = cdeti I + V k+2 v. i =
.i
X X .i
(k) (k)
= cdeti I + V k+2 . l wli =
v cdeti I + V k+2 . l w
v li
.l .i
l l
(6.28)
Drazin and W-Weighted Drazin Inverses Over the Quaternion ... 251
On the other hand, having changed the order of summation, for all s = 1, n 1
we have
X X
ds = det V k+2 = cdeti V k+2 =
Js, n {i} Js, n {i}
X X
(k)
cdeti V k+2 v. l wl i =
.i
Js, n {i} l
X X
(k)
cdeti V k+2 v. l w
l i. (6.29)
.i
l Js, n {i}
l i when
By substituting (6.28) and (6.29) in (6.27), and equating factors at w
l = j, we obtain (6.26).
0 1
L11 L21 ... Lm1
1 B L12
1 B L22 ... Lm2 C
C,
Im + (AW)k+2 =
det (Im + (AW)k+2 ) @ . . . ... ... ... A
L1m L2m ... Lmm
If s > r +1, thendet M = 0 and p, (p < r), columns are basis in the both
k+2 (k)
matrices M and (AW). i v
. Therefore, by Theorem 2.8, we
.j
(k)
also have cdeti (AW)k+2 .i v
.j = 0.
(k)
Thus, in all cases we have cdeti (AW)k+2 .i v
.j = 0, when Js,m {i}
and r + 1 s < m. From here if r + 1 s < m, then
X
(ij) k+2 (k)
cs = cdeti (AW). i v
.j = 0,
Js, m {i}
(ij) (k)
and cm = cdeti (AW)k+2 .i v .j = 0 for all i, j = 1, n.
(k) (ij) (ij)
Hence, cdeti I + (AW)k+2 . i v . j = c1 m1 + c2 m2 + . . . +
(ij)
cr mr for i = 1, m and j = 1, n. By substituting these values in the matrix
from (6.31), we obtain
254 Ivan Kyrchei
P
(ij) (k)
Here cr = cdeti (AW)k+2
.i v.j and dr =
P Jr, m {i}
k+2
(AW) . Thus, we have the determinantal representation of
Jr, m
Ad,W by (6.30).
6.4. An Example
Let us consider the matrices
0 i 0
k 1 k 0 i 0
i
A= 1 0
, W = j k 0 1 .
0
0 1 0 k
1 k j
Drazin and W-Weighted Drazin Inverses Over the Quaternion ... 255
Then
k j 0 i
1 j i + k j 1 + j i j 0
V := AW =
k
, U := WA = 0 k 0 ,
0 i 0
0 0 0
i + k 1 j i i k
and rank W = 3, rank V = 3, rank V 3 = rank V 2 = 2, rank U2 =
rank U = 2. Therefore, Ind V = 2, Ind U = 1, and k =
max{Ind(AW), Ind(WA)} = 2.
Its evident that obtaining the W-weighted Drazin inverse of A with respect
to W by using the matrix U by (6.19) is more convenient. We have
0 1 0 1 0 1
1 i+k 0 i 2 + 3j 0 i 0 0
U2 = @ 0 1 A , U5 = @ 0 k A , (U5 ) = @2 3j k A,
0 0 0 0 0 0 0 0 0
0 1 0 1
` 5 5 1 2i 3k 0 i 1+j 0
U U = @2i + 3k 14 A,U = (U ) U = @2 + 3j i + 6k
5 2 A,
0 0 0 0 0 0
0 1 0 1
k j 0 2 i j j
B 0 k 1C B 0 2k C
W = B C , W W = B i 2 C,
@ i 0 0A @j 0 1 0 A
0 1 k j 2k 0 2
0 1
k 1 2j 0
B 0 i + k 0C
W=W U =@ 2 B C.
i 1 + j 0A
0 1 0
By (6.19),
ad,W
11 =
3
P P P
cdet1 ((W W).1 (w
.t )) cdett U5 U5 .t
(
u.1 )
t=1 I3, 4 {1} 2, 3 J {t}
P
P
,
(W W) (U5 ) U5
J3, 4 J2, 3
where
X `
cdet1 (W W).1 (w
.1 ) =
I3, 4 {1}
0 1 0 1 0 1
k i j k i j k j j
cdet1 @ 0 2 0 A + cdet1 @ 0 2 2k A + cdet1 @ i 1 0A = 0,
i 0 1 0 2k 1 0 0 2
256 Ivan Kyrchei
X ` X `
cdet1 (W W).1 (w
.2 ) = 2j, cdet1 (W W).1 (w
.3 ) = 0,
I3, 4 {1} I3, 4 {1}
X
(W W) = 2,
J3, 4
and
X `
cdet1 U5 U5 u.1 ) =
(
.1
J2, 3 {1}
i 2i 3k i 0
cdet1 + cdet1 = i,
2 + 3j 14 0 0
X ` 5 5
cdet2 U U u.1 ) = 0,
(
.2
J2, 3 {2}
X ` X ` 5 5
cdet3 U5 U5 u.1 ) = 0,
( U U = 1.
.3
J2, 3 {3} J2, 3
Therefore,
(0 i) + (2j 0) + (0 0)
ad,W
11 = = 0.
21
Continuing in the same way, we finally get,
0 i 0
0 0 0
Ad,W = 1 5i 2k 0 .
(6.33)
0 0 0
By (3.11), we obtain
0 1 0 1
` i 3 + 2j 0 ` 5 + 2 i 5 0
5 + 2
U =@0 k A , (AW) = U = U U
D D
U =@0 k A.
0 0 0 0 0 0
then we have
0 i 0
2 k 6 + 5i 0
Ad,W = A (WA)D ) =
1 5i + 5k
, (6.34)
0
1 5i + 6k 0
The W-weighted Drazin inverse in (6.34) different from (6.33). It can be ex-
plained that the Jordan normal form of WA is unique only up to the order of
the Jordan blocks. We get their complete equality, if Ad,W from (6.34) be left-
multiply by the nonsingular matrix P which is the product of multiplication of
the following elementary matrices,
1 0 0 0
0 1 0 k
P = P2,4 (k) P4,3 (1) P3,4 (6) P4,1 (j) = 0 0 7 6 .
j 0 1 1
where A Cmn , W Cnm with Ind (AW) = k1 , Ind (WA) = k2 and
rank (AW) = r1 , rank (WA) = r2 . He proofed if b R (W)k2 A and
k1 k2
where U1 Cnnr
nr2
2 mmr1
, V1 Cmr 1
are matrices whose columns form
bases for N ((WA) 2 ) and N ((AW)k1 ), respectively.
k
over the quaternion skew and presented Cramers rule of the restricted matrix
equation,
W1 AW1 XW2 BW2 = D, (7.3)
Rr (X) Rr (AW1 )k1 Nr (X) Nr (W2 B)k2 ,
(7.4)
Rl (X) Rl (BW2 )k2 , Nl (X) Nl (W1 A)k1 ,
where A Hmn , W1 Hnm , B Hpq , W2 Hqp ,
and D Hnp with k1 = max {Ind(AW1), Ind (W1A)}, k2 =
max {Ind(BW2 ), Ind (W2B)}, and rank (AW1 )k1 = s1 , rank (BW2 )k2 =
s2 . He proved that if
Rr (D) Rr (W1 A)k1 , (W2 B)k2 , Rl (D) Rl (AW1 )k1 , (BW2 )k2
where (vilD ) = V D is the Drazin inverse of V = AW1 and (vilD )(2) can be
obtained by (6.10), and (uD D
f j ) = U is the Drazin inverse of U = W2 B and
(uD (2) can be obtained by (6.8).
qj )
ii) If AW1 Hmm and W2 B Hqq are Hermitian, then
P
cdeti (AW1 )k. 1i +2 dB
.j
Js1, m {i}
xij = P , (7.7)
k +2 P k +2
(AW1 ) 1 (W2 B) 2
Js1, m Is2 , q
or P
rdetj (W2 B)kj 2. +2 (dA
i. )
Is2,q {j}
xij = P , (7.8)
k1 +2 P k2 +2
(AW1 ) (W2 B)
Js1, m Is2 , q
where
X
dB t. ) Hn1 , t = 1, n
rdetj (W2 B)jk2. +2 (d
.j = (7.9)
Is2,q {j}
X
dA
i. =
cdeti (AW1 )k. 1i +2 d
. l
H1q , l = 1, q (7.10)
Js1 , m {i}
are the column vector and the row vector, respectively. d i. and d
.j are the i-th
for all i = 1, n, j = 1, p.
row and the j-th column of D
Proof. The existence and uniqueness of the solution (7.5) can be proved similar
as in ( [14], Theorem 5.2).
To derive Cramers rule (7.6) we use (6.1). Then, we obtain
2 2
X = (AW1 )D ) ADB (W2 B)D ) . (7.11)
260 Ivan Kyrchei
Denote ADB =: D = dlf Hmq , V := AW1 , and U := W2 B. The
equation (7.11) can be written component-wise as follows
! 0 1
p X
X n p X
X n m
X q
X
1 D (2) (2) A
xij = (ad,W
it )dts (bd,W
sj
2
) = (vil ) alt dts @ bsf (uD
fj)
s=1 t=1 s=1 t=1 l=1 f =1
P
n n
cdeti (AW1 )k. 1i +2 (
v.t )
X 1
X Js1 , m {i}
ad,W dts = P dts =
it k +2
t=1 t=1 (AW1 ) 1
Js1 , m
P P
n P
cdeti (AW1 )k. 1i +2 (
v.t )
dts cdeti (AW1 )k. 1i +2 d
. s
Js1 , m {i} t=1 Js1 , m {i}
P = P
k +2 k +2
(AW1 ) 1 (AW1 ) 1
Js1 , m Js1 , m
(7.15)
Suppose es. and e. s are respectively the unit row-vector and the unit column-
vector whose components are 0, except the s-th components, which are 1. Sub-
stituting (7.15) and (7.13) in (7.14), we obtain
P P
p
cdeti (AW1 )k.1i +2 d. s
rdetj (W2 B)kj 2. +2 (
us. )
X Js1 , m {i} Is2 ,q {j}
xij = P P .
k +2 k +2
s=1 (AW1 ) 1 (W2 B) 2
Js1 , m Is2 , q
Since
n
X q
X p
X
. s =
d e. t dts , u
s. = sl el. ,
u dtsu
sl = dtl , (7.16)
t=1 l=1 s=1
then we have
xij =
P
p Pn P q P P
cdeti (AW1 )k.1i +2 (e. t ) dts u
sl rdetj (W2 B)jk2. +2 (el. )
s=1 t=1 l=1 Js1 , m {i} Is2 ,q {j}
P =
k +2 P k +2
(AW1 ) 1 (W2 B) 2
Js1 , m Is2 , q
P
n P
q P P
cdeti (AW1 )k. 1i +2 (e. t ) dtl rdetj (W2 B)kj 2. +2 (el. )
t=1 l=1 Js , m {i} Is2 ,q {j}
1
P .
k +2 P k +2
(AW1 ) 1 (W2 B) 2
Js1 , m Is2 , q
(7.17)
262 Ivan Kyrchei
Denote by
X `
dA
il := cdeti (AW1 )k. 1i +2 d
. l
=
Js1 , m {i}
n
X X
cdeti (AW1 )k. 1i +2 (e. t) dtl
t=1 Js1 , m {i}
q
P
Since dA A
il el. = di . , then it follows (7.8).
l=1
If we denote by
q
X X
dB
tj := dtl rdetj (W2 B)kj 2. +2 (el. ) =
l=1 Is2 ,q {j}
X
t. )
rdetj (W2 B)kj 2. +2 (d
Is2 ,q {j}
n
P
Since e.tdB B
tj = d. j , then it follows (7.7).
t=1
Remark 7.2. To establish the Cramer rule of (7.3) we shall not use the de-
terminantal representations (6.30) and (6.30) for (7.5) because corresponding
determinantal representations of its solution will be too cumbersome. But they
are suitable in the following corollaries.
Drazin and W-Weighted Drazin Inverses Over the Quaternion ... 263
or P
k2 +2 A
(W2 B)j . (di . )
Is ,q {j}
xij = P 2 ,
k +2 P k +2
(AW1 ) 1 (W2 B) 2
Js1, m Is2 , q
where
X
k +2 n1
dB
.j =
(W2 B)j 2. (dt. ) C , t = 1, n
Is2 ,q {j}
X
k +2
dA
i. =
(AW1 ) . 1i d. l C1q , l = 1, q
Js1, m {i}
i. and d
are the column vector and the row vector, respectively. d .j are the i-th
row and the j-th column of D for all i = 1, n, j = 1, p. These determinantal
representations are most applicable for the complex case.
Corollary 7.1. Suppose the following restricted matrix equation is given,
WAWX = D, (7.18)
Rr (X) Rr (AW)k , Nl (X) Nl (WA)k , (7.19)
X = Ad,W D, (7.20)
264 Ivan Kyrchei
where D )(2)
(viq can be obtained by (6.10) and AD = R = (rqj ) Hmp .
iii) If AW Hmm is Hermitian, then
P
cdeti (AW)k+2
.i (f .j )
Jr, m {i}
xij = P , (7.23)
k+2
(AW)
Jr, m
= (AW)k AD.
where f.j is the j-th column of F = VD
Proof. To derive a Cramers rule (7.21), we use the determinantal representation
(6.19) for Ad,W . Then
p
X
xij = ad,W
is dsj =
s=1
2P
n P P `
3
2k+1
p 6
.t )
cdeti (W W).t (w cdett U U 2k+1
(
u.s )
X .t 7
6 t=1 Jr1 , m {i}
Jr, n {t}
7
6 P 7dsj
4 P 5
s=1 (W W) ((U2k+1 ) U2k+1 )
Jr1 , m Jr, n
(7.24)
Denote D = UD = (U2k+1 k
) U D, where D = dsj Hnp
. Since
p
X
.j ,
.s dsj = d
u
s=1
Drazin and W-Weighted Drazin Inverses Over the Quaternion ... 265
= (AW)k AD.
where f.j is the j-th column of F = VD
XWBW = D, (7.25)
Rl (X) Rl (BW)k , Nr (X) Nr (BA)k , (7.26)
X = DBd,W , (7.27)
xij =
P
p P ` P
rdetl V2k+1 V2k+1 i .)
(d rdetj (WW )j . (w
l. )
l=1 Ir, p {l} l. Ir1 ,q {j}
P P
|(V2k+1 (V2k+1 ) )
| |(WW )
|
Ir, m Ir1 , n
(7.28)
266 Ivan Kyrchei
Proof. The proof is similar to the proof of Corollary 7.1 in the point i), and
follows from Theorem 7.1 by putting W2 = W, AW1 = I.
we substitute usual determinants for all corresponding row and column deter-
minants in (7.28), (7.29), and (7.30). Herein the condition WB Cnn be
Hermitian is not necessary, then in the complex case (7.30) can be represented
as follows,
P
(WB)k+2
j. (gi. )
Ir,q {j}
xij = P
k+2
.
(WB)
Ir, q
7. 3. Examples
1. Let us consider the matrix equation
WAWX = D (7.31)
Drazin and W-Weighted Drazin Inverses Over the Quaternion ... 267
with the restricted conditions (7.19), where W and A are the same as in Exam-
ple 64., and
k i
D = i j .
1 i
Therefore, the matrices V = AW, U = WA, U5 U5 , W , W W, W =
2
W U are the same that in Example 64. as well, and
ijk j
D = (U5 ) U2 D = 1 + 3i + 6j 2k 4i 2k .
0 0
So, by (7.21)
P
3 P ` P ``
cdet1 (W W).1 (w
.t ) cdett U5 U5 .1 )
(d
.t
t=1 I3, 4 {1} J2, 3 {t}
x11 = P
P
,
(W W) ((U5 ) U5 )
J3, 4 J2, 3
where
X
.1 )) =
cdet1 ((W W).1 (w
I3, 4 {1}
k i j k i j k j j
cdet1 0 2 0 + cdet1 0 2 2k + cdet1 i 1 0 = 0,
i 0 1 0 2k 1 0 0 2
X
.2 )) = 2j,
cdet1 ((W W).1 (w
I3, 4 {1}
X X
.3 )) = 0,
cdet1 ((W W).1 (w (W W) = 2,
I3, 4 {1} J3, 4
and
X
cdet1 U5 U5 .1 )
(d
=
.1
J2, 3 {1}
ijk 2i 3k ij k 0
cdet1 +cdet1 = 2ijk,
1 + 3i + 6j 2k 14 0 0
268 Ivan Kyrchei
X
cdet2 U5 U5 (d .1 ) = j,
.2
J2, 3 {2}
X X
5 5 .1 ) 5 5
cdet3 U U (d = 0, U U = 1.
.3
J2, 3 {3} J2, 3
Therefore,
0 (2i j k) + (2j) j + 0 0
x11 = = 1,
21
0 (2 + 2j) + (2j) i + 0 0
x12 = = k,
21
2j (2i j k) + (10i 4k) j + 0 0
x21 = = 1 + i + 7k,
21
k j 0
k 0 i 0 0 k k i
1
A = j k 0 1 , W1 =
i 0 , W2 = j 0 ,
0
0 1 0 k 0 1
0 1 k
Drazin and W-Weighted Drazin Inverses Over the Quaternion ... 269
i 1
k j 0 k 0
B= ,D =
.
j 0 1 0 j
1 0
Since the following matrices are Hermitian
2 i 0 0 i i
V = AW1 = i 1 0 , U = W2 B = i 1 0 ,
0 0 0 i 0 1
then we can find the W-weighted Drazin inverse solution of (7.32) by its deter-
minantal representation (7.7). We have
Therefore,
2i + j 7 + k 5 + 2k
= AW1 ADBW2 B = 1 + k 5i j 4i 2j .
D
0 0 0
By (7.9), we can get
36i 9j 27 9 9k
dB.1 =
27 9k , dB
.2 =
18i , dB
.3 =
9i + 3j .
0 0 0
Since
36i 9j 8i 0
(AW1 )3. 1 dB
.1 =
27 9k 5 0 ,
0 0 4
270 Ivan Kyrchei
Similarly,
27 8i 9 9k 8i
cdet1 cdet1
18i 5 1 9i 3j 5 9 7k
x12 = = , x13 = = ,
27 3 27 9
13 36i 9j 13 27
cdet2 cdet2
8i 27 9k 7 5k 8i 18i 2i
x21 = = , x22 = = ,
27 3 27 3
13 9 9k
cdet2
8i 9i + 3j 15i 11j
x23 = = , x31 = x32 = x33 = 0.
27 9
So, the W-weighted Drazin inverse solution of (7.32) are
12i + 9j 3 9 7k
1
X = 21 15k 6i 15i 11j .
9
0 0 0
Conclusion
In this chapter, we have obtained determinantal representations of the Drazin
and W-weighted Drazin inverses over the quaternion skew field. We have de-
rived determinantal representations of the Drazin inverse for both Hermitian and
arbitrary matrices over the quaternion skew field by the theory of column-row
determinants recently introduced by the author. Using obtained determinantal
representations of the Drazin inverse we have get explicit representation formu-
las (analogs of Cramers rule) for the Drazin inverse solutions of the quater-
nionic matrix equations AXB = D and, consequently, AX = D, XB = D
in both cases when A and B are Hermitian and arbitrary. We also have obtain
determinantal representations of solutions of the differential quaternion-matrix
equations, X0 + AX = B and X0 + XA = B, where A is noninvertible.
Also, we have obtained new determinantal representations of the W-
weighted Drazin inverse over the quaternion skew field. We have gave de-
Drazin and W-Weighted Drazin Inverses Over the Quaternion ... 271
References
[1] R. E. Cline, T. N. E. Greville, A Drazin inverse for rectangular matrices.
Linear Algebra Appl. 1980, 29, 5362.
[2] I. Kyrchei, Cramers rule for quaternion systems of linear equations. Fun-
dam Prikl Mat. 2007, 13 (4), 6794. (Russian)
[3] I. Kyrchei, The theory of the column and row determinants in a quater-
nion linear algebra. In: Albert R. Baswell (Ed.), Advances in Mathematics
Research 15, Nova Sci. Publ., New York, 2012, pp. 301359.
[4] J. Dieudonn`e, Les determinants sur un corps non-commutatif. Bull Soc
Math France. 1943, 71, 2745.
[5] E. Study, Zur Theorie der linearen Gleichungen. Acta Math. 1920, 42, 1
61.
[6] E. H. Moore, On the determinant of an hermitian matrix of quaternionic
elements. Bull Amer Math Soc. 1922, 28, 161162.
[7] F. J. Dyson, Quaternion determinants. Helv Phys Acta 1972, 45, 289302.
[8] L. Chen, Definition of determinant and Cramer solution over the quater-
nion field. Acta Math Sinica (N.S.) 1991, 7, 171180.
[9] I. M. Gelfand, V. S. Retakh, Determinants of matrices over noncommuta-
tive rings. Functional Anal Appl. 1991, 25, 91102.
[10] I. Kyrchei, Determinantal representations of the Moore-Penrose inverse
over the quaternion skew field and corresponding Cramers rules. Linear
Multilinear A. 2011, 59, 413431.
272 Ivan Kyrchei
[11] I. Kyrchei, Explicit representation formulas for the minimum norm least
squares solutions of some quaternion matrix equations. Linear Algebra
Appl. 2013, 438, 136152.
[12] G. Song, Q. Wang, H. Chang, Cramer rule for the unique solution of re-
stricted matrix equations over the quaternion skew field. Comput Math.
Appl. 2011, 61, 15761589.
[20] R. M. W. Wood, Quaternionic eigenvalues. Bull Lond Math Soc. 1985, 17,
137138.
[36] I. Kyrchei, Analogs of the adjoint matrix for generalized inverses and cor-
responding Cramer rules. Linear Multilinear A. 2008, 56, 453469.
274 Ivan Kyrchei
[37] X. Liu, G. Zhu, G. Zhou, Y. Yu, An analog of the adjugate matrix for the
(2)
Outer Inverse AT ,S . Math Probl Eng. 2012, Article ID 591256, 14 pages.
[38] I. Kyrchei, Cramers rule for generalized inverse solutions. In: Ivan
I.Kyrchei (Ed.): Advances in Linear Algebra Research; New York: Nova
Sci. Publ., 2015, pp. 79132.
[39] C. D. Meyer Jr., Limits and the index of a square matrix. SIAM J Appl
Math. 1974, 26, 506515.
[45] Y. Wei, C.-W. Woo, T. Lei, A note on the perturbation of the W- weighted
Drazin inverse. Appl Math Comput. 2004, 149, 423430.
C
A
CDC, 198
accelerometers, 86, 91 challenges, 90
ADA, 217, 218 children, 146
AHRS (Attitude and Heading Reference China, 197, 199, 200
System), vii, ix, 86, 90, 91, 102, 103 color, ix, 110, 111, 112, 133, 134, 136, 137,
algorithm, 139, 155, 157, 181 138, 139, 140, 141, 142, 143, 144, 145,
angular velocity, ix, 85, 86, 91, 95, 97, 98, 146, 147, 148, 149
103, 155, 160, 161, 164, 166, 168 color image, ix, 110, 111, 112, 133, 134,
anomalous magnetic moment, vii, viii, 11, 137, 138, 139, 140, 142, 143, 147, 148,
12, 14, 42, 44, 48, 49, 51, 71, 74, 80, 82 149
arithmetic, 111 color matching, 110, 112, 141, 142, 147,
articulation, 91, 105 149
asymptotics, 36 communication, 154, 155, 167, 197
attitude, ix, 86, 87, 89, 90, 91, 93, 95, 97, complex number(s), vii, ix, 1, 2, 3, 7, 8, 49,
99, 101, 103, 105, 106, 107 89, 112, 114, 115, 117, 153, 159, 202
attitude control, ix, 85, 86, 88, 200 complexity, ix, 85, 111, 115, 116, 124
attitude estimation, ix, 85, 86, 87, 90, 102, compliance, 94
104, 107, 108 composition, 156
automotive industry, ix, 85 computation, 14
avoidance, 155, 156, 197 computer, 89, 110
computing, 161, 245
conference, 274
B congress, 197, 198
conjugation, 37, 75, 145
base, 91, 94
consensus, 155, 156, 197
Belarus, 11, 44, 47, 82
constituents, 26
278 Index
fuel consumption, 156 inequality, x, 32, 34, 41, 69, 70, 114, 153,
fusion, 99, 104 170, 171, 249
inertia, 100, 107, 185
integral backstepping (IBS), 154
G integration, 95, 97
inverse solution, x, 202, 226
Gauss-Laguerre circular harmonic function Iraq, 153
(GLCHF), ix, 109, 110, 111, 112, 117, Islam, 196
119, 120, 123, 124, 125, 128, 129, 133, Israel, 273
136 issues, 87, 90
general relativity, 49, 74, 80 Italy, 197, 198
generalized, x, 7, 12, 13, 48, 49, 75, 79, 151, iteration, 193, 194, 195, 196
201, 202, 203, 204, 210, 241, 244, 273, iterative Linear Quadratic Regulator
274, 275 (iLQR), 154
generalized inverse, x, 201, 202, 203, 204,
210, 244, 273, 274, 275
geometry, 93 J
gimbal lock, x, 103, 153
graph, 155, 197 joints, 93, 94
gravitational constant, 96 Jordan, 244, 257
gravitational field, 44, 79, 82
gravity, 86, 96, 158, 164
guidance, 90 K
kinematic, ix, 86, 87, 90, 91, 92, 93, 94,
H 157, 161, 162
knots, 100
H8 controllers, x, 187
Hamiltonian, 170, 173, 182
Hamilton-Jacobi inequality, x, 153, 171 L
hue, 146
human, 94, 142, 145, 146, 147, 149, 156 l-algebra, 1, 2, 3, 5, 6
hybrid, 104, 156, 198 lattice order(s), vii, 1, 2, 3, 5, 8
hypergeometric functions, viii, 42, 48, 80 laws, ix, 86, 100, 180, 184
leader-follower formation control, 154
least squares solution, 202
I linear model, 182
linear systems, 232, 274
IBS controllers, x, 154, 190 low-cost inertial sensors, ix, 86
identity, 2, 4, 15, 17, 18, 22, 24, 59, 61, 66, Lyapunov function, 98, 180
88, 202
iLQR controllers, x, 154, 185
image(s), ix, 110, 111, 112, 133, 134, 136, M
137, 138, 139, 140, 141, 142, 143, 144,
146, 147, 148, 149, 150, 151 magnetic field, vii, viii, 48, 49, 51, 70, 80,
incidence, 166, 178, 185, 190 96
independence, 215, 253 magnetic moment, vii, viii, 11, 12, 14, 42,
induction, 121 44, 48, 49, 51, 71, 74, 80, 82
industry, ix, 85 magnitude, 33
management, 154
manipulation, 24, 29, 61
280 Index
mass, 12, 48, 88, 90, 91, 157, 158, 162, 166, orthogonality, 78
169, 185 oscillation, 187, 190
mathematics, 110, 151
matrix, vii, viii, x, xi, 3, 5, 11, 14, 31, 42,
47, 51, 67, 72, 75, 76, 78, 79, 80, 87, 88, P
89, 92, 93, 96, 103, 110, 111, 114, 117,
118, 136, 160, 166, 179, 184, 201, 202, parallel, 94, 108, 156
203, 204, 205, 206, 207, 208, 209, 210, path tracking, 155, 200
211, 212, 214, 215, 216, 217, 218, 221, physics, 44, 82
222, 225, 227, 229, 232, 233, 234, 236, pitch, x, 86, 153, 157
237, 238, 242, 244, 248, 249, 250, 251, pitch angles rotation, x, 153
252, 253, 255, 257, 258, 263, 265, 266, plane waves, 40
268, 270, 271, 272, 273, 274 plants, 86
matrix algebra, 3, 5, 75 polar, 143, 156
measurement(s), ix, 85, 86, 91, 96, 97, 99, polarizability, 12, 44, 48, 82
104 Prediction of the Movement, 86
medium access control, 155 programming, 90
MEMS, 102 projective operators, vii, 12, 14, 48, 51, 53
methodology, 103, 107 propagation, 274
Mexico, ix, 85 proposition, 112, 218, 232, 233
minors, 209, 213, 215, 250, 252 prototype, 102
mixing, 31, 67, 72
modelling, 90, 157
models, 90, 156 Q
modulus, 113
molecular dynamics, 89 quadrotor dynamic model, x, 153
Moore-Penrose inverse, xi, 202, 203, 204, quadrotor UAVs, 154
205, 218, 242, 244, 271 quadrotors, vii, ix, x, 153, 154, 155, 156,
Moscow, 45, 83 157, 159, 161, 163, 165, 167, 169, 171,
motion control, 155, 179, 197 172, 173, 175, 177, 179, 181, 183, 185,
MPC, 155, 156, 157 187, 189, 190, 191, 193, 195, 197, 198,
multidimensional, 109, 110, 150, 151 199, 200
multiplication, 2, 88, 110, 112, 113, 114, quaternion(s), vii, ix, x, xi, 1, 2, 3, 5, 7, 8,
116, 126, 159, 257 86, 87, 88, 89, 90, 91, 92, 93, 95, 97, 98,
99, 100, 101, 103, 107, 109, 110, 111,
112, 113, 115, 116, 117, 118, 119, 120,
N 121,, 133, 136, 138, 139, 150, 151, 153,
154, 155, 156, 157, 158, 159, 160, 161,
National Academy of Sciences (NAS), 11, 162, 164, 165, 166, 167, 168, 177, 187,
44, 47, 82, 201 188, 189, 190, 191, 192, 193, 195, 196,
neutral, viii, 12, 40, 42, 48, 49, 72, 74, 80 200, 201, 202, 203, 204, 205, 206, 208,
Newtons second law, 163 209, 210, 218, 221, 232, 233, 234, 241,
nonlinear systems, 199 258, 270, 271, 272, 273, 274
null, 204, 221, 258 quaternion skew field, vii, x, xi, 201, 202,
210, 221, 232, 270, 271, 272
quaternion system, ix, 153, 271
O quaternion matrix equation(s), 202, 204,
205, 270, 271, 272
operations, 89, 95, 110, 115, 116
optimization, 99
Index 281
U W
UAVs, 154, 163, 164, 199 Washington, 197, 199
Ukraine, 201 wave equation, viii, 11, 12, 42, 43, 44, 45,
uniform, vii, viii, 11, 12, 14, 31, 42, 48, 49, 48, 49, 78, 80, 81, 82, 83
51, 80 wave function, viii, 11, 13, 14, 42, 48, 49,
uniform electric field, vii, viii, 11, 12, 14, 51, 54, 74, 80
31, 42 wavelet, 110, 150
USA, 196, 197, 198, 199 weakness, 87, 99
USSR, 43, 81 web, 93
weighted Drazin, vii, x, xi, 201, 202, 248,
270, 274
V weighted Drazin inverse, vii, x, xi, 201, 202,
248, 270, 274
variables, viii, 11, 42, 48, 54, 79, 80 weighted Moore-Penrose inverse, 202
vector, vii, viii, 11, 13, 42, 44, 47, 49, 72, W-weighted, xi, 202, 203, 204, 205, 241,
79, 80, 82, 86, 88, 89, 91, 96, 97, 100, 242, 244, 245, 246, 247, 248, 251, 254,
112, 150, 151, 157, 158, 159, 160, 161, 255, 256, 257, 258, 260, 269, 270, 271,
166, 169, 170, 175, 177, 181, 182, 206, 272, 274
209, 222, 223, 224, 259, 261, 262, 263 W-weighted Drazin inverse, xi, 202, 204,
vehicles, 197, 198 205, 241, 242, 245, 246, 247, 248, 251,
velocity, ix, 85, 86, 91, 95, 97, 98, 103, 155, 254, 255, 256, 257, 258, 260, 269, 270,
160, 161, 164, 166, 167, 168, 177, 179, 271, 272, 274
181
vision, 89, 155, 198