Beruflich Dokumente
Kultur Dokumente
discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/293821984
CITATION
READS
29
2 authors:
Hakan imek
Mustafa zdemir
Akdeniz University
Akdeniz University
13 PUBLICATIONS 17 CITATIONS
SEE PROFILE
SEE PROFILE
Mustafa Ozdemir
Hakan Simsek
Abstract
Hyperbolic rotation is hyperbolically the motion of a smooth object on general
hyperboloids given by a1 x2 + a2 y 2 + a3 z 2 = , R+ . In this paper, we
investigate the hyperbolical rotation matrices in order to get the motion of a point
about a fixed point or axis on the general hyperboloids by defining the Lorentzian
Scalar Product Space R2,1
a1 a2 a3 such that the general hyperboloids are the pseudospheres of R2,1
a1 a2 a3 . We adapt the Rodrigues, Cayley, and Householder methods to
R2,1
a1 a2 a3 and define hyperbolic split quaternions to obtain an hyperbolical rotation
matrix.
Keywords : Hyperbolical Rotation Matrix, Hyperbolic Split Quaternion, Rodrigues Formula, g-Cayley rotation matrix, g-Householder Transformation.
MSC 2010: 15A63, 15A66, 53A17, 53A35, 53B30, 70B05 , 70B10, 70E17.
Introduction
A rotation is a rigid body movement around a fixed point or an axis, which unchanges
the distance between any two points. All rotations are also a linear maps form a group
under the composition called the rotation group of a particular space. We can represent
them with the orthonormal matrices, called rotation matrices, so that the rotation group
corresponds to a special non-abelian orthogonal matrix group, denoted by SO(3) in 3
dimensional space. Rotation matrices have a lot of application areas in geometry, kinematics, physics, computer graphics, animations, and optimization problems involving the
estimation of rigid body transformations and other disciplines. Therefore, studying the
rotation matrices is an important problem in mathematics.
The rotation matrices can be obtained by using the various methods such as unit
quaternions, the Rodrigues formula, the Cayley formula, and the Householder transformation. Each unit quaternion and unit timelike split quaternion correspond to a rotation
matrix by means of a special linear map in the Euclidean and Lorentzian 3-space, respectively ([18, 19, 22, 26]). The Rodrigues formula says us that the exponential map eA
Department
of
Mathematics,
Akdeniz
University,
Antalya,
TURKEY,
e-mail:
mozdemir@akdeniz.edu.tr
generates a rotation matrix owing to the fact that this map defined from the Lie algebra
so(n) of skew-symmetric 3 3 matrices to the Lie group SO(3) is surjective , where A
is a skew symmetric matrix and is the rotation angle. In this method, only three numbers are needed to construct a rotation matrix in the Euclidean and Lorentzian 3-space
[3, 10, 12]. The Cayley formula C = (I + A) (I A)1 gives a rotation matrix without
using trigonometric functions, where A is a skew symmetric matrix [5, 14, 20, 23]. The
Householder transformation gives us a reflection matrix so that we can obtain a rotation
matrix by taking the composition of two Householder transformations [1, 11, 21, 25].
In recent times, some authors have formed the various studies related to the rotation
matrices. For example, These matrices about arbitrary lightlike axis in Minkowski 3space were studied in [13] by deriving the Rodrigues rotation formula and using the
corresponding Cayley map. In Euclidean 4-space and Minkowski space-time, they were
obtained by means of the corresponding decomposition in [7, 17]. On the other hand,
the paper [16] examined the elliptical rotation which states the motion of a point on an
ellipse or ellipsoid by defining the elliptical product space. The main idea of this paper
is to research the hyperbolical rotations occuring on the general hyperboloids a1 x2 +
a2 y 2 + a3 z 2 = r2 and find the corresponding hyperbolical rotation matrices.
The paper is organized as follows. Firstly, we define an indefinite scalar product space
2
2
denoted by Rp,q
a1 a2 an whose pseudo-spheres are the general hyperboloids a1 x1 a2 x2 +
p+q
2
2
2
= r . Then, in section 3, we find the hyperbolical
aq xq + aq+1 xq+1 + + ap+q x
rotation matrix using the Clifford algebra and the classical way in the plane Ra1,1
. Later,
1 a2
we generate the hyperbolical rotation matrices in the Lorentzian scalar product space
R2,1
a1 a2 a3 via the Rodrigues, Cayley and Hoseholders methods. Finally, we construct the
hyperbolic split quaternions and give some properties of this quaternions so that each
H-timelike hyperbolic split quaternion corresponds to an hyperbolical rotation matrix in
Ra2,1
.
1 a2 a3
Preliminaries
A scalar product g on an n-dimensional real vector space V is a nondegenerate symmetric bilinear form on V. Let B = {v1 , ..., vn } be an arbitrary basis of V. Then, g can
be expressed by n n matrix G = (gij ) , where
gij = g (vi , vj ) ,
1 i, j n.
is an indefinite scalar product called the semi-Euclidean g-inner product. The scalar
product g can be written as g (u, w) = ut w where associated matrix is given by
a1 0
0
0
0 0
..
0 a
0
.
0 0
..
...
0
0
0
.
0
..
.
.
.
.
= .
.
.
0
a
0
.
q
..
.
0 aq+1
.
.
..
.. 0
0
0
0
0 an
(1)
p
The number |det | is called constant of the scalar product and denoted by in the
rest of the paper. The Euclidean space Rn furnished with the indefinite g-inner product
is denoted by Rp,q
a1 a2 an , where p + q = n. If q = 0, then the scalar product g is reduced
to the elliptical inner product (see [16]) and if q = 1, then g is called Lorentzian g-inner
product. We say that a vector u in Rp,q
a1 a2 an is called spacelike, null (lightlike, isotropic)
or timelike if g (u, u) > 0, g (u, u) = 0 or g (u, u) < 0, respectively.
The norm of a vector
p
associated with the scalar product g is defined as kukg = |g (u, u)|. Two vectors u
and w are called g-orthogonal vectors if g (u, w) = 0. In addition, if their norms are
1, then they are called g-orthonormal vectors. If {u1 , u2 , ..., un } is an g-orthonormal
1
. In the rest of paper, we will focus the
base of Rp,q
a1 a2 an , then det (u1 , u2 , ..., un ) = ( )
2,1
n1,1
Lorentzian scalar product spaces Ra1 a2 an and Ra1 a2 a3 .
can be thought as a semi-Riemann
The Lorentzian scalar product space Ran1,1
1 a2 an
manifold or Lorentzian manifold with the scalar product g, which is indefinite metric
tensor for this manifold so that all ideas and notions in Lorentzian-Minkowski space
Rn1,1 also can be applied to Rn1,1
a1 a2 an . One can find more details about semi-Riemannian
geometry in Oneills textbook [15]. For example, a g-timecone of Rn1,1
a1 a2 an is defined as
Cg (u) = {v F : g (u, v) < 0}
for u F, where F is a set of all timelike vectors in Rn1,1
a1 a2 an . The pseudo-spheres
n1,1
of Ra1 a2 an can be described as follows: The g-hyperbolic (n 1)-space and g-de Sitter
(n 1)-space are defined by
n1,1
n1,1
Han1,1
= u Rn1,1
a1 a2 an : g (u, u) = 1 and Sa1 a2 an = u Ra1 a2 an : g (u, u) = 1 ,
1 a2 an
respectively. The pseudo-spheres in R2,1
a1 a2 a3 correspond to one sheeted and two sheeted
2
2
general hyperboloids a1 x + a2 y + a3 z 2 = 1. The hyperbolic angle between two
timelike vectors u and v which is in the same g-timecone of Rn1,1
a1 a2 an is stated as
cosh =
g (u, v)
kukg kvkg
where is compatible with the parameters of the angular parametric equations of pseudospheres of Rn1,1
a1 a2 an .
Let e1 , e2 , ..., en be standard unit vectors of Rn . For any ui = (ui1 , ui2 , ..., uin ) Rn ,
is described as
i = 1, 2, ..., n 1 the g-vector product in Ran1,1
1 a2 an
n1,1
n1,1
n1,1
Rn1,1
a1 a2 an Ra1 a2 an Ra1 a2 an Ra1 a2 an ,
(u1 , u2 , ..., un ) Vg (u1 u2 u3 un1 )
u1n
u21
u
u
u2n
22
23
Vg (u1 u2 u3 un1 ) = det
..
..
..
..
..
.
.
.
.
.
u(n1)1 u(n1)2 u(n1)3 u(n1)n
(2)
The vector Vg (u1 u2 u3 un1 ) is g-orthogonal to each of the vectors u1 , u2 ,
u3 , ..., un1 .
Let R, A, T Rnn be any matrices. Now, we will give the definitions of some kinds
of matrices in Rn1,1
a1 a2 an .
If g (Ru, Rw) = g (u, w) for all vectors u, w Rn , then R is called a semi gorthogonal matrix. These matrices preserve the norm of vectors and satisfy the
matrix equality Rt R = . Also, all rows (or columns) are g-orthogonal to each
other. We denote the set of semi g-orthogonal matrices by Og (n). It is sometimes
called the isometry group of Rn associated with scalar product g and a subgroup of
Glg (n) . If det R = 1, then we call it an hyperbolical rotation matrix or g-rotation
matrix. If det R = 1, we call it an g-reflection matrix. Although the set Og (n)
is not a linear subspace of Rnn , it is a Lie group [11]. The set of the hyperbolical
rotation matrices of Rn can be stated as
SOg (n, 1) = {R Rnn : Rt R = and det R = 1},
which is a subgroup of Og (n) .
If g (Au, w) = g (u, Aw) for all vectors u, w Rn , the matrix A is called a semi gsymmetric matrix and satisfies At = A. The set of semi g-symmetric matrices,
defined by
J = A Rnn : g (Au, w) = g (u, Aw) for all u, w Rn
is a Jordan algebra [11]. Any g-symmetric matrix in Rn1,1
a1 a2 an can be defined as
aij
(3)
A=
ai nn
where a1k = ak1 for k 6= 1, aij = aji and aij R.
If g (T u, w) = g (u, T w) for all vectors u, w Rn , then T is called a semi gskew-symmetric matrix. Also, T t = T . The set of semi g-skew symmetric
matrices, defined by
L = T Rnn : g (T u, w) = g (u, T w) for all u, w Rn
is a Lie algebra [11]. Any semi g-skew-symmetric matrix in Ran1,1
can be given
1 a2 an
as
aij
i>j
ai
aij
T = [tij ]nn with tij =
(4)
i<j
0
i=j
where a1k = ak1 for k 6= 1, aij = aji and aij R.
For the Lorentzian scalar product space Ra2,1
, the semi g-symmetric and g-skew sym1 a2 a3
metric matrices are
3.1
Rotation by Spinors
p,q
q
X
and Q (x) = x2t +
t=1
x2t for x = (x1 , ..., xp+q ) . We can express the inner product and outer product in terms
t=q+1
1
(xy + yx)
2
1
x y = (xy yx) .
2
xy =
Also, the inverse of any vector x can be defined in the Clifford algebra as the following
x1 =
x
.
x2
More information about the Clifford algebras can be found in [8, 9].
a1 a2
= gen {i, j} defined by the
In this section, we introduce the Clifford algebra Cl1,1
geometric product rules
i2 = a1 , j2 = a2 and ij = ji
. The
where i1 , j1 is the standard basis of Lorentzian scalar product plane Ra1,1
1 a2
geometric product becomes
x2 = g (x, x) ,
xy = g (x, y) + x y,
where g (x, y) = a1 x1 y1 + a2 x2 y2 for x = x1 i + x2 j and x = y1 i + y2 j, x1 , x2 , y1 , y2 R.
a1 a2
Any element of Cl1,1
, called a multivector or geometric number, has the form
s + x + tij,
a1 a2
where s, t R. In other words, the multivectors in Cl1,1
are linear combinations of
scalars (0-vector) s, vectors (1-vector) i, j, bivector (2-vector) ij.
a1 a2
by means of the Clifford algebra Cl1,1
We can study the Lorentzian g-plane Ra1,1
1 a2
by defining as R1,1
a1 a2 = {x1 i + x2 j :x1 , x2 R}. The vector x is called a spacelike vector,
lightlike (or null) vector and timelike vector if x2 > 0 or xp= 0, x2 = 0 or x2 < 0,
respectively. The norm of the vector x is described by kxk = |x2 |.
a1 a2
The hyperbolical rotation in R1,1
a1 a2 can be expressed with a spinor of Cl1,1 , is a
linear combination of a scalar and a bivector. Indeed, if we take any vector v = v1 i + v2 j
2
and a spinor B = 1 + J, where J = ji, then the geometric product of v and B is
equal to
a2
1
2
a2
a1
a1
v1
q
vB = v1 1 + v2 2 i +
v1 2 + v2 1 j =
,
a
1
v2
2
1
a2
v through . The geometric product of two spinor gives a new spinor. Thus, the spinors
a1 a2
form a subgroup of Cl1,1
.
3.2
by using (4) for the plane. The characteristic polynomial of T is P (x) = x2 1 and
T 2 = I.
Theorem 1 Let T be a semi g-skew symmetric matrix. Then, the matrix exponential
a2
sinh
cosh
a1
g
T
2
R = e = I + (sinh ) T + (cosh 1) T = a
1
cosh
sinh
a2
gives an hyperbolical rotation along the hyperbola -a1 x2 + a2 y 2 = , , a1 , a2 R+ . That
.
is, Rg is an hyperbolical rotation matrix in Ra1,1
1 a2
Example 2 Lets consider the hyperbola (H) :
() = (2 cosh , 5 sinh ). Lets take the points
x2 y 2
+
= 1 with the parametrization
4
25
u1 v1 u2 v2
+
4
25
where u = (u1 , u2 ) and w = (w1 , w2 ) so that the hyperbola (H) is the pseudo-sphere of
R1,1
1 1 . Then, = 1/10 and the semi g-skew symmetric matrix has the form
4 25
T =
0 2/5
5/2 0
cosh
Rg = eT = 5 sinh
2
.
2 sinh
5
.
cosh
such that det Rg = 1 and (Rg )t (Rg ) = are satisfied. For cosh 0 = 2, we get
2
2
3/5
g
R0 =
.
5 3/2
2
So, if we rotate the point A hyperbolically on (H), we get B = Rg0 (A) = (4, 5 3) (see
Figure 1). Thus, we get the same result using the hyperbolical rotation matrix for (H) .
g (x, y)
= 2.
kxkg kykg
It can be seen that the hyperbolical rotation matrix Rg can be also used to interpret the
x2
y2
motion on a similar hyperbola
+
= 1 to (H)
16
100
4
4.1
where =
a1 a2 a3 . The matrix
0
u3 /a1 u2 /a1
0
u1 /a2
A = u3 /a2
u2 /a3 u1 /a3
0
(5)
2,1
is semi g-skew symmetric matrix in R2,1
a1 a2 a3 . So, the g-vector product in Ra1 a2 a3 can be
viewed as a linear transformation, which corresponds to multiplication by a semi g-skew
symmetric matrix. Since the characteristic polynomial of A is P (x) = x3 hu, ui x, its
eigenvalues are x1 = 0, x2,3 = 1 and A3 = A if the vector u is a unit spacelike vector,
and x1 = 0, x2,3 = i and A3 = A if the vector u is a unit timelike vector. Therefore,
.
we can state the Rodrigues formula in Ra2,1
1 a2 a3
u3 sinh
u2 sinh
-a1 u21 + (a1 u21 + 1) cosh
-a
u
u
(cosh
-1)
-a
u
u
(cosh
-1)
2
1
2
3
1
3
a
a
1
1
u1 sinh
2
2
u3 sinh + a1 u1 u2 (cosh -1)
a
u
+
(1-a
u
)
cosh
u3 sin
u2 sin
-a
u
u
(1cos
)
-a
u
u
(1cos
)
a1 u21 + (1-a1 u21 ) cos
2
1
2
3
1
3
a
a
1
1
u3 sin + a1 u1 u2 (1- cos )
- u1 a2sin -a3 u2 u3 (1- cos )
a2 u22 + (1-a2 u22 ) cos
a2
2!
3!
4!
5!
6!
3 5
2 4 6
2
=I +A +
+
+A
+
+
3!
5!
2!
4!
6!
= I + (sinh ) A + (cosh 1) A2 .
Expanding this equation, the hyperbolical rotation matrix can explicitly written as (6).
Similarly, we get the hyperbolical rotation matrix (7).
9
x2 y 2
+
+ z 2 = 1 parameterized by
9
4
B = ln(2 + 3) + ln 4 3 + 7 ,
= 3 3, 2, 3
3
on this hyperboloid. We will find the hyperbolicalrotation
matrix which rotates the point
A to B through the hyperbolic angle 1 = ln 4 3 + 7 . First, using the vector product
of x = OA and y = OB in R2,1
1 1 , we find
1
9 4
9i 4j k
1
Vg (x y) = 3 3 2 3 = 0, 12, 2 3 .
6
3 3 2
3
(xy)
So, the unit spacelike vector can be stated u = kVVgg (xy)k
in R2,1
1 1 , which is the Lorentzian
1
9 4
spacelike rotation axis. This gives the Lorentzian hyperbolical rotation matrix
3 3
3
cosh
sinh
sinh
4
2
by using (6). This matrix describes an hyperbolical rotation on a great general hyperbola
which is intersection of the hyperboloid and
through the origin and g the plane passing
g,u
orthogonal to u. This plane is given by 2 3y = z. Then, R represents an hyperbolical
y2
x2
= 1, 2 3y = z. If we take
rotation over the great general hyperbola given by +
9
4/49
the hyperbolical rotation angle 1 , then we find
7
3
3
18
(8)
Rg,u
= 4 3/3
5/2
3 3 .
1
2
3 3/4 11/2
The matrix (8), rotates the point A to the point B hyperbolically over the great hyperbola
x2
y2
+
= 1, 2 3y = z (see Figure 2).
9
4/49
Remark 5 The eigenvalues of the matrix (6) are x1 = e , x2 = e and x3 = 1. Also, the
eigenvector corresponding to 1 is the vector u, the rotation axis of the motion. Similarly,
The eigenvalues of the matrix (7) are x1 = ei , x2 = ei and x3 = 1 and the eigenvector
corresponding to 1 is the vector u.
10
4.2
Cayleys Method
In this section, we study the hyperbolical rotation matrix obtained by Cayley transformation in R2,1
a1 a2 a3 with the scalar product g.
The Cayley transformation gives a parameterization of rotation matrices without
using trigonometric functions. Let A be an n n semi g-skew symmetric matrix given
by (5). The determinant of I A and I + A are 1 g ( u, u) . Thus, we have to suppose
g (u, u) 6= 1 in order that I A or I + A is invertible so that we can define the Cayley map
which transforms the matrix A into (I + A) (I A)1 or (I A) (I + A)1 . SOg (1, n) is
isomorphic to sog (1, n) via the Cayley formula where sog (1, n) denotes the space of semiskew symmetric matrices usually associated with the Lie algebra of the transformation
group defined by rotations in SOg (1, n). The g-Cayley map is eventually defined as
Cayg : sog (1, n) SOg (1, n)
A Cayg (A) = (I + A) (I A)1 .
or its equivalent form
Cayg (A) = (I A) (I + A)1 .
Theorem 6 Let A be a semi g-skew symmetric matrix in the form (5) such that a nonnull
vector u = (u1 , u2 , u3 ) R2,1
a1 a2 a3 satisfies g (u, u) 6= 1. Then,
Rg,u = (I + A) (I A)1
is an hyperbolical rotation matrix on the hyperboloids given by a1 x2 + a2 y 2 + a3 z 2 = 1
where u is the rotation axis. Furthermore, the matrix Rg,u can be written in the form
2 u3
a1 u21 + a2 u22 + a3 u23 + 1
2a2 u1 u2
2a1u2 2a3 u1 u3
a
1
2 u3
1
.
+ 2a1 u1 u2
a1 u21 a2 u22 + a3 u23 + 1
2a2u1 2a3 u2 u3
a2
1g(u,u)
2 u1
2a3u2 + 2a1 u1 u3
2a2 u2 u3
a1 u21 + a2 u22 a3 u23 + 1
a3
(9)
11
and
3 = 1.
Also, the eigenvector corresponding to 1 is u, which is the rotation axis. If the rotation
axis u is a timelike vector, the eigenvalues can be given by
1 =
1 + i kukg
,
1 i kukg
2 =
1 i kukg
1 + i kukg
and
3 = 1.
The matrix (9) rotates a vector hyperbolically on the hyperboloid about the axis u =
(u1 , u2 , u3 ). The hyperbolic rotation angle is given by
cosh =
1 + g (u, u)
1 g (u, u)
(10)
cos =
1 + g (u, u)
1 g (u, u)
(11)
if u is spacelike vector, or
if u is timelike vector.
Example 8 Lets find the Cayley hyperbolical rotation matrix representing an hyperbolical
x2
z2
rotation on the hyperboloid
+ y2 +
= 1 about the axis u = (4, 1, 3) . Using the
4
9
matrix (9), we find
7 4 4
1
3 3 2 .
3
3 6 3
The hyperbolical rotation angle corresponding to this matrix can be found as cos = 1/3
from (11).
12
4.3
Householders Method
2uut
v.
ut u
2a2 u1 u2
2a3 u1 u3
a1 u21 + a2 u22 + a3 u23
1
.
2a1 u1 u2
a1 u21 a2 u22 + a3 u23
2a3 u2 u3
Hu =
g (u, u)
2
2
2
2a1 u1 u3
2a2 u2 u3
a1 u1 + a2 u2 a3 u3
(12)
The matrix Hu is a semi g-symmetric, semi g-orthogonal and involutory matrix; namely,
it is satisfied the equations Hut = Hu , Hut Hu = and Hu2 = I, respectively.
Also, the determinant of Hu is 1 so that it describes an hyperbolical g-reflection about
a plane passing through the origin g-orthogonal to v. Using the matrix (12), we can
express hyperbolical g-reflections on any hyperboloid.
Example 9 Lets find the hyperbolical g-reflection matrix which reflects a point hyperbolically on the hyperboloid x2 + 3y 2 + 2z 2 = 1, about the plane which is passes through the
origin and is g-orthogonal to u = (1, 2, 1) . Using the matrix (12), we obtain
15/13 12/13
4/13
Hu = 4/13 11/13 8/13 .
2/13 12/13 9/13
The g-reflection of A (1, 0, 1) associated with u is found as B(11/13, 4/13, 11/13).
Lets check it : The middle point of [AB] is (12/13, 2/13, 12/13) and the equation of
the plane is g (u, x) = 0, which is x + 6y + 2z = 0 where x = (x, y, z). Then, the middle
point lies on this plane and the vector AB is g-orthogonal to the plane.
Theorem 10 Let a, b R2,1
a1 a2 a3 be any two nonnull vectors where a 6= b and kakg = kbkg
. Then, there exists a unique g-Householder transformation H such that H (a) = b.
2,1
Proof. Consider the map H : R2,1
a1 a2 a3 Ra1 a2 a3 defined as
if b = a
Ha (v)
Hab (v)
if b 6= a and a b is nonnull .
H (v) =
Ha+b (v)
if b =
6 a and a b is null
13
If b = a, we can find
H (a) = Ha (a) = a
2aat
a = a = b.
at a
Similarly, we can get Ha+b (a) = b for the last case which is b 6= a and a b is null.
Consequently, the g-Householder transformation H is unique map satisfies H (a) = b.
Proposition 11 Product of two g-reflection matrices is an hyperbolical rotation matrix.
Proof. Let Hu and Hv be two g-reflection matrices with the reflection planes u and
v , respectively. For the matrix R = Hu Hv , we have det R = 1 and
Rt R = Hvt Hut Hu Hv = Hvt Hv = ,
which means that R is an hyperbolical rotation matrix.
Theorem 12 Let a, b R2,1
a1 a2 a3 be any two nonnull vectors where a 6= b and kakg =
kbkg . Then, there exists a unique hyperbolical rotation matrix R constructed by two greflection matrices such that R (a) = b.
2,1
Proof. Consider the map R : R2,1
a1 a2 a3 Ra1 a2 a3 defined as the following
if b = a
Hb (Ha (v))
H
(H
(v))
if b 6= a and a b is nonnull .
R (v) =
b
ab
Hb (Ha+b (v))
if b 6= a and a b is null
(13)
6 3 4
17 0 24
1
Hab = 1 2 4 and Hb = 0 1 0
5
2 6 3
12 0 17
14
54 93
1
1 2
Hb Hab =
5
38 66
matrix is
4
4
3
4.4
Quaternions were discovered by Sir William R. Hamilton in 1843 and the theory
of quaternions was expanded to include applications such as rotations in the early 20th
century. The most important property of the quaternions is that every unit quaternion
represents a rotation and this plays an important role in the study of rotations in 3dimensional vector spaces. Using unit quaternions is a useful, natural, and elegant way to
perceive rotations. Quaternions are used especially in computer vision, computer graphics,
animation, and kinematics.
b is an associative, non-commutative non-division
The algebra of split quaternions H
ring with four basic elements {1, i, j, k} satisfying the equalities i2 = 1, j2 = k2 = ijk =
1. Any split quaternion q is represented by q = (q0 , q1 , q2 , q3 ) = q0 + q1 i + q2 j + q3 k or
q = Sq + Vq where the symbols Sq = q0 and Vq = q1 i + q2 j + q3 k denote the scalar and
vector parts of q, respectively. If Sq = 0 then q is called a pure split quaternion. The
conjugate of q is denoted
p defined as q = Sq Vq . The norm of a quaternion
pby q, and
Nq is defined by Nq = |qq| = |q02 + q12 q22 q32 |. If Nq = 1, q is called unit split
quaternion and q0 = q/Nq is a unit split quaternion if Nq 6= 0. A split quaternion is
called spacelike, timelike or null (lightlike) if Iq < 0, Iq > 0 or Iq = 0, respectively where
Iq = q02 + q12 q22 q32 . The set of timelike split quaternions are denoted by
b = {q = (q1 , q2 , q3 , q4 ) : q1 , q2 , q3 , q4 R, Iq > 0} .
TH
Any timelike split quaternion can be represented in polar form similar to quaternions as
follows:
i) Every timelike quaternion with spacelike vector part can be written in the form
q = Nq (cosh + ~0 sinh )
where cosh =
q1
,
Nq
sinh =
q2 i+q3 j+q4 k
~0 =
is a spacelike unit vector
2
2
2
q2 +q3 +q4
where cos =
q1
,
Nq
sin =
2 i+q3 j+q4 k
~0 = q
is a timelike unit vector in R2,1
2
2
2
q2 q3 q4
Let q and r be two timelike split quaternions. Then, the linear transformation Rq :
b
b defined by Rq (r) = qrq 1 is a timelike split quaternion whose norm and scalar
TH TH
are the same as for r. Since the scalar part of the quaternion r doesnt change under
Rq , we will only examine how its vector part Vr changes under the transformation Rq .
We can interpret the Lorentzian rotation of a vector in the Minkowski 3-space using the
quaternion product qVr q 1 . Namely, if q is a timelike split quaternion, the corresponding
Lorentzian rotation matrix can be found as
2
q1 + q22 + q32 + q42
2q1 q4 2q2 q3
2q1 q3 2q2 q4
q12 q22 q32 + q42 2q3 q4 2q2 q1 .
(4)
Rq = 2q2 q3 + 2q4 q1
2
2
2
2
2q2 q4 2q3 q1
2q2 q1 2q3 q4
q1 q2 + q 3 q4
This rotation matrix represents a rotation through angle 2 about the axis = (q1 , q2 , q3 )
(see [19]).
4.4.1
We define the set of hyperbolic split quaternions suitable for the hyperboloid a1 x2 +
a2 y 2 + a3 z 2 = 1 in order to get an hyperbolical Lorentzian rotation matrix. Lets take
four basic elements {1, i, j, k} satisfying the equalities
i2 = a1 ,
j 2 = a2 ,
k2 = a3
and
ij =
k = ji,
a3
where a1 , a2 , a3 R+ and =
jk =
i = kj,
a1
ki =
j = ik
a2
a1 a2 a3 .
b a a a . This set is an
The set of hyperbolic split quaternions will be denoted by H
1 2 3
associative, non-commutative non-division ring with four basic elements {1, i, j, k}. Scalar
and vector parts of hyperbolic split quaternion q = q0 + q1 i + q2 j + q3 k are denoted by
Sq = q0 and Vq = q1 i + q2 j + q3 k, respectively. If we take a1 = a2 = a3 = 1, we get the
usual split quaternion algebra. The hyperbolic quaternion product table is given as the
following
1
i
j
k
1
i
1
i
i
a1
j k/a3
k
j/a2
j
j
k/a3
a2
i/a1
k
k
j/a2
i/a1
a3
16
(14)
where g (Vp , Vq ) and Vg (Vp Vq ) are the g-scalar product and the g-vector product,
respectively. If p and q are pure hyperbolic split quaternion, then
pq = g (Vp , Vq ) + VL (Vp Vq )
i/a1 j/a2 k/a3
p2
p3 .
= a1 p1 q1 + a2 p2 q2 + a3 p3 q3 + p1
q1
q2
q3
b a a a can be expressed
The hyperbolic quaternion product for H
1 2 3
p0 a1 p1 a2 p2
a3 p 3
p3 p2 q
p1
0
p
0
a
a
1
1
q1
p1
p3
pq =
p0
p2
q2
a2
a
2
q3
p2 p1
p3
p0
a3
a3
as
p0 2p1 2p2 p3
q0
p1
p0
p3 p2
q1 .
pq =
p2
p3
p0 p1 q2
p3 2p2 2p1 p0
q3
For p = 1 + 3j 4k and q = 2 + 4i + 3k, we get pq = (38, 2, 25, 11) .
Remember that the algebra is formed by a vector space V equipped with a quadratic form
Q with the following equalities
v2 = Q (v) ;
uv + vu = 2BQ (u, v)
is called a Clifford algebra and is denoted by C` (V, Q) . If {e1 , ..., en } is a base for an
n-dimensional vector space V, then C` (V, Q) is formed by the multivectors
{1} {ei1 ei2 ...eik : 1 i1 ... ik n, 1 k n}
with dim (C` (V, Q)) = 2n . Since the Clifford product of two even multivectors is an even
multivector, they define an even subalgebra of C` (V, Q). The even subalgebra of an ndimensional Clifford algebra is isomorphic to a Clifford algebra of (n 1) dimensions and
b is isomorphic with the even
it is denoted by C`+ (V, Q) . The split quaternion algebra H
+
2,1
2
2
2
subalgebra C`2,1 = C` (R , Q = x1 + x2 + x3 ) by {1, e2 e3 i, e3 e1 k, e1 e2 j}
[22]. Similarly, the hyperbolic split quaternion algebra is an even subalgebra of the Clifford
algebra
C` (R3 ) = {q = q0 + e1 q1 + e2 q2 + e3 q3 : e21 = a1 , e22 = a2 , e23 = a3 , ei ej + ej ei = 0}
17
associated with the nondegenerate quadratic form Q (x) = a1 x21 + a2 x22 + a3 x23 and is
b a a a is isomorphic
.H
denoted by C`+ (R3 , a1 x21 + a2 x22 + a3 x23 ) , or shortly C`+ Ra2,1
1 2 3
a
a
1 2 3
a
a
a
1
2
3
+
2,1
to C` Ra1 a2 a3 with {1, e2 e3 i, e3 e1 k, e1 e2 j}.
The quadratic form Q = a1 x21 + a2 x22 + a3 x23 gives the g-inner product by using the
equality
1
BQ (x, y) = [Q (x + y) Q (x) Q (y)] .
2
Namely , we get BQ (x, y) = g (x, y) for x, y R3 . Thus, we can construct an hyperbolic
quaternion algebra for any g-inner product space.
Conjugate, norm and inverse of an hyperbolic split quaternion q = q0 + q1 i + q2 j + q3 k
can be defined similar to usual split quaternions :
q
p
p
q
q = q0 q1 i q2 j q3 k, Nq = |qq| = |qq| = |q02 + a1 q12 a2 q22 a3 q32 |, q 1 =
Jq
where Jq = q02 + a1 q12 a2 q22 a3 q32 . A hyperbolic split quaternion is called H-spacelike,
H-timelike or H-null if Jq < 0, Jq > 0 or Jq = 0, respectively.
The vector part of any H-spacelike quaternion is spacelike since q02 < a1 q12 + a2 q22 +
a3 q32 = g (Vq, Vq) . But, the vector part of any H-timelike quaternion can be spacelike,
timelike or null. Any hyperbolic split quaternion q can be expressed in polar form similar
to split quaternions as the following:
i) Every H-spacelike hyperbolic split quaternion can be witten in the form
q = Nq (sinh + 0 cosh )
where sinh =
q0
,
Nq
cosh =
q1 i+q2 j+q3 k
a1 q12 +a2 q22 +a3 q32
0 =
is a spacelike
vector in
R2,1
a1 a2 a3 .
The vector 0 is called the axis of the hyperbolical rotation. For example, if q =
b 212 , then q is a H-timelike hyperbolic split quaternion with timelike
1 + 2i + j k H
q=
!
(2, 1, 1)
1 (2, 1, 1) 5
+
= 6 cos +
sin
6
5
6
5
1
where 0 = (2, 1, 1) is a unit timelike vector in R3212 .
5
The set of H-spacelike split quaternions is not a group since it is not closed under
multiplication. On the other hand, the set of H-timelike hyperbolic split quaternions
denoted by
b a a a = {q = (q1 , q2 , q3 , q4 ) : q1 , q2 , q3 , q4 R, Jq > 0}
TH
1 2 3
forms a group under the hyperbolic split quaternion product.
Theorem 14 Each a H-timelike unit hyperbolic split quaternion q generates to an hyperbolical rotation on the an general hyperboloid. The hyperbolical hyperbolical rotation
matrix to corresponding to the hyperbolic split quaternion q is
q0 q3
2q0 q2
2
2
2
2
2
2a2 q1 q2
a1 2a3 q1 q3
q0 +a1 q1 + a2 q2 + a3 q3
a
1
q0 q3
2q0 q1
2
2
2
2
Rq = 2a1 q1 q2 + 2
a2 2a3 q2 q3
q 0 a1 q 1 a2 q 2 + a3 q 3
a2
q0 q2
q0 q1
2
2
2
2
2a1 q1 q3 2
2
2a2 q2 q3
q0 a1 q1 + a2 q2 a3 q3
a3
a3
(15)
Proof. Let q be a H-timelike hyperbolic split quaternion. The transformation Rq :
b a a a TH
b a a a defined by Rq (u) = quq 1 is a linear transformation preserves the
TH
1 2 3
1 2 3
norm and scalar part of u. Therefore, we examine only how the vector part of H-timelike
hyperbolic split quaternion u change under R. Using the equalities,
r
r
a1 a3
a1 a2
2
2
2
2
j+2 q1 q3 a1 q0 q2
k,
R (i) = q0 +a1 q1 + a2 q2 + a3 q3 i + 2 a1 q1 q2 +q0 q3
a2
a3
q0 q3
R (j) = 2
a1 a2 a3 a2 q1 q2 i+ q02 a1 q12 a2 q22 + a3 q32 j
a1
q0 q1
a1 a2 a3 a2 q2 q3 k,
+2
a3
q0 q 2
q0 q 1
R (k) = 2 a3 q1 q3 +
a1 a2 a3 i2 a3 q2 q3 +
a1 a2 a3 j
a1
a2
+ q02 a1 q12 + a2 q22 a3 q32 k,
we can obtain (15). Since the equalities det Rq = 1 and Rtq Rq = are satisfied, the
matrix (15) is an hyperbolical hyperbolical rotation matrix on the hyperboloid a1 x2 +
a2 y 2 + a3 z 2 = 1. Also, note that, if we take a1 = a2 = a3 = 1, the standard rotation
matrix is obtained in Lorentzian 3-space R2,1 .
Theorem 15 i) Let q = cosh + 0 sinh be a unit H-timelike hyperbolic split quaternion
with spacelike vector part and v be an hyperbolic Lorentzian vector. Then, Rq (v) = qvq 1
is a hyperbolical rotation through hyperbolical angle 2 about the H-spacelike axis 0 .
ii)Let q = cos + 0 sin be a unit H-timelike hyperbolic split quaternion with timelike
vector part and v be a hyperbolic Lorentzian vector. Then, Rq (v) = qvq 1 is a hyperbolical
rotation through 2 about the timelike axis 0 .
19
2
22q0 q2 8q1 q3
q0 +2q12 + q22 + 4q32 2 2q0 q3 2q1 q2
Rq = 4q1 q2 + 4 2q0 q3 q02
2q12 q22 + 4q32 4 2q0 q1 8q2 q3 .
2q0 q1 2q2 q3
q02 2q12 + q22 4q32
4q1 q3 2q0 q2
For example, the unit H-timelike split quaternion q = (0, 1, 0, 1/2) represents an hyperbolical rotation on the hyperboloid 2x2 + y 2 + 4z 2 = 1 through the angle /2, about the
timelike axis (1, 0, 1/2) . Then, the hyperbolical rotation matrix is given by
3 0 4
Rq = 0 1 0 .
2 0 3
20
Concluding Remarks
The present article introduces the methods of obtaining the hyperbolical rotation
matrices which give the rotation motion on any general hyperbolas a1 x2 + a2 y 2 = 1
and hyperboloids a1 x2 + a2 y 2 + a3 z 2 = 1. These methods are the generalization of
the Rodrigues, Cayley and Householder methods to the Lorentzian scalar product space
1,1
R2,1
a1 a2 a3 . Besides, for the Lorentzian scalar product plane Ra1 a2 , the rotation is constructed
a1 a2
by spinors using the geometric product in Cl1,1
. On the other hand, the notion of
hyperbolic split quaternion are introduced and the hyperbolical rotation matrices are
generated by using H-timelike unit hyperbolic split quaternions. In all methods, the
rotation axis is thought as a nonnull vector.
The rotation matrices about arbitrary null axis in Lorentzian 3-space were examined
by [13] using the Rodrigues rotation formula and the corresponding Cayley map. Thus,
these matrices can be studied about the null axis over the pseudo-spheres of the Lorentzian
scalar product space R2,1
a1 a2 a3 .
References
[1] Aragon-Gonzalez G., Aragon J.L., Rodrguez-Andrade M. A., The decomposition of
an orthogonal transformation as a product of reflections, J. Math. Phys. 47, Art. No.
013509 (2006)
[2] Aragon-Gonzalez G., Aragon J.L., Rodrguez-Andrade M. A., Verde Star L., Reflections, Rotations, and Pythagorean Numbers. Adv. Appl. Clifford Algebras 19, 1-14
(2009)
[3] Brackett R. W., Robotic Manipulators and the Product of Exponentials Formula.
Mathematical Theory and Networks and Systems. Proceeding of International Symposium. Berlin, Springer-Verlag, 120-127 (1984)
[4] B
ukcu
B., On the Rotation Matrices in Semi-Euclidean Space. Commun. Fac. Sci.
Univ. Ank. Series A1. 55, 7-13 (2006)
21
[5] Cayley A., Sur Quelques Proprietes des Determinants Gauches. The Collected Papers
of Arthur Cayley SC.D.F.R.S. Cambridge University Press (1889)
[6] Gallier J. H., Geometric Methods and Applications, For Computer Science and Engineering. Texts in Applied Mathematics 38, 680p (2011)
[7] Gallier J., Xu D., Computing exponetials of skew symmetric matrices and logarithms
of orthogonal matrices. Int. J. Robot. Autom. 17(4), 111 (2002)
[8] Hestenes D., New Foundations for Classical Mechanics, Kluwer Academic Publisher,
Second Edition (1999)
[9] Hestenes D., Sobczyk G., Clifford Algebra to Geometric Calculus: A Unified Language for Mathematics and Physics, Kluwer Academic Publishing, Dordrecht (1987)
[10] Kula L. ,Karacan M.K., Yayl Y., Formulas for the Exponential of Semi Symmetric Matrix of order 4. Mathematical and Computational Applications 10, 99-104m
(2005)
[11] Mackey D. S., Mackey N., Tisseur F., G-reflectors : Analogues of Householder transformations in scalar product spaces, Linear Algebra and its Applications Vol. 385,
187213 (2004)
[12] Murray M. R., Li Z.,Sastry S.S.i, A Mathematical Introduction to Robotic Manipulation. Boca Raton F.L. CRC Press (1994)
[13] Nesovic E., On Rotation About Lightlike Axis in Three-Dimensional Minkowski
Space, Adv. Appl. Clifford Algebras (2015). doi:10.1007/s00006-015-0601-6
[14] Norris A. N., Euler-Rodrigues and Cayley Formulae for Rotation of Elasticity Tensors. Mathematics and Mechanics of Solids 13, 465-498 (2008)
[15] ONeill B., Semi-Riemannian Geometry with Application to Relativity, Pure and
Applied Mathematics, vol. 130, Academic Press (1983)
[16] Ozdemir
M., An Alternative Approach to Elliptical Motion. Adv. Appl. Clifford
Algebras, doi:10.1007/s00006-015-0592-3 (2015)
[17] Ozdemir
M., Erdogdu M., On the Rotation Matrix in Minkowski Space-time. Reports
on Mathematical Physics 74, 27-38 (2014)
[18] Ozdemir
M., Erdogdu M., Simsek H., On the Eigenvalues and Eigenvectors of a
Lorentzian Rotation Matrix by Using Split Quaternions. Adv. Appl. Clifford Algebras
24, 179-192 (2014)
[19] Ozdemir
M., Ergin A.A., Rotations with unit timelike quaternions in Minkowski
3-space. Journal of Geometry and Physics 56, 322-336 (2006)
[20] Ozkald
S., G
undogan H., Cayley Formula, Euler Parameters and Rotations in 3Dimensional Lorentzian Space. Advances in Applied Clifford Algebras 20, 367-377
(2010)
22
[21] Rodrguez-Andrade M.A., Aragon-Gonzalez G., Aragon J.L., Verde-Star L., An algorithm for the Cartan-Dieudonne theorem on generalized scalar product spaces,
Linear Algebra and Its Applications, Vol. 434, Issue 5, 1238-1254 (2011)
[22] Schmidt J., Nieman H., Using Quaternions for Parametrizing 3-D Rotations in Unconstrained Nonlinear Optimization, Vision Modeling and Visualization. Stuttgart,
Germany, 399-406 (2001)
[23] Selig J. M., Cayley Maps for SE(3). 12th IFToMM World Congress, Besancon, 18-21
(2007)
[24] Tarrida A. R., Affine Maps, Euclidean Motions and Quadrics, Srpinger, 411p (2011)
[25] Uhlig F., Constructive ways for generating (generalized) real orthogonal matrices as
products of (generalized) symmetries, Linear Algebra Appl. 332334, 459467 (2001)
[26] Vicci L., Quaternions and Rotations in 3-Space, The Algebra and Geometric Interpretation. Microelectric Systems Laboratory, Department of Computer Sciences,
UNC, TR01-014 (2001)
23