Advanced Dynamics

J. Kovecses

Dynamics of Mechanical Systems

Part 1
Jozsef Kovecses
Department of Mechanical Engineering and Centre for Intelligent Machines, McGill University
Classical Mechanics is a field of physical science, which deals with the motion of objects in our
three-dimensional physical space, and the interaction among these objects. This field characterizes
the phenomena observed at our usual scale of space and time, ranging from the scale of molecules
to planetary objects. It is also important to understand that just like any other field of physics
it only provides approximations for the descriptions of physical phenomena; it relies on a set of
principles and axioms that cannot be exactly proven, but have not been contradicted by physical
experiments. These principles and axioms allow for the development of models that capture certain
features of observed physical phenomena, and also make it possible to introduce mathematical
representations for analysis.

Fundamental Observations About Motion

Geometry is a pre-requisite to Mechanics. The understanding of the geometry related to our

three-dimensional physical space is necessary to interpret the motion of physical objects and the
interaction between them. We intend to build up a framework that relies on basic everyday
observations understandable to everybody, and these observations can be reproduced by simple
experiments in common everyday physical settings.
The way how the motion of an object is seen in physical space depends on the position and
motion of the observer. This requires the interpretation of the observers frame of reference; this is
a key notion to describe motion. Each observer can be attached to a different frame of reference.
A frame of reference is an infinite dimensional geometric object, with no particular shape.1 We
will use notation F with appropriate index to represent a frame of reference.2 To parameterize
the frame of reference a coordinate system can be defined with origin at an arbitrary point of the
frame. Such a coordinate system can be used to identify the location of any point of the frame
of reference and also its orientation. An infinite number of coordinate systems can be defined for
the parameterization of a frame of reference. These coordinate systems can all be transformed to
each other via constant transformations. We will see this later in more detail.
The notion of a coordinate system is not essential to the physical meaning of a frame of
reference. It is only needed to parameterize the original geometric object and make it possible
to establish algebraic representations to describe the geometric concepts in a processable form.
But, often it is easier to think about a coordinate system than the more abstract geometric concept
of the frame of reference.
There are two important types of motion we normally identify in everyday life: the motion of a
point relative to a reference frame; and the motion of two reference frames relative to each other.
1

For example, four non-coplanar points that are always constant distance of each other can interpret such frame
of reference.
2
In the following we will often refer to a frame of reference simply as a frame.

Advanced Dynamics

J. Kovecses

To observe the position and motion of a point with respect an observer two types of information
are needed in general: magnitude and direction. For example, in the case of the position, the
distance and the direction along which the distance is measured represent these two pieces of
information. In the case of velocity and acceleration, the two pieces of information needed to
completely define them are their magnitudes and directions. Therefore, such physical quantities
cannot mathematically be represented by scalars alone. They can be represented by physical
vectors; such a quantity has two parts: a scalar magnitude and a unit vector representing its
direction.
A physical vector can be considered as a geometric object and can be denoted as ~v. It can be
decomposed in different ways. One way is to directly represent the magnitude and the direction
as
~v = v~ev
(1)
where v represents the magnitude, which is a scalar value, and ~ev indicates the geometric unit
vector giving the direction of the physical vector. This is illustrated in Fig. 1, the orientation of
the unit vector also defines the line of action of the vector. There is also a third part of a physical
line of
action
~v = v~ev
~ev

Figure 1: Representation of a physical vector

vector: the physical unit. For now we assume that if not indicated separately then the physical
unit is part of the magnitude. Later on we will revisit this. The concepts of physical and geometric
vectors are only different in that a physical vector carries a physical unit while a geometric vector
may not need to carry unit and may only indicate a direction in space.

Physical Vectors

Here we will use arbitrary physical vectors, ~a, ~b, ~c, ~d for illustration. In each subsections these
may carry different meaning.

2.1

Addition

Let us think of a displacement of a point as a typical representation for physical vectors. We

consider two displacements represented by vectors ~a and ~b as shown in Fig. 2. The addition of
these two vectors, ~c = ~a + ~b is naturally, geometrically, defined as shown in the figure. Based on
this ~a and ~b can also be called the vector components of ~c.

Advanced Dynamics

J. Kovecses
~b
~a

~a
~c =

~
+b

Figure 2: Representation of vector addition

2.2

Concepts of Product

Let us consider again two such physical vectors ~a and ~b, which may also be represented with
displaying the magnitude and direction explicitly as ~a = a~ea and ~b = b~eb . The concept of product
between geometric vectors can be interpreted in different ways. Let us look at the geometric
representations shown in Figs. 3-5. The scalar product results in a scalar value, cross (or vector)
product gives a vector, and the dyadic product produces a higher order geometric object that will
be referred to as a dyadic or second order tensor.
2.2.1

Scalar Product

Let us go back to the geometric representation, Fig. 3, and consider ~a and unit vector ~eb . Vector
~a can be decomposed to two vector components, one along ~eb and another perpendicular to ~eb .
Based on the geometry of triangles we can immediately see that the magnitude of the component
along ~eb equals a cos where is the angle between ~a and ~eb . This gives the possibility to

~a

~eb

a cos =

~a ~eb

Figure 3: Scalar product

define a product concept, the scalar product, between ~a and ~eb where the result is a cos , i.e., the
magnitude of the vector component of ~a along ~eb , as
~a ~eb = a~ea ~eb = a cos

(2)

where indicates the scalar product operator. This product is also often called the inner or dot
product. We can also see that the scalar product operation essentially comes down to operations
between the geometric unit vectors and the angle defined by them,
~ea ~eb = cos
and based on simple geometric interpretation, Fig. 3, the order can be changed, i.e., the product
is commutative
~eb ~ea = cos
3

Advanced Dynamics

J. Kovecses

Based on this the scalar product between two arbitrary vectors can be interpreted as
~a ~b = ~b ~a = ab~ea ~eb = ab cos

(3)

Using this interpretation the scalar product of two vectors is zero if they are perpendicular, or
orthogonal in other words, to each other, and it equals to the product of their magnitudes if they
are collinear.
2.2.2

Cross Product

Let us again consider two physical vectors ~a and ~b as shown in Fig. 4. The two vectors define a
plane. Based on this the concept of the cross, or vector, product
~a ~b = ~c

(4)

can be interpreted to result in a third vector, ~c that is perpendicular to the plane so that the three
vectors ~a, ~b, ~c form a right-handed vector triad, and the magnitude of ~c equals the area of the
parallelogram ~a and ~b span
~c = ab sin ~ec
(5)
where c = ab sin is the magnitude of the vector ~c and ~ec is the unit vector normal to the plane
defined by ~a and ~b. The cross product of two collinear vectors is zero due to that the area
area of parallelogram: c
~c = ~a ~b
= c ~ec

~b

~a

Figure 4: Cross product

spanned by the two vectors is zero. Using the right-hand rule we can see that the cross product
is anti-commutative
~b ~a = ~d
(6)
Orthogonal Unit Vectors: A particularly important case when we consider the cross product
of two perpendicular unit vectors, ~g1 and ~g2 . Based on the above interpretation and considering
that the two unit vectors form a right angle we obtain
~g1 ~g2 = ~g3

(7)

where ~g3 is also a unit vector and the three unit vectors form an right-handed orthonormal vector
triad that is often used to define an orthogonal Cartesian coordinate system. Considering ~g1 , ~g2 ,
~g3 it can be easily shown that
~g2 ~g3 = ~g1 ,

~g3 ~g1 = ~g2 ,

~g2 ~g1 = ~g3 ,

4

~g3 ~g2 = ~g1 ,

~g1 ~g3 = ~g2

(8)

Advanced Dynamics

J. Kovecses

These relations can serve as basis to evaluate the cross product between two vectors when the
vectors are decomposed with respect to an orthonormal set of base vectors.
We note that in the interpretation of the above operations and vector quantities we have
completely followed geometric ideas motivated by our simple observations in our physical space
to interpret these concepts of addition and products.
2.2.3

Mixed and Triple Products

Based on the previous considerations it can be shown that the following triple products also hold.
Scalar triple product:
~a (~b ~c) = ~b (~c ~a) = ~c (~a ~b)
(9)
Vector triple product:

~a (~b ~c) = ~b(~a ~c) ~c(~a ~b)

(10)

(~a ~b) ~c = ~c (~a ~b) = (~c ~b)~a + (~c ~a)~b

(11)

and

2.2.4

Dyadic Product

Let us now consider a third important concept of product for physical vectors, the dyadic product.
This is not very intuitive geometrically compared to the previous two product concepts; it more
relates to the case when a physical vector is defined as a linear function. The dyadic product is
also often called general product.
Let us consider again two vectors, ~a and ~b and also a third one ~c. We consider the case when
~c is transformed into a vector ~d the direction of which is defined by ~a and the magnitude is given
by the scalar product of ~c with ~b scaled with the magnitude of ~a. This result can be given as
~d = (~b ~c)~a. This is a so-called linear transformation, that is essentially represented by the dyadic
product, ~a ~b, and illustrated in Fig. 5
~a

~c

~b

~eb

~d = (~c ~b) ~a

~c ~b

(1)

(2)

Figure 5: Illustration of the dyadic product

The dyadic product of vectors ~a and ~b can be defined as
b
~a ~b = D

(12)

b is a geometric object on its own representing the dyadic, a second order tensor, that can
where D
act on/transform any vector ~c as




b ~c = ~b ~c ~a = ~d
~a ~b ~c = D
(13)
5

Advanced Dynamics

J. Kovecses

which operation is described above and represented in Fig. 5. We have already seen that the vector
product operations essentially come down to operations among the unit vectors as the magnitudes
act only as simple multipliers. Let us now consider a unit dyadic ~e1 ~e2 , the dyadic product of
two unit vectors (generally not perpendicular to each other) in the above formula as it operates
on vector ~c
(~e1 ~e2 ) ~c = (~e2 ~c) ~e1
(14)
We can see that the geometric meaning can be interpreted as taking the scalar component of vector
~c along ~e2 via the scalar product, and determining a new vector that has this scalar component
as magnitude but its direction is along ~e1 .
Transpose of the Dyadic: The transpose of the dyadic defined in Eq. (12) can be interpreted
as

T
bT
~a ~b = ~b ~a = D
(15)

and we can also see based on the interpretation of the operation of the dyadic on a vector, Eq. (13),
that



T




b = ~b ~c ~a = ~d
~a ~b ~c = ~c ~a ~b = ~c ~b ~a = ~c D
(16)

Representation of Physical Vectors and Their Transformations

As physical vectors and their addition are fundamentally geometric concepts the analysis of the
geometry and related pictures, as shown in Fig. 6, can tell us that the geometric decomposition of
an arbitrary vector ~a with respect to a set of three base vectors ~e1 , ~e2 , ~e3 defining three directions
in space3

~a
a3~e3
a2~e2
~e3 ~e2

a1~e1

~e1

Figure 6: Representation/decomposition of a physical vector

~a = a1~e1 + a2~e2 + a3~e3
3

These vectors are unit vectors most of the time. But, they do not necessarily have to be unit vectors.

(17)

Advanced Dynamics

J. Kovecses

should hold as long as the volume of the parallelepiped defined by ~e1 , ~e2 , ~e3 is not zero. In this
decomposition ~a1 = a1~e1 , ~a2 = a2~e2 , ~a3 = a3~e3 are the vector components and a1 , a2 , a3 are
the so-called scalar components that are also often called coordinates. Given ~a and ~e1 , ~e2 , ~e3
the decomposition of Eq. (17) can always be determined by geometric methods such as direct
geometric construction. To investigate the decomposition in another way, more quantitatively, the
scalar components need to be determined. Let us consider the scalar products of Eq. (17) by ~e1 ,
~e2 and ~e3 as
~a ~e1 = a1~e1 ~e1 + a2~e2 ~e1 + a3~e3 ~e1
(18)
~a ~e2 = a1~e1 ~e2 + a2~e2 ~e2 + a3~e3 ~e2

(19)

~a ~e3 = a1~e1 ~e3 + a2~e2 ~e3 + a3~e3 ~e3

(20)

then Eqs. (18)-(20) form a system

~e1 ~e1
~e1 ~e2
~e1 ~e3
|

of linear equations

~e2 ~e1 ~e3 ~e1

~e2 ~e2 ~e3 ~e2
~e2 ~e3 ~e3 ~e3
{z
}
H

for a1 , a2 and a3 and can be written as

~a ~e1
a1
a2 = ~a ~e2
(21)
~a ~e3
a3

where H is the 3 3 coefficient matrix, the inner product matrix; this system of equations can be
solved for the scalar components as long as H is an invertible, i.e., the rows of H are independent
of each other. This means, based on Eqs. (17)-(21), that ~e1 , ~e2 and ~e3 are linearly independent of
each other, i.e., if ~a = ~0 is the zero vector then the composition of that using ~e1 , ~e2 and ~e3 is only
possible if all scalar components in Eq. (17) are equal to zero. The geometric interpretation of
linear independence, as mentioned above, is that the parallelepiped defined by ~e1 , ~e2 and ~e3 has
nonzero volume. Based on the above reasoning we also note that orthogonality among the base
vectors is not required for the above representation and decomposition. We can see from Eq. (21)
and using the concept of the scalar product that H is a symmetric matrix.
The above considerations also imply that the decomposition given by Eq. (17) is always
possible for any triads of linearly independent vectors; they do not have to be unit vectors. Such
linearly independent set of three vectors can form a basis for the representation of quantities in
space, and they are called base vectors. Often, the more typical case when the base vectors are
also unit vectors.
Orthogonal Unit Vectors - Orthonormal Basis: A special, but particularly important case
when the base vectors are unit vectors and they are also mutually perpendicular to each other,
which represents an orthonormal basis. For that case matrix H in Eq. (21) becomes the 3 3
identity matrix, and the scalar components a1 , a2 and a3 can simply be determined as
a1 = ~a ~e1 ,

3.1

a2 = ~a ~e2 ,

a3 = ~a ~e3

(22)

Representation of Vector Operations

In the following we will represent the various vector operations when the vectors are decomposed/represented with respect to the same base unit vectors.

Advanced Dynamics

3.1.1

J. Kovecses

Vector Addition

We consider ~c = ~a + ~b as was discussed in section 2 and illustrated in Fig. 1. Vectors ~a and ~b

are given with respect to base unit vectors ~e1 , ~e2 , ~e3 as
~a = a1~e1 + a2~e2 + a3~e3

(23)

~b = b1~e1 + b2~e2 + b3~e3

(24)

and
then the result of the vector addition can be written as
~c = ~a + ~b = (a1 + b1 ) ~e1 + (a2 + b2 ) ~e2 + (a3 + b3 ) ~e3
| {z }
| {z }
| {z }
c1

c2

(25)

c3

where c1 , c2 , c3 are the scalar components of the resultant vector ~c.

3.1.2

Scalar Product

Let us consider again these two arbitrary vectors, ~a and ~b. Using Eqs. (23) and (24) and the
concept of scalar product introduced in the previous section
~a ~b = a1 b1~e1 ~e1 + a1 b2~e1 ~e2 + a1 b3~e1 ~e3 + a2 b1~e2 ~e1 + a2 b2~e2 ~e2
+a2 b3~e2 ~e3 + a3 b1~e3 ~e1 + a3 b2~e3 ~e2 + a3 b3~e3 ~e3

(26)

which can be arranged as

~a ~b =

a1

~e1 ~e1 ~e1 ~e2 ~e1 ~e3

b1
a2 a3 ~e2 ~e1 ~e2 ~e2 ~e2 ~e3 b2 = aT Hb
{z
} ~e ~e ~e ~e ~e ~e
b3
3
1
3
2
3
3
aT
|
{z
} | {z }
H
b


(27)

where matrix H is the so-called inner product matrix associated with base vectors ~e1 , ~e2 , ~e3 , this
matrix already appeared earlier in Eq. (21), 31 arrays, a and b contain the scalar components of
~a and ~b with respect to the selected base vectors; we will term such arrays as algebraic vectors. We
will often work with algebraic vectors after we transform the geometric/physical representation of
a system to an algebraic form. However, it is important to understand that an algebraic vector
do not completely represent a physical vector; it only contains its scalar components with respect
to a set of base vectors, but from the algebraic vector alone it is not possible to know what the
base vectors are. Therefore, it is important to indicate the difference between physical/geometric
quantities and algebraic representations in the notations used. Writing something such as

a1
~a = a2
(28)
a3
is incorrect; this is often forgotten and concepts are discussed this way incorrectly, significantly
limiting the potential generality and applicability of formulations.
8

Advanced Dynamics

J. Kovecses

Orthogonal Base Vectors: If ~e1 , ~e2 , ~e3 represent an orthogonal system of unit vectors then H
in Eq. (27) becomes the identity matrix and for such a case the representation of scalar product
with scalar components reduces to
~a ~b = aT b
(29)

We can also see here comparing Eq. (27) and Eq. (29) that just by considering algebraic vectors
a and b alone, only limited information is available about the complete physical vectors, and, for
example, we would not know how to perform the vector operations and what inner product matrix
to use.
3.1.3

Cross Product

We consider the two arbitrary vectors, ~a and ~b. Using Eqs. (23) and (24) and the concept of cross
product discussed in section 2.2.2, we can write
~a ~b = ~c = a1 b1~e1 ~e1 + a1 b2~e1 ~e2 + a1 b3~e1 ~e3 + a2 b1~e2 ~e1 + a2 b2~e2 ~e2
+a2 b3~e2 ~e3 + a3 b1~e3 ~e1 + a3 b2~e3 ~e2 + a3 b3~e3 ~e3
(30)
Considering that ~e1 ~e1 = ~e2 ~e2 = ~e3 ~e3 = ~0, this can be further reduced as
~a ~b = ~c = a1 b2~e1 ~e2 + a1 b3~e1 ~e3 + a2 b1~e2 ~e1 + a2 b3~e2 ~e3 + a3 b1~e3 ~e1 + a3 b2~e3 ~e2 (31)

For the general case when the base vectors ~e1 , ~e2 , ~e3 are linearly independent but not orthogonal
to each other the cross products among the base vectors have to be evaluated based on the
angles formed by the base vectors and the general definition of the cross product as described in
section 2.2.2.
Orthogonal Base Vectors: However, the most important case for the cross product when the
base vectors are orthogonal to each other, then based on Eq. (8) the above result can be further
simplified as
~a ~b = ~c = a1 b2~e3 a1 b3~e2 a2 b1~e3 + a2 b3~e1 + a3 b1~e2 a3 b2~e1 =
= (a2 b3 a3 b2 ) ~e1 + (a3 b1 a1 b3 ) ~e2 + (a1 b2 a2 b1 ) ~e3
|
{z
}
|
{z
}
|
{z
}
c1

c2

(32)

c3

where c1 , c2 and c3 are the scalar components of the resultant vector ~c with respect to the base
vectors, and they can also be collected in an algebraic vector c = [c1 c2 c3 ]T .
If we express the above operation only with the scalar components collected in algebraic vectors
a, b and c then based on Eq. (32) we can write

c = Ab
(33)
where

0 a3 a2
= a3
0 a1
A
(34)
a2 a1
0
is a skew-symmetric matrix, sometimes called the cross-product matrix, formed from the scalar
components of ~a. Again, we can see that if we only work with algebraic vectors, i.e., scalar
components, then we also have to know what base vectors were used to determine the scalar
components.
9

Advanced Dynamics

3.1.4

J. Kovecses

Dyadic Product and Second-Order Tensors

We use again three arbitrary physical vectors, ~a, ~b and ~c to illustrate this operation. Using Eqs.
(23) and (24) and the concept of the dyadic product discussed in section 2.2.4, we obtain
b = a1 b1~e1 ~e1 + a1 b2~e1 ~e2 + a1 b3~e1 ~e3 + a2 b1~e2 ~e1 + a2 b2~e2 ~e2
~a ~b = D
+a2 b3~e2 ~e3 + a3 b1~e3 ~e1 + a3 b2~e3 ~e2 + a3 b3~e3 ~e3

(35)

This can also be written as

b =
~a ~b = D

3 X
3
X
i=1 j=1

ai bj~ei ~ej =

3 X
3
X
i=1 j=1

Dij~ei ~ej

(36)

where Dij = ai bj , and the transpose of the dyadic can be interpreted as

3 X
3

T
X
T
~
~
b
~a b = b ~a = D =
Dji~ei ~ej

(37)

i=1 j=1

Considering now the scalar product of the dyadic ~a ~b with vector ~c = c1~e1 + c2~e2 + c3~e3 as
in Eq. (13), we can expand the expression as
b ~c = ~d =
(~a ~b) ~c = D

= a1 b1 (~e1 ~e1 ) ~e1 c1 + a1 b2 (~e1 ~e2 ) ~e1 c1 + a1 b3 (~e1 ~e3 ) ~e1 c1 + a2 b1 (~e2 ~e1 ) ~e1 c1 +
+a2 b2 (~e2 ~e2 )~e1 c1 +a2 b3 (~e2 ~e3 )~e1 c1 +a3 b1 (~e3 ~e1 )~e1 c1 +a3 b2 (~e3 ~e2 )~e1 c1 +a3 b3 (~e3 ~e3 )~e1 c1 +
+a1 b1 (~e1 ~e1 ) ~e2 c2 + a1 b2 (~e1 ~e2 ) ~e2 c2 + a1 b3 (~e1 ~e3 ) ~e2 c2 + a2 b1 (~e2 ~e1 ) ~e2 c2 +
+a2 b2 (~e2 ~e2 )~e2 c2 +a2 b3 (~e2 ~e3 )~e2 c2 +a3 b1 (~e3 ~e1 )~e2 c2 +a3 b2 (~e3 ~e2 )~e2 c2 +a3 b3 (~e3 ~e3 )~e2 c2 +
+a1 b1 (~e1 ~e1 ) ~e3 c3 + a1 b2 (~e1 ~e2 ) ~e3 c3 + a1 b3 (~e1 ~e3 ) ~e3 c3 + a2 b1 (~e2 ~e1 ) ~e3 c3 +
+a2 b2 (~e2 ~e2 )~e3 c3 +a2 b3 (~e2 ~e3 )~e3 c3 +a3 b1 (~e3 ~e1 )~e3 c3 +a3 b2 (~e3 ~e2 )~e3 c3 +a3 b3 (~e3 ~e3 )~e3 c3
(38)
Based on the definition of the dyadic product as described in section 2.2.4 and forming the scalar
products with the unit dyadics in the above expression we obtain
b ~c = ~d =
(~a ~b) ~c = D

= a1 b1 (~e1 ~e1 )~e1 c1 + a1 b2 (~e2 ~e1 )~e1 c1 + a1 b3 (~e3 ~e1 )~e1 c1 + a2 b1 (~e1 ~e1 )~e2 c1 +
+a2 b2 (~e2 ~e1 )~e2 c1 + a2 b3 (~e3 ~e1 )~e2 c1 + a3 b1 (~e1 ~e1 )~e3 c1 + a3 b2 (~e2 ~e1 )~e3 c1 + a3 b3 (~e3 ~e1 )~e3 c1 +
+a1 b1 (~e1 ~e2 )~e1 c2 + a1 b2 (~e2 ~e2 )~e1 c2 + a1 b3 (~e3 ~e2 )~e1 c2 + a2 b1 (~e1 ~e2 )~e2 c2 +
+a2 b2 (~e2 ~e2 )~e2 c2 + a2 b3 (~e3 ~e2 )~e2 c2 + a3 b1 (~e1 ~e2 )~e3 c2 + a3 b2 (~e2 ~e2 )~e3 c2 + a3 b3 (~e3 ~e2 )~e3 c2 +
+a1 b1 (~e1 ~e3 )~e1 c3 + a1 b2 (~e2 ~e3 )~e1 c3 + a1 b3 (~e3 ~e3 )~e1 c3 + a2 b1 (~e1 ~e3 )~e2 c3 +
+a2 b2 (~e2 ~e3 )~e2 c3 + a2 b3 (~e3 ~e3 )~e2 c3 + a3 b1 (~e1 ~e3 )~e3 c3 + a3 b2 (~e2 ~e3 )~e3 c3 + a3 b3 (~e3 ~e3 )~e3 c3 (39)
10

Advanced Dynamics

J. Kovecses

which can then be grouped to result in the final, detailed expression

b ~c = ~d = d1~e1 + d2~e2 + d3~e3 =
(~a ~b) ~c = D

h
a1 b1 (~e1 ~e1 )c1 + a1 b2 (~e2 ~e1 )c1 + a1 b3 (~e3 ~e1 )c1 + a1 b1 (~e1 ~e2 )c2 + a1 b2 (~e2 ~e2 )c2 + a1 b3 (~e3 ~e2 )c2 +
i
+a1 b1 (~e1 ~e3 )c3 + a1 b2 (~e2 ~e3 )c3 + a1 b3 (~e3 ~e3 )c3 ~e1 +
h
a2 b1 (~e1 ~e1 )c1 + a2 b2 (~e2 ~e1 )c1 + a2 b3 (~e3 ~e1 )c1 + a2 b1 (~e1 ~e2 )c2 + a2 b2 (~e2 ~e2 )c2 + a2 b3 (~e3 ~e2 )c2 +
i
+a2 b1 (~e1 ~e3 )c3 + a2 b2 (~e2 ~e3 )c3 + a2 b3 (~e3 ~e3 )c3 ~e2 +
h
a3 b1 (~e1 ~e1 )c1 + a3 b2 (~e2 ~e1 )c1 + a3 b3 (~e3 ~e1 )c1 + a3 b1 (~e1 ~e2 )c2 + a3 b2 (~e2 ~e2 )c2 + a3 b3 (~e3 ~e2 )c2 +
i
+a3 b1 (~e1 ~e3 )c3 + a3 b2 (~e2 ~e3 )c3 + a3 b3 (~e3 ~e3 )c3 ~e3
(40)

where the detailed expressions of the scalar components, d1 , d2 , d3 , are shown in the square
brackets. Based on this detailed expansion it can be shown that the scalar components of ~d, d1 ,
d2 , d3 , arranged in an algebraic vector d = [d1 d2 d3 ]T can be expressed as

where

DHc = d

(41)

a1 b 1 a1 b 2 a1 b 3
D = abT = a2 b1 a2 b2 a2 b3
a3 b 1 a3 b 2 a3 b 3

(42)

and H is the inner product matrix for base vectors ~e1 , ~e2 , ~e3 as was already introduced earlier for
example in Eq. (27), and c = [c1 c2 c3 ]T contains the scalar components of vector ~c.
Symmetric and Skew-Symmetric Tensors: Considering Eqs. (36) and (37) a symmetric
tensor (dyadic) can be interpreted as
b =D
bT =
D

3 X
3
X
i=1 j=1

Dij~ei ~ej =

3 X
3
X
i=1 j=1

Dji~ei ~ej

(43)

where we can see that the Dij scalar components can form a symmetric 3 3 matrix. A skewsymmetric tensor is defined as
b = D
bT =
D

3 X
3
X
i=1 j=1

Dij~ei ~ej =

3 X
3
X
i=1 j=1

Dji~ei ~ej

(44)

Based on this we can see for this case the Dii components must vanish Dii = 0 (i = 1, 2, 3), only
three nonzero scalar components exist in the matrix formed from the scalar components, and this
matrix will look like the cross-product matrix introduced in Eq. (34). Therefore, the operation of
a skew-symmetric tensor on a vector can be represented with a simple cross product operation as
b s ~c = w
~ ~c
D
11

(45)

Advanced Dynamics

J. Kovecses

~ = w1~e1 + w2~e2 + w3~e3 is the so-called vector invariant (or axial vector) of skew-symmetric
where w
b s where subscript s emphasizes that the tensor is skew-symmetric. The nonzero scalar
tensor D
~ , w1 , w2 , w3 , and their
components of the tensor are formed from the scalar components of w
negatives.
Orthogonal Base Vectors: Again, the special, but very important case is when the base vectors
are unit vectors and they are orthogonal to each other. In that case the inner product matrix H
b ~c = ~d using only scalar
becomes the identity matrix and the dyadic product of (~a ~b) ~c = D
components can be written based on Eq. (41) as
(abT )c = Dc = d

(46)

We can see that in such a case the dyadic product ~a ~b can be fully represented with a 3 3
matrix D with respect to the base vectors. This matrix contains essentially the scalar components
of the dyadic product that is an entity on its own and the scalar components represent all possible
product combinations of the scalar components of the two vectors that were used to compose the
dyadic product. We note again that matrix D itself does not carry information about the base
vectors; so formulas such as Eq. (46) can only be used if we have an understanding of what the
base vectors are and how the different geometric quantities are decomposed.
Note:
We note that all of the above considerations rely on geometric concepts at the first place, and the
assumption that a decomposition for a physical vector, e.g. ~a, is available in the form of Eq. (17).
Therefore, if ~a is only given geometrically without any prior decomposition such as Eq. (17), then
the only way to determine the scalar products ~a ~e1 , ~a ~e2 , ~a ~e3 is via geometric construction,4
as illustrated in Fig. 2 for example. Most of the time, however, a geometric construction is not
necessary to perform as the vector in question is already given in a decomposed form of Eq. (17)
with respect to a set of base vectors, a1 , a2 , a3 are known, and the problem is to transform the
representation into another set of base vectors.

3.2

Change of Basis and Transformation of Scalar Components

Let us then consider now that the decomposition Eq. (17) is already known and we would like to
change the base vectors of the representation to ~e10 , ~e20 , ~e30 , which is another set of independent
unit vectors and the decomposition of ~a can be expressed in that basis as
~a = a01~e10 + a02~e20 + a03~e30

(47)

and illustrated in Fig. 7. Therefore, based on this and Eq. (17) we can write
~a = a1~e1 + a2~e2 + a3~e3 = a01~e10 + a02~e20 + a03~e30

(48)

Let us form the scalar products of Eq. (48) with the three new base vectors, ~e10 , ~e20 , ~e30 , as
a1~e1 ~e10 + a2~e2 ~e10 + a3~e3 ~e10 = a1~e10 ~e10 + a2~e20 ~e10 + a3~e30 ~e10
4

(49)

This is an important concept, which again shows that all of these considerations stem from simple geometric
notions.

12

Advanced Dynamics

J. Kovecses

a30~e30
~a

~a

a3~e3

a20~e20

a2~e2

~e30

~e3 ~e2

~e20
~e10

a10~e10

a1~e1

~e1

Figure 7: Illustration of change of basis

a1~e1 ~e20 + a2~e2 ~e20 + a3~e3 ~e20 = a1~e10 ~e20 + a2~e20 ~e20 + a3~e30 ~e20
a1~e1 ~e30

a2~e2 ~e30

These equations can be expressed for

~e1 ~e10 ~e2 ~e10 ~e3 ~e10

~e1 ~e20 ~e2 ~e20 ~e3 ~e20
~e1 ~e30 ~e2 ~e30 ~e3 ~e30
{z
|
C

a3~e3 ~e30

a1~e10

~e30

a2~e20

~e30

a3~e30

~e30

the scalar components as

0 0
0

~e1 ~e1 ~e20 ~e10 ~e30 ~e10

a1
a1
a2 = ~e10 ~e20 ~e20 ~e20 ~e30 ~e20 a02
~e10 ~e30 ~e20 ~e30 ~e30 ~e30
a3
a03
} | {z } |
{z
} | {z }
a
a0
H0

(50)
(51)

(52)

where H0 is the inner product matrix associated with the new set of base vectors, and C is the
so-called direction cosine matrix formed from the scalar products among the two sets of base
vectors. The inner product matrix is always invertible as long as the set of base vectors are
linearly independent; it is in fact a positive definite matrix. Therefore, the scalar components
a0 = [a01 a02 a03 ]T for the new basis can be expressed as
1

(H0 )

Ca = a0

(53)

Considering that the base vectors are unit vectors, the elements of the direction cosine matrix can
be written as
cij = ~ei ~ej0 = ~ej0 ~ei = cos ij
i, j = 1, 2, 3
(54)
where ij is the angle between base vectors ~ei and ~ej0 and

~e1 ~e10 ~e2 ~e10 ~e3 ~e10

c11 c21 c31
C = ~e1 ~e20 ~e2 ~e20 ~e3 ~e20 = c12 c22 c32
~e1 ~e30 ~e2 ~e30 ~e3 ~e30
c13 c23 c33

(55)

Let us now consider again Eq. (48) but now analyze the transformation from base vectors ~e10 , ~e20 ,
~e30 to set ~e1 , ~e2 , ~e3 . Consider the scalar products of Eq. (48) with ~e1 , ~e2 , ~e, respectively, as
a1~e1 ~e1 + a2~e2 ~e1 + a3~e3 ~e1 = a1~e10 ~e1 + a2~e20 ~e1 + a3~e30 ~e1
13

(56)

Advanced Dynamics

J. Kovecses

a1~e1 ~e2 + a2~e2 ~e2 + a3~e3 ~e = a1~e10 ~e2 + a2~e20 ~e2 + a3~e30 ~e2

(57)

a1~e1 ~e3 + a2~e2 ~e3 + a3~e3 ~e3 = a1~e10 ~e3 + a2~e20 ~e3 + a3~e30 ~e3
These equations can again be

~e1 ~e1 ~e2 ~e1

~e1 ~e2 ~e2 ~e2
~e1 ~e3 ~e2 ~e3
{z
|
H

expressed for the scalar components

0

~e1 ~e1 ~e20 ~e1

~e3 ~e1
a1
~e3 ~e2 a2 = ~e10 ~e2 ~e20 ~e2
~e10 ~e3 ~e20 ~e3
~e3 ~e3
a3
} | {z } |
{z
a
CT

(58)

as

~e30 ~e1
~e30 ~e2
~e30 ~e3
}|

a01
a02
a03
{z }
a0

(59)

where H is the inner product matrix for base vectors ~e1 , ~e2 , ~e3 , and we can see using Eqs. (52),
(54) and (55) that the direction cosine matrix for this representation is simply the transpose of
what was obtained previously in Eq. (52). The scalar components a = [a1 a2 a3 ]T can then be
expressed as
H1 CT a0 = a
(60)
3.2.1

Orthogonal Base Vectors and Canonical Cases

Let us consider the special, but very important case when each set of base vectors consists of
three unit vectors that are orthogonal to each other. In that case both inner product matrices, H
and H0 , become the identity matrix, and Eqs. (53) and (60) for the transformation of the scalar
components become
Ca = a0
(61)
and
CT a0 = a

(62)

which clearly show that for this case C1 = CT , i.e., the direction cosine matrix becomes an
orthogonal matrix, this alone will represent the transformation between the two sets of scalar
components. We can also see based on Eq. (55) that in this case the columns of C contain the
scalar components of base vectors ~e1 , ~e2 , ~e3 in terms of ~e10 , ~e20 , ~e30 , and the columns of CT contain
the scalar components of ~e10 , ~e20 , ~e30 with respect to ~e1 , ~e2 , ~e3 .
We can consider three canonical cases for orthogonal base vectors, when two base vectors, one
in each set, are the same. The three cases are: (1) ~e1 = ~e10 , (2) ~e2 = ~e20 , (3) ~e3 = ~e30 .
Case 1: This is illustrated in Fig. 8 In that case any change only happens in the plane spanned
by ~e2 and ~e3 , and also ~e20 and ~e30 . Based on the figure we can see that 22 = 33 = and
32 = 90 , and 23 = 90 + . Using these, matrix C, detailed expression is in Eq. (52), can
be expanded as

1
0
0
1
0
0
1
0
0
cos
cos(90 ) = 0 cos sin (63)
C = 0 ~e2 ~e20 ~e3 ~e20 = 0
0
0
0 ~e2 ~e3 ~e3 ~e3
0 cos(90 + )
cos
0 sin cos
where trigonometric identities cos(90 ) = sin and cos(90 + ) = sin were also used.

14

Advanced Dynamics

J. Kovecses
~e30
23

33
~e3
32
22
~e20

~e2

Figure 8: Transformation for canonical case 1

~e10
31

11
~e1
13
33
~e30

~e3

Figure 9: Transformation for canonical case 2

Case 2: This case is illustrated in Fig. 9 Here the plane spanned by ~e1 and ~e3 , and also ~e10
and ~e30 is of primary interest. Based on the figure we can determine that 11 = 33 = and
13 = 90 , and 31 = 90 + . Matrix C can then be expanded as

~e1 ~e10 0 ~e3 ~e10

cos
0 cos(90 + )
cos 0 sin
= 0
(64)
1
0 =
0
1
0
1
0
C= 0
0
0
~e1 ~e3 0 ~e3 ~e3
cos(90 ) 0
cos
sin 0 cos
where again the above trigonometric identities were used.

Case 3: This is shown in Fig. 10 illustrating the plane spanned by ~e1 and ~e2 , and also ~e10 and ~e20 .
Based on the figure we can determine that 11 = 22 = and 12 = 90 + , and 21 = 90 .
Matrix C can then be expanded as

~e1 ~e10 ~e2 ~e10 0

cos
cos(90 ) 0
cos sin 0
cos
0 = sin cos 0 (65)
C = ~e1 ~e20 ~e2 ~e20 0 = cos(90 + )
0
0
1
0
0
1
0
0
1
which gives the form of the direction cosine matrix for the third important canonical case.

15

Advanced Dynamics

J. Kovecses
~e2
~e20
22
21

~e10
11

12
~e1

Figure 10: Transformation for canonical case 3

Vector Projections and Decompositions

Consider the standard three-dimensional representation shown in Fig. 11.

a

Figure 11: Definition of two complementary subspaces

We have a vector ~v and a direction in space defined by another vector ~a, which is used to
define a subspace, in this case, a single line. We can also define another subspace orthogonal to
this direction, as shown in the figure. The unit vector along the direction defined by ~a is
~ea =

~a
|~a|

(66)

where |~a| = a is the magnitude of the vector. Using ~ea , we can decompose ~v as the sum of two
vector components, one of them along the direction defined by ~a and the other one perpendicular
to it as
~v = ~va + ~va = va~ea + ~va
(67)
Multiplying Eq. (67) with ~ea
~v ~ea = (va~ea ) ~ea = va

(68)

because ~va ~ea = 0. We can express the two vector components as

~va = (~va ~ea ) ~ea ;

~va = ~v ~va

(69)

and considering the dyadic product, ~va can also be expressed as

~va = (~va ~ea ) ~ea = (~ea ~ea ) ~v
| {z }
b
P
16

(70)

Advanced Dynamics

J. Kovecses

b can be called projector tensor, and ~va as

where dyadic product P


~va = ~v (~va ~ea ) ~ea = ~v (~ea ~ea ) ~v = bI ~ea ~ea ~v
{z
}
|
bI P
b

(71)

b is the projector tensor to

where bI represents the identity tensor (dyadic), i.e., bI ~v = ~v, and bI P
the plane perpendicular to the direction of ~a, Fig. 11.

Change of Orientation and Finite Rotations

For the study of motion we have introduced two basic geometric objects: the point, and the vector.
The displacements of points can be represented with vectors and can be tracked by the addition
of such vectors as was discussed previously. On the other hand, everyday observations involve the
perception of another type of displacement: the change of orientation, which cannot be represented
with the transformation of a point; it requires the object that also includes information about
orientation, a vector. For example, here we can think about physical vectors that are interpreted
at the position level, e.g., to describe the relative position of two points.
To represent the change in orientation of a vector we can consider, without loss of generality
that one end point, the starting point, of the vector remains fixed, and the magnitude of the vector
does not change either. The change in orientation can then be considered as a rotation about that
point.
~n

~v

~p

Figure 12: General change of orientation

For any change in orientation the end point of the vector will move on the surface of a sphere
that has the radius equal to the magnitude of the vector and the centre of the sphere is at the
17

Advanced Dynamics

J. Kovecses

starting, fixed point of the vector, as illustrated in Fig. 12. We can consider that we are dealing
with two different vectors that represent the initial configuration, ~v and final configuration ~p,
respectively. To characterize the change in orientation we are interested in how these two vectors
are related to each other, i.e., how the first vector can be transformed into the second one. This
is not related to any frame of reference, passage of time, nor the actual path the end point of the
vector ~v would trace to arrive to the second configuration. We simply look at the static picture
of the two vectors with different orientations and attempt to decide what transformation ~v would
need to undergo to become ~p.
The end points of the two vectors are on the surface of the sphere, Fig. 12. Based on the
spherical geometry, we can see that it is always possible to define a circle on the sphere, which
will contain the two end points and it is the base of a right circular cone having its vertex at the
centre of the sphere. The vectors define lines that are two generators of the cone. This shows that
the first vector ~v can always be transformed into the new vector ~p via a single rotation about the
axis of the cone with a certain angle.
This represents the general case of displacement difference between two vectors due to pure
change in orientation. Hence, the most general change in orientation of a vector can be represented
as a rotation about an axis with an angle.5 The representation of such a displacement is called
a pure finite rotation that transforms one vector into another. The axis of the cone interpreted
is the axis of rotation can be indicated by a unit vector ~n pointing from the centre of the sphere
towards the centre of the circle, and the angle of rotation can be denoted by (Fig. 12).

Let us now consider this transformation in more detail. Consider that vector ~v = OA is

transformed into vector ~p = OB via a pure finite rotation about axis indicated by unit vector ~n
(Fig. 13). The angle of the rotation is . Based on the figure we can write that
B
~n
~p
~v||

~v||
B0

~n ~v

~v
O
~vcos

D0

~n~vsin
~v||
~v

A0

Figure 13: Representation of finite rotation

5

This does not completely define the change of orientation of an arbitrary extended object as a rotation can also
be possible with the vector itself being the axis. However, it generally proves the notion of how a general change
in orientation, a finite rotation, can be represented with a rotation about an axis.

18

Advanced Dynamics

J. Kovecses

~p = ~vk + OB 0 = ~vk + OD0 + D0 B 0

(72)

~vk = ~n(~n ~v) = (~n ~n) ~v

(73)

and using the properties of vector operations and dyadic products discussed in the previous sections
we can write


~v = ~v ~vk = bI ~n ~n ~v
(74)

where bI is the identity tensor as was introduced for Eq. (71). Considering the OD0 B 0 right-angled
triangle we can write that


0
OD = ~v cos = cos bI ~n ~n ~v
(75)

Based on the figure we can also see that rotating the ~v = OA0 by /2 about the ~n axis results

in the ON 0 vector. Using this we can establish that

0
ON = ~n ~v = ~n (~vk + ~v ) = ~n ~v
(76)

We can also notice using the figure that ON 0 = OB 0 and D0 B 0 = OB 0 sin = ON 0 sin . Lines
ON 0 and D0 B 0 are parallel. Considering this and Eq. (76) we can obtain
0 0
b ~v
D B = sin ~n ~v = sin N

(77)

b is a skew-symmetric cross product tensor associated with vector ~n. (Vector ~n is the
where N
b Based on the above equations and Fig. 13, the transformation
so-called vector invariant of N.)
relationship between ~p and ~v can be established as
h
i
h
i
b
b
b
b
~p = ~n ~n + cos (I ~n ~n) + sin N ~v = cos I + (1 cos )~n ~n + sin N ~v (78)
which can also be written as
where

b ~v
~p = R

b = cos bI + (1 cos )~n ~n + sin N

b
R

(79)
(80)

is the rotation tensor transforming vector ~v into vector ~p. It is interpreted for the general case of
a finite rotation.
A Property of the Rotation Tensor: Let us consider two arbitrary vectors ~r1 and ~r2 , that
b as R
b ~r1 and R
b ~r2 . The rotation
are both transformed by the same finite rotation tensor R
keeps the magnitudes and the relative orientation between the two vectors constant. Therefore,
the scalar products before and after the rotation have to be the same, and we can write

 

b
b
bT R
b ~r2
~r1 ~r2 = R ~r1 R ~r2 = ~r1 R
(81)
from this with the introduction of the identity tensor we can obtain
 T

bT R
b ~r2 ~r1 ~r2 = ~r1 R
b R
b bI ~r2 = 0
~r1 R
19

(82)

Advanced Dynamics

J. Kovecses

As ~r1 and ~r2 are two arbitrary vectors Eq. (82) can only hold if
bT R
b = bI
R

(83)

bT = R
b 1 , i.e., the inverse of the rotation tensor is its transpose. Due to this
which means that R
property the tensor is called an orthogonal tensor.
Linearization: Expanding the expression of the rotation tensor into Taylor series in terms of
for 0 = 0 we obtain
b = bI + N
b + 1 (~n ~n bI) 2 + HOT
(84)
R
2
where HOT stands for terms whose order in is cubic or higher. If the angle of rotation is small,
so that the HOT and the 2 terms are negligible compared to the first two terms, then the rotation
tensor can be approximated as
b = bI + N
b
R
(85)
where the represents that this is valid only for small rotations.

5.1

Composition of Two Finite Rotations

As we saw the transformation of physical vectors via a finite rotation can be represented by ~n and
. This finite rotation transforms vector ~v into vector ~p as
b ~v
~p = R

(86)

b a ~p
~pa = R

(87)

b a (R
b ~v) = R
ba R
b ~v
~pa = R

(88)

Consider now that the resulting vector undergoes another finite rotation represented by ~na and
a . This transformation can be interpreted as

b a is the rotation tensor associated with ~na and a

where R
We can write that

which shows that finite rotations are multiplicative, not additive like the displacements of points.

5.2

Time Rate of Change

Consider again a physical vector of constant magnitude, ~r, that is subjected to a finite rotation
b 6 The change in the orientation of this vector via the finite rotation can be understood if we
R.
consider another vector, ~p that is fixed in the absolute reference frame and in the initial instant
it is coincident with ~r. Then the finite rotation for ~r can be interpreted as

b ~p
~r = R

For example, a vector describing the position difference between two points of a reference frame.

20

(89)

Advanced Dynamics

J. Kovecses

The time rate of change of ~r can then be expressed

b
dR
b ~p
~r =
~p = R
dt

b R
b T ~r
~r = R

(90)

where it was used that ~p is fixed so that ~p = ~0, and the rotation tensor is orthogonal, i.e.,
b T ~r. Therefore
~p = R
b R
b T ~r
~r = R

b r
~r =

b =R
b R
bT

b is the angular velocity tensor.

where
b R
b T = bI and based on that
Consider the orthogonality condition R
T
b = 0
b R
bT + R
b R
b
R

b =
bT

(91)

(92)

shows that the angular velocity tensor is skew-symmetric and as such, based on the discussion in
~ so that
section 3.1.4, it has a vector invariant, or axial vector,
b ~r =
~r =
~ ~r

(93)

~ is the angular velocity vector. This also illustrates that the angular velocity is not
holds where
the derivative of a vector. It is a so-called nonholonomic quantity.

5.3

Euler Parameters

It can be seen from Eq. (80) that the rotation tensor is essentially defined by the unit vector of the
axis of rotation ~n and the rotation angle . However, other parameterizations are also possible.
We can consider for example
~s = sin

~n
2

and

s0 = cos

(94)

and using Eq. (80) the rotation tensor with these definitions can take the form
b = (s2 ~s ~s)bI + 2~s ~s + 2s0 S
b
R
0

(95)

b is the skew-symmetric cross product tensor associated with vector ~s. The three scalar
where S
components of ~s with respect to a basis plus s0 give the so-called Euler, or Euler-Rodriques,
parameters.

Frames of Reference and Their Parameterization

For the material discussed above the concept of the frame of reference was not needed. The concept
of vectors and their representation do not directly require the interpretation of frames. However,
as already discussed earlier, to describe typical modes of observed motion and generally interpret
physical quantities frames of reference are necessary, and these frames need to be parameterized
so that vectors in them can be properly represented.
21

Advanced Dynamics

J. Kovecses

Let us now consider that the base unit vectors ~e1 , ~e2 , ~e3 are attached to a frame of reference
F, and can define a coordinate system/parameterization for that frame. The same way ~e10 , ~e20 , ~e30
can be considered to be attached to frame F 0 .
We can consider two fundamental geometric objects related to a frame: an arbitrary point of
the frame, and a vector that defines the position of a point of the frame relative to another. This
vector may be represented with respect to the base vectors, but that is not always necessary.

6.1

Rotation Matrix

The rotation matrix is the coordinate-level representation of the rotation tensor, i.e., it contains
the scalar components of the rotations tensor in a given coordinate system. For example, the Euler
parameters can then be defined via expressing ~s in the given coordinate system: s = [s1 s2 s3 ]T
and s0 gives the fourth parameter. With these the rotation matrix, R can be expressed in a
coordinate system defined by a set of orthonormal base vectors as
R = (s20 sT s)I + 2ssT + 2s0 S

(96)

where I is the 3 3 identity matrix and S is the skew-symmetric cross product matrix.

6.2

Reference Frames and Transformations of Scalar Components

We will investigate it now further that how the rotation tensor can be represented with scalar
components in a coordinate system attached to a frame. Let us consider that an orthogonal triad
of unit vectors ~e1 , ~e2 , ~e3 with a common origin, attached to frame F, Fig. . These vectors and frame
F are transformed to ~e10 , ~e20 , ~e30 and frame F 0 via a rotation about axis ~n. This transformation is
b and we can write
represented by tensor R
b ~e1
~e10 = R

b ~e2
~e20 = R

b ~e3
~e30 = R

(97)

We will consider the decomposition of all these quantities, their representation with scalar components, relative to an arbitrary orthogonal coordinate system attached to frame Fa . This makes
it possible to give the following matrix representation of Eq. (97)
E0 = RE

(98)

where
E0 = [e01 e02 e03 ]

E = [e1 e2 e3 ]

(99)

b and E0 and E are

and R is the 3 3 rotation matrix containing the scalar components of R,
3 3 matrices whose columns, e10 , e20 , e30 and e1 , e2 , e3 , contain the scalar components of the
~e10 , ~e20 , ~e30 and ~e1 , ~e2 , ~e3 vector triads, respectively, e0i and ei are 3 1 algebraic vectors of scalar
components, i = 1, 2, 3. As mentioned above, all the components are expressed in the coordinate
system associated with Fa .
Based on (98)
R = E0 E1
(100)
which also gives a general expression to represent the scalar components of the rotation tensor in
any coordinate system (in this case in Fa ).
22

Advanced Dynamics

J. Kovecses

If this frame Fa is selected more specifically to be the orthogonal frame defined by the vector
triad ~e1 , ~e2 , ~e3 , i.e., Fa F then E = I is the 33 identity matrix, and consequently, the rotation
matrix representation will be R = E0 . This means that the columns of the rotation matrix will
contain the scalar components of the unit vectors of the F 0 coordinate frame expressed in terms
of the unit vectors of frame F. These quantities are also often referred to as direction cosines as
we already discussed previously in section 3.2, and R = CT is an orthogonal matrix. This gives
one possibility to express the rotation matrix and establish transformation relationships between
two frames and the attached coordinate systems.
Let us now consider a vector ~r which is decomposed in an arbitrary orthogonal coordinate system
attached to Fa as
ra = r1 e1 + r2 e2 + r3 e3 = [e1 e2 e3 ] r = Er
(101)
and
ra = r10 e01 + r20 e02 + r30 e03 = [e01 e02 e03 ] r 0 = E0 r0

(102)

where r and r 0 are 3 1 arrays that contain the scalar components of ~r with respect to the
coordinate systems attached to F and F 0 , respectively, r = [r1 r2 r3 ]T and r 0 = [r10 r20 r30 ]T . If
now again we consider the special selection for the coordinate system the one attached to F, i.e.,
Fa F, then considering Eq. (99), we can also write for Eq. (102) that
ra = r = E0 r0 = Rr0

(103)

and we arrive to the result that in the most common case the scalar components of vectors are
transforming as
r = Rr 0
r 0 = RT r
(104)
which describes the transformation of the scalar components of a vector the same way as done
employing the direction cosine matrix if orthogonal coordinate systems are used. Therefore, the
two main applications of the rotation matrix can be summarized as:
According to the original interpretation of the rotation tensor, the rotation matrix can be used
for the component-level representation of the transformation of a physical vector into another
physical vector, where all scalar components are expressed in the same coordinate system.
It can also be used for the transformation of the scalar components of a vector between two
different orthogonal coordinate systems. In that case, for these transformations the use of the
rotation matrix becomes equivalent to that of the direction cosine matrix (and its transpose).

23