bappL_epdf 12/18/2018 21:19:18 Page 1

Appendix L

Derivation of Similarity Transformations

L.1 Introduction
In Section 5.7, in the text we saw that systems can be represented with different state
variables even though the transfer function relating the output to the input remains
the same. The various forms of the state equations were found by manipulating the transfer
function, drawing a signal-flow graph, and then writing the state equations from the
signal-flow diagram. These systems are called similar systems. Although their state-space
representations are different, similar systems have the same transfer function and hence
the same poles or eigenvalues.
The question now arises whether we can make transformations among similar systems
from one set of state equations to another without using the transfer function and signal-flow
graphs. In this Appendix, we will derive this transformation.

L.2 Expressing Any Vector in Terms

of Basis Vectors
Let us begin by reviewing the representation of vector quantities in space. In Chapter 3, we
learned that the state variables form the axes of the state space. Using a second-order system as
an example, Figure L.1 shows two sets of axes, x1x2 and z1z2.1
Thus a state vector, x, in state space can be written either in terms of the state variables
or axes, x1 and x2, or if we call it z, the state variables or axes, z1 and z2. In other words, the
same vector is expressed in terms of different state variables. From this discussion, we begin
to see that the transformation from one set of state equations to another may be simply the
transformation from one set of axes to another set of axes. Let us look further into this
possibility by first clarifying the ways in which vectors can be represented in space.
Unit vectors, Ux1 , and Ux2 , which are collinear with the axes x1 and x2, form linearly
independent vectors called basis vectors for the space, x1x2. Any vector in the space can be
written in two ways. First, it can be written as a linear combination of the basis vectors. This
linear combination implies vector summation of the basis vectors to form that vector.
Second, any vector can be written in terms of its components along the axes. Summarizing
these two ways of writing a vector, we have
x1
x x1 Ux1 x2 Ux2 L1
x2

1
These axes are shown to be orthogonal (90° to each other) for clarity. In general, the axes need be only linearly
independent and are not necessarily at 90°. Linear independence precludes collinear axes.

A-166
bappL_epdf 12/18/2018 21:19:18 Page 2

L.4 Finding the Transformation Matrix, P A-167

x2

z2
x or z
z1
Ux2

Uz2
Uz1

x1
Ux1

FIGURE L.1 State-space transformations

Similarly, the same vector, which will now be called z, can be written in terms of the basis
vectors in the z1z2 space,

z1
z z 1 U z1 z 2 U z2 L2
z2

L.3 Vector Transformations

What is the relationship between the components of x and z in Eqs. (L.1) and (L.2)? In other
words, how do we transform vector x into vector z and vice versa? To begin we realize that
unit vectors Uz1 , and Uz2 , which are collinear with z1 and z2 and are basis vectors for the
space, z1z2, can be also written in terms of the basis vectors of the x1x2 space. Hence,
U z1 p11 Ux1 p21 Ux2 L 3a

U z2 p12 Ux1 p22 Ux2 L 3b

Substituting Eqs. (L.3) into Eq. (L.2), and realizing that the vectors z and x are the same,
yields x in terms of the components of z, or

x z1 p11 z2 p12 Ux1 z1 p21 z2 p22 Ux2 L4

which is equivalent to

p11 p12 z1
x Pz L5
p21 p22 z2
and

z P 1x L6

We can think of Eq. (L.5) as a transformation that takes z in the z1z2 plane and transforms it
to x in the x1x2 plane. Hence, if we can find P, we can make the transformation between the
two state-space representations.

L.4 Finding the Transformation Matrix, P

We can find the transformation matrix, P, from Eqs. (L.3). Since we know all vector
quantities in the equation, we can then solve for pij’s. Notice that the columns of P are the
coordinates of the basis vectors of the z1z2 space expressed as linear combinations of the
basis vectors of the x1x2 space as shown in Eqs. (L.3). Thus the first column of P is Uz1 and
bappL_epdf 12/18/2018 21:19:18 Page 3

A-168 Appendix L Derivation of Similarity Transformations

the second column is Uz2 . Partitioning P, we get

P Uz1 Uz2 L7
Let us look at an example of the transformation of a vector from one space to another.

Example L.1

Vector Transformations to New Basis

PROBLEM: Transform the vector

1
x 2 L8
2

expressed with its basis vectors,

1 0 0
Ux1 0 Ux2 1 Ux3 0 L9
0 0 1

to a vector expressed in the system,

0 0 1
U z1 1 2 U z2 1 2 Uz3 0 L 10
1 2 1 2 0

SOLUTION: Using Eq. (L.2) as a guide, the vector z can be written in terms of the basis
vectors, Uzi .

z z 1 U z1 z2 Uz2 z3 Uz3 L 11

Substituting the values of each Uzi given in Eq. (L.10) as components of the basis vectors,
Uxi , Eq. (L.11) is transformed to the components of x,

0 0 1 0z1 0z2 0z3

x z1 1 2 z2 1 2 z3 0 1 2 z1 1 2 z2 0z3
1 2 1 2 0 1 2 z1 1 2 z2 0z3
L 12

which can be written as,

0 0 1 z1
x 1 2 1 2 0 z2 L 13
1 2 1 2 0 z3
bappL_epdf 12/18/2018 21:19:19 Page 4

L.5 Transforming the State Equations A-169

As we predicted, the columns of P are the basis vectors of the z1z2 space [Eq. (L.10)].
Also,

0 0 707 0 707 1 2 83
z P x1
0 0 707 0 707 2 0 L 14
1 0 0 2 1

In summary, the vector x 1 2 2 T in the x1x2 space transforms into z

T
2 83 0 1 in the z1z2 space. x and z are the same vector expressed in different
coordinate systems.

Now that we are able to transform a state vector into different basis systems, let us see how to
transform the state-space representation between basis systems.

L.5 Transforming the State Equations

We have seen that the same state vector can be expressed in terms of different basis vectors.
This conversion amounts to selecting a different set of state variables to represent the same
system transfer function.
Let us now convert a state-space representation with state vector, x, into a state-space
representation with a state vector, z. Assume the state-space representation shown in
Eq. (L.15).
x Ax Bu L 15a

y Cx Du L 15b

Let x = Pz from Eq. (L.5). Hence,

Pz APz Bu L 16a

y CPx Du L 16b

Premultiplying the state equation by P 1,

z P 1 APz P 1 Bu L 17a

y CPz Du L 17b

Eqs. (L.17) are an alternate representation of a system in state space. The transformed
system matrix is P 1AP, the input coupling matrix is P 1B, the output matrix is CP, and the
feedforward matrix remains D.
We now will show that the transfer function, T(s) = Y(s) U(s), which relates the
output of the system to its input for the system represented by Eqs. (L.17), is the same as the
system of Eqs. (L.15) if, y and u are scalars, y(t) and u(t).
From Eq. (3.73), the transfer function for the system of Eqs. (L.15) is
Y s 1
T s C sI A B D L 18
U s

The transfer function of the system of Eqs. (L.17) can be found by substituting its equivalent
output, system, input, and feedforward matrices into Eq. (L.18). Hence, the transfer function
bappL_epdf 12/18/2018 21:19:19 Page 5

A-170 Appendix L Derivation of Similarity Transformations

for the system of Eqs. (L.17) is

Y s 1
T s CP sI P 1 AP P 1B D L 19
U s

1
Making successive use of the matrix inverse theorem, (MN) = N 1M 1, we find
1
T s CP P sI P 1 AP B D C P sI P 1 AP P 1
B D L 20

Since (sI P 1AP)P 1

= (sP 1
P 1AP),
1 1
T s C P sP 1
P 1A B D C sI A B D L 21

which is identical to Eq. (L.18). Since the transfer function is the same, the system’s poles
and zeros remain the same through the transformation.
We can show more formally that the eigenvalues do not change under a similarity
transformation. The characteristic equation for the system prior to the transformation is
det(sI A) = 0. After the transformation, the characteristic equation is det(sI P 1AP) = 0.
But, I = P 1P. Therefore, the characteristic equation after the transformation can be
written as

det sP 1 P P 1 AP det P 1
sI AP 0 L 22

Since the determinant of the product of matrices is the product of the determinants,

det P 1
sI AP det P 1
det sI A det P 0 L 23

But,

det P 1
det P det I 1 L 24

Hence,

det sI P 1 AP det sI A 0 L 25

Eq. (L.25) shows that the eigenvalues do not change under the transformation.
In this appendix we have shown that a vector, x, in the x1x2 basis system can be
expressed as a vector, z, in the z1z2 basis system using

p11 p12 z1
x Pz L 26
p21 p22 z2

Similarly, the inverse is

z P 1x L 27

We found that the transformation matrix, P, consists of columns, which are the coordinates
of the basis vectors of the z1z2 space expressed as linear combinations of the basis vectors of
the x1x2 space, or

P Uz1 U z2 L 28
bappL_epdf 12/18/2018 21:19:19 Page 6

Bibliography A-171

Using the previous results, the state equations can be transformed from the x state variables
to the z state variables using

z P 1 APz P 1 Bu L 29a

y CPz Du L 29b

Finally, we found that the eigenvalues of the x system are the same as those of the z system.
Hence, the transfer function calculated from either system will be the same.

Bibliography
Timothy, L., and Bona, B. State Space Analysis: An Introduction, McGraw-Hill, New York,
1968.