A-2 Complex numbers

Opposite: The opposite −s1 of a complex number s1 has to satisfy the condition that the sum of a
number and its opposite always yields zero, or

s1 + (−s1 ) = 0. (A.7)

This can be easily satisfied by taking the opposite of the real as well as the imaginary part of s1
A Complex numbers and assembling it again into a complex number, so

−s1 = −a1 + (−b1 ) i = −a1 − b1 i. (A.8)

From the introduction of Fourier series in section 2.2, complex numbers and complex functions are It can be directly seen that this is right by substitution of this result into equation (A.7).
often used in these lecture notes. For instance, the poles and the zeros of a transfer function of a system
generally appear to be complex numbers (section 5.2). The dynamical behaviour of that system depends Product: The product s1 s2 of two complex numbers can be calculated as
to a large extent on the location of those complex numbers in the complex plane (section 5.4). In a
s1 s2 = (a1 + b1 i)(a2 + b2 i) = (a1 a2 − b1 b2 ) + (a1 b2 + a2 b1 ) i, (A.9)
further analysis of the transfer function in the frequency domain, the absolute value and the phase of
complex numbers are used (section 5.5).
where use has been made of the definition of i in equation (A.4).
This appendix gives a summary of a number of those concepts. For a more extensive introduction,
the reader is referred to mathematical subjects. More information about complex numbers can be found Inverse: The inverse (reciprocal) s−11 of a complex number s1 not equal to zero has to satisfy the
on various sites on the Internet. An example is condition that the product of a number and its inverse is equal to one, or
• [Link]
s1 s−1
1 = 1. (A.10)

A complex number s1 is often written down by adding the real and the imaginary parts as in This inverse can be calculated as
1 a1 − b1 i a1 b1
s1 = a1 + b1 i. (A.1) s−1
1 = = = 2 − i, (A.11)
a1 + b1 i (a1 + b1 i)(a1 − b1 i) a1 + b21 a21 + b21
Here, a1 is the real part of s1 ,
from which the real and imaginary parts can be read off.
a1 = Re(s1 ), (A.2)
The expression for the inverse of a complex number in equation (A.11) can be written in a simpler form
and b1 is the imaginary part, by the use of a pair of properties of complex numbers that will be discussed right now. The first is the
absolute value |s1 | of a complex number s1 , which is calculated as
b1 = Im(s1 ). (A.3) q
|s1 | = a21 + b21 , (A.12)
The imaginary part i is defined as the number that yields −1 when it is squared, so
√ so as the square root of the sum of the squares of the real and imaginary parts. We shall presently return
i2 = −1 or i = −1. (A.4) to the meaning of this absolute value, but we remark already now that this calculation for real numbers
yields the same result as the usual approach.
Note that especially in electrical engineering, the letter j is also often used and that M ATLAB accepts an
Secondly, we shall often encounter complex numbers in pairs whose real parts are equal and whose
i as well as a j to denote the imaginary unit.
imaginary parts are each other’s opposites. Then, we speak of the complex conjugate s∗1 of s1 , which
The elementary arithmetic operations for the set of complex numbers are defined as extensions of
satisfies
those operations for real numbers. If we consider beside s1 from equation (A.1) a second complex
number s∗1 = a1 − b1 i, (A.13)

s2 = a2 + b2 i, (A.5) and s1 and s∗1 are then each other’s complex conjugates. Note that the absolute value of the complex
conjugate |s∗1 | is equal to |s1 |. Furthermore, it holds that
then we have the following:
s1 s∗1 = (a1 + b1 i)(a1 − b1 i) = a21 + b21 = |s1 |2 = |s∗1 |2 . (A.14)
Addition: The sum s1 + s2 is found be adding the real parts as well as the imaginary parts separately,
so The multiplication of a complex number with its complex conjugate always yields a real number. And
with this, equation (A.11) for the inverse can be written as
s1 + s2 = (a1 + a2 ) + (b1 + b2 ) i, (A.6) 1 s∗ s∗
s−1
1 = = 1 ∗ = 12 . (A.15)
s1 s1 s1 |s1 |
where the real and imaginary parts of the sum can be read off as a1 + a2 and b1 + b2 , respectively.

A-1
 A-3 A-4 Complex numbers

In other words, a multiplication (or division) of complex numbers can be calculated with the polar
Imaginary G(iω) notation by multiplying (dividing) the absolute values and adding (or subtracting) the phases.
axis
|G(iω)| Finally, we go into the solution of an equation of the nth degree in a complex variable s given by

cn sn + cn−1 sn−1 + · · · + c1 s + c0 = 0. (A.23)

G(iω)
We take real numbers for the coefficients ci in this expression (cn ̸= 0), but, as said, the solution for s
Real axis may be a complex number. It can be shown that this expression can be resolved into linear factors

Figure A.1: Graphical representation of the absolute value |G(iω)| and the phase ∠G(iω) of a complex number cn (s − α1 )(s − α2 ) · · · (s − αn ) = 0, (A.24)
G(iω).
and from this we can read off that there are n solutions (roots) αi for which the equation of the nth
degree (A.23) is satisfied. It is possible that a number of these roots are equal and in that case, one
To represent complex numbers graphically, the complex plane is drawn with the mutually perpendic- speaks of multiple roots. A further consideration of the roots teaches us that those roots either are real or
ular real and imaginary axis in it. The location of a complex number is then determined by measuring form complex conjugate pairs. In other words, if αi is a solution of equation (A.23) with real coefficients
its real part and its imaginary part along both axes. In figure A.1, the complex number s1 = G(iω) ci , then either αi is real or αi is complex and the complex conjugate αi∗ is also a solution of that equation.
has been drawn in this way, which was introduced for the Bode plots in section 5.5. This figure also In general, the solutions of an equation of the nth degree are not easily determined analytically, but
gives the polar representation of the complex number, in which it is expressed in the absolute value for a number of cases, expressions are available for those solutions. The best known example may be
|G(iω)| just introduced and the phase (or argument) ∠G(iω). That absolute value was already known the solutions of a quadratic equation. For the general expression
with equation (A.12), but it can be interpreted with the aid of figure A.1 as the distance of the number in
the complex plane to the origin. The phase is the angle with the positive real axis. If r1 = |G(iω)| and as2 + bs + c = 0 (A.25)
ϕ1 = ∠G(iω), then s1 = G(iω) satisfies with real coefficients a, b and c, the solutions are
s1 = r1 eiϕ1 . (A.16) √
−b ± D
s= , (A.26)
If we compare this polar notation with the expression in a real and an imaginary part as in equation (A.1) 2a
and we use Euler’s formula to rewrite the complex power of e, we get where the sign of the discriminant
s1 = r1 (cos ϕ1 + sin ϕ1 i), (A.17) D = b2 − 4ac (A.27)
so that the real and imaginary parts of s1 , respectively, satisfy indicates whether the solutions are different and real (D > 0), coincide and are real (D = 0) or they are
complex numbers (D < 0). To avoid the presence of a root of a negative number in equation (A.26) in
a1 = r1 cos ϕ1 and b1 = r1 sin ϕ1 . (A.18)
the last case, this expression for D < 0 can be rewritten as
‘The other way around’, these expressions can be written as √ √
−b −D −b 4ac − b2
q s= ± i= ± i. (A.28)
r1 = a21 + b21 and ϕ1 = atan2(b1 , a1 ), (A.19) 2a 2a 2a 2a
If the coefficients a, b and c are real numbers, then this expression yields two complex conjugate solu-
where the function atan2 calculates the arctangent as b1 /a1 , but with taking into account the right tions for D < 0.
quadrant of the complex plane in which the real part a1 and the imaginary part b1 are located; see, for
instance, the atan2 command in M ATLAB. In the discussion of the stability of systems, it is important to consider the sign of the real part of a
The polar notation of complex numbers is very convenient for the computation of multiplications complex number. The part of the s-plane in which the number is located is referred to for this.
and divisions. In the notation with real and imaginary parts, the multiplications can be fairly easily Re(s1 ) > 0 right halfplane RHP
calculated with equation (A.9). A division of two numbers can be worked out by the use of the inverse Re(s1 ) < 0 left halfplane LHP
according to equation (A.11) or (A.15), but this already leads to more complicated expressions. If we If we look at, for instance, the complex solutions of a quadratic equation with a negative discriminant
consider the complex number s1 as in equation (A.16) and in equation (A.28), then the location in the RHP or the LHP is determined by the sign of the real part, so
by −b/2a.
s2 = r2 eiϕ2 , (A.20)

then the product is

s1 s2 = r1 r2 ei(ϕ1 +ϕ2 ) , (A.21)

and the quotient is

s1 r1 i(ϕ1 −ϕ2 )
= e . (A.22)
s2 r2
 B-2 Operations with matrices

For a square matrix A, as the system matrix above, its inverse A−1 can be defined, which has to
satisfy the condition

AA−1 = A−1 A = I, (B.4)

where I is the identity matrix, which is a matrix with all ones on the main diagonal and zeros for all
B Operations with matrices other entries. This matrix inverse A−1 exists if and only if the A is not singular, that is, the determinant
det A differs from zero. The calculation of this determinant and the inverse is still uncluttered for a
2 × 2 matrix. If
 
a11 a12
In the use of state-space equations in the sections 4.3 and 5.2, among others, we encounter a number A= , (B.5)
a21 a22
of properties of and operations on matrices. This appendix gives a short overview of some relevant
concepts and properties of matrices. More information on these subjects can also be found on several then
sites on the Internet. An example is
det A = a11 a22 − a12 a21 , (B.6)
• [Link] on eigenvalues.
and
 
The state-space equations (4.48) for a LTI SISO system (see section 2.3) are 1 a22 −a12
A−1 = . (B.7)
det A −a21 a11
ẋ = Ax + Bu,
(B.1) These analytic expressions become rather extensive for larger matrices and they do not give much in-
y = Cx + Du,
sight. In these cases, we determine the determinant and the inverse only numerically with, for instance,
where x is the state vector with n states x1 , x2 , · · · , xn , u is the input of the system and y is the output. the M ATLAB commands det and inv.
The system matrices are
The eigenvalues and the eigenvectors of a square matrix A are, respectively, the numbers λ and the
A: n × n system matrix, vectors v not equal to the zero vector that satisfy the relation
B: n × 1 input matrix,
C: 1 × n output matrix, Av = λv. (B.8)
D: 1 × 1 feed-through matrix.
In other words, an eigenvector v is a vector that a multiplication by the matrix A transforms into a vector
If we write these as that is a multiple of it. The change of length and a possible change of sign is described by the associated

a11 a12 · · · a1n
 
b11
 eigenvalue λ. These eigenvalues can be found by rewriting equation (B.8) as
 a21 a22 · · · a2n   b21 
A= . (λI − A)v = 0. (B.9)
... .. ...  , B =  ..  ,
   
 .. .  . 
(B.2)

an1 an2 · · · ann bn1 Since the (eigen)vector v must not be the zero vector, this equation can only be satisfied if

C=

c11 c12 · · · c1n

, D=

d11

, det(λI − A) = 0. (B.10)

then the state-space equations can be written out with the help of matrix multiplications as For an n × n matrix A, the calculation of the determinant in this expression will lead to a polynomial of
 the nth degree in λ, the so-called characteristic polynomial
 ẋ1 = a11 x1 + a12 x2 + · · · + a1n xn + b11 u,
a(λ) = det(λI − A) = λn + a1 λn−1 + · · · + an−1 λ + an . (B.11)

ẋ2 = a21 x1 + a22 x2 + · · · + a2n xn + b21 u,



 .. .. ..


. = . . (B.3) By demanding that this expression is equal to zero, we obtain the characteristic equation


 ẋn = an1 x1 + an2 x2 + · · · + ann xn + bn1 u,
λn + a1 λn−1 + · · · + an−1 λ + an = 0, (B.12)




y = c11 x1 + c12 x2 + · · · + c1n xn + d11 u.

whose solutions are the eigenvalues λi of the matrix A.
These expressions illustrate how the multiplication of a matrix and a vector can be expanded. Note that In general, we find n eigenvalues for an n × n matrix A in this way, which can be real and complex
B, C and D are not ‘really’ two-dimensional matrices for SISO systems, but just a column vector, a and some of which can coincide. The utility of these for the state-space equations appears if we consider
row vector and a scalar, respectively. The coefficients are nevertheless denoted with double indices ij, the homogeneous solution for
so that the expressions can be easily generalized for MIMO systems.
ẋ = Ax. (B.13)

B-1
 B-3 B-4 Operations with matrices

We collect the eigenvalues λi and the associated eigenvectors v i of A in two matrices,

 
λ1 0 · · · 0
 0 λ2 · · · 0 
.  and V = v 1 v 2 · · · v n , (B.14)
 
Λ= . .. . .
 
 .. . . .. 
0 0 · · · λn
so we have

(B.15)
 
AV = λ1 v 1 λ2 v 2 · · · λn v n = V Λ.

If the matrix V is invertible, we also have

A = AV V −1 = V ΛV −1 . (B.16)

With this, the differential equation (B.13) can also be written as

ẋ = V ΛV −1 x, (B.17)

or

V −1 ẋ = ΛV −1 x. (B.18)

In the description of the state-space equations, it was remarked that the choice of the state vector is in
no way fixed. The most important demand on this vector is that the united states are able to describe the
state of a system completely. Then the transformed state vector

x′ = V −1 x, (B.19)

can be taken as well, by which choice the homogeneous differential equation reads

ẋ′ = Λx′ . (B.20)

From the diagonal form of the matrix Λ, the solutions of the differential equations can be easily found.
If the eigenvalue λ1 is real, for instance,

ẋ′1 = λx′1 , (B.21)

holds with the general solution

x1 (t) = x1 (0)eλ1 t . (B.22)

If all states of the transformed state vector can be solved for in this way, then the original states can
x′
be calculated with the aid of the inverse relation of equation (B.19).
 D-2 Modelling of system elements

D.2 Material properties

Material Density E G
103 kg/m3 1011 kg/m2 1011 kg/m2
Cast iron 7.2 1.7 0.64
Steel, low carbon 7.8 2.1 0.83
D Modelling of system elements Stainless steel 7.8 2.1 0.83
Aluminium 2.7 0.7 0.26

Table D.1: Overview of the properties of some materials.

D.1 Ideal elements

D.3 Various equivalent quantities
x,v In order to complement the equivalent masses, damping constants and stiffness values given elsewhere
F 1
F a
1 v 1 x
in these lecture notes, the following is presented.
Translating body 1/m
P
m ẍ = m Fext s s
1/m Integrator1 Integrator2
Dashpot damper
θ
h
M α ω θ
1 1
1
Rotating body 1/J
P
θ̈ = J Mext s s
r
J 1/J Integrator1 Integrator2 a v, Fd
x1
Dx F
x2 k

x1 x2 l
Translation spring F = k(x1 − x2 )
k Figure D.2: Dashpot damper
θ1
θ1 θ2 Δθ M The damping constant of the dashpot damper of figure D.2 is
θ2
Rotation spring M = c(θ1 − θ2 ) c
" #
c
h 2 a2 − r2
 
c Sum 6πµl
Deq = 3 a− − r2 −h , (D.1)
v1
Dv
h 2 a − h/2
Fd
v2 D

D where µ is the dynamic viscosity and the other symbols can be read from the figure (source: [4,
v1 v2 tabel 2.3.2]).
Translation damper Fd = D(v1 − v2 )
D
ω1
ω1 ω2 ω2
Δω
D
Md

Rotation damper Md = D(ω1 − ω2 )

D
D Sum

in out

ωout = n1 ωin
Transmission out M in M out
rotation–rotation
Min = n1 Mout
v ω
1/r

ω = 1r v 1/r
Transmission ω F M
translation–rotation v r
1/r
F = 1r M 1/r

Figure D.1: Overview of the ideal elements and their symbols.

D-1
 E-2 Laplace transform

E Laplace transform
x(t) x(s) = L{x(t)} (see equation (E.1))

E.1 Introduction δ(t) 1 Unit impulse

An important application of the so-called Laplace transform is in the solution of differential equations. 1
1(t) Unit step or Heaviside function Re(s) > 0
In section 4.2, the relation between the transfer function and the differential equations of a system has s
been discussed and the Laplace transform plays an important role in it. 1
t Linearly increasing function Re(s) > 0
In this appendix, an overview of some important properties of the Laplace transform that are im- s2
portant for the system analysis is given. For rigorous mathematical derivations of these properties, the n!
reader is referred to one of the books by Beerends and his co-authors [1, 2]. Also on various sites on the tn Re(s) > 0
sn+1
Internet, more information can be found, for instance on: 1
[Link] e−at Re(s) > −a
s+a
E.2 Forward Laplace transform 1
te−at Re(s) > −a
(s + a)2
E.2.1 Definition of the Laplace transform n!
tn e−at Re(s) > −a
In these lecture notes, the Laplace transform L{x(t)} of a time-dependent signal x(t) is denoted1 by (s + a)n+1
x(s) and is defined by ω
sin ωt Re(s) > 0
 s2 + ω 2
∞
s
x(s) = L{x(t)} = x(t) e−st dt, (E.1) cos ωt Re(s) > 0
0 s2 + ω 2
ω
under the condition that the integral exists. According to the Laplace existence theorem, this integral e−at sin ωt Re(s) > −a
(s + a)2 + ω 2
surely exists if u(t) is piecewise continuous in every finite interval in [0, ∞⟩ and u(t) is of exponential s+a
order for t → ∞, that is, there exists a real number σ for which constants M and N can be found such e−at cos ωt Re(s) > −a
(s + a)2 + ω 2
that
x(t − τ ) e−τ s x(s) Equation (E.4)
|x(t)| < M eσt for all t > N. (E.2)
eat x(t) x(s − a) Equation (E.5)
In that case, x(s) exists in any case for real s with s > σ. In some areas of application, only real values
of s are considered, but the definition of the Laplace transform can also be applied, as in these lecture ẋ(t) s x(s) Equation (E.6) with x(t = 0) = 0
notes, for complex s. Then it holds that if x(s) exists for any s = σ + iω, then x(s) also exists for  t
1
all s with Re(s) ≥ σ. If the smallest value of σ for which x(s) exists is equal to σ0 , then the region of x(τ ) dτ x(s) Equation (E.8)
convergence of the Laplace transform of x(t) follows from this as Re(s) ≥ σ0 . 0 s
With the help of this knowledge, Laplace transforms can, in principle, be computed. For a number
of elementary functions, these are given in table E.1. In it we find the Laplace transforms x(s) associ- Table E.1: Some functions and their Laplace transforms. The rightmost column indicates in which part of the
complex plane the Laplace transform exists. The lower part of the table shows how the Laplace transform x(s)
ated with the mentioned signals x(t) as a function of the time calculated as a function of a (complex) changes if a given signal x(t) is changed.
variable s.
To calculate the Laplace transform of other signals, it is not always necessary to go back to the
definition (E.1). A number of properties of the Laplace transform can help to combine the functions
from table E.1; see the next section.
1
Regrettably, there is no agreement in the literature on the precise notation of the Laplace transform of a signal x(t). For
instance, the Laplace transform is also denoted by X(s) or x̃(s) instead of x(s). In that case, sometimes the argument s is
dropped and the Laplace transform of x(t) is denoted by just X or x̃. In these lecture notes, we will sometimes even leave out
the argument s of x(s) to save on writing. However, the context should make clear whether we work in the time domain, so
we deal with x(t), or in the Laplace domain, and so we deal with x(s).

E-1
E.2 Forward Laplace transform E-3 E-4 Laplace transform

E.2.2 Properties of the Laplace transform Initial and final value theorem
The Laplace transform has a number of properties of which we shall gratefully make use in the system For signals for which equation (E.6) may be applied, the initial value theorem and the final value theorem
analysis. hold. The initial value theorem reads
Linearity lim sx(s) = x(0), (E.9)
s→∞
It can be easily seen from the definition (E.1) of the Laplace transform that it is linear, that is,
where x(0) is the initial value of x(t) at t = 0.
L{α1 x1 (t) + α2 x2 (t)} = α1 L{x1 (t)} + α2 L{x2 (t)}, (E.3) The final value theorem reads

for two coefficients α1 and α2 and two signals x1 (t) and x2 (t). lim sx(s) = lim x(t). (E.10)
s↓0 t→∞
Shift in the time domain and in the s domain
The theorem (E.9) as well as the theorem (E.10) is only valid if the limit exists.
For a signal shifted along the time axis, the integral of the definition (E.1) need not be calculated if
the Laplace transform x(s) of the original signal x(t) is already known. For then, the shifted x(t − τ ) Laplace transform of a convolution
satisfies Analogous to the result for the Fourier transforms in section 2.2.2, a property for the convolutions holds.
If he signals u(t) and g(t) are only defined for t ≥ 0, then we can equate them to zero for t < 0. Then
L{x(t − τ )} = e−τ s x(s). (E.4) the definition of the convolution in equation (2.53) can be rewritten as
Such a shift in the time is sometimes called a pure delay. This means that the original signal remains  t
unchanged, but it is available at a later point in time (if τ > 0). A physical interpretation of this shift is, (u ∗ g)(t) = u(t − τ )g(τ ) dτ. (E.11)
0
for instance, a conveyor belt, but it can also be a sensor with a camera that first has to record a picture
and then analyse it before the data are available. It can be shown that the Laplace transform of the convolution satisfies
An inverse relation exists for a shift in the s domain, namely
L{u ∗ g} = L{u} L{g}. (E.12)
L{eat x(t)} = x(s − a), (E.5)
In other words, the Laplace transform of the convolution u ∗ g is equal to the product of the Laplace
where the region of convergence of the shifted Laplace transform x(s−a) is also shifted over a distance a transforms of the separate signals u and g. This property is of great value for the relation between the
in the direction of the real axis. impulse response of a system and the transfer function in section 5.4.

Laplace transforms of derivatives of x(t) and integrals of x(t) E.3 Inverse Laplace transform
For a signal x(t) and its derivative ẋ(t) that are continuous for t ≥ 0 and whose Laplace transforms can The inverse Laplace transform L−1 {x(s)} of a function x(s) in the s domain yields the signal x(t) and
be calculated, it holds that is defined1 as
 γ+∞ i
L{ẋ(t)} = s L{x(t)} − x(t = 0) = s x(s) − x(t = 0). (E.6) 1
x(t) = L−1 {x(s)} = x(s) est ds, (E.13)
2πi γ−∞ i
Expressed in words, this means that the Laplace transform of the derivative L{ẋ(t)} is found by multi-
plying the Laplace transform of the signal itself L{x(t)} by s and possibly taking into account the initial provided that the integral exists. Furthermore, the integration is along a vertical γ in the complex plane
value of x(t) at t = 0, here written as x(t = 0). In the remainder of this appendix, this is written as such that all singularities of x(s) are on the left-hand side of γ.
x(0). So to determine the original function x(t) from x(s) with the above integral, a complex integral has
The property can be extended to higher-order derivatives. If we denote the nth-order derivative of to be calculated. However, the inverse Laplace transform is often found more easily by looking up the
x(t) by x(n) (t), then we have signal x(t) associated with the Laplace transform x(s) in table E.1.
In the study of linear differential equations, often Laplace transforms x(s) are found consisting of a
L{x(n) (t)} = sn x(s) − sn−1 x(0) − sn−2 ẋ(0) − · · · − sx(n−2) (0) − x(n−1) (0), (E.7) fraction of a numerator polynomial and a denominator polynomial in s. If higher powers of s appear in
them, this kind of expression cannot be directly looked up in the table, but partial fraction decomposition
in which the initial values for t = 0 of the signal x(0) and the n − 1 derivatives ẋ(0), . . . , x(n−1) (0) can be used to divide the expression for x(s) in parts that are indeed in the table. This approach is
appear. We gratefully make use of this property in section 4.2, to derive the transfer function from a described, for instance, in the books by Beerends and his co-authors [1, 2] and also in the description of
differential equation describing a system. the M ATLAB command residue. This method is not further discussed here, but see section 5.4.6.
Vice versa, integrals satisfy More or less related to the method of partial fraction decomposition is the so-called method of
 t residuals. With this method, the inverse Laplace transform of an expression of the form
1
L{ x(τ ) dτ } = x(s). (E.8)
0 s (s − z1 )(s − z2 )...(s − zm )
y(s) = (E.14)
In other words, integration of x(t) in the time domain yields a multiplication of the Laplace transform (s − p1 )(s − p2 )...(s − pn )
x(s) by 1/s in the s domain. We also gratefully make use of this property in section 4.2, but also in 1
Just as for the Fourier transforms, regrettably, there does not exist an agreement in the literature on the precise notation of
chapter 5. the inverse Laplace transform. This affects, for instance, the way in which the factor 1/(2πi) is included.
E.4 Laplace transforms and the frequency domain E-5 E-6 Laplace transform

can be calculated. In other words, the Laplace transform y(s) consists of a fraction with a numerator The input is offered a harmonic signal
and a denominator that are both polynomials in s and are factorized. We know from appendix A that this
is possible. Moreover, if the denominator is a polynomial of the nth degree with real coefficients, then u(t) = u0 sin ωt. (E.18)
pi in these expressions are real or occur as pairs of complex conjugates. An analogous consideration With the aid of table E.1, we find that the Laplace transform of it is
applies for the numerator.
ω
The method of residuals gives the inverse for every expression in the form of equation (E.14), but in u(s) = u0 . (E.19)
s2 + ω 2
this appendix, we restrict ourselves to those cases in which none of the pi coincide, that is, pi ̸= pj for
all i ̸= j. In that case, the inverse Laplace transform can be written as With this, the Laplace transform of the input becomes
n ω (s − z1 )(s − z2 )...(s − zm ) ω
u(s) = G(s)u(s) = G(s) u0 = Ku0 , (E.20)
X
y(t) = L−1 {y(s)} = [(s − pi )y(s)est ]s=pi . (E.15) s2 + ω 2 (s − p1 )(s − p2 )...(s − pn ) (s2 + ω 2 )
i=1

An application of this expression will follow on page E-6 in the calculation of the inverse Laplace where G(s) is given by equation (E.17). At first sight, nothing remarkable can be seen in equation (E.20),
transform of equation (E.20). but with the method of residuals for the inverse Laplace transform, conclusions can be attached to it. In
comparison with the expression (E.14), the Laplace transform in equation (E.20) has two additional
E.4 Laplace transforms and the frequency domain factors in the denominator. If we assume for the moment that all pi are different and also differ from
In a comparison of the definitions of the Laplace transform and the Fourier transform in the equa- ±iω, then with the aid of equation (E.15), we can write for the response y(t)
tions (E.1) and (2.46), and also of the inverse Laplace transform (E.13) and the inverse Fourier trans- n
X
form (2.47), some similarities catch the eye. If one takes a signal that differs from zero only for t ≥ 0 y(t) = [(s − pi )y(s)est ]s=pi + [(s − iω)y(s)est ]s=iω + [(s + iω)y(s)est ]s=−iω . (E.21)
and one considers the Laplace transform for the special case in which s is a purely imaginary number i=1

s = ±iω, the agreement is perfect: Substitution of equation (E.20) and further expansion yields
U (iω) = L{u(t)}|s=iω = F{u(t)}. (E.16) G(s)(s − p1 )u0 ωest G(s)(s − p2 )u0 ωest
y(t) = + + ···
Both transforms also appeared to have a comparable property for the convolutions, equations (E.12) s2 + ω 2 s=p1 s2 + ω 2 s=p2
and (2.54). Before a further discussion of the connections of the two transforms, it will not hurt to re- (E.22)
G(s)(s − pn )u0 ωest G(s)u0 ωest G(s)u0 ωest
member the differences. For instance, the Fourier transform (equation (2.46)) and the inverse transform + + + ,
(equation (2.47)) are only defined for signals for which the concerned integrals exist. The analysis could s2 + ω 2 s=pn s + iω s=iω s − iω s=−iω
be restricted to signals that are absolutely integrable on the whole real axis, that is, signals for which and so
 ∞
|u(t)| dt [G(s)(s − p1 )]s=p1 u0 ω p1 t [G(s)(s − p2 )]s=p2 u0 ω p2 t
y(t) = e + e + ···
−∞ p21 + ω 2 p22 + ω 2
exists, and for which the property lim u(t) = 0 also holds. A signal that is constant and not equal to (E.23)
t→±∞ [G(s)(s − pn )]s=pn u0 ω pn t G(iω)u0 ω iωt G(−iω)u0 ω −iωt
zero for t ≥ 0, so u(t) = h, does not satisfy these conditions. + e + e − e .
p2n + ω 2 2iω 2iω
The integral for the Laplace transform according to equation (E.1) exists for a larger class of signals,
for instance for all signals of exponential order conformal to the demand of equation (E.2). The constant Since all pi are different and in addition are different from ±iω, the first n terms of this expression can be
signal mentioned above, for instance, does satisfy this condition. calculated, because the denominators of the form p2i + ω 2 are not equal to zero and also the expressions
The difference between both transforms is also conspicuous in the area of application. Because the such as
Laplace transform demands relatively few conditions on the signal u(t), it can be used (in principle) to (pi − z1 )(pi − z2 ) · · · (pi − zm )
analyse the behaviour of a system for a broad gamut of input signals. This concerns, among others, the [G(s)(s − pi )]s=pi = K (E.24)
(pi − p1 )(pi − p2 ) · · · (pi − pi−1 ) (pi − pi+1 ) · · · (pi − pn )
response to periodic signals, but also explicitly the study of transient phenomena.
Fourier transforms, on the other hand, are especially useful to describe a stationary situation in which can simply be evaluated for all pi . If it is also the case that all pi have a negative real part, then it can be
a system is indefinitely periodically driven. The Fourier transform can be used in that case to calculate noted that the powers of e in the first n terms of equation (E.23) all approach zero for large t. In other
from which frequencies the signal is built up. The word ‘stationary’ in this context does not mean that words, after the fading away of the so-called transient effects, that is, the first n terms, the response y(t)
the system is driven by a constant signal and nothing changes. This notion ‘stationary’ can positively is equal to
be interpreted in the frequency domain as a system that is driven with an indefinitely long signal with a G(iω)u0 ω iωt G(−iω)u0 ω −iωt G(iω)eiωt − G(−iω)e−iωt
constant frequency content. Subsequently, with the description in the frequency domain, the response of y(t) = e − e = u0 . (E.25)
2iω 2iω 2i
the system for each of these contributions with a particular frequency can be determined and with them,
the complete response can be reconstructed. Here, G(iω) and G(−iω) are ordinary complex numbers calculated by substituting s = ±iω into the
This last-mentioned result can, of course, also be found with a Laplace transform. To illustrate this, expression for the transfer function (E.17). Since the pi and zi in that expression are real or pairs of
we consider a system whose transfer function can be written as (see also section 5.3.3) complex conjugates, G(iω) and G(−iω) are also each other’s complex conjugates. If we write G(iω)
in the polar representation (A.16) as
(s − z1 )(s − z2 )...(s − zm )
G(s) = K . (E.17) G(iω) = |G(iω)|eiϕ (E.26)
(s − p1 )(s − p2 )...(s − pn )
E.4 Laplace transforms and the frequency domain E-7

with the phase

ϕ = ∠G(iω), (E.27)

then

G(−iω) = G(iω)∗ = |G(iω)|e−iϕ . (E.28)

With this representation, the response becomes

ei(ωt+ϕ) − e−i(ωt+ϕ)
y(t) = u0 |G(iω)| = u0 |G(iω)| sin(ωt + ϕ). (E.29)
2i
We see from this that the harmonic input signal of equation (E.18) with the circular frequency ω and
the amplitude u0 results, after the dying away of the transient effects, in an output signal with the same
circular frequency. The original amplitude is multiplied by a factor |G(iω)| and there is a difference in
phase the size of ϕ given by equation (E.27).
In summary, the response of the system to a harmonic signal after the fading away of the transient
effects is completely described by G(iω), that is, the transfer function G(s) evaluated at s = iω. This
property is used in section 5.5.
With this approach, we have provided the intuitive suggestion at the beginning of this section to make
a connection between the Laplace transform and the Fourier transform through the substitution s = ±iω
with a foundation. From this approach, also the mentioned differences for the typical application of
both transforms appear. In the Fourier or the frequency domain, we especially look at the behaviour of
one or more contributions in the input signal consisting of harmonic signals with certain frequencies.
This is the stationary situation mentioned earlier. In the analysis of the Laplace transform, we obtain
the complete solution including the transient effects of the response of the system on a harmonic signal
switched on at t = 0. We see that, on the condition that the transient effects die away, after some time
indeed the response considered in the frequency domain remains.
This condition of the fading away of the transient effects is essential and non-trivial for this. We saw
that this means that all pi must have a negative real part. This is not the case for all systems, however.
We shall encounter systems in these lecture notes for which a pair of pi are purely imaginary and equal
to ±iω1 . This ω1 will then appear to be an undamped eigenfrequency of the considered system. If
such a system is provided with a harmonic signal with a (circular) frequency ω on its input, then the
expected response at that frequency ω will be found in the output signal. Besides that, however, an
oscillation at the eigenfrequency ω1 can be excited and, because it is undamped, this oscillation will not
fade away. Even more dramatic is the situation if at least one of the pi has a positive real part. Then,
the system is unstable and for a system of this kind, a solution exists with an exponentially increasing
term in equation (E.23). Also in this case, a fading transient behaviour is out of question and indeed,
the response on the harmonic input signal will be almost invisible in the output signal becoming larger
and larger. It is tempting to overlook this possibility if only G(iω) is considered in the analysis in the
frequency domain. These examples make it clear that this only gives a useful picture if all pi have
negative real parts.