Index

Chapter 2 The Theory of Systems in Automation

............................................ 1

Chapter 3 Representations with the state of systems

linear, stationary in finite dimension
12
3.1 Some simple models of continuous time systems
13
3.1.a A simple mechanical system .................. 13
3.1.b An elementary electrical network . . . . . . . . . . . . . . . . . . . . . 16
3.1.c Electric motor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2 Equivalent representations: time systems
continue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.3 Equivalent models of different physical systems ................. 22
3.4 Linear approximations of nonlinear systems:
continuous time systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.4.a Two examples of application . . . . . . . . . . . . . . . . . . . . . . 25
ii Index

3.5 Further examples of continuous-time systems . . . . . . . . . . . . . 27

3.5.a Circuit with Tunnel Diode . . . . . . . . . . . . . . . . . . . . . . . 28
3.5.b Newtonian dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.5.c Satellite in circular orbit . . . . . . . . . . . . . . . . . . . . . . 33
3.5.d Vertical dynamics of a missile . . . . . . . . . . . . . . . . . . 34
3.5.e Direct current electric motor . . . . . . . . . . . . . . . 36
3.5.f Magnetic levitation . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.6 Discrete time systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.6.a Linear time-invariant systems in discrete time . . . . . . . . . . . 41
3.6.b Discrete-time representations of systems
continuous time - discretization - . . . . . . . . . . . . . 42
3.6.c Some simple examples of time systems
discrete
3.6.d Linear approximations of nonlinear systems
46
3.6.e Equivalent representations . . . . . . . . . . . . . . . . . . . . . 47

Chapter 4 The Stationary Linear Representations

in the domain of time . . . . . . . . . . . . . . . . . . . . . . . 49

4.1 Continuous-time systems: implicit representations

and explicit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.1.a The transition matrix in the state . . . . . . . . . . . . . . . 53
4.1.b The free response in the state . . . . . . . . . . . . . . . . . . . . . 58
4.1.c the natural ways . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.1.d The natural modes: the general case * . . . . . . . . . . . . . . . . . 65
4.2 Discrete-time systems: transition to the model
explicit
4.2.a The transition matrix . . . . . . . . . . . . . . . . . . . . . . . . 69
4.2.b The natural modes for discrete-time systems
...
 Index iii

4.2.c The natural modes under discretization . . . . . . . . . . . . . . 73

4.3 Further observations * . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.4 Natural ways in the state and out...
4.4.a the case of irregular . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.4.b the general case * . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.5 Appendix 3 ........................................................ 79
4.5.a Spectral form of a matrix . . . . . . . . . . . . . . . . . . . 79
4.5.b The Jordan form * . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

Chapter 5 The stationary linear representations

in the complex domain . . . . . . . . . . . . . . . . . . . . . . . 86

5.1 The Laplace transform in the study of systems

at a continuous tempo
5.2 The transition matrix in the complex domain . . . . . . . . . . . 89
5.3 Natural modes in the complex domain . . . . . . . . . . . . . . . . . . 92
5.4 The transfer function . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.4.a Poles and zeros of the transfer function . . . . . . . . . . . 98
5.4.b The representations of the transfer function
100
5.5 The steady-state response . . . . . . . . . . . . . . . . . . . . . . . 102
5.5.a The permanent regime with periodic entries . . . . . . . . . . . 105
5.5.b The permanent regime with canonical entries . . . . . . . . . . . . 107
5.5.c The index response . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
5.6 The laws of motion in the case of multiple eigenvalues . . . . . . . . . . . . 114
5.7 Discrete-time systems: analysis in the domain
of the complex variable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
5.7.a The Z-transform in the study of systems
a discrete time.........................................................................116
iv Index

5.7.b The natural modes in the complex domain . . . . . . . . . . . . . 119

5.7.c The steady-state response . . . . . . . . . . . . . . . . . . 123
5.7.d The steady-state response to periodic inputs . . . . . . . . . . . 123
5.7.e The steady-state response to canonical inputs . . . . . . . . . . . . 126
5.7.f The laws of motion in the general case . . . . . . . . . . . . . . . . . 129
5.7.g Discretization of the transfer function
of a continuous system . . . . . . . . . . . . . . . . . . . . . . . . . . 131
5.7.h From the forced model install the representation
with the state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
5.8 Appendix 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
5.8.a The Laplace transform . . . . . . . . . . . . . . . . . . . . . . . 134
5.8.b The Bode diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
5.8.c The polar diagrams . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
5.8.d On the calculation of the transition matrix* . . . . . . . . . . . . 161
5.8.e The Z Transform .................................................................................. 163

Chapter 6 The geometric properties of space

of state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171

6.1 The properties of the internal structure: Observability

171
6.2 The properties of the internal structure: Accessibility
179
6.3 The Kalman decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . 191
6.3.a An example of application . . . . . . . . . . . . . . . . . . . . . . 196
6.4 The properties of the state space . . . . . . . . . . . . . . . . . . . . . . 199
6.4.a Unobservability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
6.4.b Accessibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
 Index v

Chapter 7 Linear input-output models and representations

with the state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
7.1 The problem of realization . . . . . . . . . . . . . . . . . . . . . . . . . 204
7.2 The minimum realizations...
7.2.a A minimum realization technique in
case dipeqmaggiori than 1 * . . . . . . . . . . . . . . . . . . . . . 219
7.3 Introduction to the Study of Interconnected Systems
221
7.3.a Interconnected systems: model calculation . . . . . . . . . . . 229

Chapter 8 Elements of stability theory . . . . . . . . . . . . . . 236

8.1 Stability: definitions and conditions . . . . . . . . . . . . . . . . . . . . . . . 237

8.2 The Routh Criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
8.3 The general method of Lyapunov* . . . . . . . . . . . . . . . . . . . . . . . 251
8.3.a Lyapunov for stationary linear systems . . . . . . . . . . . . 255
8.4 Stability through linearization . . . . . . . . . . . . . . . . . . . . . . 259
8.5 The Stability of Entry - Exit . . . . . . . . . . . . . . . . . . . . . . . . . . 266
8.6 The stability of interconnected systems . . . . . . . . . . . . . . . . . . . . 274
8.7 The stability of discrete-time systems . . . . . . . . . . . . . . . . . 285
8.7.a Conditions and criteria for linear systems . . . . . . . . . . . . . 285
8.7.b The Lyapunov criterion . . . . . . . . . . . . . . . . . . . . . . . . . . 288
8.7.c The Lyapunov criterion for linear systems
discrete-time stationary . . . . . . . . . . . . . . . . . . . . . . 289
8.7.d Study of stability through linearization
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
Elements of Systems Theory

The present notes introduce the study of stationary linear dynamic systems.
Classes of models that can be assumed to represent, possibly in
approximate way, the behaviors of various phenomena and natural processes
and artificial. The proposed analysis methods are preparatory to the techniques of
design of automatic systems and control.

After a brief introduction intended to succinctly introduce

from the perspective of Automation, the second chapter introduces the class of
models that are the subject of the study: representations with linear state,
stationary at finite dimension. These are systems of differential equations or
all differences suitable for representing continuous-time or discrete-time systems
Creation. The third and fourth chapters present the results of the analysis in the domain
of time and the complex variable, up to the characterization of the com-
frequency portamento, an approach that is typical of engineering. The
The fifth chapter is dedicated to the study of the properties of state space: the
reachability and observability, two properties that play a particular role
important mind in the project of control systems. The sixth chapter deals with
of dynamic systems resulting from the interconnection of multiple subsystems; to
2 1. Elements of Systems Theory

starting from basic connections, it is shown how to calculate the model

of the resulting system in the linear case and the study methods are introduced
of the properties of the interconnected system. The seventh chapter is dedicated to the
study of stability; that property which expresses the system's ability to
to confine, possibly compensate, the effect of the disturbances.

Salvatore MONACO
professor of Systems Theory
Department of Computer, Automation and Management Engineering 'Antonio
Ruberti
University of Rome 'La Sapienza'
via Ariosto 25 - 00185 Rome
[Link]: [Link]@[Link]
2. System Theory in Automation

In these notes, in support of the teaching of Systems Theory at

second year of the degree course in Computer Engineering and Automation of the
Sapienza - BIAR - introduces the basic methods for studying systems.
stationary linear dynamics in finite dimensions - LSdF -.
It is a particular class of systems, but of great importance
in applications as the models that describe them can be imp-
used in numerous contexts to represent the behaviors of many
phenomena and real processes. This derives from the fact that this class of systems captures
one of the fundamental components of the complexity of the processes that operate
real-time: the dynamics; that characteristic based on which the
evolutions depend not only on interactions with the environment, but also on
context, a set of factors that define what we will call the 'state'.
del sistema.
The analysis methods introduced here are preparatory to the proposed developments.
in the subsequent teachings of the Automation program at BIAR.
This is why the following will attempt to present the con-
the theory of systems, automation, control systems, ..

1.1 The Theory of Systems ..

2 2. System Theory in Automation

Theory, delegation, long line of theories

to examine, to observe) sent to the great celebrations
religious, to consult the oracles. Just as the set of acquisitions of
the theory was accepted to interpret and understand the phenomena, a theo-
ria in the current sense consists of a logically coherent formulation
(in terms of concepts and more or less abstract formulations) of a set of
definitions, principles and general laws that allow to describe, interpret,
classify, explain, at various levels of generality, aspects of natural reality and
social, and the various forms of human activity.
System, daσισθηµα(sistema): "meeting, complex". In the scientific field
scientific, any object of study that, although made up of different elements
reciprocally interconnected and interacting with each other or with the external environment,
reacts or evolves as a whole with its own general laws. It is about
more or less complex aggregates assumed to represent processes or phenomena
that works (evolves over time) by interacting with the environment.
The Theory of Systems consists of a body of methodologies for analysis of the
real phenomena and processes with the aim of providing support for understanding
based on their behavior, develop models that can reproduce it
the functioning and developing analysis methods in order to interpret and where
possible to predict its behavior.
But these same methods are also necessary for designing systems and
"automatic" devices.

.. in Automation

Automatic: from αυτωσ (autos), 'self', µατωσ (matos), 'task'

Tiki, 'technique', should be understood as that set of methods and techniques.
for the study and design of systems capable of operating autonomously
prefissaticompiti.
Automa, from the same roots, is the product of the project: an entity that is in
degree to solve on its own.
In other words, it is that discipline which aims to
the understanding of the principles underlying autonomous behavior
 3

regarding their subsequent translation into methods and project techniques

realization of engineering products.
In the context of Automation, tasks represent behaviors or
operating modalities, predefined and/or imposed from the outside, and autonomy
is understood as the ability to function as desired by compensating in
real-time the effect of disturbances and uncertainties. All this, moreover, must not
necessarily happen by replacing human intervention, but possibly
to support or integrate it.
The objects to which automation directs its attention are therefore
those who, needing to fulfill tasks, evolve over time interacting with
the environment; in other words: they operate in real time. For example, consider the sis-
the theme of regulating the body's temperature whose functioning is
characterized by the temperature over time of the different organs: the temperature
it must be kept within defined values compensating for the effects due to
variations in external temperature or changes in health status. It
think of an information distribution system whose operation
and characterized by the quantity of information that flows over time in the
node version, in this case we want the information to reach the destination
compensating for the effects of faults and interference. Think again of a system
of automatic guidance whose operation is characterized by the position and
settling over time of the vehicle: regardless of disturbances and malfunctions
the system will have to allow reaching the set goal.
Automatic systems are extremely common: household appliances,
automobiles, the systems and networks for energy distribution, the transmission networks
sensation of information, space systems, ... , robots, financial systems and
management, .. , living beings. Each of these "entities" contains, and gives the
result, of the interaction of numerous automatic components. Beings
living beings are the seat of natural automatic processes of complexity and beauty, in
molti casi non ancora compresa, che hanno ispirato e continuano ad ispirare
the development of research in the sector.
As can be inferred from the previous considerations, Automation is a tec-
pervasive technology, yet unknown; it has been baptized for this reason the
hidden technology par excellence. This is a consequence of the fact that automation,
4 2. System Theory in Automation

the automatic behavior occurs 'de facto' through procedures of

processing of the informational content related to the operation of the system,
procedures that can then be implemented using different technologies
second of the application sector: mechanical technology, electronic technology,
.. , computer science.

The architecture of automatic systems

In an attempt to understand architecture from a conceptual point of view

we can refer to a conceptual scheme for an automatic system
how well it interprets its functioning. In fact, the functioning of a sis-
an automated system designated for a predetermined task is based on the ability
decompose the activity in terms of elementary actions, organize it
sequence of actions and coordinate the execution, carry out the elementary actions.
These three phases correspond to a sequencing that is not only conceptual,
but also procedural, can be understood as fundamental components
of a hierarchical intervention logic at three levels in which the results of the processing
reactions at one level activate the processing procedures at the lower level.
This means that from a procedural point of view we can think of the system
agent according to a scheme of the type indicated in the figure.

In the diagram, the downward arrows represent interaction.

between the levels of intervention, those upwards an inverse flow of information
necessary to ensure the established modes of operation in
conditions of autonomy.
 5

Planning

supervision

intervention

Figure 2.1

A decision-making and intervention process that does not consider this flow
the inverse of information would imply a perfect correspondence between
planning and execution of the task which is not possible in the presence of
uncertainties and disturbances. In reality, in fact, the knowledge of the way of
the functioning of the process is not free from uncertainties (model uncertainties) and
Interaction with the environment can give rise to unforeseen disturbances.
It is therefore necessary to foresee that during normal operation
the decision-making and intervention process should be able to adapt, with elaborations
ad hoc that use measures of the deviation between expected behavior and
the effective feedback that has been mentioned.
This is represented in the diagram of figure (1.1) where inside each
stage, and in the connection between them, the interventions (u(t)) are activated not only
from the information of the descending flow, v(t) and z(t), but also from information
in feedback, m(t) (feedback). To better understand the three-tier structure
the characteristic of an automatic system, think of a robotic system for
the exploration of a planet. Imagine a mobile platform that needs to
carry out excursions, collect soil samples, capture images, carry out
elementary analysis on site and transmit the results... In accordance with the intuition
we can imagine the completion of an exploration mission like the
result of a complex procedure composed of the three indicated phases.
6 2. System Theory in Automation

An initial planning will have to define the path that the robot will have to follow.
seguire e il sequenziamento delle operazioni; le elaborazioni necessarie saranno
activated by external inputs that represent the desired objectives (number of
samples to be taken, experiments, ..), and will take into account the characteristics
of the soil detected through images or other types of data. This corresponds
to the first phase of the proposed plan. The path and sequencing have been planned.
During the operation, the devices of the robotic system must ensure
to the organization of the sequencing of activities, to activate and coordinate them
the execution; what corresponds to the second and third phase of the scheme with-
proposed settlement. The close interaction in the execution of the
verses phases and the need to reactivate processes based on the results that
Obstacles and unforeseen barriers are detected along the way.
It may be necessary at a high level to redefine the trajectory or to replan.
that the execution of scheduled soil survey experiments, as well as
the malfunction of an actuator or the modification of the characteristics of
soil makes it necessary to modify the execution interventions at the level
lower according to automatic modes. In this context, and with reference-
related to the architecture indicated, in the teachings of Automatic Controls
The Foundations of Automation specifically study the methods that
they are at the core of the execution level project. The methodological approach
of the project, and many of the concepts used are common to the three phases, that is
gives the study characteristics of adequate generality.

.. and control systems

With specific reference to the lowest level of automation, and imagine

even though it represents the entire project, we will try to address it in the following.
in the light of the approach that is followed in the design of an automatic system
that must ensure the execution of predefined elementary actions to ensure
the desired automatic behavior based on a given process.
The intervention strategies devised to ensure the desired level
in automation are also referred to as control strategies - from the English meaning
The term 'to control' means 'to intervene to' -. Such strategies, together with the tec-
niche of realization through appropriate components define the sector of
 7

competence of automatic control.

The project naturally unfolds in multiple phases that involve collaboration.
collaboration of industry experts with automation engineers. It
they distinguish between conceptual phases and phases related to technology.

The problem arises from the definition of automation objectives,

from the identification of external variables, inputs, and disturbances, from the in-
division of uncertainties regarding the behavior of the process and the measurements
available for real-time verification of the system's behavior. A
starting from these elements, the specifications of the com-
overall system performance, the control system.
In this first phase, from an abstract point of view, the process can be
represented by a black box that represents the functional link
between the variables represented by the incoming arrows, entries and disturbances, and
variables represented by the outgoing arrows, the outputs and the measures: the variables
through which the automation objectives are expressed, the first,
which represent the observations on the behavior of the system, the second.
The problem consists of the design and implementation through appropriate dis-
positives of implementation, processing and measurement of the control subsystem that
connected to the process as indicated in the conceptual diagram of figure (1.2),
It creates as a whole the control system that, in response to external stimuli
of reference, meets the desired automation objectives.
z(t)

v(t) u(t) y(t)

C P m(t)

Figure 2.2

What basic examples come to mind regarding the problem of maintaining a preset?
temperature profile in an oven or refrigerator, compensating for the effect of
8 2. System Theory in Automation

perturbations caused by the introduction of other bodies or heat exchanges unim-

visits, starting from measurements of the temperature itself; think, again about the prob-
the aim of ensuring a predetermined motion trajectory for a mobile platform
despite obstacles and roughness of the terrain starting from position measurements.
In the diagram, the control system, C, takes care of the calculation.
of the control action based on measures of the trend of the variables of
process (the outputs among these) and any measurements of the disturbances. Pos-
We can therefore imagine the calculation of the control action as the result
of the two distinct actions: a first one related to the measurement of the variables of the pro-
process that we will define as interaction feedback because it acts based on
of feedback information on the behavior of the system; a second component
go to the measure of the disturbances that we will define as forward compensation
(feedforward) as it is intended to produce a pre-calculated intervention that
compensate the effect of the disturbances themselves.
The diagram, at a greater level of detail regarding
regarding the connections and components, it can be used for the project
of the control subsystem based on quantitative relationships (the mod-
the mathematical equations that describe the relationships between the variables of the components themselves.

Since the models of the system and the various components are an approximation
of physical reality and do not describe with absolute precision the links between the vari-
abilities, the control subsystem calculated from this schema must
garantire il mantenimento delle modalita` di comportamento richieste entro
predefine variability margins of component models. This pro-
the robustness of the control system is indicated as necessary
would ensure the fulfillment of the project specifications of the system
physical.

The project of a control system

The project arises from the need to create a system

of control that ensures an automatic behavior of a phenomenon or
process that we can think of as pre-assigned.
The following steps characterize its development:
 9

identify the quantities that can be used to express the

desired automation objectives and the factors that influence them: who-
we will love the outputs of the system, the variables that represent the primes, the
y(t), the inputs representing the seconds, u(t) and d(t).
In doing this we will have had the foresight to distinguish the inputs in
two categories: the effective inputs, u(t), through which we will be able to ex-
to exercise the intervention to ensure the desired behavior, the disturbances,
d(t), which represent variables that we can possibly measure,
but through which we cannot intervene;
identify a model, preferably mathematical, suitable for describing the
cause-effect link that allows for quantitative expression
the trend of outflows in relation to inflows. It is a model
of the oriented system that represents the process;
to conceive, based on an ad hoc methodology that allows for an intervention:
necessary to dismantle the individual
the design methods are based on those of analysis;
implement the intervention: in this phase the colleague is particularly evident
in connection with the application sector and technology.
The components of the control system are oriented dynamic processes.
contexts in which a natural cause-and-effect relationship exists, consider the
portamento in response to external stimuli, and the dependence is dynamic in the sense
that behavior is conditioned not only by the present cause, but also
from the past. In short, in the figure schema, the complex is a dynamic system
oriented just as its parts are. In the diagram, the following can be distinguished:
- the process, the system that one wants to intervene on, represented by the block
marked with P, in which it is possible to identify the control magnitudes.
u, I disturb, which act on the process and which, in general, are magnitudes
non-manipulable, the controlled variables that represent the magnitudes of
whom one wishes to influence the course to ensure the satisfaction of the
specific, and the measured variables, which represent the measures that are
used to develop the intervention strategy;
- the controller, marked with C, which processes a suitable
control law based, on one hand, on the measured quantities
10 2. System Theory in Automation

of any measurable disturbances and, on the other hand, of the reference quantities
v, which represent the external commands.
The theory of systems and control encompasses a wide range of
project methodologies with reference to predefined families of models as-
it's used to represent the plant or process. If the plant is described by a...
Linear systems of differential equations are well documented.
synthesis methodologies of controllers. In many cases, it is possible to refer to
to these methodologies even if the process is not linear, using its
approximate representation.
The main analysis methods for this class of systems are the subject of
queste note preliminari allo studio dei metodi dell’automatica.
An implicit representation with the state of a system of dimension
finished, adopting the vector notation, is of the type

∆x(t) = f(x(t), u(t), z(t)) m(t) = w(x(t), z(t)) y(t) = h(x(t), z(t)) (1)

in which the time can be assumed continuous (coinciding with the set
of real numbers) or discrete (coinciding with the set of relative numbers); in
d(x(t)
correspondence will be the derivative operator dt or the operator of
unitary advance x(t+!). The behavior of the system with respect to time t is
therefore described by five types of variables: u(.) is the vector of the variables of
control, they represent the channels through which it is possible to modify
the behavior of the process; z(.) that of the disturbances, magnitudes that influence
they characterize the process behavior, they are sometimes measurable, but they cannot
be modified; m(.) represents the measured quantities; y(.) the outputs of
process, these are a subset of the measured variables and represent the
quantities on which control is to be exercised; x(.), finally, are the variables
of the state and represent, with their evolution over time, the behavior
of the system in its most complete form. The function f, the generating function,
h, the output transformation, ew, the transformation of the measures, have a
structure corresponding to classes of processes and depend on parameters that
we will support notes in an initial phase. If, for example, such functions are linear
 11

the mathematical model represents a linear process. The mathematical model-

The introduced attic can be taken to represent a vast class of systems.
they are characterized by continuous evolution over time,
of a finite number of variables. In the terminology of Systems Theory
(1) constitute the differential representation with the state of a s
stationary continuous-time theme, finite dimension. A formulation
equivalent is used to describe discrete-time systems in which the
evolutions occur in a discontinuous manner at intervals of amplitude
constant assumed for unitary convention; with the conventions adopted and ...
by taking the derivative with respect to time at the state value in (1)
time (t+ 1), x(t+ 1), a representation is obtained with the state of a
discrete-time system, stationary, of finite dimension.
With reference to the diagram in figure 2.2, control theory studies
the conception of the mathematical model of the subsystem that interacts with
the process in the indicated manner ensures the fulfillment of preset
specifics of the overall system. According to this scheme, the control law
it depends on the measurable variables and additional external variables, v(t); m(t)
ev(t) are the inputs of the controller, v(t) are the inputs of the system
control. Note that if often only the outputs are measurable and the scheme
Figure 1.1 is modified accordingly. If it is assumed that the inputs are
deterministic variables and that the measurements are not affected by noise
the control problem is called deterministic. If, on the other hand, it is assumed that
some or all inputs are random variables, of which some are known
statistics, and that the measurements are affected by noise the control problem is
stochastic saying.
3. Representations with the state
of linear, stationary systems
of finite dimension

In this chapter, we want to show, with the help of simple examples, that
the behavior of numerous physical systems, phenomena and processes, from different
disciplinary sectors can be described by systems of differential equations,
continuous-time systems, or prime difference systems, discrete-time systems,
type

∆x(t) = f(x(t), u(t)) y(t) = h(x(t), u(t)) (1)

The interaction with the environment, which is realized through the inputs, u(t),
and the outputs, y(t), give life to a cause-and-effect relationship that is expressed completely
using a set of auxiliary variables, the state variables.
In the hypothesis of linearity of the functions: Rn xRp→Rn edh:Rn xRp→
Rq on their respective domains of definition, the representation with the state (1)
take on the form

∆x(t) = f(x(t), u(t)) = Ax(t) + Bu(t)x(t)0 ) =x0∈Rn(2.a)

 3.1. Some simple continuous-time system models 13

y(t) = h(x(t), u(t)) = Cx(t) + Du(t) (2.b)

where the dot indicates the derivative with respect to time, A,

(n×n), is called dynamic matrix, B, (n×p), input matrix,
C, (q×n), output transformation matrix and D, (q×p),
matrix of direct connection input output. It is a representation
with the finite state size equal to adn, the dimension of the state vector x,
that describes the behavior of a given process with pressure, the size
of the vector, execute, the size DIY.
A mathematical representation of this kind is sometimes intrinsic, and
that is consequent to specific properties of the physical system, but, as we will see,
It can also be used to roughly describe behavior.
system's mento. In other words, it can be the result of an operation of
approximation based on a more complex model.

3.1. Some simple models of time systems

I continue
The construction of representations with the state of processes and phenomena
falls into a broader context, that of modeling, which involves the
skills of experts from various application sectors. It represents
for itself a problem of considerable interest, but it still presents itself as a
preliminary step for the application of the techniques of System Theory that
develop general analysis methods for fixed classes of representations.
The stationary linear representations, that is, those of the type (2.a)−(2.b),
are suitable for describing, possibly in an approximate way, a vast
gamma of phenomena of interest in the field of Engineering.
The following shows, with reference to simple phenomena, how
the knowledge of the physical laws that underlie the application sector
naturally lead to the formulation of the model 'representation with the
state space
14 3. Representations with the state of linear, stationary systems of finite dimension

A simple mechanical system

The mathematical description of the motion of a constrained cart.

in a predetermined direction due to the effect of acting forces (external or internal) and
connected to a fixed wall by means of a spring and a damper, consisting of
is a significant example for the variety of situations that, although in a way
approximated, it can represent.
If we assume that external forces are the cause of motion, the model
The mathematician in question expresses the cause-effect relationship that exists between forces.

Eastern and the progress over time of the displacement. It provides a

oriented mathematical representation (input-output) of the simple system
mechanic in training.
Consider the mechanical system in the figure where the spring and the
dampers are two devices capable of generating internal forces
in the presence of cart movement: a force that shifts slightly
The force with respect to an equilibrium position can be assumed to be proportional.
to the same displacement, the spring, a force proportional to the speed of the
movement, the damper.

x
Figure 3.1

The motion of the cart due to external forces acting along

the direction of motion is for small displacements with respect to the equilibrium position
book, can be described by imposing the balance of the external forces acting,
u(t), internal forces (generated by the motion, the spring, and the damper). For small-
in fact, the spring exerts a force opposite to the displacement,
 3.1. Some simple models of continuous-time systems 15

y(t) is proportional to its modulus according to an elastic constant k, while

the damper exerts a force in the same direction and proportional to the ve-
locality(t) according to a damping coefficient b. If u(t) is indicated with
force applied to stimulate movement, according to Newton's second law that
express the equilibrium of the forces acting on the center of mass of the entire device
it is supposed to be rigid, there is

My(t) + ky(t) + by˙(t) = u(t)

A representation with the state can easily be associated with prayer.

differential equation that describes the dynamics of the mechanical system.
It is sufficient to set the position and the state variables as
speed of the mass. This assumption can also be understood by a
intuitive viewpoint if one observes that the state is that variable that has in
At the time all the necessary information to describe the behavior
of the datin poi system: in a rigid mechanical system, composed thus of
parts that can be assimilated to material points, with respect to inputs that express forces
external agents, the position and velocity of individual material points at time t
completely characterize the subsequent evolution of the system due to
each fixed entrance. With this clarification, we can place

x1 (t) = y(t), x2 (t) = y˙(t),

deriving and taking into account the balance of forces, one obtains

ẋ(t)
1 = x2 (t)

ẋ(t)
2 = -k x1 (t) + -b x2 (t) + u(t)
M M M
y(t) = x 1 (t)

che assume la forma (2.a)−(2.b) con

0 1 0
A= ! -k -b " B= ! 1 " C= ( 1 0 )
M M M
16 3. Representations with the state of finite-dimensional linear stationary systems

It is a stationary linear representation, in continuous time of

two-dimensional, at an input, the applied force u(t), and an output, it moves
mentoy(t). This simple mechanical system is suitable for representation
in the first approximation, various phenomena, even complex ones, of interest in
applications of different nature; among these: the dynamics of devices of as-
vibration absorption, vehicle suspensions, simplified dynamics of
structures, etc.

With reference to this type of model, the study of steady-state behavior

in the presence of periodic stresses, it takes on particular interest: think of,
With this purpose, it represents the mechanical systems that have been mentioned.
studio, carried out in the third chapter, allows to establish a precise connection
between time domain analysis methods and those more traditional based on,
Indeed, on the frequency response.

3.1.b. A basic electrical network

It is an electric network powered by voltage u(t), composed of two

circuits: the first with generator, resistance R1the capacitor C on the branch
common, the second with capacitor C, inductance L, and resistance R2 .
R1 L
+ +

u R2 y
C
- -
Figure 3.2

If we assume the voltage at the terminals of the resistance R as the output2 , a

representation with the state can be calculated by imposing the equilibrium of
tensions in each link and currents at the nodes. This is obtained from the balance of
tensions at the first stitch
 3.1. Some simple models of continuous-time systems 17

u(t) − vc (t)
u(t) = R1 I1 (t) +vc (t)=⇒i1 (t) =
R1
from the sum of the currents at the node

dvc (t) dvc (t) 1 1 u(t)

i1 (t) = C +il (t)=⇒ =− vc (t)−il (t) +
dt dt R1 C C R1 C

from the equilibrium of the tensions to the second loop:

ofl (t) ofl (t) 1 R2

vc (t) = L +R2 Il (t)=⇒ = vc (t)− il (t)
dt dt L L
Post

x1 (t) = vc (t) x2 (t) = il (t)

the following representation is obtained with the stationary linear state

1 1 1
=)tẋ( − R1 C −C R1 C
! 1 R2 " x(t) + ! 0 "
u(t)
L − L
y(t) = ( 1 0 )x(t)
It should be noted that it is always possible to associate with an electrical network
Post of resistors, capacitors, inductors, transformers, and ideal rotators
a linear representation of size equal to the number of elements with
memory (capacitors, inductances, transformers and rotors) having fixed
as state variables the currents and/or the voltages on such elements.
Characteristic problems of electrical network theory are those that
they can be related in systematic terms to the properties of the states. In fact
useful in the theory of electrical networks to determine if for a given configuration
preset voltage and/or current values can be reached in some
network branches; whether or not there are redundant elements in the circuit that are unnecessary for the

characterization of input-output behavior. Such problems may

to be formulated and resolved based on the results of chapter four.
18 3. Representations with the state of finite-dimensional linear stationary systems

3.1.c. Electric motor

Si consideri un motore in corrente continua alimentato in tensione. Tale

device, as is known transforms electric power into mechanical power
generating a couple on the motor axis due to magnetic induction
The electrical part can be represented by a resistance, R, and an inductance.
Led a force generator against electromotive, andc The mechanical part
can be schematized with an inertial flywheel driven by a torque
Cm (t) generated by induction. It is generally assumed that a coefficient is present.
dynamic friction coefficient F. The input is the voltage in the armature circuit
u(t) is the output and angular velocity of the motor shaft. The electrical part is
described by the bond trau(t) ei(t) which results:

to
u(t) = Ri(t) + ec (t) +L
dt
dovec (t) = kc ω(t). The electrical and mechanical transduction occurs for generating
action of a couplem , of intensity proportional to the intensity of the current
i(t)

Cm (t) = km i(t)
the dynamics of the mechanical part is described by

dω(t)
J +Fω(t) = Cm (t)
dt
It is ultimately obtained

R kc 1
i̇(t) = -i(t) - ω(t) + u(t)
L L L
km F
=)t(ω̇ i(t) - ω(t)
J J
We ultimately obtain a stationary linear system of size
two with statox(t) = (i(t) ω(t) ′ .
R Kc 1
A= ! − L − L
B= ! L C= (O1 )
km -F " 0"
J J
 3.2. Equivalent representations: continuous-time systems 19

where the angular velocity was chosen for the output.

The system is inherently interconnected, that is, made up of different subsystems.
systems as it results from the previous modeling considerations and is placed
highlighted in the following figure.

u I Cm ω
EE km
m
M

kc
c
Figure 3.3

The study of interconnected systems, the methods of constructing mod-

she starting from the descriptions of the parts as well as the study of the properties a
starting from the properties of the parts are important topics of the theory of
systems that we will face for the class of systems under study and limited
mind some aspects in the fifth chapter.
20 3. Representations with the state of finite-dimensional linear stationary systems

3.2. Equivalent representations: time systems

continuous
The concept of equivalence between representations with state is connected to
non-uniqueness in the choice of the state to describe a given dynamic phenomenon
ico. With reference to the simple examples presented, it is immediate to realize
the fact that different choices are possible and that there correspond relationships
different presentations from each other. Just think of the electric circuit: the state is
represented by the variables of memory elements and we have chosen
tension and current in the case of the studied example. An alternative choice and
equivalent for the purpose would have been the electric charge on the capacitor and the
flow in the inductance.
Limiting ourselves to consider linear transformations of state variables.
We will show how to calculate the corresponding representations with the state.
Assigned the system

A
=
+
B)xutẋ(
y(t) =Cx(t) +Du(t)
with a linear transformation of state variables through a matrix T
non-singular constant
z=Tx |T|≠0
one obtains the representation

ż()=
tTẋ()=
tTAx()+ tTAT−1 z(t) + TBu(t)
tTBu()=

y(t) = CT −1 z(t) + Du(t)

there is still a system of the same type with dynamic matrix TAT−1 , ma-
triple of inputs TB, and matrix of outputs CT−1 The student will not have
difficulty in applying the previous considerations to some of the previous examples
teeth.
An important observation that helps in the calculation of a transformation of
The state variable is the following: if you remember that x is a n-tuple that represents
a vector, siav, with respect to a reference frame, that is
 3.3. Equivalent models of different physical systems 21

0
1 0
⎛ 1⎞
⎛ 0⎞ ⎛ ⎞...
x = Σxi eI= . x1+ ⎜ 0⎟ x2+...+ xn
⎜ ⎟. ⎜ ⎟.. ⎜ 0⎟
⎝.⎠
⎜ 0⎟ ⎜ ⎟ ⎜ 1⎟
⎝.⎠
⎜ 0⎟ ⎝ ⎠

people

x=T−1 z
0
⎛ .. ⎞
.
⎜ 0⎟
⎜ ⎟
(T−1 )i=T−1 · ⎜⎜ 1 ⎟ i−ma
⎟
⎜ 0⎟
⎜ .. ⎟⎟
⎜ .⎟
⎜
⎜ 0 ⎟⎠
⎝
(T−1 )i , the i-th column of (T −1 and the representation of the i-th vet-
tower of the new base (the one in which the vector is represented by z) with respect to
to the old base (the one in which the vector is represented by dax). This means
to perform a linear transformation of the state variable of the type
z=T means, from an operational point of view, choosing a new basis.
∼ ∼
of the state space that is represented by1 , ..., e n in the coordinates ten;
that is
∼ ∼
T −1= (e 1 , ..., e n )

As an example, the reader calculates the coordinate transformation that

invert the order of the state variables.
A concluding observation aims to clarify, with a simple example, that
There are nonlinear representations of linear systems. Assigned for this purpose,
the following scalar linear system, the following calculations are immediate:

+
b)ut)=
(axt()tẋ( z=ex x=lnz

ż()=
t ex (ax(t) + bu(t)) = az(t)lnz(t) + bz(t)u(t)
22 3. Representations with the state of finite-dimensional linear stationary systems

and clarify what has been stated, namely that there are non-linear representations
of linear systems.

3.3. Equivalent models of different physical systems

It is true that there are different mathematical models (the representations

equivalents) that describe the same physical system, it is also true that one
the same model can describe different physical systems. It is precisely this aspect,
typical of the mathematical abstraction to which we refer, which guides towards
a classification into model classes and the study of the properties of the class.
This aspect is highlighted with a simple example.
Consider the two physical systems in the following figure; the first related to
on the dynamics of temperatures T1eT2in the two isolated internal rooms
a component as the temperature of the assumed external shell varies
of metal; the second related to the dynamics of the tensions at the ends of the two con-
dissipators of the electric circuit in figure. This circuit can be considered
the electrical equivalent of the thermal system.
If it is assumed that the electric voltage v(t) corresponds to the tem-
the temperature T(t) and the current i(t) correspond to the amount of heat q(t), the
equations of conduction and thermal storage

1 dT(t) q(t)
q(t) = (T2 (t)−T1 (t)) =
R dt C
trovano una corrispondenza immediata nelle equivalenti elettriche

1 dv(t) i(t)
i(t) = (v2 (t)−v1 (t)) =
R dt C
where it is assumed that they represent resistance and thermal capacity
electric power, expressed in degree/calorie and calorie per degree, ohm farad,
respectively.
With the established correspondences (and0 (t) correspond to aT0 (t)) it's not difficult
check that the following equations define a different representation
circuit branch valid for the thermal system
 3.4. Linear approximations of nonlinear systems: continuous time systems 23

1 1 1 1 1
v̇(t)
1 =− ( + )v1 (t) + v2 (t) + e0 (t)
C1R1 R2 C1 R3 R1 C1

1 1 1 1 1
v̇(t)
2 = v1 (t)− ( + )v2 (t) e0 (t)
C2 R3 C2R2 R3 C2 R2

3.4. Linear approximations of nonlinear systems:

continuous-time systems
So far we have constructed linear representations of linear systems. It is
It is common in practice to use linear representations to describe in a
approximated nonlinear dynamic systems around fixed points of functions
mento (linear approximations around equilibrium points).
This aspect is the generalization to the case of dynamic systems, of the sem-
please concept of approximation of a curve in the vicinity of a fixed point
through the tangent at that point. With reference to the generic function not
linear f(x) around a fixed value f(x)e the series development halted at
first order, provides

df
y = f(x) = f(xe ) + (x−xe ) +....
dx ))xe
)
)
df
the postoxa=x−x e ,ya=y−y e ,m= dx a linear relationship is obtained
next that approximates the curve around adxe

ya=mxa .

The same idea can be applied to the generating function of a check

nonlinear continuous time system

u)x(=
,fẋ f ( x e , ue ) = 0

y=h(x, u) h(xe , ue ) =he

obtaining:
24 3. Representations with the state of finite-dimensional stationary linear systems

∂f∂f x−xe
f(x, u) = f(xe , ue ) + ! ∂x∂u " ) ! +...
)xe ,ue u−ue "
)
)
∂f ∂f
=f(xe , ue ) + (x−x ) + (u−ue ) +...
∂x ))xe ,ue ∂u ))xe ,ue
e
) )
and place ) )
z = x - xe , v = u - ue

∂f ∂f
A= , B=
∂x ))xe ,ue ∂x ))xe ,ue
) )
if so: ) )

ż=
()tAz+
()tBv+
()t.
In the same way for the exit

∂h∂h x−xe
h(x, u) = h(xe , ue ) + ! ∂x∂u " ) ! +...
)xe ,ue u−ue "
)
)
∂h ∂h
=hand+ (x−xe ) + (u−ue ) +...
∂x ))xe ,ue ∂u ))xe ,ue
) )
the place ) )
∂h ∂h
C= , D=
∂x ))xe ,ue ∂x ))xe ,ue
) )
)
the linear approximation is obtained )

ya (t) =Cz(t) +Dv(t).

This is the linearized model that approximates around the pair (xe , ue ) he
∂f ∂f they are the derivatives of the vector
behavior of the nonlinear system. ∂x e ∂u
compared to the state vector and the output vector, respectively.
If for the first, called the Jacobian matrix,
∂f 1 ∂f 1
... ∂x
∂f ∂x 1
= ⎛ ⎞
n
J(x) = ..
∂x ∂f n ∂f n
⎝ ∂x 1 ... ∂x n ⎠
 3.4. Linear approximations of nonlinear systems: continuous-time systems 25

The Jacobian matrix (n×n), calculated at xethe doge is a constant matrix

that we indicated with A; the derivative with respect to u is a matrix (n×p)
that we have indicated with B; the derivatives of respect to x in (xe ue ) I am
the constant matrices CeDdi dimensions (q×n) and (q×p) respectively.
These considerations extend the possibility of using linear models.
Therefore, linear theory allows for the local study of systems not
linear. The interest in switching to the approximated model does not lie, however,
Tavia, in using the model to describe the behavior of the given system
how in its use to infer some properties related to behavior
of the given non-linear system. Among these, the property of stability will be
shown in the sixth chapter.

3.4.a. Two examples of application

Pendulum
Si consideri un pendolo di massam, sospeso ad un’asta rigida di peso
negligible in length on which it acts according to the tangent to the motion a
external force. A mathematical model is to be written that represents
feel the movement on the horizontal, originating from the equilibrium point, of the
projection along the vertical of the mass of the pendulum. The balance of forces
along the tangent to the motion allows us to write

¨ + mg sin θ(t) + klθ(t)

mlθ(t) ˙ = u(t)

Where represents the external force acting, and it is an efficient coefficient of attraction.
˙ concerning the
dynamic. Setting x1 (t) = θ(t) ex2 (t) = θ(t) if it is,
equations of state,

ẋ(t)
1 = x2 (t)

k
-g thank you1 (t)−x 1
ẋ(t)
2 = 2 (t) + u(t)
l m ml
that is a nonlinear representation of the type
26 3. Representations with the state of linear, stationary systems of finite dimension

0
+)xt(=
f)tẋ( ! 1 " u(t)
ml
in which the output is equal to

y(t) = lsenx 1 (t).

The calculation of equilibrium pairs corresponding to null input, u.e= 0,

providese2= 0,xe1=hπcio`e all the states of equilibrium with the mass
in the vertical positions above and below the attachment point. Around the states of
equilibrium above the attachment point has the linearized model:

ẋ(t)
1 = x2 (t)
g k 1
ẋ(t)
2 = x1 (t)−x2 (t) + u(t)
l m ml
y(t) = lx 1 (t)

for others:

ẋ(t)
1 = x2 (t)
g k 1
dx2ot(t) = -x1 (t)−x2 (t) + u(t)
l m ml
y(t) = lx 1 (t)

Dynamics of two interacting species

The following model describes the dynamics of two interacting species, prey and
predator, it was introduced in 1926 by the Italian mathematician Vito Volterra and
takes its name. This model is able to interpret the most salient aspect.
of such a phenomenon: it concerns the presence of equilibrium situations that, if
perturbate, they see the emergence of oscillation phenomena (alternation of development
among the species).
It is assumed, in the formulation of the model, that:
The prey grows, in the absence of predation, according to the so-called equation
logistics
 3.5. Further examples of continuous-time systems* 27

dx1ot(t) = ax1 (t)−kx12(t)

it is a differential equation that clearly highlights both a
exponential growth trend in the early phase of evolution and for
a
modest density values, an asymptotic tendency towards a limiting value, k
(carrying capacity), which takes into account limiting factors such as, for example, the
limited resources
the predator relies solely on its prey and in the absence of
this decreases according to an exponential trend governed by the following
differential equation

ẋ(t)
2 = -cx2 (t)
the predation rate is proportional to the product of the individuals of
two species
With these assumptions, the equations that describe the evolution of species
can be easily deduced from the previous equations. It is obtained:

ẋ(t)
1 = ax1 (t)−bx1 (t)x2 (t)
dx2ot(t) = -cx2 (t) +dx1 (t)x2 (t)
y1 (t) = x1 (t) y2 (t) = x2 (t)
So once again there is a nonlinear model of the same type
from the previous one. It is easy to verify that in this case there are two points of
equilibrium: one trivial, corresponding to the absence of species, the other
c
x2 (-c+dx1 = 0 → x1e =
d
ad−kc
x1 (a−kx1−bx2 0 = x2e =
bd
Around such a point of equilibrium, the evolution is approximated by a sis-
linear theme in free evolution with
kc bc
A= ! − d − d
ad−kc
b 0 "
28 3. Representations with the state of finite-dimensional stationary linear systems

3.5. Further examples of continuous-time systems*

3.5.a. Circuit with Tunnel Diode

With reference to the circuit in Figure 5, the conditions of equilibrium of the

tensions and currents give

u(t) = Ril (t) +vc (t) +vl (t)

iC (t) = il (t)−iT(t)

The current flow in the diode depends on the voltage across its terminals.
according to a nonlinear relationship, IT =h(vTreported in the graph of Figure
6. From the previous equalities, with obvious steps, we obtain
R L
iT
vC
C

Figure 3.4

dvc (t)
C =il (t)−h(vC (t)
dt
ofl (t)
L =−vC(t)−Ril (t) + u(t)
dt
With positions similar to the previous ones

x1 (t) = vc (t) x2 (t) = il (t)

the following representation is obtained with the state

 3.5. Further examples of continuous-time systems* 29

h(x1 (t))x2 (t)

ẋ(t)
1 =− +
C C
x1 (t) R 1
dx2ot(t) =− x2 (t) + u(t)
L − L L
The equilibrium conditions can be calculated by solving the system.
of equations

h(x1 )x2
0 =− +
C C
x1 R 1
0 =− x2+ U
L − L L
Solving with respect to x2one obtains the equality

U 1
x1=h(x1 )
R − R
which for fixed values of UedRammette has one or three solutions as it turns out
evident from Figure 6

iR

e1
e2
e3
vR
Figure 3.5

The circuit under examination from an experimental point of view shows a behavior
so-called bi-stable actuator on the two equilibrium states 1ede3in response to
sufficiently large impulsive stresses.
How it can be verified based on the theory outlined in chapter 7,
such behavior depends on the properties of the equilibrium states resulting
e1 ede3establish2unstable.
30 3. Representations with the state of finite-dimensional linear stationary systems

Reserving the right to resume the study with one of the investigation techniques
what we will study, the trend of effective behavior can be verified
cat through numerical simulation. To this end, one can assume the
following parameter values: u = 1.2V, R = 1.5Kohm, C = 2pF, L = 5µH
scaling the time to nanoseconds and measuring the currents in
milliamps, the circuit equations become

dx1ot(t) = 0.5[-h(x1 ) +x2 ]

ẋ(t)
2 = 0.2(-x1-1.5x2+ 1.2)
Perh(x1 it can be assumed

3 5
h(x1 ) = 17.76x1-103.79x1 2+ 229.62x1-226.31x1 4+ 83.72x1

Under these conditions, equilibrium points can be calculated.1(0.063, 0.758)

e2(0.285,0.61) ee3(0.884, 0.21). The simulations were carried out starting from
different initial states (x1 (0), x2 (0)) show the presence of two families of
converging trajectories of hers1ede2is separated by a curve, called a separator,
passing pears2it consists of two converging trajectories2 The whole pi-
it is therefore divided into two regions of attraction1ede3 This allows you to
understand the actual behavior of the circuit in which starting from the gen-
the initial condition the circuit passes into the corresponding condition of
balance, the condition2not being able to be maintained due to the presence
from physical noise. The transition from one equilibrium to another occurs as a result
of the presence of an excitation that leads the initial state into the region of
opposite attraction.

3.5.b. Newtonian dynamics

The dynamics of the rectilinear translational motion of a rigid body under-

Posted to an external force(t), it can be obtained from the expression
following that expresses the balance of the acting forces
 3.5. Further examples of continuous-time systems* 31

my¨(t) +fa (t) + fe (t) +fv (t) = u(t)

in whichfrepresents Coulomb and dynamic friction, f,e a possible

elastic force edfva viscous damping.
Diverse mathematical representations of the forces acting can be
used to describe the phenomenon.
As for elastic force, it depends on the displacement.
The first approximation is well represented for small displacements from one
linear relationship of the type

feky

For significant movements, one of the following two relationships can be

employee

k(1−a2 y 2 )y |ay|<1
which models a decrease in elastic force as the displacement increases
corresponding to a failure of the elastic characteristics; or
a relationship of the type

k(1 + a2 y 2 )y

which models an increase in elastic force as displacement increases.

corresponding to a stiffening of the elastic properties.
As for the viscous frictionv=h(y˙), h(0) = 0 and for small
speed can be assumedv=
Combining linear viscous damping with a restoring elastic force
rigid and a periodic external stress Acosωt, the equation is obtained.
of Duffy

my¨(t) + cy˙(t) + Ky(t) + Ka3 y 3=Acosωt

classic in the study of periodic excitation of nonlinear dynamics.

32 3. Representations with the state of finite-dimensional linear stationary systems

Infinite, as far as the frictional force is concerneda , this manifests a com-

portamento that depends on the conditions of motion. In resting conditions
if there is a force, f,s parallel to the surface that takes any values between

−µs mg,µsstatic friction coefficient ranging from 0 to 1. The motion occurs

in the presence of an acting force greater than the aforementioned limit value. In the absence of

external forces the mass remains stationary until when|f e |µs mg. In motion the
friction force takes the value of µk mg, µk dynamic friction coefficient.
the conclusion is reached

fa=µk mgsign(y˙) per|y˙|>0 efa=fs per y = 0.

The dynamics of displacement in the regime of small displacements, in pre-

without linear viscous damping and static and dynamic friction assumes,
in the state variables1=yex2=y˙:

ẋ=x
1 2

ẋ= -k x1 (t) + -b x2 (t) + u(t)

M -η(x1 , x2 )
2
M M
in which the nonlinear function η is equal to µk mgsign(y˙) per|y˙|>0,−kyper
ẏ=0eyµmg/kg−µ
|s s mgsign(y) pery˙ = 0 e|y|> µs mg/kg.

It is possible to simplify the analysis to study the situations in which2e

different from zero. Perx2>0 is obtained

ẋ=x
1 2

ẋ= -k x1 (t) + -b x2 (t) + u(t)

M −µk g
2
M M
and forx2<0

ẋ=x
1 2

ẋ=
2
-k x1 (t) + -b x2 (t) + u(t) +µk g
M M M
 3.5. Further examples of continuous-time systems* 33

It should be noted that different approximate linear models can be calculated.

to describe the dynamics under different operating conditions. Modeling
piecewise linear function.

3.5.c. Satellite in circular orbit

u1
r u2
ϑ

Figure 3.6

Consider the dynamics of an artificial satellite in polar coordinates.

onsiqeutalainetreffdidreorocnedshetbydesexpreranad

˙2 β
r¨(t) = r(t)θ(t)− +u1 (t)
r(t) 2

ṙ()tθ()tu2 (t)˙
θ̈(t) = -2 +
r(t) r(t)
In the absence of radial thrust1and the ring road2the solutions are ellipses,
hyperbolas or parabolas. The simplest orbit is a circle

r(t) = c 1 θ(t) = c 2
1
Conr(0) = r0 r˙(0) = 0, θ(0) = θ0 ,θ(0) =ω0˙eω0= ( )
β 2
r30
The nominal orbit is thus described by

rr (t) = r0 θr (t) = ω0 t+θ0

34 3. Representations with the state of finite-dimensional linear stationary systems

˙
Postox1=r,x2=r˙,x3=θ,x4the following representation is obtained
station with the state

ẋ(t)
1 = x2 (t)
β
ẋ(t)
2 = x1 (t)x42(t)− +u1 (t)
x12(t)
ẋ(t)
3 = x4 (t)
x2 (t)x4 t u2 (t)
ẋ(t)
4 = -2 +
x1 (t) x1 (t)
The linearization of the dynamics, that is, the description of the deviations
compared to the nominal orbit

r0
0 0 ⎛ ⎞
ur (t) = ! " x(t)
r =
0 ω0 t+θ0
⎜ ⎟
⎝ ω0 ⎠
from the following stationary linear representation

0 1 0 0 0 0
⎛ 3ω20 0 0 2r0 ω0 ⎞ ⎛ 1 0⎞
ż()=
t z(t) + u(t)
0 0 0 1 0 0
⎜ ⎟ ⎜ ⎟
⎝ 0 -2ωr
0
0
0 0 ⎠ ⎝ 0r0 ⎠

3.5.d. Vertical dynamics of a missile

h0

Figure 3.7
 3.5. Further examples of continuous-time systems* 35

The balance of forces acting on the missile in the figure is written

m(t)v˙(t) = −m(t)g + )vte(ṁ

doveverepresents the relative velocity of the expelled mass, assumed

constant. Place x1=h,x2=vex3msi obtains the following rep-
presentation with the state

x2
⎛ ve u(t) ⎞
=)tẋ(-g+ x3
⎝ u(t) ⎠
.
In the particular case, very frequent in practice, where one assumes a
mass variation, u0 constant, that which corresponds to assuming that

m(t) = m0+u0 t

it is obtained
m(t)v˙(t) = -m(t)g + ve u0

and, postox1=h,x2=v

0 1 0
=)tẋ( ! x(t) + !
0" "
ve u0
0 - (m0 +u0 t)

y(t) = ( 1 0 )x(t)x(0) = 0

The variations of the developments compared to a reference trend,

those in which the variation of mass is assumed to be constant are represented
from the linearization around the following reference trajectory

x3r (t) = m0+u0 t

u0
x2r (t) = -gt + veln(1 + t)
m0
g m0 ve u0 u0 u0
x1r (t) = -t2+ 1+ tln 1 + t− t
2 u0 *+ m0 , + m0 , m0 -
In the variable z = x - xrev = u - ur , it is obtained:
36 3. Representations with the state of finite-dimensional linear stationary systems

0 1 0 0 0
ż()=
t ⎛ 0 0−(m+you teav)e2uz(t)+
0 ⎞ ⎛ ve ⎞ v(t)z(0) = x(0) - ⎛ 0 ⎞
0 0 (m0 +u0 t)
⎝ 0 0 0 ⎠ ⎝ 1 ⎠ ⎝ m0 ⎠

3.5.e. Direct current electric motor

iR

Rs
ω Ls vs

Figure 3.8

One wants to calculate the mathematical model that describes the dynamics of
motore elettrico in corrente continua di figura. Si assume che l’ingresso sia la
excitation circuit voltage, u(t) = vs (t). For the excitation circuit
the following equation holds for the loop that expresses the balance of tensions

ofs (t)
Vs (t) = Rs Is (t) +Ls
dt
The generated flow can be expressed by

Φ(t) = Ls is (t)

For the armature circuit, the following equation holds that expresses the equilibrium
of the tensions:

ofr (t)
Vr (t)−ec (t) = Rr ir (t) +Lr
dt
in which the electromotive force is equal to
ec (t) = kc Φ(t)ω(t).
Finally, regarding the mechanical part, there is
 3.5. Further examples of continuous-time systems* 37

dω(t)
J +Fω(t) = Cm (t)
dt
in which the drive pair, Cm , and the date of

Cm (t) = km Φ(t)ir (t)

In the case of lossless transformation of electric power into

mechanical powerKc=Km=K, and later1=is ,x2=ir ex3=ωSi
obtains the mathematical model

Rs 1
ẋ(t)
1 = -x1 (t) + u(t)
Ls Ls

Rr Vr KL s
dx2ot(t) = -x2 (t) + − x1 (t)x3 (t)
Lr Lr Lr

F KL s
dx3ot(t) = -x3 (t) + x1 (t)x2 (t)
J J

It is once again a nonlinear model that can be lin-

created around equilibrium conditions to describe approximately
I kill the behavior of the system for small variations compared to that situation.
equilibrium action.
To conclude, it should be noted that if the circuit current is assumed to be constant
the excitation obtains the linear model introduced in paragraph 2.1.c..

3.5.f. Magnetic levitation

Consider the device in the figure where an electromagnetic coil

is powered by current to support the weight of a sphere and keep it in place
equilibrium at a distance xefrom the coil.
38 3. Representations with the state of finite-dimensional linear stationary systems

Figure 3.9

The attractive force exerted on a metallic body as a function of

distance for different current values is represented in the graphs of the figure.

Figure 3.10

From this, it can be deduced that a ball weighing 8210−3Newton can be

maintained in equilibrium at a distance of 2.7 mm with a current of 600
mA.
Denoted by M the mass of the ball and conf(x, i) the force exerted by the
Bobina, the balance of forces at a point on the vertical at a distance x
 3.5. Further examples of continuous time systems* 39

writes
Mx¨(t) = f(x(t), i(t)) − Mg
which indicates the acceleration due to gravity.
Around axleethe linear approximation perf(x, i)

∂f ∂f
f(x, i) = f(xe , ie ) + (xe , ie )(x−xe ) + (xe , Ie )(i−ie )
∂x ∂i
The derivatives with respect to ax,kx , and adi,ki they are calculated in the following [Link] the
slope at constant current equal to adieat the point of x-axis e ; in the
the case in question is equal to 14 N/m. kicorresponds to the variation in strength,
inxe with respect to the current; in the case at hand peri1= 700mA the force is equal
a 12210−3Don't worry3= 500mA The force is equal to 4210−3N. So it has therefore

80−3
k i= = 0.4N/A
700−500
By substituting the numerical values, the following differential equation is obtained,
which governs the motion around the equilibrium (ζ=x−x eeu=δi)

¨ = 14ζ(t) + 0.4δi(t)
(3210−3)ζ(t) ζ̈(t) = 1667ζ(t) + 47.6u(t)

To avoid that in the execution of numerical simulation procedures and

the values of the variables treated are too different from each other (or too
large or small absolutely in relation to the characteristics of the processing system
prayer), which would imply a non-homogeneous propagation of errors
of rounding (or saturation phenomena), it is good practice to ensure
that the coefficients of the equations assume comparable values. Values between 0.1 and
10 represent a desirable situation. To achieve this, one can proceed
to a scaling of the variables, which also has a physical interest as
corresponds to a selection of units of measurement (scaling of variables) and
of the execution speed (scaling of time) more suitable for the study of
phenomenon. In the case in question, for example, in the previous equation the dis-
dance are expressed in meters, while a unit in mm is certainly more suitable
to describe the phenomenon.
40 3. Representations with the state of finite-dimensional linear stationary systems

At a first level of generality, we can say that a scaling of the

variables correspond to assuming a new variable, v,n , starting from a
assigned, v, and a new temporary variable, tn , instead of, according to the
following expressions

v tn=ω0
vn =
v0 t
If it is observed that the effect of scaling with respect to time means that

d(·) d(·) d(·)

= =ω0
dt d(tn /ω0 ) dtn

substituting in the motion equation in the study, one obtains

ζn (t)
(ω20 ζ0 d2 ) = (1667ζ0)ζn (t) + (47.6u0 )u(t)
dtn2
that is
d2 ζn (t) 1667 47.6u0
2 = 2
ζn + u(t)
dtn ω0 ω20 ζ0
Choosingω20= 1667 (ω0 approximately equal to 40) the coefficient of ζ(t) is even
1
to one that corresponds to taking a timeline equal to 40 seconds.
If moreover the displacements are measured in cm and the currents in Amperes,
that is chosen0= 0.01 eu0= 1, the equation is obtained

d2 ζn (t)
=z(t) + 2.86u(t)
dtn 2
Set in scale
Regarding the general case of a system of differential equations
first-order linear in the state variable x ∈ R nand entered∈Rp ,
the scaling operation leads to the following system

ẋ(t)
n = 1/ω0 S−1 −1
x ASx+ 1 over ω0 Sx BS u un

in which the real number ω0represents the variation of the temporal scale and Sx
edSuthey are square diagonal matrices, (n×n) and (p×p), of the coefficients of
scalature associated with transformation
 3.6. Discrete time systems 41

xn=S−1
x x un=S−1
u u.

3.6. Discrete time systems

If the time variable evolves at regular intervals of fixed width,

Assumed for unitary convention, the dynamic system is discrete time.
The study of discrete-time linear systems presents a strong analogy
with what has been done for continuous time systems and the results very often
they allow the same formulation. It will therefore be possible, following the same
introductory presentation line to introduce the main results of this chapter
analysis.

3.6.a. Discrete-time stationary linear systems

If the time variable is discrete and under the assumption of linearity of the function
generator of the product space (X×U) the representation takes on the
shape.

x(t+1) = f(x(t), u(t)) = Ax(t) + Bu(t)x(t)0 ) =x0Element of Real Numbersn (9.1.a)

y(t) = η(x(t), u(t)) = Cx(t) + Du(t) (9.1.b)

The implicit model introduced immediately suggests a

physical schema through which to realize the system using devices
which generate a delay, sum instant by instant of functions and the
multiplication by constants. The scheme, shown in Figure 2.11, includes
blocks,A,B,C,D, consisting of multipliers for constants and operators
of delay, highlights the possible structure of the input-output connection; there is
a part that highlights an iterative calculation is typical of the part
purely dynamic and an instantaneous part.
42 3. Representations with the state of finite-dimensional linear stationary systems

D
u (t ) x (t+ 1) x(t ) +
1
B + z C + and
( )t
+

A
Figure 3.11

3.6.b. Discrete-time representations of systems

continuous time - discretization -
Discrete-time linear representations can be used for
describe or approximate continuous-time systems as specified in
followed.
Assigned a continuous-time system where the inputs are piecewise constant.
on intervals of fixed width, if you are interested in calculating the evolutions in the
state and outgoing at the moments when the input changes it is possible to refer to
to a discrete-time system. The situation is represented in the figure
2.12; the calculation of the equivalent discrete model is known as the problem of
discretization. The time instants t=kT are called sampling instants.
sampling, and the sampling devices (on the output and the evolution of
they are synchronous with each other and with the input generation devices
continuous and constant in parts. This problem finds application in the study
of real systems connected to digital devices; in such circumstances the input is
constant in parts (and possibly deducted from the sampling of an input
(continued) and the magnitudes are all reduced to a discrete time scale.
multiple of the elementary calculation cycle.
A different circumstance that leads to referring to the problem of discretization.
...and when the continuous inputs are sampled, as indicated in the
 3.6. Discrete-time systems 43

figure 2.13; in that case the discrete model approximately describes the
behavior of the continuous-time system championship.

H S
TC
u y
Clock
S TD

Figure 3.12

The discrete-time representation that describes the input-state relationships

The output can be calculated easily by computing the evolution in the
state is exiting between two sampling moments and remembering that the input
m
etysnuosiocntheom
trfA
[Link]
cahoinspm
oifrestni nattocnssi tI

A
=
+
B)xutẋ(
. x(t0 =x0
y(t) = Cx(t)

t t0=kT
x(t) = eA(t−t ) x00+ / eA(t−τ ) Bu(τ)dτ
t0 t = (k + 1)T
(k+1)T
x(k+1)T=e ATx(kT)+ / eA(k+1)T−τ
+ dτBu(kT)
, (k+1)T - τ = ξ
+ , kT
T
x(k + 1)T = eATx(kT) + / eAξ dξBu(k)
+ , 0

If by convention we assume that the time interval of width T is

if it coincides with the unit step of the discrete-time system being considered, here it is
that the expression represents a discrete-time system.

x(k + 1) = AD x(k) + BD u(k)

.
y(k) = Cx(k)
44 3. Representations with the state of finite-dimensional linear stationary systems

ove
T
AD=eAT BD= / eAξ dξB CD=C
0
Some considerations on the calculation of the discretized system starting from the
representation with the state. In practical cases it is common to calculate AD , BD
by numerical means starting from A, BeTed using the expression of the
series development dieAT .

T2 2 Tn n
eATI AT A +. . .+ A+
2 n!
It is clear that stopping the calculation at an insufficient point
the mistake that is made is small.
A systematic procedure for performing the approximation is the one here.
as indicated below. It is defined
TA TA TA TA TA T2 A2
ψn=I+ 0 I+ ... I+ 1
. . . =I+ + +. . .
2 3 ! n−1 ! n "" 2 3!
T∥A∥
in cuin`e s c e lt o in m o d o c h e s ia s o d d is fa t t a la s e g u e n t e d is e g u an g lianza
<
It is easy to verify that:
TA T2 A2
AD=I+TA(I+ + +. . .)
2 3!
and therefore with good approximation
ADI+TA·ψ
=
∼ n

and, in the same way

BDψ·B
=
∼ n

An approximation that is often found in applications is the classical one.

so-called Euler method that corresponds to approximating the derivative with its
incremental report; this approximation, valid for very small T, gives:

A
=ẋ+
B
xu

↓
x(k+ 1) = x(k) + TAx(k) + TBu(k)
 3.6. Discrete time systems 45

3.6.c. Some simple examples of discrete-time systems

Some simple examples below highlight that normally it is the

implicit representation the starting point in the study of phenomena. In-
beyond the modeling process naturally leads to introducing the variables
of state.
Dynamics of a population.
A model designed to describe the phenomenon over time can be obtained
assumendo una suddivisione in classi di eta`. Per descriverne la dinamica
we will support the width of the unit time interval equal to the width
of the age class. This set of variables constitutes for the problem in
It is a complete set of state variables.
Indicato conxI (t) the number of people turning age i+1 in the year
t, if there is:
xi+1(t+ 1) = βi xI (t), i = 0, . . . , n−1

x0 (t + 1) = α0 x0 (t) +...+αn−1 xn−1 (t)

withIfertility coefficients eβIsurvival coefficients derived from the

statistics.
Dynamics of Gross National Product.
Let P(t) be the national product, C(t) the consumption, I(t) the investments and
G(t) government expenses. It holds:

P(t) = C(t) + I(t) + G(t)

Indicating with the marginal propensity to consume, we have:

C(t) = mP(t), 0 < m < 1

It is also common to assume that the increase in output is proportional to the investment.
secondor, growth factor

P(t+ 1)−P(t) =rI(t)

46 3. Representations with the state of finite-dimensional linear stationary systems

In conclusion, with obvious substitutions, we obtain:

P(t+ 1) = 1 + r(1−m)P(t)−rG(t)
* -
A simple model of microeconomics
The price dynamics under equilibrium conditions between demand and supply.
In the case that there is only one good indicating with unit price, condla
demand and supply, it is usual to assume that

d(p) = d0 −ap
s(p) = s0 +bp, a, b > 0

Elementary considerations suggest that supply and demand vary.

over time according to the two equations

s(t+ 1) = s0+bp(t)

d(t + 1) = d0-ap(t+ 1)
Taking the equilibrium means assuming that s(t + 1) = d(t + 1) and therefore

b d0−s0
p(t + 1) = -p(t) + .
a a
There exists an equilibrium for the price dynamics given by

p(t + 1) = p(t) = d0 -0 .
a+b
The simple model introduced can be used to predict the trend
of the price in conditions of disturbance compared to equilibrium. How easy it is
it will check if there will be a return to the state of equilibrium (stability) if < a,
a removal (instability) seb > a. It is interesting to observe that the two
The coefficients a and b represent, respectively, the propensity to consume.
(of the buyer) and the propensity (ambition) of profit of the entrepreneur.
The economic interpretation of the condition of stability of the equilibrium before
espressa leads to the following conclusion: excessive ambition for profit
the entrepreneur creates market instability.
 3.6. Discrete-time systems 47

3.6.d. Linear approximations of nonlinear systems

As for continuous-time systems, a linear representation can be

It will be used to roughly describe the behavior around.
to a balanced couple. Given the system

x(t + 1) = f(x(t), u(t)) f ( x e ue ) =xe

y(t) = h(x(t), u(t)) h(xe , ue ) =he .

By developing the calculations, one obtains:

∂f∂f x−xe
f(x, u) = f(xe , ue ) + )
! partial" derivative
! " +...
)xe ,ue u−u
of ex with respect to u
)
)
∂f ∂f
=f(xe , ue ) + (x−xe ) + (u−ue ) +...
∂x ))xe ,ue ∂u ))xe ,ue
) )
) )
∂h∂h x−xe
h(x, u) = h(xe , ue ) + ! ∂x∂u " ) ! " +...
)xe ,ue u−ue
)
)
and postoz=x−xe ,v=u−u e ,ya =y−h(x e )

∂f ∂f ∂h ∂h
A= B= C= x ,u D=
∂x ))xe ,ue ∂u ))xe ,ue ∂x | e e ∂u ))xe ,ue
) ) )
) ) )
yes, infinite:

z(t+ 1) = Az(t) + Bv(t)

ya=Cx(t) + Du(t)
48 3. Representations with the state of finite-dimensional stationary linear systems

3.6.e. Equivalent representations

Effettuando una trasformazione di coordinatez=Tx, si ottiene:z(t+

1) =Tx(t+ 1) =TAx + TBue since x=T−1 yes, it has indeed

z(t + 1) = TAT−1 z(t) + TBu(t)

y(t) = CT −1 z(t) + Du(t)
The same considerations can also be repeated regarding the possibility
to describe linear systems through non-linear representations.
4. The Linear Representations
Stationary in the domain of
time

This chapter presents the main aspects of the analysis of

linear stationary systems in the time domain. The calculation of evolutions
in the state and in the exit highlight the role played by the exponential of ma-
three times recalled in the appendix. On this basis, the analysis of the structure proceeds.
of the evolutions. This is concretized in the so-called analysis of natural ways.

4.1. Continuous-time systems: representations

implicit and explicit
It will be taken as a starting point the representation with the state.

A
=
+
Bt)(xutẋ() =x0∈Rn (3.1.a)
0x
y(t) = Cx(t) + Du(t) (3.1.b)
The implicit model introduced immediately suggests a scheme.
physical means through which to realize the system using devices that perform the
50 4. The Linear Time-Invariant Representations in the Time Domain

sum of time functions, multiplication by constants, and the integral. The

The schema, shown in Figure 3.1, includes blocks A, B, C, D, consisting of
multipliers for constants, integration operators and summators, puts in
the possible structure of the input-output connection shines; there is a dynamic part,
of an integrative type, which is typical of the purely dynamic part and a part
snapshot.

D
.
u (t ) x(t ) x(t ) +
B + +
! C + and
( )t

A
Figure 4.1

We will now verify the evolutions over time dix(t) ey(t), associated with
a fixed initial state0at the time0and a defined entrance date0from now on,
are described by a model, explicit model, which has the structure
following

t
x(t) = Φ(t−t0 )x(t0 ) + ! H(t−τ)u(τ)dτ(3.2.a)
t0

t
y(t) = Ψ(t−t 0 )x(t0 ) + ! W(t−τ)u(τ)dτ(3.2.b)
t0

where the matrices (Φ, H, Ψ, and W) take on the calculated expressions below.
To this end, let's define the following matrix function,

inf
t Ak t k
I at A2+...= =eAt(3.3)
2 "0 k!

By differentiating the series term by term, it is easy to verify that one obtains
 4.1. Continuous-time systems: implicit and explicit representations 51

d At t2
e =A(I+At+A 2+· · ·) =AeAt
dt 2
And it follows immediately that

x(t) = eAtx0

e s o lu z io n e d i

˙ = Ax(t)
x(t) x(0) = x0

This solution is, as is known, unique and globally defined in the linear case.
studio. It is therefore proven that Φ in (3.2.a) takes the expression

Φ(t) = eAt

Place, moreover,

H(t) = Φ(t)B = e AtB

Ψ(t) = CΦ(t) = CeAt

W ( t ) = C Φ ( t ) B + D δ ( t ) = C e AtB+Dδ(t)

The immediate returns (3.2) represent the solutions of (3.1). In-

facts, deriving the (3.2), for the theorem of differentiation under the sign of integral,
it is obtained

t
˙ = AΦ(t−t0 )x(t0 ) +
x(t) AH(t−τ)u(τ)dτ+H(0)u(t)
!
t0

Highlighting on the left between the first two addends and recalling that, in
Based on the definition of H(t), B := H(0), it follows that

˙ = Ax(t) + Bu(t)
x(t)

The same calculations can be repeated for (3.2.b). It results in

52 4. Linear Time-Invariant Representations in the Time Domain

t
y(t) = Ce A(t−t ) x(t
0
0) + ! (ItAtB+Dδ(t−τ))u(τ)dτ
t0

t t
=C(eA(t−t) x(t00 ) + ! eAtBu(τ)dτ) + ! Dδ(t−τ)u(τ)dτ=Cx(t) +Du(t)
t0 t0

where the property of the unit impulse δ has been taken into account: the integral of
a function multiplied by δ, within the interval of integration, is even
to the value of the function at the moment when the impulse is centered (zero value
of the topic).

The previous expressions highlight some important facts:

the responses in the state and outgoing corresponding to an assigned state in-
izialex= 0 and an ingressou(.), are the sum of two evolutions: the evolution
free, which corresponds to the initial state0 at the entrance (u(.) = 0), and the evolution
forced, which corresponds to the input datou(.) and initial state nullox0 ;
HeW, matrices of impulse responses in state and output, respect-
Again, they have columns that can be interpreted as forced responses.
to impulsive entries;
For forced responses, the principle of superposition of effects applies:
the forced response to a fixed linear combination of inputs is equal to
same linear combination of the responses to the individual inputs;
- the forced behavior of entry and exit is described by W, response
impulsive (matrix of impulsive responses); for this reason it is said that the
We have a model of forced behavior that enters us into a situation;
In the terminology of System Theory, the model (3.1) is referred to as im-
explicit or differential, the corresponding model (3.2) is called explicit;
In the calculation of the explicit model, a central role is played by the exponential.
of the dynamic matrix.
The transition from the implicit model to the explicit model, addressed in this
chapter, corresponds to the calculation of the evolutions in the state and output.
 4.1. Continuous-time systems: implicit and explicit representations 53

It is advisable to note from now on that the reverse process, which in practice-
ICA is presented as the reconstruction of a representation with the state.
(A, B, C, D) starting from a model of forced behavior, it is also,
a significant problem for applications. In fact, as already observed, to a
representation with the state corresponds to the existence of a realization scheme
simulation. The reconstruction of an implicit representation in
starting from a model of forced behavior, it is the problem of the
realization in that, once the differential representation is obtained,
the referenced schema allows for both a physical implementation using the components in-
indicate both the implementation through a computing system for simulation
of the system.

4.1.a. The transition matrix in the state

In the calculation of the matrix exponential function, it is helpful to refer to

linear coordinate transformations in state space. As stated in
evidence in the previous chapter, if with respect to fixed coordinates the system
sentadinw
oecrhetni C
,)BA
,(,xim
ratbydenetseperrsi
z=Txil system represented by the new set of matrices (TAT−1 ,TB,
eCT−1 ).
The coordinate transformations with respect to which the dynamic matrix A is...
simple forms play an important role in characterizing the
matrix function under study. In fact, if we denote with D=TAT−1
the representation of the coordinates z=Tx, and we can calculate
the exponential of the dynamic matrix D, then we can deduce the expression
of the exponential of A. It is enough to note that

A=T−1 DT
and therefore that

−1
DTt t2 tk
eA t=eT =T−1 T+tT−1 DT+ (T−1 DT) 2+. ..+ (T −1 DT) k+. . .=
2 k!
54 4. The Stationary Linear Representations in the time domain

tk
=T−1 T+tT −1 DT...+T−k D k T k= . . .=T−1 eDt T
k!

In conclusion, the calculation of the exponential of A is obtained starting from that.

by pre-multiplying by T −1and post-multiplying by T.
It is therefore important to know how to characterize the 'simplest forms' that
an operator takes on varying coordinates along with transformations
correspondents. Referring to the appendix for the details of the calculation, it is sufficient-
it is important to remember that the interesting forms for this purpose, starting from which
It is easy to calculate the exponential; they are that diagonal, the case of matrix A.
regular, up to the most general case of Jordan form.
We will first examine the case of regularity, which corresponds to
property that all eigenvalues have unit geometric multiplicity
This case includes the case of Acon distinct eigenvalues), to arrive
in the general case where the simplest form is that of Jordan (see)
the appendix).
The two situations mentioned are dealt with separately below.
which correspond to different levels of complexity.
The transition matrix with regular A
If we assume the dynamic matrix A to be regular and remember that from the point
From the perspective of calculation, distinct possible eigenvalues can be imagined that coincide.
teeth, the exponential of A can be calculated from the expression (A.3.1)
in appendix. Since the exponential of a block diagonal matrix is even
to the block diagonal matrix of the exponentials; for the calculation it is sufficient
observe that

α ω
# $t cos(ωt) sin(ωt)
e −α ω =eαt # $
-senωtcosωt

In the presence of real eigenvalues and pairs of conjugate complex numbers, we have
so pearsDtthe expression
 4.1. Continuous-time systems: implicit and explicit representations 55

eλ 1t
⎛ .. ⎞
.
⎜ e λ m 1t ⎟
⎜ ⎟
⎜ eα 1t cosine of omega1 t eα tsinω1 t
1
⎟
⎜ ⎟
⎜ 1
−eα tsinω1 t eα t cosω1 t
1
⎟
⎜ .. ⎟
⎜ . ⎟
⎜ ⎟
⎜ ⎟
⎜ eα mt cosine
2 of omegam2t eαmtsinω
2
m 2t ⎟
⎜ ⎟
⎝ ⎠
m m 2
t 2
−eα sinωmt eα t2cosine of omegam2t
and developing the calculations

m1 m2
eAt = eλ it ui v′i+ αt
j
′ ′ bj av ′ -ubv ′ )
"j=1 e cosω
+ j t(uav +ubj v a)j + sinωj j t(u bj aj
"i=1
j j
,

An example
Consider the simple mechanical system composed of mass-spring and
damping addressed in paragraph (2.2). Starting from the dynamic matrix

0 1
A= # k b $
− M − M

the calculation of the eigenvalues provides

√
λ1/2 = -b b2-4kM
2M
and for the corresponding eigenvectors, we obtain

√ √
λ1= -b + b2-4kM -b-b2-4kM
2M 2M →u1= # 2k $

√ √
λ2= −b b2-4kM -b+b2-4kM
2M − 2M →u2= # 2k $

Post
∆=b2-4kM
56 4. The Linear Time-Invariant Representations in the Time Domain

The following three different cases corresponding to parameter values are considered.
for which the value of di∆` is negative, positive, and zero.
If Δ < 0, then it has

√
ua= -b ub= # −∆
# 2k $ 0 $

and with
√
T −1 = # -b-∆ $
2k 0
it is obtained
1 √∆
TAT −1 = -b
√
2M # -∆-b $

and therefore
b cos(ωt) sin(ωt)
eAt=e− t T −12M # $T
-senωtcosωt

−bt
√∆ √∆
=e(cos
2M t(ua v′a+ub vb′) +sen ′ ′
2M 2M t(ua vb−ub va))
If ∆>0, the calculation of the eigenvectors immediately leads to:

√ √∆
T −1 = # -b-Δ-b+ $
2k 2k

1 √∆ 0
TAT −1 = -b+ √∆
2M # 0 -b-
$

and there will be no difficulty in completing the calculation of the exponential.

The case of ∆= 0 is treated in the following because it corresponds to a

irregular matrix.
The transition matrix in the general case *
With reference to the Jordan form of a matrix mentioned in (A.3)
(formula (A.3.2), (A.3.3), (A.3.4)), the exponential takes on a diagonal structure
in blocks where the generic block, of dimension k, has the shape
 4.1. Continuous time systems: implicit and explicit representations 57

tk−1
I i
eΛ tyouΛ t· · · · · · (k−1)! e Λi t
⎛ .. ⎞
.. ..
0 . . .
⎜ .. ··· .. ⎟
eB i t = ⎜ .. .. ⎟
⎜ . ··· . . . ⎟
⎜ .. .. ⎟
⎜ . . youiΛ t ⎟
⎜ ··· ··· ⎟
⎜ 0 0 eΛ it ⎟
⎝ ··· ··· ⎠
oveΛ=λ
i iin the case of real eigenvalue and

αiωi
Λi= # $
−ωiαI
for a pair of conjugate complexes.
For example, consider a four-dimensional matrix with two
pairs of complex eigenvalues coinciding with an even geometric multiplicity
due.
λ1/2=αjω

T −1= (ua1ub1 u2aub2)

u1=ua1+jub1(A−α−jω)u1= 0

u2=ua2+jub2(A−α−jω)u2=u1

⇓
α ω 1 0
⎛ −ω α 0 1⎞
TAT −1 = =ΛCj
0 0 α ω
⎜ ⎟
⎝ 0 0 −ω α⎠
cosωt sinωt tcosωt tsinωt
C ⎛ -sin(ωt)cos(ωt) - tsin(ωt)cos(ωt) ⎞
e Λj t
=eαt
0 0 cosωt senωt
⎜ ⎟
⎝ 0 0 -senωtcosωt ⎠

To conclude, let us consider again the previous example in the case where
-b
2M it has multiplicity two. It can
∆ = 0, we are in the case where the eigenvalue is
in this simple case, verify the existence of the following Jordan form
58 4. The Linear Stationary Representations in the Time Domain

1 0
T −1 = #
−-
k
M 1$

k
−- 1
TAT −1 = ⎛ ⎞
M
k
0
⎝ −- M ⎠
The calculation of the exponential can be completed without difficulty.

4.1.b. The free response in the state

As is known, the free evolution of a stationary linear system takes on

the expression:
xl (t) = eAtx(0)

It is elementary the case in which the matrix Asi reduces to a scalar. We have

A=λ

the free evolution of the state takes place according to the expression

x(t) = eλt x(0)

It is an exponential law that assumes the three qualitative trends

mind reported in the figure

> 0
e t

=0
1

<0

t
Figure 4.2
 4.1. Continuous time systems: implicit and explicit representations 59

From a qualitative point of view, the evolution over time is exponential.

divergent for λ>0, convergent for λ<0, constant and equal to one for λ=0.

To characterize the speed at which the system evolves, both growing and decreasing
crescendo, a parameter, τ, the time constant is introduced

1
τ=−
λ
With respect to this parameter
t
x(t) = e− x(0)
τ

and it is clear that it represents, in the case where the state decreases, the time
1
necessary for the state to become of the
e initial value.

4.1.c. the natural ways

The previous simple considerations made in case 1 can be

generalize by employing the calculation tools developed.
In the following section, we will address two cases: the one in which the
distinct eigenvalues of Asian and the one in which Asia is regular, i.e., admits the
diagonal form. The general case, which includes the presence of eigenvalues
a geometric multiplicity greater than one is handled separately in the
continued.
Natural modes in the presence of distinct eigenvalues

In the simplest case, where the size of the system is the matrix A.
it has real and distinct eigenvalues, as seen in the previous paragraph, we have

eλ 1t v′1 n
⎛ .. ⎞ ⎛ .. ⎞ =
eAt=T−1 eΛt T= (u1... un) . . eλ it ui v′i
"
⎠ ⎜⎝ v′n ⎟⎠
λn t
⎝ ee i=1

where to meet′i Ii′the

rows of the matrix and the columns of T are indicated.−1 ,
respectively, and where the terms of the sum are exponential, that is laws
60 4. The Linear Stationary Representations in the time domain

storms of the type we encountered in the case = 1. This expression

it allows the calculation of free evolution; in fact
n
xl (t) = eAtx(0) = eλ it ui v′ix(0)
"i=1

Note that for pert= 0 we obtain

n
x(0) = ci ui
"i=1

where is it locatedi=′ix(0).x(0) results, therefore, equal to the sum of

components in the space constituted by the eigenvectors1 , u2 , . . . u n and the coefficients
c1 , c2 , . . . , c nrepresenting the amplitudes of such components.
Ultimately, for the free evolution in the state, one obtains the expression
n
xl (t) = eAtx(0) = ci eλ ti uI
"i=1

what is equal to the sum of distinct terms, corresponding to evolutions along

the eigenvectors1 , u2 , . . . , u n The trajectory in the state space admits
projections, in the directions1 , u2 , . . . , u n , of the type

ci eλ ti

The motion law of each term is still an exponential motion law;

its width,ci eλ t it dependsi on the initial state's component in the
corresponding direction,ui .
Note that each of these individual terms could be, as it
It is enough to take an initial condition, excited individually.
which has a projection only on the eigenvectorito have an expression with the
I just finishedλ t The
1
same considerations made in the case of n = 1 can
being repeated in the general case if an initial state is assumed in the direction
of the generic eigenvectori If the corresponding eigenvalue is 0, one remains
In the initial state, if the eigenvalue is positive, the evolution is divergent.
 4.1. Continuous-time systems: implicit and explicit representations 61

subtracting in the direction of the eigenvector; if instead λinegative, it is a mutation

decreasing towards the origin. Therefore, the trajectory related to natural motion is
it maintains along the eigenvector and has these three types of behavior. In figure
possible trends are indicated.

> 0 i 0 = < 0
i x0 x0 x0 i

u ui ui
i

Figure 4.3

The previous considerations can be summarized by stating that

the free evolution in the state in the case of real and distinct eigenvalues is
linear combination, according to coefficients that depend on the initial state, of
independent revolutions that take place in one-dimensional subspaces
generated by the eigenvectors. These evolutions in the direction of the eigenvalue-
they are natural aperiodic dett modes. The temporal laws of evolution
the modes depend only on the eigenvalues λ1 ,λ2 , . . . ,λ nand we will call them distinct.
Thus, in the case of distinct real eigenvalues, there are distinct modes.
since both the law of motion and the direction of the space in which it takes place are different
the evolution.
Proceeding in a manner similar to that done in the scaling case, one can define
a time constant for each eigenvalue. Placed

1
τi=−
λi

one obtains the expression

t
ui
x(t) = ci e− τi

"
in which the time constants play the previously emphasized role.
62 4. The Stationary Linear Representations in the Time Domain

How much is the value in the case of real eigenvalues. Suppose now that there are
also complex eigenvalues, real and complex conjugates, conn=µ+2ν,

λ1 , . . . ,λ µ ,(α1+jω1 ), . . . ,(αν +jων )

indicating withkaIkbthe real and imaginary parts of the eigenvectork if

obtains

µ ν cosine of omegak tsinωk t v′ka

eAt = eλ it ui v′i + eα kt (u ka ukb )
⎛ ⎞⎛ ⎞
"i=1 " v′kb
k=1
⎝ -sinωk tcosωk t ⎠⎝ ⎠

in which convkaevkbthe rows are indicated, in the inverse of the matrix of

eigenvectors, corresponding to the columnsgoIkb .
The expression of free evolution is now obtained by multiplying by x.0 ;
it is obtained

µ ν
xl (t) = ci eλ ti ui+ mk eα tksin(ω)
# k t+ϕk )uka+ cos (ωkt+ϕk )ukb $
"i=1 "k=1

having placed
cka
cka=v′ka x0 sinφk=
2 +c2 mk
mk=cka ckb
- kb ckb=v′kb x0 cos φk =
mk

It has been shown that in the case of distinct, real, and paired eigenvalues.
of complex conjugates,

free evolution results in the linear combination of natural modes

evolutionary changes that occur in the plans generated byaIbthey take
the name of natural pseudoperiodic modes. The temporal evolution am-
put an exponential envelope, andαt , but with periodic components sine and
cosine, along two directions of the plane of motion.
The figure shows the possible trajectories of the pseudoperiodic motion.
odico
 4.1. Continuous-time systems: implicit and explicit representations 63

k > 0 k
< 0 k= 0

x0 x0
uk b x0

uk a

Figure 4.4

Perαk >0 if it has a divergent spiral trajectory, for αk= 0 an ellipse

and forαka converging spiral as illustrated in the figure.
The temporal laws of the pseudoperiodic mode are shown in the figure in
diverse situations.

k0 k= 0 k
< 0

t t t

Figure 4.5

The pulse of the temporal law is ω.k the imaginary coefficient of the eigenvalue
complex.
In conclusion, in the general case of distinct eigenvalues, the free evolution
in the state it turns out to be a linear combination of natural periodic modes,
associated with real eigenvalues, natural pseudoperiodic modes, associated
all pairs of complex eigenvalues. We will say in this case that they have modes.
64 4. The Linear Time-Invariant Representations in the Time Domain

simple natural laws of distinct motion.

The knowledge of the location of the eigenvalues in the complex plane allows
to understand the qualitative trends of the system's developments.
It is important to observe the placement of the eigenvalues with respect to the axis.
imaginary, characterizes the convergence or divergence of the ways. The axis
imaginary and in a certain sense the frontier regarding such composition
Statement: if the eigenvalues are on the left, then the motion converges to zero.
either periodically or aperiodically; if they are to the right, the motion diverges;
If I am on the imaginary axis, the motion is constant or oscillatory.

Even in the case of complex eigenvalues, it is common to introduce two parameters.

that characterize the manner in which αeω takes place. It is about the natural pulsation-
rale,ωn ,
ωn=α2+ω2
.
and damping, ξ,

ξ=sinθ= −α
√ α2+ω2

The figure highlights such parameters.

sin =
ωn

Figure 4.6

The physical meaning of natural frequency and damping can be...

taken from the expression of αeω as a function of ωn eξ

α=−ξω n e ω=ωn1 - ξ2
.
 4.1. Continuous-time systems: implicit and explicit representations 65

If α = 0, that is, if the conjugate complex values are imaginary, ξ = 0 and

ωn =ω; this means ωnand the pulsation that would have the system if there were not the part
real (natural pulsation). Upon appearance of α different from zero, the pulsation
move away from this 'ideal' pulsation, corresponding to zero damping.
it is called damping factor, because an increase in it corresponds to
a greater attenuation of the envelope of the oscillatory function.

Natural modes in the presence of non-distinct eigenvalues and

lare

In the event that the dynamic matrix is regular, that is, the eigenvalues,
currently not distinct, they have geometric multiplicity, mi , unitary, and possi-
calculate corresponding to each real eigenvalue (pair of eigenvalues
conjugate complexes) as many eigenvectors (eigenspaces of dimension two) as
and the multiplicity of the algebraic eigenvalue of this self-value (of the pair of eigenvalues).
In this situation, the matrix A is still a diagonalizable matrix, as
it can be easily verified using the same arguments as in the previous case. For
natural modes correspond to expressions analogous to the previous ones being able to
coinciding the laws of motion of different natural modes. The natural mode is, in
this case, characterized by the associated self-space.
In conclusion, if the dynamic matrix A has coincident eigenvalues, but
regularly correspond to each eigenvalue with algebraic multiplicity
greater than one, equally simple natural ways, characterized by
from different self-spaces, but with coinciding temporal laws.

4.1.d. Natural modes: the general case *

We will now examine the case where there are eigenvalues with geometric multiplicity.
greater than one. For this purpose, the procedure is recalled below,
selection of a basis with respect to which a given linear operator takes
the Jordan form.
As is known, such a basis corresponds to a decomposition of [Link]
disjoint subspaces (autospaces) each associated with an eigenvalue and of
dimension equal to its algebraic multiplicity (of dimension equal to double
66 4. The Linear Stationary Representations in the Time Domain

of the algebraic multiplicity for a pair of complex eigenvalues). If in-

we say with Uithe generic eigenspace associated with the eigenvalue λi , it results

U0⊕. . .⊕U m =Rn

Secithey are the dix components0 with respect to such subspaces:
µ
x0 = ci
"1

ciit turns out to be a generalized eigenvector of a certain order, siak, lesser

or equal to the geometric multiplicity of the eigenvalue λi .
k:= order with whichiand the car vector regenerator, i.e.

(A−λi I) i cI̸= 0 i= 0, . . . , k−1

(A−λi I) k ci= 0

onixtalerm
It is worth remembering that each subspace ofrf tcerdiasU
i tiinad
each associated with a chain of generalized eigenvectors. Furthermore, each
of these last ones, generated by a first-order eigenvector, contains eigenspaces
of dimension one, two,..., up to the maximum order of the generalized eigenvector
because it belongs to that chain.
The two extreme situations that occur in the presence of an eigenvalue
the algebraic multiplicity greater than one is that in which the multiplicity
geometric and unitary, that is, what corresponds to the existence of many eigenspaces of
dimension one in a number equal to the algebraic multiplicity, and it is the case
Already treated previously (Aregular), or the geometric multiplicity
it corresponds to the existence of a single chain
of generalized eigenvectors and the existence of a sequence of eigenspaces.
eralizzati of increasing size up to the algebraic multiplicity. In cases
intermediates have more chains of generalized eigenvectors, each defines
a sequence of autospaces of increasing size. The number of chains,
as is known, it is defined as the index, I, associated with the fixed eigenvalue
or, defines the number of Jordan blocks that make up the most
simple of the operator.
 4.2. Discrete-time systems: transition to the explicit model 67

The free evolution is the sum of evolutions associated with it.i Each
It defines a natural multiple way of order, according to the order.
with whichiand the self-regenerating autovettore. The natural mode multiplies.
the expression

tk−1
eAtci=eAt−λ t eλ t ci i=eλi t t(A−λ#i I)ei λ t +. . .+ i
(A−λi I) k−1 eλ tci i $
(k−1)

Note that in this case the temporal law of the mode is of an exponential type
a polynomial coefficient in the time variable.
Each eigenvalue is associated with multiple natural modes, one for each
with autoplace. Therefore, there is an even number of natural ways equal to the index.
The order of the natural mode is assumed to be equal to the dimension of the eigenspace and
it corresponds to the order of the polynomial that characterizes the temporal law.
There will therefore be a set of natural ways in ascending order up to the
algebraic multiplicity. The number of ways, as mentioned, is equal to the index.
(to the number of Jordan blocks) and the size is that of the corresponding
Jordan block. In this case, we will say that we are in the presence of nat-
multiple distinct urals.
In the two limiting cases of geometric multiplicity equal to algebraic multiplicity
or equal to one we will say that there is only one natural multiple more
number equal to the algebraic multiplicity) natural multiple ways with laws
coinciding temporals.

4.2. Discrete-time systems: transition to the model

explicit

We will now verify that the evolutions over time dix(t) ey(t) associated with
a fixed initial state0 at the time0 and a defined entrance date0in
then, they are described by the following explicit model

x(t) = Φ(t−t0 )x(t0 ) +Σ(t−1 H(t−τ)u(τ)

t0 (9.2.a)
68 4. The Stationary Linear Representations in the Time Domain

y(t) = Ψ(t−t 0 x(t)0 ) +Σt W ( tt0− τ ) u ( τ ) ( 9 . 2 . b )

If we denote with Atthe power of the matrix A is easy to verify, for

successive substitutions, which the form that the solution takes is of the indicated type
with

Φ(t) = At

H(t) = Φ(t−1)B = A (t−1) B

Ψ(t) = CΦ(t) = CAt

W ( t ) = C Φ ( t − 1 ) B = C A (t−1)B t >0W(0) =D

Even in this case, the observations made for the systems can be repeated.
at a continuous tempo:

the responses in the state and outgoing result are the sum of two
evolutions: free evolution and forced evolution; forced evolution has
the form of a convolution summation;
HeW, impulse response matrices in the state and output, respect-
again, they have columns that can be interpreted as forced responses
at impulsive entrances;
For forced responses, the principle of superposition of effects applies;
The forced behavior of entry and exit is described by W, response
impulsive (matrix of impulsive responses), a model of behavior
forcibly
In the terminology of Systems Theory, the model (1.a) - (1.b) is
implicit or differential said, the corresponding model (2.a) - (2.b) is said
explicit;
In the calculation of the explicit model, power plays a central role.
of the dynamic matrix.
 4.2. Discrete-time systems: transition to the explicit model 69

The construction of a representation with the state (A, B, C, D) at par-

tire from a forced behavior model, it's an interesting problem
from a designer's point of view: this problem, known as prob-
the realization lemma allows us to identify the recursive algorithm that
corresponds to the calculation of the convolution sum. Therefore, it corresponds to
a real-time implementation of the forced input-output bond.

4.2.a. The transition matrix

The structure of the transition matrix, At plays a central role in

characterize the behavior of a discrete-time system. In fact, a
starting from a representation, linear, stationary, of finite dimension, the
calculation of evolutions consists of calculating
t−1
x(t) = A(t−t) x0+
0
A(t−τ −1)This(τ)
"
τ=t0
t−1
y(t) = CA (t−t) x00 + CA (t−τ −1)Bu(τ) + Du(t)
"
t=t0

It is also helpful to refer to the calculation of the power of a matrix.

linear coordinate transformations and remember that the matrices that describe
the system transforms into TAT−1 ,TB, eCT−1 . IndeedTAT−1there is a
structure with respect to which the calculation is simpler, and this for example
the case of the canonical form in appendix 3,D, is obtained

A=T−1 DT ⇒A=
t (T−1 DT) t. . .=T−1 D t T.

Being block diagonal, the power is equal to the diagonal of the powers.
of the individual blocks. There are no issues with the power of the diagonal part
resulting equal to the diagonal of the powers of the elements on the diagonal.
It is only necessary to understand how to calculate the power of the block (2×2)
associated with a pair of complex conjugate eigenvalues, given

λ=α+jω=σe jθ =σ(cosθ+jsenθ)
70 4. The Stationary Linear Representations in the Time Domain

⇒α=σcosθ ω=σsinθ
it results
α ω cosθsenθ
# $ =σ # $
−ω α -sin(θ)cos(θ)
it's not difficult to calculate
t
cosθsenθ cosθt sinθt
# $ = # $
-sinθcosθ -sinθcosθ
So for the generic block (2×2) associated with a complex pair
He/she is tired.
cosθt sinθt
Λt=σt # $
−sin(θ)cos(θ)
It therefore results

λt1
⎛ .. ⎞
.
⎜ λtm1 ⎟
⎜ ⎟
⎜
t ⎜ σt1cosθ1 tσt1sinθ1 t ⎟
D= ⎜ ⎟
⎟
⎜ −σt1sinθ1 tσt1cosθ1 t ⎟
⎜ .. ⎟
⎜ . ⎟
⎜ ⎟
⎜ σtmcosθ t
m2tσmsinθm2t
⎟
⎜ ⎟
⎜ ⎟
2 2

⎝ −σtmsinθ t 2
2 mtσcosθ
m m 2t 2 ⎠
and developing the calculations

m1 m2
At= λitui v′i + t ′ ′ bj av ′ −ubv ′ )
"j=1 σcosθ
j + j t(uav +ubvaj) + sinθj t(u bj aj
"i=1
j j j j
,

The student will not have difficulty revisiting, with reference to the calculation of
power of a matrix, the considerations made earlier in the case of
eigenvalues with geometric multiplicity greater than one.
For the general case, the same considerations already made for the calculation apply.
color of the exponential that in the specific case gives for the generic block, of
dimensions,
 4.2. Discrete-time systems: transition to the explicit model 71

t[k−1]
ΛItt[1]Λi t−1
· · · · · · (k−1)!Λi
t−k+1
⎛ .. ⎞
.. ..
0 . . ··· .
⎜ .. .. ⎟
B=i t ⎜ .. .. ⎟
⎜ . ··· . . . ⎟
⎜ .. .. ⎟
⎜ . . t Λi t−1
[1] ⎟
⎜ ··· ··· ⎟
⎜ 0 ΛI t ⎟
⎝ ··· ··· 0 ⎠
oveΛ=λ
i iin the case of real eigenvalue and

cosθiscentI
ΛI=σi # $
-sinθicosθI

for a pair of conjugate complexes and, furthermore, t[k] denotes the polynomial
factorial of a defined order as

t[k] t(t−1)...(t−k+1)

4.2.b. The natural modes for discrete-time systems

Similar considerations to those made so far can be repeated for

discrete-time systems. The same topics used for continuous-time systems.
discreet lead to
m1 m2
t
xℓ (t) =At x0= λitui ci+
"j=1 |λj |m j sin(θ t+ϕj )u+ cos (θ
ja j t + φj )u jb
"i=1 / j 0

love
λℓ =|λ ℓ |ejθ
ℓ
, 0<thetaℓ <π

and the parameters present are defined as for continuous-time systems.

Even in this case, the free evolution in the state is given by the combi-
linear nation of natural modes that occur along particular directions,
those associated with real eigenvalues, and in planes, those associated with pairs of auto-
complex values. A peculiarity of discrete time systems is a subdi-
view of the modes associated with real eigenvalues; in this case, in fact, there are modes
72 4. The Linear Time-Invariant Representations in the Time Domain

so-called periodic ones corresponding to positive eigenvalues and modes so-called-

the alternative details corresponding to negative eigenvalues. The motion in this case
it happens, think of the simple case of unit multiplicity, along a line
subtracting from the same side of zero, for the aperiodic mode, alternating from
one part is on one side and the other side of zero, for the alternating way. Thus, there are modes
aperiodic (λireal>0), alternating (λireal<0), pseudoperiodic (pairs
complex conjugates.
The figure shows the time trends of the laws of motion.
corresponding to the different placement of the eigenvalues.

Im [ ]

1Re [ ]

Figure 4.7

In the case of eigenvalues with geometric multiplicity greater than one, with
Similar considerations to those made in the continuous time case are calculated:

At ci= (A−λi I+λ I I) t cI=

t
t h t−h
" # h$ (A−λi I) ci λi
h=0
 4.3. Further remarks * 73

t t(t−1)...(t−h+1) t(h)
(# $ ) = :=
h h! h!

Therefore, the laws of motion are of the type power with coefficients that are polynomials.
factorials.

Finally, the same considerations can be repeated regarding excitability.

the observability of the modes and the meaning of the forced model. The interpretation
It is even more evident in this case.

4.2.c. The natural modes under discretization

An elementary observation concerns the placement of the eigenvalues of

discrete-time system in relation to those of the continuous system. One has

D λCI T
λ=e
i

n
C
AD=eAT = eλ iTui v′I
"i=1

in fact
C
eλi T
→λiD

It follows that the eigenvalues of a discrete-time system that comes from

discretization of a continuous-time system cannot take values
arbitrary, for example, they cannot be equal to zero. Furthermore, they can be
negatives only corresponding to a pair of complex eigenvalues. This
behaves, in systems that originate from the discretization of a system
in continuous time, the eventual presence of alternating modes necessarily to
couples.
74 4. Linear Time-Invariant Representations in the Time Domain

4.3. Further observations *

The following considerations aim to provide some clarification

on the complex situation that corresponds to the case in which the dynamic matrix
not be regular.
As has been observed several times, natural ways are the evolutions in
state space through which it is possible to express free evolution;
more precisely they are evolutions that occur in subspaces of space
states (autospaces), are characterized by a temporal evolution associated
to the eigenvalue and the order of the mode, every free evolution and combination
linear of such evolutions in relation to the component of the initial state in the
relative self-space.
In the presence of multiple eigenvalues, algebraic multiplicity greater than
one, if an even number is defined at index, I,i , of sequences of eigenspaces
of increasing dimensions until generating a subspace of dimension equal to the
geometric multiplicity of the eigenvalue. There will be, therefore, for the same
eigenvalue plus eigenvector. Under these conditions, we will say that to the eigenvalue λi
they are associatedinatural modes, each marked by its own
autospacio and a periodic or pseudoperiodic temporal law with a co-
polynomial efficient in degree equal to the dimension of the eigenspace minus
one. The dimension of the eigenspace is defined as the order in a certain way.
takes, therefore, how there can exist natural ways characterized by
same time law. In reality, the case in which the algebraic multiplicity, µi ,`e
greater than one and that geometric,i they are the same as one and a case parts-
to color this situation in which all the temporal geeges of theiassociated modes
at the same eigenvalue, they are the same.

4.4. The natural modes in the state and out

The concept of natural mode now introduced lends itself to carrying out a first
investigation into what was mentioned earlier regarding the representativeness of
model consisting only of forced response (impulse response matrix).
 4.4. Natural modes in the state and outgoing 75

To this end, referring to the presence of the temporary law of

mode in the columns of the matrix of impulsive responses in the state and in the
output transition matrix, it is common to introduce in systems theory the
definition of excitable natural mode and that of observable natural mode.

4.4.a. the irregular case

Let us first consider the case in which the natural ways are distinct; that is,
which corresponds to the coincidence between the algebraic and geometric multiplicities of
an eigenvalue. In this case, in fact, the temporal laws associated with the modes are
distinct.
It is evident that while changing dix0in the free evolution of
the system will be able to show all the temporal laws of natural ways, such as
it is said that free evolution is characterized by all natural ways, for
As for the forced evolution in the state, that is no longer true. In fact,
the forced response is characterized by all and only the 'temporal' laws that
they appear ineAtB.
Such modes are said to be excitability (and it is implied 'with input impulses')
as an impulsive input is applied, it turns out
t
x(t) = ! eA(t−τ ) Bδ(τ)dτ=e AtB
0
that is, the response coincides with the free evolution that moves from the initial state
x0=B, in the case of single-input systems, from an initial state equal
to a fixed linear combination of the columns of B, in the case of systems with
more entries. The mode associated with the eigenvalue λi , and therefore excitable if
influenza the evolution in the state. More precisely, we will say it is excitable if the
your temporal law appears in H(t).
The excitability condition of the generic way naturally in the case of
distinct eigenvalues are
v′iB ≠ 0
In fact, from the expression
n
eAtB= eλ it ui v′iB
"i=1
76 4. The Stationary Linear Representations in the Time Domain

it is understood that the temporary lawλ tcompare

i
a second member if and only if
if the mentioned condition is verified.
Similar considerations can be repeated for the evolving response.
release on exit. This naturally leads to the introduction of the definition of
observable natural mode as that mode whose temporal law appears in
This At(which can therefore be stimulated by varying)0 ) In this case too
an observability condition can be formulated that admits a con-
equivalent expression to the previous one. In fact, if we assume that the eigenvalues
they are distinct, the generic way is natural and observable if and only if

Cu inot equal to 0
what is evident from the expression
n
This At= eλ it Cu i v′i
"i=1

One can only wonder what are the temporal laws that characterize
W(t); and therefore the forced response at the output. It is clear that only the natural modes,
and therefore the relative temporal laws, which are simultaneously excitable and
Observables may appear in the forced response.
It obviously concerns all and only the natural ways for which it results

v′iB̸= 0 Cu i≠ 0
The previous considerations actually apply in the case where the matrix of-
she is regular with the caution of distinguishing between natural ways that have
the same temporal law. In this case, it will not be the presence (inΦ, H,Ψ,
of the law of motion corresponding to the natural way to certify it
property

4.4.b. the general case *

The geometric conditions of excitability and observability presented with

reference to the case of distinct eigenvalues and regular a admits a simple
generalization.
 4.4. The natural modes in the state and out 77

For example, if we think about the condition of excitability (with impulses in

income) and the columns of the matrix B can be seen as vectors of the space of
one realizes that the geometric condition of excitability can be
formulate in terms of a non-zero component of at least one column
the matrix compared to the generic eigenspace associated with the natural mode in
exam.
For example, the generic natural mode associated with the eigenvalue λi , tra
theIIIt is excitable if at least one column of the matrix B is associated with it.
it has a component different from zero in the associated subspace.
This condition can be expressed geometrically by means of
′
Vk iB ≠ 0
′
doveVk and i the matrix (kxn) that characterizes the transformation of coordinates
′
nate,T, he/she/she associates with the auto space of modeUk in the base
i of
autospaces of natural modes. In the particular case of the i-mo natural mode
corresponding to an eigenvalue with unit geometric multiplicity, in which
k = 1, it is expressed as indicated in the previous case′iB̸= 0.
Similarly, regarding the observability of natural modes, the
Generic in a natural way of ordering and observable, its temporary law
compare in the response in free evolution, whether for the eigenvector of order
µi-k + 1 is the condition
Cu inot equal to 0
Generalizing the concepts presented, it can be understood that in the presence
of higher order methods than one, the excitability and the observability may not
regarding maximum order methods. A geometric condition corresponding to
teeth consist in verifying that the intersection between Babbia and V is not emptyi ,
the eigenspace associated with the eigenvalue λi , o cheVIis not contained in KerC.

As for the natural ways that characterize W(t); we can assert

but the generic way of order appears in Wse only if at least one
column of matrix B, conceived as a vector of the state space, in its
expression as a sum of generalized eigenvectors includes for the i-th
autospazio an eigenvector of order, for the eigenvector of order µi-k + 1
let the condition holdinot equal to 0.
78 4. The Stationary Linear Representations in the Time Domain

As an example, consider the system described by

1 1
A= # -1 $ B= # $ C= ( 0 1 )
0 -1 0
There are two natural ways, one of first order and one of second order; that one
first order and temporary law− it can be seen, it is not observable, while
the second order is not excitable, but observable. Therefore, there is no...
simultaneously excitable and observable, which is reflected in the fact that
the impulsive response is identically zero.

e−tte−t 1
(01) # 0 =0
e−t $ # $0

The previous condition, with reference to the generic natural mode, in

a special case of an eigenvalue with unit algebraic multiplicity is expressed
come
v′iB̸= 0 e With inot equal to 0
The forced behavior model, i.e., W(t), is therefore a model
partial.
The problem at hand presents an underlying ambiguity in the case where the modes do not
they are distinct, that is what corresponds to the presence of eigenvalues with multiplicity
arithmetic greater than one and different from the geometric. In this situation
the condition of excitability, observability, cannot be formulated as pre-
without the temporal law in H(t), in Ψ(t), due to an intrinsic impossibility
to distinguish between the different ways that are, in fact, characterized by laws of
distinct motorcycles. Consider a system of size two with matrix A.
diagonal.
Excitability, observability, in this particular case can only be
it is formulated and verified through geometric formalism. In fact, in the case
general ways of natural with coinciding laws of temporal evolution the way
natural is characterized by the associated self-space. So it is the con-
statements of excitability and observability previously stated, which will be
these geometric conditions, the conditions to which one can refer. With
 4.5. Appendix 3 79

referring to the previous example, an output matrix C = (1 0) ensures

the observability of a single mode; C = (1 1) the observability of both modes
natural.

4.5. Appendix 3

[Link] form of a matrix

As is known in characterizing the simple forms of a linear operator

A central role is played by the eigenvalues and eigenvectors.
Recall that for a given matrix A, a number λ and a vector u
I am self-valuing corresponding to car vectors

(A−λI)v = 0

The previous equation has a solution if λ is zero from the characteristic polynomial.
isto, d(λ), associated with the matrix A

µr
d(λ) = det(λI - A) = (s - λ1 )µ . . .(s−λ1 r )

the multiplicity of λilike zero does(λ)`e algebraic multiplicity. The

algebraic multiplicity of the generic eigenvalue (λisara andicated withi ).
Another polynomial that plays an important role in the description of
forms of an operator as the coordinates vary and the minimum polynomial. From
from an algebraic point of view it coincides with the least common multiple of the
the polynomial in the denominator of the inverse of (λI−A) is denoted by m(λ).
It is a factor polynomial of the characteristic polynomial as it
has the same roots as the latter, but each with a multiplicityi , this
geometric multiplicity, which is less than or equal to the algebraic multiplicity (µi ).

mr
m(λ) = (λ−λ1 )m. . .(λ−λ
1
r)
80 4. The Stationary Linear Representations in the time domain

Using coordinate transformations, the simplest structure is that one

diagonal with possibly complex coefficients and, correspondingly, that
quasi diagonal with real coefficients

Almost diagonal form with real coefficients, D, (diagonal for the eigenvalues
real and block diagonal of size two for the pairs of co-complexes
A transformation to put this into canonical form exists.
the matrix is said to be regular if all the eigenvalues
they have unitary geometric multiplicity. How could one show the con-
the definition of unitary geometric multiplicity implies that in correspondence to
each eigenvalue λiof algebraic multiplicityi It is possible to calculate proprioµI
first order auto-vectors, that is, independent vectors solutions of

(A−λi I)u I= 0

A special case of a regular matrix is when its eigenvalues are the

matrices have unit algebraic multiplicity.

The transformation we are looking for is the one that corresponds to as-
to take as a new basis of the space that associated with the eigenvectors of the first
[Link] that, as can be demonstrated, are independent of each other.

Let us assume, therefore, to regular and imagine, without loss of generality

to distinctly indicate the eigenvalues, called λ1 , . . . ,λ m,α1ȷ ω1 ,. 1. .,
 4.5. Appendix 3 81

αm 2 ȷ ωm2, conm1+ 2m2and if it has

v′1
⎛ .. ⎞
.
⎜ vm′ 1 ⎟⎟
⎜
D=TAT−1 = ⎜⎜ v′1a ⎟⎟ ·A·(u 1... umu1au1bone
. . . u m a um 2 2 )
⎜ v′1b ⎟⎟
b
⎜
⎜ .. ′ ⎟
⎜ .vm2 a ⎟⎟
⎜
⎜ ′
vm ⎟
⎝ 2b ⎠

λ1
⎛ ..
. ⎞
⎜ λ m1 ⎟
⎜ ⎟
⎜ α1ω1 ⎟
= ⎜ ⎟
⎜ −ω1α1 ⎟
⎜ ⎟
⎜ .. ⎟
⎜ . ⎟
⎜ ⎟
⎜ α m2 ωm2 ⎟
⎜ ⎟
⎝ −ωmαm 2 2 ⎠

where do you go′I Ii′the rows of the matrix and the columns of T are indicated−1 ,
respectively.
The previous representation constitutes the almost diagonal form at co-
real eigenvalues of a matrix. The verification is soon done starting from the following
equality obtained from the previous by multiplying both members by
T −1
A(u1. . . u mu1au1b1 . . . u m a um b ) = 2 2

= (u1. . . u mu1au1b1 . . . u m a um b )· 2 2

λ1
⎛ ..
. ⎞
⎜ λ m1 ⎟
⎜ ⎟
⎜ α1ω1 ⎟
· ⎜⎜ ⎟
⎟
⎜ −ω1α1 ⎟
⎜ .. ⎟
⎜ . ⎟
⎜ ⎟
⎜ α m2 ωm2 ⎟
⎜ ⎟
⎝ −ωmα2m 2 ⎠
82 4. The Linear Time-Invariant Representations in the Time Domain

The equality among the primaries1columns of products in the first and second member
the assured fax to c h e λiIithey are eigenvalue and corresponding eigenvector
e, therefore, (A−λi I)u i= 0 that is equivalent to writing Aui=λi ui(beingAuIthe
i-ma column of the product to the first member eλi uilaima column of the product
a second member). The equality between the remaining columns is ensured by
property on the real and imaginary part to which the eigenvectors satisfy
complex. If, in fact, λ = (α + jω) is the eigenvalue associated with the eigenvector
u=ua+jub , yes it has

(A - (α + jω)I)(ua+jub ) = 0

what is equivalent to the two equalities between real and imaginary part

(A−αI)ua=−ωub (A−αI)ub=ωua

(that expresses also that (A−λ∗ I)u ∗ (A - (α - jω)I)(ua-jub ) =

0). The previous equalities are then equivalent to the following equality
matrix
α ω
A(ua ub ) = (ua ub ) # $
−ω α
This expression allows to conclude the verification under consideration.

4.5.b. The Jordan form *

In the general case of a non-regular matrix, the presence of eigenvalues with multiplicity
geometric greater than one, the simplest canonical form it takes on
the operator is the so-called Jordan blocks, J.
In that case, by appropriately choosing the new coordinates in the au-
subspaces associated with eigenvalues, eigenspaces that can be further decomposed into sub-
associated spaces to "chains of generalized eigenvectors," a form is obtained
of the type:
J1 0
J= ⎛ .. ⎞
.
⎝ 0 Jm ⎠
 4.5. Appendix 3 83

they are the eigenvalues.

The dimension of Jithe rare multitude applied algebraic λi , for a car-
real value, double the algebraic multiplicity for a pair of eigenvalues
conjugated complexes.
B1 0
Jℓ = ⎛ .. ⎞
.
⎝ 0 Bℓ i ⎠
and the block of Jordan associated with the self-evaluation.
λI 1
⎛ .. .. ⎞
. .
Bi= ⎜ .. ⎟
⎜ . 1⎟
⎜ .. ⎟
⎜ . λi ⎟⎠
⎝
for a real eigenvalue,

ΛI I
⎛ .. .. ⎞
. . αiωi
Bi= ⎜ .. ⎟ , Λi= # $
⎜ . I ⎟ −ωiαi
⎜ .. ⎟
⎜ . ΛI ⎟⎠
⎝
for a pair of complex conjugate eigenvalues. The maximum dimension
the blocksBiit coincides with the maximum order of the generalized eigenvectors
associated with the eigenvalue λi .
It is remembered that a generalized eigenvector of order k satisfies the equations

(A−λi I) i uinot equal to 0 i = 0, ... , (k - 1) (A−λi I) k ui= 0

Furthermore, the higher-order eigenvectorsijcan be calculated at

start from those of lower orderi(j−1) solving equations of the type

j (j−1)
(A−λi I) i u=u
i i

and building what are technically called chains of eigenvectors.

84 4. The Stationary Linear Representations in the Time Domain

Infinte recalls that a calculation procedure based on the calculation of all

the first-order eigenvectors, then those of second order starting from those of
order one, and so on, until calculating a number equal to the multiplicity
algebraic, allows to find the base for the Jordan form. This results in
composed of the aggregate of the bases of the eigenspaces associated with each eigenvalue
or, in turn, made up of the aggregate of the chains of associated eigenvectors
to each eigenvector of order one.
The verification that in the new coordinates, defined as the aggregate of
chains of eigenvectors, the representation assumed by the operator is that in-
dictation can be done as indicated in the regular case.
5. The linear representations
stationary in the complex domain

This chapter introduces the analysis method in the domain of

complex variable that employs the Laplace transform. The study of
stationary linear dynamic systems in the transform domain allows
to link the impulsive response to the system's behavior in response
periodic solicitations. This is a characterization of the behavior
dynamic treatment on which the classification used in the sector is based
of engineering.

5.1. The Laplace transform in the study of

continuous time systems

Recall the definition of the Laplace transform; it associates to a

function f(t), defined for t ∈ [0,∞), the complex variable function F(s)

L[f(t)] =F(s) := ! e−stf ( t ) d t

0
 5.1. The Laplace transform in the study of continuous-time systems 87

which is defined for Re[s] > α,f being αf a real number associated
adf(t) is called the convergence abscissa. This operation is linear and holds true.
important properties that are the basis of the calculation of transformations of classes of
functions. Among these, the following, known as the comet theorem of derivation, connects
the transform of the derivative of a function to its transform. Deriving
both members of the previous identity, the derivative's transform takes on
the following expression

d
L " dt f ( t ) #= s F ( s ) - f ( 0 ) .

with the convergence abscissa coinciding with αf

These simple calls allow for the calculation of the Laplace transform of
both members of the equations that describe a linear representation,
stationary, finite size, regular. It is obtained, indicating with the letters
capital letters the transforms of the corresponding time functions,

sX(s) - x 0=AX(s) + BU(s)

Y(s) = CX(s) + DU(s)

and, with simple manipulations,

X(s) = (sI−A)−1 x0+ (sI−A)−1 BU(s) := φ(s)x 0+H(s)U(s)

Y(s) = C(sI - A) −1 x0+ (C(sI−A)−1 B+D)U(s) :=ψ(s)x0 +W(s)U(s)

The previous relations express the Laplace transform of the evolutions

nello stato e in uscita. Esse dunque non sono altro che la rappresentazione es-
explicit of the system in the domain of the complex variable, that is the transform
of Laplace of

t t
x(t) = eAtx0+ ! eA(t−τ ) Bu(τ)dτ=φ(t)x0 + ! H(t−τ)u(τ)dτ
0 0
t t
y(t) = Ce Atx0+ ! This A(t−τ )Bu(τ)dτ + Du(t) = ψ(t)x0+ ! W(t−τ)u(τ)dτ
0 0
88 5. The stationary linear representations in the complex domain

For the identification of the corresponding terms in the previous equalities.

si ottengono le seguenti identit`a:

φ(s) = L[φ(t)] = LeAt= (sI−A)−1 H(s) = L[H(t)] = (sI - A) −1 B

$ %
ψ(s) = L[ψ(t)] = C(sI−A)−1 W ( s ) = L [ W ( t ) ] = C ( s I − A ) −1 B+D

e
t

L "! w(t−τ)u(τ)dτ=W(s)U(s)
#
0

The matrix W(s), (q×p), Laplace transform of W(t), is the matrix

of transfer functions or simply transfer function.
The previous identities express two properties of the Laplace transform.
they explain the interest in the use in the study of linear dynamic systems. The
transformed into exponential functions (thus also constants, periodic ones
The pseudoperiodic ones) are strictly proper rational functions; the transfor-
the solution of a convolution integral (and it is in this type that the forced response that
weigh the values of the input at τ with those of the impulse response at (t−τ))`e
equal to the product of the transforms, convolution theorem.
Regarding the transfer function W(s), it should be noted that the
their elements are proper rational functions, i.e., the ratio of polynomials in the
complex variables. Furthermore, the polynomial in the denominator has degree N ≤ n,
essendon the size of the dynamic matrix.
This fact, which will be clarified in the following chapter, depends on events-
removing factors from the numerator and denominator in the calculation of the inverse
of (sI−A) and in the subsequent multiplication operations.
Following from the previous considerations, the transfer matrix of
a stationary linear system with finite dimensional input-output relationships
assume the form

B0+B1 s+B2 s2+. . .+B n−1 sN −1

W(s) =D+
a0+a1 s+a2 s2+ . . . + a n−1 sN −1+sN
for suitable constant matrices DeBi ,i= 0, . . . , N −1, (q×p).
 5.2. The transition matrix in the complex domain 89

What has been presented so far provides an alternative method for calculating the
answers or, what is equivalent, for the transition to the representation ex-
explicit. Once an input is assigned, and its Laplace transform is calculated, the
calculation of the evolution at output starting from an initial state0behave
the calculation of (sI−A) −1 , simple multiplications and an operation of anti-
transformation. The antitransformation operation is particularly sem-
splice in the presence of inputs that admit rational transforms; in that case
the free and enforced response are themselves rational functions that can
to be anti-transformed without difficulty once the corresponding ones are calculated
expansions in partial fractions.
Consider, as an example, a system that has a transfer function.
1
W(s) =
s+ 1
and calculate the forced response to the unit step input, u(t) = δ−1 (t).
1
After all, how easy it is to calculate L[δ1 (t)] = , if
s has

-1 1 1 1
yf (t) = L [ ] =L−1 [− ] = (1−e−t )δ−1 (t)
s(s + 1) s s+ 1
It is observed that in the calculation it is good practice in the operation of anti-transfer
definition, to emphasize that the function obtained is defined from zero in
then multiply on the right by the function δ−1 (t).
The following delves into the aspects of computation in the domain com-
the junction and the connections are established with the analysis conducted in the chapter
precedent.

5.2. The transition matrix in the complex domain

Let it be assumed that the matrix A, (n×n), has eigenvalues (λ1 , . . . ,λ r ) with
algebraic multiplicity (µ1 , . . . , µ r ) (µ1+. . .+µ r=n). Since the calculation of the
transition matrix in the complex variable reduces to the calculation of the inverse
by (sI−A), one obtains
T
(sI−A)a E(s) E(s)
(sI−A)−1 = = &1 ' mr
= & '
|sI−A| (s−λ1 )m. . .(s−λr) m(s)
90 5. The stationary linear representations in the complex domain

in which the polynomial is indicated

mr
m(s) = (s−λ1 )m ...(s−λ
1
r)

least common multiple of the polynomials in the denominator of the elements of the
matrix (sI−A) −1 Thsi poylnomail si ceald hte mnm
i al [Link] tI
it deals with a polynomial that is a factor of the characteristic polynomial

µr
d(s) = (s−λ1 )µ ...(s−λ
1
r)

since it has the same roots as the latter, but each with a
multiplicityi geometric multiplicity, which is less than or equal to
the algebraic (µi An indirect demonstration of the fact that all self-
values appear between the zeros of the minimum polynomial is observed that if so
if it weren't for a fixed eigenvalue then the corresponding exponential law
would not appear in the matrix's exponential; what we know not to be
possible.
In the particular case of unitary geometric multiplicity, there is

E(s)
(sI−A)−1 = & '
(s−λ1 . . .(s−λr )

which admits the expansion into simple fractions

µ ν
RI Rk R∗
k
= + ) + * µ + 2ν = r
(i=1 s−λi (k=1 s−λk s−λ∗k

Such expression, placed

Rk=Rka+jRkb λk=αk+jωk

it is rewritten

µ ν
RI 2Rka (s−αk )−2Rkb ωk
(sI−A)−1 = + ) *
(i=1 s−λI (
k=1
(s−αk )2+ω2k
 5.2. The transition matrix in the complex domain 91

and returns, through anti-transformation, the expression of the exponential of

matrix in the case of already calculated in the previous chapter.
When the multiplicity of some eigenvalues is not unitary, then, as
It can be understood from the calculation, we will have the opportunity to examine more closely.

Precision in the following, the laws of motion of natural modes are more complex.
in fact, the expansion in partial fractions is composed not only of addends with
first degree polynomials in the denominator (simple fractions), but also from
fractions with a denominator of degree greater than one (compound fractions) that are
the transforms of exponential functions to polynomial coefficients int. Is this
the case that arises when the matrix is regular, that is, it does not admit
the diagonal form, but that of Jordan. In this case the size of the
The Jordan block of the largest size coincides with the geometric multiplicity.
To verify with a simple example how it can happen that in the expression
the polynomial in the denominator does not coincide with the characteristic one and
if degree < n, consider

0 1 s+1 0
A= + -1 , (sI−A)−1 = + =
0 -1 (s + 1) 2 0 s+ 1 ,
1
(s+1) 0 1
= + 1 , = (s+ 1) I
0 (s+1)
It is a matrix of size two with the only eigenvalue -1 that pre-
algebraic multiplicity two and geometric multiplicity one. The geometric multiplicity
unitary accounts for the presence of a simple exponential law in the
exponential of this matrix.
This does not happen in this other simple case.
1 1 s+1 1
A= + -1 , (sI−A)−1 = =
0 -1 (s+ 1)2 + 0 s+ 1 ,
1 1
(s+1) (s+1)2 1 (s + 1) 1
= + , = (s+ 1)2 +
0 1
(s+1)
0 (s + 1) ,
in which the geometric multiplicity of the eigenvalue is equal to the algebraic one and
ofm
egnechertsesxim
rhetat ofphispornshet ,esacshitnI [Link]
una legge esponenziale non semplice nell’elemento fuori dalla diagonale.
92 5. The stationary linear representations in the complex domain

In the presence of eigenvalues with unit geometric multiplicity, this is the

in the case of the first example, the laws of motion of the natural modes are those simple ones,
exponential or pseudoperiodic. In the case of the example, it is easy to calculate
by means of antitransformation of the elements of the matrix the temporal law
et As is known, this case corresponds to the existence of the diagonal form.
regularMatrix.
In the presence of eigenvalues with geometric multiplicity greater than one, it is
this is the case of the second example, the laws of motion of natural modes are
what complex, with integer polynomial coefficients. In the case of the example, it calculates
they easily flow through anti-transformation the laws t summert As is well known
this is the case in which the matrixAnon admits the diagonal form and the
irregular matrix.

5.3. The natural modes in the complex domain

The case of real eigenvalues with unit geometric multiplicity

It is assumed that the dynamic matrix A, (n×n), has real eigenvalues.

λi , with algebraic multiplicityiindefinite and geometric multiplicityi
unitary.
In that case, you get

λ1·µ1
E(s)
A mi= 1⇔(sI−A)−1 = & '
- (s−λ1 ...(s−λr )
-
λr·µr
r
R1 R2 Rr Ri
= + +. . .+ = =L(eAt)
s−λ1 s−λ2 s−λr (i=1 s−λI

I love the residuesiThey

are square matrices (n×n) of rank equal to the multiplicity.
algebraic. The residues can be calculated using

Ri= lim (s−λ i ((sI−A)−1

s→λI & '
 5.3. The natural modes in the complex domain 93

If it has indeed

r
L−1(sI−A)−1 = Ri eλ ti
& ' (i=1

If you remember the expression of the transition matrix in the domain of

time, it is not difficult to verify that

µi
Ri= uij v′ij=Ui Vi′
(j=1

doveij(j= 1, . . . , µI they are the right eigenvectors, all of the first order in
this case, associated with the eigenvalue λi . The matrixRi (n×n), has rank µI
being the vectorsijindependent.
A basic example:

1
A= + -1 , (sI−A) = +
s+ 1 -1
0 -2 0 s+2 ,

1 s+2 1 R1 R2
(sI−A)−1 = = +
(s + 1)(s + 2) + 0 s+ 1 , s+ 1 s + 2
conR1eR2residue matrices (2×2) of unit rank (distinct eigenvalues)
obtained by expansion into simple fractions

1 1 0-1
R1 R2 + 0 0, +0 1 ,
φ(s) = + = +
s+ 1 s + 2 s+ 1 s+2

We can now anti-transform and obtain the transition matrix.

1 1 0-1
φ(t) = + e−t + + e−2t
0 0, 0 1 ,

φ(t) = e−t u1 v′1+e−2t u2 v′2 u1 v′1=R1 u2 v′2=R2

The case of regular dynamic matrix

94 5. The stationary linear representations in the complex domain

If there are also conjugate pairs of complex eigenvalues, λk=αk+jω k

eλk ∗=αk-jωk it is obtained
µ ν
Ri Rk R∗
k
(sI−A)−1 = + + s−λ + s - λ∗ ,
(i=1 s−λi (k=1 k k

µ ν
RI 2Rka (s−αk )−2Rkb ωk
(sI−A)−1 = +
(i=1 s−λI (k=1 (s−αk )2+ω2k
Recalling the expression of the transition matrix int, it is not difficult to verify
make sure that
′ ′
vka vkb
Rk=Rka+jRkb= (uka+jukb ) -j 2
&2 '
and, developing the calculation, we have

1 1
Rka=(uka v′ka+ukb v′kb ) Rkb(ukb v2′ka-uka v′kb ).
2
Multiplying by the initial state:

Ri x0=ci ui

2Rka (s−αk )−2Rkb ωk

(s−αk )2+ω2k ·x0

= mk(s−α
& k
)senϕk +ωk cos phikuka+mk(s−αk )cosϕk −ωk senϕkukb
' & '
(s−αk )2+ω2k
One has taken into account the definitions of the parameters. In fact, by antitransforming
according to Laplace, aperiodic and pseudoperiodic modes are obtained

ci eλ ti uImk eα tksin(ω)k t+ϕk )uka+ cos (ωk t + φk )ukb

& '
Regarding the time constants, natural frequencies, and damping
it is obtained, with simple calculations:

µ ν 2Rka s 2Rkb
RI τi ωnkωnk( +ζk )− ωnk 1−ζk2
(sI−A)−1 = + 2zk s2
.
(i=1 1 + τi s (k=1 1+ s+
ωnk ω2nk
 5.4. The transfer function 95

As an example, consider the model of a simple electrical circuit.

studied in the third chapter
1
-1 -1 B= + R1 C
A= + R1 C
1
C
−R2 , 0 ,
L L
C= ( 0 R2)
R2 1
1 s+ L −C
φ(s) = +
s 2+ R2
L + 1
R1 C s+ 1
LC
1
L s+ R11C ,
& '
from which
R2
R1 CL
W(s) = R2 1 1
s 2+ L + R1 C s+ LC
& '
R1 R2 C+L
e conLC= 1, R1 LC = 1, the transfer function becomes
R2
R1
W(s) =
1 +s+s2
with pole
1 1 1 √3
s1/2=− / j
2 4 -1 = - 2 2

The case of eigenvalues with geometric multiplicity greater than one

In the general case, indicated withithe geometric multiplicity occurs for

the transition matrix the expression
r mI
RI
(sI−A)−1 =
(i=1
( k=1
(s−λi )k

1 dm−k
i
−1 m
RI= lim + (sI−A) (s−λI ) i ,
s→λi i
(mi-k)!dsm−k

Antitransforming according to Laplace

r mi
tk−1
L−1 (sI−A)−1 = RI eλ it
(i=1
( k=1
(k−1)
96 5. The stationary linear representations in the complex domain

By identity with the calculated expressions, one obtains the link between the
residualIand the vectorsicomponents ten0in the autospacesi ) It is obtained

ci=Ri1 x0 (A−λi I)c i=RI2 x0 (A−λi I) m−1 cim i−1 =RI amxI0

5.4. The transfer function

Come piu` volte sottolineato la funzione di trasferimento, essendo la trasfor-

The Laplace transform of the impulse response is a model of the behavior.
forced from a stationary linear dynamic system. Its knowledge with-
I felt the need to calculate any forced response to a given input.
through the simple relationship

yf (t) = L−1 (Yf (s)) =L−1 (W(s)U(s))

Based on what has been observed, the transfer function can be calculated
side, in a so-called black box modeling approach in which it is assumed
of not having other information about the system other than the input measurements and
corresponding outputs, starting from the knowledge of the transform of a
any forced input-output pair. In fact, for a system with a single input
and an exit one has

Yf(s)
W(s) =
U(s)
and if the input is the unit impulse δ(t), since L(δ(t)) = 1, the transform
the forced exit is precisely the transfer function.
For systems with more inputs and more outputs, the characterization of components
the matrix of transfer functions allows to define the experts-
minds that it is necessary to conduct in order to calculate the W. From the expression of the
matrix of transfer functions
w11 (s)... w1p (s)
W(s) =⎛ ... ... ... ⎞
⎝ wq1 (s). . . wqp s ⎠
 5.4. The transfer function 97

it is deduced
yi
yi=wi1 u1+. . .+wipup⇒wij= uj --uk =0k ≠ j
-
-
In reality, the identification procedure outlined now consists of the calculation-
create the transfer function from input-output experiments,
corresponds, in the time domain, to calculating the average impulse response
before deconvolution operations. Such an identification procedure and
not easy to apply in practice due to the intrinsic complexity of the measurements in
dynamic regime and the related elaborations, particularly in pre-
without errors and measurement uncertainties. We will see in the following of the chapter how
the values that the transfer function of a linear system takes
They can be directly connected to a part of the answer in
time, the steady-state response. This allows us to enable
a simple and robust experimental procedure regarding uncertainties
of measure for the identification of the transfer function.
Before delving into the details of these aspects that are at the core of a
characterization of the system behavior in relation to its behavior
In frequency, it is important to observe that the transfer function model
This can also be obtained directly through the so-called approach
a transparent box in which it is assumed to have a model of the compo-
focus on the system and its connections are known. Procedures of this type,
based on graph manipulation, are available for modeling of
mechanical and electrical systems. Among these, the most well-known in engineering is the
method of complex impedances characteristic of circuit theory. With
reference to this method is indeed known that associated with each component
a complex impedance is constructed the graph, which expresses the connections of
circuit, appropriate manipulations define a systematic procedure of
calculation of the complex impedance of the circuit, which is nothing more than the function
of transfer.
Since the model in the domain of the complex variable can, on par with the
representation with the state, being the starting point in the study of a
assigned physical system and it is necessary to specify its representativeness.
A simple example that illustrates to what extent the function
98 5. The stationary linear representations in the complex domain

the transfer is a partial model of the dynamic behavior of the system

The topic is the following. Consider a simple linear electric network composed of
from the series connection of a resistor with two capacitors; the input
both the supply voltage and the output are the voltage across the second
capacitor. It is not difficult to verify that while the representation with the
the state has dimension two (since there are two components with memory)
Transfer function has a single pole. By developing the calculations, it is verified that
from the point of view of the forced input-output connection, the circuit is equivalent
to one composed of the same resistance series with a capacitor of
capacity equal to the parallel in which the output is reduced, through a divider, of
a value equal to the capacity of the first divided by the sum of the capacities of the two
capacitors. Thus, there is a different electrical network that has only one condenser
satire and also admits the same forced behavior as before; it
It is equivalent to the first regarding forced behavior.
This network, however, is not equivalent from a point of view of points to the one assigned
born as it cannot reproduce all the answers from conditions
non-zero initials.
From the point of view of the formulas, the previous considerations find a
finding in the study conducted on the excitability and observability of natural modes
being an urali, the transfer function, the Laplace transform of
impulsive response. In summary, and with reference to a system with
distinct autovalues, the transfer function is representative of all and only the
modes that are simultaneously excitable and observable. This understands an -
that observing that under the transformation operation linear combinations of
exponential functions transform into polynomial ratios and the zeros of the denominator
they coincide with the exponents of the temporal laws; therefore the zeros of the denominators
the natural frequencies coincide with the eigenvalues associated with the excitable modes and
observable.

5.4.a. Poles and zeros of the transfer function

The zeros of the polynomial in the denominator of the transfer function in

casop=q= 1 (the zeros of the least common multiple polynomial of the polynomials
 5.4. The transfer function 99

the poles of the transfer function (at the denominator in the general case)
the poles of the transfer function matrix). In light of what
Observed, the poles are a subset of the eigenvalues of the system.
The zeros of the polynomial in the numerator in the case p=q=1 (the common zeros
determinants of polynomial matrices, of size equal to the minimum trap
equivalent, which are obtained by considering all the combinations of columns among the ep
available seq < p, which are obtained by considering all the combinations of dip
lines between the possible sep < q), are called zeros of the transfer function.
matrix of transfer function transmission zeros.
The meaning of the poles is known. The poles are the eigenvalues that appear in the
transfer function: the corresponding laws of motion, aperiodic or pseudo-
periodic, convergent, constant or divergent, characterize the behavior of
forced amendment.
The zeros of the transfer function also allow for an interpretation.
in terms of behavior input output. Siazuno zero (z1ez2a couple
of complex conjugate zeros) and siau(t) an input that admits transformed of
Laplace
1 1
( )
(s−z) (s−z1 ) (s−z2 )

Now calculate the forced response and observe that, following the cancellation-
the relationship between the transformed input and zero (the pair of zeros)
the trend over time of such a response does not contain the terms that correspond
both in the development in partial fractions at the poles that belong to the transform
of the entrance. In other words, that type of entrance generates a forced response.
which boils down to a free evolution. As we will have the opportunity to better c
subsequently taking the presence of one or more zeros translates into the ability of the
system for filtering the effect of a class of input functions. This characteristic
characteristic, which can be thought of as a lack of reactivity of the system
with respect to certain stresses, it is even more evident (and in some ways
surprising) when one thinks of such a filtering action due to the
presence of zeros in the positive real part. In this case, in fact, the filter action
occurs in the presence of an entrance whose width tends to infinity as it increases
of time.
100 5. The stationary linear representations in the complex domain

It will return to these aspects in the next chapter where, with reference
AI systems with multiple inputs and outputs, the study will be resumed based on the
following definition of zero based on state representation. The
The transmission frequencies of the system coincide with the values that cancel the
determinant of the so-called system matrix. These are the values
such

s∗ I−A−B
det ) *=0
C 0
As we will have the opportunity to show, such a definition is equivalent to the prayer.
tooth.

5.4.b. The representations of the transfer function

In conclusion of the paragraph, three different expressions are to be highlighted.

sioni of the transfer function: ratio of polynomials; factorization
zation (poli - zeros); Bode form.

Ratio of polynomials
This denomination is taken from the representation already introduced in the para-
graph 5.1.1
B0 +B1 s+. . .+Bm sm
W(s) =
a0 +a1 s+. . .+sN
oveBi , (q, p), are appropriate coefficient matrices.
As an example, consider

s(s + 2) (s2-1)
1 s−1 + s(s + 1) 0 ,
s+1 s2 +2s
W(s) =+ 1 , =
s+2 0 s(s + 1)(s + 2)

0-1 2 0 1 1 2
+ 0 0 , + + 1 0 , s+ + 1 0,
s
W(s) =
s3+ 3s2+ 2s

Factorized form (poles - zeros)

 5.4. The transfer function 101

In this form, each element of the matrix W(s) is written as

m
(s−zi )
N(s) 4
i=1
Wij (s) = =k′ n
D(s)
(s−pj )
4
j=1

Bode's form
Let us initially consider the following factorization.
real coefficients of a polynomial:

p(s) = h(s5- λi ) (s−α5 k-jωk ) (s - αk+jωk )

i k

=h(s−λ i ) (s−αk )2+ω2k

5i 5k & '

=h(s−λ
5I i ) (s2-2α 2 2
5 k s+αk+ωk)
k

=h(−λ
5II ) (1 5 +τi s) (s2-2α5k s+α2k+ω2k)
i k

2zk s2
p(s) = h(-λi ) (ω2nk ) (1 +τi s) ) 1+ s+ 2 *
5i 5k 5i 5k ωnk ωnk

αk2+ω2k=ω2nk
ωnk ζk=−αk
Starting from the previous expression, one obtains

2zk′ s2
h′ (1 + τi′ s) ) 1+ ′
ωnk s+ ′2
ωnk *
4i 4k
W(s) =
2zk s2
h (1 + τi s) ) 1+ ωnk s+ ω2nk
*
4i 4k

what is the expression used to characterize each element of the matrix

of the transfer function.
102 5. The stationary linear representations in the complex domain

In a factorization with real coefficients of this type, it is highlighted,

the time constants τi the damping ζ associated with the real eigenvaluesk e
the natural pulsation ωnk associate with each pair of complex eigenvalues
conjugates, the presence of a possible excess of poles in zero,r and the earnings,
k

ν′ µ′
2z k′ s2 ′
(1+τi′ s) ) 1+ ′
ω nk s+ 2
ω nk
*
4
i=1 4
k=1
W(s) = k
ν µ
2zk s2
sr (1+τi s) ) 1+ ωnk s+ ω 2nk *
4
i=1 4
k=1

5.5. The steady-state response

The study in the complex domain allows us to understand how the characteristics
characteristics of the response of a dynamic system are related to the behavior
treatment regarding a particular class of solicitations: the solicitations
periodic. These aspects are the basis of the so-called frequency analysis
of a dynamic system; an approach that engineers use in
study of dynamic systems. The starting point is represented by the charact-
characterization of the physical meaning of the transfer function evaluated on
points of the imaginary axis and rests on the concept of steady-state response
none.
The steady-state response to a given input is defined as
that time function at which, regardless of the initial state,
the response tends to increase over time.
If you remember that the output response is composed of free evolution and
forced, it is understood that in order to achieve independence from the initial state
it is necessary that you read the code that accompanies inCeAttend to zero at
growing of time. In other words, the condition of independence from the state
Initially, it is reduced to requiring that the observable modes are associated with eigenvalues.
a negative real part. In fact, under this condition, in
t
y(t) = Ce A(t−t) x00+ ! W(t−τ)u(τ)dτ
t0
 5.5. The steady-state response 103

the generic free evolution for a fixed point0tends to zero as the increases
time. In reality, for a physical system, the existence of the steady state does not
can do without an additional property that makes correct function possible
system zoning: it concerns the limitation of all evolutions
To understand this, as can be grasped in light of the...
God led in natural ways and as we will have the opportunity to specify in the
successive phases of our study, having already imposed that the eigenvalues asso-
If observable modes have a negative real part, it is necessary that the others
eigenvalues have strictly negative real part if with geometric multiplicity
metric greater than one, real part less than or equal to zero if with multiplicity
unitary geometry. In other words, that property which is known is necessary.
as internal stability of the system. We will therefore assume internal stability
In addition, to ensure the aforementioned independence from the initial state as-
we will assume that the eigenvalues associated with the observable modes have a real part
strictly negative regardless of their multiplicity.
Limited in this way to the class of systems to which we will refer, those
which have eigenvalues with negative real part and possibly zero those do not
observable with unit geometric multiplicity, we will define the steady-state response
permanent like that function obtained by taking the limit as it approaches.0how'sitgoing
to the response of the system.
More precisely, we define
t
yr (t) = limit ! W(t−τ)u(τ)dτ
t0 →−∞ t0

if such a function of time exists for a fixed we will call this

response to the permanent regime corresponding to the fixed input.
From the definition, it is understood that the existence of the permanent regime
It depends not only on the properties of the system, the fixed ones, but also on the input;
in fact, the summability of the integrand function must be ensured.
where members of the previous expression are fixed class of
inclusions, periodic and polynomial, the steady state exists and has the same
shape of the entrance. The steady-state response is therefore characterized by some
parameters that specify the change compared to the input.
104 5. The stationary linear representations in the complex domain

To understand that the function calculated coincides with the trend

towards which the response of the system tends to stabilize, it is enough to observe that
For the system's stationarity, the observation of the time response (t+
t′ ) is equivalent to the observation having applied the same input to
start from the moment (t0−t′ So observing as time grows is equivalent
to observe at the temple responses to the sequence of entries obtained for
translation towards negative infinity.
We will show that the now defined response represents that function to the
what tends, in the usual sense of the limit, to the forced response of the system. To this
It will be shown that

For every epsilon, there exists a T.a:||y(t)−y r (t)||<ϵ,∀t≥Ta .

Tait takes the name of adjustment time and depends on the eigenvalues of A.
In this regard, it should be noted that based on the definition given and by virtue
properties of regularity of the impulse response and for input functions
the limit operation can be moved to the endpoint of integration
and, following a change of variable, we obtain
t ∞
yr (t) = ! W(t−τ)u(τ)dτ= ! W(ξ)u(t−ξ)dξ
−∞ 0
what allows to express the difference y− yrin higher instances of time
aTa , come
Ta +t
y(T a+t)−yr (Ta+t) =CeA(T+t) x0+ a
! W ( T a+t
0
∞

-u(τ)dτ- ! W ( ξ ) u ( T a+t−ξ)dξ
0
from which follows
∞

y(T a+t)−yr (Ta+t) =CeA(T+t) x0 −

a
! W ( ξ ) u ( T a+t−ξ)dξ
Ta +t

what can be less than any set ε for Ta sufficiently large.

That is a consequence of the fact that the elements ofAteW(t) are combi-
linear combinations of decreasing exponential functions that become infinitesimal
as time goes by.
 5.5. The steady-state response 105

Per concludere`e opportuno sottolineare che a seguito della introduzione

of the steady-state response, the overall output response can be
it will be interpreted, by classes of inputs that allow the regime and for systems
asymptotically stable, like the sum of two responses called transient
and permanent
y(t) = y t (t) +yr (t)
Some study and examination methods rely on this decomposition of the response.
project of stationary linear dynamic systems; for this reason it represents
a decomposition that is in a certain sense complementary to that known in
response in free and forced evolution.
It is worth noting that since the regime coincides with the trend
limit, as time goes on, of the response and the conditions of existence, as
we have seen, they require that the free evolving response tends to zero,
we can conclude that the permanent regime is the persistent part of the
forced response while the free response is part of the transitional.

5.5.a. The permanent system with periodic entries

That said, it is important to characterize the steady-state response.

at the entrances of pure periodic type. Hence, let u(t) = eȷ ωtyes, it has

t
yr (t) = lim ! W(t−τ)eȷ ωτ dτ
t0 →−∞ t0

and put (t−τ) = ξ, developing the calculation remembering that due to the regularity of the
the functions involved the limit shifts to the boundary of integration, it is obtained
∞
yr (t) = eȷ ωt
! W ( ξ ) e −ȷ ωξ dξ=eȷ ωt W ( s )
0 -s=ȷ ω
-
The last equality is guaranteed by the fact that being the abscissa of convergence
the Laplace transform is negative being W(t) a linear combination
of exponentials with negative real part. By exploiting linearity and observing that
W(ȷ ω) in the quality of the transform of a real function verifies

W ( - j ω ) = M ( ω ) e −ȷ φ(ω)
106 5. The stationary linear representations in the complex domain

ove
M(ω) = |W(ȷ ω)|, φ(ω) = W(ȷ
̸ ω)
it is obtained

ȷ ωt eȷ φ(ω) -M(ω)e−ȷ ωt e−ȷ φ(ω)

yr (t) = M ( ω ) e =M(ω) sin (ωt+φ(ω))
2j

Thus, we have the important result that the steady-state response to a

pure periodic input is a function of the same type as the input with the
the same frequency ω and modified in magnitude and phase of quantity M(ω) and φ(ω) that
I am the module and the phase of W(s) calculated at s=jω, ω being the input pulsation.
For this reason, we can say that the modulus and the phase of W(ȷ ω) characterize,
as ω varies and under the assumption of stability of the representation, the behavior
The system's steady-state response to pure periodic inputs. W(ȷ ω) is called
harmonic response, and matrix of harmonic responses in the general case, pe/o
q > 1.
If one thinks of the inverse Laplace transform formula in which the fun-
calculation of a complex variable along a line parallel to the imaginary axis
inario and internal to the convergence semicircle, and it is remembered at the same time
that the imaginary axis, for the hypothesis Re[λi <0, belongs to such a half-plane
one understands, from a mathematical point of view, how the behavior at
the permanent regime varying with ω (magnitude and phase of W(jω)) contains all
the information on the dynamic input-output behavior. In other words
the modulus and the phase of the harmonic response for a fixed value of pulsation
they account for the behavior at steady state at that pulsation;
the module and the phase of the harmonic response vary with the pulsation
story of the dynamic behavior of the system: therefore also of the behavior-
transitory measure that pertains to forced behavior.
For characterization of frequency behavior, from a point of
application view, refers to the behavior of M(ω) and φ(ω). These functions are
significant from the point of view of dynamic behavior and these contribute
reference methods of analysis and identification. Among these, the simplest one
the identification methods, called the harmonic response method, consist of
in the performance of multiple input-output measurements on steady-state responses to inputs
 5.5. The steady-state response 107

pure periodic with different frequencies ω1 , . . . ,ω NCalculated in such a wayMi

eφi for i = 1, ..., N, the problem is solved by calculating a rational function
interpolating such values. It involves identifying two intermediate two
polynomials of complex variable, of degree respectively,
such that postoW(s) = n(s)/d(s)

W ( s ) s=ȷ ω = M ,i ̸ W(s) =φI

- - - -s=ȷ ω
- - - -
-
Obviously, what has been -
said represents the conceptual line of the procedure.
of identification that presents complex aspects related to the choices of N,med
nor the filtering of the noise effect on the measurements.
In conclusion, the harmonic response W(jω) describes the behavior in frequency.
the response of a stationary linear dynamic system and at the same time allows
to give the transfer function an equivalent physical interpretation of
that date to the impulsive response. For this reason, they take on importance.
the graphic representations of the harmonic response; the most notable are the r a p -
Bode presentations, magnitude (in dB) and phase diagrams with respect to a
abscissa in logarithmic scale, that polar, a curve in the complex plane
describe the image according to the imaginary axis, that of
Nichols, equivalent to the polar one, but with respect to a coordinate system.
orthogonal for the module (in dB) and the phase.
An appendix presents a procedure for tracking such
graphics. The interest in this topic lies, based on the previous considerations
teeth, on the importance of having a key to understanding that allows for
take the dynamic behavior starting from the knowledge of the responses
regime under periodic stresses.

5.5.b. The permanent regime with conventional entries

It will proceed now, assuming that the system meets the aforementioned condition of
existence, to the calculation of the steady-state response to so-called inputs
canonical; that is of the type

tk
u(t) = δ−1 (t)
k!
108 5. The stationary linear representations in the complex domain

The calculation of the steady-state response can be carried out into

As for the int calculation, starting from the very definition of
response to regime is available

t
rk
yr (t) = lim ! W(t−r)dr= t−r=ξ
t0 →−∞ t0 k!
∞
(t−ξ)k
= ! W(ξ) dξ
0 k!
∞ k
t k−i i i k
= ! W(ξ)· ξ -1 + i , dξ
0 (i=0 k!
k ∞
1 k k−i i
= t -1 W ( ξ ) ξ i dξ
(i=0 k! + i , !
0

k
1
= -1i tk−i Mi
(i=0 i!(k−i)!
Sia therefore obtains a polynomial of degree k whose coefficients are the parameters.
Mi
∞
M i= ! W ( ξ ) ξ i dξ
0
determinants of the impulsive response. It should be noted that such coefficients are
I define it in terms of the assumptions about the system in the impulse response com-
They seem to be temporal laws of modes associated with eigenvalues with negative real parts;
what ensures the boundedness of the integral.
The moments of the impulsive response are connected by a simple re-
Relation to the coefficients of the Maclaurin series expansion of W(s). It follows that
in fact:
i
dW(s)
Mi-1I
dsi --s=0
-
Recalling the development of Mc Laurin W(s) = - c si , in which
i
6
i
1d(W(s)
c i= -
i!dsI -s=0
-
-
 5.5. The steady-state response 109

it is obtained

-1i Mi=i!ci

⇓
MI-1i i!cI

In short, for the steady-state response to the canonical input of

order gets
k
tk−i tk
yr (t) = ci=c0+. . .+c k
(i=0 (k−i)! k!

In the particular case of unit step input, k = 0, u(t) = δ−1 it is

obtains
yr (t) = c0=W(0)
the steady-state response is therefore constant and equal to W(0), that constant value is
detailed gain of the transfer function. The gain of the function
the transfer of a system has, in the case where the steady-state response exists
permanent, the physical meaning of the steady-state value of the step response
unitary.
As previously mentioned, a different approach in the analysis of the regime
permanent with respect to canonical entries is based on the study of the response
in the domain of the Laplace transform.
A preliminary observation can be made to justify the wording.
canonical entrances. The entrances to the studio are called canonical entrances because
in analogy to what happens with the impulsive input, the forced response to
One of these inputs is a model of the forced behavior of the system. More
specifically, if the convolution product of two functions is indicated with '⋆'
of time,
L(W⋆u) = W(s)U(s) =

=sW−1 (s)U(s) = W−1 (s)sU(s)

=L(W−1⋆u′ )
110 5. The stationary linear representations in the complex domain

where the prime indicates the derivative with respect to time and

W(s)
W−1 (s) =
s
So the forced response, equal to the convolution of the impulse response with W.
siva), pu`o also be obtained as the convolution of W−1(forced response)
to the input at unit step) with the derivative of the input if, as occurs
In the case of the inputs being considered here, the derivative at zero is zero.
These topics can be generalized to higher order derivatives.
or the inputs and to the responses to canonical inputs of higher order if so
Note that the Laplace transform of the canonical input of order k:

tk
u(t) = δ−1 (t)
k!
vale
tk 1
L + k! δ−1 (t) =, sk+1
In fact
1 1
Lδ−1 (t) = Ltδ−1 (t) =
& ' s & ' s2
in fact−1 (t)`e the derivative of ditδ−1 (t) and therefore
1 1
Lδ−1 (t) = s· s2 −0 = s
& '
Generalizing
t (k−1) 1 tk 1
L (k−1)! δ−1 (t) = sk ⇒L &k!
δ−1 (t) =
s( k+ 1)
& ' '
Returning to the study of the steady-state response, consider the response
forced at the entrance of ordinek,

W ( s ) A0+A1 s+. . .+Ak sk B1 Bn

yf (s) = = ++ +. . .+
sk+1 sk+1 s−p1 s−pn
n
-1 A0+...+Ak sk p it
yf (t) = L + , + ( Bi e
sk+1 i=1
 5.5. The steady-state response 111

From this, it is evident that, having the polypsinegative real part, as time increases
the response tends to take on an even trend

-1 A0+. . .+Ak sk
yr (t) = L + ,
sk+1

Let's calculate this antitransform:

W ( s ) k+1
A0= s - W(0) = k gain
sk+1 -s=0
-
-
d W ( s ) k+1 dW
A1= s s=0 =
+
ds s k+1 , | ds --s=0
-
-
|

1dk
Ak= W ( s )-
kids k -s=0
-
-
↓

tk tk−1
yr (t) = A0+A1 +. . .+Ak
k! (k−1)!
Ultimately, the previous expression is obtained where with AiI am here
indicate the coefficientsiof the series expansion of Mc Laurin. To the calculation of such
the coefficients reduce the calculation of the steady-state response.

5.5.c. The indicator response

Let's examine in detail the forced response to step input; how

thus said such a response is called indicative response. Let it be

b0+b1 s+. . .+bm sm

W(s) =
a0+a1 s+. . .+an−1 sn−1+sn
112 5. The stationary linear representations in the complex domain

postou(t) = δ−1 (t), the forced response is data given

W ( s ) A0 A1 An
y(s) = = + +. . .+
s s s−p1 s−pn

where doveiThe poles of the transfer function have been indicated. Again a
turn, sincei )<0 it is understood that the anti-transformed of the response
forced is a function that tends to the constant value A as time increases0
=k, profit. In fact:
n
yf (t) = A0 δ−1 (t) + A1 eptδi −1 (t)
(i=1

A0 =W(0) can also be calculated by applying the final value theorem

in the domain of transforms

limf(t) = limsF(s)
t→∞ s→0

In conclusion, following the application of the single step entry

u(t) = δ−1 (t)
yr (t) = W(0)δ−1(t) = Kδ−1 (t)

The steady-state response tends to a constant value equal to the gain.

The behavior of the impulse response around zero can be set

in relation to the parameters of W(s). If one remembers that for the theorem of
initial value:
limf(t) = limsF(s)
t→0 s→∞

it is obtained immediately

y(0) = W(∞) = 0 n>m

moreover
y ′ (0) = lims sy(s)−y(0) = limsW(s) =
s→∞ & ' s approaches infinity
=bn−1let m = n - 1 = 0 if m < n−1
 5.5. The steady-state response 113

..
.

y (n−m) (0) = bm

In conclusion: the response is zero up to the derivative of order (n−m−1).

nothing. The derivative of order (n−m) is equal to the coefficient of maximum order.
of the numerator.
Example of mass - spring - damper

ẋ=x
1 2
7 k b u(t)
̇ -x1−
2x
= x2+
M M M
y=x1
-1
s -1 0
W(s) = ( 1 0 +
) k
M s+ Mb , +
M
1 ,
b
1 s+ m 1 0
= (10) +
s s + b
M + k
M −
k
M s ,+ 1
M
,
& '
1
M
⇒W(s) = s2+ Ms+ b k
M

The gain, the steady-state displacement under the action of a unit force, is
to Paris
1
A0=
k
while the behavior around zero is characterized by

1
y(0) = 0 y ′ (0) = 0 y ′′(0) =
M

The paragraph concludes by highlighting two parameters that are gen-

generally assumed to characterize the impulse response:
overelongation, s*: deviation from the maximum normalized value
to the operating value
s∗= yM-k
k
114 5. The stationary linear representations in the complex domain

climbing time, ts time needed to reach for the first time the
steady state value (in the oscillating case); time to go from 0.1 to 0.9 of
final value in the non-oscillating case.

ts they are linked to the parameters in the frequency domain, resulting in

transfer functions over a wide range: B3 ts = constant and1+s*
∼ Mr =
∼
constant. In fact, as intuition suggests, a wider bandwidth
health corresponds to a greater readiness to respond. Furthermore, if one thinks about the
the spectral content of the unit step signal has led us to think of a
such an input as 'stimulating' behavior across the entire spectrum
delle frequenze (da zero, per la componente continua, fino all’infinito, per
the initial discontinuity); this naturally leads to connecting the presence
of overstretching in the impulse response to resonances in behavior in
frequency.

5.6. The laws of motion in the case of multiple eigenvalues

As is well known, the general expression of the expansion in partial fractions of

a matrix of strictly proper rational functions is the following
T r mi
(sI−A)a E(s) RI
φ(s) = = =
|sI−A| m(s) (i=1
( k=1
(s−λi )k

where the eigenvalues of A have geometric multiplicities, mi , and algebraic, µi ,

for which the relationships apply

λ1m1≤µ1
.. .. ..
. . .
λrmr≤µr
The calculation of motion laws in the general case involves the anti-transformation.
the second member in which RIthey have expressions

1 dm −k
i
−1 m
Rik= lim + (sI−A) (s−λI )
I
,
s→λi I
(mI-k)!dsm−k
 5.7. Discrete-time systems: analysis in the complex variable domain 115

It is obtained
r mi k
1
φ(t) = Rik L−1 ) *
(i=1i=1
( s−λi
Because
tk−1 1
L) eλ it * =
(k−1)! (s−λI )k
yes, ultimately
r mi
tk−1
φ(t) = RI eλ it
(i=1i=1
( (k−1)!
The temporary laws are thus of the type

(α0+α1 t+. . .αk−1 tk )eλ t i

k = 0, ..., mi-1

The temporal laws that appear in φ(t) with varying multiplicity

metrics have similar trends as time increases if the real part
the eigenvalue is positive or negative, respectively converging to zero or -
diverging, which deviate in the case where the real part of the eigenvalue is zero.
If, in fact, the geometric multiplicity and the unitary laws of motion are constants
or otherwise limited, if it is greater than one they are divergent.

5.7. Discrete-time systems: analysis in the domain

of the complex variable
For stationary linear representations in discrete time, one can
develop similar considerations based on the Z-transform.
Given a function f(t) defined in Z+ the transformed function is defined as
function of complex variable
∞
Z[f(t)] = F(z) = f ( t ) z −t
(t=0

which is defined for |z|>ρ f,ρf radius of convergence associated with the fun-
zionef.
116 5. The stationary linear representations in the complex domain

As shown in the appendix, this is a linear transformation and holds true.

property, known as the translation theorem to the right

Z[f(t+ 1)] = zF(z) - zf(0).

The application to an implicit representation at discrete time gives,

simple calculations

X(z) = (zI−A)−1 zx0+ (zI−A)−1 BU(z)

Y(z) = C(zI - A)−1 zx0+C(zI−A) −1 B+D
& '
The Z-transform of the transition matrix is therefore given by

A Zt = (zI−A)−1 z
$ %
it is also implicitly demonstrated that the Z-transform of a con-
evolution is equal to the product of the transforms

Z At−1∗Bu(τ) = (zI−A)−1 BU(z)

& '

5.7.a. The Z transform in the study of time systems

discreet
For stationary discrete-time linear representations, it is possible to
develop similar considerations to those made using the Z-transform.
Assigned a function f(t) defined in Z+ ∪0 is defined Z transformed
dif(t) the function of complex variable
∞

Z[f(t)] = F(z) = f ( t ) z −t
(t=0

which is defined for |z|>ρ f,ρfradius of convergence associated with the function
zionef.
The transformation induced by such an operation is linear and holds true in-between
essential properties that are the basis for the calculation of transforms of classes of
 5.7. Discrete-time systems: analysis in the domain of complex variables 117

functions and their application to the study of discrete-time dynamic systems

what concerns here. Among these, the following is remembered, noted as the comet theorem of
left translation

Z[f(t+ 1)] = zF(z) - zf(0).

These simple calls allow for the calculation of the zeta transform of
both the members of the equations that describe a linear representation,
stationary, at a finite size. It is obtained, indicating with capital letters
the transforms of the corresponding functions

zX(z) - zx 0=AX(z) + BU(z) Y(z) =CX(z) +DU(z)

and with simple manipulations

X(z) = (zI−A)−1 zx0+ (zI−A)−1 BU(z)

Y(z) = C(zI - A) −1 zx0+C(zI−A) −1 B+DU(z)
& '
The previous relationships express the zeta transform of the evolutions
in the state and outgoing; they are therefore nothing more than the representation ex-
explicit of the system in the domain of the complex variable, that is, the transform
zeta of

x(t) = At x0+Σt−1
0 A
(t−τ −1) Bu(τ)

y(t) = A t x0+Σt−1
0 CA
(t−τ −1) Bu(τ) + Du(t)

Thus, the system is defined by the matrices

φ(z) = (zI - A)−1 z H(z) = (zI−A)−1 B

ψ(z) = C(zI - A)−1 z W ( z ) = C ( z I - A ) −1 B+D
W(z), (q×p), this matrix of transfer functions is the transform
zeta diW(t).
For the identification of the corresponding terms in the previous equalities
the following identities are obtained:

ZAt(zI−A)−1 z
$ %
118 5. The stationary linear representations in the complex domain

e
ZΣ0t w(t−τ)u(τ) := (A(t−1) ∗Bu(t)) = W(z)U(z)
$ %
These identities express two properties of the Laplace transform.
which explain the interest in the use in the study of linear dynamic systems.
transformed power functions, including constant functions
periodic and quasi-periodic) are proper rational functions; the transform
of a convolution integral, and this is the type of forced response that weighs
the values of the input at τ compared to those of the impulse response at −τ are equal to
product of transforms, convolution theorem.
In the last equality, the well-known property has been taken into account.
rema of the translation to the right

F(z)
Z[f(t−1)] = z

The proof is elementary and is provided in the appendix.

Even in this case, with a perfect analogy to time-based systems with-
continuously, the elements of W(z) are proper rational functions; that is, a ratio of
polinomi nella variabile complessaz. Inoltre il polinomio a denominatore ha
the degree is equal to the dimension of the dynamic matrix.
Following the previous considerations, the transfer matrix of
a stationary linear system with finite dimensional input-output equations
take on the form

B0 +B1 z+B2 z 2...+Bn−1 z N −1

W(z) = D +
a0+a1 z+a2 z 2+. . .+a n−1 z N −1+zN
for suitable constant matrices DeBi i = 0, ..., N - 1, (q × p).

Based on what has been outlined, an alternative method is therefore identified.

native for calculating responses or, what is equivalent, for the passage
day to the explicit representation. Given an input, and calculated the
Laplace transform, and an initial state the solution requires the calculation of
z(zI−A) −1 , simple multiplications and an anti-transformation operation.
The anti-transformation operation is particularly simple in the presence of
 5.7. Discrete time systems: analysis in the complex variable domain 119

entries that admit rational transforms; in this case the expressions from an-
Titrasformare are rational functions that can be antitransformed without
It's difficult once the corresponding expansions in partial fractions are calculated.

5.7.b. The natural modes in the complex domain

The same considerations made for continuous time systems apply where
please keep in mind that
Z(A t ) = (zI−A)−1 z

Let's begin our analysis with the free response in the state:

xl (t) = At x0
what it implies, in
Xl (z) = (zI - A)−1 zx0

The same considerations made for continuous-time systems apply to the

same algorithms for calculating the transition matrix φ(z) = (zI−A) −1 z.
The only difference is the multiplication by z.

As an example, consider the matrix

1 0
A= + ,
-1 0.5
in the calculation, it is advisable to refer to

aT
z−1 0 z−0.5 0
φ(z) + 1 z−0.5 , + z−1 ,
(zI - A)−1 = = -1
z (z−1)(z−0.5) (z−1)(z−0.5)

and do the expansion into partial fractions; in fact, multiplying by will have
expressions of the known type, z , of which the inverse can be calculated. In the example
z−λ
Hello studio,
1 0 0 0
+ , + +2 1 ,
φ(z) = −2 0 +
(z−1) (z-0.5)
120 5. The stationary linear representations in the complex domain

from which it is calculated

1 0 0 0
z+ , z+
-2 0 +2 1 ,
φ(t) = Z ) + * (R1+R20.5t )δ−1 (t)
(z−1) (z-0.5)

For calculating the output response, which takes the form

Y(z) = C(zI - A)−1 zx0+C(zI−A)−1 BU(z)

hay
1
C= ( 1 0 ) B= + 1,
and it is obtained through simple calculations

1
W(z) =
z−1
While the system is characterized by two natural modes and therefore two laws of
motto, only one is excitable and observable and that is the one associated with the way
constant with eigenvalue 1. Ultimately, we obtain a transfer function
ment characterized by a single pole.

How easy it is to verify this transfer function is that which characterizes

characteristic of a device called numerical integrator and represented by the
following equations:
t−1
x(t + 1) = x(t) + u(t)
8 x(0) = 0 ⇒ y(t) = u(τ)
y(t) = x(t) (t=0

The numerical integrator performs a sum of the values of the input signal.
just like the continuous integrator, characterized by a transfer function
1
imentoW(s) = , performs
s the integral

1 )=
ut()tẋ(
W(s) = ⇒ 8 W(t) = 1
s y(t) = x(t)
 5.7. Discrete-time systems: analysis in the complex variable domain 121

t
y(t) = ! u(τ)dr
0

One observes the correspondence between the operations of discrete integration and with-
continue: the transfer function of a discrete integrator (summator)
at a pole equal to 1, the transfer function inside an integrator has a
polo in 0.
The following example shows that the same applies for discrete-time systems.
the considerations already made for those with continuous time regarding the features
filtering characteristics of a given system with respect to inputs whose transform
admits poles coinciding with some of the zeros of the transfer function.
Consider the system with transfer function
z+2
W(z) =
(z−0.5)(z+1)

and the entrance = at

yf(z) 1 z+2
=
z z - a(z - 0.5)(z + 1)

and, following an expansion into partial fractions and an anti-transformation, gives

a response of the type

yf(t) = r1 at +r2 0.5t+r2 -1t

It remains evident that part of the response has the same time trend.
rale dell’ingresso. Ci`o non accade solo se si verifica una cancellazione con uno
zero of the transfer function. If it were, for example
z
u(t) = (−2)t⇒U(z) = z+2
it would be, coming out
yf (t) = r1 0.5t+r2 -1t

Thus, the zeros of a transfer function potentially represent

filtering actions against predefined classes of inputs or parts thereof; it
122 5. The stationary linear representations in the complex domain

it deals with those parts of the entries whose transformation has one or more poles
coinciding with one or more zeros of the transfer function.

As a further example, the forced response to the unit step will be calculated.
the initial answer,
z
yf(z) = W(z)U(z) U(z) = u(t) = δ−1 (t)
z−1

yf(z) W(z) K r1 rn
= = + +. . .+
z z−1 z−1z−p 1 z−pn

N
|pi |<1 W=
D
n
yf(t) = Kδ−1 (t) + rI pit yr (t) = Kδ−1 (t)
(i=1
Similar behavior to continuous time: as time increases, if the poles
dellaW(z) all have modulus <1, the answer settles around
a constant trend,K,

W(z)
K= (z−1)|z=1=W(1)
z−1
detailed gain of the discrete-time system, that is, the value to which it tends
response to the growth of time when the transfer function has all the
police with module<1.
Perfect parallel therefore with continuous time systems where the gain is
W(0) and we find the most cited correspondence between 0 and 1 at the end.
Let's proceed with our analysis by noting that assigned

b0+b1 z+. . .+bm z m

W(z) = n is greater than or equal to m
a0+a1 z+. . .+zn

The value at zero can be calculated using the initial value theorem.
(in appendix)
f(0) = lim F(z)
|z| → ∞
 5.7. Discrete-Time Systems: Analysis in the Complex Variable Domain 123

For the impulse response, we obtain

W(z) =0n>m
yf(z) = = yf (0) = W(∞) 7
(z−1) =bmm=n
from which it follows that the value at zero of the unit step response is not equal to 0 if
it is only if sem=n, otherwise the first value≠ 0 occurs at time t=n−me
its width is equal to abm In fact, recalling the translation theorem, one...
calculate with ease
yf(0) =yf...f(n−m−1) = 0 yf(n−m) = bm
In summary, from a qualitative analysis of W(z), it emerges that the delay in response
initially even adn−m, the excess poly - zeros; the amplitude of the first sample
not null in the index response and is even abm As time goes by, the response
the initial tends to take on a constant value equal to W(1).

5.7.c. The steady-state response

How well it has been highlighted with reference to time systems.

Continuing, another way to proceed with the calculation of W(s) is to do so by...
behavior in steady state towards example of
periodic type inputs characterizing the modulus and phase of W(s) itself.
Similar considerations can be made for discrete-time systems at
light of the meaning that the steady-state response takes to inputs
periodics. Such a topic is succinctly presented below.

5.7.d. The steady-state response to periodic inputs

Assigned the input = sin(θt)

u(t) = sinθt
yr (t) = M(θ)sin(θt + φ(θ))
M ( θ ) = | W ( e jθ )|
φ(θ) = W(ejθ )
124 5. The stationary linear representations in the complex domain

We know that the steady-state response is that behavior around

to which the behavior of the system tends to stabilize, a trend that
we want it not to depend on the initial state. Impose independence from the state
the initial corresponds, similarly to what has been seen for time systems with-
tinuo:
0
CA t−t x0→0 For all x0⇒|λi |<1
In other words, the condition of independence from the initial state is reduced to
require that the observable modes are associated with eigenvalues with real parts
negativa. Anche in questo caso c’`e da osservare che l’esistenza del regime
permanence cannot disregard an additional property that makes it possible
the correct functioning of the system itself: it is about the limitation of
all the internal evolutions. In order for this to happen, as one can understand
in light of the study conducted on natural ways and how we will have the opportunity to
it is necessary that the eigenvalues be specified in the subsequent phases of our study
that the unitary geometric multiplicity has a modulus less than or equal to one and
those with geometric multiplicity greater than one have strictly modulus
less than one. In other words, that property is necessary which is known as
internal stability of the system. We will therefore assume internal stability and in
addition, to ensure the aforementioned independence from the initial state we will assume
that the eigenvalues associated with the observable modes have strictly modulus
less than one regardless of their multiplicity.
Assuming such a condition on the eigenvalues of the system and taking the limit
pert0→ −∞, the expression of the steady-state response is obtained
t
yr (t) = W(t−τ)u(τ)
(
τ=−∞

that exist for definite classes of input functions. Let's consider classes of
special entries
u(t) = sin(θt) = ejθt-−jθt
2j
it is obtained
t
yr (t) = W ( t − τ ) e jθτ
(
τ=−∞
 5.7. Discrete time systems: analysis in the complex variable domain 125

and place
t-τ=ξ

∞ ∞
yr (t) = W ( ξ ) e jθ(t−ξ) =ejθt W ( ξ ) e −jθξ
(ξ=0 (ξ=0

∞
W ( ξ ) ( e jθ )−ξ =W(z)|ejθ
(xi=0

So the steady-state response to this input is:

yr (t) = ejθt W(ejθ )

The same calculations pair−jθtdamage

e−jθt W ( e jθ )

Note that the W(z) for z=ejθ it is in fact to be treasured most rare.
the radius of convergence of W(z) coincides with the maximum of the moduli of the
eigenvalues. Assuming these are all less than 1, on the circumference of
the transfer function is well defined.
From the previous calculation, adopting the polar representation for W

W ( e jθ ) W ( e −jθ )

⇓ ⇓
M ( θ ) e jφ(θ) M ( − θ ) e jφ(−θ)

and observing that they are even and odd functions of theta

φ(θ) =−φ(−θ) M(θ) = M(−θ)

it is obtained

ejθt ·W(ejθ )−e−jθt ·W(e−jθ ) jθt jφ(θ) -e−jθt.e−jφ(θ) )

= M ( θ ) ( e .e
2j 2j
126 5. The stationary linear representations in the complex domain

that is
yr (t) = M(θ)sin(θt + φ(θ))

In conclusion, the steady-state response to a pure periodic input.

and the selection of the type of entry and the stress pulsation; it is modified
in modulus and amplitude of quantities that are equal to the modulus and the phase of W
calculated inz=ejθ For this reason, it is said that the modulus and the phase of
W(z) perz = ejθ the variation of θ between 0 and π characterizes the behavior in
frequency.

5.7.e. The steady-state response to canonical inputs

Siat(k) a factorial polynomial of order (k)

t(k) (t(t-1)(t-2)...(t-k+1))

and let us consider the canonical input of order (k) for discrete-time systems

t(k)
u(t) =
k!

t
yr (t) = lim W(t−r)u(r)
t0 →−∞ (
r=t 0
t
r (k)
= lim W(t−r) t-r=θ
t0 →−∞ (
r=t 0
k!
∞
(t−θ)(k)
= W(θ)
(θ=0 k!

Since for the factorial polynomial of order (k) of a binomial the expansion holds
following
k
k
(t−θ)(k) = i (k−i)
+ i , -1 t (θ+i−1)
(i)
(i=0
 5.7. Discrete-time systems: analysis in the complex variable domain 127

yes it has

k ∞
1 k
yr (t) = -1i t(k−i) W ( θ ) ( θ + i − 1 ) (i)
(i=0 k! + i , (θ=0
k i
-1 t(k−i)
= Mi
(i=0 i! (k−i)!

in which it is placed

dI W ( z )
Mi-1i
dzi |
z=1

in fact ∞ ∞
di W ( z ) di W(t) equals negative onei(t+i−1)(I)
W(t)
=
dz I dz (t=0 z t
i (t=0 z t+1

Also, if we consider the power series development of W(z) around z = 1,

i
1d W I
W(z) = z=1 (z−1)
(i ≥ 0 i!dzI |

it is deduced that
Miequals negative oneI i!cI

In conclusion, for the steady-state answer, the expression is obtained

k
t(k−1)
yr (t) = ci
(i=0 (k−1)

which shows a close formal analogy with that obtained for the systems at
continuous time.
The steady-state response can also be calculated as the part
persistent forced response. In this case, it is useful to refer to the
domain of transforms.
A first aspect concerns the calculation of the transform of the input canon-
factorial icon
t[k] t(k−1)...(t−k+1)
=
k! k!
128 5. The stationary linear representations in the complex domain

results
t[k] z
Z" =
k! # (z−1)k+1
To verify what has been stated, it is necessary to premise a property of the
Zeta transformation. More precisely

d
Z tf(t) = -zF(z)
& ' dz
indeed

d f(1)f(2) f(1) 2f(2) 3f(3)

) f(0) + *
+ 2 +. . . =−+ ) + + *=
dz z z z2 z3 z4

and multiplying by -z

d f(1) 2f(2)
-zF(z) + +. . .=Z t ·f(t)
dz = z z2 & '
We can now calculate the Z-transform.
d z z
Z[t] = -z =
dz z^{-1} (z−1)2
and then the transformation of the factorial polynomial

t[k] =t(t-1)...(t-k+1)

d 1 2z
Z[t [2]=Z[t·(t−1)] =−z =
dz(z−1)2 (z−1)3

t[2] t·(t−1) d 1 z
Z9 : =Z 9 : =−z · =
2! 2 dz(z−1) 22 (z−1)3

t(t−1)(t−2) z
Z9 :=
3! (z−1)4
and, in general:
t[k] z
Z9 :=
k! (z−1)k+1
 5.7. Discrete-time systems: analysis in the complex variable domain 129

The steady-state response to such an input can be calculated.

easily; indeed it has

yf(z) W(z) N(z)

= =
z (z−1)k+1 D(z)(z−1) k+1
and developing the calculations

n
yf(z) c0 c1 ck rI
= + +. . .+ +
z (z−1)k+1(z−1)k (z−1) (i=1 z−pi

n
t(k) t(k−1)
yf (t) = c0 +c1 +. . .+c k + rI pit
k! (k−1)! (i=1

and infinite
t(k) t(k−1)
yr (t) = c0 +c1 +. . .+ck
k! (k−1)!
with
W(z)
c0=K= (z−1)k+1 - =W(1)
(z−1)k+1 -z=1
-
-
gain of the discrete system

d W(z)
c 1= ) (z−1)k+1 *-
dz(z−1) k+1 -z=1
-
-
.....

1di
c i=
W ( z )-
i!dzi -z=1
-
-
Even in the case of the discrete-time system, we can repeat the same with-
previous considerations; the coefficients are nothing but the coefficients
of the Taylor series development of W(z) around the point z = 1.
130 5. The stationary linear representations in the complex domain

5.7.f. The laws of motion in the general case

We can repeat here the same considerations already made in the case of time.
continuo
r mi
φ(z) RI
=
z (i=1
( k=1
(z−λi )k

z
φ(t) = Rik Z −1 ) *
( (i k
(z−λI )k

To antitransform, it is necessary to observe a property of the transform.

zeta:
Z[λ t f(t)] = F(λ−1 )

in fact
λ λ2
Z[λ t f(t)] = f(0) + f(1) + f(z) + . . . 2
z z
from which
t(k−1) z/λ z zλ (k−1)
Zλ9t := = λ
=
(k−1)! (z/λ−1)k λk (z−λ)k (z−λ)k

Again, one has

tk−1 zλ k−1
Zλ9−k+1λt := ·λ−k+1
(k−1)! (z−λ)k

that is
t(k−1) z
Zλ9t−k+1 :=
(k−1)! (z−λ)k
e
r mi
t(k−1)
φ(t) = RI λt−k+1
(i=1
( k=1
(k−1)! i

These are the laws of motion that appear; therefore no longer just laws of motion.
of the type (λt ), but also laws of motion with coefficients that are polynomials in
(t). It corresponds to the terms that characterize the power of a matrix A
when this is not regular (the diagonal form does not exist). The same applies
 5.7. Discrete-time systems: analysis in the complex variable domain 131

considerations made regarding continuous-time systems, remembering that

In this case, the diversity of behavior as time goes by is observed.
in the case of eigenvalues with unit modulus.

5.7.g. Discretization of the transfer function

of a continuous system
Let's assume that the continuous-time system is described by the function
of transfer
Stc=W(s)
The system we obtain is a linear discrete-time system, therefore the link
Forced will be characterized by aW(z). It will now be shown how to calculate
directly the W(z) from the W(s).

Real. ↑CD (zI−AD )−1 BD

(A, B, C) −→ (AD , BD , CD )
D
The solution is the following:

z−1 W(s)
W(z) = Z 9L−1 ) *- :
z s -
-t=KT
-
To understand this expression, note that the system is- time-discrete.
linear criterion, it is sufficient to calculate W(z) to express the ratio between
a forced exit and the corresponding entry; moreover, the step response to
discrete time coincides with the sampling of the unit step response
continuous-time system behavior (that is because the response to a discrete step,
from a continuous step); therefore the sampled continuous impulse response:

W(s)
L−1 ) *-
s -
-kT
-
coincides with the system's impulse response in- discrete time. Making it
Z-transform
W(s)
Z 9L−1 ) *- :
s -
-kT
-
-
132 5. The stationary linear representations in the complex domain

and dividing by the Z-transform of the discrete-time step function, we obtain, for
as observed, the sought transfer function

yf(z) z−1 −1 W(s)

W(z) = = 9 L )
From *- :
uf(z) z s -
-kT
-
-
Example:
1
W(s) =
s2
0 1 0
⎧ =ẋ + 0 0 , x+ + 1, u
⎪⎨
y = (1 0)x
⎪⎩
implementation of the model W(s) = s12 .
This is a very common model, a model on which many considerations are based.
actions of both analysis and synthesis, intervention and control strategies.
Calculation:
W(s) 1
= 3
s s

W(s) t2
L−1 ) *
s 2

W(s) k2T 2
L−1 ) *-
s - 2
-t=kT
-
-
−1 W ( s ) T2
9 L )
From *- := z(k 2 )
s - 2
-t=kT
-
- z
z(k) =
(z−1)2
d z
z(k·k) = -z
dz(z−1)2

−1 W(s) T 2z(z + 1)
9 L )
From *- :=
s - 2 (z−1) 3
-t=kT
-
-
 5.7. Discrete-time systems: analysis in the complex variable domain 133

z−1 −1 W(s) T 2z(z + 1)z - 1

9 L )
From *- :=
z s - 2 (z−1) 3 z
-t=kT
-
-
z−1 From L−1 W ( s ) T 2(z+ 1)
9 ) *- := =W(z)
z s - 2 (z−1) 2
-t=kT
-
In the solution, there is a double pole at- one:
1
(z−1)2 , just like in the W(s) = s2
where there is a double pole at zero.

Zero di W(s) under discretization:

While the eigenvalues, and thus also the poles of the transfer function,
they transform according to the correspondencesT, the situation presents itself
complex regarding the zeros. The first aspect concerns their
number;`is generically (n−1) regardless of how many zeros there are
diP(s). Quindi sotto discretizzazioneappaiononuovi zeri; piu` precisamente
The SeW(s) has m zeros, W(z) has (n−m−1) additional zeros. It could be proven.
sufficiently small

zT
zDi =
∼e i i = 1, ..., m

1
i = m + 1, ... , n - 1 zDI discretized zero of sn−m

1
it is important to note that for (n−m)>2 the zeros of the discretized sn−m

they are unstable (|zD

i 1).
In other words, for Tpiccolo (as usually happens) the transfer function
element, of the discretized version of a continuous process with excess pole-zero >2,
presents unstable zeros, even if the process does not have them.

Discretization of systems with delay

The last consideration concerns the discretization of systems with delay.

W ( s ) = e −λs H(s)

λT = eT · mT 0<m<1
134 5. The stationary linear representations in the complex domain

W ( s ) −It is emT s a
=e
s s(s+a)

z−1 1 mT s emT s
W(z) = z 9 L−1 ) e *- :
z zl s − s+a -
-t=KT
-
-
z−1
= l+1 zδ) −1 (kT+mT)−e−a(kT+mT) *
z

z−1 z e−amT z
W(z) = ) *
z l+1z−1z−e −aT −

Thus, one obtains a W(z) which is a ratio of polynomials, in reality one obtains
that the degree in the denominator is equal to (λ + 1). Therefore, there is a function of
transfer that is a ratio of polynomials and therefore a representation with
the state that is of finite dimension, which in reality would not occur in this
case for the continuous-time system.

5.7.h. From the forced model, install the representation

with the state

The forced model coincidence in the complex domain means that the
the same solutions identified for continuous-time systems can be
apply AI systems to discrete time.

5.8. Appendix 4
 5.8. Appendix 4 135

5.8.a. The Laplace transform

Recall the definition of the Laplace transform that associates to a function

f(t) defined for t ∈ [0, ∞) as a complex variable function
∞

L[f(t)] = F(s) := ! e−stf ( t ) d t

0
which is defined for Re[s] > α, α associated with f(t) called the abscissa of convergence
This operation is linear; moreover, it can be verified by deriving both.
members of the previous identity that the transform of the derivative takes on the
the following expression with the abscissa of convergence coinciding with that of the dif
derivative theorem
d
L " dt f ( t ) #= s F ( s ) - f ( 0 ) .
Definition, properties, residue decomposition, elementary formulas.
held in multiple ways

5.8.b. The Bode diagrams

It is common to refer to a particular representation of poles-zeros to co-

real efficiencies of the transfer function. In this regard, it is recalled that
the factored form is
(s−zI )
N(s) ′ 4 I
p=q= 1 W(s) = =k
D(s) (s−pj )
4j
To bring it back to the desired form, proceed as indicated below for
a generic polynomial.

p(s) = h(s5- λi ) (s−α5 k-jωk ) (s - αk +jωk )

i k

=h(s−λ i ) (s−αk )2+ω2k

5i 5k & '

=h(s−λ
5i i ) (s2-2α 2 2
5 k s+αk+ωk)
k

=h(-λ5i ) (1 5 +τi s) (s2-2α5k s+α2k+ω2k)

I i k
136 5. The stationary linear representations in the complex domain

2zk s2
p(s) = h(-λi ) (ω2nk ) (1 + τI s) ) 1+ s+ 2 *
5i 5k 5i 5k ωnk ωnk

αk2+ω2k=ω2nk
ωnk ζk=−αk
Starting from the form of the generic factorization with real coefficients of a
polynomial(s), it is obtained

2zk′ s2
h′ (1 +τi′ s) ) 1+ ′
ωnk s+ ′2
ωnk
*
4i 4k
W(s) =
2zk s2
h (1 +τi s) ) 1+ ωnk s+ ω2nk *
4i 4k

what is the expression used to characterize an element of the matrix of

transfer function.
In a factorization with real coefficients of this type where they are placed in
evidence, the time constants τi associated with real eigenvalues, it dampens-
mentoζkand the natural pulsation ωnkassociate with each pair of eigenvalues
conjugated complexes, the presence of a possible excess of poles in zero,srand the
gain,K

2zk′ ′
s2
K (1 + τi′ s) ) 1+ ′
ωnk s+ '2
ωnk
*
4i 4k
W(s) =
2zk s2
sr(1 +τi s) ) 1+ ωnk s+ ω2nk *
4i 4k

Starting from such a factorized form with real coefficients, the so-called form
Bode's factored form, it is relatively easy to understand what the behavior is
in frequency of the well-represented system, as noted from the response ar-
monica, W(jω). It is indeed necessary to remember that its modulus and its phase
they represent, in the case of all eigenvalues having negative real parts, the
modification in amplitude and phase that a pure periodic signal undergoes while passing through
 5.8. Appendix 4 137

through the system.

The constant K is the gain of the system, which is exactly the W(s) calculated
ins= 0 once any poles at zero in the function have been eliminated
of transfer.
The physical meaning associated with gain, such as the value to which the response tends.
a permanent regime at the entrance at a unit step can be attributed
only in the absence of poles at zero and with all other poles having a strictly real part
very negative.

2z s2
K s 1 + τs 1+ s+ 2
ωn ωn

there are the four different types of terms that characterize the W(s); the term
s2
costanteK, monomios, binomio 1 +τs, trinomio 1 +2ζ s+ . The ωn product
ω2n
of this type, in the numerator and denominator, gives rise to a
genericW(s).
The Bode diagrams

Post
|W(jω)|=M(ω)

phaseW(jω) = φ(ω)

Bode plots are a representation of

M(ω)
7
φ(ω)

Particular representation because a scale is used on the ω axis.

Logarithmic representation, therefore, the modulus and phase φ vary.
ωma according to a scale, on the x-axis, which is indeed logarithmic in
base 10. If you need to report the true values of ω on the axis. Compared to ω, the phase φ
is represented in degrees or radians and the magnitude in dB, MdB ,

MdB= 20 log10Logarithmic representation of the modulus

138 5. The stationary linear representations in the complex domain

The justification for such a choice is provided in the following steps where we
note that the diagrams of a product transfer function, this that
responds to the cascade of two systems, obtained by summing the diagrams
the indBe modules of the phases

W ( s ) = W 1 (s)W2 (s)
1 2
W=Mejφ =M1 ejφ ·M2 ejφ
1 2
W=M1·M2 ej(φ +φ )

and with the representation in dB

MdB= (M1 M2 )dB=M1dB+M2dB

So, starting from the graphic representations of modules and phases, the rep
The overall graphic presentation is obtained by summing for.
Based on the previous consideration, the opportunity of choosing the methods
the representation is validated by the fact that, as already noted, the W(s)
the product of four terms, constant, monomial, binomial and trinomial.

Bode diagram tracking

To represent the frequency behavior of a W(s) it will be enough
to sum the trends of the individual terms.
Let's review the graphical representations of the individual terms.
The first representation concerns the constant term K.

KdB= 20 log10|K|

its value is positive if |K|>1; negative if the gain |K|<1

La fase sar`a costantemente nulla seK`e p o s it iv o e s a r`a −πseK`e n e g a t iv o .
The following figure illustrates two different situations.

K= 100

K=−1000
 5.8. Appendix 4 139

100
K = -1000
50
K = 100
0

-50

-100
10-2 10-1 100 101 102
pulsation

200

100

K = 100
0

-100
K = - 1000
-200
10-2 10-1 100 101 102
pulsation

Figure 5.1

For the term monomio, we have

y=MdB= 20 log10ω = 20x

that is the module is a line with a slope of 20dB per decade, with a semi-
π
pre-increasing. The phase, however, is constantly equal to. A
2
monomial factor at
the denominator is represented by opposing trends. The figure shows
the trends
140 5. The stationary linear representations in the complex domain

100

AT THE DENOMINATOR
50

-50
A NUMERATOR

-100
10-2 10-1 100 101 102
pulsation

100 AT THE NUMERATOR

A DENOMINATOR
-100

10-2 10-1 100 101 102

pulsation

Figure 5.2

Now consider the binomial term 1 + jωτ

MdB= 20 log101 + ω2 τ 2
.
1 √
ω= ⇒MdB= 20 log102 = +3dB
|τ|
1 √
ω≪ =
⇒MdB20 ∼
log 10 1=0
|τ|

1
ω≫ = omega +1020 log tau
∼
⇒MdB20 log 10
|τ|
1
Perωmuch less than |τ| the asymptotic diagram coincides with the axis of
abscissa; a line with a slope of 20dB per decade represents the value of
module perω≫|τ | Thsinilein1etsrecst atω= 1
|τ| the x-axis. The
The approximate representation made up of a broken line on two sides is called
asymptotic diagram.
 5.8. Appendice 4 141

1 1
The phase perω≪ |τ| it is practically null, for ω>>> |τ| e p r a t ic a m e n t e
π
the fairy conτ>0
2 is equal to− 2conτ<0. π
The modulus and phase of binomial terms are represented in the figure.
numerator and denominator for positive values of τ.
100

At the numerator
50

At the denominator
-50

-100
10-2 10-1 100 101 102
pulsation

200

AT THE NUMERATOR
100

-100 On the Denominator

-200
10-2 10-1 100 101102
pulsation

Figure 5.3

1
With reference to a binomial term in the denominator, in ω = |τ| the module
presents a variation of (1/2)dB−3d√ relative to the zero value of
asymptotic diagram
As observed in the figure, the trend corresponding to values of τ
negatives do not change regarding the module, while it becomes opposite
regarding the phase. With constant term tracking
and of the term binomial we are able to characterize the behavior in
frequency of a system that, for example, has a transfer function type
K
W(jω) =
1 + jωτ
Based on what has been stated, the trend of the module and the phase is characterized by
from the sum of the trends modulo - phase of the two factors that in this case
142 5. The stationary linear representations in the complex domain

1
soneKe1+jωτ. The frequency ω= |τ| and that pulsation corresponds
of which the value of the gain of the module is −3dB compared to the initial value,
this pulsation is normally referred to as the bandwidth, B3 .
We have yet to address the representation of a trinomium, which
we will support the denominator, corresponding therefore to the transfer function
mind of a dynamic system characterized by a pair of eigenvalues com-
conjugated pairs. This will allow assigning a physical meaning to parameters
specify what the natural frequency and the damping are
1
(jω)2
1 +2ζωjω+
n ω2n

ω2 4z2
MdB-20 log10 ? + 1− , + 2 ω2
ωn
2 ωn
At ω=ωn The value of the module depends on the second:

MdB-20 log102|z|

and for the unit module, that which corresponds to the cancellation of the part
imaginary and the reduction of the trinomial square of a term
binomial, the value of the modulus results in −20 log 102 = −6

As for asymptotic behaviors, for low and high values of ω

respect to ωn ,
ω≪ωn⇒MdB = 0
∼
ω2
=
∼
ω≫ωnMdB-20 log 10
ω2n
MdB-40= log ω + 40
∼ 10 log ω 10 n

yes, therefore a line that has a slope of -40dB per decade, which crosses
the axis of literally at the pointn . It could be verified that the trend
effectively coincident with the asymptotic one at a decade of
distance from ωn .

ωn=α2+ω2natural pulsation or resonance pulsation

7 .
ζ= −α damping
ωn
 5.8. Appendix 4 143

If the damping is low, it corresponds to omega.n a resonance,

an amplification of the response. The deviation of trends as they change
reporting the following figure in accordance with a standardized census
compared to the value ωn

60

40
SMORZAMENTO = 0.01

20

0
SMORZAMENTO = 1

-20

-40

-60
10-1 100 101

pulsation

Figure 5.4
144 5. The stationary linear representations in the complex domain

0
SMORZAMENTO = 0.01

-50

-100
SMORZAMENTO = 1

-150

-200
10-2 10-1 100 101 102

Figure 5.5

Similar considerations can be made for the trends of the phase of

a trinomial term in the denominator.

2 live
ωn
φ = arctan
1− ωω2n
2

But depending on the assumptions initially that this is positive

phase is about zero, as it tends to ωn from values
I have a much smaller amount than [Link]
the lower tends to -∞ and as ω increases it tends to zero again from negative values,
it corresponds to a trend between zero and -π as indicated in the figure for
different positive values.
 5.8. Appendix 4 145

100

50 A NUMERATOR

AT THE DENOMINATOR
0

-50

-100
10-2 10-1 100 101 102
pulsation

200
A NUMERATOR
100
AT THE DENOMINATOR
0

-100

-200
10-2 10-1 100 101 102
pulsation

Figure 5.6

Obviously, the trends of the phase for negative values of ζ are opposite.
to those indicated.
It is interesting to examine the frequency behavior of one of the systems.
physicists considered in the previous lessons.
Bode diagrams of the mechanical damper
1/K
W(s) = Ms2
k + bsk + 1
1 M
=
ω2n k

k b
ω n= / ζ= √
M 2kM
Seb >2kMlo√ damping would result in 1 which corresponds to eigenvalues
reality.
We evaluate such parameters as the mass and the elastic constant vary.
molla and the coefficient of dynamic friction.
146 5. The stationary linear representations in the complex domain

M= 104Kg k= 106N m → ω n = 10 rad/sec

104 1
b = 104⇒ζ= √ = = 0.05
2kM 20

102
b= 102⇒ζ= √ 0.0005
2kM
Let's examine what happens in the two cases when the system is stressed with a
own periodic pulsation signal = ωn= 10 rad/sec

1
1
- (j10)2 10
6
|W(j10)|= -= = 10−1−1b
- +j+ b
1 - 10b
- 100 k -
- -
- -
That is, if we stimulate with a pulse of 10 radians per second with a
a force of 100 Newtons results, at steady state, in an oscillation of 10−3 m, 1 mm
(millimeter), in the first case, 10 centimeters, in the second.

A dynamic system characterized by a pair of complex eigenvalues

conjugated can be used to describe, albeit with extreme simplicity
causing, the dynamics of the central part of a building structure, such as that one
of a bridge, subjected to vertical loads. It can be assumed, in fact that
in first approximation the effect of the connected parts has, in the
stress fronts acting vertically, an effect of 'pulling' equal to
that of a spring and a damper. Such a model, in light of the
considerations made, lends itself to interpreting some characteristic phenomena
of the dynamics of structures; in fact, as the model suggests, in pre-
without non-elevated dampening (large span/flexible structures) there exists
a frequency, characteristic of the structure itself, at which
Periodic stresses can produce amplifications of the response. The
previous considerations are not far from reality as the dynam-
The performance of a flexible mechanical structure can always be described by
the interconnection of dynamic systems of this type. The resulting model
presents the phenomenon of resonance at different frequencies;
the first of them, which is assumed to characterize the structure, corresponds to
 5.8. Appendix 4 147

a simplified modeling and uses a mass-spring type model

dampers.
As for the tracking of Bode diagrams starting from the dia-
The grammars of the individual factors, so far discussed, hold the considerations already made;
The overall diagram is obtained by summing the individual factors.
As an example, consider the following transfer function

(jω)2
100+ 1 +2ζ
ωn(jω) + jω2n ,
W(jω) =
jω(1 + jωτ1 (1 + jωτ2 )

the harmonic response in the factored form of Bode is characterized by the

following parameters

τ1= 10 τ2= 0.1 z = 0.1 ωn= 10

The trend is reported in the following figures.

100

50

-50

-100
10-2 10-1 100 101 102
pulse

200

100

-100

-200
10-2 10-1 100 101 102
pulsation

Figure 5.7
148 5. The stationary linear representations in the complex domain

100

50

-50

-100
10-2 10-1 100 101 102
pulse

200

100

-100

-200
10-2 10-1 100 101 102
pulsation

Figure 5.8

The considerations made regarding the plotting of Bode diagrams are ...
important not for the actual tracking of the diagrams themselves, what can
to be entrusted to a very simple calculating tool, but to understand
for inspection on the factors of the transfer function what will be the behavior
tamento in frequenza e, quindi nel tempo, del sistema stesso. Innanzitutto
it is necessary to remember that the corresponding elements depend on various factors
standard: to a monomial term, which corresponds to an integral action-
In the perspective, a fixed trend is associated that translates into concrete terms in a
constant decrease in gain at different frequencies and a phase shift
constant; to a binomial term corresponds a standard trend that
translate into a maintenance of profit and a subsequent constant decrease
decrease with a corresponding increase in phase shift; the time constant does not
it modifies the device's behavior, but only its 'placement' in fre-
similar considerations apply for a trinomial term that also
can exert a resonance or a filtering action (think of a term
trinomio a numeratore).
 5.8. Appendix 4 149

One last consideration concerns the low-pass behavior of a system.

dynamic linear strictly causal. Based on what has been observed, as the increase of
ω, the module always goes to zero (−∞ in dB), every time the degree of
the denominator is greater than the degree of the numerator (i.e., the system is strict-
causal mind). Furthermore, the phase as ω grows tends always to multiple values.
positive or negative π/2
Alcuni esempi di comportamento a regime permanente ad ingressi periodici
are proposed below.
Consider the following transfer function

1
W(s) =
s2
) 1 + 2ζωns+ ω2n *

ζ= 0.1 ω n= 2

The figure shows the trend of the steady state response.

INGRESSO = SIN( t )
5

2
Y(t)
1

0
U(t)
-1

-2

-3

-4

-5
20 21 22 23 24 25 26 27 28 29 30

Figure 5.9
150 5. The stationary linear representations in the complex domain

INGRESSO = SIN( 2t )
5

4
Y(t)
3

1
U(t)
0

-1

-2

-3

-4

-5
20 21 22 23 24 25 26 27 28 29 30

Figure 5.10

INGRESSO = SIN( 5t )
5

1 U(t)
Y(t)
0

-1

-2

-3

-4

-5
20 21 22 23 24 25 26 27 28 29 30

Figure 5.11
 5.8. Appendix 4 151

We are around the natural pulsation of the system as shown in the diagrams.
in Bode reported below
BODE DIAGRAM
20

10

-10

-20
10-1 100 101
OMEGA

-50

-100

-150

-200
10-1 100 101
OMEGA

Figure 5.12
152 5. The stationary linear representations in the complex domain

BODE DIAGRAM
6

0
10-1 100 101
OMEGA

-50

-100

-150

-200
10-1 100 101
OMEGA

Figure 5.13

In the module diagram, there are two parameters that are often
assumptions to characterize the frequency behavior: the passband B3 ,
that pulsation from which there is an attenuation in the response to
permanent regime higher than 0.707 compared to the value in ω= 0; the modulus at
resonanceMr that is, the maximum value of the modulus divided by the value of the modulus
in zero.
Pronounced resonances highlight the fact that there are frequencies
which are exalted, while a wide passband characterizes the ra-
small time constants with which the system responds.
parameters and the previous simple considerations find a correspondence in-
mediated in the behavior of the response to the unit step. In fact, a wide
passband, response speed, translates into a rapid tendency to the val-
regime hours (low response time), while the enhancement of signals in a
certain frequency range translates into the presence of an enhancement of the
response in the presence of the initial discontinuous stimulus and for this reason rich
in harmonic content.
 5.8. Appendix 4 153

As an example, and to verify the aspects just highlighted, let us consider

the following two transfer functions
K
W1 (s) = τ2≪τ1
1 + τ1 s
K
W2 (s) =
1 + τ2 s

calculate the impulse response and the bandwidth in both cases.

As an additional exercise, calculate the steady-state response to a periodic input.
the wave square of a system that has a transfer function
1
W(s) =
s2
) 1 +2ζωs+
n ω2n *

ωn = 5 z = 0.5
the result is shown in the following figure. The previous considerations
They intend to convey the qualitative trend in relation to the different values.
the parametersn .
FORCED RESPONSE WITH INPUT = SQUARE WAVE
1.5

0.5

EXIT
0

-0.5

ENTRY
-1

-1.5
0 1 2 3 4 5 6 7 8 9 10

TIME IN SECONDS

Figure 5.14
154 5. The stationary linear representations in the complex domain

Similarly, it is easy to understand that adding a direct link

unitary input output, the trend shown in the following figure is obtained.
FORCED RESPONSE WITH INPUT = SQUARE WAVE
2.5

2
EXIT
1.5

0.5

-0.5

-1 ENTRY

-1.5

-2

-2.5
0 1 2 3 4 5 6 7 8 9 10

TIME IN SECONDS

Figure 5.15

5.8.c. The polar diagrams

An additional graphical representation of the harmonic response useful for study

of dynamic systems and in the design of control systems and that which is described
polar.
Formally, the polar diagram is the image according to the function W(s)
of the imaginary axis. It is a transformation of the complete plane in itself
the same that transforms the points of the imaginary axis into a curve in the comp-plane
lesson that for reasons of regularity (image of an holomorphic function) describes
a closed curve in the complex plane if there are no singularities on the axis.
imaginary (such singularities can only be potential zeros of the polynomial
at the denominator of the transfer function).
The qualitative trend of the polar diagram can easily be
understood from the qualitative trend of the Bode plots.
 5.8. Appendix 4 155

An important first observation concerns the symmetry that the dia-

grammar manifests itself in relation to the real axis. This can be easily understood.
if you remember that the W(jω) is the Laplace transform of a function of
time calculated as s=jω. Writing this transform and using the formula of
Euler to represent the exponential, ejω =cosωt + jsinωt, and indeed I am-
to check that a complex function is obtained whose real part (and
modulo) is an even function of ω and the imaginary part (and phase) is a function
odd of ω; that which corresponds to the mentioned symmetry.
Thus, the trend for positive values is traced in the polar diagram.
The figure is obtained immediately since for negative values of ω the figure is symmetrical.
with respect to the real axis. As an example, let us consider the transfer functions
the following components and their respective Bode and polar diagrams.

10
τ 1= 1
(1 + jωτ1 )

20

10

0
10-1 100 101
Frequency (rad/sec)

-50

-100

-150

-200
10-1 100 101
Frequency (rad/sec)

Figure 5.16
156 5. Stationary linear representations in the complex domain

0 OMEGA = + INFINITY OMEGA = 0

-1

-2

-3

-4

-5

-6

-7
-2 0 2 4 6 8 10

REAL PART

Figure 5.17

K
τ1 τ2>0
(1 + τ1 s)(1
+τ2 s)
5.8. Appendix 4 157

20

10

-10

-20

-30
10-2 10-1 100 101
Frequency (rad/sec)

-50

-100

-150

-200
10-2 10-1 100 101
Frequency (rad/sec)

Figure 5.18

OMEGA = + INFINITY OMEGA = 0

0

-1

-2

-3

-4

-5

-6

-7
-2 0 2 4 6 8 10

REAL PART

Figure 5.19
158 5. The stationary linear representations in the complex domain

10
τ1= 1 τ2= 2
(1 + jωτ1 (1 + jωτ2 )

10
τ 1= 1 τ 2= 2 τ 3= 2
(1 + jωτ1 (1 + jωτ2 (1 + jωτ3 )

20

-20

-40

-60
10-2 10-1 100 101
Frequency (rad/sec)

-100

-200

-300
10-2 10-1 100 101
Frequency (rad/sec)

Figure 5.20
5.8. Appendix 4 159

0 OMEGA = + INFINITY OMEGA = 0

-1

-2

-3

-4

-5

-6

-7
-2 0 2 4 6 8 10

REAL PART

Figure 5.21

(s−0.5)
W(s) = 40
(s+ 1)(s+ 4)
160 5. The stationary linear representations in the complex domain

20

10

-10
10-2 10-1 100 101 102
Frequency (rad/sec)

200

100

-100

-200
10-2 10-1 100 101 102
Frequency (rad/sec)

Figure 5.22

6 OMEGA POSITIVES

OMEGA = 0 OMEGA = INFINITE

0

-2

-4

-6 OMEGA NEGATIVES

-8
-6 -4 -2 0 2 4 6 8 10

REAL PART

Figure 5.23
 5.8. Appendix 4 161

In conclusion, note that if there are singularities on the imaginary axis, the
The polar diagram is open, meaning there are branches that go to infinity.

5.8.d. On the calculation of the transition matrix*

Indifferently for a continuous or discrete time system, there is:

(sI−A)aT
(sI−A)−1 =
|sI−A|

E0+. . .+En−1 sn−1

=
a0+. . .+an−1 sn−1+sn
in which0 En−1they are real matrices nxn. From the previous relation it follows
that the inverse of (λI−A) is a matrix (nxn) whose elements are ratios of
polynomials in λ with real coefficients, where the degree of the polynomial in the numerator
and, in the case of the polynomial in the denominator
Valid for the calculation of E0 ,· · ·En−1and for the calculation of the coefficients of the poly-
characteristic nominee0 ,···, an−1the procedure illustrated below is known
as Soriau's algorithm.

|sI−A|I= (a0 +. . .+s n )I= (sI−A)(E0. . .+E n−1 sn−1 )

and for the identification of the coefficients, we obtain:

En−1I

En−2=AEn−1+an−1 I

En−3=AEn−2+an−2 I
..
.

E0=AE1+a1 I

0 =AE0+a0 I
162 5. The stationary linear representations in the complex domain

The latest report is redundant and can be used for verification.

The discovered relationships define an algorithm particularly useful if one takes into account
present that
an−1=−tr(AE n−1 )
1
an−2=−tr(AE n−2 )
2
..
.
1
aI=− tr(AEI )
n−i
it indicates the trace of a matrix, that is, the sum of the eigenvalues or, that
It is the same, the sum of the elements on the diagonal. For the verification of the
previous equalities it should be remembered that they areithe sum of the powers
The zeros of a polynomial satisfy Newton's relations.

an−1=−s1

2an−2−(s2+an−1s1 )

3an−3=−(s 3+an−1s2+an−2s1 )

...

na0=−(sn +an−1sn−1+· · ·+a1 s1 )

Furthermore, since the eigenvalues of the power of a matrix are the eigen-
the eigenvalues themselves raised to the power, and the sum of the eigenvalues is equal to the trace
of the matrix, it follows that
si=tr(Ai )

It follows that the i−th element of Newton's relations can be expressed in the
following mode

theren−i=−tr(A n−i−1+an−1 An−i−2+· · ·+an−I A)

Infinite the matricesn−ithey satisfy the relationships

En−i=An−i−1+an−1 An−i−2+· · ·+an−i+1A+an−I i = 0, ... n−1

 5.8. Appendix 4 163

From the last two equalities, it is deduced

1
ai=− tr(AEI )
n−i

In conclusione vale la seguente procedura per il calcolo della funzione di

transfer
i=n−1

⇓
EiI

⇓
Ci=AEi

⇓
1
ai= tr(AEi )
n−1
⇓
Ei−1=Ci+ai I

⇓
PRINT(i, E i , ai )

i= 0 ⇒AE+a I 0Stop
0 =
0 ∼

⇓
i = i - 1 ⇒ ITERA

If less than or equal to 10, good results are generally obtained.

subsection Appendix 9
 164 5. The stationary linear representations in the complex domain

5.8.e. The Z Transform

The Z-transform, which associates with a discrete time function defined

pert≥0,f(t), a complex variable function, F(z) = Z[f(t)], and one
a tool of great utility in the study of stationary linear systems over time
discrete. The study of the system in the complex domain not only allows for
with computational simplifications, as we will have the opportunity to verify, but
important interpretative insights on the physical behavior of the system.
This appendix is dedicated to introducing this mathematical tool,
its main properties and some elementary transformations.
The Z-transform, associated with a function f(t) defined for t ≥ 0, is
∞
f(1)f(2)
Z[f(t)] = f ( t ) z −t=f(0) + + 2 +. . .=F(z)
(t=0 z z

ρf>|z| orfndeidfeebalivroacm
xehplt ofoniuntcfasi tI
(ρfradius of convergence). The class of functions for which such an operation of
The transformation is defined and extremely broad, and it is indeed intuitive to understand.
It is said that for sufficiently large values of the modulus, the series is convergent.
Being the samples of the function multiplied by z. t

The analogy with the Laplace transform is evident.

∞
L[f(t)] = ! f ( t ) e −stdt=F(s)
0

defined for Re(s) > α(α

f fabscissa of convergence
In the transform, the values of the function are multiplied by the power
in z1 the Laplace transform, the values of the function are multiplied by
the exponential of -s. This is a correspondence already noted in the
study of natural ways and considerations regarding the convergence of laws
di moto.

Very often, to indicate that the function which is being transformed is defined
pert≥0 if you use the convention to represent the function conf(t)δ−1(t)
 5.8. Appendix 4 165

being−1the unit step, the even function equals one for t≥0. A prime
The aspect we consider is the calculation for elementary functions. Among these
let's consider the temporal expressions of natural modes that, as we know,
characterize the explicit description of the system. The law of motion of a
aperiodic or alternating mode of type λtand if there is

λ λ2 1 z
Z[λ t δ−1 (t)] = Z[λt ] = 1 + + 2 +. . .= λ =
z z 1− z
z−λ

defined by
λ
- z -<1⇔|z|>|λ|
- -
- -
- -
The verification is simple if one observes that the function in question is the sum of
a geometric series with reason and λzfor a geometric series with ratio

S = 1 + a + a2+a3+. .. (geometric series)

post
Sn= 1 + a + a2+a3+...+an

yes
1−an+1
S n=
1−a
and therefore

1-an+1 1
S=S∞ = limSn= lim = |a|<1
n approaches infinity 1−a
n approaches infinity 1−a

A special case is represented by the unit step. It is obviously

z
z[δ −1 (t)] = |z|>1
z−1

As a second example, consider a periodic temporal law.

scent= ejθt-−jθt
2j
166 5. The stationary linear representations in the complex domain

With simple steps you obtain

jθt −jθt 1
Z[sinθt] = Z " e −e jθt −jθt
#= 2j Z[e ]−Z[e ] =
2j

1 z z 2 −jθ 2 jθ
= ) − z-e−jθ * = 1z -ze -z +ze =
2j z -ejθ 2j z −z(e +e ) + 1
2 jθ −jθ

zsenθ
Z[senθt] = |z|>1
z 2-2zcosθ +1
where the property of linearity was used.

Is this a fundamental property of the Z-transform that formally we

express
Z[k 1 f1 (t) +k2 f2 (t)] =k1 F1 (z) +k2 F2 (z) For all k1 , k2

ρf=max{ρf,ρf} 1 2

Further properties of the Z-transform that will now be highlighted are

the following
Translation (to the right and to the left)
Convolution
-Valore iniziale e finale(nel tempo)

Right translation
1
Z[f(t−1)] = F(z)
z
In fact, from the definition:
∞
f(0)f(1)
Z[f(t−1)] = f ( t − 1 ) z −t = + 2 +. . .
(t=0 z z

it can be calculated

f(0)f(1)
zZ[f(t−1)] =z ) *
+ 2 +. . . =F(z)
z z
 5.8. Appendix 4 167

and therefore
1
Z[f(t−1)] = F(z)
z
that is, shifting to the right by one step in time is equivalent to multiplying by in 1zthe
domain of the complex variable.

Left translation

Z[f(t+ 1)] = zF(z) - zf(0)

In fact, from the definition

f(2)f(3)
Z[f(t+ 1)] = f(1) + + 2 +. . .
z z
it can be calculated

1 f(1)f(2)f(3)
Z[f(t+ 1)] =−f(0) + f(0) + 2 + +. . .
z z z z3
and therefore

Z[f(t + 1)] = zF(z) - zf(0)

Note the analogy with the theorem of differentiation in the Laplace domain.

Lf(t)˙
+ =, sF(s)−f(0)

Convolution
t
Z" f(τ)g(t−τ) =
# F(z)·G(z)
(
t=0

which is expressed by saying that convolution in the time domain corresponds

to the product in the domain of transforms. In fact

t
Z) f(τ)g(t−τ) =
,
(
t=0
168 5. The stationary linear representations in the complex domain

1 1
=f(0)·g(0)+f(0)·g(1)+f(1)·g(0) + f(0)g(2)+f(1)g(1)+f(2)g(0) + . .,.
z+ , z2 +
g(1)g(2) f(1) g(1)g(2)
=f(0)g(0)
+ + + 2 +. . . +, g(0) + + 2 +. . . ,
z z z + z z
f(2) g(1)
+ + g(0) + +. . . ,
z 2 z
f(1) f(2)
=f(0)G(z) + G(z) + G(z). . .
z z2
f(1)f(2)
=f(0)
+ + + 2 G(z) =, F(z) · G(z)
z z

It should be noted that this property involves a simplification in the calculation of

response, indeed the forced response in the state.

t−1
H(t−τ)u(τ)
(
τ=0

and parallel convolution in the field of time translates into the product of
transform yourself.

Initial value theorem

f(0) = limF(z)
|z|→∞

In fact
f(1)f(2)
F(z) = f(0) + + 2 +. . .
z z
e
limF(z) = f(0)
|z|→∞

final value theorem

f(∞) = lim(z−1)F(z)
z→1
 5.8. Appendix 4 169

This result can be demonstrated with the following calculations.

N
Zf(t+
+ 1)−f(t) = zF(z)−zf(0)−F(z)
, = lim f(t+ 1)−f(t)z−t
N→∞ (t=0 & '
and so
N
limzF(z)−zf(0)−F(z) = lim lim f(t + 1) - f(z)z−t
z→1 & ' z→1 N→∞ (t=0 & '
that is
N
lim(z−1)F(z)−f(0) = lim f(t+ 1)−f(t) = f(∞)−f(0)
z→1 (t=0 +infinity
N approaches ,

In conclusion:
lim(z−1)F(z)−f(0) =f(∞)−f(0)
z→1

To conclude, let's calculate the Z-transform of [Link] calculation follows the steps
turned in the case of scaling
A A2 A3
Z[A t ] =I+ + 2 + 3 +. . .
z z z
and place
A An
Sn=I+ + + . . .+
z zn ,
results
A A A An An+1
+ I− z,
Sn=I− +
z ,+
I+ + . . .+
z zn ,
I−
z n+1
that is
-1
A An+1
SnI− + + I−
z, z n+1 ,
and, in conclusion
-1 -1
A zI−A
Z[A t ] =limn⇒∞ Sn=I− + = + =z(zI-A)−1
z, z ,
which converges for

A
@z <1⇒|z|>∥A∥
@
@ @
@ @
@ @
6. The geometric properties of
state space

This chapter introduces and studies two properties of states

that allow to highlight the internal structure of the system.
The first property pertains to the state-output relationship and consists of the existence
of states which correspond to coincident outputs. As intuition suggests,
and we will have the opportunity to understand better, from an external point of view this
It constitutes an obstacle to the reconstruction of internal evolutions starting from
from the observations (unobservability of the system).
The second concerns the entry state behavior and more specifically
la possibilita` di raggiungere prefissati stati. Da un punto di vista esterno ci`o
express the possibility of modifying the dynamic behavior of part of the
system.

6.1. The properties of the internal structure: Observability

The property of observability concerns the state-output behavior; more

precisely this property takes into account the fact that all the information
Regarding the internal evolution in the state, it should be included in the report. How
172 6. The geometric properties of the state space

The intuition suggests this should allow for reconstructing behavior

in the state from the observation of the exit. As we will see in the following
characterization of the most intuitive hindrance to the reconstruction of the behavior
internal treatment from the observations indicates the way and naturally leads to the
solution. The starting point is represented by the definition of indis-
Tinguibili; observability in fact materializes in the absence of indistinguishable states.
In fact, indistinguishable states, different states corresponding to the same us-
for each entry, represent evident limitations of information for
reconstruct internal behavior (redundancy in the description of the state
the ineffectiveness of observation channels
Two staticaexb they are indistinguishable at the time0if for each function
of entry data0from then on, the corresponding outputs are coincident. The presence
of indistinguishable states can be seen as an ineffectiveness of observation channels.
consideration, if we think about the possibility of reconstructing internal behavior in
starting from the observation of the exit.
The presence of indistinguishable states is equivalent, for stationary linear systems,
in the presence of unobservable states. They are called those states to which
it corresponds to a freely evolving state that is identically null (indistinguishable states
gauge from the zero state). In fact, if for every fixed input function u(·)
the two exits
t
ya (t) =CeAtxa+ ! This A(t−τ ) Bu(τ)dτ
0
e
t
yb (t) =CeAtxb+ ! This A(t−τ ) Bu(τ)dτ
0

they coincide, necessarily

This At(xa-xb ) = 0

So, xa`e in d is t in g u ib ile d a xbit implies that (xa -x b ) it goes out identically
Nothing is in free evolution, that is to say, it is unobservable.
It is easy to show that the inverse also holds; in fact, if there exists an observation x¯
able, then ∀xa , xbthat
xa-xbx bar
 6.1. The properties of the internal structure: Observability 173

xaindistinguishableb
Therefore, it is enough to study the existence of statix¯ that give free evolution nothing.
in output, unobservable states, to obtain indistinguishable states.
The questions to which we want to respond are:
Under what conditions is an assigned state x¯ observable?
How to characterize the set of unobservable states?
Sotto quali condizioni il sistema`e tutto osservabile (assenza di stati inosserv-
able) ?
If we denote with I the set of unobservable states, a first result
the following is important.
The set of unobservable states

I = x ∈ RnThisAtx = 0, for all t ≥ 0

" #
coincides with the set
C C
⎛ CA ⎞ ⎛ CA ⎞
I≡ker .. x * in
is R:n .. x=0+
⎜ ⎟
. ⎜ . ⎟
⎜ CA n−1 ⎟ ⎜ CA n−1 ⎟
⎝ ⎠ ⎝ ⎠
In fact, if
dk
This Atx = 0 ⇒ , This Atx- =0
dtk
t=0
mom
dk
This Atx=CAk eAtx
dtk
and so:

dk At
This xt=0CAk eAtt=0x=CAk x k = 0, 1, ...
dtk . .
. .
That is ifAtx = 0 then CAk x = 0 for each flame, and in particular
C
⎛ CA ⎞
.. x=0
⎜ ⎟.
⎜ CA n−1 ⎟
⎝ ⎠
 174 6. The geometric properties of state space

Or vice versa,
CA k x = 0 k= 0, . . . , n−1
from the Cayley-Hamilton theorem it follows that:

CA k x = 0 For all k, k is greater than or equal to n

In fact, for Cayley-Hamilton's theorem: A n-a0 I−a 1 A . . .−a n−1 An−1

and if this expression is premultiplied by it is post-multiplied by x:
CA n x = -a0 Cx−a1 CAx...-a n−1 CA n−1 x. The same calculation can be
repeated by iterating the procedure to show that the generic powerCAkyes
can be expressed as a linear combination of the first (n−1) powers and leads
to the desired result.
But if CAk x= 0 For all k≥0, then:

tk
This Atx= CA k x= 0
/
k≥0
k!

The exposed shows the following result

Theorem 1. The set of unobservable elements coincides with

C
⎛ CA ⎞
ker ..
⎜ ⎟
.
⎜ CA n−1 ⎟
⎝ ⎠

The set of unobservable states is therefore a subspace of R.n .

The matrix (nq×n),
C
⎛ CA ⎞
..
⎜ . ⎟
⎜ CA n−1 ⎟
⎝ ⎠
.ytilpbaicngiom
reteronitm
estys'The
 6.1. The properties of the internal structure: Observability 175

Returning to the previous questions, it is now clear that it is observable whatever.

C
⎛ .. ⎞0= ̅x
.
⎝ CA n−1 ⎠
Moreover, the set of unobservable states is the subspace of [Link] by
C
⎛ CA ⎞
I=ker ..
⎜ ⎟.
⎜ CA n−1 ⎟
⎝ ⎠
The system is fully observable if and only if there are no unobservable states.
that is, the observability matrix has full rank
C
⎛ CA ⎞
ρ .. =n
⎜ . ⎟
⎜ CA n−1 ⎟
⎝ ⎠
The set of unobservable states is a fixed possible subspace.
highlight an internal decomposition into subsystems that takes into account
of such property of the states.
To that end, let us suppose that

C
⎛ CA ⎞
ρ= .. m<n
⎜ ⎟.
⎜ CA n−1 ⎟
⎝ ⎠
if this is true it can be demonstrated that:

There exists a set T.n×nz =Tx

and, in these new coordinates, the matrices that represent the system hands-
making a block partition of the following type:

A 0 B1
TAT −1 = 0 11 1 TB= 0 B2 1 C T −1=10 )
A21A22
176 6. The geometric properties of the state space

with

A11(m×m) B1(m×p) C1(q×m)

Furthermore, the triplet11 B1 , C1the observable status.

The tale structure corresponds to the internal decomposition that highlights

the property of the system. If we consider the partition diz

m
z1
z=Tx 0 z2 1
m−n

ż1=A11 z1+B1 u
⎧
⎪⎨ ż2=A21 z1+A22 z2+B2 u
y=C1 z1
⎪⎩

and it is evident the composition in two under system is1 ,S2with state variables
z1this2 each of the subsystems is influenced by the subsystem S1no
the influenza today2while influences2the exit is characterized only by the
subsystemS1 .
So in these coordinates the internal structure is that of the figure
 6.1. The properties of the internal structure: Observability 177

z 10

S
z 1
S 1

u
z y
2
S 2

z 20

Figure 6.1

where it appears that only the subsystemS1influence the exit. Moreover, since the
ternaA11, B1 C1S available as to t u t e1it seems to be in a state of use.
this breakdown clearly highlights the internal structure of the system, which is
the interconnection of the two subsystems, one fully observable and the other not
The observable has a dimension equal to the rank of the observability matrix.

The coordinate transformation that highlights the structure in-

it dictates that corresponds to taking as ultimin-mvectors of the
new base a base of the set of the unobservable and the primimun its gener-
ico completion. It is assumed that

m n−m
..
T −1 = . base in I 7
6 ∗
178 6. The geometric properties of state space

An example is the following

1
A= 0 -2 C=(01)
0−1 1

C 0 1
0 CA 1 = 0 0−1 1

0 1 1
0 0−1 1
x= 0 implies I=gen 80 1 9
0

and choosing
..
T −1 = , 0.1 -
.
1.0 .
TAT matrices are obtained−1eCT−1of the desired form.
Another aspect that is wanted to be highlighted is that the bond
forced input-output is characterized by the only observable subsystem. This
It is evident from the figure, but it is directly obtained through calculation. We have
in fact
¯ eAt¯ B¯
W ( t ) = C eAt B = C
k
t k A11 0 B1
(C10 ) 1 0 B2 1 =
/ k! 0 A21A22
t k A11
k 0 B1
= 10 ) 0 k 10 B 1 =
/ k! ∗ A22 2

=C1 eA tB111

All and only the observable modes characterize the observable subsystem.
In fact, if we remember that the observable modes are those whose laws are...
they belong to AceAt, that is λoi

¯ At¯ (C1 eA
Ce= 11 t
0) =ψ(t)

and that it was only a coordinate transformation z = Tx, and therefore

the eigenvalues λiwhich characterized are the same that characterize the
 6.2. The properties of the internal structure: Accessibility 179

TATMatrix−1 in reality, this is enough to understand that the subsystemS1

It is characterized by temporal legacies that are those of observable modes.
it could also demonstrate in this case that if a system is everything or is observed -
Able then, and only then, all the ways are observable.

6.2. The properties of the internal structure: Accessibility

Definition of deductible earnings in tax0set0 And an entrance

on the time interval [t0 The door carries the status0ax.
Starting from this definition, the investigation that will be conducted focuses on the
development of methods that allow to understand if a fixed state can be rag-
reached independently from calculating the input that allows to reach it;
to characterize the entire set of states that can be reached; the con-
conditions under which the system is entirely accessible.
In the case of a stationary linear representation, it is common to refer to the
statox(t0 ) = 0. The zero state is indeed always maintained in the absence of
input (state of equilibrium).
If we denote with R(T) the set of reachable at T, we have

T
*
R(T) = x:x= ! eA(T−τ ) Bu(τ)dτ, for someone0 , u [t0 ,T) +
t0

which can also be described as

T
*
R(T) = x:x= ! eA(T−τ ) Bu(τ)dτ, u(−∞,T ) +
−∞

From the previous expressions, it is clear that linear combinations of states

reachable are reachable, and therefore that R(T) is a subspace of Rn .

Per arrivare ad una descrizione dell’insieme degli stati raggiungibili che ne

To allow the calculation, it is necessary to refer to a known result of the
theory of linear operators.
180 6. The geometric properties of the state space

It is necessary to preliminarily recall the definition of an operator ag-

Arrived. Assigned L: A → B linear, L:∗ B → A and that operator that
satisfies
<L(a), b >B =< a,L(b)>A∗
where <·,·> indicates the operation of inner product.

Lemma 1. Select a linear operator

∗ ⊥
R(L) = Ker(L)
that is, its image coincides with the orthogonal complement of the kernel of the operator
∗
addedL.

∗ ⊥
Having stated what is asserted in the lemma, being R(L) and Ker(L) are...
tospazi, is equivalent to

∗
R(L)⊥ =Ker(L)
what is true in that nonetheless presobinR(L)⊥ and for eachainA

< b,L(a)>B = 0⇒<L(b), a >∗A = 0

∗
cio´eb`e inKer(L). Similarly, one proceeds for the vice versa.

The result now recalled can be applied to the case under study by observing
that the function
∗

B ∗ eA(T−τ ) x:Rn →Cω (−∞,T)

the additional operator of
T

! eA(T−τ ) Bu(τ)dτ
−∞

It is also observed that the adjoint operator of a matrix is the matrix ag-
matrix (which in the real case coincides with the transpose) and it is observed that for the
previous theorem theKerCeAtcoincides with the kernel of the observation matrix
Ability, it is not difficult to verify that the following result holds.
 6.2. The properties of the internal structure: Reachability 181

Theorem 2.

T ⊥
∗

R(T) = R , ! - B∗ eA(T−τ )
eA(T−τ ) Bu(τ)dτ=Ker x =
−∞ 6 7
⊥
B∗
⎛ B ∗ A∗ ⎞
Ker .. =R(B AB· · · An−1 B)
⎜ . n−1 ⎟
⎜ B ∗ A∗ ⎟⎠
⎝

The previous considerations lead to a simple and complete characterization.

characterization of the set of reachable states. It coincides with the sub-
space generated by the linearly independent columns of the matrix, (n×
np), ( B AB· · · An−1 B) this reachability matrix.
If this matrix has rank, all states are reachable and the system
same said completely achievable. The highlighted result for-
a simple and very effective test arises. Given a system, it is indeed possible
with a simple criterion to investigate the "potential" for intervention through the
input channels. The set of reachable states is a subspace of the
state space; it coincides with the subspace generated by the linear columns-
independent minds of the reachability matrix.
A numerical example:

1−2 -1
00 =A 1 =B
31 0
-1

..
R=Im ⎛ -1 . 1 ⎞
..
⎝ -1 . 3 ⎠
In this case, the system is fully accessible and therefore one can go anywhere.
any state.
The verification of the reachability of a state has become very simple, because
In reality, given a state: is x reachable? It was necessary to see if x ∈ R (if not.
182 6. The geometric properties of state space

it belongs to Rnon and cannot be reached.

Another example:
1−2 -1
00 =A =B
31 0 01

R=Im -1-1
0 0 01
in this case the set of reachable states is a space of one dimension
sione, this means that for this system all and only the states of the type:

α
0 01 =xR

they are reachable and only these.

Therefore, the set of reachable states is a subspace, and it is exactly
The image of the considered matrix can be said that the system is all rag-
reachable if the rank of the matrix is.

You have thus answered the three initial questions.

How to determine if a state isRis it possible to arrange?

the r aggregating ibles sep uo es ser erees press o come combination linear of the columns
diR.

How is the set of reachable states made?

´It is the linear combination of the columns of R.

When is the system all reachable?

When in the set of columns there are independent ones, that is when
The rank of the matrix is equal to the determinant.

The reachability is differential for continuous time systems

R=Im .. ..
n−1
6 B.. . . .A B7

∥
 6.2. The properties of the internal structure: Accessibility 183

¯
R(n)xR∋t≤n
For a continuous-time system, S, entirely reachable if we denote with x Rhe
generic state, this can be achieved starting from, statox(0) = 0 in
an arbitrary time, T. In fact, it can be verified that with the entry
′
uR (t) = B′ e−AtC −1 e−ATxR

in which
T
′
C= ! e−AtBB ′ e−Atdt
0

and in ver t ib le n e ll’ip o t e s i d i c o m p le t a r a g g iu n g ib ilita, we arrive at the timeTallo

statoxR In fact
T
′

! eA(t−τ ) BB ′ e−Aτ C −1 e−ATxR dτ

0

T
′
eAT ( ! e−Aτ BB ′ e−Aτ dτ)C−1 e−ATxR
0

quella tra le parentesi`e la matriceCe si ottiene

eATe−ATxR=xR

therefore the state reached at time Ts under the action of the indicated input is
proprioxR .
Being arbitrary, it is said that the property of reachability is different -
exercises for continuous time systems. Obviously, the less time it takes to
go from 0 axRThe entrance to be applied will be wider.
Another observation concerns the fact that the states reachable from
zeros can also be forced to zero (controllability). A state, x, is
controllable aTse there exists an input that brings it to zero at time T. For
it can be shown that linear and stationary representations are controllable
all and only the reachable states.
If indeed a state, xTand reachable at T, being the set of states
reachable a subspace of the state space invariant with respect to A,
184 6. The geometric properties of the state space

even the state−e ATxTit is dragging; when you are in a bad mood,Ttale
what
T
eATxT+ ! eA(T−τ ) ThisT(τ)d(τ) = 0
0

which proves the controllability aTdixT.

The following considerations aim to highlight how the
the property of reachability is important in the project.
m

θ
l

u
Figure 6.2

The interest of the study conducted is often related to equivalence now.

shown. The possibility of bringing a system to the zero state, the control-
instability is often the goal of an intervention on a physical system. The simple
The test developed can be used for this purpose. It could be
furthermore, to show that the property of controllability is maintained under approximation
linear mapping. This implies that if the property does not hold on the model
linear approximating the actual behavior, then there is no existence on the sis-
the same physical theme. Consider, by way of example, an inverted pendulum on
an uncontrolled mass at the base due to the action of the force (see the figure).
θ is the angle formed by the pendulum with the vertical; if the forces are projected
on the tangent of the trajectory of the pendulum and the equilibrium of forces is imposed
agents, the equation of motion is obtained:

¨
mgsinθ = mlθ + ucosθ

The problem being considered is the maintenance of the pendulum in equilibrium. It

it essentially deals with a state controllability problem
 6.2. The properties of the internal structure: Reachability 185

at the vertical position with zero speed.

θ
0 θ̇1 = 0

The study can be conducted on the linearized model, to see initially

instance, if the linearized system is controllable. In fact, it can be shown that
If this were not the case, that is to say, if it were not possible to bring the state to zero
in its linearized model, certainly one could not solve the problem on
non-linear model and therefore in reality on the effective structure.
Starting from the representation with the system state

x1=θ
*
x2=θ ˙

=
xẋ 1 2
⎧ g 1
⎨ cosx1 u
dx2ot= sin x1−
I ml
⎩
calculate the linear approximation and the set of controllable states.

0 1 0
A= , g - . =J(xe ) B= , 1 cosx - ..
cosx10. . − ml 1
l . xe =0 . xe =0
. .
.
0 1 0
˙ , g
ξ= - ξ+ , 1 - u
0 − ml
l
0 1
− ml ⎞
R= = ⎛
6B AB 7 1 0
⎝ − ml ⎠
This matrix has a rank of two, so the set of reachable states is everything.
space, therefore also the zero state with zero angular velocity is controllable.
What implies the zero controllability of the physical system for small displacements
thoughts around balance (for more insight, refer to what has been said)
with stability analysis through linearization).
186 6. The geometric properties of the state space

The same topics can be used to study the structure, known

with the name of inverted double pendulum, shown in the following figure

u
Figure 6.3

The study of linear approximation allows us to verify, in this

case, the incomplete controllability that could be shown leads to the
non-controllability of the real system.
Another aspect that is now intended to highlight concerns the
the possibility of choosing the coordinates, and thus representing the system, in a way
so that the composition of the system itself and the interconnection become evident
of a reachable part and an unreachable part.
To this end, suppose that the system is not fully reachable

..
ρ [Link].. ..
6 . . n−1A B 7 =m < n

In such circumstances, it can be demonstrated that

There exists a non-zero |T|, such that z equals Tx:

e, in these new coordinates, the matrices that represent the system

they show a block partitioning of the following type:

A A B1
TAT −1 = 0 11 12 TB= C T −1= (C1C2)
0 A22 1 0 0 1
 6.2. The properties of the internal structure: Reachability 187

with
A11(m×m) B1(m×p) C1→(q×m)
Furthermore, the triplet11 , B1 , C1a ggingittuert
This block structure highlights the fact that the system in
realt`a`e consists of two subsystems, one fully reachable and the other not reachable.
reachable.

z 10

S
u z 1
S 1
+
y
+
z 2
S 2

z 20

Figure 6.4

The structure of the matrices in the new coordinates corresponds to the structure
of the system highlighted in the figure. In fact, to a partition
m
z1
z=Tx= 0 z2 1
n−m

the equations correspond

{"text":"188"} 6. The geometric properties of state space

ż1=A11 z1+A12 z2+B1 u

⎧
⎪⎨ z2.=A22 z2
y=C1 z1+C2 z2
⎪⎩
If we denote with S1s2two dynamic systems, one corresponding to evolution
of the variabilizing1 , the other to those2 , it is noted that1and influence to document
S2and autonomy, and in influencing the city, when the destruction corresponds
the tooth is precisely that of the figure. This structure highlights that the subsonic
themes1the potential to be reached, S2instead no; in reality the subsystem
S1and it is attainable. Because

ρ(B· · ·An−1 B) =m
n−1
11 B1
¯ ¯ ¯ A ¯n−1 B̄) = , B1A11 B1· · · A
ρ(B|AB|... - =m
0 0 ··· 0
with this it is demonstrated that the pair (A11, B1 )and it is all reachable. So if from
given overall system (A, B, C), the triplet (A) is extracted11, B1 , C1 the new
representation thus obtained,

A,B,C(n)
⎧
⎪⎪ T
⎪⎪
⎪⎪ ¯ ¯ ¯
A,B,C(n)
⎪⎨

⎪⎪ ⇓
⎪⎪ A11 B1 C1 (m < n)
⎪⎪
⎪⎩
it is degrading.
As for the calculation of the transformation T
To this end, it is sufficient to choose the prime new coordinates a basis in the set
the reachable ones, and the remaining ones - any, but obviously such as to generate
independent vectors.

base ∀
T −1 = 0 .. 1
R .
m-column n-m
 6.2. The properties of the internal structure: Accessibility 189

Consider the following example:

1 -1
B= 0 1 A= 0
1 0 1 1

..
⎛ 1.−1 ⎞ =gen 1
R=Im .. : 1;
⎝ 1 - 1 ⎠
..
1 0
T −1 = , 1.0 . - T=
0 1
1 .. 1 −1 1

1
TAT −1 = 0 −1−3 TB=
0 -1 1 0 01

dx1ot-x1-3x2+u
⎧
⎪⎨ ẋ−x
2 2

y=C1 x1+C2 x2
⎪⎩
Another observation concerns the fact that the forced link entry - us-
The cited is represented by the only reachable subsystem. This is also what
suggests the intuition in observing the figure. The calculation of W(t) confirms
such intuition. In fact
¯
¯ , ¯B ,¯ C ) ⇒
W ( t ) = C eAt B = ( A ¯ At
C e¯ B

ma
∞ k
t k A11 A12 B1
(C1 , C2 ) =
/
k=0
k! 0 0 A22 1 0 0 1
∞ k
t k A11
k
∗ B1
(C1 , C2 ) 0 k 1 0 0 1 =
/ k! 0 A22
k=0

(C1 eA tB111) =W(t) =CeAtB

and this is the impulsive response of the subsystem S1 .

Another observation is also suggested by the figure; the subsystem rag-
reachable should be characterized only by excitable modes. This fact
190 6. The geometric properties of the state space

establishes a precise link between the reachability of the states and their excitability
natural methods.
As already mentioned, they are excitable with input impulses those natural modes whose
exponential laws appear in:

H(t) = e AT B

If one takes into account the fact that the exponential laws that characterize the
two subsystems1eS2 what inTAT1 they are the same as those of the matrix
A
A A
A→λI TAT −1 = 0 11 12
0 A22 1 →λI
It is understood that the eigenvalues of A are partly in A.11 , in number dim, e
part inA22 , in number of (n−m); furthermore from the calculation

¯ eA 11tB1
H(t) = e AtB ¯= 0 1
0

cio`e inH(t) ci stanno soltanto le leggi esponenziali che stanno ineA tB1 11

those that correspond to the eigenvalues of A11 , that is to say those that correspond
to the subsystems1 It could be shown that since the subsystem
S1everything is reachable all its ways are excitable. This allows to affir-
but the reachable subsystem is characterized by all and only the modes
excitable.
A final observation is needed to clarify what is meant when one states
that the properties of reachability and observability are dual to each other.
This stems from the fact that if one considers the two systems

A
=ẋ+
B
xu
*
y=Cx

̇′
=zAx+C′u
*
y=B′ u
the duality of each other, the complete reachability of one is equivalent
the complete observability of the other and vice versa. Therefore, the existence of a
 6.3. The Kalman decomposition 191

Property data on one of them implies the existence of dual property on the
dual system.

6.3. The Kalman Decomposition

In this lesson it will be shown how based on the properties of unobservability

abilities and reachability can be made possible by a change of coordinates,
highlight the internal structure of the system resulting from the interconnection
of four subsystems for which the two properties are alternatively true
certificate. It generalizes what has already been studied in the previous lessons.
In fact, with reference to accessibility

R=Im(B ··· An−1 B)

If the system is not entirely reachable, we know that a transformation exists.

in coordinates, T, such that

A A B1
TAT −1 = 0 11 12 TB=
0 A22 1 0 0 1

e, to this matrix structure corresponds a decomposition into two sub-

Systemisto1eS2 which highlights how the subsystemS2do not be
reachable.
With reference to unobservability

C
⎛ CA ⎞
I=ker ..
⎜ .⎟
⎜ CA n−1 ⎟
⎝ ⎠

if the system is not fully observable, we know that there exists a transformation
in coordinates, T, such that

A 0
TAT −1 = 0 11 1 C T −1=10 )
A21A22
192 6. The geometric properties of state space

and this matrix structure corresponds to a decomposition of the system

in two subsystems1eS2 both potentially modifiable by the entry,
but the systemS2not observable in output.

Regarding the two properties, reachability and observability, we can

classify the states into:

reachable and non-observable

reachable and observable

not reachable and not observable

unreachable and observable

It will be shown below that it is possible to select the coordinates in such a way that
The properties of the states should be highlighted.

The Kalman decomposition

It is a decomposition into four subsystems each characterized

generated from a state space obtained from the linear subspaces of
reachable states and unobservable states. Let χ1 ,χ2 ,χ3eχ4the subspaces
from the state spacen , defined as follows

χ1=R∩Reachable and unobservable states

χ2 :χ1⊕χ2Reachable and observable states

χ3 :χ1⊕χ3Unreachable and unobservable states

χ2and is a complement to the space1inRche has in common with

χ1only the origin, this allows us to express through the operation of addition
live. The same consideration applies to χ.3 .
With the three subspaces introduced in the case of the figure below, everything is generated.
how much the state space, in this case the spaceR3 .
 6.3. The Kalman decomposition 193

3
X = R
I

Figure 6.5

In general, this is not the situation, so it is necessary to introduce a fourth.

subspace, χ4 , aR complementn , that is how much longer is needed to obtain
everythingn :

χ4 :χ1⊕χ2⊕χ3⊕χ4=Rnstates not reachable and observable

194 6. The geometric properties of the state space

At this point, a coordinate transformation that corresponds should be performed.

to take on as primin1vectors a base in χ1 coconut dessert2vectors that
they define a base in χ2 , poi,n3vectors that define a basis in χ3 ,
infinite, n4=n−n 1-n2-n3vectors
that define a basis in χ4 .
This corresponds to setting a transformation in T: in which the T−1the columns
the selected vectors.

T −1(baseχ1 |baseχ 2 baseχ3 |baseχ 4 )

How does the representation appear in these new coordinates? Place
z1
⎛ ⎞
z2
⎜ ⎟
z=Tx= ⎜ ⎟
⎜ ⎟
⎜ z3 ⎟
⎜ ⎟
⎜ ⎟
⎜ ⎟
⎝ 4⎠
z
sheepithe dimensionsi , if there is the following structure of the matrices A, B, and C.
A11A12A13A14 B1
⎛ 0 A22 0 A24 ⎞ ⎛ B2 ⎞
TAT −1 = TB=
0 0 A33A34 0
⎜ ⎟ ⎜ ⎟
⎝ 0 0 0 A44 ⎠ ⎝ 0 ⎠
C T −10 C20C4)
to which the system of equations corresponds:
ż1=A11 z1+A12 z2+A13 z3+A14 z4+B1 u
ż2 =A22 z2+A24 z4+B2 u
ż3=A33 z3+A34 z4
ż4=A44 z4
y=C2 z2+C4 z4
If we indicate with Sithe subsystem characterized by the variables inzi , not
it results in the decomposition into four subsystems (S1 , S2 , S3 , S4 Messi at home
denotes in the following figure.
 6.3. The Kalman decomposition 195

z S
u 1
S 1

z 2
S 2

+
z 3
S 3 y
+

z 4
S 4

z
Figura 6.6

The subsystemS1and influence today2 , daS3and also toS4and infinite from the entrance
u. The subsystemS2and influenza today4 , and from the entry. The subsystemS3
and influence today4 , edS4The autonomous modal evolves into the method of achievements.
The output is conditioned by S2eS4 .
In conclusion, this is the structure that appears from the outside in the coordinates.
S1It is a system that can be reached, but it is not servable;2 the suns are shining
theme characterized by reachable and observable states; The subsystem S3no
reachable and unobservable; finally, S4it is a non-reachable system
but observable.
It appears from this decomposition that well manifests the internal structure of the
196 6. The geometric properties of the state space

state, that the forced connection input output is characterized by only the sub-dose
themes2 .

This can be verified directly with the calculation. In fact, in the coordinates
z
W ( t ) = C e AtB=C2 eA tB222
and this for the block structure of matrices. Therefore, the impulse response
of the overall system is the response of the subsystem S2what is a system
everything is reachable and observable.

It is easily understood, and it would not be difficult to prove, that the trio
(A22 , B2 , C2 ), the one that characterizes the subsystem S2 , has the largest dimensions
small among all the triples of matrices that have fixed impulse response and
equal to W(t). Taking advantage of the connections that exist between property of rag-
reachability and observability of states, and excitability and observability of modes,
despite the considerations made regarding the implementation, the subsystem
S2extract a minimal achievement from a possible realization.

6.3.a. An example of application

Consider the bridge circuit powered by a current generator indicated

in figure
+
R C
i
y
C
R C
-
Figure 6.7

Equations that govern the circuit can be obtained by imposing equilibrium.

all currents in the nodes AeBe in loops 1 and 2

A u = iR1+iC 1 C1 V˙C=−i
1 R+u 1
 6.3. The Kalman decomposition 197

1 VC1=VR+V1C⇒VR=V 3
C+VC1 1 3

VR1
iR 1 = = VC1−VC 3

R1 R1
VC1 VC3
C1 V˙C=−
1 + +u
R1 R1

˙ 1 VC1 VC3 u
→VC=− + +
C1 R1 C1 R1 C1
+
+ i +
i v
i
- +v - -
- y
+
i v
- + -
Figure 6.8

Assuming the voltage across the capacitor C as state variables1 ,

of the capacitor C2eC3 .

B iR=i
2 C+u 2

˙ 2
C2 VC=iR −u 2
VR2
VC3=VR+V2C 2 ⇒ = VC3-VC 2

R2 R2
VC2 VC3
C2 V˙C=− +
2
R2 R2 -u

˙ 2 VC2 VC3 u
→VC=− +
C2 R2 C2 R2 − C2
Infinite perx3you can write the current balance at node C:

C iR=i
1 C+iR 3 2

i C3 =−i+iR2 R1
198 6. The geometric properties of state space

˙
-iR+i2R=C3 V1 C 3

VR1
iR 1 = = VC1-VC 3

R1 R1

iR 2 = VC3-VC 2

R2

˙ VC1 VC2 VC3 VC3

−iR+i2 R=C3 V
1 3 = + − R1 − R2
C R1 R2

VC1 VC2 VC3 VC3

V̇C3 = +
C3 R1 C3 R2 − C3 R1 − C3 R2

˙ 3 VC1 VC2 VC3 1 1

→VC=− + , + -
C3 R1 C3 R2 − C3R1 R2

In conclusion, a linear, stationary system of dimension three is obtained:

VC1 VC3 u
V =− + +
⎧ C1 C1 R1 C1 R1 C1
⎪⎪ VC2 VC3 u
⎪⎪ V =−
⎪⎪ C2 +
⎪⎨ C2 R2 C2 R2 − C2
VC1 VC2 VC3 1 1
⎪⎪ VC3 = + , + -
⎪⎪ C3 R1 C3 R2 − C3R1 R2
⎪⎪
⎪⎩

y=VC-V1 C 2

described, therefore by the matrices

1 1
− R1 C1 0 R1 C1
⎛ 0 1 1 ⎞
− R2 C2 R2 C2
A=
⎜ ⎟
⎜ 1 1 1
, 1
+ 1
- ⎟⎟
⎜ R1 C3 R2 C3 − C3 R1 R2
⎝ ⎠
1
C1
C= ( 1 −1 0 ) B= ⎛ − C12 ⎞
⎝ 0 ⎠
 6.4. The properties of the state space 199

The reachability matrix is

1 1
C1 −R1 C21 ...
⎛ 1 1 ⎞
R=Im − C2 R2 C22 ...
⎜ 0 1 1
. . . ⎟⎠
⎝ R1 C1 C3 − R2 C2 C3
and, for example, it is assumed
C1=C2 R1=R2
it is obtained
..1 1
. −
⎛ C1 ⎞ R1 C21
R=Im 1 1
..
⎜ − C2 .⎟
⎜ R2 C 2
⎟
⎝ 0 0 0⎠
in which the third column, not shown here, is proportional to the previous ones.
So
1
R=gen−1 ⎛ ⎞
⎝ 0 ⎠
Similarly, one can conduct the study regarding observability.
1 -11 0
I=ker ⎛ 1 1 1 ⎞
− R1 C1 R2 C2 R1 C1 − R2 C2
⎝ ⎠
If it is assumed that
R1 C1=R2 C2
the set of unobservable states coincides with the kerdic, thus:
0 1
I=ker(1 −1 0) =gen * ⎛ 0⎞ ⎛ · ⎞ 1 +
⎝ 1⎠ ⎝ ⎠ 0
The calculation of the four subspaces1 ,χ2 ,χ3 ,χ4of the
χ1={0}
⎧
⎪⎪ χ2=R
⎪⎪
⎨ x=I
3

⎪⎪ χ4=0
⎪⎪
⎩
200 6. The geometric properties of state space

6.4. The properties of the state space

6.4.a. Unobservability

The same considerations can be repeated for time systems.

discreet, withAtinstead of dieAtthey have the same results.

6.4.b. Accessibility

The same considerations can be repeated and are simplified.

for discrete-time systems with Atinstead of [Link]
naw
ebsohnathaiptexcerarheB
tut
about the links between controllability and reachability and the "necessary" time to
reaching a state. It is very easy to understand for the recursive structure
of the equations that the set of reachable states in a time interval
fixed, mesh, re B · · · At−1 B , therefore a reachable state can be
< =
obtained0 = 0 at0= 0) in the shortest amount of time.

Equivalence between controllability and reachability only if it is invertible.

Reachability implies controllability, but not vice versa.

Study of accessibility property

Definition:xRit's crying out for a state0at0if there exists a

¯ admission0(t ,t) such
temporary 0 ¯as:

¯ 0 , x0 ,u¯¯ )[t=x
ϕ(t−t 0 ,t) R t¯ < ∞

We will assume that

t0= 0 x0= 0

so:
¯
φ(t,0,u¯ [t 0t̄,) ) =xR
 6.4. The properties of state space 201

To describe the set of reachable states, reference will be made to,

explicit treatment, to the class of discrete-time systems. For the systems at
Continuous time produces, in fact, the same result.
A first result concerns the calculation of the set of reachable states.
A joint is a subspace of the state space and can be easily
calculated from the matrices AaB.
Consider the discrete-time system

x(t + 1) = Ax(t) + Bu(t)

x(0) = 0
x(1) = Bu(0)

In one step, at time 1 there is

x(1) = Bu(0) implies R(1) = ImB

so it can go, varying(0), in all the states that are in the image
diB; still

x(2) = ABu(0) + Bu(1)

by varying (0) I (1) in all possible ways, we obtain all the states that
they are in the image of Be of AB, essentially:

n×p n×p
..
R(2) = Im 6 B . AB 7

If you continue with this reasoning, you realize that

R(k) = Im(B AB . . . Ak−1 B)

In this procedure, it is not necessary to go to infinity, at a certain

One can stop at this point, this by virtue of a theorem; Cayley's theorem.
Hamilton. Based on this theorem, every matrix A is null and of its own.
characteristic polynomial

A d(λ) = |λI − A| ⇒ d(A) = 0

202 6. The geometric properties of the state space

That is, the matrix that is obtained when in the characteristic polynomial to the variable
If a matrix A is replaced, it is equal to the null matrix.

The discussion conducted so far can be repeated for time systems.

discreet with the sole warning to replace Atinstead of dieAtthe results do not
they change.
The only differences concern the time dependency of the reach.
the ability of a state and the corresponding observability of the initial state as follows
as the link between controllability and reachability.
As for accessibility, it is very easy to understand for the
recursive structure of the equations that the set of reachable states in a
fixed time interval, beT,`eR(B· · · AT−1 B), therefore a state
reachable can be "obtained" (dax0= 0 at0= 0) in the most unfortunate of
time. Similar considerations can be made regarding observability
from an initial state based on the observations of the output.

Reachability implies controllability, but not vice versa. How

for example consider the pair of matrices

1 1 1
A= 0 , B = 0 1
1 11 1
that characterize a system certainly not entirely reachable, but control-
unstable. It is indeed evident that the input(0) = x1 (0) +x2 (0) brings the state to
zero in a single moment in time. Even in this case, the two properties are
equivalents are in the solar.
7. Linear input-output models
and representations with the state

7.1. The problem of realization

The problem we will study, the realization, consists in associating with

a matrix of complex variable functions, which is assumed to de-
write an input-output connection, a representation with the state.
The problems addressed so far are mainly aimed at clarifying the ...
sliding starting from a representation. The calculation of responses, also
the forced in the domain of the complex variable are of this type. The
the problem we will address is in a certain sense the opposite: what is the possible relationship
presentation with the state, A, B, C, D, associated with a given function of
complex variable?

This problem is interesting because as we know in representation-

The state is associated with a schema called implementation that allows
to simulate the behavior of the system in real time
 7.1. The problem of implementation 205

We will imagine, therefore, that we have somehow identified the connection.

functional that exists between input and forced output of an assigned process; under
under what conditions can this functional bond be reproduced by the behavior
forced confinement of a linear representation?
The formulation is as follows: given a defined functional link
from the convolution integral with a kernel K(t)
t
y(t) = ! K(t−τ)u(τ)dτ
0

under what conditions does this functional connection coincide with the forced response
of a finite-dimensional, stationary linear system, causal?
Since we know the constrained bond of a linear, stationary system to
finite dimension, characterized by:
t
yf(t) = ! A(t−τ ) B + Dδ(t - τ)u(τ)dτ
" This #
0

the formulated problem admits a solution if the nucleus, K(t), coincides with the
impulse response matrix of a system.

One thus arrives at the following reformulations of the problem in the domain
of time and the complex variable, respectively.

Problem of realization
AssignedK(t) (q×p) find (n,A,B,C,D) such that

This AtB + Dδ(t) = K(t)

Equivalently: assigned K(s) (q×p) find (n,A,B,C,D)

such that
C(sI−A) −1 B+D=K(s)

The quaterna (A, B, C, D) is called realization of K(t) or of K(s).

A condition certainly necessary for feasibility is that K(s)
It is a matrix of proper rational functions. In fact, it has this characteristic.
206 7. Linear input-output models and state representations

the expression of the matrix of transfer functions of a system of-

linear dynamic. This means that if the functional relationship between input - output
is characterized by a nucleus OK(t) whose Laplace transform, K(s), is not
a matrix of proper rational functions, this input-output relationship does not
can be implemented through a linear, stationary system of size
finished.

The question arises spontaneously: is this condition also sufficient?

The condition is also sufficient as shown below.

The constructive proof, that is, it will be shown how starting from a matrix
of proper rational functions, it is possible to construct, by inspection of the coefficients
of the matrices that describe it, a realization of it.
Assume, without loss of generality, that K(s) is a matrix of functions.
strictly proper rational functions are those rational functions when the degree
the numerator is lower in degree than the denominator. It is said without
loss of identity because one can always shift from a proper form to
a strictly proper, in fact, a rational eigenmatrix can always
scrivere come
K(s) = K 0+K(s)
conK0constant matrix eK(s) strictly proper rational matrix. Given
D=K0 the problem of realization reduces to the calculation of din, A, B, C for
a strictly proper rational matrix of the type

B0 +B1 s+. . .+Bn−1 sn−1

K(s) =
a0+a1 s+. . .+an−1 sn−1+sn
love leBIthey are (q×p) matrices of constants.
As an example, the realization of the following nucleus

D K(s)
1 1
s+1 0 s+1
K(s) = " s+1 # = $ %+ $ 1
%
s 1 s

it refers to the realization of a strictly proper rational K(s).

 7.1. The problem of realization 207

Two implementations in canonical form

Various implementations can be obtained directly for is-
collection of the coefficients of K(s) in the given form.

Realization in canonical form achievable

A first so-called canonical form achievable, of
dimensions, indicates the second dimension of Kenil degree of
least common multiple of polynomials in the denominator.

0 I 0 ··· 0 0
⎛ .. .. .. .. ⎞
. . I . . ⎛ ⎞..
.. .
.. ..
AR= ⎜⎜ . . . 0 ⎟
⎟ B=R ⎜ ⎟...
⎜ ⎟ ⎜ ⎟
⎜ 0 0 I ⎟ ⎜ 0⎟
⎜ ······ ⎟ ⎜ ⎟
⎜ ⎜ I⎟
⎝ -a0 I ··· −an−1 I ⎟⎠ ⎝ ⎠

CR= (B0. . . . . . B n−1 )

The identity matrices, I, are matrices that have dimensions (p×p). and, of
consequence, the matricesAR ,BR eCRthey have size (np×np), (np×p)
e (q×np), respectively.
Example:

s 0 1
1 " s+ 1 # " 1# +" #1 s
s+1
K(s) = " 1 # = s2+s =
s s2+s

0 1 0 0 1
AR= " # BR= " 1# CR= " 1
0-1 1#
in fact, it is easy to verify that

CR (sI−AR )−1 BR=K(s)

Another example:
1 0 1 0
" # + " 1 2#
s
K(s) = -1 1
1 + 3s + s2
208 7. Linear input-output models and state representations

in which
p= 2 n= 2

and the realization, of size 4, is

0 0 1 0
0 I ⎛ 0 0 0 1⎞
AR= " # =
-a0 I -a1 I ⎜ -1 0 -3 0⎟
⎜ ⎟
⎝ 0−1 0−3 ⎠

0 0
⎛ 0 0⎞ 1 0 1 0
B=R CR = " #
1 0 -1 1 1 2
⎜ ⎟
⎝ 0 1⎠
The demonstration of the correctness of the procedure followed to calculate
The implementation consists of verifying that the matrix of transfer functions
The measure associated with such representation coincides with the assigned K(s).
In this regard, it is necessary to calculate

T
(sI−AR )a
CR (sI−AR )−1 BR=CR BR
(sI−AR )
-1
s -1 0 ··· 0
⎛ .. ⎞ 0
.. ⎛ ⎞..
0 . . .
⎜ .. ⎟
= (B0 . . . B n−1 ) ⎜ .. ⎟ ⎜ ⎟...
⎜ . . 0 ⎟ ⎜ ⎟
⎜ ⎟ ⎜ 0⎟
⎜ ⎟ ⎜ ⎟
⎜ 0 -1 ⎟ ⎜ 1⎟
⎜ ⎟ ⎝ ⎠
⎜ a0an−2s+an−1 ⎟
⎝ ⎠
We refer to case p=q=1 for simplicity of notation. So
observe, first of all that the calculation of the determinant det(I−A) Rd'a

a0+a1 s+. . .+sn

What is always true for a matrix that has the structure of AR the
The elements of the last row are precisely the coefficients of the characteristic polynomial.
 7.1. The problem of realization 209

The calculation of the transfer function is significantly simplified by the

structure of BRwhich has all zeros and a 1 in the last position. Therefore, it will be enough to
calculate only the last column of the addition - transposed; developing the calculations

∗ ∗ 1 0
⎛ ∗ ∗ s ⎞ ⎛ ⎞..
.. .
(B0. . . B n−1 )
⎜ ∗ ∗ . ⎟ ⎜ ⎟... =
a0+a1 s+. . .+sn ⎜ .. ⎟ ⎜ ⎟
⎜ . ⎟ ⎜ 0⎟
⎜ ∗ ∗ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ 1⎟
⎝ ∗ ∗ sn−1 ⎠ ⎝ ⎠

0
⎛ ⎞..
.
B0+B1 s+. . .+Bn−1 sn−1
$∗ ∗ % ⎜ ⎟... =
a0 +a1 s+. . .+sn ⎜ ⎟
⎜ 0⎟
⎜ ⎟
⎜ 1⎟
⎝ ⎠

and so
B0+B1 s+. . .+Bn−1 sn−1
K(s) =
a0+a1 s+. . .+sn

The implementation/simulation scheme associated with this representation

the registration with the state involves operations of integration, multiplication by
constant and sum as highlighted in the following figure in which it is
subject for simplicity n = 2.

0 1 0
AR= " # BR= " 1# CR= (b0. . . b 1)
-a0−a1

ẋ=x
1 2
⎧
⎪⎨ dx2ot=−a 0 x1−a1 x2+u
y=b0 x1+b1 x2
⎪⎩
210 7. Linear input-output models and state representations

b1

. .
u x2 x2 =x1 x1 + y
+ - ! ! b0
+
-

a1

a2
Figure 7.1

Note that the achievement identified according to this technique has diminished.
sione (np×np); therefore every time the number of inputs is one, in-
depending on the number of outputs, the size of this achievement
ne.
The realization (AR , BR , CR this in canonical form reachable in
how much the state space associated with it is all reachable. In fact, to
start from the expression

0 1 0 0
⎛ .. .. ⎞ ⎛ ⎞...
AR= . . BR =
⎜ 0 1 ⎟ ⎜ 0⎟
⎜ an−1 ⎟⎠ ⎜ 1⎟
⎝ -a0 ··· ⎝ ⎠
CR= (b0. . . b n−1 )
the reachability matrix is easily calculated
0 0 0 1
⎛ .. .. .. ⎞
. . .
(B|. . . A n−1 B) = ⎜ ... 0 1 ⎟
⎜ ⎟
⎜ 0 1* ⎟
⎜ ⎟
⎜ 1* * ⎟
⎝ ⎠
that has a determinant different from zero and therefore full rank and even adn. This
prove that the realization (AR , BR , CR ) it is all reachable.
 7.1. The problem of realization 211

Note that if the acquired results are linked to what is shown in the
study of reachability and observability properties, we have a
calculation procedure that allows us, given a generic realization of
a given nucleus, to calculate a new one that satisfies the property of being entire
reachable. If in fact (A, B, C) is a not fully reachable realization,

ρ(B ... An−1 B)< n

performing a decomposition with respect to the property of reachability and ex-

taking the reachable part, a new set (A) is obtained11, B1 , C1 ) that
has the same matrix of impulse responses, and therefore it is still a realization-
the given nucleus organization is entirely reachable.

Realization in observable canonical form

A second realization that we will consider, of dimension q, is that so-called -
this observable canonical information,Ao , Bo , Co .

0· · · · · · -a
0 0I
⎛ .. ⎞ B0
I . ⎛ .. ⎞
Ao .. Bo .
= ⎜⎜ 0 . . . .
⎟
⎟ = ..
(nq×nq) ⎜ . ⎟ (nq×p) ⎜ ⎟
⎜ .. .. .. ⎟ ⎜ . ⎟
⎜ . . ⎟ ⎜ ⎟
⎜ 0· · · 0 ⎝ Bn−1 ⎠
⎝ I -an−1 I ⎟⎠
Co
(0 · · · · · · 0I)
(q×nq)

Co (sI−Ao )−1 Bo =K(s)

-1
s 0 ··· 0 +a0
⎛ .. .. .. ⎞ B0
s . . .
.. ⎛ .. ⎞
⎜-1 0 .. .. ⎟ .
=(0······01) . . 0 . ⎟ ..
⎜ .. .. .. .. .. ⎟ ⎜ ⎟
⎜ . . . . . ⎟ ⎜ . ⎟
⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎝ Bn−1 ⎠
⎜ ⎟
⎝ 0· · · 0−1s+a n−1 ⎠

|sI−Ao |=a 0+a1 s+. . .+sn

212 7. Linear input-output models and state representations

aT
s 0 ··· 0 +a0
⎛ .. .. .. ⎞ B0
s . . .
.. ⎛ .. ⎞
(0···01) ⎜-1 0 .. .. ⎟ .
. . 0 . ⎟ ..
a0+· · ·+sn ⎜ .. .. .. .. .. ⎟ ⎜ ⎟
⎜ . . . . . ⎟ ⎜ . ⎟
⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎝ Bn−1 ⎠
⎜ ⎟
⎝ 0· · · 0−1s+a n−1 ⎠

by virtue of the structure of the matrix Co the only thing we want to calculate
and the last term, as the preceding ones multiplied by zeros.

B0
∗ ∗ ∗ ⎛ .. ⎞
(0···01) ⎛ ⎞ .
a0+. . .+sn ∗ ∗ ∗ n−1 ⎜ .. ⎟
⎝ 1 · · · s ⎠ ⎜ . ⎟
⎜ Bn−1 ⎟
⎝ ⎠

The overall product is just the K(s)

Example:
1−1 2 0
" 0 + s
K(s) 1# " 1 0#
= p=q= 2
q times p 3 + 2s + s2

0 0 -3 0 1-1
⎛ 0 0 0−3 ⎞ ⎛ 0 1⎞
Ao= Bo=
⎜ 1 0 -2 0⎟ ⎜ 2 0 ⎟
⎜ ⎟ ⎜ 1 0 ⎟⎠
⎝ 0 1 0−2 ⎠ ⎝
0 0 1 0
Co = "
0 0 0 1#
(q×p) q= 1

( 1 −1 ) + ( 2 0 )s
K(s) =
3 + 2s + s2

0−3 1−1
Ao= " # Bo= " Co( 0 1 )
1−2 2 0#
 7.1. The problem of realization 213

The implementation/simulation scheme associated with this representation

estimation with the state involves operations of integration, multiplication for
constant and sum as highlighted in the following figure in which it is
Assume for simplicity n = 2.

0−a0 b
Ao = " # Bo = " 0 # Co= ( 0 1 )
1−a1 b1

dx1ot-a0 x2+b0 u
⎧
⎪⎨ dx2ot=x1−a1 x2+b1 u
y=x2
⎪⎩
u

b0 b1
. .
+ x1 x1 + x2 x2 y
! !
- -

a0 a1

Figure 7.2

The realization now introduced is said to be in canonical form observable as it is

from the shape of the matrices0eC0it is obtained

0 0 1
Co ⎛ 0 0 1 ∗⎞
⎛ Co Ao ⎞ 0
.. = ⎜ ⎟
⎜ . ⎟ ⎜ ⎟
⎜ ⎟ ⎜ ⎟
⎝ Co Aon−1 ⎠ ⎜ ⎟
⎜ ⎟
⎝ 1 ∗ ⎠
The observability matrix is upper triangular and is therefore non-singular.
the realization is therefore entirely observable.
 214 7. Linear input-output models and state representations

As much as studied provides, implicitly, a reduction technique

with respect to this property. In fact, given a realization A, B, C, even-
Currently the realization of a given nucleok(t) = CeAtB, if this isn't all
observable, it can be extracted following a transformation, T, a new re-
realizationA11 , B1 , C1 , of dimension < n. It is a realization
of the same core of smaller size and all observable.

7.2. The minimum realizations

The problem of realization, as will be remembered, studies under what conditions

and in what way a functional input-output bond described by the con-
evolution of an assigned core K(t) (in the continuous integral time of with-
evolution, in discrete time convolution summation), can be realized
driven by the forced response of a dynamic, linear stationary system,
a finite dimension.
DataK(s), which can be assumed as elements without loss of generality
proper rational functions, one must find a dimension of the matrices A,
B, and eCche define a representation with the state of the dynamic system.
which has a transfer function coincident with K(s).
It has been shown as a necessary and sufficient condition for feasibility.
how
nesw
hvaeeM
,[Link]
rietansi K
)se(hce
build two different realizations. The one in canonical form is achievable:

B0+· · ·+Bn−1 sn−1

K(s) =
a0+· · ·+sn

0 I 0
.. .. 0
AR ⎛ ⎞
= . . BR= ⎛ ⎞..
(np×np) ⎜ 0 I ⎟ ⎝.⎠
⎜ ⎟ I
⎝ -a0 I · · · · · · ⎠ I
−an−1

CR= (B0· · · Bn−1 )

 7.2. The minimum achievements 215

and that in observable canonical form:

0· · · · · · -a
0 0I
⎛ .. .. ⎞
I . . B0
Ao .. .. .. ⎞
= ⎜⎜ 0
..
. . .
⎟
⎟ Bo = ⎛ .
(nq×nq) ⎜ .. .. .. ⎟
⎜ . . 0 . ⎟ ⎝ Bn−1 ⎠
⎜ ⎟
⎜ 0· · · 0 I -an−1 I ⎟⎠
⎝

Co(0 ... 0I)

having these realizations with dimensions equal to when one goes to realize a
nucleus that characterizes an input-output relationship with a single input p = 1
or with a single output.q= 1
Further elementary examples. The supplement:

1
K(s) = ⇒A=
s 0 B= 1 C= 1

=ẋyux

The function with a time constant τ and gain K:

K K/τ 1 K
K(s) = = 1 ⇒A=− B= C= 1
1 + τs τ +s
τ τ

1 K
=ẋ−x+=
yux
τ τ
the double integrator:
1
K(s) =
s2
0 1 0
A= " B= " # C= ( 1 0 )
0 0# 1
dx1ot=x22ẋ=u y =x1

The following considerations concern the size of the achievements;

minimum size should be preferred as highlighted in
continue. Remember, first of all that different realizations of a fixed K(s)
216 7. Linear input-output models and state representations

can be calculated from any (A, B, C) through transformation

coordinate versions

|T|n×n = 0 ⇒ (TAT−1 TB, CT−1 )

But there are also realizations of different sizes, in fact we know

that the matrix of impulse responses contains information about the laws of
motto of ways that are simultaneously excitable and observable, therefore the
W(s) does not retain information on all the eigenvalues of A. Consider
for example,

1 0 1
A= " # B= " # C= ( 1 0 )
-1-2 -1
the transfer function is:

1
W(s) =
s−1

and the construction of its realization according to one of the two techniques already
you see from
Ā= 1 B̄= 1 C̄= 1

Yes, therefore, a different realization A, B, ¯C ¯of ¯size 1 that is not

equivalent to the previous one of size 2.
Since the dimensions can vary, there is a natural need to
to have techniques that provide realizations with the size of the
state as small as possible. The reduction in size corresponds to,
make a lower complexity in terms of calculation procedures, an economy
of devices in terms of physical realization. And, it is ´ important to go
to build minimum realizations that have size of the state
as small as possible.

A first observation concerns the proposed techniques which, as already pointed out,
highlighted provide realizations of dimensions in particular conditions. In
in this case such achievements are also minimal.
 7.2. The minimum achievements 217

If we indeed denote the degree of the polynomial in the denominator in K(s),

It is simple to understand how to carry out dimensions of the lower Adnonn.
There are. Just remember that the polynomial in the denominator of a function
the transfer has a size less than or equal to that of the matrix; this
it behaves like a dynamic array of lower dimension that does not exist;
because the calculated one has this size, it is minimal.

q= 1⇒(Ao , Bo , Co minimum
p = 1 ⇒ (AR , BR , CR minima

So we already have a minimal implementation technique for nuclei, K,

which are row or column vectors. In the general case, this is not true.
example the system with two inputs and two outputs:

s2+ 2s 0
1
0 " 2s + 2s2+ 1 #
s+1
K(s) = " 2 1 # = =
s(s+2) s+2 s3+ 3s2+ 2s

0 0 2 0 1 0 2
" 2 + s+ " s
1# " 2 0# 0 1#
2s + 3s +s
2 3

p=q= 2
0 >N= 6
n= 3

admit, according to the proposed techniques, implementations of size 6.

´ is possible to develop production techniques that allow
It
In the general case of identifying realizations in minimal form? That is
possible.
Preliminarily and with reference to the example, it will be highlighted how it is
It is possible, even with the information we have, to identify a realization.
of size less than 6.
From a first point of view, the given matrix defines two input connections -
output through row vectors with two elements, Ki , between each exit and the two
entrances
218 7. Linear input-output models and state representations

1 0
s+1
⎛ ⎞ K (s)
K(s) = = " 1
2 1 K2 (s) #
⎜ s(s + 2) s+ 2 ⎟⎠
⎝
There are therefore two connections

(10)
K1 (s) =
s+ 1
realizable through matrices

Ao1-1 Bo1( 1 0 ) Co1 = 1

( 2 0 ) ( 0 1 )s
K2 (s) =
s2+ 2s
achievable through matrices
0 0 2 0
Ao2= " Bo2= Co2= ( 0 1 )
1−2 # " 0 1#
Starting from the realizations of K1eK2 , realizations of the lines, we can cal-
to color a realization of dimension 3 of the given nucleus through the following
matrices
A1 0 B1 C1 0
A= " o B= " o2 # C= " o
0 A2o # Bo 0 Co2 #
it has been taken into account that the system results from the connection of two
subsystems that have the same inputs and each defines an output. A
the alternative technique consists of creating each column. In this case

1 ..
0 .
⎛ s+1 ⎞ ⎛ . ⎞
K(s) =
2 1
= K1 (s) .K2. (s)
⎜ s(s + 2) ⎜ ⎟
⎝ s+ 2 ⎟⎠ ⎜ .. ⎟
⎝ . ⎠
To each of these columns, I can associate a realization of size
3 perK1
s(s + 2)
" 2(s + 1) #
K1 (s) =
s(s + 1)(s + 2) →n1= 3
 7.2. The minimum achievements 219

and 1 perK2
0
" 1#
K2 (s) =
s + 2 ⇒n2= 1
the overall realization will have dimension N=n1+n24 ed`e of the type

A1 0 B1 0
A= " o B= " o C= (Co1Co2)
0 A2o # 0 Bo2 #

It has been taken into account that the system results from the connection of two
subsystems that have the same outputs and each has a single input.
With reference to the example of multiple inputs and outputs under examination, the technique of
The implementation by rows is more convenient than that by columns.
In conclusion, with the information known to us, we are able to build
of achievements of a size smaller than what would turn out to be the
size of the realization built directly on the K(s).

7.2.a. A minimal realization technique in the case of

greater than 1*

The technique in question is based on some elementary considerations.

starting from the expansion of K(s) into partial fractions.

1 0 0 1 0 0 0
0 " 1 0# " 0 0# " #
s+ 1
K(s) = ⎛ 2 1
⎞ = + + -1 1
s s+1 s+ 2
⎝ s(s + 2) s+2 ⎠
r
Ki
K(s) =
1i=1 s−pi

we will support characterized by the presence of simple poles: polynomial at

common denominator with zeros of unitary multiplicity. Kithey are the residues but-
trials of size (q×p).
Sia
ρi=range Ki
220 7. Linear input-output models and state representations

C isieBitwo matrices of internal rank ρithat characterize a factor-

minimum leveling of the residue itself

Ki=Ci Bi

(q×ρi )(ρi×p)
A realization that turns out to be minimal is that of dimension N =
r
i=1ρiwith
2 Diagonal to elements coinciding with the poles of the expansion in fractions
partial with multiplicities each equal to the rank of the related residue
BeC, the aggregate of matrices BieCi

B1
⎛ ··· ⎞
. .. ..
B= ⎜ .. ⎟ C=
⎜ ⎟ 3 C1.· · · Cr .4
⎜ ··· ⎟
⎜ B ⎟
⎝ r ⎠
In the case of the previous example:

0 0 0
K 1= " = " # (10)
1 0# 1

1 0 1
K 2= " = " # (10)
0 0# 0
0 0 0
K 3= " # = " 1# (−1 1)
-1 1
and therefore
0 0 0
A= 0⎛−1 0⎞
⎝ 0 0-2 ⎠
1 0
0 1 0
B= ⎛ 1 0⎞ C= "
1 0 1#
⎝ -1 1 ⎠
This technique allows us, in the case of simple poles of the expansion in
partial fractions of K(s), to identify a minimal realization for another
 7.3. Introduction to the study of Interconnected Systems 221

in a simple formula, with diagonal.

One last aspect that wants to be highlighted is that the achievements min-
they have dimensions equal to the rank of the following matrix, H, called the matrix of
Hankel. The result, which would not be difficult to prove, is important in the
applications as it allows to detect the order of the minimum realizations
eventually starting from the examination of entrance - exit behavior.

K(t)
Si=q×p
(q×p)

S 0S 1S 2· · ·
⎛ ⎞
S1S2S3· · ·⎟
H= S S S · · ·⎟ ..
⎜2 3 4 ⎟
S 3 S4 .
⎜ . .. .. ⎟
⎜ . ⎟
⎝ . . . ⎠
infinite matrix, in blocks, where
I
dK(t)
S i= 5
dtI 5t=0
5
5
The Hankel matrix from a realization (A, B, C) of a system
dynamic, can be obtained by replacing at K(t) the expression
of the impulsive response, obtaining for the coefficients:

Si→CAi B

7.3. Introduction to the study of Interconnected Systems

The interconnected systems, that is, made up of several parts, of several subsystems among
They connected in various ways are the subject of this lesson.
The first problem that arises is the calculation of the model that describes how
specifically the interconnected system starting from the knowledge of the models
222 7. Linear input-output models and state representations

of the individual parts that make it up. The following introduces the
elementary connections: in series, in parallel and feedback.
Consider two systems, the first S1who has entered1the exit1 ; it
seconds2conu2hey2 A first connection that can be considered is that
the exit of the first becomes the entrance of the second; thus there is an interconnection
specified by the following conditions:

u2=y1 u=u1y=y2

ẋ=A
1 1 x1+B1 u12
ẋ=A2 x2+B2 y1
0 0
y1=C1 x1 y2=C2 x2
Therefore, it is possible to find the overall model by introducing into the descriptions of
each of the two systems the condition of interconnection, in this case:

dx1ot=A1 x1+B1 ux˙2=A2 x2+B2 C1 x1

0 0
y1=C1 x1 y=C2 x2

It is therefore natural to consider as a state variable of the system comp-

lexical the aggregate of the two state variables1 , x2 One obtains, then the expression:

ẋ1 A1 0 x1 B1
" ẋ2 # = " B2 C1A2 # " x2 # + " 0 # u

x1
y = ( 0 C2) " x2 #

It can therefore be noted that the aggregate of the two systems is still a system, and it
state space can be considered through the aggregate of the components
of the state space of the first system and the second. This is a connection
important the cascaded connection or series connection is reported in
figure
u1 y1 u2 y2
S1 S2
Figure 7.3
 7.3. Introduction to the Study of Interconnected Systems 223

Another type of connection is the parallel connection shown in the figure.

following
y1
u1
S1
+ y
u
+
u2 y2
S2
Figure 7.4

It is called parallel connection when the input is common to both.

In systems, the output instead becomes the sum of the two outputs. The conditions of
interconnection in this case is:

u1=u2=u y=y1+y2

By imposing the constraints, we obtain:

ẋ=A
1 1 x1+B1 ux˙2=A2 x2+B2 u
0 0
y1=C1 x1 y2=C2 x2

y=y1+y2

ẋ1 A1 0 x1 B1
" ẋ2 # = " 0 +
A2 # " x2 # " B2 #
u

x1
y = (C1 C2) " x2 #

So constructing the overall model is quite easy, just consider

the two representations and introduce the interconnection constraints
Another type of connection is the feedback connection reported in
figure
224 7. Linear input-output models and state representations

u + u1 y1 y
S1
-

y2
S2 u2
Figure 7.5

It is about connecting the two systems.1 edS2in a different way. One takes the
systems1with entered1he/she/it exited1and the systems2conu2hey2in sense
reverse; start from a defined external entry regardless and proceed
to subtract the exit from it2of the second system:

u1=u−y 2

Then consider the overall output of the system as the input of the system.
S2the exit1 , that is:
u2=y1 y=y1

An interconnection is built in which the first system is acted upon by the ...
the difference between an external input and its output brought back and subtracted,
after passing through the system2 at the entrance. There is a cycle that
it is called counterreaction, because the negative sign is introduced, meaning we subtract
draw the exit1upon entry. This feedback or cycle system
closed, or closed-loop is very important especially in the description of
natural phenomena, as well as in the technological and application fields.

The overall system, starting from the descriptions of the two systems.1 , S2 ,
here are the equations:
ẋ=A
1 1 x1+Bu−B1 C2 x2
0
y=C1 x1

dx2ot=A2 x2+B2 C1 x1
0
y2=C2 x2
 7.3. Introduction to the study of Interconnected Systems 225

ẋ1 A1 -B1 C2 x1 B1
" 2ẋ # = " B2 C1 +
A2 # " x2 # " 0 #
u

x1
y= (C10 ) " x2 #
Three emblematic cases have been examined, a system with two parts in
cascata o in serie, un sistema con due parti in parallelo, e uno con due parti in
a counter-reaction scheme, it has been verified that assuming it as a state
The aggregate of the states of the component parts can describe a system.
of the same nature, with the finite, linear, stationary, regular dimension.
Ultimately, it has been found how the representation of an interconnected system
more parts can be identified starting from the models of the parts that constitute it
of course the interconnections can also be more complex,
So perhaps it is appropriate to introduce simpler graphic notations of
which of the block diagrams used so far.
Representations of interconnected systems through flow graphs
In the block diagrams considered so far, the variables are fast-
presented by arches that connect blocks; the block represents the system that
connect the two input - output variables; the connections occur for col-
ligament, derivation, and sum of arches. There are four symbols, four
notations through which the block diagram is constructed: arcs, blocks, sum-
mothers and letters.
Flow diagrams represent an alternative tool for describing
the interconnected systems. In them, the variable is represented by a point, a
nodou, the system through a segment that links two variables u−y, there is
a certain dualism this time the trait represents the system, ultimately the fact
that the same variable can be input or output of multiple systems, represents
simply the branching from a generic variable. The sum instead
is assumed following the convergence on a node, establishing that the vari-
able to be found at the point of convergence is the sum of the effects of the two
systems represented by the two segments. The notation is therefore very simple.
and allows for quick representation of interconnections. It has nothing of
conceptually different from the symbolism of the block diagram, but it is very
sleeker.
226 7. Linear input-output models and representations with state

The symbols have well-established names, the points are called nodes and represent
variable feelings, the traits are called branches, and the set of nodes and branches
it constitutes a flow graph, that is, the graphical representation of the system. In
In some cases there are open paths, parallel paths, however there are -
three characteristic configurations, namely rings, closed loops. Open paths
the rings that in a flow graph allow for the development of a theory for
describe the overall system in a very simple way.
The transfer function of the overall system
Cascade system case: be W1the transfer function that connects
entry - exit, eW2entrance - exit of the second system.

y1=W1 u1 y2=W2 u2

This is not a complete description like that of the state variable.

because it is a description that represents only the part made up of the ways
observable dynamics and reachable modes, however when the system
the representation through the transfer function has these characteristics
usual maintenance. Following the interconnection:

y=W2 W1 u=Wu

W=W1 W2

For the parallel:

y1=W1 u+W2 u

y= (W1+W2 )u=Wu

W=W1+W2

Thus, in cascade, the transfer function is the product, in the connection.

in parallel is the sum.
Infinite, for the counter-reaction:

y=W1 (u−W2 y)

y(1 +W1 W2 ) =W1 u

 7.3. Introduction to the Study of Interconnected Systems 227

W1
y= u=Wu
1 +W1 W2
W1
W=
1 +W1 W2
This last expression is very interesting because1and the function of
transfer of the direct path connecting entrance - exit; 1 +W1 W2the
The denominator can be interpreted as 1 minus the product of the functions of
transfer of the ring (−W1 W2 Therefore, two concepts can be introduced.
important:
P=W1

L = -W1 W2

D = 1 - L = 1 + W1 W2

W1 P P
W= = =
1 +W1 W2 1−L D
The function has a direct interpretation. Suppose we divide the node,
from the case to counter-reaction, in two parts, a node where there are starting branches
in this case one, and the other with branches arriving, that is to cut the knot where
there is the loop, moreover it is assumed to insert a signal equal to 1 in this loop
interrupted. What is the signal at the end of the interrupted cycle?

1·(-W1 W2 ) =L

So if a node is cut in a cycle and a unit signal is inserted.

what returns, which is the measure of how the cycle behaves.
1−L`e the difference between what it puts out, that is 1, and what it retains. This fun-
The concept of difference has a precise meaning. Therefore, the transfer functions
in this case, they appear with the transfer function in the numerator of
direct relationship input - output1 , and in the denominator the difference function.
From this elementary consideration, it can be generalized.
way of considering feedback systems also in more complex structures
less.
One last observation that is important to make concerns a deepening.
on the counter-reaction scheme. A counter-reaction scheme with: a block
228 7. Linear input-output models and state representations

P, in direct chain, represents the process that one wants to evolve.

according to a desired trend; with organsG, in direct chain before the
process, and a mechanism that measures the output and reports the indications,
at the entrance, represents a characteristic scheme in the design of systems
of control and intuitively lends itself to representing the modalities
according to which an intervention action is exercised on an assigned system
to make it behave as desired. It represents the system that on which it
wants to intervene, our model of reality that we intend to intervene on,
It represents the measures, the evaluations that consider the effects of our
intervention, it represents the control law, the intervention strategy with
which based on the measures we believe we can modify as desired the
behavior of the system. The scheme therefore conceptually expresses a
intervention philosophy on a system not directly on the input but on the
difference between the measured input and output. In this it represents in some
in a way the desired value that one would like as output.
What is the operating principle of this system?
If the output is in the desired correspondence with the input, and if:

Hy=u,

So the input signal is zero and there is no further indication.

through Gsu because the output corresponds to the input. However, there is one
shift, if it changes, the error signal changes as well, the difference between
if it decreases, there is an action that tends to increase the output, if it increases
There is an action that tends to reduce it.
Thus, this system acts automatically regardless of the cause that
it causes the deviation of the output from the desired value. So think about
example of a disturbance that acts on the output, if this disturbance acts
Externally, the exit will deviate from its current trend and there will again be a signal.
which at the entrance will tend to compensate for this disturbing action. This happens
Even if there are modifications to the process, the process will vary its
caratteristiche, questa variazione provocher`a una variazione in uscita, questa
it will resonate again when entering and the system will intervene again. The great potential
This mechanism is to react to any cause of disturbance.
 7.3. Introduction to the study of Interconnected Systems 229

´ is a mechanism capable of functioning very satisfactorily to make

It
correspond the output to the input. All natural processes that work in
in real time, they act according to this scheme. This scheme is a counter-reaction
the fundamental basis of the theory of control, and also of the theory of
automatic adjustment.
What is presented in this lesson lays the foundation for the study of systems.
interconnected. This is a very extensive study that goes well beyond the
objectives of the Systems Theory course as suggested by the latest con-
Considerations made. Some topics will be addressed in the next lesson.
with the aim of providing a systematic procedure for the description of the sis-
overall theme and some elements of analysis of stability based on the
property of the subsystems that make it up.

7.3.a. Interconnected systems: model calculation

This paragraph is dedicated to the presentation of two methods for calculation

of the mathematical model of an interconnected system; the transfer function
Mento (Mason's formula) and the input-state-output representation.
For the calculation of the transfer function of the overall system if
it is indicated with Sithe transfer function of the generic, i-th subsystem
(this branch transfer in the flow graph), the use of it is useful
of the following expression known as Mason's formula

i∆i Pi
S= 2 ∆
in which
PItransfer of′ I walk from u to y
∆⇒1−(−1)k+1 Pjk
1k1 j

Pjkproduct of k transfers d′ ring of rings that do not touch

∆i⇒∆ deprived of all transfer products that are touched by PI

SoPiand the transfer of the i- and similar walking to operate in Uruguay.
230 7. Modelli ingresso uscita lineari e rappresentazioni con lo stato

With reference to the flow graph represented in the figure

S4

u 1 S1 S2 S3 1 y

S5

S6
Figure 7.6

Dauaysi has two paths.

P1=S1 S2 S3
0
P2=S3 S4

∆= 1−( sum transfers d′ ring)+

"1 of transfer products of pairs of rings = #

−" 1 products of triple rings = + ... #

Returning to the case under consideration:

∆= 1−S2 S5−S2 S3 S6

in this case one must stop, because the rings touch, they have a branch in
common, but just one shared node is enough to no longer be considered
the product of the pairs, which therefore becomes zero, just like all those that
they follow.
∆ideprived of all those products of transfer functions that
involve some transfer function that is already in PI

∆1 = 1
0
∆21 - S2 S5
 7.3. Introduction to the study of Interconnected Systems 231

In conclusion, regarding the transfer function of the example.

Taking into consideration, the following expression will be obtained:

P1 ∆1+P2 ∆2
S= = S1 S2 S3+S3 S4-2 S3 S4 S5
∆ 1−S2 S5−S2 S3 S6
Another example that is discussed is shown in the figure
S6 S8

u one S1 S2 S3 S4 S5 1 y

S7

S9

Figure 7.7

P1=S6 S3 S4 S5 ∆= 1−S2 S3 S7−S3 S4 S8

∆1=
P2=S1 S2 S3 S4 S5
∆2=
P 3= 9
∆3=
After calculating ∆1 ,∆2 ,∆3apply the formula.

Now it will address the calculation of the input-state-output representation.

ẋi=Ai xi+Bi ui
S i: 0
yi=Ci xI

the calculation boils down to imposing the topological constraints established by the connections between
the various subsystems. As already seen, in the case of the cascading connection, one
ha
232 7. Linear input-output models and state representations

ẋ=A
1 1 x1+B1 u
⎧
⎪⎨ ẋ=A
2 2 x2+B2 C1 x1

y=C2 x2
⎪⎩
in which the state is the aggregate of the states of the subsystems. The heuristic procedure
which consists of enforcing compliance with constraints is simple in elementary cases,
but when connections with multiple subsystems are considered, with multiple
reverse reaction connections, it is necessary to refer to a procedure
systematic based on the analysis of the flow diagram of the interconnected system
connection. This procedure is based on the fact that in the absence of cycles, the calculation of
representation does not present difficulties. It is therefore a matter of releasing some
connections in order to obtain an acyclic graph, calculate its model and
then restore the open connections.
Consider the flow graph and make, possibly in number
at least, separations of nodes so that there are no more in the resulting graph
cycles. The separation must be done in such a way as to leave all branches
incoming on one side and outgoing branches on the other. For a graph of this type,
without loops, the representation with the state is immediately calculated, to
starting from that of the subsystems.
In fact, the state vector is the aggregate of the state of the individual subsystems.
the representation can be simply written by inspection on the graph
of flow. This will be highlighted by referring to the example in the figure

S4

u 1 S1 S2 S3 1 y
S5
S6

Figure 7.8

p= well, the part of the node separated where all the branches arrive
 7.3. Introduction to the study of Interconnected Systems 233

s = source, the part from which only branches start

So we get:

A=
+
B
xẋu1 1 1 1
⎧
⎪⎪ ẋ=A 2 2 x2+B2 s
⎪⎪
⎪⎪ dx3ot=A3 x3+B3 C2 x2+B3 C4 x4
⎪⎪
⎪⎪
⎨Axuẋ 4
=
+
B 4 4 4

⎪⎪ ẋ=A
⎪⎪ 5 5 x5+B5 C2 x2+B5 C4 x4
⎪⎪
⎪⎪ ẋ=A x +B C x
⎪⎪ 6 6 6 6 3 3
⎩
y=C3 x3 p=C1 x1+C5 x5+C6 x6

From which it is obtained

A
=
+
B
xẋu1 1 1 1
⎧
⎪⎪ ẋ=A 2 2 x2+B2 s
⎪⎪
⎪⎪ ẋ=A
⎪⎪ 3 3 x3+B3 C2 x2+B3 C4 x4
⎪⎪
⎨ A
=
+
B
xẋu4 4 4 4

⎪⎪ ẋ=A
⎪⎪ 5 5 x5+B5 C2 x2+B5 C4 x4
⎪⎪
⎪⎪ ẋ=A x +B C x
⎪⎪ 6 6 6 6 3 3
⎩
y= ( 0 0C30 0 0 )x
p= (C10 0 0C5C6)x
To restore the original graph, it is necessary to impose that it coincides with p.
By doing this, the representation with the state of the system is obtained.
detergent.
In conclusion, the overall representation is obtained: separating
some nodes of the graph, calculating the associated representation, imposing the
coincidence constraint of separate nodes
So two methods: one for calculating the transfer function, the other
for the calculation of the representation with the state, they put, among other things, in evi-
it is assumed that the state of the overall system is the aggregate of the states of the sub-
systems.
234 7. Linear input-output models and state representations

In this regard, it is worth noting that following the connection

there may be a loss of reachability and/or observability.
for example, the subsystems1 , S2characterized by their respective functions
of transfer:
s+1 1
" s−1s+#1 " #

the transfer function of the overall system is

1
W(s) =
s−1

and admits the following achievement with the state

+
=
uxẋ
0
y=x

A system of dimension one has been obtained while the size of the system
comprehensive, interconnection of two subsystems of size one and two.
This fact can be explained in terms of loss of reachability or
of system observability, because the transfer function of the system
overall and influenced only by the fully reachable subsystem and everything os-
servable.
In reality, the case being considered corresponds to a loss of reason.
accessibility of the system. This can be verified by realizing the two subsystems,
finding the representations with the state, the overall representation and
Finally verifying that the system is not all reachable.
An opposite cancellation will correspond to a loss of observability.

In conclusion, an interconnected system made with subsystems reaches

and observable variables show a loss of reachability and/or observability
the denominator of the transfer function has a degree lower than the sum
of the sizes of the subsystems.
8. Elements of stability theory

Stability is that property by virtue of which small deviations of

evolutions correspond to small disturbances; where disturbances are in-
there are various changes in the conditions that generate the different developments. So much
The smaller the disturbances, the smaller their effect.
If it is observed that for a given system an evolution, in the state or in
output corresponds to an initial state and an input function, as well as
all fixed functions that define its representation, it is understood that a
Disturbance can materialize in changes to the initial state, of the input.
and of the model. The effect of model perturbations defines stability.
structural; a property that will not be considered in this context.
The study of stability is one of the methods that define qualitative analysis.
stability of dynamic systems; these are the methods that allow us to study
the properties of the solutions without proceeding to their calculation.

The following discussion is primarily dedicated to the study of stability.

of internal evolutions following disturbances to the initial state or to states
of equilibrium of an assigned dynamic system.
Starting from the main definitions, the study leads to the identification
of conditions and the establishment of criteria and methods) of verification.
 8.1. Stability: definitions and conditions 237

8.1. Stability: definitions and conditions

Stability is the most important among the properties of a dynamic system.

ico as representative of behaviors (the stable ones) that we could
to define 'auspicati'. In a general sense, we can understand that it expresses
a sort of 'continuity' of the cause-effect relationship that the system intrinsically
mind represents.
Consider the continuous-time system described by the following representation
presentation with the state, stationary at finite size

p,(7.1)
)ut(,)xt(=
f)tẋ(
! "
y(t) = ηx(t), u(t), p (7.2).
! "

This is a nonlinear representation in which the generating function f

express the dynamic link between the input and the state, while the function
express the dependency of the output on the state and on the input; the functions
they do not depend on time (stationarity), but can depend on parameters
p.
With reference to the system (7.1)-(7.2), we will define 'stable' those that involve
mind (evolutions in the state)0 (·) and in exit0 corresponding to fixed0 ,
u0 (·),p∗) that vary little following disturbances to the initial state,x0p
(referred to as internal stability with respect to ax)0 (·)), following disturbances
on the input function, u0p (•) (referred to as entry-state stability if referenced
ax0 (·), entry-exit if referred to ay0 (·)), following disturbances on the value
of parameters (structural stability, in the state or in output).
The following discussion is mainly dedicated to
study of internal stability, the results of this study are the basis of the
diverse properties of stability and will allow us to easily deduce in the case
the stability conditions of input-output for linear systems.
One last aspect worth highlighting is represented by
an additional characteristic that can be found in the study of compositional
effects of perturbations on the state of a dynamic system; they
238 8. Elements of stability theory

it deals with a strengthening of the property of stability that leads to

thinking about the growing effect of the disturbance over time. The system,
In other words, it is capable of compensating (offsetting over time).
the effect of disturbances. In this case, we will talk about asymptotic stability.
We will assume hereinafter, unless explicitly stated otherwise,
action, that the system is represented by (7.1); we are therefore interested in
behavior in the state.
A statusandit is in a state of equilibrium, if the system is in that state
It remains with null entry; this happens if and only if

0 = f(xe ,0)

In the case of linear systems, where:

A
=
+
B)xutẋ(

The equilibrium states are those that satisfy the following equation:

0 = Axe

From this, it is evident that (xe=

0), that is the origin of the state space, is always
a state of equilibrium; in general, it is not unique, it is only if the determinant of
A`e d iv e r s o d a z e r o.
Consider, for graphical representation purposes, a state space
two-dimensional 1 , x2and consider the state of equilibriumefor example
the origin of the state space.
xeit is said to be locally stable if a small ε is defined at will and drawn
a circle with radius ϵ around dxethere exists a δ that depends on the ϵ
that if a disturbance of the initial state with respect to the equilibrium point is
within the circle of radius δ, the trajectory that originates does not diverge from xe ,
or rather does not exit the circle of radius ϵ.
So ultimately, once it has been disturbedanda movement is born that
it does not stray too far from the state of equilibrium. This is written in the following
mode:
 8.1. Stability: definitions and conditions 239

xandand it is stable

∀ϵ∃δ(ϵ) :

∥x(t0 )−xe ∥<δ(ϵ)⇒ ∥x(t)−xe For all t greater than or equal to t0

Essentially, the property of stability is a condition of the limitation of the effect.
of disturbances. We have a system in a state of equilibrium, it is
turbato,x(t0 ) , therefore a transitory is born, if this free evolution is
! "
is maintained locally limited, then it must be said that that state of equilibrium
it is the stable element, vice versa it is unstable.
To formally express these concepts, it is necessary to remember that the
the norm of a vector x(t) can be defined in different ways:
1
n n 2

$ x2i %
#i=1 |xi | #i=1 max|x
i i|

and as for convergence, in finite-dimensional spaces all

the standards are equivalent. Therefore, one can refer to any of
these definitions.
A more stringent property is asymptotic stability: xeand the s in to t ic a m e n t e
locally stable if the previous condition is met, that is
limitedness of the effects of the disturbance, but also the convergence towards xe
of evolution. That is ultimately:

∃δa (ϵ) : ∥x(t0 )−xe ∥<δa⇒ lim∥x(t)−xe ∥= 0

t→∞

And fifteene and the s in the static elements is a consequence of a perturbation,

after a certain time, one returns to that state.
The definite properties have a local significance if one imagines that in a
system there are multiple equilibrium points, locally around each one it is possible to
to think of defining stability. A very strong property is global stability
xethe global environment is a stable systemeis table

For all k > 0, ||x(t0 )−xe ∥< k⇒lim∥x(t)−x

t→∞ e ∥= 0
240 8. Elements of stability theory

however great the disturbance of the state of equilibrium, a tran- arises

converging ax supporte .
In general, we therefore have local properties, more or less strong, and properties
global, but the interesting thing is that in linear systems, the local properties
they imply the global ones. In fact, for a linear system

A
=)xtẋ(

Local properties imply global ones.

Consider, for this purpose, the transformation

z(t) = αx(t)

arbitrary constant; it is obtained

ż()=
tαẋ()=
tαAx()=
tAz()t

therefore the new variable satisfies the same dynamic equation that it satisfied
lax, then having the same dynamic properties, I will be satisfied, for example,
the same condition of local asymptotic stability namely:

∥z(t0 )−ze ∥<δa⇒ lim∥z(t)−z

t→∞ e ∥= 0

in this condition, it is possible to replace again αxe, so we have:

δa
∥x(t0 )−xe ∥< α ⇒ limα∥x(t)−x e= 0
t approaches infinity

which proves the global asymptoticity for arbitrariness.

In reality, a clarification is needed: the equilibrium point has been [Link]
general terms, in this case the equilibrium point xe if the system is linear
it can only be the zero state, the only possible state of isolated equilibrium,
because if there were equilibrium pointsedifferent from the origin, moving with
the initial disturbance daxein one of these equilibrium points it would not be
then satisfied the condition that once disturbed the initial state asin-
totally you come back inxe .
 8.1. Stability: definitions and conditions 241

Another aspect concerns the stability of a motion: let us suppose, again for
graphic representation requirements, the two-dimensional state space1 , x2
one starts from an initial state x(t)0 ), and imagine fixing an entrance; one will have
a motion that satisfies the equation:

)ut(,)xt(=
f)tẋ(
! "
if the initial state is disturbed, a different trajectory is born and a motion for-
disturbed,xp (t), which satisfies the same equation:

ẋ(t)
p = f xp (t), u(t)
! "
it indicates the evolution of the deviation between the two trends

ξ(t) = x p (t)−x(t)

it is obtained

˙ = x˙ p (t)-x˙(t) = f x(t) + ξ(t), u(t) - f x(t), u(t)

ξ(t)
! " ! "
this last one is the equation that the new variant satisfies. It is noted that if
it takes ξ = 0, that is, the origin of the state space of the new variable ξ, the
the second member of the equation becomes zero. So when ξ = 0, ξ(t) = ˙0,
That is, the origin of the state space of the variable ξ is a point of equilibrium.

So just study the stability of the origin of the state space.

variable to know the stability of the motion that was present in the space.
This shows that with a change of variable, the study of stability
a motion transforms into the study of the stability of an equilibrium point,
more precisely of the origin in the variable ξ.

Naturally, in the general case now being discussed, the system is not stationary.
because it depends on the particular input u(t) and the motor x(t). But in the case of the
stationary linear systems do not happen. In fact:

A
=
+
B)xutẋ(
 7.3. Introduction to the Study of Interconnected Systems 223

Another type of connection is the parallel connection shown in the figure.

following
y1
u1
S1
+ y
u
+
u2 y2
S2
Figure 7.4

It is called parallel connection when the input is common to both.

In systems, the output instead becomes the sum of the two outputs. The conditions of
interconnection in this case is:

u1=u2=u y=y1+y2

By imposing the constraints, we obtain:

ẋ=A
1 1 x1+B1 ux˙2=A2 x2+B2 u
0 0
y1=C1 x1 y2=C2 x2

y=y1+y2

ẋ1 A1 0 x1 B1
" ẋ2 # = " 0 +
A2 # " x2 # " B2 #
u

x1
y = (C1 C2) " x2 #

So constructing the overall model is quite easy, just consider

the two representations and introduce the interconnection constraints
Another type of connection is the feedback connection reported in
figure
 8.1. Stability: definitions and conditions 243

and that they are all equivalent in finite-dimensional spaces.

That being said, it is easy to show that the expressed condition is necessary and
sufficient because if it is assumed to be satisfied, one can write:

||x(t)|| ≤ ||φ(t−t0 )∥·∥x(t0 )∥

||x(t)|| ≤ k||x(t0 )∥
ϵ
||x(t0 )∥< k
ϵ
||x(t)|| ≤ k k
So if the matrix φ is limited to evolution free limited; furthermore,
se∥φ(t−t 0 if it were not limited, there would be at least one element of the matrix
not limited to the changing of:

xi (t) = φi (t−t0 )xj (t0 )

With the same arguments, it is easy to verify that a necessary condition is...
aware of asymptotic stability and that

lim∥φ(t−t 0 )∥= 0
t approaches infinity

the demonstration is analogous both for sufficiency and for necessity

This condition can be expressed as:

A
=)xtẋ(

φ(t) = eAt →λi<0

for the expression that φ(t) takes as a function of the eigenvalues; in fact the laws
The temporal φ(t) are increasing or decreasing depending on whether the real part of the
eigenvalues of Asia to the right or to the left of the imaginary axis. In conclusion,
the condition ∥φ(t)∥ < k requires the eigenvalues to be to the left of the axis
imaginary, eigenvalues can be admitted on the imaginary axis only if
simple. If it also requires that:

lim|φ(t−t 0 = 0
t→∞
244 8. Elements of stability theory

the eigenvalues must be strictly to the left.

The imaginary axis is the boundary that divides the eigenvalues of the matrix.
There are two categories, those that give rise to converging natural ways and those
which give rise to divergent natural ways. In conclusion, the condition of
stability can be expressed by saying that the eigenvalues of A must not
positive, as long as they are simple; for asymptotic stability, they must all be with
Negative real part. Therefore, the imaginary axis is the stability boundary.
So given the dynamic matrix that characterizes the system, the stability
asymptotic or not, it is linked to the sign of the real part of the eigenvalues of A. This
and the condition that must be verified through some methods, as can be seen
they do not require the calculation of eigenvalues, but only the sign of their part
real.
The stability criteria, which we will discuss in the upcoming lessons, are used to verify
such allocation.

8.2. The Routh criterion

As established in the previous lesson, the internal stability of a linear system

that is, of the stability of equilibrium states and motions, is equivalent to the
stability of the zero state. The conditions for the latter are as follows
remember.
simple stability
Re(λi1)≤0
&
Re(λi1>)<0
asymptotic stability
Re(λI )<0

where it is indicated with the prime 1 or >1 the unitary or greater multiplicity
of an eigenvalue.
 8.2. The Routh criterion 245

The topic of this lesson is the development of criteria for verification.

satisfaction of such conditions. Given the dynamic matrix, one must know how to
evaluate the sign of the real part of the eigenvalues, in the case of continuous time, the
module, in the case of discrete time. One could go on to calculate the eigenvalues,
but fortunately this is not necessary; in fact, calculating the zeros of a
A polynomial is one of the most complex problems in mathematics when the degree
the superior to 4.
The criteria we will discuss allow without solving the characteristic equation
to have information about the real part or the modulus of the zeros.
A preliminary consideration regarding the study of the part
It is true that among the zeros of a polynomial there are some with a positive real part if ...
the coefficients do not all have the same sign. Indeed, it is not difficult to understand
there, if we think of the polynomial in its factored form, that the presence of
coefficients of different signs can only be the result of zeros apart
positive and negative real. The coefficients all have the same sign and therefore a
necessary condition for all zeros to have negative real part.

The Routh criterion

With reference to the generic polynomial of degree n,

an λn+an−1 λn−1+. . .+a0= 0

The Routh criterion allows us to determine if all the zeros have a real part.
negative and potentially how many are the zeros in the positive real part. The criterion
it consists of constructing a table and then verifying the coefficients
degli elementi della prima colonna della tabella. Il criterio in esame estende
the Cartesian rule on the correspondence of solutions with positive real parts and
negative to variations and permanences of the signs of the coefficients of a polynomial
Of second degree. As it will be remembered, a variation corresponds to a zero.
a positive real part, to a permanence one with a negative real part.

ax2+bx+c= 0

Construction of the Routh table

246 8. Elements of stability theory

n an an−2 an−4
n−1 an−1 an−3 an−5
bn−2 bn−3 ...
cn−3 cn−4 ...

anan−2
' an−1an−3 '
bn−2 '
= ' '
'
' -an−1 '

anan−4
' an−1an−5 '
bn−3 = '' '
'
' -an−1 '
The following line, of order -3, is still constructed with the same procedure.
but with reference to the two previous lines; and so on
Example:
λ5+ 3λ4+ 2λ3-2λ2+ 2λ + 4

5 1 2 2

4 3 -2 4

3 8 2 0
3 3
2 11 4
− 4

1 50
11
0 16
In the construction of the Routh table, the result does not change if it is modified
you can multiply the rows by the same positive number; this allows for
simplify the calculations in the construction of the table.
Once the construction of the table is completed, the first column is considered.
table and they count the number of sign changes that occur in the transition
from one coefficient to another. The number of variations is equal to the number of zeros
the positive real part of the given polynomial.
 8.2. The Routh criterion 247

In the case under study, the fifth degree polynomial has two zeros with real parts.
positive, the other three have negative real parts.
It is important to note that one may encounter obstacles in construction.
of the table; this occurs when an element of the first column becomes zero.
This can happen for only the first element of the row or for all the elements.
of the line. When one of these situations arises, stability is not achieved.
asymptotic.

Consider the following situation

1 2 3
2 4
0 3

The impediment can be overcome in three different ways as explained in the se-
guito.
Replacing the symbol ϵ with zero to represent a positive number and 'small-
color

1 2 3
2 4
ϵ 3
-4ε-6
ϵ
3

Once the table is constructed, the sign changes are counted. In the case
There are two sign changes in the exam, so it is concluded that there are two.
zero positive real part.
Multiplying the given polynomial by a binomial with a negative zero;

d1 (λ) =d(λ)(λ+ 1)

a new polynomial of degree +1 is obtained with new coefficients if to this

new polynomial we apply the Routh criterion with probability one, not
I will find zero on the first column. For example, if one studies

λ4+ 2λ3+ 3λ2+ 6λ+ 4

248 8. Elements of Stability Theory

1 3 4
2 6
0 4

from which

(λ + 1)(λ4+ 2λ3+ 3λ2+ 6λ + 4) =

λ5+ 3λ4+ 5λ3+ 9λ2+ 10λ+ 4

1 5 10
3 9 4
6 26
-1 1
32
1

A third way to overcome the obstacle encountered, and it is in effect

di una procedura priva di ambiguita` ed applicabile in presenza di piu` elementi
none in the same row, consists of replacing the row in question with the string
of numbers obtained by adding to the i-th element of the row the element at position
(i+j) in the same row multiplied by (-1)j being the number of primes
null elements. For example, consider the polynomial

λ5+ 5λ3+ 10λ+ 4

With the indicated changes, the following table is reached.

1 5 10
4 0 4
4 9
-5 4
32
61 divided by 5

A different situation that can occur when building the table is nullification.
say an entire line. This can only happen in correspondence with a line of
odd order, in fact the two previous rows must be proportional and
 8.2. The Routh Criterion 249

so they must have the same number of elements (note that in passing from
an odd line to an even one underlying the non-changing elements.

λ6+ 2λ5+ 3λ4+ 4λ3+λ2+λ+ 1

6 1 3 1 1
5 2 4 1
4 x x x
3 x x
2 x x
1 x
0 x

In this circumstance, it can be concluded that the polynomial under consideration is the
product of two polynomials; the first will have zeros with a real part characterized
given the sign variations of the elements in the first column of the table
so far built (the zeros of)1the positive real part are as many as the variables
sign actions that appeared in the first column of the constructed table
up to that moment); the second polynomial is of even degree and equal to the index
of the line preceding the line that was canceled, it only has even degree powers
and its coefficients are in order, from the one of maximum degree to that of
zero degree, the coefficients of the row that precedes the one that has been annulled. If
for example consider
5 1 3 1
4 1 3 1
3 0 0
The given polynomial is therefore equal to the product of two polynomials.

d(λ) = d1 (λ)d2 (λ)

the first of first degree with a negative zero, the second of degree four
characterized by the presence of powers of λ only of even order

d2 (λ) =λ4+ 3λ2+ 1

For such a polynomial, we can calculate the zeros, but in the case of the construction of the
if the table had stopped at a higher level, there could be some
250 8. Elements of stability theory

difficulty in seeing the sign of zeroes did2 (λ). Then the construction of the
Table can be resumed in another way. The polynomial is derived.2

d 4
(λ + 3λ2+ 1) = 4λ3+ 6λ
dλ
and the coefficients of the null row (that of order three in the case under study) are
sostituiscano i coefficienti di questo nuovo polinomio. Si pu`o or a c on t in u ar e
the construction of the table.
5 1 3 1
4 1 3 1
3 2 3
2 3 2
5
1 3
0 2

Upon completing the table, the analysis of the coefficients is carried out. In the case under study, it does not

there are sign variations, which guarantees stability in the case in question
simple, as it could be verified by calculating the zeros. It should be noted
the zeros of the polynomial2 (λ), with powers of only even degree, have a
double symmetry, with respect to each axis of the complex plane. This axis-
care that if there are no zeros on the positive real axis, all are found on the axis
imaginary.

Some concluding thoughts concern the potential of the method.

It is possible to determine not only how many zeros a certain polynomial has that
I am to the left of the imaginary axis, but also how many zeros there are
to the left of a predetermined axis that has coordinates on the real axis equal to -α.
Such research is of interest to understand the response readiness of a
dynamic system. To do this, it is enough to make a translation of axes and
bring the imaginary axis in−αed and express the polynomial d(λ) in the new ones
coordinated(λ−α)

Example:
λ2+ 9λ+ 20 = 0
 8.3. The general method of Lyapunov* 251

It has a zero on the negative real part as a result of the elementary application of
Routh's criterion or, which is the same in the case of a second-degree polynomial
degree, of the Cartesian criterion. Are the zeros with real part less than -3?
A tale question it is still possible to respond by applying the Routh criterion to
polynomial obtained by substituting aλ, (λ−3)

(λ−3)2+ 9(λ−3) + 20 = 0⇒λ2+ 3λ + 2 = 0

The latter has zeros with negative real parts, which implies zeros with real parts.
less than -3 in the original polynomial.

With similar considerations, the Routh criterion could also be used.

it is used to verify if the zeros not only have a real part less than a
certain value, but even if they are found in a prescribed region of the plane com-
plexus, for example in a conical sector. In reality, verify if the zeros of a
polynomials contained in a region of that type is equivalent to verifying that
the damping of a possible pair of complex eigenvalues is greater
to a certain damping fixed and equal to sin θseθ`e the angle between the axes of
the ordinate and the edge of the cone. The verification of the belonging of the zeros to such
region can be done using variable transformations

λ←λejθ lambda equals lambda e−jθ

and applying the Routh criterion to the following polynomial with real coefficients

D(λ) = d(λe jθ )d(λe−jθ )

the number of sign changes divided by two indicates the zeros that are located at
outside of our sector.
252 8. Elements of stability theory

8.3. The general Lyapunov method*

Regarding the system

)xt(=
f)tẋ( f ( x e) = 0

The Lyapunov criterion extends, in a purely mathematical context, a

concept that is connected to energy considerations. If one imagines that...
Assigning an energy level to each state of the system, stability is associated.
to the decrease in the energy level along the evolutions of the system. Therefore
if you imagine associating with the statoxea zero energy level and furthermore
to associate with the states all around adxea higher energy level, to the
free evolutions that start from the states around adxandyou must correspond to a
decrease in energy.

The formalization of these aspects is based on the concept of function.

positive and negative definite.

A function is defined as positive in a neighborhood of x.e ,V(x)>0 inIx , oneself e

V(xe ) = 0 V(x) > 0 For all x in I, xe

a definite negative, V(x) < 0, if -V(x) > 0.

Example of positive definite functions in the vicinity of a point:

xe1
.. ⎞
xe= ⎛ .
⎝ xin ⎠
n
V(x) = 2
#i=1 (xI-xhi)
In the plane, n = 2,

(x1-xand)2+1 (x2-xe )2 2

and a positive definition in all space.

 8.3. The general method of Lyapunov* 253

If the strict inequality does not hold, but

V(x) ≥ 0

the function is said to be positive semidefinite

Statement of the criterion: the statoxeequilibria and asymptotically, locally

Stable if there exists a positive definite function V(x) in a neighborhood Ix, so that e

its derivative along the motion

n
∂V
V̇(x) = ẋi<0
#i=1 ∂x i

the negative definition.

d ∂V dx
V(x(t)) =
dt ∂x dt·
ẋ1 n
∂V∂V ∂V . ∂V
$ ... % ⎛ .. ⎞ V̇(x) = fi (x)<0
∂x 1∂x 2 ∂x n · #i=1 ∂x I
⎝ dxnot ⎠
The Lyapunov criterion says that essentially, if you are able to find a
function V(x) defined positive in Ixsuch that
e its derivative along the motion is
definite negativity in the same neighborhood then one can conclude about stability
local asymptotic of the equilibrium state.
If you have talked about asymptotic stability, in reality the criterion also allows for
establish under what conditions the equilibrium state is locally simple
˙
stable mind, in this case it is sufficient that (V) is negative semidefinite.

For example, consider

ẋ-x 2
1 1 x2
&
ẋ=−x 3
2 2

↓
 254 8. Elements of stability theory

x
xe= , -
0

0
, 0- =xestate of equilibrium

V(x) = x12+x22

V̇(x) = 2x1 ẋ+ ̇

1 2x22x

-2x21x22-2x42=−2x22(x12+x22)≤0
A positive semidefinite function is obtained, therefore the equilibrium state xe
baitssiehretytilaernI .ebaltsm
uym
ispilbrilqeuiofetatsasi ti
simple global of this state, because it is a generalization of the criterion
following: if the function V(x) is positive defined throughout the space, also the
function V(x) is radially unbounded, that is, when x tends to infinity
the function tends to infinity, and the function (V) is˙ defined as negative or semi-negative
negative in all space, then there is also simple, non-local stability
my global.
Another physical example we will address concerns the dynamics of the configuration.
a rigid body in the absence of gravity. This model is assumed to describe the
variation of a satellite's orientation with respect to an inertial reference.
We denote with ω1the angular velocity around the axis x, ω2that around
to the assembly, eω3quella attorno all’asseze immaginiamo che l’orientamento del
satellite can change as a result of pairs, τ1 ,τ2 ,τ3 , which are generated
around the principal axes of inertia, which we will assume to coincide with the axes
of the reference system. The equation of dynamics can be written

J1 ω̇=
1 (J2-J3 )ω2 ω3+τ1

J2 ω̇=(J
2 3−J1 )ω3 ω1+τ2

J3 ω̇(J
3 1-J2 )ω1 ω2+τ3

oveJ1 J2 J3These are the moments of inertia around the axes x, y, z. This mod-
Hello generalizes orientation with respect to an axis: one can imagine having a
 8.3. The general Lyapunov method* 255

metallic square in the plane, free from friction and that can be operated with a
pair around their center of mass, the equation of dynamics is

τ=ω̇

The problem to be solved is the following: let us imagine such an object in

rotation, how to stop it? It's easy to verify that with

τ = -kω

yes it has
−
=ω̇
kω
& ω(t)→0
ω(t) = e −kt ω(0)
So it is enough to apply a torque proportional to the angular velocity,
A dynamic friction would be sufficient.
The same result occurs in the general case. Indeed, if one sets

τi =−kI ωii= 1,2,3

ω→0. The verification of what is stated is easy if one substitutes into the equations of the
dynamics of the rigid body
τi=−k i ωi

and if we consider the function

1 1 1
V(x) = J1 ω21+J2 ω22+J3 ω23
2 2 2
Taking the derivative, it is easily verified that it is a defined function.
negative throughout the space with reference to the state ω = 0. In conclusion
V(x) is positive definite in the whole space with respect to the state ω = 0 and is
radially unlimited, and the (V) ˙that is calculated is a negative defined function
in all space, and this is enough to guarantee global asymptotic stability
of the zero state. Therefore, the body tends to stop.
256 8. Elements of stability theory

8.3.a. Lyapunov for stationary linear systems

With reference to the system

A
=)xtẋ( xe= 0

we want to verify the conditions of asymptotic stability, therefore with xe = 0

the only state of equilibrium.
As is known, the criterion requires that there exists
n
∂V
V(x) > 0 with V(x) =˙ ẋi<0
#i=1 ∂x i

In the study of stationary linear systems, it is sufficient to refer to

positive definite quadratic forms:

V(x) = x′ P x > 0

In that case, the matrix P, which without loss of generality can be

A symmetric matrix is called positive definite and is denoted by P > 0. The verification
To ensure that a matrix is defined as positive, one can use the cri-
Sylvester's theorem: necessary and sufficient condition for a matrix P
symmetric and positive definite, and that all its principal minors are positive
The main minors of a symmetric matrix are the determinants that
they obtain by considering all the submatrices constructed with respect to the diagonal
mainly in the following way:
.. .. ..
⎛ p11 . p12 . p13 . ... ⎞
......
⎜ .. .. ⎟
P= ⎜⎜ p21 p22 . p23 . ... ⎟
⎟
⎜ ............ ⎟
⎜ .. ⎟
⎜ p p23 p33 . ... ⎟
⎜ 13 ⎟
⎜ .................. ⎟
⎝ ⎠
One might wonder whether it is necessary to consider only similar matrices.
metrics?
 8.3. The general Lyapunov method* 257

In reality, when considering quadratic forms, if one does not assume the matrix
symmetric and it can be easily verified that the same result is obtained that
it would obtain if the symmetric matrix obtained fromPaccording to the
so-called symmetrization procedure:

P+P T
P̃=
2

in fact
P+P T
x′ P x = x ′ x
2
As an example, it is immediately verified that:

1
1 1 1 −2 x
x′ , - x=x
′
,
-2 2 −
1
2 2 -

This explains why one limits oneself, in quadratic forms, to assuming matrices.
symmetric. The positivity is then verified with Sylvester's criterion.

With the previous premises on quadratic forms, we can state the cri-
Lyapunov's theorem for stationary linear systems that provides a condition
necessary and sufficient for the verification of asymptotic stability.
The condition requires that the matrix equation

A′ Q + QA = -P

the symmetric and positive definite unknown has a unique solution for each
A symmetric and positive definite matrix.
So fixed to referee P, that in the applications of the criterion is assumed
the identity I, the equation in the unknown Q must be solved; such a solution must
to be unique, symmetric, and positive definite. If so, it can be affirmed
that the system is stable, otherwise it can be asserted that the system is not stable.
The consideration is therefore stronger than in the non-linear context in which
the impossibility of concluding from a positive definite function does not
allows declaring the absence of ownership.
258 8. Elements of stability theory

The sufficiency of the condition is demonstrable in a constructive way, in fact

for a fixed P, there exists the solution Q, we have

V(x) = x′ Qx > 0
′
V̇(x) = x˙Qx + x Qx˙′
V̇(x) = x A′ Qx′ + x QAx′
V̇(x) = x(A′ Q +′ QA)x
-x′ P x < 0

Therefore,˙V(x) is a defined negative function, and it is so throughout the space. Furthermore

˙
The radial VeV are unlimited and the zero state is asymptotically stable.
globally.

The demonstration of necessity, assuming that the eigenvalues of A have

negative real part, Re(λ)i )<0 consists in showing that the following matrix
and the sought solution ∞
′
Q= 0 eAtP e Atdt
0

The solution of the equation, as can be verified for direct substitution,

it is also symmetric, because it coincides with its transpose, in fact
∞
′
Q′ = 0 eA T P e ATdt
0

Furthermore, it is also positive because if we consider the associated quadratic form:

∞ ∞ ∞
′ ′
x′ Qx = x′ 0 (eA T P e AT)dtx= 0 x′ (eA TP e AT)xdt= 0 z′ P z d t
0 0 0

Where have you been?

z=eAT x

Since it is positive, there is also a unique solution; this follows from

gives a property of equations of the type that we are studying that
and so it is that all the authors have a real part
negative.
 8.4. Stability through linearization 259

The criterion provides a necessary and sufficient condition based on the resolution.
n(n+1)
of a system of 2 equations. There are as many equations as there are
unknown in a symmetric matrix of dimension n. For example, with

0 1
A= , -
-5-6
the matrix equation

0−5 q11q12 q11q12 0 1 -1 0

, 1−6 -, q12q22 - + , q12q22 -, - = , 0 -
-5-6 -1
it is rewritten

−1 =−5q12-5q12

0 = -5q22+q11-6q12

-1 = 2q12-12q22
These are three equations, in general.n(n+1), pern=
2 2.

Results equivalent to those studied so far for continuous-time systems

they apply to discrete-time systems.

8.4. Stability through linearization

The method relates to the use of linear approximation around one.

equilibrium state to study the stability of the equilibrium state non-system
linear data.
As is known, the linearization of a given system around a state of
balance
)xt(=
f)tẋ( f ( x e) = 0
260 8. Elements of stability theory

post
ξ(t) = x(t) - x e

provides
∂f 1 ∂f 1
⎛ ∂x 1 ··· ∂x n ⎞
˙ = Aξ(t)
ξ(t) A=J(x) =e ..
⎜ . ⎟
⎜ ∂f n ∂f n ⎟
⎜ ··· ⎟
⎝ ∂x 1 ∂x n ⎠

As we will see, the asymptotic stability of the linear approximation implies the
local asymptotic stability of the statee . It follows that it is necessary to verify
preliminarily the state of equilibriumehe is not late.
It is possible to demonstrate that the state of equilibriumefor the nonlinear system
he is this or that, so he
|J(xe )|̸= 0
This condition is not necessary as shown by the following simple example:

-3ẋ
x=

love, despitee = 0, the equilibrium state is unique (in reality the solution
it is the algebraic multiplicity of the eigenvalue.

To show that the asymptotic stability of the linear system implies stability
local asymptotics of the stateeof the given nonlinear system, let V(ξ) = ξ′ Qξ,
with solution of A′ Q + QA = -P, a Lyapunov function for the system
linear approximating. The proof of the local asymptotic stability of the system
nonlinear is based on the use of the same function

V(x) = (x−xe )′ Q(x−xe )

positively defined function with respect to xe , which has a definitively negative derivative in
an appropriate contextx. e

This means that the asymptotic stability of the linear system implies that one
local asymptotic stability of the equilibrium state for the nonlinear system. Furthermore, it
it can show that the instability of the linear system due to the presence of at least one
 8.4. Stability through linearization 261

Eigenvalue with a positive real part implies instability of the equilibrium state.
in the non-linear system, while nothing can be said in the case where the system
linear and in stability limit conditions or unstable due to mere presence
of eigenvalues with zero real part and geometric multiplicity greater than one.
This result concludes, as far as we are concerned, sufficient conditions.
stability coefficients. For example, if the linear approximation had a matrix

0
A= , -1
1−3 -

in this case the equilibrium state would be locally asymptotically stable;

if it were
J ( x e ) =A= -1 0
, 1 3-

the state of equilibrium would be unstable; if it were

-1 0 0 0
J ( x e ) =A= , orJ(xe ) =A=
1 0- , 1 0-

nothing could be said about the stability of the equilibrium state regarding
to nonlinear dynamics.
Here are some simple application examples.
Dynamics of the pendulum

Consider an inextensible pendulum of length l and mass m; the equation

that describes the dynamics is easily obtained by imposing the balance of the
forces along the tangent to the trajectory

mlθ¨ + klθ ˙+ mgsinθ = 0

post
θ=x1 θ̇=x2

the representation is obtained with the following state:

ẋ=x
1 2
262 8. Elements of stability theory

k g
ẋ=−x
2 2+ sin x1
m l
The calculation of equilibrium states of a

x2e= 0 sinx1e= 0 x1e=hπ

0
xe= ,
hπ -
It is noted that the equilibrium states 2hπ correspond to the pendulum in the
vertical position downward, while the equilibrium states (2h+ 1)π
correspond to the pendulum in the upper vertical position in a condition of
´ that the two types of equilibrium states have
initial velocity zero. It is clear
different properties: the vertical one, when disturbed, tends to be abandoned
Nato, the other conversely tends to be maintained if disturbed. The method
the linear approximation provides this result to support intuition.
The linear approximation corresponding to each of these states of
equilibrium can be calculated from the Jacobian matrix

0 1
J ( x ) = $ -g %'
cosx1 ' -k
l ' xe m
'
'
In the equilibrium states corresponding to the upward vertical position, one
ha
0 1
A= , g
-k -
l m
in the others, however:
0 1
Ā= $ -g %
-k
l m

With the dynamic matrix of this form, the study of stability is immediate.
as the elements of the last row represent the coefficients of the
characteristic polynomial changed signs. Therefore, in the first case

0 1 k g
A= $ g % →|λI−A|=λ 2+λ− characteristic polynomial
-k m l
l m
 8.4. Stability through linearization 263

With a variation in the coefficients, there is a separate eigenvalue.

positive real, which implies the instability of the equilibrium state with respect to the
nonlinear dynamics. In the second

0 1 k g
Ā= $ -g ¯
% →|λI−A|=λ+λ+ 2
-k m l
l m

with two eigenvalues with negative real parts, i.e., asymptotic stability of the approximation
linear and thus local asymptotic stability of the equilibrium position
responding to the pendulum downward.

Logistic equation
The following nonlinear differential equation

2-)=
kx(t)
axt()tẋ(

is taken on to represent the growth of a population in the presence of

limitations on resources. The solution to an equation of this type well
represent this phenomenon as intuition suggests because it predicts a
exponential growth trend in an initial phase, dominated by a rate
of constant growth and not conditioned by the limitation of resources, limited
in the continuous expansion and converging towards a steady value that depends
from the resources. In this equation, there are two equilibrium values, zero and the
value; akthis value is called load capacity.

x(a−kx) = 0

xe10
⎧ a
⎨ xe2 =
k
⎩
The stability analysis method through linearization can be used.
to study the stability of these two equilibrium states. The Jacobian matrix
in this case it is the scaling:

J ( x e ( ) = (a−2kx)x e
 264 8. Elements of stability theory

and calculated in the two equilibrium states becomes

for xe1if haJ(xe1 ) =a

e
per xe2if you want(xe2 ) =−a

The result is now evident: being a positive constant, the first state of
a
The equilibrium is unstable, while the state of equilibrium
k is stably asymptotic.
The previous considerations find further confirmation in the example that
next.
Predator-prey model
As already seen, the model suitable for representing the interaction between two species.
ypethet
ẋ(t)
1 = ax1 (t)−bx1 (t)x2 (t)
&
dx2ot(t) =−cx2 (t) +dx1 (t)x2 (t)
dovex1ex2they represent the density of the two populations and it has been assumed that
the prey population,x1 , grow, in the hypothesis of unlimited resources according to a
exponential growth with exponent a, while the predator population,
x2 , since the prey is its only nourishment, it decreases in the absence of interaction.
with the prey itself according to an exponential trend characterized by
coefficient - c. The coefficients a, b, c are all positive.

x1 (a - bx2 ) = 0
&
x2 (-c+dx1 ) = 0

The equilibrium states are easily calculated and equal to

a c
xe1 = xe2=.
b d

The application of the linearization method does not allow one to conclude,
how the eigenvalues of the approximation can be easily verified
lines have zero real part. Yes, therefore in a condition where one cannot
 8.4. Stability through linearization 265

infer nothing about the stability of the equilibrium state of the nonlinear system.

A different situation arises if the growth of the prey is described using...

solving the logistic equation. In this case,

dx1ot(t) = ax1 (t)−bx1 (t)x2 (t)−kx12(t)

&
ẋ(t)
2 = -cx2 (t) + dx1 (t)x2 (t)
with equilibrium states solutions of the system

& x1 (a−bx2-kx1 ) = 0
x2 (-c+dx1 ) = 0

⎧ (a−bx2-kx1 ) = 0 c
⎨ (−c+dx1 ) = 0→x1=
d
⎩
c
, a−bx2-k =0
d-

a kc
x2= −
b db
c
d
, a kc - =xe
b − bd
therefore, a new state of equilibrium appears which is positive when ad - kc > 0
or, what is the same, ka > cdThsi conoidnti si expersed by saynig htat hte
the carrying capacity of the prey species must be sufficiently high to
can react to the predator's prey capacity. This state of balance is
positive and has a physical meaning, in the context being examined, on
the reader is invited to reflect.
The application of the method with respect to this state of equilibrium allows for verifying
about its local asymptotic stability. The steps are left for exercise.
mathematically explains well a phenomenon that occurs in the presence of
Competition between two species, prey - predator, occurs in reality.
Following a disturbance from a given equilibrium condition
an oscillatory phenomenon arises that converges to the state of equilibrium.
266 8. Elements of stability theory

Two examples to solve

Leslie model
price dynamics
Leslie, 1948, represents the interaction between two species and uses the following
equations:
ẋ=ax 2
1 1-kx −bx1 x12
⎧ 2
⎨ ẋ=−cx ′ x2
2 2+d
x1
⎩
self
x1 d′
=
x2 c
The predator is in a state of equilibrium.
If it is verified that if
x1
x2
and the large predator grows, according to an exponential amendment,
it is very small, decreases, according to an exponential trend.

8.5. The Stability of Entry - Exit

At the completion of the study conducted, after having developed criteria and
internal stability methods, let's consider the input behavior
exit.
Under what conditions on the limited input system does it correspond
a limited exit? This property is also called external stability

With reference to a linear system

A
=
+
B)xutẋ(
&
y(t) = Cx(t) x(0) = x 0
 8.5. The Entry-Exit Stability 267

it is requested that any limited input corresponds to a limited output,

that is:

For all M,
0 there exists an N.
x ,M:|u(t)|< M⇒0 ∥y(t)∥< Nx ,M

two cases are distinguished: the first of stability in the zero state, related to
forced response when that ex0 = 0; a second of external stability in
any state, if this property is true ∀x0 this simply stability
external.
The conditions

1. ||ψ(t)|| < k1For all t

t
2. 0 ∥W(τ)∥dτ< k2For all t
0

they are necessary and sufficient for external stability.

The second sole necessary and sufficient condition for external stability in
zero state.
The limitedness of ψ(t) imposes that the free evolutions are limited, here
because this property comes into play only for external stability in each state.

The sufficiency of the indicated conditions is easily demonstrated starting from the expression
of the answer
t
y(t) = ψ(t)x 0+ 0 W(t−τ)u(τ)dτ
0

if you go to the norm

t

||y(t)|| = ||ψ(t)x||0 + 0 W(t−τ)u(τ)dτ∥

0

based on the elementary properties of the norms,

||a+b|| ≤ ||a||+||b||
268 8. Elements of stability theory

it is obtained:
t

||y(t)|| ≤ ||ψ(t)x||0 ∥+∥ 0 W(t−τ)u(τ)dτ∥

0

Still,
The norm of the dot product of a and b is less than or equal to the product of the norms of a and b.
implies
t
||y(t)|| ≤ ||ψ(t)|| ||x||0 ∥+ ∥W(t−τ)u(τ)∥dτ
0
0

And, infinite, with the simple further increase

t
||y(t)|| ≤ ||ψ(t)|| ||x||0 ∥+ ∥W(t−τ)∥∥u(τ)∥dτ
0
0

remembering that
||u(τ)|| < M
it concludes
t
||y(t)|| < ||ψ(t)|| ||x0 ∥+ ∥W(t−τ)∥dτ·M
0
0

and, for 1 and 2:

||y(t)|| < k1 ||x0 ∥+k2 M=Nx ,M 0

one obtains the limitation of the response for each initial state. If the state
initial is null the first part is not there and the same applies to the limitation of the y, only
for hypothesis 2.. Therefore condition 2 is sufficient for external stability
in a zero state.
The necessity is demonstrated with similar arguments, fundamentally the
The demonstration consists of verifying that, if conditions 1 and 2 are not met.
they are satisfied, unlimited output evolutions can be obtained. As far as
Note that if the matrix ψ(t) does not have a bounded norm, some among
its elements are not limited to what corresponds to the existence of evolutions
unlimited free. Therefore condition 1.`is also necessary. Similarly
If it is assumed that hypothesis 2 is not satisfied, it can be demonstrated that there exists
a limited entrance, and it is that entrance that manages to take those terms
which are possibly not limited, corresponding to an evolution
 8.5. The Stability of Entrance - Exit 269

forced in unlimited exit. Essentially, the two conditions are necessary and
sufficient. The formulation is given in terms of matrix norms. It is possible
understand how such conditions reflect on the characteristics of the system
particularly on the eigenvalues, this allows for establishing a connection between the different
types of stability studied.

External stability condition in the zero state

The condition in terms of eigenvalues is expressed

ΛEo ⊂C−

dove:C− indicates the open lower half-plane of complex numbers andΛ oE

indicate the set of eigenvalues that can be simultaneously excited with in-
impulsive and observable behaviors in output. To understand this formulation
just remember that inCeATThere are exponential laws that are associated
natural modes that are simultaneously excitable with input pulses,
that is, those contained in itATand observable outputs, that is those with-
tied inCeAT. Since W(t) = CeATBe a matter of the ice element is on or
linear combinations of exponential laws; the integral over an interval
infinite and finite only if such exponential laws have an exponent with a real part
strictly negative. The condition therefore states that the 'eigenvalues' that are
They must have a negative real part, that is, they must belong to the semi-
open negative piano.
An elementary example of an externally unstable system in state
1
zero`e constituted by the system characterized by the transfer function, s
an integrator, which has an equation of the type in differential terms

=
uẋ
&
y=x

In fact: ∞
W ( t ) = 1 →0 dt=∞
0

This responds to the fact that if a constant input is allowed into the integrator,
the integral tends to infinity as the transfer function has a zero
270 8. Elements of stability theory

in zero.

Condition of external stability in every state

The condition is expressed, in terms of eigenvalues, as

Λ1o⊂Ce− o ⊂C−
Λ>1

e
ΛEo ⊂C−
oveCe− indicate the closed negative complex half-plane and the subscript 1 e>1 the
one is greater than one respectively. The second condition has already been
Commentary corresponds to 2.. The first corresponds to 1. How easy it is.
check remembering that in free evolution out comes the auto-
values associated with observable exponential laws.

Λ={λ1 , . . . ,λ n }
This ATΛo={λo1, . . . ,λ or }
eATBΛE={λE1 , . . . ,λ Es }
This ATBΛEo =ΛE∩Λo
To understand how the first of the given conditions corresponds to the [Link]
note that the boundedness of the norm of ψ(t) corresponds to the boundedness
of its elements. These elements are a linear combination of exponential functions.
essentially, possibly with polynomial coefficients in the presence of eigenvalues
a geometric multiplicity greater than one. They are limited only if all the
the eigenvalues that appear to you, those that are observable, are with strictly real part
negative, if the geometric multiplicity is greater than one, with a negative real part
or nothing for others.
So condition 1 is equivalent to the following

Λ1o⊂Ce− o ⊂C−
Λ>1
 8.5. Entry-Exit Stability 271

These considerations made for continuous-time systems can be

this relates directly to discrete-time systems in terms of boundedness of
norms.
With reference to ψ(t), the condition is the same; the second has the same
This way, the summation replaces the integral. The conditions in terms
the eigenvalues are expressed in the same way when the modulus is substituted for the
real part with reference to 1.

Relations between internal and external stability

As is well known, the following conditions exist

L is a subset of C−

for internal asymptotic stability

ΛE0⊂C−

for external stability in the zero state

Λ10⊂Ce− Λ>1
0 ⊂C− Λ0 ⊂C−
E

for external stability

The obvious implications are:
internal asymptotic stability always implies external stability
external stability implies external stability in the zero state
Furthermore: the external stability inx0= 0 implies asymptotic internal stability if

ΛE0≡Λ
That is, all modes are excitable and observable. In fact, in such a case, the eigenvalues
which appear in W(t), which by hypothesis have a negative real part, are all
those of the dynamic matrix which is, therefore, asymptotically stable.
External stability0= 0 implies external stability if

ΛE≡Λ
272 8. Elements of stability theory

That is, the ways are all excitable. In fact:

ΛE≡Λ

ΛE0=Λ∩Λ0

so if all the eigenvalues are in C− and since Λ∩Λ0=Λ0 ,

Then all the conditions for external stability are met because the
Observable eigenvalues are all in C− so, in reality, she is satisfied
an even stronger condition.
In conclusion, this implication holds under the assumption that all modes are
excitable.
External stability implies asymptotic internal stability if

ΛE0≡Λ

that is, if all the modes are excitable and observable. If we limit ourselves to considering
the implication between external stability and non-asymptotic internal stability, then
the condition of observability of the modes is sufficient. In fact, if all the modes are
observables may eventually exist in the spectrum of eigenvalues at
real part zero, and therefore of constant or periodic motion laws. And this
it implies simple internal stability. In conclusion, external stability implies
the asymptotic internal stability is

ΛE0≡Λ

the simple internal stability if

Λ0≡Λ

It will conclude with an illustrative example. Consider the system:

-1 2 0 1
A= ⎛ 0 0 0⎞ B= ⎛ 0⎞ C= ( 1 1 0 )
⎝ -1 1 1 ⎠ ⎝ 0⎠
 8.5. The Stability of Ingress - Egress 273

λ1-1

λ2= 0

λ3equals 1

and carry out the analysis of the ways

0 2 0
(A−λ1 I)u 1 ⎛ 0 1 0 ⎞ u1= 0
⎝ -1 1 2 ⎠
0 2 0 2 2
⎛ 0 1 0⎞⎛⎞ 0 0 u1= ⎛ 0⎞
⎝ -1 1 2 ⎠⎝⎠ 1 ⎝ 1⎠
0 2 0
v′1 ⎛ 0 1 0 ⎞= 0 v′1(+21-1 0 )
⎝ -1 1 2 ⎠
perλ2
−a+ 2b= 0
-a + b + c = 0
-1 2 0
⎛ 0 0 0 ⎞ u 2= 0 a = 2b
⎝ −1 1 1 ⎠ -2b + b + c = 0
c=b

2
-1 2 0 u2= ⎛ 1⎞
v′2 ⎛ 0 0 0⎞
⎝ 1⎠
⎝ -1 1 1 ⎠ v′2( 0 1 0 )
perλ3
-2 2 0 0
⎛ 0−1 0 u⎞3= 0 u3= ⎛ 0⎞
⎝ -1 1 0⎠ ⎝ 1⎠

-2 2 0
v′3 ⎛ 0−1 0 = 0⎞ v3′ = (−20 1
1 )
⎝ -1 1 0⎠
274 8. Elements of Stability Theory

So:
Λ={−1; 0; 1}

ΛE{-1; 1}

Λ0 ={−1; 0}

ΛE0-1

Therefore, the system is internally unstable, because there is an eigenvalue at

positive real part, and stable input-output in the zero state because the only
The excitable and observable eigenvalue is equal to -1; it is infinitely externally stable.
being in L0 a zero eigenvalue but with unit multiplicity.

8.6. The stability of interconnected systems

The present lesson is dedicated to introducing a criterion for the study of

stability of interconnected systems.
To understand the generality of the criterion, one must refer to the calculation.
of the representation with the state of a system Starting from the representations-
design of subsystems. As will be remembered, starting from the flow graph of S,
opened some nodes to make the flow graph acyclic, a is identified
representation with the state of the system corresponding to the modified graph,
further income and further expenses

A
=ẋ+
B
x+
Bs
us
⎧
⎪⎨ y=Cx
p=Cp x
⎪⎩
therefore the overall representation with the state is:

A
=ẋ+
B
x+
BC
us p x
&
y = Cx
 8.6. The stability of interconnected systems 275

obtaining a system of the type:

A
=
(ẋ+BC
s p x + Bu
&
y = Cx

Starting from the dynamic matrices of the subsystemsi that is starting from the
(AI , Bi , Ci ), the matrix A is first identified and then, starting from this,
the dynamic matrix of the interconnected system is (A+B s Cp ).
Considering that an interconnected system is characterized
A directed acyclic flow graph has as eigenvalues the aggregate of the eigenvalues.
of the subsystems that make it up, which is equivalent to saying that the property of
equivalence of stability to the stability property of subsystems, is understood
that the change in stability is connected to the transition from A to A (A+B s Cp ).

The previous considerations suggest that from a point of view of the

stability of the interconnected system is equivalent to a system of the type:

A
=ẋ+
Bu
xs
&
y = -Cp x

in which a unit counter-reaction is made, that is, it is stated:

u=v-y

If, in fact, we set u = v - y, we obtain a system with a dynamic matrix.

(A+Bs Cp x). Therefore, the study of the stability of the overall system can
being referred to the study of the stability of a feedback system
unitary. This suggests the generality of the study.
The criterion we will address pertains to the investigation of the stability of the system.
pleasing starting from the knowledge of the stability properties of the system.
The prior information on the stability of the direct chain system is related
I had the number of eigenvalues with positive real parts. The investigation is conducted
using the polar representation, or rather the Nyquist diagram of
direct chain system. The method in question is known as the criterion of
Nyquist.
276 8. Elements of stability theory

It is necessary to make some observations regarding the following

stationary linear system that is assumed to be strictly causal (D= 0)

A
=ẋ+
B
xu
S: &
y = Cx

postou=−ysi obtains

A
=
(ẋ−BCx)+BV
S f: &
y = Cx

and it is put

dAP(s) =|sI−A|⇒characteristic polynomial in open loop

dC H (s) =|sI−A+BC|⇒characteristic polynomial in closed loop

The overall system will be stable when the second polynomial has no zeros.
the positive real part.
Observation 1.
dC H (s)
=|1 +F(s)|=|I+C(sI−A) −1 B|.
dAP(s)

F(s) transfer function of the system

|I−A+BC|=|sI−A||1 + (sI−A)|−1 BC
=|sI−A||1 +C(sI−A) −1 B|
dC H (s) =|sI−A+BC|
dAP(s) = |sI - A|
dCH(s) −1
= 1 + C(sI−A) B
dAP(s) ! "

Observation 2. Given a polynomial of complex variable, p(s), the vari-

phase action of the complex number that it represents when s = jωe
omega varies from -∞ to +∞ and is a pair (n-n) p -π-np π.
 8.6. The stability of interconnected systems 277

If, in fact, one poses

p(jω) = k(jω - p1 ) . . . (jω - p n )

it is understood that

∆ϕ (p) = overall phase variation

sdiolshetnihncagseephsaofonsitaivrhaow
sngiofsoecprhetnieram
sesnetaousveiprhetnad
we must distinguish two situations: the case of a zero with a real part
positive and the case of a zero with a negative real part.

The vector (jω−p i ) it admits indeed opposite phase variations perpIe−πe,

for the zeros with negative real part, it is easy to understand that the variation of
phase e^{-πi} clockwise.
So, if the polynomial is of degree np they are the zeros with positive real part,
yes, it has

∆ϕ (p) = (n−np )pi−np π

Because it turns out

dCH(s)
= 1 + C(sI - A)−1 B
dAP(s) ! "
yes it has
∆ϕ (1 +F) =∆ϕ + (dC H )−∆ϕ (dAP) =
(n−zp )π−zp π−(n−pp )p+pp π=
= (-zp+pp 2π
where it is indicated withpthe number of eigenvalues with a positive real part
disf and with conppthe number of eigenvalues with a real part greater than zero of S.
Obviously, the overall system is asymptotically stable if and only ifp =
0. Indicating with Nil the number of turns that the representative vector of 1 +
F(jω) completes around the zero of the complex plane, that is, the number of turns that
they relate to its phase change,

N=∆ϕ (1 + F)
278 8. Elements of stability theory

the Nyquist criterion is written

N=pp

and it is stated as a necessary and sufficient condition for asymptotic stability

The characteristic of a unit feedback system is that the number of revolutions that the
representative vector of 1 + F(jω) rotates around zero, or what it is
the same number of turns that the representative vector of F(jω) makes around
at the point (-1, j0), when ω varies from minus infinity to plus infinity it is equal
to the number of eigenvalues with positive real parts of the direct chain system.

Note that the criterion allows for evaluating the stability of the overall system.
starting from the knowledge of the stability properties of the chain system
direct and from the graphic examination of the frequency behavior.
In the particular case where in the direct chain system there isp= 0,
if it has the criterion in its so-called reduced form

N= 0.

Recalling the considerations made regarding the study of stability

one will not find it difficult to check that with reference to a system inter-
connected that presents more subsystems in a single feedback cycle, the
the study of stability is reduced to that of a system with unit feedback
in which it is assumed that cascading interconnection is present in the direct chain of
subsystems present in the cycle.

Some simple examples and the extension of the criterion to the case where
All eigenvalues have zero real part.
Consider the system described by the following transfer function
with gain, k, and time constants, τi , positive.

k
F(s) =
(1 + τ1 s)(1 +τ2 s)(1 +τ3 s)

The examination of the polar representation of F(jω) shown in the following figure
it allows to understand that the unit feedback system is stable if the point
 8.6. The stability of interconnected systems 279

(1 , j0) is external to the curve, as one would have N = 0 = pp , unstable at-

otherwise, as one would have N=−2̸=p p= 0. Note that it surrounds-
the limit at the point (-1, j0) by the curve depends, for fixed
time constant values, from the gain value k. In particular, at
As it grows, the module increases, which makes crossing possible. It is
usual in the use of the method refers to a drawn diagram and
evaluate the stability as the gain varies by modifying by an inverse factor
the scale on the axes, which corresponds to moving the point (−1, j0).

(-1, j0) +∞ 0
-∞

Figure 8.1

In many real-life situations, one deals with transfer functions that

they have singularities on the imaginary axis, in this case the diagram
polar cannot be closed, and if the diagram is not closed it cannot be
calculate the number of turns around the point (-1;j0)? Consider for example
the system:
k k>0
F(s) τ>0
(1 + τs)s

Seen in the complex plane, there will be a pole at zero and one at−1if one
τ builds
the image according to the polar diagram, as it approaches zero the curve
to infinity. But how do you calculate the number of turns around the point
Is the diagram open at (−1; j0)?
280 8. Elements of stability theory

-
0

(-1, j0) -∞
+∞

+
0
Figure 8.2

So the diagram needs to be closed and the result is different depending on the direction.
the diagram is closed. To overcome this ambiguity, one must proceed
as follows:
the polar diagram is no longer considered, but the Nyquist diagram, that is to say the
diagram showing how to construct the image according to the function F no longer
of the imaginary axis, but of the so-called Nyquist path: a path un-
it coincides with the imaginary axis at almost all points, but that around
At the singular points, the poles of the function are located on the imaginary axis, it leaves
of hooks that is to say it avoids them, and by convention it is established that this path
you leave the poles that have zero real part to the left, that is, you "consider" them
a negative real part. With this convention, even the eigenvalues have a part
Nothing will be considered in the negative complex half-plane. Note that
that, furthermore, with such an agreement,p= 0 does not correspond to a system in
a directly stable chain and not at the limit of stability.

Now consider this hooked path, to this corresponds an image.

according to the function F, which is a closed curve because of the singularity point
Is this the summary of the curve?
One must refer to the study of analytic functions and the fact that the
functionF(s) is a ratio of polynomials; therefore an analytic function that
 8.6. The stability of interconnected systems 281

induce conformal transformations, that is, those that preserve angles and directions
distance. So if in the domain to go from point A to point B, you
a hook traverses and a hook corresponds to a phase shift of π,
the same thing will happen in the image as well, so a rotation must be done
Moreover, the travel lines are maintained, which in this case leave
on the left the singular point and thus also the corresponding point, that is the
the image point must be left on the left. The image point is the point
improper that goes to infinity, this means that the closure of the diagram of
Nyquist must be done leaving the improper point on the left which is the image.
from the singular point.

Let's go back to the example; in this case, after the closure, it can be seen that the
Point (−1;j0) remains external, regardless of the gain value.

N=pp= 0

and the system is asymptotically stable.

Nyquist diagrams and case studies

In summary, we recall the results obtained so far regarding the Nyquist criterion.

If the system is connected directly characterized by its presentation with

the state, A, B, C, or from the transfer function F(s); it is knownp , the number
of eigenvalues with positive real part of S.

N=pp

the number of rotations the function F(jω) performs around the point
(−1;j0), when ω varies from −∞ to +∞, expresses the necessary and sufficient condition
asymptotic stability coefficient of the unity feedback system that has
in a direct chain.
We also saw: how the criterion allows for evaluation
stability as the gain of the direct chain system varies; in what way
It is possible to study stability in cases of non-unitary feedback.
282 8. Elements of stability theory

Example:
SystemS:
k(s + 1)
F(s) =
s(s−1)(s+10)
and, obviously
pp = 1

To construct the Nyquist diagram, reference will be made to the diagrams of

Bode, that which simplifies study. It is assumed, based on what has already been observed,
the unit gain, and consider the assigned transfer function in the
Bode form:
(s+ 1)
F(s) =−
10s(1−s)(1 + 10s )
These are three binomial terms with break frequencies of 1 and 10. The trend
the module is always decreasing as can be easily verified. As far as
regarding the phase, there are two contributions between zero and π2 with a pulse of
breakω= 1; one between zero and −π2 with break frequency ω = 10 and a
constant contribution equal to−3π This is 2 enough to understand that
−3π
2 e−πe which, for the placement of the pole at +1 and the zero
the phase varies between
in−1, a decade before the second pole, the phase reaches greater values
di−π. From the trend shown in the figure
+
0

(-1, j0) -∞
+∞

-
0
Figure 8.3
 8.6. The stability of interconnected systems 283

recalling the considerations made in the last lesson regarding the study
As the profit varies, it is understood that regarding stability
the unit feedback system will be unstable for gain values
lower than a critical value, k,∗ , stable for higher values.
How to evaluate∗ ?
For this purpose, the Routh criterion can be applied to the characteristic polynomial.
closed-loop system:

k(s + 1)
F(s) =
s(s−1)(s+10)
F(s)
W(s) =
1 + F(s)
k(s + 1)
W(s) =
s3+ 9s2+ (k−10)s+k
If a preliminary check is noted >10 it is necessary. The critical value may be
10 or any other potential greater value corresponding to the sign change
of some coefficient of the first column of the table. It is:

1 k−10
9 k
8k−90
9
k

and therefore the following conditions:

90
k>0 k>
8
what it implies
90
k∗ =
8
The unit feedback system for this value is at the limit of stability.

Further examples of application can be carried out for practice.

s+ 1
F(s) = unstable system in direct chain p p= 0
s2
284 8. Elements of stability theory

s+1
F(s) =
s 2+ 1
Applying the Nyquist criterion, it will be checked that for every positive value,
both unit feedback systems are asymptotically stable.
The following considerations are useful for deciding on stability in the case where
there are eigenvalues with zero real partAPe/o indCH. Sia jω0 a point
of the imaginary axis and indicate withAPedmCHthe multiplicity with which
that point and possible zero of the aforementioned polynomials. It is also assumed to be known
mAP. If 1 + F(jω0 ) =c̸= 0,mCH=mAPit will be simply established if
mAP= 1, instability itselfAP1. If F(jω0 ) =−1,m CHmAPand if it has
instability weAP≥0. If, infinite, F(jω) = ∞, mAP> mCHand it has been established
if the closure involves a rotation dimAPπ.

So far it has been seen how to decide on stability starting from the dia-
Nyquist diagram, but this diagram has been deduced from Bode's.
So some information about the stability of the feedback system is already
contained in the Bode diagram of the direct chain system. In fact, it is pos-
you can evaluate the stability conditions on the Bode diagrams. This will
we will refer to a direct chain system free of poles except for
positive real.
With reference to the case where the feedback system is stable, and therefore
that the point (−1;j0) is to the left of the crossing; it is possible to introduce
two parameters that account for the stability of the system.
One of them is the module of the F when the phase is -π; if it is less than one,
Negative dB, the system will be stable. The smaller it is than one, the greater.
it goes to stability. It is customary to give this parameter the value of the gain when
the phase e−π, the name margin of gain as it tells us how much we
can modify the system's gain to reach a situation that is
stability limit.
The other parameter that is perhaps more common to refer to is the so-called
phase margin corresponds to the phase variation necessary to lead to
stability limit.
mϕ =π−phaseF |F|=1
 8.7. The stability of discrete-time systems 285

With the introduction of these last aspects, the study of stability of a

the counter-reaction system is traced back, albeit in a particular case, to
an investigation on the Bode plots of the transfer function in the chain of-
Straight, therefore diagrams that characterize the frequency behavior.
of direct chain systems. Furthermore, the two introduced parameters measure, if
if there is stability, a margin that represents a specific of
bustiness as the parameters vary in relation to the stability property. by how much
the system is stable, more precisely it is

The profit margingquantitatively express the maintenance of the

stability in varying profit.

The phase marginϕ quantitatively express the maintenance of the sta-

varies with the open-loop cycle phase. This connection with the com-
Frequency shifting is very important from an engineering point of view.

8.7. The stability of discrete-time systems

8.7.a. Conditions and criteria for linear systems

It is understood that the definitions remain unchanged, the study conducted so far
It allows us to understand that the condition of asymptotic stability of systems
discrete-time linear and that the eigenvalues have a modulus less than one.
The verification can be conducted using different criteria; among these, those here
below exposed.
Stability conditions for discrete-time systems.

x(t+ 1) = Ax(t) + Bu(t)

xand=Axe

(A−I)xe = 0
286 8. Elements of stability theory

As for the condition of stability and asymptotic stability, its

the formulation is the same on the transition matrix

||φ(t)|| < k

e
lim∥φ(t)∥= 0
t approaches infinity

Since the transition matrix is:

φ(t) = At

in this case the limitation and the convergence to zero of the norm impose
that the eigenvalues belong to the unit circle.
So for the eigenvalues that are outside, there are divergent motions.
for those inside, there are converging motions, on the boundary there are modes
which can have constant amplitude if the multiplicity is one, if instead it is
greater than one, divergent motions occur again. The stability condition
Then it will be that the eigenvalues are all inside the circle, at most.
Simple eigenvalues are allowed at the border. For asymptotic stability
they must be strictly inside. In conclusion, even in this case
the stability condition corresponds to the allocation of the eigenvalues in a
fixed region of the complex plane. simple stability

1
& |λi|≤1
|λi1>|<1

asymptotic stability
|λI |<1
where it is indicated with the apostrophe 1 or >1 the unitary multiplicity or greater than one
of the self-value.
A first criterion is represented by Routh's criterion itself.
With reference to the polynomial

d(z) = an z n+an−1 z n−1+...+a0

 8.7. The stability of discrete-time systems 287

variable substitution
s+ 1
z←
s−1
it corresponds to performing a transformation from the unit circle to the half-plane
negative of the complex plane. On the polynomial it is possible to apply the
Routh's criterion and obtaining information on stability; in fact, the zeros in
Zeros with real part greater than zero correspond to zcon modulo>1.

A second criterion is the so-called Jury criterion, which consists, just like
Routh's, in verifying that among the coefficients of a table constructed at
This requires the verification of precise conditions.

Regarding the construction of the table

a0 a1 ... ... an
⎛ an an−1. . . . . . a0 ⎞
b0 b1 . . . . . . bn−1
⎜ ⎟
⎜ bn−1 ... ... b0 ⎟
⎜ ⎟
⎜ c0 . . . . . . c n−2 ⎟
⎜ ⎟
⎝ cn−2 ... ... c0 ⎠

..
.

t 0t 1t 2

where the coefficients are calculated as indicated below:

a0 a1
b0← ' anan−1 '
' '
' '
' '
aa
bn−1 ← ' 0 n '
' n a0
a '
' '
in general ' '
a0ak+1
bk← '' anan−k−1 '
'
' '
' '
The statement of Jury's criterion is as follows: necessary conditions and
sufficient for asymptotic stability are
288 8. Elements of stability theory

d(1) > 0
-1n d(-1) >0

|an |>|a 0 |
|b0 |>|bn−1 |
..
.

|t0 |>|t2 |

Example:
z 2+z+ 0.5

The first three conditions are preliminarily verified that do not depend on
coefficients of the table; thus, the table is constructed characterized by the
single line
1 1 0.5

and the satisfaction of the stability conditions is immediately verified.

As for continuous-time systems, an additional criterion for investigation on the
stability is represented by Lyapunov's criterion. It is a criterion of
great generality and potential for use not limited to the linear context.

8.7.b. The Lyapunov criterion

With reference to the discrete-time system

x(t+ 1) = f(x(t)) f ( x e ) =xe

the state of equilibriumandand the c a lm e n t e s t a b ile, as in t o t ic a m e n t e s t a b ile, s e e s is t e

a function V(x) defined positive in a certain neighborhood of the equilibrium state,
Ix,e and its difference before:

∆V(x) = V(x(t+1)) - V(x(t))

 8.7. The stability of discrete-time systems 289

e d e fi n it a n e g a t iv a in
e Ix.

The criterion is therefore stated in the same way and this time one must find
a function whose first derivative is defined (negative semidefinite). An-
once again the criterion of life under conditions that are only sufficient, in the
I feel that if in correspondence with V(x) defined as positive, the ∆V(x) is not the case.
Negative definite (semidefinite) cannot decide the stability of the state.
of balance.
Example:
x2 (t)
x (t + 1) =
⎧ 1 1 + x2 (t)2
⎪⎪
⎨ x1 (t)
x2 (t+ 1) =
⎪⎪ 1 +x2 (t)2
⎩
the state of equilibriume= 0 and posted:
V(x) = x12+x22

∆V(x) = Vf(x(t)) − V(x(t))

! 2 "
x2 x21
Vf(x(t)) = 2
+ =
! " (1 + x2 ) 2 (1 + x22)2
x2+x2 V(x)
= 1 222 =
(1 + x2 ) (1 + x22)2
V(x)
∆V(x) = -V(x)
(1 + x22)2

1
∆V(x) = $ % ≤0
-1V(x)
(1 + x22)2

This last quantity is negative semi-definite and therefore the equilibrium state.
qeuiebaltsyloblaglasi tiytilaerni ,etatsebaltsasi ti
the stated conditions of unlimited validity of the conditions apply throughout
the space.
Also with reference to discrete-time systems in the linear case, the cri-
terio is specified by creating necessary and sufficient conditions that are easy
verify.
290 8. Elements of stability theory

8.7.c. The Lyapunov criterion for stationary linear systems

discrete time

x(t+ 1) = Ax(t) xe=Axe(A−I)x e= 0

Necessary and sufficient condition for the asymptotic stability of the system
data, that the following equation

A′ QA−Q=−P

admit a unique solution in asymmetric and positive definite, for every P-

it is also symmetric and positive definite.
So the formulation is identical to the previous case, what changes is the
structure of the equation. Also in this case, it is generally fixed P=I; if
this equation does not have a solution, if the solution is not unique, if
not definitively positive, it can be concluded that the system is not asymptotic
stable.

For the demonstration, we proceed as in the case of time systems with-

continuous. For the sufficiency, it is fixed

V(x) =x′ Qx

∆V(x) = V(x(t + 1)) - V(x(t))

=x′ (t + 1)Qx(t + 1) - x′ (t)Qx(t)
=x′ A′ QAx−x′ Qx
=x′ (A′ QA−Q)x
-x′ P x

defining a positive quadratic equation′ It is a negative evaluation.

demonstrate the global asymptotic stability of the system.
 8.7. The stability of discrete-time systems 291

For the necessity, one proceeds in a formally similar manner to what has been seen.
in the case of continuous time showing that

∞
Q= A′k PA k
#0

the solution of the unique positive definite symmetric equation. Also in

n(n+1)
in this case the criterion refers to the resolution of 2 linear equations
dear.

The comparison between the Routh criterion and the Lyapunov criterion is interesting in
due to the complexity of calculations in terms of the number of operations. Obvious-
Unfortunately, everything depends on the structure of the dynamic matrix; in the presence of
A high number of null elements in the matrix can be advantageous.
apply the Routh criterion as it should not be forgotten that for ap-
To apply the Routh criterion, it is necessary to first calculate the polynomial.
characteristic.

You have seen in this paragraph how the Lyapunov criterion is applied.
larizzi in the linear context, continuous time and discrete time, giving life to
necessary and sufficient conditions. This does not happen in the non-linear context where
the possibility of achieving a positive verification of stability is linked to
calculation of a function that satisfies the known conditions. This aspect makes
the application of the method is difficult in cases where the verification has not occurred
positive outcome using the defined positive elementary quadratic functions.
What to do when you cannot directly study stability
of a nonlinear system applying Lyapunov's criterion? Are there others?
methods?
In reality, another method, moreover with a complementary role to that
carried out from the Lyapunov criterion, it is based on the study of stability
of the nonlinear system based on the linear approximation. This will be
the subject of the next lesson.
292 8. Elements of stability theory

8.7.d. Study of stability through linearization

The method is related to the use of linear approximation around

a state of equilibrium to study the stability of an equilibrium state of a
non-linear system.
As is known, the linearization of a given system around a state
of balance
Algorithm for calculating the square root
The following equation
x2 -a = 0
It can be solved for a fixed point using the following iterative algorithm.

x(t + 1) = x(t) + a - x2 (t)

According to this algorithm, the state

xe = √ a

and the balance, as immediate verification. The stability of this verified

equilibrium state is equivalent to verifying the convergence of the algorithm for the
√ ensures that
calculation of the square root. If there is stability around a, this
the implementation of the algorithm allows for convergence towards the solution:
√ a.

Such issues are referred to as stability problems of a

algorithm.
If one proceeds with the use of the study method through linearization,
it is obtained
J ( x e ) = (1−2a) √

⇓
0 < a < 1 if and only if |1 - 2a|√< 1

From this analysis it is derived that for 0 < a < 1 the solution xeto be √
certainly calculated with the given algorithm.
 8.7. The stability of discrete-time systems 293

The obtained result has a local validity and the maximum remains undefined.
Deviation allowed from the initial condition with respect to the true value (that one
of equilibrium). More precise information about any deviation,
stability region, can be obtained using the Lyapunov criterion with

V(x) = |a - x|2 |

The calculation of V(x(t+ 1)) and ∆(V(x)) allows for easy verification.
if we assume 0 < a < 1 that
0≤x≤1
defines the region of stability.
Two examples to be carried out
Leslie model
price dynamics
The second example generalizes the example of price dynamics in...
Equilibrium conditions in the presence of a single good. Let us consider a market
characterized by the presence of two assets
d(t+ 1) = Ap(t+ 1) + d0
&
o(t+ 1) = Ep(t) + o0
doveA,E,d0eo0they are matrices and vectors of dimension two whose coefficients
they allow for simple interpretations in terms of production and demand.
equilibrium conditions it is possible to establish a dynamics equation of
price according to a standard report.

p(t + 1) = A−1 Ep(t) + A−1 (o0-d0 )

d β
d= , 1 - A= , −α -
d2 γ −δ
α, β, γ, δ > 0
Assuming E=Study the previous equation and identify which
these are the conditions for the equilibrium to be asymptotically stable. It is about
of a stability analysis of a discrete-time linear system, and admits
an interpretation in terms of the coefficients α, β, γ, δ that have an interest-
health meaning in economic terms.