Communications and Control Engineering

Published titles include:

Stability and Stabilization of Innite Positive 1D and 2D Systems
Dimensional Systems with Applications Tadeusz Kaczorek
Zheng-Hua Luo, Bao-Zhu Guo and
Identication and Control Using
Omer Morgul
Volterra Models
Nonsmooth Mechanics (Second edition) Francis J. Doyle III, Ronald K. Pearson
Bernard Brogliato and Bobatunde A. Ogunnaike
Nonlinear Control Systems II Non-linear Control for Underactuated
Alberto Isidori Mechanical Systems
Isabelle Fantoni and Rogelio Lozano
L2 -Gain and Passivity Techniques in Nonlinear
Control Robust Control (Second edition)
Arjan van der Schaft Jrgen Ackermann
Control of Linear Systems with Regulation Flow Control by Feedback
and Input Constraints Ole Morten Aamo and Miroslav Krstic
Ali Saberi, Anton A. Stoorvogel and
Learning and Generalization
Peddapullaiah Sannuti
(Second edition)
Robust and H Control Mathukumalli Vidyasagar
Ben M. Chen
Constrained Control and Estimation
Computer Controlled Systems Graham C. Goodwin, Mara M. Seron
Em N. Rosenwasser and Bernhard P. and Jos A. De Don
Lampe
Randomized Algorithms for Analysis
Control of Complex and Uncertain Systems and Control of Uncertain Systems
Stanislav V. Emelyanov and Sergey K. Roberto Tempo, Giuseppe Calaore
Korovin and Fabrizio Dabbene
Robust Control Design Using H Methods Switched Linear Systems
Ian R. Petersen, Valery A. Ugrinovski Zhendong Sun and Shuzhi S. Ge
and Andrey V. Savkin
Subspace Methods for System Identication
Model Reduction for Control System Design Tohru Katayama
Goro Obinata and Brian D.O. Anderson
Digital Control Systems
Control Theory for Linear Systems Ioan D. Landau and Gianluca Zito
Harry L. Trentelman, Anton Stoorvogel
Multivariable Computer-controlled Systems
and Malo Hautus
E.N. Rosenwasser and B.P. Lampe
Functional Adaptive Control
Simon G. Fabri and Visakan Kadirkamanathan
Bernard Brogliato, Rogelio Lozano,
Bernhard Maschke and Olav Egeland

Dissipative Systems
Analysis and Control
Theory and Applications
2nd Edition

With 94 Figures

123
Bernard Brogliato, PhD Rogelio Lozano, PhD
INRIA Rhone-Alpes Heuristique et Diagnostic des
BIBOP Project Systmes Complexes
ZIRST UMR-CNRS 6599
Montbonnot Universit de Technologie de Compigne
38334 St. Ismier Cedex Centre de Recherche de Royalieu
France BP 20529
60200 Compigne
Bernhard Maschke, PhD France
Laboratoire dAutomatique et de
Gnie des Procds Olav Egeland, PhD
Universit Claude Bernard Lyon1 Department of Engineering Cybernetics
UFR Gnie Electrique et des Procds NTNU
CNRS UMR 5007 Faculty of Information Technology,
CPE Lyon Mathematics and Electrical Engineering
69622 Villeurbanne Cedex NO-7491 Trondheim
France Norway

Series Editors
E.D. Sontag M. Thoma A. Isidori J.H. van Schuppen

British Library Cataloguing in Publication Data

Dissipative systems analysis and control : theory and
applications. - 2nd ed. - (Communications and control
engineering)
1. System analysis 2. Automatic control - Mathematics
I. Brogliato, Bernard, 1963-
629.8
ISBN-13: 9781846285165
ISBN-10: 184628516X
Library of Congress Control Number: 2006932284
Communications and Control Engineering Series ISSN 0178-5354
ISBN-10: 1-84628-516-X 2nd edition e-ISBN 1-84628-517-8 2nd edition Printed on acid-free paper
ISBN-13: 978-1-84628-516-5 2nd edition
ISBN 1-85233-285-9 1st edition
Springer-Verlag London Limited 2007
First published 2000, Second edition 2007
MATLAB and Simulink are registered trademarks of The MathWorks, Inc., 3 Apple Hill Drive, Natick,
MA 01760-2098, U.S.A. http://www.mathworks.com
Apart from any fair dealing for the purposes of research or private study, or criticism or review, as
permitted under the Copyright, Designs and Patents Act 1988, this publication may only be reproduced,
stored or transmitted, in any form or by any means, with the prior permission in writing of the
publishers, or in the case of reprographic reproduction in accordance with the terms of licences issued
by the Copyright Licensing Agency. Enquiries concerning reproduction outside those terms should be
sent to the publishers.
The use of registered names, trademarks, etc. in this publication does not imply, even in the absence of
a specic statement, that such names are exempt from the relevant laws and regulations and therefore
free for general use.
The publisher makes no representation, express or implied, with regard to the accuracy of the infor-
mation contained in this book and cannot accept any legal responsibility or liability for any errors or
omissions that may be made.
987654321
Springer Science+Business Media
springer.com
Preface

Thank you for your interest in the second edition of our book on dissipative
systems. The rst version of this book has been improved and augmented in
several directions (mainly by the rst author supported by the second and
third authors of the second version). The link between dissipativity and op-
timal control is now treated in more detail, and many proofs which were not
provided in the rst edition are now given in their entirety, making the book
more self-contained. One diculty one encounters when facing the literature
on dissipative systems is that there are many dierent denitions of dissi-
pativity and positive real transfer functions (one could say a proliferation),
many dierent versions of the same fundamental mathematical object (like
the Kalman-Yakubovich-Popov Lemma), and it is not always an easy task to
discover the links between them all. One objective of this book is to present
those notions in a single volume and to try, if possible, to present their rela-
tionships in a clear way. Novel sections on descriptor (or singular) systems,
discrete-time linear and nonlinear systems, some types of nonsmooth systems,
viscosity solutions of the KYP Lemma set of equations, time-varying systems,
unbounded dierential inclusions, evolution variational inequalities, hyper-
stability, nonlinear H , input-to-state stability, have been added. Conditions
under which the Kalman-Yakubovich-Popov Lemma can be stated without
assuming the minimality of the realization are provided in a specic section.
Some general results (like well-posedness results for various types of evolution
problems encountered in the book, denitions, matrix algebra tools, etc.) are
presented in the Appendix, and many others are presented in the main text
when they are needed for the rst time. We thank J. Collado and S. Hadd
who made us some remarks, and we remain of course open to any comments
that may help us continue to improve our book.

M ontbonnot, April 2006 Bernard Brogliato

Contents

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Example 1: System with Mass Spring and Damper . . . . . . . . . . . 2
1.2 Example 2: RLC Circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Example 3: A Mass with a PD Controller . . . . . . . . . . . . . . . . . . . 5
1.4 Example 4: Adaptive Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Positive Real Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.1 Dynamical System State-space Representation . . . . . . . . . . . . . . 10
2.2 Denitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Interconnections of Passive Systems . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 Linear Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.5 Passivity of the PID Controllers . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.6 Stability of a Passive Feedback Interconnection . . . . . . . . . . . . . . 24
2.7 Mechanical Analogs for PD Controllers . . . . . . . . . . . . . . . . . . . . . 25
2.8 Multivariable Linear Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.9 The Scattering Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.10 Impedance Matching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.11 Feedback Loop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.12 Bounded Real and Positive Real Transfer Functions . . . . . . . . . . 36
2.13 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.13.1 Mechanical Resonances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.13.2 Systems with Several Resonances . . . . . . . . . . . . . . . . . . . . 50
2.13.3 Two Motors Driving an Elastic Load . . . . . . . . . . . . . . . . . 51
2.14 Strictly Positive Real (SPR) Systems . . . . . . . . . . . . . . . . . . . . . . 53
2.14.1 Frequency Domain Conditions for a Transfer Function
to be SPR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.14.2 Necessary Conditions for H(s) to be PR (SPR) . . . . . . . 56
2.14.3 Tests for SPRness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.14.4 Interconnection of Positive Real Systems . . . . . . . . . . . . . 57
2.14.5 Special Cases of Positive Real Systems . . . . . . . . . . . . . . . 58
2.15 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
viii Contents

2.15.1 SPR and Adaptive Control . . . . . . . . . . . . . . . . . . . . . . . . . 62

2.15.2 Adaptive Output Feedback . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.15.3 Design of SPR Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

3 Kalman-Yakubovich-Popov Lemma . . . . . . . . . . . . . . . . . . . . . . . . 69
3.1 The Positive Real Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.1.1 PR Transfer Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.1.2 A Digression to Optimal Control . . . . . . . . . . . . . . . . . . . . 76
3.1.3 Duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.1.4 Positive Real Lemma for SPR Systems . . . . . . . . . . . . . . . 79
3.1.5 Descriptor Variable Systems . . . . . . . . . . . . . . . . . . . . . . . . 91
3.2 Weakly SPR Systems and the KYP Lemma . . . . . . . . . . . . . . . . . 95
3.3 KYP Lemma for Non-minimal Systems . . . . . . . . . . . . . . . . . . . . . 100
3.3.1 Spectral Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
3.3.2 Sign-controllability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
3.3.3 State Space Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . 106
3.3.4 A Relaxed KYP Lemma for SPR Functions with
Stabilizable Realization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.4 SPR Problem with Observers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
3.5 The Feedback KYP Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
3.6 Time-varying Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
3.7 Interconnection of PR Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
3.8 Positive Realness and Optimal Control . . . . . . . . . . . . . . . . . . . . . 119
3.8.1 General Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
3.8.2 Least Squares Optimal Control . . . . . . . . . . . . . . . . . . . . . . 120
3.8.3 The Popov Function and the KYP Lemma LMI . . . . . . . 125
3.8.4 A Recapitulating Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 129
3.8.5 On the Design of Passive LQG Controllers . . . . . . . . . . . . 130
3.8.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
3.8.7 A Digression on Semidenite Programming Problems . . 134
3.9 The Lure Problem (Absolute Stability) . . . . . . . . . . . . . . . . . . . . 135
3.9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
3.9.2 Well-posedness of ODEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
3.9.3 Aizermans and Kalmans Conjectures . . . . . . . . . . . . . . . 140
3.9.4 Multivalued Nonlinearities . . . . . . . . . . . . . . . . . . . . . . . . . . 142
3.9.5 Dissipative Evolution Variational Inequalities . . . . . . . . . 152
3.10 The Circle Criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
3.10.1 Loop Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
3.11 The Popov Criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
3.12 Discrete-time Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
3.12.1 The KYP Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
3.12.2 The Tsypkin Criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
3.12.3 Discretization of PR Systems . . . . . . . . . . . . . . . . . . . . . . . 175
 Contents ix

4 Dissipative Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177

4.1 Normed Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
4.2 Lp Norms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
4.2.1 Relationships Between L1 , L2 and L Spaces. . . . . . . . . 180
4.3 Review of Some Properties of Lp Signals . . . . . . . . . . . . . . . . . . . 180
4.3.1 Example of Applications of the Properties of Lp
Functions in Adaptive Control . . . . . . . . . . . . . . . . . . . . . . 186
4.3.2 Linear Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
4.3.3 Induced Norms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
4.3.4 Properties of Induced Norms . . . . . . . . . . . . . . . . . . . . . . . . 188
4.3.5 Extended Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
4.3.6 Gain of an Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
4.3.7 Small Gain Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
4.4 Dissipative Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
4.4.1 Denitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
4.4.2 The Signication of . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
4.4.3 Storage Functions (Available, Required Supply) . . . . . . . 201
4.4.4 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
4.4.5 Regularity of the Storage Functions . . . . . . . . . . . . . . . . . . 217
4.5 Nonlinear KYP Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
4.5.1 A Particular Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
4.5.2 Nonlinear KYP Lemma in the General Case . . . . . . . . . . 223
4.5.3 Time-varying Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
4.5.4 Nonlinear-in-the-input Systems . . . . . . . . . . . . . . . . . . . . . . 230
4.6 Dissipative Systems and Partial Dierential Inequalities . . . . . . 231
4.6.1 The linear invariant case . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
4.6.2 The Nonlinear Case y = h(x) . . . . . . . . . . . . . . . . . . . . . . . 235
4.6.3 The Nonlinear Case y = h(x) + j(x)u . . . . . . . . . . . . . . . . 238
4.6.4 Recapitulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
4.6.5 Inverse Optimal Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
4.7 Nonlinear Discrete-time Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 247
4.8 PR tangent system and dissipativity . . . . . . . . . . . . . . . . . . . . . . . 249
4.9 Innite-dimensional Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
4.9.1 An Extension of the KYP Lemma . . . . . . . . . . . . . . . . . . . 252
4.9.2 The Wave Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
4.9.3 The Heat Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
4.10 Further Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255

5 Stability of Dissipative Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257

5.1 Passivity Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
5.1.1 One-channel Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
5.1.2 Two-channel Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
5.1.3 Lossless and WSPR Blocks Interconnection . . . . . . . . . . . 263
5.1.4 Large-scale Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
5.2 Positive Deniteness of Storage Functions . . . . . . . . . . . . . . . . . . 266
x Contents

5.3 WSPR Does not Imply OSP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270

5.4 Stabilization by Output Feedback . . . . . . . . . . . . . . . . . . . . . . . . . 272
5.4.1 Autonomous Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
5.4.2 Time-varying Nonlinear Systems . . . . . . . . . . . . . . . . . . . . 273
5.4.3 Evolution Variational Inequalities . . . . . . . . . . . . . . . . . . . 274
5.5 Equivalence to a Passive System . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
5.6 Cascaded Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
5.7 Input-to-State Stability (ISS) and Dissipativity . . . . . . . . . . . . . . 282
5.8 Passivity of Linear Delay Systems . . . . . . . . . . . . . . . . . . . . . . . . . 288
5.8.1 Systems with State Delay . . . . . . . . . . . . . . . . . . . . . . . . . . 288
5.8.2 Interconnection of Passive Systems . . . . . . . . . . . . . . . . . . 290
5.8.3 Extension to a System with Distributed State Delay . . . 291
5.8.4 Absolute Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
5.9 Nonlinear H Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
5.9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
5.9.2 Closed-loop Synthesis: Static State Feedback . . . . . . . . . . 300
5.9.3 Closed-loop Synthesis: PR Dynamic Feedback . . . . . . . . . 302
5.9.4 Nonlinear H . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
5.9.5 More on Finite-power-gain Systems . . . . . . . . . . . . . . . . . . 307
5.10 Popovs Hyperstability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310

6 Dissipative Physical Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315

6.1 Lagrangian Control Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
6.1.1 Denition and Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
6.1.2 Simple Mechanical Systems . . . . . . . . . . . . . . . . . . . . . . . . . 324
6.2 Hamiltonian Control Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326
6.2.1 Input-output Hamiltonian Systems . . . . . . . . . . . . . . . . . . 326
6.2.2 Port Controlled Hamiltonian Systems . . . . . . . . . . . . . . . . 331
6.3 Rigid JointRigid Link Manipulators . . . . . . . . . . . . . . . . . . . . . . . 340
6.3.1 The Available Storage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
6.3.2 The Required Supply . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342
6.4 Flexible JointRigid Link Manipulators . . . . . . . . . . . . . . . . . . . . 343
6.4.1 The Available Storage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346
6.4.2 The Required Supply . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346
6.5 A Bouncing System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
6.6 Including Actuator Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
6.6.1 Armature-controlled DC Motors . . . . . . . . . . . . . . . . . . . . . 349
6.6.2 Field-controlled DC Motors . . . . . . . . . . . . . . . . . . . . . . . . . 354
6.7 Passive Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
6.7.1 Systems with Holonomic Constraints . . . . . . . . . . . . . . . . . 358
6.7.2 Compliant Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
6.8 Nonsmooth Lagrangian Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
6.8.1 Systems with C 0 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . 363
6.8.2 Systems with BV Solutions . . . . . . . . . . . . . . . . . . . . . . . . . 365
 Contents xi

7 Passivity-based Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373

7.1 Brief Historical Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
7.2 The Lagrange-Dirichlet Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 375
7.2.1 Lyapunov Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375
7.2.2 Asymptotic Lyapunov Stability . . . . . . . . . . . . . . . . . . . . . 376
7.2.3 Invertibility of the Lagrange-Dirichlet Theorem . . . . . . . 378
7.2.4 The Lagrange-Dirichlet Theorem for Nonsmooth
Lagrangian Systems (BV Solutions) . . . . . . . . . . . . . . . . . . 379
7.2.5 The Lagrange-Dirichlet Theorem for Nonsmooth
Lagrangian Systems (C 0 Solutions) . . . . . . . . . . . . . . . . . . 384
7.2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
7.3 Rigid JointRigid Link Systems: State Feedback . . . . . . . . . . . . . 386
7.3.1 PD Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 386
7.3.2 PID Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
7.3.3 More about Lyapunov Functions and the Passivity
Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393
7.3.4 Extensions of the PD Controller for the Tracking Case . 398
7.3.5 Other Types of State Feedback Controllers . . . . . . . . . . . 405
7.4 Rigid JointRigid Link: Position Feedback . . . . . . . . . . . . . . . . . . 408
7.4.1 P + Observer Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 408
7.4.2 The Paden and Panja + Observer Controller . . . . . . . . . . 410
7.4.3 The Slotine and Li + Observer Controller . . . . . . . . . . . . 412
7.5 Flexible JointRigid Link: State Feedback . . . . . . . . . . . . . . . . . . 414
7.5.1 Passivity-based Controller: The Lozano and Brogliato
Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 414
7.5.2 Other Globally Tracking Feedback Controllers . . . . . . . . 418
7.6 Flexible JointRigid Link: Output Feedback . . . . . . . . . . . . . . . . 422
7.6.1 PD Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
7.6.2 Motor Position Feedback . . . . . . . . . . . . . . . . . . . . . . . . . . . 424
7.7 Including Actuator Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
7.7.1 Armature-controlled DC Motors . . . . . . . . . . . . . . . . . . . . . 426
7.7.2 Field-controlled DC Motors . . . . . . . . . . . . . . . . . . . . . . . . . 428
7.8 Constrained Mechanical Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 428
7.8.1 Regulation with a Position PD Controller . . . . . . . . . . . . 429
7.8.2 Holonomic Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 430
7.8.3 Nonsmooth Lagrangian Systems . . . . . . . . . . . . . . . . . . . . . 431
7.9 Controlled Lagrangians . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 432

8 Adaptive Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435

8.1 Lagrangian Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 436
8.1.1 Rigid JointRigid Link Manipulators . . . . . . . . . . . . . . . . 436
8.1.2 Flexible JointRigid Link Manipulators: The Adaptive
Lozano and Brogliato Algorithm . . . . . . . . . . . . . . . . . . . . 442
8.1.3 Flexible JointRigid Link Manipulators: The
Backstepping Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . 452
xii Contents

8.2 Linear Invariant Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 456

8.2.1 A Scalar Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 456
8.2.2 Systems with Relative Degree r = 1 . . . . . . . . . . . . . . . . . . 457
8.2.3 Systems with Relative Degree r = 2 . . . . . . . . . . . . . . . . . . 460
8.2.4 Systems with Relative Degree r 3 . . . . . . . . . . . . . . . . . . 461

9 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 467

9.1 Flexible Joint Manipulators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 467
9.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 467
9.1.2 Controller Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 468
9.1.3 The Experimental Devices . . . . . . . . . . . . . . . . . . . . . . . . . . 469
9.1.4 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473
9.1.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483
9.2 Stabilization of the Inverted Pendulum . . . . . . . . . . . . . . . . . . . . . 496
9.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 496
9.2.2 Systems Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 497
9.2.3 Stabilizing Control Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . 500
9.2.4 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 503
9.2.5 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 503
9.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 504

A Background Material . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 507

A.1 Lyapunov Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 507
A.1.1 Autonomous systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 507
A.1.2 Non-autonomous Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 511
A.2 Dierential Geometry Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515
A.2.1 Normal Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 517
A.2.2 Feedback Linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 518
A.2.3 Stabilization of Feedback Linearizable Systems . . . . . . . . 519
A.2.4 Further Reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 520
A.3 Viscosity Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 520
A.4 Algebraic Riccati Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 523
A.4.1 Reduced Riccati Equation for WSPR Systems . . . . . . . . . 525
A.5 Some Useful Matrix Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 531
A.5.1 Results Useful for the KYP Lemma LMI . . . . . . . . . . . . . 531
A.5.2 Inverse of Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 533
A.5.3 Jordan Chain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 534
A.5.4 Auxiliary Lemmas for the KYP Lemma Proof . . . . . . . . . 534
A.6 Well-posedness Results for State Delay Systems . . . . . . . . . . . . . 537

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 539

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 571
Notation

IR: the set of real numbers; C the set of complex numbers; IN : the set of
nonnegative integers.
IRn (Cn ): the set of ndimensional vectors with real (complex) entries.
AT : transpose of the matrix A IRnm or Cnm .
A: conjugate of the matrix A Cnm .
A
: conjugate transpose matrix of the matrix A Cnm .
A > 0 ( 0): positive denite (semi positive denite) matrix.
(A): an eigenvalue of A IRnm .
(A) the set of eigenvalues of A IRnm (i.e. the spectrum of A).
max (A), min (A): the largest and smallest eigenvalue of the matrix A,
respectively.
max (A) (min (A)): largest (smallest) singular value of A.
(A): the spectral radius of A, i.e. max{|| : (A)}.
tr(A): the trace of the matrix A.
A : the Moore-Penrose inverse of the matrix A.
ODE: Ordinary Dierential Equation; PDE: Partial Dierential equation.
BV, LBV, RCLBV: Bounded Variation, Local BV, Right Continuous LBV.
AC: Absolutely Continuous.
In the n n identity matrix, On the n n zero matrix.
f
x (x) IRmn : the jacobian of the function f : IRn IRm at x.
f (x) IRnm : the euclidean gradient of the function f : IRn IRm at
f T
x ( f (x) = x (x)).
x iv Notation

A function is said to be smooth if it is innitely dierentiable; C 0 denotes

the set of continuous functions; C r denotes the set of r-times dierentiable
functions f () with f (r) () continuous.
f (t+ ): right-limit of the function f () at t; (f (t ): left-limit).

|| ||: Euclidean norm in IRn (||x|| = xT x for all x IRn ).
||f ||p : Lp -norm of a Lebesgue integrable function f ().
 H(j): the phase of H(j) C.
LTI: Linear Time Invariant (system).
Ker(A): kernel of A IRnm ; Im(A): image of A IRnm .
dom(f ): domain of a function f .
K: closure of a domain K IRn (K = K if and only if K is closed).
Int(K): interior of a domain K IRn (Int(K) is always open), i.e. the set
of interior points of K (points x of K such that there is a neighborhood
of x inside K).
Re[] denotes the real part and Im[] denotes the imaginary part.
a.e.: almost everywhere (usually in the Lebesgue measure sense).
1
Introduction

Dissipativity theory gives a framework for the design and analysis of control
systems using an input-output description based on energy-related considera-
tions. Dissipativity is a notion which can be used in many areas of science, and
it allows the control engineer to relate a set of ecient mathematical tools to
well known physical phenomena. The insight gained in this way is very useful
for a wide range of control problems. In particular the input-output descrip-
tion allows for a modular approach to control systems design and analysis.
The main idea behind this is that many important physical systems have
certain input-output properties related to the conservation, dissipation and
transport of energy. Before introducing precise mathematical denitions we
will somewhat loosely refer to such input-output properties as dissipative
properties, and systems with dissipative properties will be termed dissipa-
tive systems. When modeling dissipative systems it may be useful to develop
the state-space or input-output models so that they reect the dissipativity of
the system, and thereby ensure that the dissipativity of the model is invariant
with respect to model parameters, and to the mathematical representation
used in the model. The aim of this book is to give a comprehensive presenta-
tion of how the energy-based notion of dissipativity can be used to establish
the input-output properties of models for dissipative systems. Also it will be
shown how these results can be used in controller design. Moreover, it will
appear clearly how these results can be generalized to a dissipativity theory
where conservation of other physical properties, and even abstract quantities,
can be handled.
Models for use in controller design and analysis are usually derived from
the basic laws of physics (electrical systems, dynamics, thermodynamics).
Then a controller can be designed based on this model. An important problem
in controller design is the issue of robustness which relates to how the closed
loop system will perform when the physical system diers either in structure
or in parameters from the design model. For a system where the basic laws of
physics imply dissipative properties, it may make sense to dene the model so
that it possesses the same dissipative properties regardless of the numerical
2 1 Introduction

values of the physical parameters. Then if a controller is designed so that sta-

bility relies on the dissipative properties only, the closed-loop system will be
stable whatever the values of the physical parameters. Even a change of the
system order will be tolerated provided it does not destroy the dissipativity.
Parallel interconnections and feedback interconnections of dissipative sys-
tems inherit the dissipative properties of the connected subsystems, and this
simplies analysis by allowing for manipulation of block diagrams, and pro-
vides guidelines on how to design control systems. A further indication of
the usefulness of dissipativity theory is the fact that the PID controller is
a dissipative system, and a fundamental result that will be presented is the
fact that the stability of a dissipative system with a PID controller can be
established using dissipativity arguments. Note that such arguments rely on
the structural properties of the physical system, and are not sensitive to the
numerical values used in the design model. The technique of controller design
using dissipativity theory can therefore be seen as a powerful generalization
of PID controller design.
There is another aspect of dissipativity which is very useful in practical
applications. It turns out that dissipativity considerations are helpful as a
guide for the choice of a suitable variable for output feedback. This is helpful
for selecting where to place sensors for feedback control.
Throughout the book we will treat dissipativity for state space and input-
output models, but rst we will investigate simple examples which illustrate
some of the main ideas to be developed more deeply later.

1.1 Example 1: System with Mass Spring and Damper

Consider a one-dimensional simple mechanical system with a mass, a spring
and a damper. The equation of motion is

mx(t) + Dx(t) + Kx(t) = F (t), x(0) = x0 , x(0) = x0

where m is the mass, D is the damper constant, K is the spring stiness, x

is the position of the mass and F is the force acting on the mass. The energy
of the system is
1 1
V (x, x) = mx2 + Kx2
2 2
The time derivative of the energy when the system moves is
d
V (x(t), x(t)) = mx(t)x(t) + Kx(t)x(t)
dt
Inserting the equation of motion we get
d
V (x(t), x(t)) = F (t)x(t) Dx2 (t)
dt
 1.2 Example 2: RLC Circuit 3

Integration of this equation from t = 0 to t = T gives

T
T
V [x (T ) , x (T )] = V [x (0) , x (0)] + F (t) x (t) dt Dx2 (t) dt
0 0

This means that the energy at time t = T is the initial energy plus the energy
supplied to the system by the control force F minus the energy dissipated
by the damper. Note that if the input force F is zero, and if there is no
damping, then the energy V () of the system is constant. Here D 0 and
V [x (0) , x (0)] > 0, and it follows that the integral of the force F and the
velocity v = x satises

T
F (t) v (t) dt V [x (0) , x (0)] (1.1)
0

The physical interpretation of this inequality is seen from the equivalent in-
equality

T
F (t) v (t) dt V [x (0) , x (0)] (1.2)
0
T
which shows that the energy 0 F (t) x (t) dt that can be extracted from
the system is less than or equal to the initial energy stored in the system. We
will show later that (1.1) implies that the system with input F and output v
is passive. The Laplace transform of the equation of motion is
 2 
ms + Ds + K x(s) = F (s)

which leads to the transfer function

v s
(s) = 2
.
F ms + Ds + K
It is seen that the transfer function is stable, and that for s = j the phase
of the transfer function has absolute value less or equal to 90 , that is,
 v  v 
 
 (j) 90 Re (j) 0 (1.3)
F F
for all [, +]. We will see in the following that these properties of
the transfer function are consequences of the condition (1.1), and that they
are important in controller design.

1.2 Example 2: RLC Circuit

Consider a simple electrical system with a resistor R, inductance L and a
capacitor C with current i and voltage u. The dierential equation for the
circuit is
4 1 Introduction

di
L (t) + Ri(t) + Cx(t) = u(t)
dt
where
t
x(t) = i (t ) dt
0
The energy stored in the system is
1 2 1 2
V (x, i) = Li + Cx
2 2
The time derivative of the energy when the system evolves is
d di
V (x(t), i(t)) = L (t)i(t) + Cx(t)i(t)
dt dt
Inserting the dierential equation of the circuit we get
d
V (x(t), i(t)) = u(t)i(t) Ri2 (t)
dt
Integration of this equation from t = 0 to t = T gives

T
T
V [x (T ) , i (T )] = V [x (0) , i (0)] + u (t) i (t) dt Ri2 (t) dt
0 0

Similarly to the previous example, this means that the energy at time t = T
is the initial energy plus the energy supplied to the system by the voltage u
minus the energy dissipated by the resistor. Note that if the input voltage u
is zero, and if there is no resistance, then the energy V () of the system is
constant. Here R 0 and V [x (0) , x (0)] > 0, and it follows that the integral
of the voltage u and the current i satises

t
u (s) i (s) ds V [x (0) , i (0)] (1.4)
0

The physical interpretation of this inequality is seen from the equivalent in-
equality
t
u (s) i (s) ds V [x (0) , i (0)] (1.5)
0
t
which shows that the energy 0 u (s) i (s) ds that can be extracted from the
system is less than or equal to the initial energy stored in the system. We will
show later that (1.4) implies that the system with input u and output i is
passive. The Laplace transform of the dierential equation of the circuit is
 2 
Ls + Rs + C x(s) = u (s)

which leads to the transfer function

 1.3 Example 3: A Mass with a PD Controller 5

i s
(s) = .
u Ls2 + Rs + C
It is seen that the transfer function is stable, and that, for s = j, the phase
of the transfer function has absolute value less or equal to 90 , that is,
  
 i 
 (j) 90 Re i (j) 0 (1.6)
 u  u

for all [, +]. We see that in both examples we arrive at transfer

functions that are stable, and that have positive real parts on the j axis. This
motivates for further investigations on whether there is some fundamental
connection between conditions on the energy ow in equations associated
with the control equations (1.1) and (1.4) and the conditions on the transfer
functions (1.3) and (1.6). This will be made clear in chapter 2.

1.3 Example 3: A Mass with a PD Controller

Consider the mass m with the external control force u. The equation of motion
is
mx(t) = u(t)
Suppose that a PD controller

u = KP x KD x

is used. Then the closed loop dynamics is

mx(t) + KD x(t) + KP x(t) = 0

A purely mechanical system with the same dynamics as this system is called
a mechanical analog. The mechanical analog for this system is a mass m with
a spring with stiness KP and a damper with damping constant KD . We see
that the proportional action corresponds to the spring force, and that the
derivative action corresponds to the damper force. Similarly, as in Example
1, we can dene an energy function
1 1
V (x, x) = mx2 + KP x2
2 2
which is the total energy of the mechanical analog. In the same way as in
Example 1, the derivative action will dissipate the virtual energy that is ini-
tially stored in the system, and intuitively, we may accept that the system
will converge to the equilibrium x = 0, x = 0. This can also be seen from the
Laplace transform  2 
ms + KD s + KP x(s) = 0
6 1 Introduction

which implies that the poles of the system have negative real parts. The point
we are trying to make is that, for this system, the stability of the closed
loop system with a PD controller can be established using energy arguments.
Moreover, it is seen that stability is ensured for any positive gains KP and
KD independently of the physical parameter m. There are many important
results derived from energy considerations in connection with PID control,
and this will be investigated in Chapter 2.

1.4 Example 4: Adaptive Control

We consider a simple rst order system given by

x(t) = a
x(t) + u(t)

where the parameter a
is unknown. An adaptive tracking controller can be

designed using the control law

u = Ke ax + xd , e = x xd

where xd is the desired trajectory to be tracked, a is the estimate of the

parameter a
, and K is the feedback gain. The dierential equation for the
tracking error e is
de
dt (t) = a
x(t) + u(t) xd (t)
= a
x(t) Ke(t) a(t)x(t) + xd (t) xd (t)
= Ke(t) a(t)x(t)
where a = a a
is the estimation error. We now dene

(t) = a(t)x(t)

which leads to the following description of the tracking error dynamics

de
(t) + Ke(t) = (t)
dt
We dene a function Ve which plays the role of an abstract energy function
related to the tracking error e:
1 2
Ve (e) = e
2
The time derivative of Ve along the solutions of the system is

Ve (e(t)) = e(t)(t) Ke2 (t)

Note that this time derivative has a similar structure to that seen in Examples
1 and 2. In particular, the Ke2 term is a dissipation term, and if we think of
 1.4 Example 4: Adaptive Control 7

as the input and e as the output, then the e term is the rate of (abstract)
energy supplied from the input. We note that this implies that the following
inequality holds for the dynamics of the tracking error:

T
e(t)(t)dt Ve [e (0)]
0

To proceed, we dene one more energy-like function. Suppose that we are

able to select an adaptation law so that there exists an energy-like function
Va (a) 0 with a time derivative

Va (a(t)) = e(t)(t) (1.7)

We note that this implies that the following inequality holds for the adaptation
law:
T
[(t)] e(t)dt Va [a(0)]
0
Then the sum of the energy functions

V (e, a) = Ve (e) + Va (a)

has a time derivative along the solutions of the system given by

V (e(t), a(t)) = Ke2 (t)

This means that the energy function V (e, a) is decreasing as long as e() is
nonzero, and by invoking additional arguments from Barbalats Lemma (see
Chapter A), we can show that this implies that e(t) tends to zero as t +.
The required adaptation law for (1.7) to hold can be selected as the simple
gradient update
da
(t) = x(t)e(t)
dt
and the associated energy-like function is
1 2
Va (a) = a
2
Note that the convergence of the adaptive tracking controller was estab-
lished using energy-like arguments, and that other adaptation laws can be
used as long as they satisfy the energy-related requirement (1.7).
2
Positive Real Systems

The notion of Positive Real system may be seen as a generalization of the

positive deniteness of a matrix to the case of a dynamical system with in-
puts and outputs. When the input-output relation (or mapping, or operator)
is a constant matrix, testing its positive deniteness can be done by sim-
ply calculating the eigenvalues and checking that they are positive. When
the input-output operator is more complex, testing positive realness becomes
much more involved. This is the object of this chapter which is mainly devoted
to positive real linear time-invariant systems. They are known as PR transfer
functions.
The denition of Positive Real (PR) systems has been motivated by the
study of linear electric circuits composed of resistors, inductors and capaci-
tors. The driving point impedance from any point to any other point of such
electric circuits is always PR. The result holds also in the sense that any PR
transfer function can be realized with an electric circuit using only resistors,
inductors and capacitors. The same result holds for any analogous mechanical
or hydraulic systems. This idea can be extended to study analogous electric
circuits with nonlinear passive components and even magnetic couplings as
done by Arimoto [24] to study dissipative nonlinear systems. This leads us
to the second interpretation of PR systems: they are systems which dissipate
energy. As we shall see later in the book, the notion of dissipative systems,
which applies to nonlinear systems, is closely linked to PR transfer functions.
This chapter reviews the main results available for PR linear systems. It
starts with a short introduction to so-called passive systems. It happens that
there has been a proliferation of notions and denitions of various kinds of
PR or dissipative systems, since the early studies in the 1960s (to name a few:
ISP, OSP, VSP, PR, SPR, WSPR, SSPR, MSPR, ESPR; see the index for
the meaning of these acronyms). The study of their relationships (are they
equivalent, which ones imply which other one?) is not so easy and we bring
some elements of answers in this chapter and the next ones. This is why we
introduce rst in this chapter some basic denitions (passive systems, positive
real systems, bounded real transfer functions), their relationships, and then
10 2 Positive Real Systems

we introduce other rened notions of PR systems. The reason why passive

systems are briey introduced before bounded real and positive real transfer
functions, is that this allows one to make the link between an energy-related
notion and the frequency domain notions, in a progressive way. This, however,
is at the price of postponing a more rigorous and general exposition of passive
systems until later in the book.

2.1 Dynamical System State-space Representation

In this book various kinds of evolution, or dynamical systems will be analyzed:
linear, time invariant, nonlinear, nite-dimensional, innite-dimensional, dis-
crete time, non-smooth, standard dierential inclusions, unbounded or
maximal monotone dierential inclusions etc. Whatever the system we shall
be dealing with, it is of utmost importance to clearly dene some basic ingre-
dients:
A state vector x() and a state space X
A set of admissible inputs U
A set of outputs Y
An input/output mappping (or operator) H : u
y
A state space representation which relates the derivative of x() to x()
and u()
An output function which relates the output y() to the state x() and the
input u()
Such tools (or some of them) are necessary to write down the model, or
system, that is under examination. When one works with pure input/output
models, one doesnt need to dene a state space X; however U and Y are
crucial. In this book we will essentially deal with systems for which a state
space representation has been dened. Then the notion of a (state) solution is
central. Given some state space model under the form of an evolution problem
(a dierential equation or something looking like this), the rst step is to
provide informations on such solutions: the nature of the solutions (as time-
functions, for instance), their uniqueness, their continuity with respect to the
initial data and parameters, etc. This in turn is related to the set of admissible
inputs U. For instance, if the model takes the form of an ordinary dierential
equation (ODE) x(t) = f (x(t), u(t)), the usual Caratheodory conditions will
be in force to dene U as a set of measurable functions, and x() will usually
be an absolutely continuous function of time. In certain cases, one may want
to extend U to measures, or even distributions. Then x may also be a measure
or a distribution. Since it is dicult (actually impossible) to provide a general
well-posedness result for all the systems that will be dealt with in the rest
of the book, we will recall the well-posedness conditions progressively as new
models are introduced. This will be the case especially for some classes of
 2.2 Denitions 11

nonsmooth systems, where solutions may be absolutely continuous, or of local

bounded variation.
From a more abstract point of view, one may dene a general state-space
deterministic model as follows [364, 510, 512]:

There exists a metric space X (the state space), a transition map :

IR IR X U X, and a readout map r : X IRm IRp , such that:
(i) The limit x(t) = limt0 (t0 , t, 0, u) is in X for all t IR and all
u U (then x(t) is the state at time t)
(ii) (Causality) (t0 , t1 , x, u1 ) = (t0 , t1 , x, u2 ) for all t1 t0 , all x X,
and all u1 , u2 U such that u1 (t) = u2 (t) in the interval t0 t t1
(iii) (Initial state consistency) (t0 , t0 , x0 , u) = x0 for all t0 IR, u U,
and all x0 X
(iv) (Semigroup property) (t1 , t2 , (t0 , t1 , x0 , u), u) = (t0 , t2 , x0 , u) for
all x0 X, u U, whenever t0 t1 t2
(v) (Consistency with input-output relation) The input-output pairs (u, y)
are precisely those described via y(t) = r (limt0 (t0 , t, 0, u), u(t))
(vi) (Unbiasedness) (t0 , t, 0, 0) = 0 whenever t t0 and r(0, 0) = 0
(vii) (Time-invariance) (t1 + T, t2 + T, x0 , u1 ) = (t1 , t2 , x0 , u2 ) for all
T IR, all t2 t1 , and all u1 , u2 U such that u2 (t) = u1 (t + T )
Clearly item (vii) will not apply to some classes of time-varying systems,
and an extension is needed [512, 6]. There may be some items which do not
apply well to dierential inclusions where the solution may be replaced by a
solution set (for instance the semigroup property may fail). The basic fact that
X is a metric space will also require much care when dealing with some classes
of systems whose state spaces are not spaces of functions (like descriptor vari-
able systems that involve Schwarz distributions). In the innite-dimensional
case X may be a Hilbert space (i.e. a space of functions) and one may need
other denitions, see e.g. [39, 507]. An additional item in the above list could
be the continuity of the transition map () with respect to the initial data
x0 . Some nonsmooth systems do not possess such a property, which may be
quite useful in some stability results. A general exposition of the notion of a
system can be found in [467, Chapter 2]. We now stop our investigations of
what a system is since, as we said above, we shall give well-posedness results
each time they are needed all through the book.

2.2 Denitions
In this section and the next one, we introduce input-output properties of a
system, or operator H : u
H(u) = y. The system is assumed to be well-
posed as an input-ouput system, i.e. we may assume that H : L2,e L2,e
1
.
1
More details on Lp spaces can be found in Chapter 4.
12 2 Positive Real Systems

Denition 2.1. A system with input u() and output y() where u(t), y(t)
IRm is passive if there is a constant such that

t
y T ( )u( )d (2.1)
0

for all functions u() and all t 0. If, in addition, there are constants 0
and 0 such that

t
t
t
y T ( )u( )d + uT ( )u( )d + y T ( )y( )d (2.2)
0 0 0

for all functions u(), and all t 0, then the system is input strictly passive
(ISP) if > 0, output strictly passive (OSP) if > 0, and very strictly passive
(VSP) if > 0 and > 0.

Obviously 0 as the inequality (2.1) is to be valid for all functions

u() and in particular the control u(t) = 0 for all t 0, which gives 0 =
t T
0
y (s)u(s)ds . Thus the denition could equivalently be stated with
0. The importance of the form of in (2.1) will be illustrated in Examples 4.59
t t
and 4.60; see also Section 4.4.2. Notice that 0 y T (s)u(s)ds 12 0 [y T (s)y(s)+
uT (s)u(s)]ds is well dened since both u() and y() are in L2,e by assumption.

Theorem 2.2. Assume that there is a continuous function V () 0 such that

t
V (t) V (0) y(s)T u(s)ds (2.3)
0

for all functions u(), for all t 0 and all V (0). Then the system with input
u() and output y() is passive. Assume, in addition, that there are constants
0 and 0 such that

t
t
t
V (t) V (0) y T (s)u(s)ds uT (s)u(s)ds y T (s)y(s)ds (2.4)
0 0 0

for all functions u(), for all t 0 and all V (0). Then the system is input
strictly passive if there is a > 0, it is output strictly passive if there is an
> 0, and very strictly passive if there is a > 0 and an > 0 such that the
inequality holds.

Proof: It follows from the assumption V (t) 0 that

t
y T (s)u(s)ds V (0)
0

for all functions u() and all s 0, so that (2.1) is satised with := V (0)
0. Input strict passivity, output strict passivity and very strict passivity are
shown in the same way.
 2.2 Denitions 13

This indicates that the constant is related to the initial conditions of the
system; see also Example 4.59 for more informations on the role played by .
It is also worth looking at Corollary 3.3 to get more informations on the real
nature of the function V (): V () will usually be a function of the state of the
system. The reader may have guessed such a fact by looking at the examples
of Chapter 1.

Corollary 2.3. Assume that there exists a continuously dierentiable func-

t
tion V () 0 and a measurable function d() such that 0 d(s)ds 0 for all
t 0. Then
1. If
V (t) y T (t)u(t) d(t) (2.5)
for all t 0 and all functions u(), the system is passive.
2. If there exists a > 0 such that

V (t) y T (t)u(t) uT (t)u(t) d(t) (2.6)

for all t 0 and all functions u(), the system is input strictly passive
(ISP).
3. If there exists a > 0 such that

V (t) y T (t)u(t) y T (t)y(t) d(t) (2.7)

for all t 0 and all functions u(), the system is output strictly passive
(OSP).
4. If there exists a > 0 and a > 0 such that

V (t) y T (t)u(t) uT (t)u(t) y T (t)y(t) d(t) (2.8)

for all t 0 and all functions u(), the system is very strictly passive
(VSP).

t
If V () is the total energy of the system, then u, y = 0 y T (s)u(s)ds can be
seen as the power supplied to the system from the control, while d(t) can be
seen
 t as the power dissipated by the system. This means that the condition
0
d(s)ds 0 for all t 0 means that the system is dissipating energy. The
term w(u, y) = uT y is called the supply rate of the system.

Remark 2.4. All these notions will be examined in much more detail in Chap-
ter 4; see especially Section 4.5.2. Actually the notion of passivity (or dissi-
pativity) has been introduced in various ways in the literature. It is some-
times introduced as a pure input/output property of an operator (i.e. the
constant in (2.1) is not related to the state of the system) [125, 499, 500],
and serves as a tool to prove some bounded input/bounded output stabil-
ity results. Willems has, on the contrary, introduced dissipativity as a notion
14 2 Positive Real Systems

which involves the state space representation of a system, through so-called

storage functions [510, 511]. We will come back to this subject in Chapter 4.
Hill and Moylan started from an intermediate denition, where the constant
is assumed to depend on some initial state x0 [206209]. Then, under some
controllability assumptions, the link with Willems denition is made. In this
chapter and the next one, we will essentially concentrate on linear time in-
variant dissipative systems, whose transfer functions are named positive real
(PR). This is a very important side of passivity theory in Systems and Control
theory.

2.3 Interconnections of Passive Systems

A useful result for passive systems is that parallel and feedback interconnec-
tions of passive systems are passive, and that certain strict passivity properties
are inherited.

u1 - y1 u - hu 1 - y1 y-
h1 h1
6
u ?
h y-
6
u2 - y2 y2  u2
h2 h2

Fig. 2.1. Parallel and feedback interconnections.

To explore this we consider two passive systems with scalar inputs and out-
puts. Similar results are found for multivariable systems. System 1 has input
u1 and output y1 , and system 2 has input u2 and output y2 . We make the
following assumptions:
1. There are continuous dierentiable functions V1 (t) 0 and V2 (t) 0.
t t
2. There are functions d1 () and d2 () such that 0 d1 (s)ds 0 and 0 d2 (s)ds
0 for all t 0.
3. There are constants 1 0, 2 0, 1 0 and 2 0 such that

V1 (t) = y1 (t)u1 (t) 1 u21 (t) 1 y12 d1 (t) (2.9)

V2 (t) = y2 (t)u2 (t) 2 u22 (t) 2 y22 d2 (t) (2.10)

Assumption 3 implies that both systems are passive, and that system i is
strictly passive in some sense if any of the constants i or i are greater than
zero. For the parallel interconnection we have u1 = u2 = u, y = y1 + y2 , and
 2.4 Linear Systems 15

yu = (y1 + y2 )u = y1 u + y2 u = y1 u1 + y2 u2 (2.11)
By adding (2.9) (2.10) and (2.11), there exists a V () = V1 () + V2 () 0 and
t
a dp = d1 + d2 + 1 y12 + 2 y22 such that 0 dp (t )dt 0 for all t 0, and

V (t) = y(t)u(t) u2 (t) dp (t) (2.12)

where = 1 + 2 0. This means that the parallel interconnection system

having input u and output y is passive and strictly passive if 1 > 0 or 2 > 0.
For the feedback interconnection we have y1 = u2 = y, u1 = u y2 , and

yu = y1 (u1 + y2 ) = y1 u1 + y1 y2 = y1 u1 + u2 y2 (2.13)

Again by adding (2.9) (2.10) and (2.11) we nd that there is a V () = V1 () +

t
V2 () 0 and a df b = d1 + d2 + 1 u21 such that 0 df b (s)ds 0 for all t 0
and
V (t) = y(t)u(t) y 2 (t) df b (t) (2.14)
where = 1 + 2 + 2 . This means that the feedback interconnection is
passive, and in addition output strictly passive if 1 > 0, 2 > 0, or 2 > 0. By
induction it can be shown that any combination of passive systems in parallel
or feedback interconnection is passive.

2.4 Linear Systems

Let us now deal with linear invariant systems, whose input-output relation-
ships takes the form of a rational transfer function H(s) (also denoted as h(s)),
s C, and y(s) = H(s)u(s) where u(s) and y(s) are the Laplace transforms
of the time-functions u() and y(). Parsevals Theorem is very useful in the
study of passive linear systems, as shown next. It is now recalled for the sake
of completeness.

Theorem 2.5 (Parsevals Theorem). Provided that the integrals exist, the
following relation holds:



1
x(t)y (t)dt = x(j)y (j)d (2.15)
2

where y denotes the complex conjugate of y and x(j) is the Fourier trans-
form of x(t), where x(t) is a complex function of t, Lebesgue integrable.

Proof: The result is established as follows: the Fourier transform of the time
function x(t) is

x(j) = x(t)ejt dt (2.16)

16 2 Positive Real Systems

while the inverse Fourier transform is

1
x(t) = x(j)ejt d (2.17)
2

Insertion of (2.17) in (2.15) gives



1
x(t)y (t)dt = x(j)e d y (t)dt
jt
(2.18)
2

By changing the order of integration this becomes



1 jt
x(t)y (t)dt = x(j) y (t)e dt d (2.19)
2

Here


jt
y (t)e jt
dt = y(t)e dt = y (j) (2.20)

and the result follows.

u(s) y(s)
- h(s) -

Fig. 2.2. Linear time-invariant system

We will now present important properties of a linear time-invariant passive

system, which link the input-output passivity property to frequency-domain
conditions, using Parsevals Theorem. These notions will be generalized later
in the book, both in the case of LTI and nonlinear systems. Their usefulness
will be illustrated through examples of stabilization.

Theorem 2.6. Given a linear time-invariant linear system with rational

transfer function h(s), i.e.

y(s) = h(s)u(s) (2.21)

Assume that all the poles of h(s) have real parts less than zero. Then the
following assertions hold:
1. The system is passive Re[h(j)] 0 for all [, +].
2. The system is input strictly passive (ISP) There exists a > 0 such
that Re[h(j)] > 0 for all [, +].
3. The system is output strictly passive (OSP) There exists an > 0 such
that
 2.4 Linear Systems 17

Re[h(j)] |h(j)|2

 
1 2
 1 2
Re[h(j)] 2 + (Im[h(j)])2 2

Remark 2.7. A crucial assumption in Theorem 2.6 is that all the poles have
negative real parts. This assures that in Parsevals Theorem as stated in The-
orem 2.5, the integrals exist.
Proof: The proof is based on the use of Parsevals Theorem. In this Theorem
the time integration is over t [0, ). In the denition of passivity there is
an integration over t [0, T ]. To be able to use Parsevals Theorem in this
proof we introduce the truncated function

u( ) when t
ut ( ) = (2.22)
0 when > t

which is equal to u( ) for all less than or equal to t, and zero for all greater
than t. The Fourier transform of uT (t), which is denoted uT (j), will be used
in Parsevals Theorem. Without loss of generality we will assume that y(t)
and u(t) are equal to zero for all t 0. Then according to Parsevals Theorem

t


1
y( )u( )d = y( )ut ( )d = y(j)ut (j)d (2.23)
0 2

Insertion of y(j) = h(j)uT (j) gives

t

1
y( )u( )d = h(j)uT (j)ut (j)d, (2.24)
0 2

where

h(j)ut (j)ut (j) = {Re[h(j)] + jIm[h(j)]}|ut (j)|2 (2.25)

The left hand side of (2.24) is real, and it follows that the imaginary part on
the right hand side is zero. This implies that

t

1
u( )y( )d = Re[h(j)]|ut (j)|2 d (2.26)
0 2

First, assume that Re[h(j)] 0 for all . Then

t

t

u( )y( )d |ut (j)|2 d = u2 ( )d (2.27)
0 2 0

The equality is implied by Parsevals Theorem. It follows that the system

is passive, and in addition input strictly passive if > 0.
18 2 Positive Real Systems

Then, assume that the system is passive. Thus there exists a 0 so that

t
t

2
y(s)u(s)dsz u (s)ds = |ut (j)|2 d (2.28)
0 0 2
for all u(), where the initial conditions have been selected so that = 0. Here
= 0 for a passive system, while > 0 for a strictly passive system. Then



1 2
Re[h(j)]|uT (j)| d |uT (j)|2 d (2.29)
2 2
and


1
(Re[h(j)] )|uT (j)|2 d 0 (2.30)
2

If there exists a 0 so that Re[h(j0 )] < , then inequality will not hold
for all u because the integral on the left hand side can be made arbitrarily
small if the control signal is selected to be u(t) = U cos 0 t. The results 1 and
2 follow.
To show result 3 we rst assume that the system is output strictly passive,
that is, there is an > 0 such that

t
t

2
y(s)u(s)ds y (s)ds = |h(j)|2 |ut (j)|2 d. (2.31)
0 0 2

This gives the inequality (see (2.26))

2
Re[h(j)] |h(j)| (2.32)
which is equivalent to
 
2 2
(Re[h(j)]) + (Im[h(j)]) Re[h(j)] 0 (2.33)

and the second inequality follows by straighforward algebra. The converse

result is shown similarly as the result for input strict passivity.
Note that according to the theorem a passive system will have a transfer
function which satises

| h(j)| 90 for all [, +] (2.34)

In a Nyquist diagram the theorem states that h(j) is in the closed half
plane Re [s] 0 for passive systems, h(j) is in Re [s] > 0 for input
strictly passive systems, and for output strictly passive systems h(j) is inside
the circle with center in s = 1/ (2 ) and radius 1/ (2 ). This is a circle that
crosses the real axis in s = 0 and s = 1/ .
 2.4 Linear Systems 19

Remark 2.8. A transfer function h(s) is rational if it is the fraction of two

polynomials in the complex variable s, that is if it can be written in the form
Q(s)
h(s) = (2.35)
R(s)

where Q(s) and R(s) are polynomials in s. An example of a transfer function

that is not rational is h(s) = tanh s which appears in connection with systems
described by partial dierential equations.

Example 2.9. Note the dierence between the condition Re[h(j)] > 0 and
the condition for input strict passivity in that there exists a > 0 so that
Re[h(j0 )] > 0 for all . An example of this is
1
h1 (s) = (2.36)
1 + Ts
We nd that Re[h1 (j)] > 0 for all because
1 1 T
h1 (j) = = j (2.37)
1 + jT 1 + (T )2 1 + (T )2

However there is no > 0 that ensures Re[h(j0 )] > 0 for all

[, +]. This is seen from the fact that for any > 0 we have

1 1 1
Re[h1 (j)] = 2
< for all > (2.38)
1 + (T ) T
This implies that h1 (s) is not input strictly passive. We note that for this
system
2 1
|h1 (j)| = = Re[h1 (j)] (2.39)
1 + (T )2
which means that the system is output strictly passive with = 1.

Example 2.10. Consider a system with the transfer function

s+c
h2 (s) = (2.40)
(s + a)(s + b)
where a, b and c are positive constants. We nd that

j+c
h2 (j) = (j+a)(j+b)

(c+j)(aj)(bj)
= (a2 + 2 )(b2 + 2 )

abc+ 2 (a+bc)+j[(abacbc) 3 ]
= (a2 + 2 )(b2 + 2 ) .
From the above it is clear that
20 2 Positive Real Systems

1. If c a + b, then Re[h2 (j)] > 0 for all IR. As Re[h2 (j)] 0 when
, the system is not input strictly passive.
2. 
If c > a + b, then h2 (s) is not passive because Re[h2 (j)] < 0 for >
abc/(c a b).
Example 2.11. The systems with transfer functions

h3 (s) = 1 + T s (2.41)
1 + T1 s
h4 (s) = , T1 < T2 (2.42)
1 + T2 s
are input strictly passive because

Re[h3 (j)] = 1 (2.43)

and
1 + 2 T1 T2 T1
Re[h4 (j)] = ,1 (2.44)
1 + (T2 )2 T2
2
Moreover |h4 (j)| 1, so that
T1 T1 2
Re[h4 (j)] |h4 (j)| (2.45)
T2 T2
which shows that the system is output strictly passive with = T1 /T2 . The
reader may verify from a direct calculation of |h4 (j)|2 and some algebra that
2
it is possible to have Re[h4 (j)] |h4 (j)| , that is, = 1. This agrees with
the Nyquist plot of h4 (j).

Example 2.12. A dynamic system describing an electrical one-port with resis-

tors, inductors and capacitors is passive if the voltage over the port is input
and the current into the port is output, or vice versa. In Figure 2.3 dierent
passive one-ports are shown. We consider the voltage over the port to be the
input and the current into the port as the output. The resulting transfer func-
tions are admittances, which are the inverses of the impedances. Circuit 1 is
a capacitor, circuit 2 is a resistor in parallel with a capacitor, circuit 3 is a
resistor in series with a inductor and a capacitor, while circuit 4 is a resistor
in series with a parallel connection of an inductor, a capacitor and a resistor.
The transfer functions are
h1 (s) = Cs (2.46)
1
h2 (s) = (1 + RCs) (2.47)
R
Cs
h3 (s) = (2.48)
1 + RCs + LCs2
1 1+ R L
s + LCs2
h4 (s) = (2.49)
R1 1 + ( RL + R
1
L
)s + LCs2
 2.4 Linear Systems 21

1) h -i
+
u C
h

2) h -i
+
u R C
h

R L C
3) h -i 


+
u
h

R1
4) h -i
+ 


u L
 C R

h

Fig. 2.3. Passive electrical one-ports

Systems 1, 2, 3 and 4 are all passive as the poles have real parts that are
strictly less than zero, and in addition Re[hi (j)] 0 for all [, +]
and i {1, 2, 3, 4} (the fact that all the poles are in Re[s] < 0 is important; see
Theorem 2.14). It follows that the transfer functions have phases that satisfy
| hi (j)| 90 . In addition system 2 is input strictly passive as Re[h2 (j)] =
1/R > 0 for all . For system 4 we nd that
2
2 2 2
1 (1 LC) + R1 (R1 +R)
L
1
Re[h4 (j)] = (2.50)
R1 (1 2 LC)2 + 2 L2 2 R1 + R
(R +R)1

which means that system 4 is input strictly passive.

So far we have only considered systems where the transfer functions h(s)
have poles with negative real parts. There are however passive systems that
22 2 Positive Real Systems

have transfer functions with poles on the imaginary axis. This is demonstrated
in the following example:
Example 2.13. Consider the system y(t) = u(t) which is represented in transfer
function description by y(s) = h(s)u(s) where h(s) = 1s . This means that the
transfer function has a pole at the origin, which is on the imaginary axis. For
this system Re[h(j)] = 0 for all . However, we cannot establish passivity
using Theorem 2.6 as this theorem only applies to systems where all the poles
have negative real parts. Instead, consider

t
t
y(s)u(s)ds = y(s)y(s)ds (2.51)
0 0
A change of variables y(t)dt = dy gives

t
y(t)
1 1
y(t )u(t )dt = y(t )dy = [y(t)2 y(0)2 ] y(0)2 (2.52)
0 y(0) 2 2
and passivity is shown with = 12 y(0)2 .
It turns out to be relatively involved to nd necessary and sucient con-
ditions on h (j) for the system to be passive when we allow for poles on
the imaginary axis. The conditions are relatively simple and are given in the
following Theorem.
Theorem 2.14. Consider a linear time-invariant system with a rational
transfer function h(s). The system is passive if and only if
1. h(s) has no poles in Re [s] > 0.
2. Re[h(j)] 0 for all [, +] such that j is not a pole of h(s).
3. If j0 is a pole of h(s), then it is a simple pole, and the residual in s =
j0 is real and greater than zero, that is, Ress=j0 h(s) = limsj 0 (s
j0 )h(j) > 0.

The above result is established in Section 2.12. Contrary to Theorem 2.6,

poles on the imaginary axis are considered.
Corollary 2.15. If a system with transfer function h(s) is passive, then h(s)
has no poles in Re [s] > 0.
Proposition 2.16. Consider a rational transfer function
(s + z1 )(s + z2 ) . . .
h(s) = (2.53)
s(s + p1 )(s + p2 ) . . .
where Re[pi ] > 0 and Re[zi ] > 0 which means that h(s) has one pole at
the origin and the remaining poles in Re [s] < 0, while all the zeros are in
Re [s] < 0. Then the system with transfer function h (s) is passive if and only
if Re[h(j)] 0 for all [, +].
 2.4 Linear Systems 23

Proof: The residual of the pole on the imaginary axis is

z1 z2 . . .
Ress=0 h(s) = (2.54)
p1 p2 . . .
Here the constants zi and pi are either real and positive, or they appear in
complex conjugated pairs where the products zi zi = |zi |2 and pi pi = |pi |2 are
real and positive. It is seen that the residual at the imaginary axis is real and
positive. As h(s) has no poles in Re [s] > 0 by assumption, it follows that the
system is passive if and only if Re[h(j)] 0 for all [, +].
Example 2.17. Consider two systems with transfer functions
s2 + a2
h1 (s) = , a = 0, 0 = 0 (2.55)
s(s2 + 02 )
s
h2 (s) = 2 , 0 = 0 (2.56)
s + 02
where all the poles are on the imaginary axis. Thus condition 1 in Theo-
rem 2.14 is satised. Moreover,
a2 2
h1 (j) = j (2.57)
(02 2 )

h2 (j) = j 2 (2.58)
0 2
so that condition 2 also holds in view of Re[h1 (j)] = Re[h2 (j)] = 0 for all
so that j is not a pole in h (s) . We now calculate the residual, and nd
that
a2
Ress=0 h1 (s) = 2 (2.59)
0
02 a2
Ress=j 0 h1 (s) = (2.60)
202
1
Ress=j 0 h2 (s) = (2.61)
2
We see that, according to Theorem 2.14, the system with transfer function
h2 (s) is passive, while h1 (s) is passive whenever a < 0 .
Example 2.18. Consider a system with transfer function
1
h(s) = (2.62)
s
The transfer function has no poles in Re [s] > 0, and Re[h(j)] 0 for all
= 0. However, Ress=0 h(s) = 1, and Theorem 2.14 shows that the system
is not passive This result agrees with the observation

t
y(t)
1
y(s)u(s)ds = y(s)dy = [y(0)2 y(t)2 ] (2.63)
0 y(0) 2
where the right hand side has no lower bound as y(t) can be arbitrarily large.
24 2 Positive Real Systems

2.5 Passivity of the PID Controllers

Proposition 2.19. Assume that 0 Td < Ti and 0 1. Then the PID
controller
1 + Ti s 1 + Td s
hr (s) = Kp (2.64)
Ti s 1 + Td s
is passive.

This follows from Proposition 2.16.

Proposition 2.20. Consider a PID controller with transfer function

1 + Ti s 1 + Td s
hr (s) = Kp (2.65)
1 + Ti s 1 + Td s
where 0 Td < Ti , 1 < and 0 < 1. Then the controller is passive
K
and, in addition, the transfer function gain has an upper bound |hr (j)| p
and the real part of the transfer function is bounded away from zero according
to Re [hr (j)] Kp for all .

It follows from Bode diagram techniques that

1 Kp
|hr (j)| Kp 1 = (2.66)

The result on the Re [hr (j)] can be established using Nyquist diagram, or
by direct calculation of Re [hr (j)].

2.6 Stability of a Passive Feedback Interconnection

y0 (s)
- h - - y(s)-
h1 (s) h2 (s)

6 
h0 (s)  

Fig. 2.4. Interconnection of a passive system h1 (s) and a strictly passive system
h2 (s)

Consider a feedback loop with loop transfer function h0 (s) = h1 (s)h2 (s)
as shown in Figure 2.4. If h1 is passive and h2 is strictly passive, then the
phases of the transfer functions satisfy

| h1 (j)| 90 and | h2 (j)| < 90 (2.67)

It follows that the phase of the loop transfer function h0 (s) is bounded by
 2.7 Mechanical Analogs for PD Controllers 25

| h0 (j)| < 180 (2.68)

As h1 and h2 are passive, it is clear that h0 (s) has no poles in Re [s] >
0. Then according to standard Bode-Nyquist stability theory the system is
asymptotically stable and BIBO stable 2 . The same result is obtained if instead
h1 is strictly passive and h2 is passive.
We note that, in view of Proposition 2.20, a PID controller with limited
integral action is strictly stable. This implies that
A passive linear system with a PID controller with limited integral action
is BIBO stable.
For an important class of systems passivity or strict passivity is a structural
property which is not dependent on the numerical values of the parameters
of the system. Then passivity considerations may be used to establish sta-
bility even if there are large uncertainties or large variations in the system
parameters. This is often referred to as robust stability. When it comes to
performance it is possible to use any linear design technique to obtain high
performance for the nominal parameters of the system. The resulting system
will have high performance under nominal conditions, and in addition robust
stability under large parameter variations.

2.7 Mechanical Analogs for PD Controllers

In this section we will study how PD controllers for position control can be
represented by mechanical analogs when the input to the system is force and
the output is position. Note that when force is input and position is output,
then the physical system is not passive. We have a passive physical system if
the force is the input and the velocity is the output, and then a PD controller
from position corresponds to PI controller from velocity. For this reason we
might have referred to the controllers in this section as PI controllers for
velocity control.
We consider a mass m with position x() and velocity v() = x(). The
dynamics is given by mx(t) = u(t) where the force u is the input. The desired
position is xd (), while the desired velocity is vd () = xd (). A PD controller
u = Kp (1 + Td s) [xd (s) x(s)] is used. The control law can be written as

u(t) = Kp (xd (t) x(t)) + D(vd (t) v(t)) (2.69)

where D = Kp Td . The mechanical analog appears from the observation that

this control force is the force that results if the mass m with position x is
connected to the position xd with a parallel interconnection of a spring with
stiness Kp and a damper with coecient D as shown in Figure 2.5.

2
Bounded Input-Bounded Output.
26 2 Positive Real Systems

Kp
.. .. ..
............ ....... .........
.. .. ..
u- h h
- -
x D x0
Fig. 2.5. Mechanical analog of PD controller with feedback from position

If the desired velocity is not available, and the desired position is not
smooth a PD controller of the type

u(s) = Kp xd (s) Kp (1 + Td s)x(s), s C

can be used. Then the control law is

u(t) = Kp (xd (t) x(t)) Dv(t) (2.70)

This is the force that results if the mass m is connected to the position
xd with a spring of stiness Kp and a damper with coecient D as shown in
Figure 2.6.

Kp
.. . .
..................................
. . ..
u- h .. h
......
...................
...........
.. ... ..
...........
.....
- D -
x x0
Fig. 2.6. Mechanical analog of a PD controller without desired velocity input

If the velocity is not measured the following PD controller can be used

1 + Td s
u(s) = Kp [xd (s) x(s)] (2.71)
1 + Td s
where 0 1 is the lter parameter. We will now demonstrate that this
transfer function appears by connecting the mass m with position x to a
spring with stiness K1 in series with a parallel interconnection of a spring
with stiness K and a damper with coecient D as shown in Figure 2.7.
To nd the expression for K1 and K we let x1 be the position of the
connection point between the spring K1 and the parallel interconnection. Then
the force is u = K1 (x1 x), which implies that x1 (s) = x(s) + u(s)/K1 . As
 2.8 Multivariable Linear Systems 27

K
.. . .
.................................
K1 . . ..
u- h .. .. ..
............. ....... ......... h
. .. ..

D
- - -
x x1 x0
Fig. 2.7. Mechanical analog of a PD controller without velocity measurement

there is no mass in the point x1 there must be a force of equal magnitude in

the opposite direction from the parallel interconnection, so that

u(s) = K[xd (s) x1 (s)] + D[vd (s) v1 (s)] = (K + Ds)[xd (s) x1 (s)] (2.72)

Insertion of x1 (s) gives

1
u(s) = (K + Ds)[xd (s) x(s) u(s)] (2.73)
K1
We solve for u(s) and the result is

+K+Ds [xd (s) x(s)]

u(s) = K1 K1K+Ds

1+ K
D
s
= K1 K
K1 +K 1+ K K+K KD [xd (s)
s
x(s)]
1

We see that this is a PD controller without velocity measurement where

Kp = KK11+K
K



D
Td = K





= K1K+K [0, 1)

2.8 Multivariable Linear Systems

Theorem 2.21. Consider a linear time-invariant system

y(s) = H(s)u(s) (2.74)

with a rational transfer function matrix H(s) Cmm , input u(t) IRm and
input y(t) IRm . Assume that all the poles of H(s) are in Re [s] < 0. Then,
1. The system is passive min [H(j)+H (j)] 0 for all [, +].
28 2 Positive Real Systems

2. The system is input strictly passive There is a > 0 so that min [H(j)
+H (j)] > 0 for all [, +].

Remark 2.22. Similarly to Theorem 2.6, a crucial assumption in Theorem 2.21

is that the poles have negative real parts, i.e. there is no pole on the imaginary
axis.
Proof: Let A Cmm be some Hermitian matrix with eigenvalues i (A).
Let x Cm be an arbitrary vector with complex entries. It is well-known
from linear algebra that x Ax is real, and that x Ax min (A)|x|2 . From
Parsevals Theorem we have
 T m 
0
y (s)ut (s)ds = i=1 0 yi (s)(ui )t (s)ds
m 1

= i=1 2 yi (j)(ui )t (j)d

1

= 2
y (j)ut (j)d
where we recall that ut () is a truncated function and that s in the integrand is
a dumb integration variable (not to be confused with the Laplace transform!).
This leads to
t T  T
0 y (s)u(s)ds = 0 y (s)ut (s)ds

1

= 2 y (j)ut (j)d

1

= 4
[uT (j)y(j) + y (j)ut (j)]d

1

= 4 ut (j)[H(j) + H (j)]ut (j)d
Because H(j) + H (j) is Hermitian we nd that

t

1
y T ( )u( )d min [H(j) + H (j)]|ut (j)|2 d (2.75)
0 4

The result can be established along the lines of Theorem 2.6.

2.9 The Scattering Formulation

By a change of variables an alternative description can be established where
passivity corresponds to small gain. We will introduce this idea with an ex-
ample from linear circuit theory. Consider a linear time-invariant system de-
scribing an electrical one-port with voltage e, current i and impedance z (s)
so that
 2.9 The Scattering Formulation 29

e(s) = z(s)i(s) (2.76)

Dene the wave variables

a = e + z0 i and b = e z0 i (2.77)

where z0 is a positive constant. The Laplace transform is

a(s) = [z(s) + z0 ]i(s)

b(s) = [z(s) z0 ]i(s)
Combining the two equations we get

b(s) = g(s)a(s) (2.78)

where
z(s)
z(s) z0 1
g(s) = = z0 z(s) (2.79)
z0 + z(s) 1 + z0
is the scattering function of the system. The terms wave variable and scatter-
ing function originate from the description of transmission lines where a can
be seen as the incident wave and b can be seen as the reected wave.
If the electrical circuit has only passive elements, that is, if the circuit is an
interconnection of resistors, capacitors and inductors, the passivity inequality
satises
t
e( )i( )d 0 (2.80)
0
where it is assumed that the initial energy stored in the circuit is zero. We
note that
a2 b2 = (e + z0 i)2 (e z0 i)2 = 4z0 ei (2.81)
which implies

t
t
t
b2 ( )d = a2 ( )d 4z0 e( )i( )d (2.82)
0 0 0
From this it is seen that passivity of the system with input i and output e
corresponds to small gain for the system with input a and output b in the
sense that

t
t
2
b ( )d a2 ( )d (2.83)
0 0
This small gain condition can be interpreted loosely in the sense that the
energy content b2 of the reected wave is smaller than the energy a2 of the
incident wave. For the general linear time-invariant system

y(s) = h(s)u(s) (2.84)

introduce the wave variables

30 2 Positive Real Systems

a = y + u and b = y u (2.85)
where, as above, a is the incident wave and b is the reected wave. As for
electrical circuits it will usually be necessary to include a constant z0 so that
a = y + z0 u b = y z0 u so that the physical units agree. We tacitly suppose
that this is done by letting z0 = 1 with the appropriate physical unit. The
scattering function is dened by

b yu h(s) 1
g(s) = (s) = (s) = (2.86)
a y+u 1 + h(s)
Theorem 2.23. Consider a system with rational transfer function h(s) with
no poles in Re[s] 0, and scattering function g(s) given by (2.86). Then
1. The system is passive if and only if |g(j)| 1 for all [, +].
2. The system is input strictly passive, and there is a so that |h (j)|
for all [, +] if and only if there is a  (0, 1) so that |g(j)|2
1 .
Proof: Consider the following computation
|h(j)1|2
|g(j)|2 = |h(j)+1|2

|h(j)|2 2Re[h(j)]+1
= |h(j)|2 +2Re[h(j)]+1
(2.87)

4Re[h(j)]
= 1 |h(j)+1|2

It is seen that |g(j)| 1 if and only if Re[h(j)] 0. Result 1 then

follows as the necessary and sucient condition for the system to be passive
is that Re[h(j)] 0 for all [, +]. Concerning the second result,
we show the if part. Assume that there is a so that Re[h(j)] > 0
and a so that |h (j)| for all [, +]. Then
4
|g(j)|2 1 2 (2.88)
( + 1)
 
4
and the result follows with 0 <  < min 1, (+1)2 . Next assume that
g(j)|2 1  for all . Then
 
4Re [h(j)]  |h(j)|2 + 2Re[h(j)] + 1 (2.89)

and strict passivity follows from


Re [h(j)] >0 (2.90)
4 2 
Finite gain of h (j) follows from
 2.10 Impedance Matching 31

 |h(j)|2 (4 2  ) Re[h(j)] +  0 (2.91)

which in view of the general result |h(j)| > Re[h(j)] gives the inequality
(4 2  )
|h(j)|2 |h(j)| + 1 0 (2.92)

This implies that
(4 2  )
|h (j)| (2.93)


We shall come back on the relationships between passivity and bounded

realness in the framework of dissipative systems and H theory; see Sec-
tion 5.9. A comment on the input-output change in (2.85): the association
of the new system with transfer function g(s) merely corresponds to writing
t
down uy = 14 (a + b)(a b) = 14 (a2 b2 ). Thus if 0 u(s)y(s)ds 0 one gets
t 2 t 2
0
a (s)ds 0 b (s)ds: the L2 -norm of the new output b(t) is bounded by
the L2 -norm of the new intput a(t).

2.10 Impedance Matching

In this section we will briey review the concept of impedance matching.
Again an electrical one-port is studied. The one-port has a voltage source e,
serial impedance z0, output voltage v and current i. The circuit is coupled to
the load which is a passive one-port with driving point impedance zL (s) as
shown in Figure 2.8.

z0 (s) - i
h
+ +



e u zL (s)

Fig. 2.8. Impedance matching

The following problem will be addressed: suppose z0 (s) is given and that
e (t) = E sin e t. Select zL (s) so that the power dissipated in zL is maximized.
The current is given by
e(s)
i(s) = (2.94)
z0 (s) + zL (s)
32 2 Positive Real Systems

while the voltage over zL is

u(s) = zL (s)i(s) (2.95)

The power dissipated in zL is therefore

P (e ) = 12 Re[uL (je )i (j e )]

= 12 Re[zL (j e )]i(j e )i (j e )

= 1 Re[zL (j e )] 2
2 [z0 (j e )+zL (j e )] [z0 (j e )+zL (j e )] E

where () denotes the complex conjugate. Denote

z0 (j e ) = 0 + j0 or zL (j e ) = L + jL (2.96)

This gives
1 L E 2
P = (2.97)
2 (0 + L )2 + (0 + L )2
We see that if L = 0, then P = 0, whereas for nonzero L then |L | ,
gives P 0. A maximum for P would be expected somewhere between these
extremes. Dierentiation with respect to L gives

P E2 2L(0 + L )
= (2.98)
L 2 [(0 + L )2 + (0 + L )2 ]2
which implies that the maximum of P appears for L = 0 . Dierentiation
with respect to L with L = 0 gives

P E2 20 L 2
= (2.99)
a 2 [(0 + L )2 + (0 + L )2 ]2

and it is seen that the maximum is found for L = 0 . This means that the
maximum power dissipation in zL is achieved with

zL (je ) = z0 (je ) (2.100)

This particular selection of zL (j e ) is called impedance matching. If the

voltage source e(t) is not simply a sinusoid but a signal with a arbitrary spec-
trum, then it is not possible to nd a passive impedance zL (s) which satises
the impedance matching condition or a general series impedance z0 (j). This
is because the two impedances are required to have the same absolute values,
while the phase have opposite signs. This cannot be achieved for one particular
zL (s) for all frequencies.
However, if z0 (j) = z0 is a real constant, then impedance matching at
all frequencies is achieved with zL = z0 . We now assume that z0 is a real
constant, and dene the wave variables to be
 2.11 Feedback Loop 33

a = u + z0 i or b = u z0 i (2.101)

Then it follows that

a=e (2.102)
for the system in Figure 2.8. A physical interpretation of the incident wave a is
as follows: let u be the input voltage to the one-port and let i be the current
into the port. Consider the extended one-port where a serial impedance z0
is connected in to the one-port as shown in Figure 2.9. Then a is the input
voltage of the extended one-port.

z0 (s) - i
h h
+ +

a u zL (s)

h h

Fig. 2.9. Extended one-port with a serial impedance z0

z0 (s) - i z0 (s)
h h h
+ + +
a u b z
h h h

Fig. 2.10. Physical interpretation of the reected wave b where z = zL (s) z0 (s)

The physical interpretation of the reected wave b is shown in gure 2.10.

We clearly see that if zL = z0 , then

u = z0 i b=0 (2.103)

This shows that if impedance matching is used with z0 being constant, then
the scattering function is
b(s)
g(s) = =0 (2.104)
a(s)
34 2 Positive Real Systems

uf
y0 e ut ? u y
- j - hr (s) - j - h(s) -
6

h0 (s)  

Fig. 2.11. Feedback interconnection of two passive systems

2.11 Feedback Loop

A feedback interconnection of two passive linear time-invariant systems is
shown in Figure 2.11 where signals are given by

y(s) = h(s)u(s), ut (s) = hr (s)e(s) (2.105)

u(t) = uf (t) + ut (t), e(t) = y0 (t) y(t) (2.106)

We can think of h(s) as describing the plant to be controlled, and hr (s)
as describing the feedback controller. Here ut is the feedback control and
uf is the feedforward control. We assume that the plant h(s) and that the
feedback controller hr (s) are strictly passive with nite gain. Then, as shown
in Section 2.6 we have  |h0 (j)| < 180 where h0 (s) := h(s)hr (s) is the loop
transfer function, and the system is BIBO stable.
A change of variables in now introduced to bring the system into a scat-
tering formulation. The new variables are

a = y + u and b = y u

for the plant and

ar = ut + e and br = ut e
for the feedback controller. In addition input variables

a0 = y0 + uf and b0 = y0 uf

are dened. We nd that

ar = u t + y 0 y = u u f + y 0 y = b 0 b (2.107)

and
b r = u t y 0 + y = u u f y 0 + y = a a0 (2.108)
The associated scattering functions are

h(s) 1 hr (s) 1
g(s) = and gr (s) =
1 + h(s) 1 + hr (s)
 2.11 Feedback Loop 35

Now, h(s) and hr (s) are passive by assumption, and as a consequence they
cannot have poles in Re [s] > 0. Then it follows that g(s) and gr (s) cannot
have poles in Re [s] > 0 because 1 + h(s) is the characteristic equations
for h(s) with a unity negative feedback, which obviously is a stable system.
Similar arguments apply for 1 + h(s). The system can then be represented as
in Figure 2.12 where

b(s) = g(s)a(s), br (s) = gr (s)ar (s) (2.109)

a(t) = br (t) + a0 (t), ar (t) = b0 (t) b(t) (2.110)

a0 = y0 + uf
b0 = y0 uf ar br ? a b
- j - gr (s) - j - g(s) -
6

g0 (s)  

Fig. 2.12. Equivalent system

In the passivity setting, stability was ensured when two passive systems were
interconnected in a feedback structure because the loop transfer function
h0 (j) had a phase limitation so that  h0 (j) > 180 . We would now
like to check if there is an interpretation for the scattering formulation that is
equally simple. This indeed turns out to be the case. We introduce the loop
transfer function

g0 (s) = g(s)gr (s) (2.111)
of the scattering formulation. The function g0 (s) cannot have poles in Re [s] >
0 as g(s) and gr (s) have no poles in Re [s] > 0 by assumption. Then we have
from Theorem 2.23:
1. |g(j)| 1 for all [, +] because h(s) is passive.
2. |gr (j)| < 1 for all [, +] because hr (s) is strictly passive with
nite gain.
As a consequence of this,

|g0 (j)| < 1 (2.112)

for all [, +], and according to the Nyquist stability criterion the
system is BIBO stable.
36 2 Positive Real Systems

2.12 Bounded Real and Positive Real Transfer Functions

Bounded real and positive real are two important properties of transfer func-
tions related to passive systems that are linear and time-invariant. We will in
this section show that a linear time-invariant system is passive if and only if
the transfer function of the system is positive real. To do this we rst show
that a linear time-invariant system is passive if and only if the scattering
function, which is the transfer function of the wave variables, is bounded real.
Then we show that the scattering function is bounded real if and only if the
transfer function of the system is positive real. We will also discuss dierent
aspects of these results for rational and irrational transfer functions.
We consider a linear time-invariant system y(s) = h(s)u(s) with input u

and output y. The incident wave is denoted a = y + u, and the reected wave

is denoted b = y u. The scattering function g(s) is given by

h(s) 1
g(s) = (2.113)
1 + h(s)

and satises b(s) = g(s)a(s). We note that

1 2
u(t)y(t) = [a (t) b2 (t)] (2.114)
4
For linear time-invariant systems the properties of the system do not depend
on the initial conditions, as opposed to nonlinear systems. We therefore as-
sume for simplicity that the initial conditions are selected so that the energy
function V (t) is zero for initial time, that is V (0) = 0. The passivity inequality
is then

t

1 t 2
0 V (t) = u(s)y(s)ds = [a (s) b2 (s)]ds (2.115)
0 4 0
The properties bounded real and positive real will be dened for functions
that are analytic in the open right half plane Re[s] > 0. We recall that
a function f (s) is analytic in a domain only if it is dened and innitely
dierentiable for all points in the domain. A point where f (s) ceases to be
analytic is called a singular point, and we say that f (s) has a singularity at
this point. If f (s) is rational, then f (s) has a nite number of singularities, and
the singularities are called poles. The poles are the roots of the denominator
polynomial R(s) if f (s) = Q(s)/R(s), and a pole is said to be simple pole if
it is not a multiple root in R(s).
 2.12 Bounded Real and Positive Real Transfer Functions 37

Denition 2.24. A function g(s) is said to be bounded real if

1. g(s) is analytic in Re[s] > 0.

2. g(s) is real for real and positive s.
3. |g(s)| 1 for all Re[s] > 0.

This denition extends to matrix functions G(s) as follows:

Denition 2.25. A transfer matrix G(s) Cmm is bounded real if all

elements of G(s) are analytic for Re[s] 0 and the H norm satises
||G(s)|| 1 where we recall that ||G(s)|| = supIR max (G(j)). Equiva-
lently the second condition can be replaced by: Im GT (j)G(j) 0 for all
IR. Strict Bounded Realness holds when the 0 inequalities are replaced
by > 0 inequalities.

Theorem 2.26. Consider a linear time-invariant system described by y(s) =

h(s)u(s), and the associated scattering function a = y + u, b = y u and
b(s) = g(s)a(s) where
h(s) 1
g(s) = (2.116)
1 + h(s)
which satises b(s) = g(s)a(s) a = y + u and b = y u. Then the system
y(s) = h(s)u(s) is passive if and only if g(s) is bounded real.

Proof: Assume that y(s) = h(s)u(s) is passive. Then (2.115) implies that

t
t
a2 ( )d b2 ( )d (2.117)
0 0

for all t 0. It follows that g(s) cannot have any singularities in Re[s] > 0 as
this would result in exponential growth in b(t) for any small input a(t). Thus,
g(s) must satisfy condition 1 in the denition of bounded real.
Let 0 be an arbitrary real and positive constant, and let a(t) = e0 t 1(t)
where 1(t) is the unit step function. Then the Laplace transform of a(t) is
1 g(s)
a(s) = s 0
, while b(s) = s 0
. Suppose that the system is not initially
excited so that the inverse Laplace transform for rational g(s) gives

n  
g(s) si t g(s)
b(t) = Ress=si e + Ress=0 e0 t
i=1
s 0 s 0

g(s)
where si are the poles of g(s) that satisfy Re [si ] < 0, and Ress=0 s 0
=
g(0 ). When t the term including e will dominate the terms including
0 t
38 2 Positive Real Systems

esi t , and b(t) will tend to g(0 )e0 t . The same limit for b(t) will also be found
for irrational g(s). As a(t) is real, it follows that g(0 ) is real, and it follows
that g(s) must satisfy condition 2 in the denition of bounded realness.
Let s0 = 0 + j0 be an arbitrary point in Re[s] > 0, and let the input
be a(t) = Re[es0 t 1(t)]. Then b(t) Re[g(s0 )es0 t ] as t and the power
1 2
P (t) := [a (t) b2 (t)] (2.118)
4
will tend to
1 20 t
P (t) = [e cos2 0 t |g(s0 )|2 e20 t cos2 (0 t + )]
4
where = arg[g(s0 )]. This can be rewritten using cos2 = 12 (1 + cos 2), and
the result is
8P (t) = (1 + cos 20 t)e20 t |g(s0 )|2 [1 + cos(20 t + 2)]e20 t
 
= [1 |g(s0 )|2 ]e20 t + Re[ 1 g(s0 )2 e2s0 t ]

In this expression s0 and 0 are constants, and we can integrate P (t) to get
the energy function V (T ):
t
V (t) = P (s)ds

1 1 1
= 160 [1 |g(s0 )|2 ]e20 t + 16 Re{ s0 [1 g(s0 )2 ]e2s0 t }
First it is assumed that 0 = 0. Then Re{ s10 [1 g(s0 )2 ]e2s0 t } will be a
sinusoidal function which becomes zero for certain values of t. For such values
of t the condition V (t) 0 implies that
1
[1 |g(s0 )|2 ]e20 t 0
160
which implies that
1 |g(s0 )|2 0
Next it is assumed that 0 = 0 such that s0 = 0 is real. Then g(s0 ) will
be real, and the two terms in V (t) become equal. This gives
1
0 V (t) = [1 g 2 (s0 )]e20 t
80
and with this it is established that for all s0 in Re[s] > 0 we have

1 |g(s0 )|2 0 |g(s0 )| 1

To show the converse we assume that g(s) is bounded real and consider
 2.12 Bounded Real and Positive Real Transfer Functions 39

g(j) = lim g( + j) (2.119)

0
>0

Because g(s) is bounded and analytic for all Re [s] > 0 it follows that this
limit exists for all , and moreover

|g(j)| 1

Then it follows from Parsevals Theorem that with aT being the truncated
version of a we have
1
 2 
0 8 t
|a (j)| 1 |g(j)|2 d

1
t
= 4 0
[a2 (s) b2 (s)]ds
t
= 0
u(s)y(s)ds
which shows that the system must be passive.

Im

Re

Fig. 2.13. Contour in the right half plane.

Dene the contour C which encloses the right half plane as shown in Figure
2.13. The maximum modulus theorem is as follows. Let f (s) be a function
that is analytic inside the contour C. Let M be the upper bound on |f (s)|
on C. Then |f (s)| M inside the contour, and equality is achieved at some
point inside C if and only if f (s) is a constant. This means that if g(s) is
bounded real, and |g(s)| = 1 for some point in Re[s] > 0, then |g(s)| achieves
its maximum inside the contour C, and it follows that g(s) is a constant in
Re[s] 0. Because g(s) is real for real s > 0, this means that g(s) = 1 for all
40 2 Positive Real Systems

s in Re[s] 0. In view of this [1 g(s)]1 has singularities in Re[s] > 0 if

and only if g(s) = 1 for all s in Re[s] 0.
If g(s) is assumed to be a rational function the maximum modulus theorem
can be used to reformulate the condition on |g (s)| to be a condition on |g (j)|.
The reason for this is that a rational transfer function satisfying |g(j)| 1
for all will also satisfy

lim |g(j)| = lim |g(s)| (2.120)

|s|

Therefore, for a suciently large contour C we have that |g(j)| 1 implies

|g(s)| 1 for all Re[s] > 0 whenever g(s) is rational. This leads to the
following result:

Theorem 2.27. A rational function g(s) is bounded real if and only if

1. g(s) has no poles in Re[s] 0.
2. |g(j)| 1 for all [, +].

Let us now state a new denition.

Denition 2.28. A transfer function h(s) is said to be positive real

(PR) if
1. h(s) is analytic in Re[s] > 0
2. h(s) is real for positive real s
3. Re[h(s)] 0 for all Re[s] > 0

The last condition above is illustrated in Figure 2.14 where the Nyquist
plot of a PR transfer function H(s) is shown. The notion of positive realness
extends to multivariable systems:

Denition 2.29. The transfer matrix H(s) Cmm is positive real if:
H(s) has no pole in Re[s] > 0
H(s) is real for all positive real s
H(s) + H
(s) 0 for all Re[s] > 0

An interesting characterization of multivariable PR transfer functions is

as follows:

Theorem 2.30. Let the transfer matrix H(s) = C(sIn A)1 + D Cmm ,
where the matrices A, B, C, and D are real, and every eigenvalue of A has
a negative real part. Then H(s) is positive real if and only if y
[H
(j) +
H(j)]y = y
(j)y 0 for all IR and all y Cm .
 2.12 Bounded Real and Positive Real Transfer Functions 41

Im[H(j)]

()

Re[H(j)]
0

Fig. 2.14. Positive real transfer function

This result was proved in [8, p.53]. The rational matrix (s) = c(sIn
A)1 B B T (sIn + AT )1 C T + D + DT is known as the Popov function of the
system. It is a rational spectral function, since it satises (s) = T (s). The
introduction of the spectral function (s) allows us to state a result on which
we shall come back in Section 3.3. Let : L2,e L2,e be a rational input-
output operator u()
y() = (u()). Assume that the kernel of has a
minimal realization (A, B, C, D). In other words, the operator is represented in
the Laplace transform space by a transfer matrix H(s) = C(sIn A)1 B + D,
where (A, B) is controllable and (A, C) is observable. The rational matrix (s)
is the spectral function associated to .

Proposition 2.31. The rational operator is non-negative, i.e.

t
u( )(u( ))d 0
0

for all u L2,e , if and only if its associated spectral function (s) is non-
negative.

Proof: We assume that u(t) = 0 for all t < 0 and that the system is causal.
Let the output y() be given as

t
y(t) = Du(t) + CeA(t Bu( )d (2.121)
0
42 2 Positive Real Systems

Let U (s) and Y (s) denote the Laplace transforms of u() and y(), re-
spectively. Let us assume that (s) has no pole on the imaginary axis. From
Parsevals Theorem one has

+
+
1
[y T (t)u(t) + uT (t)y(t)]dt = [Y
(j)U (j) + U
(j)Y (j)]d
2
  (2.122)
One also has Y (s) = D + C(sIn A)1 B U (s). Therefore

+
+
1
[y T (t)u(t) + uT (t)y(t)]dt = U
(j)(j)U (j)d. (2.123)
2

It follows that:
 +
(j) 0 for all IR implies that [y T (t)u(t) + uT (t)y(t)]dt 0
for all admissible u().
Reciprocally, given a couple (0 , U0 ) that satises U0T (j0 )U0 < 0, there
exists by continuity an interval 0 such that U0T (j)U0 < 0 for all
0 . Consequently the inverse Fourier transform v0 () of the function

U0 if 0
U (j) = (2.124)

0 if  0
1
 T
makes the quadratic form 2 0 U0 (j)U0 d < 0. Therefore positivity
of and of its spectral function are equivalent properties.
If (s) has poles on the imaginary axis, then Parsevals Theorem can be
used under the form

+
+
2at T T 1
e [y (t)u(t)+u (t)y(t)]dt = U
(a+j)S(a+j)U (a+j)d
2
(2.125)
which is satised for all real a, provided the line a + j does not contain any
pole of (s).

Remark 2.32. It is implicit in the proof of Proposition 2.31 that the initial
data on y() and u() and their derivatives, up to the required orders, are
zero. Consequently, the positivity of the operator (), when associated to a
state space representation (A, B, C, D), is characterized with the initial state
x(0) = 0. Later on in Chapter 4, we shall give a denition of dissipativity,
which generalizes that of positivity for a rational operator such as (), and
which precisely applies with x(0) = 0; see Denition 4.22.

It is sometimes taken as a denition that a spectral function (s) is non-

negative if there exists a PR function H(s) such that (s) = H(s) + H T (s)
 2.12 Bounded Real and Positive Real Transfer Functions 43

[145, Denition 6.2]. We shall make use of Proposition 2.31 in Section 5.10
on hyperstability. Notice that Proposition 2.31 does not imply the stability
of the above mentioned operator (provided one has associated a state space
realization to this operator). The stability is in fact obtained if one makes
further assumptions like the observability and controllability. We shall come
back on these points in the next chapters on dissipative systems and their
stability, via the Kalman-Yakubovich-Popov Lemma; see Remark 3.32.
The next theorem links bounded realness with positive realness.

Theorem 2.33. Consider the linear time-invariant system y(s) = h(s)u(s),

and the scattering formulation a = y + u, b = y u and b(s) = g(s)a(s) where

h(s) 1
g(s) = (2.126)
1 + h(s)

Assume that g(s) = 1 for all Re[s] > 0. Then h(s) is positive real if and only
if g(s) is bounded real.

Proof: Assume that g(s) is bounded real and that g(s) = 1 for all Re[s] > 0.
Then [1 g(s)]1 exists for all s in Re[s] > 0. From (2.126) we nd that

1 + g(s)
h(s) = (2.127)
1 g(s)

where h(s) is analytic in Re[s] > 0 as g(s) is analytic in Re[s] > 0, and [1
g(s)]1 is nonsingular by assumption in Re[s] > 0. To show that Re[h(s)] 0
for all Re[s] > 0 the following computation is used:

2Re[h(s)] = h (s) + h(s)

1+g (s) 1+g(s)

= 1g (s) + 1g(s) (2.128)

1g (s)g(s)
= 2 [1g (s)][1g(s)]

We see that Re[h(s)] 0 for all Re[s] > 0 whenever g(s) is bounded real.
Next assume that h(s) is positive real. Then h(s) is analytic in Re[s] > 0,
and [1 + h(s)] is nonsingular in Re[s] > 0 as Re[h(s)] 0 in Re[s] > 0.
It follows that g(s) is analytic in Re[s] > 0. From (2.128) it is seen that
|g(s)| 1 in Re[s] > 0; it follows that g(s) is bounded real.
It is noteworthy that Theorem 2.33 extends to multivariable systems:

Theorem 2.34. Let H(s) Cmm be a square transfer function, with

det(H(s) + H(s)) = 0 for Re[s] 0, and H(j) + H(T (j) 0. Then
the bounded realness of G(s) = (G(s) Im )(G(s) + Im )1 implies that H(s)
is positive real.

From Theorem 2.26 and Theorem 2.33 it follows that:

44 2 Positive Real Systems

Corollary 2.35. A system with transfer function h(s) is passive if

and only if the transfer function h(s) is positive real.

Example 2.36. A fundamental result in electrical circuit theory is that if the

transfer function h(s) is rational and positive real, then there exists an electri-
cal one-port built from resistors, capacitors and inductors so that h(s) is the
impedance of the one-port [126, p. 815]. If e is the voltage over the one-port
and i is the current entering the one-port, then e(s) = h(s)i(s). The system
with input i and output e must be passive because the total stored energy of
the circuit must satisfy
V (t) = e(t)i(t) g(t) (2.129)
where g(t) is the dissipated energy.

Example 2.37. The transfer function

1
h(s) = (2.130)
tanh s
is irrational, and positive realness of this transfer function cannot be es-
tablished from conditions on the frequency response h(j). We note that
tanh s = sinh s/ cosh s, where sinh s = 12 (es es ) and cosh s = 12 (es + es ).
First we investigate if h(s) is analytic in the right half plane. The singularities
are given by

sinh s = 0 es es = 0 es (1 e2s ) = 0

Here |es | 1 for Re[s] > 0, while

es (1 e2s ) = 0 e2s = 1

Therefore the singularities are found to be

sk = jk, k {0, 1, 2 . . .} (2.131)

which are on the imaginary axis. This means that h(s) is analytic in Re[s] > 0.
Obviously, h(s) is real for real s > 0. Finally we check if Re [h(s)] is positive
in Re[s] > 0. Let s = + j. Then

cosh s = 12 [e (cos + j sin ) + e (cos j sin )]

= cosh cos + j sinh sin

while
sinh s = sinh cos + j cosh sin (2.132)
This gives
 2.12 Bounded Real and Positive Real Transfer Functions 45

cosh sinh
Re[h(s)] = > 0, Re [s] > 0 (2.133)
| sinh s|2
where it is used that = Re [s], and the positive realness of h(s) has been
established.

Consider a linear system represented by a rational function H(s) of the

complex variable s = + j:
b m sm + + b 0
H(s) = (2.134)
sn + an1 sn1 + + a0
where ai , bi IR are the system parameters n is the order of the system and
r = n m is the relative degree. For rational transfer functions it is possible
to nd conditions on the frequency response h(j) for the transfer function
to be positive real. The result is presented in the following theorem:

Theorem 2.38. A rational function h(s) is positive real if and only if

1. h(s) has no poles in Re[s] > 0.
2. Re[h(j)] 0 for all [, +] such that j is not a pole in h(s).
3. If s = j0 is a pole in h(s), then it is a simple pole, and if 0 is nite,
then the residual

Ress=j 0 h(s) = lim (s j 0 )h(s)

sj 0

is real and positive. If 0 is innite, then the limit

h(j)
R := lim
j
is real and positive.

Proof: The proof can be established by showing that conditions 2 and 3 in

this Theorem are equivalent to the condition

Re[h(s)] 0 (2.135)

for all Re[s] > 0 for h(s) with no poles in Re[s] > 0.
First assume that conditions 2 and 3 hold. We use a contour C as shown
in Figure 2.15 which goes from j to j along the j axis with small
semicircular indentations into the right half plane around points j 0 that
are poles of h(s). The contour C is closed with a semicircle into the right
half plane. On the part of C that is on the imaginary axis Re[h(s)] 0 by
assumption. On the small indentations

Ress=j 0 h(s)
h(s) (2.136)
s j0
46 2 Positive Real Systems

Im

Re
r

Fig. 2.15. Contour C of h(s) in the right half plane.

As Re[s] 0 on the small semi-circles and Ress=j 0 h(s) is real and positive
according to condition 3, it follows that Re[h(s)] 0 on these semi-circles.
On the large semi-circle into the right half plane with radius we also have
Re[h(s)] 0 and the value is a constant equal to lim Re[h(j)], unless
h(s) has a pole at innity at the j axis, in which case h(s) sR on the
large semi-circle. Thus we may conclude that Re[h(s)] 0 on C. Dene the
function
f (s) = eRe[h(s)]
Then |f (s)| 1 on C, and in view of the maximum modulus theorem, |f (s)|
1 for all s Re[s] > 0. It follows that Re[h(s)] 0 in Re[s] > 0, and the
result is shown.
Next assume that Re[h(s)] 0 for all Re[s] > 0. Then condition 2 follows
because
h(j) = lim h( + j)
0
>0

exists for all such that j is not a pole in h(s). To show condition 3 we
assume that 0 is a pole of multiplicity m for h(s). On the small indentation
with radius r into the right half plane we have s j 0 = rej where /2
/2. Then

Ress=j 0 h(s) Ress=j 0 h(s) jm

h(s) = e (2.137)
rm ejm rm
Clearly, here it is necessary that m = 1 to achieve Re[h(s)] 0 because
the term ejm gives an angle from m/2 to m/2 in the complex plane.
Moreover, it is necessary that Ress=j 0 h(s) is positive and real because ejm
gives an angle from /2 to /2 when m = 1. The result follows.
 2.13 Examples 47

The foregoing theorem extends to multivariable systems:

Theorem 2.39. The rational function H(s) Cmm is positive real if and
only if:
H(s) has no poles in Re[s] > 0
H(j) + H
(j) 0 for all positive real such that j is not a pole of
H()
If i0 , nite or innite, is a pole of H(), it is a simple pole and the
corresponding residual K0 is a semi positive denite Hermitian matrix.

2.13 Examples
2.13.1 Mechanical Resonances

Motor and Load with Elastic Transmission

An interesting and important type of system is a motor that is connected to

a load with an elastic transmission. The motor has moment of inertia Jm , the
load has moment of inertia JL , while the transmission has spring constant K
and damper coecient D. The dynamics of the motor is given by

Jm m (t) = Tm (t) TL (t) (2.138)

where m () is the motor angle, Tm () is the motor torque, which is considered

to be the control variable, and TL () is the torque from the transmission. The
dynamics of the load is
JL L (t) = TL (t) (2.139)
The transmission torque is given by
 
TL = D L m K (L m ) (2.140)

The load dynamics can then be written in Laplace transform form as

 
JL s2 + Ds + K L (s) = (Ds + K) m (s) (2.141)

which gives
L 1 + 2Z s1
(s) = 2 (2.142)
m 1 + 2Z s1 + s 2
1

where
K
12 = (2.143)
JL
and
48 2 Positive Real Systems

2Z D
= (2.144)
1 K
By adding the dynamics of the motor and the load we get

Jm m (t) + JL L (t) = Tm (t) (2.145)

which leads to
1 + 2Z s1
Jm s2 m (s) + JL s2 s2
m (s) = Tm (s) (2.146)
1 + 2Z s1 + 12

and from this 2

m 1 + 2Z s1 + s 2
1
(s) = 2 (2.147)
Tm Js2 (1 + 2 s1 + s 2 )
1

where
J = Jm + JL (2.148)
is the total inertia of motor and load, and the resonant frequency 1 is given
by
1 J 2
12 = 12 = (2.149)
1 J JL Jm 1
while the relative damping is given by

J
= Z (2.150)
Jm
We note that the parameters 1 and depend on both motor and load pa-
rameters, while the parameters 1 and Z depend only on the load.
The main observation in this development is the fact that 1 < 1 . This
means that the transfer function m (s)/Tm (s) has a complex conjugated pair
of zeros with resonant frequency 1 , and a pair of poles at the somewhat
higher resonant frequency 1 . The frequency response is shown in Figure 2.16
when K = 20, Jm = 20, JL = 15 and D = 0.5. Note that the elasticity does
not give any negative phase contribution.
By multiplying the transfer functions L (s)/m (s) and m (s)/Tm (s) the
transfer function
L 1 + 2Z s1
(s) = 2 (2.151)
Tm Js2 (1 + 2 s1 + s 2 )
1

is found from the motor torque to the load angle.

The resulting frequency response is shown in Figure 2.17. In this case the
elasticity results in a negative phase contribution for frequencies above 1 .
Example 2.40. Typically the gear is selected so that Jm = JL . This gives
1
1 = 1 = 0.7071 (2.152)
2
 2.13 Examples 49

40
T m (j Z)
20
amplitude (dB)
Tm

-20

-40
10 -1 10 0 10 1
Z [rad/s]

0
Tm (j Z)
-50 Tm
fase

-100

-150

-200
10 -1 10 0 10 1
Z [rad/s]

Fig. 2.16. Frequency response of m (s)/Tm (s)

40
T L (j Z)
20
amplitude (dB)

Tm

-20

-40
10 -1 10 0 10 1
Z [rad/s]

-150

-200
T L (j Z)
fase

-250 Tm

-300

-350
10 -1 10 0 10 1
Z [rad/s]

Fig. 2.17. Frequency response of L (s)/m (s)

50 2 Positive Real Systems

Example 2.41. Let Z = 0.1 and Jm = JL . In this case

s
L 1 + 3.535 1
(s) = s2
(2.153)
Tm Js2 (1 + 2 s1 + 12
)

Passivity Inequality

The total energy of motor and load is given by

1 2 1 2 1
V (m , L , L , m ) = Jm m + JL L + K[L m ]2 (2.154)
2 2 2

where m (t) = m (t) and L (t) = L (t). The rate of change of the total energy
is equal to the power supplied from the control torque Tm (t) minus the power
dissipated in the system. This is written

V (t) = m (t)Tm (t) D[L (t) m (t)]2 (2.155)

We see that the power dissipated in the system is D[L (t) m (t)]2 which is
the power loss in the damper. Clearly the energy function V (t) 0 and the
power loss satises D[(t)]2 0,. It follows that

t
t
m (s)Tm (s)ds = V (t) V (0) + D[(s)]2 ds V (0) (2.156)
0 0

which implies that the system with input Tm () and output m () is passive.
It follows that

Re[hm (j)] 0 (2.157)

for all [, +]. From energy arguments we have been able to show
that
m
180  (j) 0 . (2.158)
Tm

2.13.2 Systems with Several Resonances

Passivity

Consider a motor driving n inertias in a serial connection with springs and

dampers. Denote the motor torque by Tm and the angular velocity of the
motor shaft by m . The energy in the system is
 2.13 Examples 51

V (m , m , Li ) = 12 Jm m
2
+ 12 K01 (m L1 )2

+ 12 JL1 L1
2
+ 12 K12 (L1 L2 )2 + . . .

+ 12 JL,n1 L,n1
2
+ 12 Kn1,n (L,n1 Ln )2

+ 12 JLn Ln
2

Clearly, V () 0. Here Jm is the motor inertia, Li is the velocity of inertia

JLi , while Ki1,i is the spring connecting inertia i 1 and i and Di1,i is
the coecient of the damper in parallel with Ki1,i . The index runs over
i = 1, 2, . . . , n. The system therefore satises the equation

V (t) = Tm (t)m (t) d(t) (2.159)

where

d(t) = D12 (L1 (t)L2 (t))2 +. . .+Dn1,n (L,n1 (t)Ln (t))2 0 (2.160)

represents the power that is dissipated in the dampers, and it follows that the
system with input Tm and output m is passive. If the system is linear, then
the passivity implies that the transfer function
m
hm (s) = (s) (2.161)
Tm
has the phase constraint
| hm (j)| 90 (2.162)
for all [, +]. It is quite interesting to note that the only information
that is used to nd this phase constraint on the transfer function is that the
system is linear, and that the load is made up from passive mechanical com-
ponents. It is not even necessary to know the order of the system dynamics,
as the result holds for an arbitrary n.

2.13.3 Two Motors Driving an Elastic Load

In this section we will see how passivity considerations can be used as a

guideline for how to control two motors that actuate on the same load through
elastic interconnections consisting of inertias, springs and dampers as shown
in Figure 2.18.
The motors have inertias Jmi , angle qmi and motor torque Tmi where
i {1, 2}. Motor 1 is connected to the inertia JL1 with a spring with stiness
K11 and a damper D11 . Motor 2 is connected to the inertia JL2 with a spring
with stiness K22 and a damper D22 . Inertia JLi has angle qLi . The two
inertias are connected with a spring with stiness K12 and a damper D12 .
The total energy of the system is
52 2 Positive Real Systems

Motor 1 Tm1 Tm2 Motor 2

K 11, D 11 K 12, D 12 K 22,D 22
Jm1 JL1 JL2 Jm2
q m1 q L1 q L2 q m2

Fig. 2.18. Two motors actuating on one load

V (qm1 , qm2 , qLi ) = 12 [Jm1 qm1

2 2
+ Jm2 qm2 2
+ JL1 qL1 2
+ JL2 qL2

+K11 (qm1 qL1 )2 + K22 (qm2 qL2 )2 + K12 (qL1 qL2 )2 ]

and the time derivative of the energy when the system evolves is

V (t) = Tm1 qm1 (t) + Tm2 qm2 (t) D11 (qm1 (t) qL1 (t))2

+D22 (qm2 (t) qL2 (t))2 +D12 (qL1 (t) qL2 (t))2

It is seen that the system is passive from (Tm1 , Tm2 )T to (qm1 , qm2 )T . The
system is multivariable, with controls Tm1 and Tm2 and outputs qm1 and qm2 .
A controller can be designed using multivariable control theory, and passivity
might be a useful tool in this connection. However, here we will close one
control loop at a time to demonstrate that independent control loops can be
constructed using passivity arguments. The desired outputs are assumed to
be qm1 = qm2 = 0. Consider the PD controller

Tm2 = Kp2 qm2 Kv2 qm2 (2.163)

for motor 2 which is passive from qm2 to Tm2 . The mechanical analog of this
controller is a spring with stiness Kp2 and a damper Kv2 which is connected
between the inertia Jm2 and a xed point. The total energy of the system
with this mechanical analog is
V (qm1 , qm2 , qL1 , qL2 ) = 12 [Jm1 qm1
2 2
+ Jm2 qm2 2
+ JL1 qL1 2
+ JL2 qL2

+K11 (qm1 qL1 )2 + K22 (qm2 qL2 )2

+K12 (qL1 qL2 )2 + Kp2 q22 ]

and the time derivative is

V (t) = Tm1 (t)qm1 (t) D11 (qm1 (t) qL1 (t))2 + D22 (qm2 (t) qL2 (t))2

+D12 (qL1 (t) qL2 (t))2 Kv2 q22 (t)

 2.14 Strictly Positive Real (SPR) Systems 53

It follows that the system with input Tm1 and output qm1 is passive when
the PD controller is used to generate the control Tm2 . The following controller
can then be used:
1 + Ti s 1
T1 (s) = Kv1 q1 (s) = Kv1 [1 + ( 1) ]sq1 (s) (2.164)
1 + Ti s 1 + Ti s
This is a PI controller with limited integral action if q1 is considered as the
output of the system. The resulting closed loop system will be BIBO stable
independently from system and controller parameters, although in practice,
unmodelled dynamics and motor torque saturation dictate some limitations
on the controller parameters. As the system is linear, stability is still ensured
even if the phase of the loop transfer function becomes less that 180 for
certain frequency ranges. Integral eect from the position can therefore be
included for one of the motors, say motor 1. The resulting controller is
1 + Ti s
T1 (s) = Kp1 q1 (s) + Kv1 sq1 (2.165)
Ti s
In this case the integral time constant Ti must be selected e.g. by Bode dia-
gram techniques so that stability is ensured.

2.14 Strictly Positive Real (SPR) Systems

Consider again the denition of Positive Real transfer function in Denition
2.28. The following is the standard denition of Strictly Positive Real (SPR)
transfer functions.

Denition 2.42 (Strictly Positive Real). A rational transfer func-

tion H(s) Cmm that is not identically zero for all s, is strictly
positive real (SPR) if H(s ) is PR for some > 0.

Let us now consider two simple examples:

Example 2.43. The transfer function of an asymptotically stable rst order
system is given by

1
H(s) = (2.166)
s+
where > 0. Replacing s by + j we get
1 + j
H(s) = = (2.167)
( + ) + j ( + )2 + 2

Note that Re[s] = > 0 we have Re[H(s)] 0. Therefore H(s) is PR.

Furthermore H(s ) for = 2 is also PR and thus H(s) is also SPR.
54 2 Positive Real Systems

Example 2.44. Consider now a simple integrator (i.e. take = 0 in the previ-
ous example)
1 1 j
H(s) = = = 2 . (2.168)
s + j + 2
1
It can be seen that H(s) = s is PR but not SPR.

In view of Theorem 2.6, one may wonder whether an SPR transfer function
is ISP, OSP. See Examples 4.62, 4.64, 4.65.

2.14.1 Frequency Domain Conditions for a Transfer Function to

be SPR

The denition of SPR transfer functions given above is in terms of conditions

in the s complex plane. Such conditions become relatively dicult to be ver-
ied as the order of the system increases. The following theorem establishes
conditions in the frequency domain for a transfer function to be SPR.

Theorem 2.45 (Strictly Positive Real). [226] A rational transfer function

h(s) is SPR if
1. h(s) is analytic in Re[s] 0, i.e. the system is asymptotically stable
2. Re[h(j)] > 0, for all (, ) and
3. a) 2lim 2 Re[h(j)] > 0 when r = 1,

h(j)
b) 2lim Re[h(j)] > 0, lim j > 0 when r = 1,
||

where r is the relative degree of the system.

Proof: Necessity: If h(s) is SPR, then from Denition 2.42, h(s ) is PR

for some > 0. Hence, there exists an > 0 such that for each [ 0, ),
h(s ) is analytic in Re[s] < 0. Therefore, there exists a real rational function
W (s) such that [8]

h(s ) + h(s + ) = W (s )W (s + ) (2.169)

where W (s) is analytic and nonzero for all s in Re [s] > . Let s = + j;
then from (2.169) we have

2 Re [h(j)] = |W (j)|2 > 0, (, ) (2.170)

Now h(s) can be expressed as

bm sm + bm1 sm1 + + b1 s + b0
h(s) = (2.171)
sn + an1 sn1 + + a1 s + a0
If m = n 1, i.e., r = 1, bn1 = 0, then from (2.171) it follows that bn1 > 0
and an1 bn1 bn2 bn1 > 0 for h(s ) to be PR, and
 2.14 Strictly Positive Real (SPR) Systems 55

lim 2 Re [h(j)] = an1 bn1 bn2 bn1 > 0 (2.172)

2

If m = n + 1, i.e., r = 1, bn+1 = 0, then

1  
Re [h(j )] = 2 (bn bn+1 an1 bn+1 ) 2n + (2.173)
|a(j )|

Since Re [h(j )] 0 (, ) and

h(j )
lim = bn1 0,
|| j
then bn+1 > 0, bn bn+1 an1 bn+1 > 0, and therefore 3.b) follows directly.

Suciency; Let (A, b, c, d, f ) be a minimal state representation of h(s),

i.e.,

h(s) = c(sI A)1 b + d + f s (2.174)

From (2.174) we can write

 
h(s ) = c(sIA)1 b+d+f s+ c(sI A I)1 (sI A)1 b f (2.175)

Hence,

Re [h(j )] = Re [h(j)] + Re [g(j )] (2.176)

where g(j ) = c(jIn A I)1 (jIn A)1 bf. There exists an > 0
such that for all [0, ) and (, ), (jIn A I)1 is analytic.
Therefore for each [0, ), |Re [g(j )]| < k1 < for all (, )
and some k1 > 0. If r = 0, then Re [h(j)] > k2 > 0 for all and some
k2 > 0. Therefore

Re [h(j )] = Re [h(j)] + Re [g(j )] > k2 k1 > 0 (2.177)

for all (, ) and 0 < < min { , k2 /k1 }. Hence, h(s ) is PR and
therefore h(s) is SPR.
If r = 1, then Re [h(j)] > k3 > 0 for all || < 0 and 2 Re [h(j)] >
k4 2> 0 for all ||  0 , where 0 , k3 , k4 are nite positive constants. Similarly,
 Re [g(j )] < k5 and |Re [g(j )]| < k6 for all (, ) and
some nite positive constants k5 , k6. Therefore, Re [h(j )] > k3 k6
for all || < 0 and 2 Re [h(j )] > k4 k5 for all || 0 . Then,
for 0 < < min {k3 /k6 , , k4 /k5 } and (, ), Re [h(j )] > 0.
Hence, h(s ) is PR and therefore h(s) is SPR.
If r = 1, then d > 0 and therefore
56 2 Positive Real Systems

Re [h(j )] > d k1 (2.178)

Hence, for each in the interval [0, min { , d/k1 } ), Re [h(j )] > 0 for all
(, ). Since

h(j)
lim =f >0
j
then

h(j )
lim =f >0
j
and therefore, all the conditions of Denition 2.28 and Theorem 2.38 are
satised by h(s ); hence h(s ) is PR, i.e., h(s) is SPR and the suciency
proof is complete.

Remark 2.46. It should be noted that when r = 0, conditions 1 and 2 of

the Theorem, or 1 and Re[h(j)] > > 0 for all [, +], are both
necessary and sucient for h(s) to be SPR.

Notice that H(s) in (2.166) satises condition 3.a), but H(s) in (2.168)
does not. Let us now give a multivariable version of Theorem 2.45, whose
proof is given in [256] and is based on [226, 508].

Theorem 2.47. Let H(s) Cmm be a proper rational transfer matrix, and
suppose that det(H(s) + H T (s)) is not identically zero. Then H(s) is SPR if
and only if

H(s) has all its poles with negative real parts

H(j) + H T (j) > 0 for all IR
and one of the following three conditions is satised:
H() + H T () > 0
H() + H T () = 0 and lim 2 [H(j) + H T (j)] > 0
H() + H T () 0 (but not zero nor nonsingular)and there exist
positive constants and such that

2 min [H(j) + H T (j)] , || (2.179)

2.14.2 Necessary Conditions for H(s) to be PR (SPR)

In general, before checking all the conditions for a specic transfer function
to be PR or SPR, it is useful to check rst that it satises a set of necessary
conditions. The following are necessary conditions for a system to be PR
(SPR)
 2.14 Strictly Positive Real (SPR) Systems 57

H(s) is (asymptotically) stable.

The Nyquist plot of H(j) lies entirely in the (closed) right half complex
plane.
The relative degree of H(s) is either r = 0 or r = 1.
H(s) is (strictly) minimum-phase, i.e. the zeros of H(s) lie in Re[s]
0 (Re[s] < 0).

Remark 2.48. In view of the above necessary conditions it is clear that unsta-
ble systems or nonminimum phase systems are not positive real. Furthermore
proper transfer functions can be PR only if their relative degree is 0 or 1.
This means for instance that a double integrator, i.e. H(s) = s12 is not PR.
This remark will turn out to be important when dealing with passivity of
nonlinear systems. In particular for a robot manipulator we will be able to
prove passivity from the torque control input to the velocity of the generalized
coordinates but not to the position of the generalized coordinates.

2.14.3 Tests for SPRness

Stating necessary and sucient conditions for a transfer function to be PR or

SPR is a rst fundamental step. A second step consists in usable criteria which
allow one to determine if a given rational function is SPR or not. Work in this
direction may be found in [31,132,146,177,205,341,396,455,504,528,536]. We
can for instance quote a result from [455].

Theorem 2.49. [455] Consider H(s) = C(sIn A)1 B C. H(s) is SPR if

and only if 1) CAB < 0, 2) CA1 B < 0, 3) A is stable, 4) A(In ABC
CAB )A has
no eigenvalue on the open negative real axis (, 0). Consider now H(s) =
C(sIn A)1 B + D C, D > 0. H(s) is SPR if and only if 1) A is stable,
2) the matrix (A BC
D )A has no eigenvalue on the closed negative real axis
(, +].

Stability means here that all the eigenvalues are in the open left-half of the
complex plane Re[s] < 0, and may be called strict stability. An interpretation
of SPRness is that (A, B, C, D) with D = 0 is SPR if and only if the matrix
pencil A1 + (A BCD ) is nonsingular for all > 0 [455].

2.14.4 Interconnection of Positive Real Systems

One of the important properties of positive real systems is that the inverse
of a PR system is also PR. In addition the interconnection of PR systems
in parallel or in negative feedback (see Figure 2.19) inherit the PR property.
More specically we have the following properties (see [226]):
1
H(s) is PR (SPR) H(s) is PR (SPR).
If H1 (s) and H2 (s) are SPR so is H(s) = 1 H1 (s) + 2 H2 (s) for 1
0, 2 0, 1 + 2 > 0.
58 2 Positive Real Systems
H1 (s)
If H1 (s) and H2 (s) are SPR, so is H(s) = 1+H1 (s)H2 (s) .

Remark 2.50. Note that a transfer function H(s) need not be proper to be
PR or SPR. For instance, the non-proper transfer function s is PR.

Remark 2.51. Let us recall that if (A, B, C, D) is a realization of the transfer

function H(s) C, i.e. C(sIn A)1 B + D = H(s), and if D = 0, then
C 1 1
(A BCD , D , D , D ) is a realization of a system with transfer function H(s)
B

(see for instance [246, p.76]).

u1 y1
H1
-

y2 u2
H2
Fig. 2.19. Negative feedback interconnection of H1 and H2

2.14.5 Special Cases of Positive Real Systems

We will now introduce two additional denitions of classes of systems. Both

of them are PR systems but one of them is weaker than SPR systems and
the other is stronger. Weak SPR (WSPR) are important because they allow
the extension of the KYP Lemma presented in Chapter 3 for systems other
than PR. They are also important because they allow to relax the conditions
for stability of the negative feedback interconnection of a PR system and an
SPR system. We will actually show that the negative feedback interconnection
between a PR system and a WSPR produces an asymptotically stable system.
Both properties will be seen later.

Remark 2.52. Consider again an electric circuit composed of an inductor in

parallel with a capacitor. Such a circuit will exhibit sustained oscillatory be-
havior. If we have instead a lossy capacitor in parallel with a lossy inductor,
 2.14 Strictly Positive Real (SPR) Systems 59

it is clear that the energy stored in the system will be dissipated. However,
it is sucient that at least one of the two is a lossy element (either a lossy
capacitor or a lossy inductor) to guarantee that the oscillatory behavior will
asymptotically converge to zero. This example motivates the notion of weakly
SPR transfer function.

Denition 2.53. (Weakly SPR) A rational function H(s) C is weakly

SPR (WSPR) if
1. H(s) is analytic in Re[s] 0.
2. Re[H(j)] > 0, for all (, ).

In the multivariable case one replaces the second condition by H(j) +

H T (j) > 0 for all IR. It is noteworthy that a transfer function may be
WSPR but not be SPR; see an example below. WSPRness may be seen as an
intermediate notion between PR and SPR. See Section 5.3 for more analysis
on WSPR systems, which shows in particular and in view of Examples 4.62
and 4.64 that WSPR is not SPR.

Denition 2.54. (Strong SPR) A rational function H(s) C is strongly

SPR (SSPR) if
1. H(s) is analytic in Re[s] 0.
2. Re[H(j)] > 0, for all [, ] and some IR.

In the multivariable case the second condition for SSPRness becomes H(j)+
H T (j) > 0 for all IR and H() + H T () > 0, or as H(j) +
H T (j) > Im for all [, ]. From Theorem 2.6, it can be seen
that a SSPR transfer function is ISP. If the system has a minimal state space
realization (A, B, C, D) then H(s) + H T (s) = C(sIn A)1 B B T (sIn +
AT )1 C T + D + DT so that the second condition implies D + DT > 0
D > 0. This may also be deduced from the fact that C(sIn A)1 B + D =
 +
i=1 CA
i1
Bsi + D. The next result may be useful to characterize SSPR
functions.

Lemma 2.55. [146] A proper reational matrix H(s) Cmm is SSPR if and
only if its principal minors Hi (s) Cii are proper rational SSPR matrices,
respectively, for i = 1, ..., m1, and det(H(j)+H T (j)) > 0 for all IR.

The next lemma is close to Theorem 2.34.

Lemma 2.56. Let G(s) Cmm be a proper rational matrix satisfying

det(Im + G(s)) = 0 for Re[s] 0. Then the proper rational matrix H(s) =
(Im + G(s))1 (Im G(s)) Cmm is SSPR if and only if G(s) is strictly
bounded real.
60 2 Positive Real Systems

Let us now illustrate the various denitions of PR, SPR and WSPR func-
tions on examples.

Example 2.57. Consider again an asymptotically stable rst order system

1
H(s) = , with > 0 (2.180)
s+
Let us check the conditions for H(s) to be SPR.
1. H(s) has only poles in Re[s] < 0
2. H(j) is given by

1 j
H(j) = = 2 (2.181)
+ j + 2
Therefore,

Re[H(j)] = > 0 (, ) (2.182)
2 + 2
2
lim 2 Re[H(j)] = 2lim 2 2 =>0
2 +

1 1
Therefore s+ is SPR. However s+ is not SSPR because there does
not exist a > 0 such that Re[H(j)] > , for all [, ] since

lim 2 +
2
2 = 0.

1 s
Example 2.58. Similarly it can be proved that H(s) = s and H(s) = s2 + 2
2
s+a
are PR but they are not WSPR. H(s) = 1 and H(s) = s+b2 are both SSPR.

The following is an example of a system that is WSPR but is not SPR.

Example 2.59. Consider the second order system

s++
H(s) = ; , > 0 (2.183)
(s + )(s + )
Let us verify the conditions for H(s) to be WSPR. H(j) is given by
j++
H(j) = (j+)(j+)

(j++)(j)(j)
= ( 2 +2 )( 2 + 2 ) (2.184)

(j++)(j(+) 2 )
= ( 2 +2 )( 2 + 2 )

Thus
 2.14 Strictly Positive Real (SPR) Systems 61

2 (+)+(+)( 2 )
Re[H(j)] = ( 2 +2 )( 2 + 2 )
(2.185)
(+)
= ( 2 +2 )( 2 + 2 ) > 0, for all (, )

so H(s) is weakly SPR. However H(s) is not SPR since

2 ( + )
lim =0 (2.186)
( 2 + 2 )( 2 + 2 )
2

s+
Example 2.60. [213] The transfer function (s+1)(s+2) is

PR if 0 3
WSPR if 0 < 3
SPR if 0 < < 3

Let us point out that other denitions exist for positive real transfer func-
tions, like the following one:

Denition 2.61. [430] [-PR] Let 0 < < 1. The transfer function H(s)
Cmm is said to be -positive real if it is analytic in Re[s] 0 and satises

( 2 1)H
(s)H(s) + ( 2 + 1)(H
(s) + H(s)) + ( 2 1)Im 0 (2.187)

for all s Re[s] 0.

Then the following holds:

Proposition 2.62. [430] If a system is positive real, then it is SSPR.

Conversely, if a system is SSPR, then it is positive real for some 0 < < 1.

For single input-single output systems (m = 1) the index can be used

to measure the maximal phase dierence of transfer functions. The transfer
function H(s) C is PR if and only if the Nyquist plot of H(s) is in the
1+ 2 2
circle centered at 1 2 and radius 1 2 .

Lemma 2.63. [430] Let m = 1. If the system (A, B, C, D) with transfer

function H(s) = C(sIn A)1 B + D is PR, then

2
|arg(H(s))| arctan for all Re[s] 0 (2.188)
1 2
62 2 Positive Real Systems

Other classes of PR systems exist which may slightly dier from the above
ones; see e.g. [149,245]. In particular a system is said to be extended SPR if it
is SPR and if H(j)+H T (j) > 0. From the series expansion of a rational
+
transfer matrix one deduces that H(j) = i=1 CAi1 B(j)i + D which
implies that D + DT > 0. The denition of SSPRness in [245, Denition 3]
and Denition 2.54 are not the same, as they impose that H()+H T () 0
only, with lim 2 [H(j) + H T (j)] > 0 if H() + H T () is singular.
The notion of marginally SPR (MSPR) transfer functions is introduced in
[245]. MSPR functions satisfy inequality 2 of Denition 2.53, however they
are allowed to possess poles on the imaginary axis.

2.15 Applications
2.15.1 SPR and Adaptive Control

The concept of SPR transfer functions is very useful in the design of some
type of adaptive control schemes. This will be shown next for the control of
an unknown plant in a state space representation and it is due to Parks [394]
(see also [240]). Consider a linear time-invariant system in the following state
space representation

x(t) = Ax(t) + Bu(t)

(2.189)
y(t) = Cx(t)
with state x(t) IRn , input u(t) IR and output y(t) IR. Let us assume
that there exists a control input

u = LT x + r(t) (2.190)
where r(t) is a reference input and L IR , such that the closed loop system
n

behaves as the reference model

xr (t) = (A BLT )xr (t) + Br(t)
(2.191)

yr (t) = Cxr (t)
We also assume that the above reference model has an SPR transfer func-
tion. From the Kalman-Yakubovich-Popov Lemma, which will be presented
in detail in the next chapter, this means that there exists a matrix P > 0, a
matrix L , and a positive constant such that
T
Acl P + P Acl = L LT P
(2.192)

PB = CT
where
 2.15 Applications 63

Acl = A BLT
Since the system parameters are unknown, let us consider the following
adaptive control law:

u = LT x + r(t)
(2.193)
= LT x + r(t) LT x
where L is the estimate of L and L is the parametric error

L(t) = L(t) L
Introducing the above control law into the system (2.189) we obtain

x(t) = (A BLT )x(t) + B(r(t) LT (t)x(t)) (2.194)

Dene the state error x = x xr and the output error e = y yr . From
the above we obtain
dx
dt (t) = Acl x(t) B LT (t)x(t)
(2.195)

e(t) = C x(t)
Consider the following Lyapunov function candidate

V (x, L) = xT P x + LT PL L (2.196)
where P > 0 and PL > 0. Therefore

T dL
V (x, L) = xT (Acl P + P Acl )x 2xT P B LT x + 2LT PL
dt
Choosing the following parameter adaptation law

dL
(t) = PL1 x(t)e(t) = PL1 x(t)C x(t)
dt
we obtain
T
V (x, L) = xT (Acl P + P Acl )x 2xT (P B C T )LT x
Introducing (2.192) in the above we get

V (x) = xT (L LT + P )x 0 (2.197)

It follows that x, x and L are bounded. Integrating the above we get

t
xT (s)(L LT + P )x(s)ds V (x (0) , L (0))
0

Thus x L2 . From (2.195) it follows that dx

dt () is bounded and we conclude
that x() converges to zero.
64 2 Positive Real Systems

2.15.2 Adaptive Output Feedback

In the previous section we presented an adaptive control based on the as-

sumption that there exists a state feedback control law such that the result-
ing closed-loop system is SPR. In this section we present a similar approach
but this time we only require output feedback. In the next section we will
present the conditions under which there exists an output feedback that ren-
ders the closed loop SPR. The material in this section and the next have been
presented in [219]. Consider again the system (2.189) in the MIMO (multiple-
input multiple-output) case, i.e., with state x(t) IRn , input u(t) IRm
and output y(t) IRp . Assume that there exists a constant output feedback
control law

u(t) = Ky(t) + r(t) (2.198)

such that the closed loop system

x(t) = Ax(t) + Br(t)
(2.199)

y(t) = Cx(t)
with

A = A BKC
is SPR, i.e. there exists a matrix P > 0, a matrix L , and a positive constant
such that 3
T
A P + P A = L LT P
(2.200)

PB = CT
Since the plant parameters are unknown, consider the following adaptive con-
troller for r(t) = 0:
u(t) = K(t)y(t)
where K(t) is the estimate of K at time t. The closed loop system can be
written as

x(t) = Ax(t) B(K(t) K)y(t)

y(t) = Cx(t)
Dene

K(t) = K(t) K
3
Similarly to in the foregoing section, this is a consequence of the Kalman-
Yakubovich-Popov Lemma for SPR systems.
 2.15 Applications 65

and consider the following Lyapunov function candidate

 
V (x, K) = xT P x + tr K T 1 K

where > 0 is an arbitrary positive denite matrix. The time derivative of

V () along the system trajectories is given by

 
T 1 d
V (x, K) = x (A P + P A)x 2x P B Ky + 2tr K
T T T
K
dt

Introducing (2.189) and (2.200) we obtain


d  
V (x, K) = xT (AT P + P A)x 2tr Kyy T K T 1 K
dt

Choosing the parameter adaptation law

d  
K (t) = y(t)y T (t)
dt
and introducing (2.192) we obtain

V (x) = xT (L LT + P )x 0

Therefore V () is a Lyapunov function and thus x() and K() are both
bounded. Integrating the above equation it follows that x L2 . Since x()
is also bounded we conclude that x(t) 0 as t 0.
Hence the proposed adaptive control law stabilizes the system as long as
the assumption of the existence of a constant output feedback that makes the
closed-loop transfer matrix SPR is satised. The conditions for the existence
of such control law are established in the next section. Further work on this
topic may be found in [42] who relax the symmetry of the Markov parameter
CB.

2.15.3 Design of SPR Systems

The adaptive control scheme presented in the previous section motivates the
study of constant output feedback control designs such that the resulting
closed-loop is SPR. The positive real synthesis problem is important in its own
right and has been investigated by [179, 428, 480, 505]. This problem is quite
close to the so-called passication or passivation by output feedback [153,
156,280]. Necessary and sucient conditions have been obtained in [219] for a
linear system to become SPR under constant output feedback. Furthermore,
they show that if no constant feedback can lead to an SPR closed-loop system,
then no dynamic feedback with proper feedback transfer matrix can do it
neither. Hence, there exists an output feedback such that the closed-loop
66 2 Positive Real Systems

system is SPR if and only if there exists a constant output feedback rendering
the closed-loop system SPR.
Consider again the system (2.189) in the MIMO case, i.e., with state x(t)
IRn , input u(t) IRm and output y(t) IRp and the constant output feedback
in (2.198). The closed loop is represented in Figure 2.20 where G(s) is the
transfer function of the system (2.189). The equation of the closed-loop T (s)
of Figure 2.20 is given in (2.199).

+ u y
G(s)

Fig. 2.20. Closed-loop system T (s) using constant output feedback

Theorem 2.64. [41] Any strictly proper strictly minimum-phase system

(A, B, C) with the m m transfer function G(s) = C(sIn A)1 B and with
CB > 0 and symmetric, can be made SPR via constant output feedback.
The fact that the zeroes of the system have to satisfy Re[s] < 0 is cru-
s2 +1
cial. Consider G(s) = (s+1)(s+2)(s+5) . There does not exist any static output
feedback u = ky + w which renders the closed-loop transfer function PR.
Indeed if 2 = 9k
8k then Re[T (j)] < 0 for all k < 0. Therefore the strict
minimum phase assumption is necessary. Recall that a static state feedback
does not change the zeroes of a linear time invariant system. We now state
the following result where we assume that B and C are full rank.
Theorem 2.65 (SPR synthesis [219]). There exists a constant matrix K
such that the closed-loop transfer function matrix T (s) in Figure 2.20 is SPR
if and only if

BT C = C T B > 0
and there exists a positive denite matrix X such that
T T
C herm{B XB A}C < 0
When the above conditions hold, K is given by
 2.15 Applications 67

K = C Z(I C (C
T
ZC )1 C
T
Z)C T + S
where Z = herm{P A} and P = C(B T C)1 C T + B XB
T
, and S is an arbi-
trary positive denite matrix.

The notation used above is herm{X} = 12 (X + X ), and X is dened as

T
X X = 0 and X T
X = In , X IRnn .
In the single-input single-output case, the necessary condition B T C > 0
implies the relative degree of G(s) is one. It is noteworthy that the above two
results apply to systems with no feedthrough term, i.e. D = 0. An answer
is provided in [480, Theorem 4.1], where this time one considers a dynamic
output  feedback.
The system
 (A, B, C, D) is partitioned as B = [B1 B2 ],
C1 D11 D12
C = ,D= . It is assumed that (A, B2 ) is stabilizable and
C2 D21 0
that (A, C2 )is detectable. The closed-loop
system is said internally stable if
A + B2 DK C2 B2 CK
the matrix is stable (has eigenvalues with strictly
BK C2 AK
negative real parts), where (AK , BK , CK , DK ) is the dynamic feedback con-
troller.

Theorem 2.66. [480] There exists a strictly proper (dynamic) state feedback
such that the closed-loop system is internally stable and extended SPR if and
only if there exists two matrices F and L such that
T
D11 + D11 >0
The algebraic Riccati inequality
T 1
(A + B2 F )T P + P (A + B2 F ) + (C1 + D12 F B1T P )T (D11 + D11 ) .

.(C1 + D12 F B1T P ) < 0

(2.201)
has a positive denite solution Pf
The algebraic Riccati inequality

T 1
(A + LC2 )T G + G(A + LC2 ) + (B1 + LD12 GC1T )T (D11 + D11 ) .

.(B1 + LD12 GC1T ) < 0

(2.202)
has a positive denite solution Gf ,
The spectral radius (Gf Pf ) < 1.

The conditions such that a system can be rendered SPR via static state
feedback are relaxed when an observer is used in the control loop. However
68 2 Positive Real Systems

this creates additional diculty in the analysis because the closed-loop system
loses its controllability. See Section 3.4 for more information. Other works
related with the material exposed in this section, may be found in [49, 50,
177, 205, 330, 448, 465, 497, 515, 516]. Despite there being no close relationship
with the material of this section, let us mention [19] where model reduction
which preserves passivity is considered. Spectral conditions for a single-input
single-output system to be SPR, are provided in [455]. The SPRness is also
used in identication of LTI systems [12]. Robust stabilisation when a PR
uncretainty is considered is studied in [180].
3
Kalman-Yakubovich-Popov Lemma

The Kalman-Yakubovich-Popov Lemma (also called the Yakubovich-Kalman-

Popov Lemma) is considered to be one of the cornerstones of Control and
Systems Theory due to its applications in absolute stability, hyperstability,
dissipativity, passivity, optimal control, adaptive control, stochastic control
and ltering. Despite its broad applications the Lemma has been motivated
by a very specic problem which is called the absolute stability Lure problem
[321,408]. The rst results on the Kalman-Yakubovich-Popov Lemma are due
to Yakubovich [518,519]. The proof of Kalman [247] was based on factorization
of polynomials, which were very popular among electrical engineers. They later
became the starting point for new developments. Using general factorization of
matrix polynomials, Popov [407,409] obtained the Lemma in the multivariable
case. In the following years the Lemma was further extended to the innite
dimensional case (Yakubovich [520], Brusin [87], Likhtarnikov and Yakubovich
[300]) and discrete-time case (Szego and Kalman [483]).
The Kalman-Yakubovich-Popov Lemma (which will be shortly denoted as
the KYP Lemma) establishes an equivalence between the conditions in the
frequency domain for a system to be positive real, an input-output relationship
of the system in the time domain, and conditions on the matrices describing
the state-space representation of the system. A proof of this Lemma in the
multivariable case is also due to Anderson [11]. This result is very useful in
the stability analysis of dynamical systems and is also extensively used in
the analysis of adaptive control schemes. We will use this Lemma to prove
the Passivity Theorem which ensures the stability of a closed loop system
composed of two passive systems connected in negative feedback. Both results
are extensively used in the analysis and synthesis of dynamical systems. The
reader is referred to the survey [36] for more details on the history of the KYP
Lemma.
70 3 Kalman-Yakubovich-Popov Lemma

3.1 The Positive Real Lemma

3.1.1 PR Transfer Functions

Let us consider a multivariable linear time-invariant system described by the

following state-space representation


x(t) = Ax(t) + Bu(t)



y(t) = Cx(t) + Du(t) (3.1)





x(0) = x0

where x(t) IRn , u(t), y(t) IRm with n m. The Positive Real Lemma can
be stated as follows [8].

Lemma 3.1 (Positive Real Lemma or KYP Lemma). Let the

system in (3.1) be controllable and observable. The transfer function
H(s) = C(sIn A)1 B + D, with A IRnn , B IRnm , C
IRmn , D IRmm is PR with H(s) IRmm , s C, if and only
if there exists matrices P = P T > 0, P IRnn , L IRnm and
W IRmm such that:

P A + AT P = LLT
P B C T = LW (3.2)
D + DT = W T W

The proof will be given below.

Example 3.2. Let us point out an important fact. It is assumed in Lemma 3.1
that the representation (A, B, C, D) is minimal. Then PRness of the trans-
fer function C(sIn A)1 B + D is equivalent to the solvability of the set of
equations (3.2) with P = P T > 0. Consider now the following scalar exam-
ple, where (A, B, C, D) = (, 0, 0, 1), with > 0. The transfer function
 is
2p 0
H(s) = 0 that is PR. The set of equations (3.2) takes the form 0,
0 2
which is satised for any p 0. Obviously, however, this system is neither
controllable nor observable. This example shows that the minimality assump-
tion is not necessary for the set of equations (3.2) to possess a positive denite
solution. We shall come back on this topic in Section 3.3.

The rst equation above is known as the Lyapunov equation. Note that
LLT is not positive denite but necessarily semi-positive denite as long as
m < n. The third equation above can be interpreted as the factorization
 3.1 The Positive Real Lemma 71

of D + DT . For the case D = 0, the above set of equations reduces to the

rst two equations with W = 0. If one comes back to the frequency domain
(Denitions 2.28 and 2.29) one sees that the stability is taken care of by the
rst equation in (3.2) while the other equations rather deal with the positivity.
As recalled in the introduction of this chapter, the rst published version of
the KYP Lemma was by Yakubovich [518, 519] in 1962, with D = 0. The set
of equations (3.2) can also be written as
 
P A AT P C T P B L  T 
= L W 0 (3.3)
C B P T
D+D T
W T

From (3.2) it follows that B T P B B T C T = B T LW . So if W = 0 one

gets CB = B T P B 0. If B is full column rank then CB > 0. Thus the
rst non-zero Markov parameter of the system is CB, which means that the
uniform relative degree of the system is equal to r = (1, ..., 1)T IRm . Before
presenting the proof of the KYP Lemma, let us state a number of interesting
results, which link the set of equations (3.2), the positive realness, and a new
tool that is named a dissipation equality.

Corollary 3.3. Let the system in (3.1) be controllable and observable, and let
D = 0. Assume that C(sIn A)1 B is PR. Then

t
t
1
uT (s)y(s)ds = V (x(t)) V (x0 ) xT (s)(AT P + P A)x(s)d (3.4)
0 2 0

for all t 0, with V (x) = 12 xT P x, P satises the LMI in (3.3), and the
equality is computed along state trajectories starting at x(0) = x0 and driven
by u() on [0, t].

Proof: Positive realness and minimality imply that (3.2) is satised. By

t
simple calculation of the integral 0 uT (s)y(s)ds and using the KYP Lemma
conditions, premultiplying x(t) by P , (3.4) follows.
The same holds if D = 0, as the reader may check. We shall see in the next
chapter that V (x) is a storage function for the system (A, B, C), and that the
equality in (3.4) is a dissipation equality. One may rewrite it as follows, with
an obvious physical interpretation:

t
1
V (x(t)) = V (x0 ) + xT (s)(AT P + P A)x(s)ds +
    2 0
energy at time t initial energy  
dissipated energy
(3.5)

t
+ uT (s)y(s)ds
0
 
externally supplied energy
72 3 Kalman-Yakubovich-Popov Lemma

where we recall that AT P + P A 0. A dynamical system which satises (3.5)

along its trajectories is named dissipative. Notice that the minimality of the
triple (A, B, C) is used in Corollary 3.3, which therefore shows that PRness
implies the dissipation equality (3.4). However the following is also true.

Corollary 3.4. Let the triple (A, B, C) be given, where the matrices have ap-
propriate dimensions. Suppose that the KYP Lemma set of equations (3.2) is
solvable, i.e. there exists a triple (P = P T > 0, L, W ) that solves (3.2). Then
the dissipation equality (3.4) holds along the systems trajectories.

Proof: One has x(t) = Ax(t) + Bu(t) P x(t) = P Ax(t) + P Bu(t)

xT (t)P x(t) = xT (t)P Ax(t) + xT (t)P Bu(t) xT (t)P x(t) xT (t)P Ax(t)
xT (t)P Bu(t) + uT (t)y(t) = uT (t)y(t). Integrating between 0 and t we de-
t
duce that 12 xT (t)P x(t) 12 xT (0)P x(0) 12 0 xT ( )(P A + AT P )x( )d +
t T t
u ( )(B T P C)x( )d = 0 uT ( )y( )d . From the second equation in
0 t
(3.2) we get that 12 xT (t)P x(t) 12 xT (0)P x(0) 12 0 xT ( )(P A+AT P )x( )d =
t T
0 u ( )y( )d which is (3.4).
The interest of Corollary 3.4 is that no minimality on (A, B, C) is required
1
. We let the reader treat the case where D = 0, using Proposition A.63.
Corollary 3.3 assumes minimality but shows a stronger result, namely that
H(s) P R (3.2) (3.4). The issues linked to minimality and the KYP
Lemma are examined in Section 3.3. t
One notices from (3.4) that if x0 = 0 then 0 uT (s)y(s)ds 0: this in-
equality is always true for positive real transfer functions. This is to be linked
with Denition 2.1 (the constant is shown to be equal to V (x0 )), and
to Theorem 2.2: the function V (t) in Theorem 2.2 actually is a function of
the state x and is not an explicit function of time! As the reader may have
guessed, it plays the role of a Lyapunov function for the uncontrolled system
x(t) = Ax(t).
Corollary 3.3 proves that a minimal system satisfying (3.2) satises (3.4).
It is also of interest to show the converse: suppose that the system (3.1) with
D = 0 satises (3.4) for some positive denite quadratic function V (x). Then
does it satisfy the KYP Lemma conditions? The answer is yes. Indeed notice
rst that the dissipation equality (3.4) is equivalent to its innitesimal form
1
uT (t)y(t) = xT (t)P x(t) xT (t)(AT P + P A)x(t) (3.6)
2
since it holds for all t 0. Continuing the calculations we get

1
uT (t)Cx(t) = xT (t)P (Ax(t) + Bu(t)) xT (t)(AT P + P A)x(t) (3.7)
2

1
Let A IRnn be the transition matrix. Minimality of n is equivalent to having
(A, B) controllable and (A, C) observable.
 3.1 The Positive Real Lemma 73

so that uT (t)Cx(t) = xT (t)P Bu(t). Since this equality holds for any u() one
must have C T = P B. This shows that the second KYP Lemma condition
is true. Now suppose that more generally the system satises a dissipation
equality as

t

1 t T
u (s)y(s)ds = V (x(t)) V (x0 )
T
x (s)Qx(s)ds (3.8)
0 2 0
with Q 0 and V (x) = 12 xT P x, P = P T > 0. Then the uncontrolled system
is stable in the sense of Lyapunov since V (x(t)) V (x(0)) for all t 0. Thus
AT P + P A 0 from Lyapunovs Theorem. Using once again the innitesimal
version of the dissipation equality we get
1
uT (t)y(t) = xT (t)(P A + AT P )x(t) xT (t)Qx(t)
2
This must hold for any admissible input. Rewriting this equality with u() 0
we obtain that necessarily P A+AT P = Q = LLT for some matrix L. Thus
we have proved the following.

Corollary 3.5. Let (3.8) hold along the systems trajectories with Q 0,
V (x) = 12 xT P x, P = P T > 0. Then the KYP Lemma set of equations (3.2)
also hold.

Remark 3.6. In the case D = 0, assuming that the dissipation equality (3.8)
holds yields after time-derivation

1 1
uT (C B T P )x + uT Du xT (AT P + P A)x = xT Qx 0 (3.9)
2 2
since Q 0. In a matrix form this leads to

AT P P A C T P B 
(xT uT ) x 0 (3.10)
u
C BT P D + DT
Using Proposition A.63, (3.2) follows.

We have seen in the proofs of Theorems 2.6 and 2.21 that Parsevals Theo-
t
rem allows us to assert that if H(s) is PR then 0 uT ( )y( )d 0, where the
underlying assumption is that x(0) = 0, and conversely (see Corollary 2.35).
t
Obviously the dissipation equality implies 0 uT ( )y( )d 0 when x(0) = 0.
Therefore concatenating all these results we get the following.
74 3 Kalman-Yakubovich-Popov Lemma

KYP Lemma matrix equality (3.2)

 ((A, B, C, D) minimal)
t
PR transfer function 0 uT ( )y( )d 0 when x(0) = 0

Dissipativity with quadratic storage function

These developments and results somewhat shed new light on the relation-
ships between PR transfers, passivity, dissipation, and the KYP Lemma set
of equations. However we have not yet proved the KYP Lemma, i.e. the fact
that the frequency domain conditions for positive realness, are equivalent to
the LMI in (3.2) when (A, B, C, D) is minimal. Several proofs of the KYP
Lemma appeared in the book [8].

Proof of the KYP Lemma: The proof that is reproduced now is taken
from Andersons work [11].

Suciency: This is the easy part of the proof. Let the set of equations in
(3.2) be satised. Then

H(s) + H T (s) = DT + D + B T (sIn AT )1 C T + C(sIn A)1 B.

 
.W T W + B T (sIn AT )1 P + P (sIn A)1 B+

+B T (sIn AT )1 LW + W T LT (sIn A)1 B

 
= W T W + B T (sIn AT )1 P (s + s) P A AT P (sIn A)1 +

+B T (sIn AT )1 LW + W T LT (sIn A)1 B

= W T W + B T (sIn AT )1 LW + W T LT (sIn A)1 B+

+B T (sIn AT )1 LLT (sIn A)1 B+

+B T (sIn AT )1 P (sIn A)1 B(s + s)

  
= W T + B T (sIn AT )1 L W + LT (sIn A)1 B +

+B T (sIn AT )1 P (sIn A)1 B(s + s)

(3.11)
 3.1 The Positive Real Lemma 75

which is nonnegative denite for all Re[s] > 0.

Necessity: Suppose that rank(H(s)+H T (s)) = r almost everywhere. From

the PRness it follows that there exists an r m matrix W (s) such that

H(s) + H T (s) = W0T (s)W0 (s) (3.12)

and
(i) W0 () has elements which are analytic in Re[s] > 0, and in Re[s] 0
if H(s) has elements which are analytic in Re[s] 0.
(ii) Rank(W0 (s) = r in Re[s] > 0.
(iii) W0 (s) is unique save for multiplication on the left by an arbitrary
orthogonal matrix.
This is a Youla factorization. Suppose that all poles of H(s) are in
Re[s] < 0 (the case when poles may be purely imaginary will be treated
later). Equivalently all the eigenvalues of A have negative real parts,i.e. A is
asymptotically stable. From Lemmas A.66 and A.68 (with a slight adaptation
to allow for the direct feedthrough term) it follows that there exists matrices
L and W = W0 () such that W0 (s) has a minimal realization (A, B, L, W ),
with two minimal realizations for H(s) + H T (s) = W0T (s)W0 (s) being
given by

T
A 0 B C
(A1 , B1 , C1 , W T W ) = , , ,WTW (3.13)

0 AT CT B

and

A 0 B P B + LW
(A3 , B3 , C3 , W T W ) = , , ,WTW

0 AT P B + LW B
(3.14)
where P is the unique symmetric positive denite solution of P A + AT P =
LLT . From Lemma A.69 there exists
nonsingular
 matrices
T commuting
 T
A 0 B B 1 T C
with and such that T = and (T ) =
 0 A T
C T
P B + LW B
P B + LW
. By Corollaries A.67, A.17 and A.70 there exists T1 commuting
B
with A such that T1 B = B, and (T11 )T C T = P B + LW . Now since T1
commutes with A one has

[B, AB, ...] = [T1 B, AT1 B, ...] = [T1 B, T1 AB, ...] = T1 [B, AB, ...] (3.15)
76 3 Kalman-Yakubovich-Popov Lemma

The matrix [B, AB, ...] has rank n because of the minimality of the real-
ization. Thus T1 = In and thus P B + LW = C T . The third equation in (3.2)
follows by setting s = in H(s) + H T (s) = W0T (s)W0 (s).
In a second step let us relax the restriction on the poles of H(s). In this
case H(s) = H1 (s) + H2 (s) where H1 (s) has purely imaginary axis poles, and
H2 (s) has all its poles in Re[s] < 0, and both H1 (s) and H2 (s) are positive
real. Now from Lemma A.71 it follows that there exists P1 = P1T > 0 such that
P1 A1 +AT1 P1 = 0 and P1 B1 = C1T , where (A1 , B1 , C1 ) is a minimal realization
of H1 (s). For H2 (s) we may select a minimal realization (A2 , B2 , C2 , D2 ) and
using the material just proved above we may write

P2 A2 + A2 P2 = L2 L2
T T



P2 B2 = C2T L2 W (3.16)




T
W W = D2 + D2T
It can be veried that the KYP Lemma set of equations (3.2) is satied by
taking P = P1 +P2 , A = A1 +A2 , B T = [B1T B2T ], C = [C1 C2 ], LT = [0 LT2 ].
Moreover with (A1 , B1 , C1 ) and (A2 , B2 , C2 , D2 ) minimal realization sof H1 (s)
and H2 (s), (A, B, C, D2 ) is a minimal realization of H(s). Indeed the degree
of H(s) is the sum of the degrees of H1 (s) and H2 (s) which have no common
poles. It just remains to verify that the equations (3.2) hence constructed are
valid under any (full rank) coordinate transformation, since they have been
established for a particular form A1 + A2 .
The KYP Lemma has been derived in the so-called behavioural framework
in [162].

3.1.2 A Digression to Optimal Control

We will deal at several places in the book with optimal control and its link
with dissipativity. Let us nevertheless point out a rst relationship. Provided
D + DT is full-rank (i.e. D + DT > 0 in view of (3.2)), the matrix inequality
in (3.3) is equivalent to the following algebraic Riccati inequation:

P A AT P (C B T P )T (D + DT )1 (C B T P ) 0 (3.17)

Equivalence means that the LMI and the Riccati inequality possess the
same set of solutions P . The KYP Lemma says that if the transfer function
D + C(sIn A)1 B is PR and (A, B, C, D) is minimal, then they both possess
at least one solution P = P T > 0. Let us recall that the optimal control
problem

+
min J (x0 , u) = (xT (t)Qx(t) + uT (t)Ru(t))dt (3.18)
uU 0
 3.1 The Positive Real Lemma 77

under the constraints (3.1) and with R > 0, Q 0, has the solution
u
(x) = R1 B T P x where P is a solution of the Riccati equation P A
AT P + P BR1 B T P = Q 0. When the cost function contains cross terms
2xT Su then P is the solution of the Riccati equation P A AT P (S
B T P )R1 (S T P B) = Q 0 and the optimal control is u
(x) = R1 (S T
B T P )x. The Belmann function for these problems is the quadratic function
V (x) = xT P x and V (x0 ) = minuU J (x0 , u). If Q > 0 then P > 0 and V (x)
is a Lyapunov function for the closed-loop system x(t) = Ax(t) + Bu
(x(t)),
as can be checked by direct calculation of V (x(t)) along the closed-loop tra-
jectories.
Therefore the Riccati inequality in (3.17) corresponds to the Riccati in-
equation of an innite horizon LQ problem whose cost matrix is given by

Q CT
(3.19)
T
C D+D

where D + DT = W T W (= R) is the weighting matrix corresponding to u

in the cost function, S = C and Q = LLT 0. The equivalence between
(3.3) and the Riccati inequality also holds with strict inequality (> 0) in both
(3.3) and (3.17). To recapitulate, the positive realness of the controllable and
observable LTI system (3.1) is equivalent to the KYP Lemma conditions (3.2),
which are equivalent to the linear matrix inequality (3.3), which is equivalent
to the Riccati inequality (3.17), whose solution provides the optimal feedback
control that corresponds to the optimal control problem in innite horizon
with cost matrix (3.19). All this is relying on the condition D + DT > 0. The
controllability assumption on the system (3.1) can be interpreted in the light
of the optimal control problem, in the sense that controllability implies the
existence of some u() such that J (x0 , u) < +.
The proof of the equivalence between the Riccati inequality and the linear
matrix inequality follows from Theorem A.61, which is instrumental in char-
acterizing dissipative systems with Riccati and partial dierential inequal-
ities. The reader may have a look at Appendix A.5 where several results
of matrix
algebra are recalled. We may apply Lemma A.62 to the matrix
D + DT C BT P
M = . Then rank(M ) = m is equivalent
C P B P A A P LL
T T T

to the Riccati equality

P A + AT P + LLT + (C T P B)(D + DT )1 (C B T P ) = 0 (3.20)

which is (3.17) with = instead of . As we shall see further in the book, a

Riccati equation for a PR system corresponds in the nonlinear case to a partial
dierential inequation (Hamilton-Jacobi inequalities), whose solutions serve as
Lyapunov functions candidates. The set of solutions is convex and possesses
78 3 Kalman-Yakubovich-Popov Lemma

two extremal solutions (which will be called the available storage and the
required supply) which satisfy the algebraic Riccati equation, i.e. (3.17) with
equality, see Section 4.4.3, Lemma 4.47 and Proposition 4.48. More details
between the KYP Lemma and optimal control will be given in Section 3.8.
The case when D + DT = 0 and D + DT 0 will be treated in Section 4.6.
Such cases possess some importance. Indeed PR functions may not have a
realization with a full rank matrix D. Let us end this subsection by recalling
another equivalence: the system (A, B, C, D) with a minimal realization and
D + DT > 0 is PR if and only if the Hamiltonian matrix

A B(D + DT )1 C B(D + DT )1 B T
(3.21)
T 1 T 1 T
C (D + D ) C A + C (D + D ) B
T T T

has no pure imaginary eigenvalues. This is a way to characterize SSPR transfer

matrices. Indeed notice that
+

H(s) = C(sIn A)1 B + D = CAi1 Bsi + D
i=1

so that H() = D. The SSPRness thus implies by Denition 2.54 (2) that
D > 0 (or D + DT Im > 0 if m 2). It is noteworthy that D + DT >
0 D > 0; however D is not necessarily symmetric.

3.1.3 Duality

The linear matrix inequality (3.3) thus denes a set P of matrices P > 0.

Lemma 3.7 (duality). Let (A, B, C, D) be such that the set P is not empty.
The inverse P 1 P 1 of any element of P is a solution of the dual problem
(AT , C T , B T , D).

Remember that the adjoint system is dened as (AT , C T , B T , D).

Proof of Lemma 3.7: Clearly if P > 0 then P 1 > 0. From the following
matrix relation

AP 1 P 1 AT B P 1 C T
=
1
B CP
T
R
(3.22)
P 1 0 AT P P A CT P B P 1 0
=
0 In CB PT
R 0 In

one sees that P 1 P if P P, because the two matrices

 3.1 The Positive Real Lemma 79

AP 1 P 1 AT B P 1 C T

1
B CP
T
R
and

AT P P A CT P B

C B P
T
R
are simultaneously negative denite. The set P is the set that solves the KYP
Lemma linear matrix inequality for the dual system.

3.1.4 Positive Real Lemma for SPR Systems

Consider the set of equations in (3.2) and Denition 2.42 of a SPR transfer
function. Assume that a realization of the input-output system is given by the
quadruple (A, B, C, D), i.e. C(sIn A)1 B + D = H(s), and (A, B, C, D) is
minimal. Then H(s ) = C(sIn In A)1 B + D, and a realization of
H(s ) is given by (A + In , B, C, D). Saying that H(s ) is PR is therefore
equivalent to stating that (A + In , B, C, D) satises the KYP Lemma set of
equations (3.2), provided (A+ In , B, C, D) is minimal. Therefore (A, B, C, D)
is SPR if and only if (A + In )T P + P (A + In ) = LLT and the last two
equations in (3.2) hold, with P = P T > 0. The rst equation can be rewritten
as AT P + P A = LLT 2 P . As is well known, this implies that the matrix
A is Hurwitz, i.e. all its eigenvalues have negative real parts. Indeed consider
the Lyapunov function V (x) = xT P x. Then along trajectories of the system
x(t) = Ax(t) one obtains V (x(t)) = xT (t)(LLT 2 P )x(t) 2 V (x(t)).
Consequently the system is exponentially stable. This in particular shows
that SPR transfer functions have poles with negative real parts, and conrms
Theorem 2.45.

The Lefschetz-Kalman-Yakubovich Lemma

We now present the Lefschetz-Kalman-Yakubovich (LKY) Lemma which gives

necessary and sucient conditions for a system in state space representation
to be SPR.

Lemma 3.8 (Multivariable LKY Lemma). [485] Consider the system in

(3.1), with m 2. Assume that the rational transfer matrix H(s) = C(sI
A)T B + D has poles which lie in Re[s] < where > 0 and (A, B, C, D) is
a minimal realization of H(s). Then H(s ) for > 0 is PR if and only if
a matrix P = P T > 0, and matrices L and W exist such that
80 3 Kalman-Yakubovich-Popov Lemma

P A + AT P = LLT 2P
P B C T = LW (3.23)

D + DT = W T W.

The conditions in (3.23) are more stringent than those in (3.3). Notice
that the rst line in (3.23) can be rewritten as

P (In + A) + (AT + In )P = LLT (3.24)

which allows one to recover (3.3) with A changed to In + A. The transfer
function of the triple (In +A, B, C) precisely is H(s). Thus (3.23) exactly
states that (In + A, B, C) is PR and satises (3.3).
It is assumed in Lemma 3.8 that the system is multivariable, i.e. m 2.
The LKY Lemma for monovariable systems (m = 1) is as follows.

Lemma 3.9 (Monovariable LKY Lemma). [485] Consider the system in

(3.1), with m = 1. Suppose that A is such that det(sIn A) has only zeroes
in the open left-half plane. Suppose (A, B) is controllable, and let > 0,
L = LT > 0 be given. Then a real vector q and a real matrix P = P T > 0
satisfying

qq T L
P A + AT P =
(3.25)
P B C T = 2Dq
exist if and only if H(s) is SPR and is suciently small.

Lemma 3.8 is not an extension of Lemma 3.9 because the matrix L = LT >
0 is arbitrary in Lemma 3.9. We now state a result that concerns Denition
2.61.

Lemma 3.10. [430] Assume that the triple (A, B, C) is controllable and ob-
servable. The system whose realization is (A, B, C, D) is positive real if and
only if there exist matrices L and W such that

P A + AT P = (1 2 )C T C LT L



P B = (1 + 2 )C T (1 2 )C T D LT W (3.26)




T
W W = ( 2 1)Im + ( 2 1)DT D + ( 2 + 1)(D + DT )
 3.1 The Positive Real Lemma 81

Time Domain Conditions for Strict Positive Realness

The next result is due to J.T. Wen [508] who established dierent relationships
between conditions in the frequency domain and the time domain for SPR
systems.

Lemma 3.11 (KYP Lemma for SPR Systems). Consider the LTI, min-
imal (controllable and observable) system (3.1) whose transfer matrix is given
by
H(s) = D + C(sIn A)1 B (3.27)
where the minimum singular value min (B) > 0. Assume that the system is
exponentially stable. Consider the following statements:

1. 1) There exist P > 0, P , L IRnn , min (L) = > 0, Q IRmn ,
W IRmm that satisfy the Lure equations

AT P + P A = QT Q L (3.28)

BT P C = W T Q (3.29)

W T W = D + DT (3.30)
1 ) Same as 1) except L is related to P by

L = 2P (3.31)
for some > 0.
2) There exists > 0 such that for all IR

H(j) + H (j) Im (3.32)

3) For all IR

H(j) + H (j) > 0 (3.33)

4) For all IR

H(j) + H (j) > 0 (3.34)

and

lim 2 (H(j) + H (j)) > 0 (3.35)

5) The system can be realized as the driving point impedance of a multi-

port dissipative network.
6) The
 Lure equations with L = 0 are satised by the internal parameter
set A + In , B, C, D corresponding to T (j ) for some > 0.
82 3 Kalman-Yakubovich-Popov Lemma

7) For all IR, there exists > 0 such that

H(j ) + H (j ) 0 (3.36)
8) There exists a positive constant and a constant (x0 ) IR, (0) = 0,
such that for all t 0

t
t
2
uT (s)y(s)ds (x0 ) + u(s) ds (3.37)
0 0

9) There exists a positive constant and a constant (x0 ), (0) = 0, such

that for all t 0

t
es uT (s)y(s)ds (x0 ) (3.38)
0

10) There exists

 a positive constant such that the following kernel is
positive in L2 IR+ ; IRmm :

K (t s) = D (t s) + Ce(A+I)(ts) B.I(t s) (3.39)

where and I denote the Dirac measure and the step function, respec-
tively.  
11) The following kernel is coercive in L2 [0, T ] ; IRmm , for all T :

K (t s) = D (t s) + CeA(ts) B.I(t s) (3.40)

These statements are related as follows:

= (2) (8) (11)







(if D > 0)



(1) = (1 ) (4) (5) (6) (7) (9) (10)



=



(if D = 0)





(3)

Proof:
(2) (1)
( L2 ((, ) ; IRm ) to min-
Consider the optimization problem of nding u
imize


) *
Jf = (
x (j)F T x u (j)(
((j) + 2( y (j) d

 3.1 The Positive Real Lemma 83

where the superscript denotes complex conjugate transposition and x (, y( and

( are the Fourier transforms of the x, x and u, respectively. By writing x
u (, in
terms of the initial condition and the input, the optimization index can be
expanded as


Jf =
{((jIn A)1 x0 + (jIn A)1 B(
u(j)) F T F ((jIn A)1 x0

+(jIn A)1 B( ( (j)[(C(jIn A)1 B + D)

u(j) + u

+(C(jIn A)1 B + D)](

u(j)

u(j)C(jIn A)1 x0 }d
+2(

Consider the problem as an L2 optimization. Then

Jf = R( ( + r, u
u, u ( + k,
where the inner products are in the L2 sense. A unique solutions exists if R
is a coercive L(L2 ) (the space of bounded operators in L2 ) operator. Now,

R = H (j) + H(j) B T (jIn AT )1 F T F (jIn A)1 B

By condition (2), if
+ +2
> +F (jIn A)1 B +H
then the operator R is coercive. By Plancherels Theorem, Jf can be trans-
formed back to the time domain as


 
J= x(t)T F T F x(t) + 2uT (t)y(t) dt

Since a unique solution of the optimal control problem exists, the necessary
conditions from the maximum principle must be satised. The Hamiltonian
is given by

H(x, u) = xT F T F x + 2uT (Cx + Du) + T (Ax + Bu)

where is the costate or the Lagrange multiplier. The feedforward D in uT Du
can be regarded as the symmetrized D. Since condition (2) implies D > 0,
there exists W > 0 such that D + DT = W T W . The optimal u is obtained by
minimizing H:
1  
u = W 1 W T 2Cx + B T .
2
The costate equation is governed by
84 3 Kalman-Yakubovich-Popov Lemma

= 2F T F x 2C T u AT .
It can be shown [88] that depends linearly on x. Let = 2P x. Then

   T
P A + AT P + F T F x = C B T P u

 T  
= C B T P W 1 W T C B T P x

Since the equality holds for all x, we have

P A + AT P = F T F QT Q
(3.41)

C = BT P W T Q
The rst equation implies P > 0. By identifying L with F T F and choosing
F T F > 0 and
2
min (F ) < 2
(jI A)1 BH
condition (1) is proved.
(1) (2)
(When D > 0)
Given the Lure equations, compute the Hermitian part of the transfer func-
tion as follows :

H(j) + H (j) = D + DT + C(jI A)1 B + B T (jIn AT )1 C T

= W T W + B T P WT Q (jIn A)1 B
+B T (jIn AT )1 P B QT W
= W T W + B T (jIn AT )1 [(jIn AT )1 P
+P (jIn A)](jIn A)1 B
W T Q(jIn A)1 B B T (jI T 1 T
 T A )  Q W
T 1
= W W + B (jIn A )
T T
Q Q + L (jIn A)1 B)
1 T 1 T
W T
 Q(jI n A) B B (jI
T
 n A ) Q W 1 
T 1 T
= W B (jIn A ) Q
T T
W Q(jIn A) B
+B T (jIn AT )1 L(jIn A)1 B 0

Assume condition (2) is false. Then there exist {un }, un  = 1, and {n }
such that
1
0 (H (jn ) + H (jn )) un , un 
n
As n , if n , then

(H (jn ) + H (jn )) un , un  Dun , un  min (D) > 0

 3.1 The Positive Real Lemma 85

which is a contradiction since the left-hand side converges to zero. Hence, un

and n are both bounded sequences and therefore contain convergent subse-
quences unk and nk . Let the limits be uo and o . Then

(H (jn ) + H (jn )) un , un  = 0.
This implies

W uo Q (jo In A)1 Buo = 0

1
L1/2 (jo In A) Buo = 0
Since L > 0, the second equality implies
1
(jo In A) Buo = 0
Substituting back to the rst equality yields W uo = 0. The positive de-
niteness of W (by assumption D > 0) implies contradiction. Hence, condition
(2) is satised.
(2) = (8)
Since (2) = (1), the Lure equation holds. Let
1 T
V (x) = x Px
2
Then

V (x(t)) = x(t)P Ax(t) + x(t)T P Bu(t)

1 1+ +
= xT (t)Lx(t) +Qx(t)2 + + uT (t)Cx (t) + uT (t)W T Qx(t)
2 2
1 T 1+ +
= x (t)Lx(t) +Qx(t)2 + uT (t)Du(t) + uT (t)W T Qx(t) +
2 2
+uT (t)y(t)
2 1 2
x(t) + uT (t)y(t) Qx(t) W u(t)
2 2
2
x(t) + uT (t)y(t)
2
By integrating both sides [421] for all t 0 we get

t
uT (s)y(s)ds V (xo ) (3.42)
0

Since (3.32) remains valid if D is replaced by D for suciently small,

(3.42) holds with y replaced by

y1 = Cx + (D )u
86 3 Kalman-Yakubovich-Popov Lemma

Then (3.42) becomes

t
t
u (s)y(s)ds
T
u(s)2 ds V (xo )
0 0

Identifying V (xo ) with (xo ) and with in (3.37), condition (8) follows.
(8) = (2)
Let t in (3.37), then

t

2
u (s)y(s)ds (xo ) +
T
u(t) dt
0 0
In particular, for xo = 0,

t

uT (s)y(s)ds u(t)2 dt
0 0
By Plancherels Theorem,



u (j)y(j)d u()2 d,

for all u L2 . Suppose that for each > 0, there exists w C and o IR
such that

w H(j)w > w2

By the continuity of w H(j)w in , there exists an interval around o
of length r such that
2
w H(j)w > w
for all . Let

w if
u(j) =
0 otherwise
Clearly, u L2 . Then

2
u (j)y(j)d = u (j)T (j)u(j)d < r w ,

and


2 2
u() d = r w .

If < , this is a contradiction. Hence, there exists an interval > 0 such

that (3.32) holds.
(8) = (11)
 3.1 The Positive Real Lemma 87

Condition (11) follows directly from condition (8).

(11) = (8)
The implication is obvious if xo = 0. In the proof of (8) = (2), xo is taken
to be zero. Therefore, for x = 0 (11) = (8) = (2). It has already been
shown that (2) = (8). Hence, (11) = (2) = (8).
(1 ) = (1)
By denition.
(if D = 0)
If D = 0, then W = 0. Rewrite (3.28) as

AT P + P A = QT Q L + 2P 2P
For small enough,

QT Q + L 2P 0
Hence, there exists Q1 such that

AT P + P A = QT1 Q1 2P
Since (3.29) is independent of Q1 when D = 0, (1 ) is proved.
(1 ) = (6)
By straightforward manipulation
(6) = (7)
Same as in (1) = (2) except L is replaced 2P .
(7) = (6)
Positive Real (or KYP) Lemma
(4) = (7)
For > 0 suciently small, AIn remains strictly stable. Now, by direct
substitution

H(j ) + H (j ) = D + DT + C(jIn A In )1 B+
+ B T (jIn AT In )1 C T
= H(j)
 + H (j)
+ C(jIn A)1 (jIn A In )1 B+ 
+B T (jIn AT In )1 (jIn AT )1 C T
(3.43)
Therefore for any w Cm ,

2w H(j )w
+ 2w H(j)w
++ + (3.44)
+
2 C B (jIn A)1 + +(jIn A In )1 + w
2

Since

(jIn A)x |(|| A) x|

88 3 Kalman-Yakubovich-Popov Lemma

it follows [375]
+ + 1
+(jIn A)1 +
||| A|
Then

2 C B w2

2w H(j )w 2w H(j)w
||| A| ||| A I|

By (3.34), for all , is compact in R, there exists k > 0, k dependent

on , such that
2
2w H(j)w k w (3.45)
By (2.7b), for all suciently large, there exists g > 0 such that
g 2
2w H(j)w w (3.46)
2
Hence, there exists 1 R large enough so that (3.45) and (3.46) hold
with some g and k dependent on 1 . Then, for || 1 ,

2
2 2 C B w
2w H(j )w k w (3.47)
||| A| ||| A In |
, -
2
2 2 C B w
k w sup
||1 ||| A| ||| A In |

and for || > 1 ,

2
g 2 2 C B w
2w H(j )w w (3.48)
2 ||| A| ||| A In |
. , -/
2 2
w 2 C B w
g sup
2 ||>1 ||| A| ||| A I|

The terms in curly brackets in (3.47) and (3.48) are nite. Hence, there
exists small enough such that (3.47) and (3.48) are both non-negative,
proving condition (7).
(7) = (4)
From (7) = (6), the minimal realization (A, B, C, D) associated T (j)
satises the Lure equation with L = 2P . Following the same derivation as
in (1) = (2), for all w Cm , we have
 3.1 The Positive Real Lemma 89

w (H(j ) + H (j ))w =
= w (W T + B T (jIn AT )1 QT )(W + Q(jIn AT )1 B)w
+2w B T (jIn AT )1 P (jIn AT )1 Bw (3.49)
2w B T (jIn AT )1 P (jIn AT )1 Bw
2
2min (P )min (B)
|||
A
|2
w

Since P is positive denite and, by assumption, min (B) > 0, T (j) is

positive for all IR.
It remains to show (3.35). Multiply both side of the inequality above by
2 , then

2 2min(P )min
2
(B) 2
2 w (H(j) + H (j))w 2 w
||| A|
As , the lower bound converges to 2 2min (P )min
2 2
(B) which is
positive.
(7) = (5)
If (3.36) is satised, H(j ) corresponds to the driving point impedance
of a miltiport passive network [8]. Hence, H(j) corresponds to the impedance
of the same network, with all C replaced by C in parallel with resistor of
conductance C and L replaced by L in series with a resistor of resistance
L. Since all C, L elements are now lossy, or dissipative, H(j) is the driving
point impedance of a dissipative network.
(7) = (5)
Reversing the above argument, if H(j) is the driving point impedance
of a dissipative network, all L and C elements a lossy. Hence, by removing
suciently small series resistance in L and parallel conductance in C, the
network will remain passive. Hence, again by [8], condition (7) is satised.
(6) = (9)
Let
1 t T
V (t, x(t)) = e x Px
2
Then

V (t, x(t)) =
1 t 1 t t
= e xT (t)P x(t) + e xT (t)(P A + AT P )x(t) + e xT (t)P Bu(t)
2 2
V (t, x(t))
e Qx(t) W u(t)2 + e uT (t)y(t)
t t
V (t, x(t))
2 P 

t
V (t, x(t)) + e uT (t)y(t)
2 P 

Choose 0 < < /2 P  . Then by comparison principle, for all T 0,

90 3 Kalman-Yakubovich-Popov Lemma

t
s
e uT (s)y(s)ds xTo P xo
0

(9) = (6)
Dene


u1 (t) = e
( /2 )t
u(t)
( /2 )t (3.50)
y1 (t) = e y(t)


x1 (t) = e( ) x(t)
/2 t

where > 0 is as given in (3.38). then

 
x1 (t) = A + 2 I x1 (t) + Bu1 (t)

(3.51)

y1 (t) = Cx1 (t) + Du1 (t)
The corresponding transfer function is

 1
H1 (j) = D + C jIn A In B
  2

= H j
2
By setting t = and xo = 0 in (3.38),

t
uT1 (s)y1 (s)ds 0
0
By Plancherels Theorem,


u1 (j)(T1 (j) + T1 (j))u1 (j) 0

Since this holds true for all u1 (j) L2

H1 (j) + H1 (j) 0
Equivalently
   
H j H j 0
2 2
proving (7)
(9) = (10)
Use the transformation in (3.50), then condition (10) follows directly from
condition (9) with = /2.
(10) = (9)
 3.1 The Positive Real Lemma 91

If xo = 0, (10) = (9) is obvious. Since in the proof of (9) = (6), only the
xo = 0 case is considered, it follows, for the xo = 0 case, (10) = (9) = (6).
It has already been shown that (6) = (9). Hence, (10) = (6) = (9).
(2) = (4) = (3)
The implications are obvious.
Remark 3.12. Stating H(j) + H
(j) In for all IR = (, +),
is equivalent to stating H(j) + H
(j) > 0 for all IR {, +} =
[, +]. This is dierent from H(j) + H
(j) > 0 for all IR because
such a condition does not imply the existence of a > 0 such that H(j) +
H
(j) In for all IR.
2
s 2
Example 3.13. If H(s) = s+1 , then H(j) + H
(j) = 1+ 2 , so H(s) is not

SPR despite Re[H()] = 2. But H(0) + H (0) = 0. If H(s) = s+2

s+1 , then
2
H(j) + H
(j) = 4+
1+ 2 1 for all [, +]. This tranfer function
1 2
is SSPR. If H(s) = s+1 , then H(j) + H
(j) = 1+ 2 > 0 for all

2 2
(, +). Moreover lim+ 1+ 2 > 0, so H(s) is SPR.
Further works on the characterization of PR or SPR transfer functions can
be found in [10, 49, 50, 132, 177, 205, 339, 340, 396, 448, 455, 479, 504, 530].

3.1.5 Descriptor Variable Systems

The KYP Lemma can be extended to a class of linear systems larger than
(3.1). Let us consider the following class of linear time invariant systems


E x(t) = Ax(t) + Bu(t)



y(t) = Cx(t) + Du(t) (3.52)





x(0 ) = x0
with A, E IRnn , B, C IRnm , and D IRmm . When the matrix E is
singular (i.e., its rank is < n) then the system in (3.52) is called singular or
descriptor system. Throughout this section we shall assume that rank(E) < n
since otherwise we are back to the classical regular case. Descriptor systems
arise in various elds of applications, like for instance constrained mechanical
systems, or electrical circuits, since Kirschos laws directly yield algebraic
equality constraints on the state. The next assumption will be supposed to
hold throughout the whole of this section.
Assumption 1 The pair (E, A) is regular, i.e. det(sE A) is not identically
zero, s C.
Let us recall some facts about (3.52). If the pair (E, A) is regular, then
there exists two square invertible matrices U and V such that the system can
be transformed into its Weierstrass canonical form
92 3 Kalman-Yakubovich-Popov Lemma

E x(t) = Ax(t) + Bu(t)



y(t) = Cx(t) + Du(t) (3.53)





x(0 ) = x0

A1 0 Iq 0 B1
with A = U AV = , E = U EV = , B = U B = ,
0 Iq 0 N B2
C = CV = (C1 C2 ). The (nq)(nq) matrix N is nilpotent, i.e. N l = 0 for
some integer l 1. Generally speaking, solutions of (3.52) are not functions
of time but distributions (i.e. the general solutions may contain Dirac and
derivatives of Dirac measures). The system is called impulse free if N = 0. To
better visualize this, let us notice that the transformed system can be written
as [106]

x1 (t) = A1 x1 (t) + B1 u(t)
(3.54)

N x2 (t) = x2 (t) + B2 u(t)
and the solution of (3.52) is x = x1 + x2 . One has

x1 (t) = exp(tA1 )xs (0) + exp(tA1 )  B1 u(t)
(3.55)
l1 (i1) l1
x2 (t) = i=1 0 N i x2 (0 ) i (i)
i=0 N B2 u (t)

When N = 0 the variable x2 () is just equal to B2 u(t) at all times.

Otherwise an initial state jump may occur, and this is the reason why we
wrote the left-limit x(0 ) in (3.52). The exponential modes of the regular pair
(E, A) are the nite eigenvalues of sE A, s C, such that det(sE A) = 0.

Denition 3.14. The descriptor system (3.52) is said to be admissible if the

pair (E, A) is regular, impulse-free and has no unstable exponential modes.

Proposition 3.15. [348] The descriptor system (3.52) is admissible and

SSPR (Strongly SPR) if and only if there exists matrices P IRnn and
W IRnm satisfying

E T P = P T E 0, E T W = 0

AT P + P T A AT W + P T B C T (3.56)
<0
(AT W + P T B C T )T W T B + B T W D DT
 3.1 The Positive Real Lemma 93

When E = In then W = 0, P = P T and we are back to the classical KYP

Lemma conditions. In the next theorem PRness is understood as in Denition
2.29.
Theorem 3.16. [157] If the LMI

ET P = P T E 0

AT P + P T A P T B CT (3.57)
0
(P T B C T )T D DT

p P IR i , then the transfer matrix H(s) is PR. Conversely,

nn
has a solution
let H(s) = i= Mi s be the expansion of H(s) about s = , and assume
that D+DT M0 +M0T . Let also the realization of H(s) in (3.52) be minimal.
Then if H(s) is PR there exists a solution P IRnn to the LMI in (3.57).

Minimality means that the dimension n of E and A is as small as possible.

The main dierence between Proposition 3.15 and Theorem 3.16 is that it
is not supposed that the system is impulsive-free in the latter. When the
system is impulse-free, one gets M0 = H() = D C2 B2 , and the condition
D + DT M0 + M0T is not satised unless C2 B2 + (C2 B2 )T 0.

Proof: Let us prove the sucient part of Theorem 3.16. Let s with Re[s] > 0
be any point such that s is not a pole of H(s). The matrix sEA is nonsingular
for such a s. From Proposition A.63 it follows that we can write equivalently
the LMI in (3.57) as
T

A P + P T A = LLT






P T B C = LW
(3.58)



D + D T
W T
W




T
E P = PTE 0
for some matrices L and W . From the rst and last equations of (3.58) it
follows that

(sE A)
P + P T (sE A) = AT P P T A + sE T P + sP T E

= LLT + Re[s](eT P + P T E)
(3.59)
jIm[s](E T P P T E)

= LLT + 2Re[s]E T P
94 3 Kalman-Yakubovich-Popov Lemma

Notice that (sE A)F (s) = B where F (s) = (sE A)1 B. Thus since
H(s) = C(sE A)1 B + D and the second relation in (3.58) one has

H(s) = D + C T F (s)

= D + W T LT F (s) + B T P F (s) (3.60)

= D + W T LT F (s) + F
(s)(sE A)
P F (s)

Using now (3.60) and (3.59) and the third relation in (3.58) we obtain

H(s) + H
(s) = D + DT + W T LT F (s) + F
LW +

+F
(s)[(sE A)
P + P T (sE A)]

W T W + W T LT F (s) + F
(s)LW +

+ F
(s)(LLT + 2Re[s]E T P )F (s)

= (W + LT F (s))
(W + LT F (s)) + 2Re[s]F
(s)(E T P )F (s)

(3.61)
Since (W + LT F (s))
(W + LT F (s) 0 and since Re[s] > 0 and E T P 0,
we have Re[s]F
(s)(E T P )F (s) 0. Thus from (3.61) we obtain

H(s) + H
(s) 0 (3.62)

Recall here that s has been assumed to be any complex with Re[s] > 0
and such that it is not a pole of H(s). Now suppose H(s) has a pole s0 with
Re[s0 ] > 0. Then there exists a pointed neighborhood of s0 that is free of any
pole of H(s) and thus H(s) satises (3.62) in this domain. However this is
impossible if s0 were a pole of H(s). Therefore H(s) does not have any pole
in Re[s] > 0, and (3.62) is true for any s C with Re[s] > 0. Thus H(s) is
PR.

In the proof we used the fact that the pair (E, A) is regular (see Assump-
tion 1) which equivalently means that the matrix sE A is singular for only
nitely many s C.

Example 3.17. [541] Consider

100 1 0 0 1
E = 0 1 0 , A = 0 2 0 , B = 1
000 0 0 1 b (3.63)
1
C = [1 1 1], D = 2
 3.2 Weakly SPR Systems and the KYP Lemma 95

where b is a constant. The pair (E, A) is regular, impulse-free and stable. One
has
1 1 1
H(s) = + b+ (3.64)
s+1 s+2 2
and from
2 4
H(j) + H(j) = + 2b + 1 (3.65)
2 + 1 2 + 4
it follows that H(s) is SSPR when b = 0 and is not SSPR when b = 1.
Another example is treated in Example 4.63. Further results on positive
realness of descriptor systems and applications to control synthesis, can be
found in [157, 260, 348, 541]. The discrete-time case is analyzed in [287, 517].

3.2 Weakly SPR Systems and the KYP Lemma

A dissipative network is composed of resistors, lossy inductors and lossy capac-
itors (see Example 3.90 for the case of nonsmooth circuits with ideal diodes).
Consider the circuit depicted in Figure 3.1 of an ideal capacitor in parallel
with a lossy inductor. Even though this circuit is not only composed of dis-
sipative elements, the energy stored in the network always decreases. This
suggests that the concept of SPR may be unnecessarily restrictive for some
control applications. This motivates the study of weakly SPR systems and its
relationship with the Kalman-Yakubovich-Popov Lemma. The transfer func-
tion of the depicted circuit is LCsLs+R
2 +RCs+1 . It can be checked from Theorem

2.45 that this is not SPR, since r = 1 and lim+ 2 Re[H(j)] = 0. Lozano
and Joshi [310] proposed the following Lemma which establishes equivalent
conditions in the frequency and time domain for a system to be weakly SPR
(WSPR).
Lemma 3.18. [310] [Weakly SPR] Consider the minimal (controllable and
observable) LTI system (3.1) whose transfer function is given by

H(s) = D + C(sIn A)1 B (3.66)

Assume that the system is exponentially stable and minimum-phase. Under
such conditions the following statements are equivalent:
1. P > 0, P IRnn , W IRmm , L IRnm

P A + AT P = LLT
P B C T = LW (3.67)

D + DT = WTW
and such that the quadruplet (A, B, L, W ) is a minimal realization whose
transfer function: H(s) = W + LT (sIn A)1 B has no zeros in the j
axis (i.e. rank H(j) = m, < ).
96 3 Kalman-Yakubovich-Popov Lemma

R
C
L

Fig. 3.1. An ideal capacitor in parallel with a lossy inductor.

2. H(j) + H (j) > 0, IR.

3. The following input-output relationship holds

t
t
u (s)y(s)ds +
t
y T (s)y(s)ds, , t > 0
0 0

with = x(0)T P x(0), P > 0 and y(s) = H(s)u(s).

Proof: (1) (2)

Using (3.66) and (3.67) we obtain

H(j) + H (j)

= D + DT + C(jIn A)1 B + B T (jIn AT )1 C T

= W T W + (B T P + W T LT )(jIn A)1 B

+B T (jIn AT )1 (P B + LW )

= W T W + B T (jIn AT )1 [(jIn AT )P

+P (jIn A)](jIn A)1 B + W T LT (jIn A)1 B

+B T (jIn AT )1 LW
and so
 3.2 Weakly SPR Systems and the KYP Lemma 97

H(j) + H (j)

= W T W + B T (jIn AT )1 LLT (jIn A)1 B

+W T LT (jIn A)1 B + B T (jIn AT )1 LW

= (W + LT (jIn A)1 B)T (W + LT (jIn A)1 B)

It then follows

H(j) + H (j) = H (j)H(j) > 0 (3.68)

Since H(s) has no zeros on the j-axis, H(j) has full rank and, therefore,
the right-hand-side of (3.68) is strictly positive.
(2) (1)
In view of statement 2, there exists an asymptotically stable transfer func-
tion H(s) such that (see [406] or [145])

H(j) + H (j) = H (j)H(j) > 0 (3.69)

Without loss of generality let us assume that

H(s) = W + J(sIn F )1 G (3.70)

with (F, J) observable and the eigenvalues of F satisfying i (F ) < 0 1 i n.

Therefore, there exists P > 0 (see [272]) such that

P F + F T P = JJ T (3.71)
Using (3.70) and (3.71) we have
T
H (j)H(j) = [W + J(jIn F )1 G]T

[W + J(jIn F )1 G]
(3.72)
= W T W + W T J(jIn F )1 G

+GT (jIn F T )1 J T W + X

where
X = GT (jIn F T )1 J T J(jIn F )1 G

= GT (jIn F T )1 [P (F jIn )
(3.73)
+(F T + jIn )P ](jIn F )1 G

= GT (jIn F T )1 P G + GT P (jIn F )1 G
98 3 Kalman-Yakubovich-Popov Lemma

Introducing (3.73) into (3.72) and using (3.69):

T
H (j)H(j) = W T W + (W T J + GT P )(jIn F )1 G

+GT (jIn F T )1 (J T W + P G)

= H(j) + H T (j) (3.74)

= D + DT + C(jIn A)1 B

+B T (jIn AT )1 C T

From (3.74) it follows that W T W = D + DT . Since the eigenvalues of A

and F satisfy i (A) < 0 and i (F ) < 0, then

C(jIn A)1 B = (W T J + GT P )(jIn F )1 G (3.75)

Therefore the various matrices above can be related through a state space
transformation, i.e.

T AT 1 = F
TB = G (3.76)

CT 1 = W T J + GT P
Dening P = T T P T and LT = JT and using ( 3.71) and ( 3.76)

LLT = T T J T JT

= T T (P F + F T P )T

= T T P T T 1F T + T T F T T T T T P T

= P A + AT P
which is the rst equation of (3.67). From (3.76) we get

C = W T JT + GT P T

= W T LT + GT T T T T P T (3.77)

= W T LT + B T P
which is the second equation of (3.67). H(s) was dened by the quadruplet
(F, G, J, W ) in (3.70) which is equivalent, through a state-space transforma-
tion, to the quadruplet (T 1 F T , T 1 G, JT , W ). In view of (3.76) and since
LT = JT , H(s) can also be represented by the quadruplet (A, B, LT , W ) i.e.
 3.2 Weakly SPR Systems and the KYP Lemma 99

H(s) = W + LT (sIn A)1 B (3.78)

We nally note from (3.69) that H(j) has no zeros on the j-axis.

(1) (3)

1 T
Consider the following positive denite function: V (x) = 2 x P x. Then
using (3.67) we obtain

V (x) = 12 xT (P A + AT P )x + xT P Bu

= 21 xT LLT x + uT B T P x

= 21 xT LLT x + uT (C W T LT )x

= 21 xT LLT x + uT y 12 uT (D + DT )u uT W T LT x (3.79)

= 21 xT LLT x + uT y 12 uT W T W u uT W T LT x

1
 T T  T 
= uT y 2 L x + Wu L x + Wu

= uT y 12 y T y

where y is given by
x(t) = Ax(t) + Bu(t)
(3.80)

y(t) = LT x(t) + W u(t)
Therefore, in view of (3.69)

y(s) = H(s)u(s), (3.81)

with H(s) = W + LT (sIn A)1 B. Integrating (3.79) gives

t

1 t T
u (s)y(s)ds +
T
y (s)y(s)ds (3.82)
0 2 0
with = V (x(0)).

(3) (2)
t
Without loss of generality, consider an input u such that 0 uT (s)u(s)ds <
t
+, t 0. Dividing (3.82) by 0 uT (s)u(s)ds, we obtain
t t T
0 uT (s)y(s)ds + V (x(0)) y (s)y(s)ds
t  0t (3.83)
T
u (s)u(s)ds uT (s)u(s)ds
0 0
100 3 Kalman-Yakubovich-Popov Lemma

This inequality should also hold for t = and x(0) = 0, i.e.

 T  T
0 u (s)y(s)ds y (s)y(s)ds

T
 0 T (3.84)
0 u (s)u(s)ds 0 u (s)u(s)ds

Since H(s) and H(s) are asymptotically stable, u L2 y, y L2 and

we can use Plancherels Theorem [421], see also Sections 4.1, 4.2 and 4.3 for
Lp functions and their properties. From the above equation we obtain
R R
U (j)(H(j)+H (j))U(j)d U (j)H (j)H(j)U(j)d
R
U (j)U(j)d
R
U (j)U(j)d

Since H(s) has no zeros on the j-axis, the right-hand-side of the above
equation is strictly positive and so is the left-hand-side for all nonzero U (j)
L2 , and thus
H(j) + H (j) > 0, (, )

3.3 KYP Lemma for Non-minimal Systems

The KYP Lemma as stated above is stated for minimal realizations (A, B, C, D),
i.e. when there is no pole-zero cancellation in the rational matrix C(sIn
A)1 B. However as Example 3.2 proves, non-minimal realizations may also
yield a solvable set of equations (3.2). The KYP Lemma can indeed be stated
for stabilizable systems, or more generally for uncontrollable and/or unobserv-
able systems. This is done in [110, 150, 151, 390, 412, 444, 445]. The motivation
for such an extension stems from the physics, as it is easy to construct systems
(like electrical circuits) which are not controllable or not observable. There
are also topics like adaptive control, in which many poles/zeroes cancellation
occur, so that controllability of the dynamical systems cannot be assumed.
Let us recall a fundamental result. Consider any matrices A, B, C, D of ap-
propriate dimensions. Then the KYP Lemma set of equations (3.2) implies
that

(j) = C(iIn A)1 B B T (j + AT )1 C T + D + DT 0 (3.85)

for all IR, where the spectral density function () was introduced by
Popov, and is named Popovs function, as we already pointed out in Sec-
tion 2.12, Theorem 2.30 and Proposition 2.31. There we saw that one can
characterize a positive operator with the positivity of the associated spec-
tral function. In a word a necessary condition for the solvability of the KYP
Lemma set of equations is that the Popov function satises (3.85). The spec-
tral function satises (s) = (s) with s C. In addition, if the pair
 3.3 KYP Lemma for Non-minimal Systems 101

(A, B) is controllable, then the inequality (3.85) implies the solvability of

the KYP Lemma set of equations, i.e. it is sucient for (3.2) to possess
a solution (P = P T , L, W ). It is worth noting that, under minimality of
(A, B, C, D), that the KYP Lemma set of equations solvability and the posi-
tive realness of H(s) = C(sIn A)1 B + D are equivalent. Let us notice that
(j) = H(j) + H
(j). Let us summarize:

KYP Lemma equations solvability

(if (A, B) controllable) (for all IR | j is not a pole of (s))

(j) 0

 (if A is Hurwitz)

H(s) = C(sIn A)1 B + D is PR

(D = 0)

KYP Lemma equations solvability with P = P T > 0

The rst implication was proved by Kalman [247]. Notice that the second
equivalence is stated under no other assumption that all eigenvalues of A have
negative real parts. In particular no minimality of (A, B, C, D) is required. The
last implication shows that the KYP Lemma solvability is sucient for PRness
of the transfer matrix, without minimality assumption [445] (the proof is led
in [445] with D = 0). It is important to recall that KYP Lemma equations
solvability does not mean that P is positive denite, but only the existence
of a solution (P = P T , L, W ). When P is searched as a non-negative denite
matrix, then we have the following:

KYP Lemma equations solvability with P = P T > 0

 (if (A, B, C, D) minimal)

C(sIn A)1 B is PR

The original result of Popov, building on earlier works of Kalman and

Yakubovich, was as follows:
102 3 Kalman-Yakubovich-Popov Lemma

KYP Lemma equations solvability with P = P T 0

 (if (A, B, C) minimal)

(j) 0 for all IR.

One may have a look at Theorem 3.46 where the link between the Popov
function positivity and the KYP Lemma set of equations solvability is con-
cerned, and a complete proof is provided. In particular it then becomes clear
where the controllability assumption comes into play in this result. However
the controllability assumption is not at all necessary for the KYP Lemma set
of equations to possess a solution. It is therefore of interest to relax as much
as possible this assumption. Perhaps one of the rst, if not the rst, result
relaxing the controllability is due to Meyer [353].

Lemma 3.19 (Meyer-Kalman-Yakubovich Lemma). Given a scalar D

0, vectors B and C, an asymptotically stable matrix A, and a symmetric
positive denite matrix L, if


D 1
Re[H(j)] = Re + C(jIn A) B > 0 for all IR (3.86)
2

then there exists a scalar > 0, a vector q and P = P T > 0 such that
T
A P + P A = qq T L
(3.87)

P B CT = Dq

An application of the MKY Lemma is in Section 8.2.2.

3.3.1 Spectral Factors

The rst set of results that we present rely on the factorization of the Popov
function, and have been derived by Pandol and Ferrante [151, 390]. If (s)
is a rational matrix that is bounded on the imaginary axis and is such that
(j) 0, then there exists a matrix M (s) which is bounded in Re[s] > 0
and such that (j) = M T (j)M (j). The matrix M (s) of a spectral fac-
torization has as many rows as the normal rank of (s). The normal rank of
a polynomial matrix is dened as the rank of (s) considered as a rational
matrix. If (s) Cmm , and if det((s)) is not the zero function (for in-
stance, if the determinant is equal to s 1), (s) is said to have normal rank
m. More generally a polynomial matrix has rank q if q is the largest of the
orders of the minors that are not identically zero [246, 6.3.1].
 3.3 KYP Lemma for Non-minimal Systems 103

Let us consider an eigenvalue s0 of A and a Jordan chain of s0 , i.e. a nite

sequence of vectors satisfying Av0 = s0 v0 , Avi = s0 vi + vi1 , 0 < i r 1,
where r is the length of the Jordan chain. One has


k
ti
eAt v0 = es0 t v0 , eAt vk = es0 t vki (3.88)
i=0
i!
An eigenvalue s0 may have several Jordan chains, in general in nite
number. We suppose these chains have been ordered, and we denote the
ith one as Js0 ,i . The factor M (s) is used together with the Jordan chain
Js0 ,i = (v0 , v1 , ..., vq1 ), to construct the following matrix:

M0 0 0 .... 0


M1 M0 0 ... 0

.
Ms0 ,i =

(3.89)

.

.
Mr1 Mr2 Mr3 ... M0
One has

1 dh T 1 dh T
Mh = M (s0 ) = M (s) (3.90)
h! dsh h! dsh s0

In other words h!Mh is the hth derivative of the function M T (s) calcu-
lated at s = s0 . All the matrices Ms0 ,i as well as the rational functions (s)
and M (s) are calculable from A, B, C and D. The notation col[a0 , a1 , ..., an ]
is for the column matrix [a0 a1 ... an ]T .
Theorem 3.20. [390] Let the matrices Ms0 ,i be constructed from any spectral
factor of (s) and assume that every eigenvalue of A has a negative real part.
If the transfer function H(s) is positive real, then there exist matrices L, W
and P = P T 0 which solve the KYP Lemma set of equations (3.2), if and
only if the following conditions hold for every Jordan chain Js0 ,i of the matrix
A:

col[C T v0 , C T v1 , ...., C T vr1 ] Im(Ms0 ,i ) (3.91)

For the proof (that is inspired from [32]) the reader is referred to the paper
[390]. It is noteworthy that there is no minimality assumption in Theorem
3.20. However P is only semi-positive denite.
Example 3.21. [390] Let C = 0, B = 0, D = 0. Then (s) = 0 and the set of
equations AT P + P A = LLT , P B = C T LW is solvable. One solution is
L = 0, P = 0. This proves that Theorem 3.20 does not guarantee P > 0.
104 3 Kalman-Yakubovich-Popov Lemma

A second Theorem relaxes the Hurwitz condition on A.

Theorem 3.22. [151] Let A IRnn , B IRnm , C IRmn , and D

IRmm . Assume that (A) (AT ) = . If the KYP Lemma set of equations
(3.2) is solvable, i.e. there exist matrices P = P T , L, W which solve it, then
(j) 0 for each and the condition (3.91) holds for every Jordan chain
Js0 ,i of the matrix A. Conversely, let (j) be nonnegative for each and
let (3.91) hold for every Jordan chain of A. Then the set of equations (3.2)
is solvable. Condition (3.91) does not depend on the specic spectral factor
M (s) of (s).

A matrix A satisfying (A) (AT ) = is said unmixed.

Remark 3.23. Until now we have spoken only on controllability, and not of
observability. Thus one might think that the unobservable part has no inu-
ence neither on (3.85) nor on the solvability of (3.2). Things are more subtle
as shown in the next subsection.

3.3.2 Sign-controllability

I0
To start with, let us consider the following system [150]: A = ,C =
0I

0
(I 0), B = , D = 0. Then the KYP Lemma set of equations in (3.2) has
I
innitely many solutions, which can be parametrized as triples
  
P1 I Q1
, ,0
I 0 0
with P1 0, and Q1 QT1 = 2P1 . However the system of equations obtained
by eliminating the unobservable subspace associated to (A, C) has no solution,
because the second equation for this reduced system takes the form 0 = I
0. This example shows that unobservability is not so innocent in the KYP
Lemma solvability (which is to be understood here as the existence of a triple
(P = P T , L, W ) that solves (3.2)).

The sign-controllability of a pair of matrices is dened in Appendix A.4.

Let us assume that (A, B) is sign controllable. Then there exists a feedback
u(t) = Kx(t) + v(t) such that the new transition matrix A + BK is unmixed.
One can start from a system such that A is unmixed.
Before stating the next Lemma, let us perform a state space transforma-
tion. We assume
that(A, C) is not observable. The Kalman observability form
A1 0
reads A = , C = (C1 0). Let us dene
A21 A2
= [(A2 ) (AT1 )] [(A2 ) (AT2 ) (A1 )],
and select a basis such that
 3.3 KYP Lemma for Non-minimal Systems 105


A2 0
A2 =
0 A2
with ( = , (A2 ) = . Then A2 may be partioned conformably as
A2 ) 
A21
A2 = . Then A and C may be partitioned as
A21

A1 0
A=
A21 A2

C = (C1 0)
with

A1 0
A1 =
A21 A2

A21 = (A21 0)

C1 = (C1 0)
One may check that (A2 )(AT1 ) = . The matrix  B can be partitoned
B1
conformably with the partitioning of A as B = . The image space of
B2
the matrix (0 I), where the identity matrix I has the size of A2 , is unob-
servable for the pair (A, C) and is the largest unobservable subspace such
that the corresponding dynamics does not intersect the backwards dynamics
of the remainning part, i.e. (A2 ) (AT1 ) = . This space is named the
unmixing unobservable subspace. The system (A1 , B1 , C1 , D) obtained from
(A, B, C, D) by eliminating the part corresponding to the unmixing unobserv-
able subspace, is called the mixed+observable subsystem. When A is unmixed,
the mixed+observable subsystem is exactly the observable subsystem. In such
a case the unobservable part of the system plays no role in the solvability of
the KYP Lemma set of equations (3.2).

Theorem 3.24. [150] Given a quadruple (A, B, C, D), let A be unmixed and
(A1 , B1 , C1 , D) be the matrices associated to the observable subsystem. Then
the KYP Lemma set of equations (3.2) possesses solutions (P = P T , L, W ) if
and only if the set of equations
T

A P1 + P1 A1 = L1 LT1
1


P1 B1 = C1T L1 W1 (3.92)




T
W1 W1 = D + DT
possesses solutions (P1 = P1T , L1 , W1 ).
106 3 Kalman-Yakubovich-Popov Lemma

Once again we insist on the fact that it is not required here that P nor
P1 be positive denite or even semi positive denite matrices. The result of
Theorem 3.24 relies on the unmixity of A. However the following is true, which
does not need this assumption.

Theorem 3.25. [150] The KYP Lemma set of equations (3.2) possesses so-
lutions if and only if (3.92) possesses solutions.

3.3.3 State Space Decomposition

The result presented in this subsection also relies on a decomposition of the

state space into uncontrollable and unobservable subspaces. It was proposed
in [444]. Let us start from a system (A, B, C). The Kalman controllability
and observability matrices are denoted as Kc and Ko , respectively. The state
space of the linear invariant system (A, B, C) is given by the direct sum

X = X1 X2 X3 X4
where sp(Kc ) = X1 X2 , sp(Kc ) Ker(Ko ) = X1 , Ker(Ko ) = X1 X3 . The
notation sp(A) means the algebraic span of the column vectors of A. Then
the following holds.

Theorem 3.26. [444] Let (A, B, C) be a realization of the rational matrix

H(s). Let K be any matrix satisfying

X1 X2 sp(K)X1 X2 X3
Then H(s) is positive real if and only if there exist real matrices P = P T 0
and L such that
T
K (P A + AT P + LLT )K = 0
(3.93)
T
K (P B C T ) = 0
If B has full column rank, then H(s) is positive real if and only if there
exist real matrices P = P T and L, with K T P K 0, such that
T
K (P A + AT P + LLT )K = 0
(3.94)

P B CT = 0
 3.3 KYP Lemma for Non-minimal Systems 107

3.3.4 A Relaxed KYP Lemma for SPR Functions with Stabilizable

Realization

The next result is taken from [110]. Let us consider the system in (3.1)
and suppose (A, B, C, D) is a minimal realization, m n. Suppose that
H(s) + H T (s) has rank m almost everywhere in the complex plane, i.e. it
has normal rank m (this avoids redundant inputs and outputs). The following
Lemma gives us a general procedure to generate uncontrollable equivalent re-
alizations from two minimal realizations of a given transfer matrix H(s). The
uncontrollable modes should be similar and the augmented matrices should
be related by a change of coordinates as explained next.
Lemma 3.27. [110] Let (Ai , Bi , Ci , Di ), i = 1, 2 be two minimal realizations
of H(s), i.e. H(s) = Ci (sIn Ai )1 Bi + Di for i = 1, 2. Now dene the
augmented systems

Ai 0 Bi
Ai = Bi =
0 A0i 0 (3.95)

Ci = (Ci C0i ) Di = Di
where the dimensions of A01 and A02 are the same. Moreover there exists a
nonsingular matrix T0 such that A01 = T0 A02 T01 and C01 = C02 T01 . Then
(Ai , Bi , Ci , Di ), i = 1, 2 are two equivalent realizations.
As a dual result we can generate unobservable augmented realizations of
H(s) as established in the following Corollary.
Corollary 3.28. Let i (Ai , Bi , Ci , Di ) for i = 1, 2 be two minimal realiza-
tions of Z(s), i.e. Z(s) = Ci (sI Ai )1 Bi + Di for i = 1, 2. Now dene the
augmented systems:
 
Ai 0 Bi
Ai = Bi =
0 A0i Bi0
(3.96)
 
C i = Ci 0 D i = Di
where the dimensions of A01 and A02 are the same. Moreover, there exists a
1
nonsingular
 matrix  T0 such that A01 = T0 A02 T0 and B01 = T0 B02 . Then
i Ai , B i , C i , Di for i = 1, 2 are two equivalent realizations of H(s).

Theorem 3.29. [110] Let H(s) = C(sIn A)1 B + D be an m m transfer

matrix such that H(s) + H T (s) has normal rank m, where A is Hurwitz,
(A, B) is stabilizable, (A, C) is observable. Assume that if there are multiple
eigenvalues, then all of them are controllable modes or all of them are uncon-
trollable modes. Then H(s) is SPR if and only if there exist P = P T > 0, W ,
L and a constant > 0 such that
108 3 Kalman-Yakubovich-Popov Lemma

P A + AT P = LT L P



P B = C T LT W (3.97)




T
W W = D + DT

This theorem is interesting since it states the existence of a positive denite

solution to the KYP Lemma set of equations, and not only its solvability
with P = P T or P = P T 0. The assumption that H(s) + H T (s) has
normal rank m is in order to avoid redundances in inputs and/or outputs.
The assumption that the intersection of the set of controllable modes with
the set of uncontrollable modes is empty, is used only in the necessary part of
the proof.
Proof: Suciency: Let (0, /2) then from (3.97)

P (A + In ) + (A + In )T P = LT L ( 2)P (3.98)
which implies that (A + In ) is Hurwitz and thus Z(s ) is analytic in
Re[s] 0. Dene now for simplicity

(s) = (sIn A)1

Therefore:
T T T T
H(s ) + H T (s ) = D + D + C(s )B + B (s )C
T T T  
= W T W + B P + W T L (s )B + B (s ) P B + LT W
T T
= W T W + W T L(s )B + B (s )LT W +
T T T
+B P (s )B + B (s )P B
T T
= W T W + W T L(s )B + B (s )LT W +
T T T 1
B (s ) (s )P + P (s ) (s )B
T T T T
 )B + B (s 
3 )L W + B (s )
T T T
=
2W W + W L(s T
(s + )I A P + P (s )I A (s )B
T T
= WT W + WT2L(s )B + B (s
3 )L W +
T
T T T
B (s ) 2P A P P A (s )B
T T
= W T W + W T L(s )B + B (s )LT W +
T T ) *
B (s ) LT L + ( 2) P (s )B
T T
= W T W + W T L(s )B + B (s )LT W +
T T T T
B  (s )LT L(s )B + ( 2) B (s )P (s )B
T T 
= W + B (s )L
T T
W + L(s )B +
T T
( 2) B (s )P (s )B
 3.3 KYP Lemma for Non-minimal Systems 109

From the above it follows that H(j ) + H T (j ) 0,

[, +] and H(s) is SPR.
 
Necessity: Assume that H(s) SP R. Let A, B, C, D be a stabilizable
and observable realization of H(s) and (A, B, C, D) a minimal realization
of H(s). Given that the controllable and uncontrollable modes are dierent
we can consider that the matrix A is block diagonal and therefore H(s) can
be written as
 1 
  sIn A 0 B
H(s) = C C0 + 
D (3.99)
  0 sI A0 0
    D
C
[sIn A] B

where the eigenvalues of A0 correspond to the uncontrollable modes. As stated

in the preliminaries, the condition (A) (A0 ) = ( where (T ) means the
spectrum of the square matrix T ) means that the (C, A) and (C0 , A0 )
 pairs 
  A 0
are observable if and only if (C, A) = C C0 , is observable.
0 A0
 
We have to prove that A, B, C, D satises the KYP equations (3.97).
Note that A, A0 are both Hurwitz. Indeed A is stable because (A, B, C, D)
is a minimal realization of H(s) which is SPR. A0 is stable because the system
is stabilizable. Thus there exists > 0 such that H(s ) P R and H(s
) P R for all [0, ] . Choose now > 0 suciently small such that

U (s) = Z(s 2 ) SP R. Then the following matrices are Hurwitz:

A = A + 2 I R(n+n0 )(n+n0 )
A = A + 2 I Rnn (3.100)
A0 = A0 + 2 I Rn0 n0 .
Note that A is also block diagonal having block elements A and A0 and
the eigenvalues of A and A0 are dierent. Let  (A , B, C, D) be a mini-
mal realization of U (s) and  A , B, C, D an observable and stabilizable
realization of U (s). Therefore

U (s) = C(sIn A )1 B + D = C(sIn A )1 B + D (3.101)

Note that the controllability of the pair (A , B) follows
 from
 the control-
lability of (A, B) . Since A0 is Hurwitz, it follows that A , B is stabilizable.
From the spectral factorization Lemma for SPR transfer matrices [527], [256,
Lemma A.11, pp. 691], or [11], there exists an m m stable transfer matrix
V (s) such that

U (s) + U T (s) = V T (s)V (s) (3.102)

110 3 Kalman-Yakubovich-Popov Lemma

Remark 3.30. Here is used implicitly the assumption that Z(s) + Z T (s) has
normal rank m, otherwise the matrix V (s) would be of dimensions (r m),
where r is the normal rank of Z(s) + Z T (s).

Let V (F, G, H, J) be a minimal realization of V (s), F is Hurwitz because 

V (s) is stable; a minimal realization of V T (s) is V T F T , H T , GT , J T .
Now the series connection V T (s)V (s) has realization (see [257, p. 15] for the
formula of a cascade interconnection)

   
F 0 G  T 
V T (s)V (s) , , J H G , J J
T T
(3.103)
H T H F T HT J

Although we will not require the minimality of V T (s)V (s) in the se-
quel, it can be proved to follow from the minimality of V (F, G, H, J) ,
see [11, 256]. Let us now dene a nonminimal realization of V (s) obtained
from V (F, G, H, J) as follows:
 
F 0 G
F = , G=
0 F0 0
(3.104)
 
H = H H0 , J = J
and such that F0 is similar to A0 and the pair (H0 , F0 ) is observable, i.e.
T0 nonsingular such that

F0 = T0 A0 T01 (3.105)

This constraint will be claried later on. Since (F0 ) (F ) = then the
pair  
  F 0
(H, F ) = H H0 , (3.106)
0 F0
 
is observable. Thus the nonminimal realization V F , G, H, J of V (s) is
T
observable and  stabilizable.
 Now a nonminimal realization of V (s)V (s)
based on V F.4 , G, H, J 5 4 5 /
F 0 G    T 
T T
V T (s)V (s) T T , T , J H G , J J (3.107)
H H F H J
is (see [257, p. 15])

F 0 0 0 G
0 F0 0 0 0

HT H H T H0 F T 0 HT J
V T (s)V (s) =


(3.108)
H0T H H0T H0 0 F0T H0T J

J T H J T H0 GT 0 JT J
 3.3 KYP Lemma for Non-minimal Systems 111

From the diagonal structure of the above realization, it could be concluded

that the eigenvalues of F0 correspond to uncontrollable modes and the eigen-
values of (F0T ) correspond to a unobservable modes. A constructive proof is
given below.
Since the pair (H, F ) is observable and F is stable, there exists a positive
denite matrix 
T K r
K=K = T >0 (3.109)
r K0
solution of the Lyapunov equation
T T
K F + F K = H H (3.110)
This explains why we imposed the constraint that (H0 , F0 ) should be ob-
servable. Otherwise there
will not
 exist a positive denite solution for (3.110).
I 0 1 I 0
Dene T := ;T = and use it as a change of coordinates
KI K I
for the nonminimal realization V T (s)V (s) above to obtain

F 0 0 0
0 F0 G
0 0
0 0
V T (s)V (s) = 0 F T
0 T (3.111)
0 0 0 F0 T (JH + G K) T


T
JH + G K GT 0 JT J
Now it is clear that the eigenvalues of F0 correspond to uncontrollable
modes and the eigenvalues of (F0T ) correspond to unobservable
 modes.

From (3.101) a nonminimal realization of  U (s) is  A , B,  D . Thus
C,
T T T T
a nonminimal realization for U T (s) is  A , C , B , D . Using the
T
results in the preliminaries, a nonminimal realization of U (s) + U (s) is
.4 5 4 5 /
A 0 B    T

T
U(s)+U T (s) T , T , C B , D + D . (3.112)
0 A C
Using (3.102) we conclude that the stable (unstable) parts of the realiza-
tions of U (s) + U T (s) and V T (s)V (s) are identical. Therefore, in view of
the block diagonal structure of the system and considering only the stable
part we have
 
F 0 A 0
F = = RA R1 = R R1
0
 F0  0 A 0
G B
G= = RB = R
0 0
(3.113)
T 1
  1
JH + G K = CR = C C0 R

T
JT J = D + D
112 3 Kalman-Yakubovich-Popov Lemma

The above relationships impose that the uncontrollable parts of the real-
izations of U (s) and V (s) should be similar. This is why we imposed that F0
is similar to A0 in the construction of the nonminimal realization of V (s).
From the Lyapunov equation (3.110) and using F = RA R1 in (3.113),
we get
T T

KF + F K = H H

T T
KRA R1 + RT A RT K = H H
(3.114)


T
RT KRA + A RT KR =
T
RT H HR

T
P A + A P = LT L

where we have used the denitions P = RT KR; L = HR. Introducing (3.100)
we get the rst equation of (3.97). From the second equation of (3.113) we
have G = RB. From the third equation in (3.113) and using W = J we get
T

JH + G K = CR1

T T
J HR + G RT RT KR =C
(3.115)


T
WTL + B P =C

T
PB = C LT W
which is the second equation of (3.97). Finally from the last equation of
(3.113), we get the last equation of (3.97) because W = J.
s+a
Example 3.31. Consider H(s) = (s+a)(s+b) , for some a > 0, b > 0, b = a. Let
a nonminimal realization of H(s) be


a 0 0


x(t) = x(t) + u(t)
1
0 b (3.116)






y(t) = [ ]x(t)
with = 0 and = 0. For all < min(a, b) one has

(a+b)2 2

(2b)(2a)
P = >0
2

 
for all a > 0, b > 0, = 0, = 0. The matrix L = a+b

2b
2b and
P satisfy the KYP Lemma set of equations .

Remark 3.32. Proposition 2.31 states that positivity of an operator is equiv-

alent to the positivity of its Popovs function. There is no mention of sta-
bility. This is in accordance with the above versions of the KYP Lemma for
which the stability (i.e. solvability of the KYP Lemma set of equations with
 3.5 The Feedback KYP Lemma 113

P = P T > 0) requires more than the mere positivity of the spectral function.

3.4 SPR Problem with Observers

The KYP Lemma for non-controllable systems is especially important for the
design of feedback controllers with state observers [111, 241, 242], where the
closed-loop system may not be controllable. This may be seen as the extension
of the works described in Section 2.15.3 in the case where an observer is added
to guarantee that the closed-loop is SPR.

Theorem 3.33. [111] Consider a system with stable transfer function H(s)
Cmm , and its state space realization

x(t) = Ax(t) + Bu(t)
(3.117)

y(t) = Cx(t)
where (A, B) is stabilizable and (A, C) is observable. Then there exists a gain
observer L and an observer


x(t) = Ax(t) + Bu(t) + LC(x(t) x(t))
(3.118)

z(t) = M x(t)
such that (A LC) is in the open left-hand complex plane,and the transfer
x
function between u() and the new output z(t) = M0 = M x(t), with
x x
M = B T P , is characterized by a state space realization (A0 , B0 , M0 ) that is
SPR, where
 
A 0 B
A0 = , B0 =
0 A LC 0

The modes associated to the matrix (A LC) are non-controllable.

3.5 The Feedback KYP Lemma

The feedback KYP Lemma is an extension of the KYP Lemma, when one
considers a controller of the form u(t) = Kx(t). This is quite related to the
material of Section 2.15.3: which are the conditions under which a system can
be made passive (or PR) in closed-loop? Let us consider the system
114 3 Kalman-Yakubovich-Popov Lemma

x(t) = Ax(t) + Bu(t)
(3.119)

y(t) = Cx(t)
with the usual dimensions and where all matrices are real.

Denition 3.34. The system in (3.119) is said to be

A In B
Minimum phase if the polynomial det is Hurwitz (it has all
C 0
its zeroes in the open half plane)
Strictly minimum phase if it is minimum phase and the matrix CB is
nonsingular
Hyper minimum phase if it is minimum phase and the matrix CB is pos-
itive denite

The next Theorem is close to what is sometimes referred to as Fradkovs

Theorem [56].

Theorem 3.35. [16, 154156] Let rank(B) = m. Let Q = QT 0. Then

(A) There exists P = P T > 0 and K such that P (A+BK)+(A+BK)T P <
Q and P B = C T
if and only if
(B) the system in (3.119) is hyper minimum phase
if and only if
(C) there exists P = P T > 0 and K such that P (A + B KC) + (A +
B KC)T P < Q and P B = C T
if and only if
(D) the matrix CB is symmetric positive denite and the zero dynamics
of the system in (3.119) is asymptotically stable
Moreover the matrix K can be chosen as K = C where > 0 is large
enough. Assume that in addition Ker(C) Ker(Q). Then
(E) There exists P = P T > 0 and K such that A + BK is Hurwitz and
P (A + BK) + (A + BK)T P < Q and P B = C T
if and only if
 3.6 Time-varying Systems 115

(F) the matrix CB is symmetric positive denite,

the pair
(A, B) is stabi-
A In B
lizable, all the zeroes of the polynomial det are in the closed
C 0
left half
plane,
and allthe pure
imaginary eigenvalues of the matrix pencil
AB In 0
R() = have only linear elementary divisors j
C 0 0 0
if and only if
(G) the matrix CB is symmetric positive denite, the pair (A, B) is sta-
bilizable and the system (3.119) is weakly minimum phase.

Both matrix equations in (A) and (C) are bilinear matrix inequali-
ties (BMIs). The feedback KYP Lemma extends to systems with a direct
feedthrough term y = Cx + Du. It is noteworthy that Theorem 3.35 holds
for multivariable systems. If u(t) = Kx(t) + v(t), then (A) means that the
operator v
y is SPR. It is known that this control problem is dual to the
SPR observer design problem [22]. Related results are in [23]. We recall that a
system is said weakly minimum phase if its zero dynamics is Lyapunov stable.
The zero dynamics can be explicitly written when the system is written in a
special coordinate basis as described in [432434]. The particular choice for K
after item (D) means that the system can be stabilized by output feedback.
More work may be found in [153]. The stability analysis of dynamic output
feedback systems with a special formulation of the KYP Lemma has been
carried out in [241].

3.6 Time-varying Systems

Let us consider the linear system:

x(t) = A(t)x(t) + B(t)u(t)



y(t) = C(t)x(t) + D(t)u(t) (3.120)





x(t0 ) = x0
where the functions A(), B(), C(), D() are supposed to be piecewise con-
tinuous, and D(t) Im , 0. It is assumed that all (t, x) with t > t0 are
reachable from (t0 , 0), and that the system is zero state observable (such con-
trollability and observability conditions may be checked via the controllability
and observability grammians, see e.g. [467]). It is further assumed that the re-
quired supply is continuously dierentiable in both t and x, whenever it exists
(the required supply is a quantity that will be dened in Denition 4.36. The
116 3 Kalman-Yakubovich-Popov Lemma

reader may just want to consider this as a regularity condition on the system
(3.120)). The system (3.120) is supposed to be well-posed; see Theorem 3.55,
and it denes an operator : u(t)
y(t). The kernel of () is given by
K(t, r) = C(t)(t, r)B(r)1(t r) + B T (t)T (r, t)C T (tr)1(r t) + R(t)tr ,
where 1(t) = 0 if t < 0, 1(t) = 12 if t = 0 and 1(t) = 1 if t > 0,
R(t) = D(t) + DT (t), t is the Dirac measure at t, (, ) is the transition
matrix of A(t), i.e. (t, r) = X(t)X 1 (r) for all t and r, and dX
dt = A(t)X(t).
t
Then (u(t)) = K(t, r)u(r)dr.

Lemma 3.36. The operator () is nonnegative if and only if there exists an

almost everywhere continuously dierentiable function P () = P T () 0 such
that on (t0 , t)

Q S
0 (3.121)
T
S R
where

P (t) + AT (t)P (t) + P (t)A(t) = Q(t)
(3.122)

C (t) P (t)B(t) = S(t)
T

Nonnegativity of () is understood as in Proposition 2.31.

3.7 Interconnection of PR Systems

We will now study the stability properties of positive real or strictly positive
real systems when they are connected in negative feedback. We will consider
two PR systems H1 : u1 y1 and H2 : u2 y2 . H1 is in the feedforward path
and H2 is in the feedback path(i.e. u1 = y2 and u2 = y1 ). The stability of
the closed loop system is concluded in the following Lemma when H1 is PR
and H2 is weakly SPR.

Lemma 3.37. Consider a system H1 : u1 y1 in negative feedback with a

system H2 : u2 y2 as shown in Figure 3.2, where H1 is PR and H2 is
WSPR. Under those conditions u1 , u2 , y1 and y2 all converge to zero exponen-
tially.

Proof: Let us dene the following state-space representation for system H1 (s)
 3.7 Interconnection of PR Systems 117

u1 y1
H1
-

y2 u2
H2
Fig. 3.2. Interconnection of H1 and H2

x1 (t) = A1 x1 (t) + B1 u1 (t)
(3.123)

y1 (t) = C1 x1 (t) + D1 u1 (t)
Since H1 (s) is PR there exists matrices P > 0, P IRnn , W
IRmm
, L IRnm such that

P1 A1 + AT1 P1 = L1 LT1
P1 B1 C1T = L1 W1 (3.124)

D1 + D1T = W1T W1
Dene the following state-space representation for the system H2 (s)

x2 (t) = A2 x2 (t) + B2 u2 (t)
(3.125)

y2 (t) = C2 x2 (t) + D2 u2 (t)
Since H2 (s) is WSPR there exists matrices P > 0, IRnn , W
IRmm
, L IRnm such that

P2 A2 + AT2 P2 = L2 LT2
P2 B2 C2T = L2 W2 (3.126)

D2 + D2T = W2T W2
and
H 2 (s) = W2 + LT2 (sIn A2 )1 B2 (3.127)
has no zeros in the j-axis. Consider the following positive denite function

Vi (xi ) = xTi Pi xi , i = 1, 2.
118 3 Kalman-Yakubovich-Popov Lemma

Then using (3.124) and (3.126):

Vi (xi ) = (xTi ATi + uTi BiT )Pi xi + xTi Pi (Ai xi + Bi ui )

= xTi (ATi Pi + Pi Ai )xi + 2uTi BiT Pi xi

= xTi (Li LTi )xi + 2uTi BiT Pi xi

= xTi Li LTi xi + 2uTi (BiT Pi + WiT LTi )xi 2uTi WiT LTi xi

= xTi Li LTi xi + 2uTi [Ci xi + Di ui ] 2uTi Di ui 2uTi WiT LTi xi

= xTi Li LTi xi + 2uTi yi 2uTi Di ui 2uTi WiT LTi xi

(LTi xi + Wi ui )T (LTi xi + Wi ui ) + 2uTi yi

(3.128)
where we have used the fact that
2uTi Di ui = uTi (Di + DiT )ui = uTi WiT Wi ui
Dene yi = LTi xi + Wi ui and V (x) = V1 (x1 ) + V2 (x2 ), then
V (x1 , x2 ) = y1T y1 y2T y2 + 2(uT1 y1 + uT2 y2 )
Since u1 = y2 and u2 = y1 it follows that
uT1 y1 + uT2 y2 = y2T y1 + y1T y2 = 0
Therefore

V (x1 , x2 ) = y1T y1 y2T y2 y2T y2 ,

which implies that V () is a nondecreasing function and therefore we conclude
that xi L . Integrating the above equation:

t
V (0) V (t) V (0) y2T (s)y(s)2 ds (3.129)
0

Then
t
y2T (s)y2 (s)ds V (0) (3.130)
0
The feedback interconnection of H1 and H2 is a linear system. Since xi
L , the closed loop is at least stable, i.e. the closed-loop poles are in the left-
half plane or in the jw-axis. This means that ui , yi may have an oscillatory
behavior. However the equation above means that y 2 0. By assumption
H2 (s) has no zeros on the j axis. Since the state is bounded, u2 () can not
grown unbounded. It follows that u2 (t) 0 as t +. This in turn implies
that y2 (t) 0 since H2 is asymptotically stable. Clearly u2 (t) 0 and
y2 (t) 0 as t +.
 3.8 Positive Realness and Optimal Control 119

3.8 Positive Realness and Optimal Control

The material of this section is taken from [513, 514]. As we have already
pointed out in Section 3.1.2, strong links exist between dissipativity and opti-
mal control. In this section more details are provided. Close results were also
obtained by Yakubovich [349, 520, 523].

3.8.1 General Considerations

Let us start with some general considerations which involve some notions
which have not yet been introduced in this book, but will be introduced in
the next chapter (actually, the only missing denitions are those of a storage
function and a supply rate: the reader may thus skip this part and come back
to it after having read Chapter 4). The notions of dissipation inequality and
of a storage function have been introduced (without naming them) in (2.3),
where the function V () is a so-called storage function and is a function of
the state x() (and is not an explicit function of time). Let us consider the
following minimization problem

+

Vf (x0 ) = min w(u(s), x(s))ds (3.131)
uL2,e 0

with

w(u, x) = uT Ru + 2uT Cx + xT Qx (3.132)

with R = RT , Q = QT , subject to x(t) = Ax(t) + Bu(t), x(0) = x0 . It is

noteworthy that Vf (x0 ) is nothing else but the value function of the principle
of optimality. The set L2,e is the extended set of L2 -bounded functions; see
Section 4.3.5. If w(u, x) 0 for all x IRn and all u IRm then the value
function satises

t1
Vf (x(0)) Vf (x(t1 )) + w(u(t), x(t))dt (3.133)
0
for all t1 0, or, if it is dierentiable, the innitesimal equivalent

Vf
(x)[f (x) + g(x)u] + w(u, x) 0, x IRn , u IRm . (3.134)
x
One realizes immediately by rewriting (3.133) as the dissipation inequality

t1
Vf (x(0)) Vf (x(t1 )) w(u(t), x(t))dt (3.135)
0
120 3 Kalman-Yakubovich-Popov Lemma

that Vf () plays the role of a storage function with respect to the supply
rate w(u, x). Let us end this subsection making a small digression on the
following well-known fact: why is the optimal function in (3.131) a function
of the initial state? To see this intuitively, let us consider the minimization
problem

+
inf (u2 (t) + x2 (t))dt (3.136)
uU 0

subject to x(t) = u(t), x(0) = x0 . Let U consist of smooth functions. Then

niteness of the integral in (3.136) implies that limt+ x(t) = 0. Take any
constant a IR. Then
 +  +
0 2ax(t)u(t)dt = 0 2ax(t)x(t)dt =
(3.137)
 + d 2 2 + 2
= 0 dt [ax (t)]dt = [ax (t)]0 = ax0 .
 + 2
So indeed inf uU 0 (u (t) + x2 (t))dt is a function of the initial state.

3.8.2 Least Squares Optimal Control

We have already pointed out the relationship which exists between the linear
matrix inequality in the KYP Lemma (see Section 3.1.2) and optimal control,
through the construction of a Riccati inequality that is equivalent to the
linear matrix inequality (LMI) in (3.3). This section is devoted to deepen
such relationships. First of all, let us introduce (or re-introduce) the following
algebraic tools:
The linear matrix inequality (LMI)

GA + AT G + Q GB + C T
0 (3.138)
BT G + C R
The quadratic matrix inequality (QMI) or algebraic Riccati inequality
(ARI)

GA + AT G (GB + C T )R1 (B T G + C) + Q 0 (3.139)

The algebraic Riccati equation (ARE)

GA + AT G (GB + C T )R1 (B T G + C) + Q = 0 (3.140)

The frequency-domain inequality (FDI)

H(s, s) = R + C(sIn A)1 B + B T (sIn AT )1 C T +

(3.141)
+B T (sIn AT )1 Q(sIn A)1 B 0
 3.8 Positive Realness and Optimal Control 121

where s C and s is its complex conjugate. Notice that H(s, s) can be

rewritten as

T  
(sIn A)1 B Q CT (sIn A)1 B
H(s, s) = (3.142)
Im C R Im

Remark 3.38. Comparing (3.138) and (3.17) it is expected that G 0 in

(3.138), and in (3.139) and (3.140) as well.
Let R = D + DT . The function H(s, s) in (3.141) is also known as the
Popov function (s), and was formally introduced by Popov in [410] (the
rst time it has been introduced may even be in [86]). It is worth noting that
when Q = 0 then H(s, s) = H(s) + H T (s) where H(s) = C(sIn A)1 B + D.
Thus H(j, j) 0 a the condition for the PRness of H(s). By extension
one may also call the function in (3.141) a Popov function [391]. Notice that
H(s, s) in (3.142) is linked to the system x(t) = Ax(t) + Bu(t) as follows. For
every u Cm and every IR such that j is not an eigenvalue of A, we
have

T  
x(j, u) Q CT x(j, u)
uT H(j, j)u = (3.143)
u C R u

where x(j, u) is dened from jx = Ax+Bu, i.e. x(j, u) = (jIn A)1 Bu.
See for instance Theorem 3.46 for more information on the spectral function
and its link with the KYP Lemma
set
 of equations. One sometimes calls any
Q CT
triple of matrices A, B and a Popov triple.
C R

Remark 3.39. In the scalar case the ARE (3.140) becomes a second order
equation aG2 + bG + c = 0 with real coecients. It is clear that without
assumptions on a, b, and c there may be no real solutions. Theorem A.53 in
Appendix A.4 states conditions under which an ARE as in (3.140) possesses
a real solution.

We will denote the inequality in (3.133) as the DIE (for dissipation inequal-
ity), keeping in mind that the real dissipation inequality is in (3.135). Let us
introduce the following optimal control problems, with w(x, u) in (3.132).

+
+
V (x0 ) = min w(u(s), x(s))ds, lim x(t) = 0 (3.144)
uL2,e 0 t+

+

V (x0 ) = min w(u(s), x(s))ds, lim x(t) = 0 (3.145)
uL2,e 0 t+
122 3 Kalman-Yakubovich-Popov Lemma

t

Vn (x0 ) = min w(u(s), x(s))ds (3.146)
uL2,e ,t0 0

These four problems (i.e. (3.131), (3.144), (3.145) and (3.146)) are subject
to the dynamics x(t) = Ax(t) + Bu(t), with initial data x(0) = x0 .
Assumption 2 We assume that the pair (A, B) is controllable throughout
Section 3.8.2.
Therefore this assumption will not be repeated. One notes that the four
functions in (3.131), (3.144), (3.145) and (3.146) are quadratic functions of
the state x0 . Let us summarize few facts:
Vn () 0 (take t = 0 in (3.146) to deduce that the minimum cannot be
positive).
t
Vn () Vf () V + (): indeed, if the scalar 0 w(u(s), x(s))ds sweeps a
 +
certain domain in IR while t 0, then the scalar 0 w(u(s), x(s))ds must
 +
belong to this domain. And similarly if the scalar 0 w(u(s), x(s))ds
 +
sweeps a certain domain while u L2,e , the scalar 0 w(u(s), x(s))ds
subject to the limit condition must lie inside this domain.
Vn () < +,Vf () < +, V + () < +: by controllability the integrand
w(u, x) is bounded whatever the nal (bounded) state, so the lowerbound
is bounded.
V () > : note that

+
+
min w(u(s), x(s))ds = max w(u(s), x(s))ds.
uL2,e 0 uL2,e 0

By controllability one can surely nd a control u that drives the system

 +
from x0 to some other state, and such that the scalar 0 w(u(s), x(s))ds
is bounded. So the supremum surely cannot be .
Vf (), V + (), and Vn () satisfy the DIE (3.133). By direct inspection
t
Vf (x0 )Vf (x1 ) = t01 w(u(s), x(s))ds and similarly for the other two func-
tions.
If for all x IRn there exists a u IRm such that w(x, u) 0 then
Vn () = Vf (). A sucient condition for this is that R > 0, or that Q = 0.
If there exists a feedback controller u(x) such that w(u(x), x) 0 and
such that x(t) = Ax(t) + Bu(x(t)) has an asymptotically stable xed
point x = 0, then Vn () = Vf () = V + ().
If w(u, x) = uT y and an output y = Cx + Du is dened, then the optimal
control problem corresponds to a problem where the dissipated energy is
to be minimized.
If w(x, 0) 0 then the functions V () which satisfy the DIE in (3.133)
dene Lyapunov functions candidate since V () is then non-increasing
along the (uncontrolled) systems trajectories, as (3.135) shows.
 3.8 Positive Realness and Optimal Control 123

The second part of the last item is satised provided the system is asymp-
totically stabilizable, which is the case if (A, B) is controllable. The rst part
may be satised if R = 0, Q = 0, and the matrix A + BC is Hurwitz. The rst
part of the last-but-one item, is satised if R = 0, Q = 0 (take u = Cx).

Lemma 3.40. Let R > 0. For quadratic functions V (x) = xT Gx, G = GT ,

the DIE in (3.133) is equivalent to the LMI.

Proof: From (3.134) one obtains

2xT G[Ax + Bu] w(u, x), x IRn , u IRm (3.147)

The LMI follows from (3.147). Then the proof is as in Section 3.1.2.

Let us now present some Theorems which show how the LMI, the ARI, the
ARE and the FDI are related one to each other and to the boundedness prop-
erties of the functions Vf (), V + (). The proofs are not provided entirely for
the sake of brievity. In what follows, the notation V () > and V () < +
mean respectively that the function V : IRn IR is bounded for bounded
argument. In other words, given x0 bounded, V (x0 ) is bounded. The control-
lability of (A, B) is sucient for the optimum to be bounded [246, p.229].

Theorem 3.41. The following assertions hold:

Vf () > there exists a real symmetric solution G = GT 0 to
the LMI.
V + () > there exists any real symmetric solution G = GT to the
LMI.
Vf () > = the FDI is satised whenever Re(s) 0, s C.
V + () > = the FDI is satised along Re(s) = 0, s C.

Proof: Let us prove the last two items. If there exists a solution G = GT to
the LMI, then

(In s AT )G G(In s A) GB + C T 2P 0
(3.148)
T
B G+C R 0 0

with s = + j, IR, IR, and s = j. Postmultiplying by


(In s A)1 B
(3.149)
In

and premultiplying by [B T (In s AT )1 In ] one obtains

H(s, s) 2B T (In s AT )1 G(In s A)1 B (3.150)

124 3 Kalman-Yakubovich-Popov Lemma

From the rst item and since 0 one sees that indeed (3.150) implies
the FDI (as P is non-positive denite).

The following theorems characterize the solutions of the ARE.

Theorem 3.42. Let R = RT > 0.

The ARE has a real symmetric solution if and only if H(j, j) 0
for all real , j  (A). There is then only one such solution denoted
as G+ , such that Re((A+ )) 0, A+ = A BR1 (B T G+ + C), and
only one such solution denoted as G , such that Re((A )) 0, A =
A BR1 (B T G + C).
Any other real symmetric solution G satises G G G+ .

One recognizes that A+ and A are the closed-loop transition matrices

corresponding to a stabilizing optimal feedback in the case of A+ . G+ is
called the stabilizing solution of the ARE. V + () and V () are in (3.144)
and (3.145) respectively. It is noteworthy that if in the rst assertion of the
Theorem one looks for negative semi-denite solution of the ARE, then the
equivalence has to be replaced by only if. In such a case the positivity of
the Popov function is only a necessary condition.

Theorem 3.43. Assume that R = RT > 0. Then

V + () > and V < + there exists a real symmetric solution
to the ARE.
Moreover V + (x) = xT G+ x and V (x) = xT G x.
Vf () > there exists a real symmetric non-positive denite solu-
tion to the ARE.
Consequently Vf () > if and only if G 0. When G < 0 then
Vf () = V + () = xT G+ x.
The optimal closed-loop system x(t) = A+ x(t) is asymptotically stable if
G < 0 and G+ > G , where A+ is dened in Theorem 3.42.

One can already conclude from the above results that the set of solutions to
the KYP Lemma conditions (3.2) possesses a minimum solution P = G+
and a maximum solution P + = G when D +DT > 0, and that all the other
solutions P > 0 of the ARE satisfy G+ P G . The last two items
tell us that if the ARE has a solution G < 0 then the optimal controller
asymptotically stabilizes the system. In this case limt+ x(t) = 0 so that
indeed Vf () = V + ().
The function Vf () corresponds to what we shall call the available storage
(with respect to the supply rate w(x, u)) in Chapter 4. The available storage
will be shown to be the minimum solution to the ARE, while the maximum
solution will be called the required supply. Also dissipativity will be character-
ized by the available storage being nite for all x X and the required supply
 3.8 Positive Realness and Optimal Control 125

being lower-bounded. The material in this section brings some further light on
the relationships that exist between optimal control and dissipative systems
theory. We had already pointed out a connection in Section 3.1.2. Having in
mind that what we call a dissipative linear invariant system is a system which
satises a dissipation equality as in (3.4), we can rewrite Theorem 3.42 as
follows:

Theorem 3.44. [511] Suppose that the system (A, B, C, D) in (3.1) is

controllable and observable and that D + DT is full rank. Then the ARE
P A + AT P + (P B C T )(D + DT )1 (B T P C) = 0 has a real sym-
metric non-negative denite solution if and only if the system in (3.1)
is dissipative with respect to the supply rate uT y. If this is the case
then there exists one and only one real symmetric solution P such that
Re((A )) 0, A = A + B(D + DT )1 (B T P C), and one and
only one real symmetric solution P + such that Re((A+ )) 0, A+ =
A + B(D + DT )1 (B T P + C). Moreover 0 < P P + and every real
symmetric solution satises P P P + . Therefore all real symmet-
ric solutions are positive denite. The inequalities H(j) + H T (j) > 0
for all IR, Re((A )) < 0, Re((A+ )) > 0, and P < P + , hold
simultaneously.

It will be seen later that the matrices P + and P play a very particular role
in the energy properties of a dynamical system (Section 4.4.3, Remark 4.37).
Theorem 3.44 will be given a more general form in Theorem 4.58. The matrix
P is the stabilizing solution of the ARE. Algorithms exist that permit to
calculate numericallt the extremal solutions P and P + ; see [145, Annexe
5.A] where a Fortran routine is proposed.
Remark 3.45. Let us study the case when C = 0 and Q = 0, with R = Im
without loss of generality. The ARE then becomes

AT G + GA GBB T G = 0 (3.151)
and obviously G = 0 is a solution. It is the solution that yieldsthe free terminal
+
time optimal control problem of the optimization problem 0 uT (t)u(t)dt.
If the matrix A is Hurwitz, G = 0 is the maximum solution of (3.151). If A
is Hurwitz, G = 0 is the minimum solution to the ARE.
Extensions towards the singular case (R 0) can be found in [506]; see
also Remark 4.94.

3.8.3 The Popov Function and the KYP Lemma LMI

We did not provide most of the proofs of the results of this section, and in
particular Theorem 3.42. Let us end this section with a result that links the
126 3 Kalman-Yakubovich-Popov Lemma

positivity of the Popov function, and a KYP Lemma LMI, and its complete
proof.
Theorem 3.46. [145] The spectral function

Q S 
(sIn A)1 B
(s) = [B T (sIn AT )1 Im ] (3.152)
Im
ST R
where the pair (A, B) is controllable, is non-negative if and only if there exists
P = P T such that

Q AT P P A S P B
0
S B P
T T
R

Before passing to the proof we need some intermediate results.

Lemma 3.47. [145] Let (s) be the spectral function in (3.152), which we
say is described by the ve-tuple (A, B, Q, S, R). Then
i) (s) is also described by the ve-tuple (A2 , B2 , Q2 , S2 , R2 ) where
A2 = A
B2 = B
Q2 = Q AT P P A
S2 = S P B
R2 = R
where P = P T is any matrix.
ii) For H(s) = Im C(sIn A + BC)1 B where C is any m n ma-
trix, the spectral function H T (s)(s)H(s) is described by the ve-tuple
(A1 , B1 , Q1 , S1 , R1 ) where
A1 = A BC
B1 = B
Q1 = Q + C T QC SC C T S
S1 = S C T R
R1 = R.

Proof: i) Let 2 (s) be the Popov function described by the ve-tuple

(A2 , B2 , Q2 , S2 , R2 ). Then
2 (s) (s) = B T (sIn AT )1 (AT P + P A)(sIn A)1 B
B T (sIn AT )1 P B B T P (sIn A)1 B

= B T (sIn AT )1 [AT P + P A + P (sIn A)+ (3.153)

+ (sIn AT )P ](sIn A)1 B

= 0.
 3.8 Positive Realness and Optimal Control 127

ii) Notice that (sIn A)1 BH(s) = (sIn A + BC)1 B. The Popov
function H T (s)(s)H(s) can be written as

H T (s)(s)H(s) =

Q S (sIn A + BC)1 B
= [B T (sIn AT + C T B T )1 H T (s)]
ST R H(s)

Q 1 S1 (sIn A1 )1 B
= [B T (sIn AT1 + C T B T )1 Im ]
S1T R1 Im
(3.154)
which ends the proof.

Lemma 3.48. Let A IRrr , B IRss , C IRrs . The solution of the

equation AP + P B = C is unique if and only if the set of eigenvalues of A
and the set of eigenvalues of B, have no common element.
(C,P )
In the next proof the notation (A, B, Q, S, R) (A , B  , Q , S  , R )
means that one has applied the two transformations of Lemma 3.47 succes-
sively. The two Popov functions which correspond one to each other through
such a transformation are simultaneously non-negative.

Proof of Theorem 3.46: Let C be a matrix such that (A BC) is asymp-

totically stable. Let J be the unique solution of

(A BC)T J + J(A BC) = Q + C T RC SC C T S T (3.155)

(C,J)
One checks that (A, B, Q, S, R) (A BC, B, 0, H T , R) with H =
S JB C T R. Under these conditions the positivity of (s) is equivalent to
that of

 (s) = H(sIn A + BC)1 B + B T (sIn AT + C T B T )H T + R (3.156)

i.e. is equivalent to the existence of a matrix G = GT > 0 such that

(A BC)T G G(A BC) H T GB
(3.157)
H B G T
R
But for P = G + J a direct computation shows that
128 3 Kalman-Yakubovich-Popov Lemma

Q AT P P A S P B
=
ST BT P R

In C T (A BC)T P P (A BC) H T P B In 0
=
0 In H BT P R C In
(3.158)
which ends the proof.
It is noteworthy that the matrix P in Theorem 3.46 is not necessarily
positive denite. We will need those results when we deal with hyperstability.
We notice that Theorem 3.46 states an equivalence under a controllability as-
sumption of the pair (A, B). But it does not say that it is necessary that (A, B)
be controllable for the result to hold; see Section 3.3 for more informations on
this point.

Popovs Function and Triples

Remember that given a Popovs function as in (3.142) we call (A, B, Q, C, R)
a Popov triple.
Denition 3.49. Two Popov triples (A, B, Q, C, R) and (A, B, Q, C, R) are
called (X, F )-equivalent if there exist matrices F IRmn and X = X T
IRnn such that


A = A + BF, B = B



Q = Q + LF + F T LT + F T RF + AX + XA (3.159)





L = L + F T R + XB R = R
One then writes (A, B, Q, C, R) (A, B, Q, C, R). Two Popov triples
(A, B, 0, C, R) and (A, B, 0, C, R) are called dual if A = AT , B = L,
L = B, R = R.
From the material which is presented above, it should be clear that a Popov
triple can be seen as the representation of a controlled dynamical system
x(t) = Ax(t) + Bu(t) together with a functional with a quadratic cost as in
(3.132). To a Popov triple one can therefore naturally associate a Popovs
function (3.142), a Riccati equality, and an extended Hamiltonian pencil

In 0 0 A 0 B



M N = 0 In 0 Q A C T T
(3.160)

0 0 0 C BT R
which shall be denoted as EHP().
 3.8 Positive Realness and Optimal Control 129

Lemma 3.50. [225] (a) If = (A, B, Q, C, R) = (A, B, Q, C, R), then

(s) = SF
(s) (s)SF (s), where SF (s) = F (sIn A)1 B + Im .
(b) If = (A, B, 0, C, R) and = (A, B, 0, C, R) are two dual Popov
triples, then (s) = (s).

The following holds:

Lemma 3.51. [224, 225] Let = (A, B, Q, C, R) be a Popov triple; the fol-
lowing statements are equivalent:
There exists an invertible block 2
2 matrix V
with upper right block zero,
Im1 0
such that R = V T JV , where J = , and the Riccati equality
0 Im2
AT P + P A (P B + C T )R1 (B T P + C) + Q = 0 has a stabilizing solution
P.
has a Jspectral factorization = G
JG, with G, G1 being ratio-
nal m m matrices with all poles in the left open complex plane.

These tools and results are useful in the H theory; see [224, Lemma 2,
Theorem 3].

3.8.4 A Recapitulating Theorem

Let us state a Theorem proved in [349] and which holds for stabilizable sys-
tems (there is consequently also a link with the material of Section 3.3). This
theorem summarizes several relationships between the solvability of the KYP
Lemma set of equations and the regular optimal control problem, under a
stabilizability assumption only.

Theorem 3.52. Let the pair (A, B) be stabilizable. Then the following asser-
tions are equivalent:
(i) The optimal control problem: (3.131) and (3.132) subject to x(t) =
Ax(t)+Bu(t), x(0) = x0 , is regular, i.e. it has a solution for any x0 IRn ,
and this solution is unique.
(ii) There exists a quadratic Lyapunov function V (x) = x P x, P = P ,
such that the form V + w(u, x) = 2x P (Ax + Bu) + w(u, x) of the variables
x Cn and u Cm is positive denite.
(iii) The condition w(u, x) (x x+u u) for any value of IR, x Cn ,
u Cm satisfying jx = Ax + Bu, holds for some > 0.
(iv) The matrix R = RT in (3.132) is positive denite and the set of
equations P A + A P + Q = kRk , P B + C = kR, possesses a solution
in the form of real matrices P = P T and C, such that the controller
u = Cx stabilizes the system x(t) = Ax(t) + Bu(t).
130 3 Kalman-Yakubovich-Popov Lemma

0 In
(v) R > 0 and det(jJ K) = 0 for all IR, with J = ,
In 0

C T R1 C Q AT CR1 B T
K= .
A BR1 C T BR1 B T
(vi) R > 0 and there exist a quadratic form V = x P x, P = P , and
1
a matrix k IRnm , such that V + w(u, x) = |R 2 (u k x)|2 and the
controller u = k x stabilizes the system x(t) = Ax(t) + Bu(t).
(vii) The functional Vf () in (3.131) is positive denite on the set M(0) of
processes (x(), u()) that satisfy x(t) = Ax(t) + Bu(t) with x(0) = x0 = 0,
i.e. there exists > 0 such that

+
+
w(u(t), x(t))dt (xT (t)x(t) + uT (t)u(t))2 dt
0 0

for all (x(), u()) M(0), where M(x0 ) is the set of admissible processes.
Let at least one of these assertions be valid (which implies that they are all
valid). Then there exists a unique pair of matrices (P, k) which conforms with
the requirements of item (iv). In the same way there is a unique pair which
complies with the requirements of item (vi) , and the pairs under consideration
are the same. Finally any of the items (i) through (vii) implies that for any
initial state x0 IRn one has V (x0 ) = xT0 P x0 = minM(x0 ) Vf (x(), u()).

The set M(x0 ) of admissible processes consists of the set of pairs (x(), u())
which satisfy x(t) = Ax(t) + Bu(t) with x(0) = x0 , with u L2 . If (A, B) is
controllable then M(x0 ) = for any x0 IRn .

3.8.5 On the Design of Passive LQG Controllers

The Linear-Quadratic-Gaussian (LQG) controller has attained considerable

maturity since its inception in the 1950s and 1960s. It has come to be generally
regarded as one of the standard design methods. One attribute of LQG-type
compensators is that, although they guarantee closed-loop stability, the com-
pensator itself is not necessarily stable. It would be of interest to characterize
the class of LQG-type compensators which are stable. Going one step further,
if the LQG compensator is restricted to be not only stable, but also passive,
this would dene an important subclass. The importance of such compensators
is that they would not only be passive, but would also be optimal with respect
to an LQG performance criteria. One reason for considering passive compen-
sators is that, when used to control positive real plants, they oer excellent
robustness to modeling errors as long as the plant is PR. An important appli-
cation of passive compensators is vibration suppression in large exible space
structures (LFSS), which are characterized by signicant unmodeled dynam-
ics and parameter errors. The linearized elastic-mode dynamics of LFSS [253]
 3.8 Positive Realness and Optimal Control 131

with compatible collocated actuators and sensors are PR systems regardless

of the unmodeled dynamics or parameter uncertainties can, therefore, be ro-
bustly stabilized by an SPR compensator.
The objective of this section is to investigate the conditions under which
an LQG-type compensator is SPR, so that one can simultaneously have high
performance and robustness to unmodeled dynamics.
Consider a minimal realisation of a PR system expressed by the following
state space representation:

x(t) = Ax(t) + Bu(t) + v(t)
(3.161)

y(t) = Cx(t) + w(t)

where v() and w() are white, zero-mean Gaussian noises. Since the system
is PR, we assume, without loss of generality (see Remark 3.54 at the end of
this section), that the following equations hold for some matrix Qa 0:

A + AT = QA 0 (3.162)
and

B = CT (3.163)
The above conditions are equivalent to the Kalman-Yakubovich-Popov
Lemma. The LQG compensator for the system (3.161), (3.162) and (3.163) is
given by (see [9])

u(t) = u (t) (3.164)

 

x(t) = A BR1 B T Pc Pf BRw
1 T 1
B x(t) + Pf BRw y(t) (3.165)

u (t) = R1 B T Pc x(t) (3.166)

where Pc = PcT > 0 and Pf = PfT > 0 are the LQ-regulator and the
Kalman-Bucy lter Riccati matrices which satisfy the algebraic Riccati equa-
tions
Pc A + AT Pc Pc BR1 B T Pc + Q = 0 (3.167)

1 T
Pf AT + APf Pf BRw B Pf + QV = 0 (3.168)

where Q and R are the usual weighting matrices for the state and input, and
QV and RW are the covariance matrices of v and w. It is assumed that Q > 0
1/2
and that the pair (A, QV ) is observable. The main result is stated as follows:
132 3 Kalman-Yakubovich-Popov Lemma

Theorem 3.53. [312] Consider the PR system in (3.161), (3.162)

and (3.163) and the LQG-type controller in (3.164) through (3.168). If
Q, R, Qv and Rw are such that

Qv = Qa + BR1 B T (3.169)

Rw = R (3.170)
and
Q BR1 B T = QB > O

(3.171)
then the controller in (3.165) through (3.166) (described by the transfer

function from y to u ) is SPR.

Proof: Introducing (3.162), (3.169), (3.170) into (3.168), it becomes clear that
Pf = I is a solution to (3.168). From (3.167) it follows:

Pc (A BR1 B T Pc BR1 B T ) + (A BR1 B T Pc BR1 B T )T Pc

= Q Pc BR1 B T Pc Pc BR1 B T BR1 B T Pc

= Q (Pc + I)BR1 B T (Pc + I) + BR1 B T

= QB (Pc + I)BR1 B T (Pc + I) < 0

where QB is dened in (3.171). In view of (3.163 ), (3.170) and the above, it
follows that the controller in (3.165) and (3.166) is strictly positive real.
The above result states that, if the weighting matrices for the regulator
and the lters are chosen in a certain manner the resulting LQG-type compen-
sator is SPR. However, it should be noted that this compensator would not be
optimal with respect to actual noise covariance matrices. The noise covariance
matrices are used herein merely as compensator design parameters and have
no statistical meaning. Condition (3.171) is equivalent to introducing an addi-
tional term y T R1 y in the LQ performance index (since Q = QB + CR1 C T )
and is not particularly restrictive. The resulting feedback conguration is
guaranteed to be stable despite unmodeled plant dynamics and parameter
inaccuracies, as long as the plant is positive real. One application of such
compensators would be for controlling elastic motion of large exible space
structures using collocated actuators and sensors. Further work on passive
LQG controllers has been carried out in [99, 160, 165, 179, 237, 238].
 3.8 Positive Realness and Optimal Control 133

Remark 3.54. Consider a positive real system expressed as

z(t) = Dz(t) + F u(t)
(3.172)

y(t) = Gz(t)
Then, there exists matrices P > 0 and L such that

P D + DT P = LLT
(3.173)

P F = GT
1 1
Dene x = P 2 z, where P 2 is a symmetric square root of P [272]. Intro-
ducing this denition in (3.172), we obtain a state space representation as the
1 1 1 1
one in (3.161), but with A = P 2 DP 2 , B = P 2 F , C = GP 2 . Multiplying
1
the rst equation in (3.173) on the left and on the right by P 2 we obtain
1 1 1
(3.162) with QA = P 2 LLT P 2 . Multiplying (3.173) on the left by P 2 we
obtain (3.163).

3.8.6 Summary
Let us recapitulate some of the material in the previous subsections. We con-
sider the two matrix polynomials

R(P ) = AT P + P A + (C B T P )T (D + DT )1 (C B T P )
(3.174)
S(G) = AG + GAT + (B GC T )(DT + D)1 (B GC T )T
and the linear invariant system () : x(t) = Ax(t) + Bu(t), y(t) = Cx(t) +
Du(t) which is controllable and observable.
Then all the following statements are equivalent one to each other [480]:
1) The transfer function of () is extended SPR.
2) There exists a positive denite matrix P such that
T
A P + P A CT P B
<0 (3.175)
C B T P (D + DT )
3) D + DT > 0 and the ARI R(P ) < 0 has a positive denite solution Pi .
4) D + DT > 0 and the ARE R(P ) = 0 has a solution Pe such that
A + (D + DT )PE has all its eigenvalues with strictly negative real parts.
5) There exists a positive denite matrix G such that

AG + GAT B GC T
<0 (3.176)
B CG (D + D)
T T
134 3 Kalman-Yakubovich-Popov Lemma

6) D + DT > 0 and the ARI S(G) < 0 has a positive denite solution Gi .
7) D + DT > 0 and the ARE S(G) = 0 has a solution Ge such that
A + (D + DT )Ge has all its eigenvalues with strictly negative real parts.
In addition, assume that any of the above statements 1)7) holds. Then:
8) If the matrix P (resp. Pi ) solves the inequality (3.175) (resp. R(P ) <
0) then its inverse P 1 (resp. Pi1 ) solves the inequality (3.176) (resp.
S(G) < 0), and vice-versa.
9) The inequalities 0 Pe < Pi and 0 Ge < Gi hold.

Lemma 3.7 is used to prove some of the above equivalences. More on Ric-
cati equations can be found in [273, 416]; see also Appendix A.4. Point 2)
above and Theorem A.61 show that extended SPR functions and SSPR func-
tions can be tested with the same LMI conditions and are therefore equivalent
notions.
Let us recall a fundamental result which is also closely linked to the KYP
Lemma solvability under no-controllability assumption of (A, B). Given A
IRnn , B IRnm , M = M T IR(n+m)(n+m) , with det(jIn A) = 0 for
IR (A does not have imaginary eigenvalues) and (A, B) controllable, the
next two statements are equivalent [145, 412]:


(jIn A)1 B (jIn A)1 B
M 0 for all [, +].
Im Im
There exists a matrix P = P T IRnn such that
T
A P + PA PB
M + 0 (3.177)
BT P 0

Q CT
When M = , one recovers the KYP Lemma set of equations.
C D + DT
When Q 0 then P 0 and A is Hurwitz. The corresponding equivalence
with strict inequalities holds even if (A, B) is not controllable. This equivalence
therefore somewhat generalizes Proposition 2.31. The generalization of this
equivalence for a limited range of frequencies || , has been proposed
in [229, 230]. This has important practical consequences.

3.8.7 A Digression on Semidenite Programming Problems

The above equivalence makes a nice transition to the relationships between

semidenite programming problems (SDP) and the KYP Lemma. Let us con-
sider a SDP of the form
 3.9 The Lure Problem (Absolute Stability) 135
L
minimize q T x + k=1 Tr(Qk Pk )

ATk Pk + Pk Ak Pk Bk p
subject to +
i=1 xi Mki Nk , k = 1, ..., L
BkT Pk 0
(3.178)
where the variables (unknowns) are x IRp and Pk = PkT IRnk nk , the
problem data are q IRp , Qk = QTk IRnk nk , Ak IRnk nk , Bk IRnk mk ,
Mki = MkTi IR(nk +mk )(nk +mk ) , and Nk = NkT IR(nk +mk )(nk +mk ) . Such
a SDP is named a KYP-SDP [498] because of the following. As seen just above
the KYP Lemma states a frequency domain inequality of the form


(jIn A)1 B (jIn A)1 B
M 0 (3.179)
Im Im
for all [, +], with M symmetric and A has no imaginary eigenvalue
(equivalently the transfer function C(sIn A)1 B + D has no poles on the
imaginary axis). And (3.179) is equivalent to the LMI in (3.177). The con-
straints in the KYP-SDP in (3.178) possess the same form as (3.177) where
M is replaced by an ane function of the variable x. Let us take Qk = 0, then
the KYP-SDP can equivalently be rewritten as

minimize q T x


(jIn Ak )1 Bk (jIn Ak )1 Bk
subject to (Mk (x) Nk ) 0
Im Im

k = 1, ..., L
p (3.180)
where the optimization variable is x and Mk (x) = i=1 xi M ki . Applica-
tions of KYP-SDPs are in optimization problems with frequency-domain in-
equalities, linear systems analysis and design, digital lter design, robust con-
trol analysis using integral quadratic constraints, linear quadratic regulators,
quadratic Lyapunov functions search, etc. More details may be found in [498].
We do not provide more details on this topic since this would bring us too far
away from our main interest in this book.

3.9 The Lure Problem (Absolute Stability)

3.9.1 Introduction

In this section we study the stability of an important class of control systems.

The Lure problem has been introduced in [321], it was very popular in the
136 3 Kalman-Yakubovich-Popov Lemma

1950s 2 and can be considered as the rst steps towards the synthesis of
controllers based on passivity. For a complete account on the Russian school
to the Lure problem, [524] is mandatory reading. Consider the closed-loop
system shown in Figure 3.3. We are interested in obtaining the conditions
on the linear system and on the static nonlinearity such that the closed-loop
system is stable. This is what is called the Lure problem.

y
0 + u
Linear
system

I (t,y)
z

Fig. 3.3. The Lure problem

The linear system is given by the following state-space representation:

x(t) = Ax(t) + Bu(t)
() (3.181)

y(t) = Cx(t) + Du(t), x(0) = x0
with x(t) IRn , u(t), y(t) IRm , m < n. The static nonlinearity : IR
IRm IRm is possibly time-varying and described by

z(t) = (t, y(t))

(3.182)
u(t) = z(t) (interconnection relation)

The linear system is assumed to be minimal, i.e. controllable and observ-

able which means that
 
rank B AB ... An1 B = n,

2
Of the 20th century.
 3.9 The Lure Problem (Absolute Stability) 137

and

C
CA
rank

= n.

:
n1
CA
The nonlinearity is assumed to belong to the sector [a, b], i.e.:
i) (t, 0) = 0 t0
T
ii) [ (t, y) ay] [by (t, y)] 0 t 0, y(t) IRm
In the scalar case (m = 1), the static nonlinearity is shown in Figure 3.4.

Fig. 3.4. Static nonlinearity for n = 1

3.9.2 Well-posedness of ODEs

The function (, ) must be such that the closed-loop system possesses a

unique solution. For an ordinary dierential equation x(t) = f (x(t), t), the
so-called Caratheodory conditions are as follows:

Theorem 3.55. [107] Let I = {(x, t) | ||xx0 || b, |t | a, a IR+ , b

IR+ }, and let us assume that f : I IR satises:
(i) f (x, ) is measurable in t for each xed x
(ii) f (, t) is continuous in x for each xed t
(iii) there exists a Lebesgue integrable function m() on the interval |t |
a such that |f (x, t)| m(t) for all (x, t) I
138 3 Kalman-Yakubovich-Popov Lemma

Then for some > 0 there exists an absolutely continuous solution x() on
some interval |t | , 0, satisfying x( ) = x0 .

One notices that, due to the absolute continuity of the solution x(), it
follows that the equality x(t) = f (x(t), t) is satised almost everywhere in the
Lebesgue measure (i.e. for all t in the said interval, except on a set of zero
Lebesgue measure). When f (, ) satises ||f (t, x)f (t, y)|| (|t |, ||xy||)
where (, ) is continuous and non-negative, then uniqueness of the solution
starting at x0 is guaranteed (and its derivative is unique up to a set of zero
Lebesgue measure in the said interval of time). When f (, ) is a C r function
of both x and t, then local existence and uniqueness of a solution which is also
a C r function of both x and t, is guaranteed [28]. The basic and classical
well-posedness results for an ordinary dierential equation x(t) = f (t, x(t))
are as follows:

Theorem 3.56 (Local Existence and Uniqueness [96]). Let f (t, x) be

continuous in a neighborhood N of (t0 , x0 ) IR IRn , and be locally Lipschitz
with Lipschitz constant k. Then there exists > 0 such that the ODE x(t) =
f (t, x(t)) possesses in the interval I = [t0 , t0 +] one and only one solution
x : I IRn such that x(0) = x0 .

The denition of Lipschitz functions is in Denitions 4.2 and 4.3.

Theorem 3.57 (Global Uniqueness [96]). Let f (t, x) be locally Lipschitz.

Let I IR be an interval (I may be open, closed, unbounded, compact, etc).
If x1 () and x2 () are two solutions of x(t) = f (t, x(t)) on I and if they are
equal for some t0 I, then they are equal on the whole I. If in addition f (t, x)
is continuous in some domain U IR IRn and if (t0 , x0 ) U , then there
exists a maximum interval J  t0 in which a solution exists, and this solution
is unique.

Theorem 3.58 (Continuous Dependence on Initial Data). Let f :

W IRn , W IRn an open set, be Lipschitz with constant k. Let x1 ()
and x2 () be solutions of x(t) = f (x(t)) on the interval [t0 , t1 ]. Then for all
t [t0 , t1 ], one has ||x1 (t) x2 (t)|| ||x1 (t0 ) x2 (t0 )|| exp(k(t t0 )).

The proof of Theorem 3.58 is based on Gronwalls Lemma which is re-

called later in the book (Lemma 3.68). It is noteworthy that some of the
nonsmooth dynamical systems which are studied in this book do not enjoy
the continuity in the initial data property, like Lagrangian systems subject to
complementarity conditions (unilateral constraints).
In Section 3.9.4, well-posedness will be extended to multivalued and nons-
mooth feedback nonlinearities. Then new tools for studying the well-posedness
are required. Concerning the closed-loop system (3.181) and (3.182), one has
f (x(t), t) = Ax(t) B(t, Cx(t)) when D = 0, and the conditions on (t, y)
which assure that the vector eld ts within the conditions of Theorems 3.55,
 3.9 The Lure Problem (Absolute Stability) 139

3.56 or 3.57, are easily deduced. It is worth noting that when D = 0 some
care is needed. Indeed one obtains

y = Cx D(t, y), (3.183)

and the output mapping makes sense only if Equation (3.183) has a unique
solution y = h(x) for all t 0 and all x IRn . A single-valued mapping ()
is monotone if x x , y y   0 whenever x = (y) and x = (y  ). It is
strongly monotone if x x , y y   ||y y  ||2 for some > 0.

Lemma 3.59. Let D 0 and : IRm IRm be monotone. Then the equation

y = Cx D(y) (3.184)
possesses a unique solution y = h(x) for all x IRn .

Proof: The proof uses the fact that the generalized equation 0 F (x) pos-
sesses a unique solution provided the mapping F () is strongly monotone
on IRn [137, Theorem 2.3.3]. We are thus going to show that the mapping
y
y + D(y) is strongly monotone. Take two couples (x, x ) and (y, y  ) in
the graph of this mapping, i.e. x = x + D(x) and y  = x + D(y). Then

(x y)T (x y  ) = (x y)T (x y + D(x) D(y))

= (x y)T (x y) + (x y)T D((x) (y))

(x y)T (x y) + min (D)(x y)T ((x) (y))

(x y)T (x y)
(3.185)
This inequality precisely means that y
y + D(y) is strongly monotone
[137, Denition 2.3.1]. Thus y
y + D(y) + for some IRm is strongly
monotone as well.
The proof of the above fact applies to generalized equations of the form
0 F (x) + NK (x), where NK () is the normal cone to the closed convex
set K IRn (we shall come back on convex analysis later in this chapter).
It happens that NIRn (x) = {0} for all x IRn . But it is worth keeping in
mind that the result would still hold by restricting the variable y to some
closed convex set. Coming back to the Lure problem, one sees that a direct
feedthrough of the input in the output is allowed, provided some conditions are
respected. Positive real systems with D > 0 (which therefore have a uniform
vector relative degree r = (0, ..., 0)T IRm ), or with D 0, meet these
conditions.
140 3 Kalman-Yakubovich-Popov Lemma

3.9.3 Aizermans and Kalmans Conjectures

Lure problem in Figure 3.3 can be stated as follows: Find the conditions
on (A, B, C, D) such that the equilibrium point x = 0 of the closed-loop
system is globally asymptotically stable for all nonlinearities () in the sec-
tor [a, b]. Then the system is said to be absolutely stable. Another way to
formulate it is as follows: suppose the nonlinearity (, ) belongs to the sec-

tor [0, k]. The absolute stability problem is to nd the value k
= inf{k
0 | there exists
() in the sector [0, k] for which the feedback system
(3.181) (3.182) is not asymptotically stable}. Equivalently, the feedback sys-
tem (3.181) (3.182) is asymptotically stable for any nonlinearity in the sector
[0, k
]. In the next sections, we shall rst review three celebrated conjectures
which happen to be true only in very specic cases. Then we shall see what
happens when the feedback nonlinearity (, ) is no longer a function but a
multivalued function. This demands new mathematical tools to be correctly
handled, and we shall spend some time on this. Then two celebrated results,
the circle criterion and the Popovs criterion, will be presented.

Fig. 3.5. Linear system with a constant output feedback

Conjecture 3.60 (Aizermans conjecture). If the linear subsystem with D = 0

and m = 1 in Figure 3.5 is asymptotically stable for all (y) = ky, k [a, b],
then the closed loop system in Figure 3.6 with a time-invariant nonlinearity
() in the sector [a, b] is also globally asymptotically stable.

Aizermans conjecture states that if the vector eld Ax + b(y) is Hurwitz

for all linear characteristic functions (), then the xed point x = 0 should be
globally asymptotically stable for any continuously dierentiable () whose
slope remains bounded inside [a, b].
 3.9 The Lure Problem (Absolute Stability) 141

Conjecture 3.61 (Kalmans conjecture). Consider the system in Figure 3.6

with a nonlinearity such that (t, y) = (y) (i.e. a time-invariant and con-
tinuously dierentiable nonlinearity), m = 1, (0) = 0 and a d dy (y) b.
Then the system in (3.181) with D = 0 is globally asymptotically stable if it
is globally asymptotically stable for all nonlinearities (y) = ky, k [a, b].

Fig. 3.6. Linear system with a sector nonlinearity in negative feedback

Thus Kalmans conjecture says that if AkBC is Hurwitz for all k [a, b],
x = 0 should be a globally stable xed point for (3.181) (3.182) with ()
as described in Conjecture 3.61. However it turns out that both conjectures
are false in general. In fact, the absolute stability problem, and consequently
Kalman conjecture, may be considered as a particular case of a more general
problem known in the Applied Mathematics literature as the Markus-Yamabe
conjecture (MYC in short). The MYC can be stated as follows [350]:

Conjecture 3.62 (Markus-Yamabes conjecture). If a C 1 map f : IRn IRn


satises f (0) = 0 and if its Jacobian matrix f
x  is stable for all x0 IRn ,
x0
then 0 is a global attractor of the system x(t) = f (x(t)).

In other words, the MYC states that if the Jacobian of a system at any
point of the state space has eigenvalues with stricty negative real parts, then
the xed point of the system should be globally stable as well. Although this
conjecture seems very sound from an intuitive point of view, it is false for
n 3. Counter examples have been given for instance in [104]. It is however
true in dimension 2, i.e. n = 2. This has been proved in [175]. The proof
is highly technical and takes around 40 pages. Since it is, moreover, outside
the scope of this monograph dedicated to dissipative systems, it will not be
reproduced nor summarized here. This is however one nice example of a result
142 3 Kalman-Yakubovich-Popov Lemma

that is apparently quite simple and whose proof is quite complex. The Markus-
Yamabe conjecture has been proved to be true for gradient vector elds, i.e.
systems of the form x(t) = f (x(t)) with f () of class C 2 [334]. It is clear that
the conditions of the Kalmans conjecture with f (x) = Ax + b(y), (0) = 0,
make it a particular case of the MYC. In short one could say that Kalmans
conjecture (as well as Aizermans conjecture) is a version of MYC for control
theory applications. Since, as we shall see in the next subsections, there has
been a major interest in developing (sucient) conditions for Lure problem
and absolute stability in the Systems and Control community, it is also of
signicant interest to know the following result:

Theorem 3.63. [35, 46] Kalmans conjecture is true for dimensions n =

1, 2, 3. It is false for n > 3.

Since it has been shown in [175] that the MYC is true for n = 1, 2, it
follows immediately that this is also the case for the Kalmans conjecture.
Aizermans conjecture has been shown to be true for n = 1, 2 in [163], proving
in a dierent way that Kalmans conjecture holds for n = 1, 2. The following
holds for the case n = 3:

Theorem 3.64 (n = 3 [35]). The system

x(t) = Ax(t) + b(y(t))
(3.186)

y(t) = cT x(t)

with x(t) IR3 , y(t) IR, b IR3 , c IR3 , miny d d

dy (y) = 0, maxy dy (y) =
k (0, +), (0) = 0, is globally asymptotically stable if the matrices A +
dy (y)c IR are Hurwitz for all y(t) IR.
d T nn

3.9.4 Multivalued Nonlinearities

It is of interest to extend the Lure problem to the case where the static
nonlinearity in the feedback loop is not dierentiable, or even not a single-
valued function (say, a usual function), but is a multivalued function. The
material in this section is taken from [81]. Before stating the main results, we
need to introduce some basic mathematical notions from convex analysis. The
reader who wants to learn more on convex analysis and dierential inclusions
with monotone mappings, is invited to have a look at the textbooks [66, 168,
210, 359].

Basic Facts on Convex and Nonsmooth Analysis

Let K IRn denote a convex set. Its indicator function is dened as

 3.9 The Lure Problem (Absolute Stability) 143

Fig. 3.7. Tangent and normal cones

0 if x K
K (x) = (3.187)

+ if x  K

A convex function f : IRn IR satises f (x+(1)y) f (x)+(1)f (y)

for all 0 < < 1, and for all x and y in its (convex) domain of denition.
The indicator function K () is convex if and only if K is convex. A convex
function is not necessarily dierentiable, so that a more general notion of a
derivative has to be introduced. The subdierential of a convex function f ()
at y is denoted as f (y) and is the set of subgradients, i.e. vectors IRn
satisfying
f (x) f (y) T (x y) (3.188)
for all x IRn . Geometrically, (3.188) means that one can construct a set of
ane functions (straight lines) y
(x y)T + f (x) whose slope is a
subgradient of f () at x. The set f (y) may be empty, however if f () is convex
and f (y) < + then f (y) = [359]. The simplest example is f : IR IR+ ,
x
|x|. Then


1 if x < 0



f (x) = [1, 1] if x = 0 (3.189)





1 if x > 0
One realizes in passing that |x| is the so-called relay characteristic and
that 0 |0|: the absolute value function has a minimum at x = 0. The
144 3 Kalman-Yakubovich-Popov Lemma

subdierential of the indicator of K (which is convex if K is convex) is given

by

{0} if x Int (K)



K (x) = NK (x) if x K (3.190)





if x  K
where K is the boundary of K, and

NK (x) = {z | z T ( x) 0, K} (3.191)
is the outwards normal cone to K at x. Notice that 0 NK (x) and that
we have drawn the sets x + NK (x) rather than NK (x) in Figure 3.7. Also
NK (x) = {0} if x Int(K), where Int(K) = K \ K. The set in (3.190) is
the subdierential from convex analysis.

Example 3.65. If K = [a, b] then NK (a) = IR and NK (b) = IR+ .

Remark 3.66. The symbol is used in three dierent meanings in this section:
boundary of a set, subdierential and partial derivative. Since this notation
is classical we choose not to change it.

Denition 3.67. Let K be a convex cone. Its polar cone (or negative cone)
is

K
= {s IRn | s, x 0 for all x K} (3.192)

The inwards tangent cone TK (x) is the polar cone to NK (x) and is dened
as TK (x) = {z | NK (x), T z 0}. Both the normal and the tangent
cones are convex sets. If the set K is dened as {x | h(x) 0} for some
dierentiable function h : IRn IRm , then an alternative denition of the
tangent cone at x is [358]

TK (x) = {v IRn | v T hi (x) 0, i J(x)} (3.193)

with J(x) = {i {1, ..., m} | hi (x) 0}. One notes that this denition
coincides with the rst one as long as x K, and that K needs not be convex
to dene TK (x) in (3.193). Some examples are depicted in Figure 3.7; see
also [69].
A mapping () from X to Y is said to be multivalued if it assigns to
each element x of X a subset (x) of Y (which may be empty, contain just
one element, or contain several elements). The graph of a mapping () is
dened as gph() = {(x, y) | y (x)}. The mappings whose graphs are in
Figure 3.8 (cf) are multivalued. A multivalued mapping () is monotone if
(x x )T (y y  ) 0 for any couples (x, y) and (x , y  ) in its graph, i.e. x
 3.9 The Lure Problem (Absolute Stability) 145

(y  ) and x (y). When n = 1 monotone mappings correspond to completely

non-decreasing curves. When () is single-valued, monotonicity simply means
((y) (y  ))T (y y  ) 0 for all y and y  . Let dom() = {x|x X, (x) = }
be the domain of (). Recall that the domain of a (single-valued) function f ()
is dom(f ) = {x | f (x) < +}. A monotone mapping () is maximal if for
any x X and any y Y such that y y1 , x x1  0 for any x1 dom()
and any y1 (x1 ), then y (x). Complete nondecreasing curves in IR2
are the graphs of maximal monotone mappings. Another interpretation is
that the graph of a maximal monotone mapping cannot be enlarged without
destroying the monotonicity (hence the notion of maximality). Examples of
monotone mappings (n = 1) are depicted in Figure 3.8. They may represent
various physical laws, like dead-zone (a), saturation or elasto-plasticity (b),
corner law unilateral eects, ideal diode characteristic (c), Coulomb friction
(d), MOS transistor ideal characteristic (e), unilateral and adhesive eects (f).
Maximal monotone mappings play an important role in the study of innite
dimensional systems. As is illustrated next, they also nd nice application
in the Lure problem. One can see easily that if an operator H : u
y is
monotone, then it is also passive.

Fig. 3.8. Monotone mappings (one-dimensional case)

We nally end this section by recalling classical tools and denitions which
we shall need next:
146 3 Kalman-Yakubovich-Popov Lemma

Lemma 3.68 (Gronwalls Lemma). Suppose f : IR+ IR+ is a con-

tinuous function, and b 0, c 0, are some constants. Then, if f (t)
t
b + 0 cf (s)ds for all t 0, one has f (t) b exp(ct) for all t 0.
We recall the denition of an absolutely continuous function.
Denition 3.69. Let < a < b < +. A function f : [a, b] IRn is
absolutely continuous if for all > 0 there exists a > 0 such that for all
n IN and any n(1 , 1 ), (2 , 2 ), ....,(n , n ) in
nfamily of disjoint intervals
IR satisfying i=1 (i i ) < , one has i=1 |f (i ) f (i )| < .
In fact absolutely continuous (AC) functions are usually better known as
follows:
Theorem 3.70. An AC function f : [a, b] IR is almost x everywhere dier-
entiable with derivative f() L1 and f (x) f (a) = a f(t)dt for any a x.

Theorem 3.70 can also be stated

 as: there exists a Lebesgue integrable
function g() such that f (t) = g( )d (d being the Lebesgue measure).
In a more sophisticated and pedantic language, df = g(t)dt as an equality
of measures, which means that f(t) = g(t) almost everywhere. A function is
Lipschitz continuous if and only if it is absolutely continuous and its derivative
f is essentially bounded in the sense that there exists a compact set K such
that f(t) K for almost all t [a, b]. All continuously dierentiable (C 1 )
functions dened on a compact interval of IR, are AC. AC functions are of
bounded variation (see Denition 6.58) on such an interval and possess a
derivative almost everywhere. For functions dened on IR one then may dene
the notion of local AC, which simply means that the restriction of the function
on any bounded interval of IR is AC. Let us now introduce the notion of lower
semi-continuity.
Denition 3.71. Let f : X IRn IR. The function f () is said lower
semi-continuous (lsc) at x
X if lim inf xx f (x) f (x
).
Obviously a continuous function at x
is also lsc at x
. But the contrary
is false (otherwise both properties would make one!). An lsc function can be
discontinuous. The sublevel sets are dened as Sr (f ) = {x X | f (x) r}.
Proposition 3.72. A function f : IRn IR{+} is lower semi-continuous
on IRn if and only if the sublevel-sets Sr (f ) are closed (possibly empty) for all
r IR.
The subdierential () of a convex lower semicontinuous function on
IRn is a maximal monotone mapping, and (x) is a convex closed do-
main (possibly empty) of IRn . One has for instance (x) = IR (x) in Fig-
x2
ure 3.8 (c), (x) = |x| + 2 for (d), (x) = (,a] (x) [a,+) (x) +
 3.9 The Lure Problem (Absolute Stability) 147

ab
(x b)2 if |x| b
2 for (e). If (x1 , , xm ) = 1 |x1 | + + m |xm | +
0 if |x| < b
1 T
2 x x, then (0) = ([1 , 1 ], , [m , m ]) . Let us now state a classical
T

result of convex analysis, which is a generalization of the chain rule [168].

Proposition 3.73. Assume that f : Y (, +] is convex and lower

semi-continuous. Let A : X Y be a linear and continuous operator. Assume
that there exists a point y0 = Ax0 at which f () is nite and continuous. Then

(f A)(x) = AT f (Ax) (3.194)

for all x X.

Further generalizations exist, see [415, 10.B]. Let us now state a gen-
eralization of the existence and uniqueness results (Theorems 3.55 to 3.57).
The next theorem is known as the Hille-Yosida Theorem when the operator
A : x
Ax is linear.

Theorem 3.74 (Existence and uniqueness of solutions of monotone

inclusions). [66, Theorem 3.1] Let A be a maximal monotone operator map-
ping IRn into IRn . Then for all x0 dom(A) there exists a unique Lipschitz
continuous function x() on [0, +) such that

x(t) + Ax(t)  0
(3.195)

x(0) = x0
almost everywhere on (0, +). The function satises x(t) dom(A) for
all t > 0, and it possesses a right-derivative for all t [0, +). If x1 ()
and x2 () are two solutions then ||x1 (t) x2 (t)|| ||x1 (0) x2 (0)|| for all
t [0, +). In case the operator A is linear then x() C 1 ([0, +), IRn )
C 0 ([0, +), dom(A)). Moreover ||x(t)|| ||x0 || and ||x(t)|| ||Ax(t)||
||Ax0 || for all t 0.

It is noteworthy that the notion of an operator in Theorem 3.74 goes

much further than the mere notion of a linear operator in nite dimension. It
encompasses subdierentials of convex functions, as will be seen next. It also
has important applications in innite-dimensional systems analysis.

+1 if x > 0
Example 3.75. Let Ax = [0, 1] if x = 0 . Then the solution is

1 if x < 0

(x0 t)+ if x0 0
x(t) =
x0 if x0 < 0
148 3 Kalman-Yakubovich-Popov Lemma

The Multivalued Absolute Stability Problem

It is of interest to extend the absolute stability problem with a single-valued

feedback nonlinearity, to the case where the operator : y
yL = (y) is
a maximal monotone operator. The state space equations of the system are
given by a.e.

x(t) = Ax(t) ByL (t)



y(t) = Cx(t) (3.196)





yL (y),
where y(t), yL (t) IRm , x(t) IRn and a.e. means almost everywhere in the
Lebesgue measure. The xed points of (3.196) can be characterized with the
generalized equation
0 {Ax0 } B(Cx0 ).
One notices that the system in (3.196) is a dierential inclusion, due to
the multivalued right-hand-side. Indeed the subdierential (y) is in gen-
eral multivalued. What is the dierence between the dierential inclusion in
(3.196) and, say, Filippovs systems, which readers from Systems and Control
are more familiar with? The main discrepancy between both is that the right-
hand-side of (3.196) need not be a compact (bounded) subset of the state
space X IRn , for all x X. It can for instance be a normal cone, which is
usually not bounded (the normal cone at a of the interval [a, b], a < b, is the
half line IR ; see Example 3.65). Of course there is a nonzero overlap between
the two sets of inclusions: If the feedback loop contains a static nonlinearity
as in Figure 3.8 (d), then the inclusion (3.196) can be recast either into the
maximal monotone formalism, or the Filippov formalism. Actually, Fil-
ippovs systems are in turn a particular case of what one can name standard
dierential inclusions, i.e. those inclusions whose right-hand-side is compact,
convex, and possesses some linear growth property (see [124] for more details).
To summarize, the basic assumptions on the right-hand-sides of both types of
inclusions dier so much that their study (mathematics, analysis for control)
surely dier a lot as well.
Let us assume that
a) G(s) = C(sI A)1 B, with (A, B, C) a minimal representation, is a SPR
transfer matrix. In particular from the KYP Lemma this implies that
there exists positive denite matrices P = P T and Q = QT such that
P A + AT P = Q and B T P = C.
b) B is full column rank, equivalently Ker(B) = {0}. Thus CA1 B +
B T AT C T is negative denite 3 .
c) : IRm IR {+} is convex lower semi-continuous, so that is a
maximal monotone multivalued mapping.
3
Indeed B T AT C T + CA1 B = B T AT QA1 B < 0.
 3.9 The Lure Problem (Absolute Stability) 149

Lemma 3.76. [81] Let assumptions a)c) hold. If Cx(0) dom , then
the system in (3.196) has a unique absolutely continuous (AC) solution on
[0, +).

Proof: Let R be the square root of P , i.e. R = RT > 0, RR = P . Consider the

convex lower semi-continuous function f : IRn IR{+} dened by f (z) =
(CR1 z). Using a) shows that Ker(C T ) = {0} so that Im(CR1 ) = Im(C) =
IRm . From Proposition 3.73 it follows that f (z) = R1 C T (CR1 z). Let
us prove that the system

z(t) RAR1 z(t) f (z(t))
(3.197)

z(0) = Rx(0)

has a unique AC solution on [0, +). First, to say that Cx(0) dom is
to say that CR1 z(0) dom , and this just means that z(0) dom f .
Second, it follows from the KYP Lemma that RAR1 + (RAR1 )T is nega-
tive denite. Therefore the multivalued mapping RAR1 + f is maximal
monotone [66, Lemma 2.4]. Consequently the existence and uniqueness result
follows from Theorem 3.74.
Now set x(t) = R1 z(t). It is straightforward to check that x(t) is a
solution of the system in (3.196). Actually the system in (3.197) is deduced
from (3.196) by the change of state vector z = Rx.

As an example, let us consider dissipative Linear Complementarity Sys-

tems (LCS) [83, 94]:

x(t) = Ax(t) + B(t)
(3.198)

0 y(t) = Cx(t) 0
where (A, B, C) satises a) and b) above, y(t), (t) IRm , and Cx(0) 0.
The second line in (3.198) is a set of complementarity conditions between y and
, stating that both these terms have to remain non-negative and orthogonal
one to each other. The LCS in (3.198) can be equivalently rewritten as in
(3.197) with (y) = (IR+ )m (y), noting that

0 y 0 (IR+ )m (y) (3.199)

which is a basic result in convex analysis, where () is the indicator function

in (3.187). Lemma 3.76is extended in [85] to the case of non-autonomous
systems with both locally AC and locally BV inputs, both in the linear and
nonlinear cases 4 . The non-autonomous case yields another, more complex,
type of dierential inclusion named rst order Moreaus sweeping process.

4
Linearity refers in this context to the vector elds, not to the system itself that
is nonlinear as it is unilaterally constrained.
150 3 Kalman-Yakubovich-Popov Lemma

Remark 3.77. Let usnote in passing that Lemma 3.76 applies to nonlinear
systems as x(t) = k=0 x2k+1 (t) yL (t), y = x, yL (y), x IR. Indeed
n
2
the dynamics yL
y is strictly dissipative with storage function V (x) = x2 ,
so that P = 1 and z = x.

Let us notice that y dom .

 Finally there exists a Lebesgue integrable
function w(t) such that x(t) = w( )d , where d is the Lebesgue measure.
Hence dx = w(t)dt as an equality of measures.

Lemma 3.78. [81] Let assumptions a)c) hold, the initial data be such that
Cx(0) dom , and assume that the graph of contains (0, 0). Then: i)
x = 0 is the unique solution of the generalized equation Ax B(Cx) ii)
The xed point x = 0 of the system in (3.196) is exponentially stable.

Proof: The proof of part i) is as follows. First of all notice that x = 0 is

indeed a xed point of the dynamics with no control, since 0 B(0).
Now Ax B(Cx) P Ax P B(Cx) xT P Ax = xT g(x), where
g(x) = (Cx) (use Proposition 3.73 to prove this), g() is convex as it is
the composition of a convex function with a linear mapping, and we used
assumption a). The multivalued mapping g(x) is monotone since g() is
convex. Thus xT g(x) 0 for all x IRn . Now there exists Q = QT > 0
such that xT P Ax = 12 xT Qx < 0 for all x = 0. Clearly then x satises the
generalized equation only if x = 0.
Let us now prove part ii). Consider the candidate Lyapunov function
W (x) = 12 xT P x. From Lemma 3.76 it follows that the dynamics in (3.196)
possesses on [0, +) a solution x(t) which is AC, and whose derivative x(t)
exists a.e.. The same applies to W () which is AC [421, p.189]. Dierentiating
along the closed-loop trajectories we get

d(W x) a.e.
dt (t) = xT (t)P w(t)

= xT (t)P (Ax(t) ByL (t)) = xT (t)Qx(t) xT (t)P ByL (t)

= xT (t)Qx(t) xT (t)C T yL (t)

(3.200)
where yL is any vector that belongs to (Cx). The equality in the rst
line means that the density of the measure d(W x) with respect to the
Lebesgue measure dt (which exists since W (x(t)) is AC) is the function
xT P w. Consequently d(Wdtx) + xT Qx xT C T (Cx) = xT g(x) a.e.,
where d(Wdtx) is computed along the systems trajectories. Let us consider
a.e.
any z g(x). One gets d(Wdtx) = xT Qx xT z xT Qx from the
property of monotone multivalued mappings and since (x, z) = (0, 0) be-
longs to the graph of g(x). The set of time instants at which the inequality
d(W x)
dt xT Qx is not satised is negligible in the Lebesgue measure. It
 3.9 The Lure Problem (Absolute Stability) 151

follows that the function of time W (), which is continuous, is nonincreasing.

t t
Thus one has W (t) W (0) = 0 (xT Qx xT z)d 0 xT Qxd . Conse-
t
quently 12 min (P )xT x W (0) 0 min (Q)xT xd , where min () is the small-
est eigenvalue.
 By the Gronwalls Lemma 3.68 one gets that 12 min (P )xT x
(Q)
W (0) exp 2 min
min (P )
t which concludes the proof.
It is worth noting that part i) of Lemma 3.78 is a particular case of gen-
eralized equation 0 F (x), where F () is a maximal monotone operator.
Example 3.79. Let us consider a one degree-of-freedom mechanical system
with Coulomb friction

mq(t) = sgn(q(t)) + u(t) (3.201)
where q(t) is the position of the system, is the friction coecient and the
control is given in Laplace transform by u(s) = H(s)q(s). Dening x1 = q
and x2 = q and u = q + q we obtain
 

0 1 0

x(t) = x(t) yL (t)


m

(3.202)

yL (t) |q(t)|





y(t) = x2 (t)
s
The transfer function of the triple (A, B, C) is G(s) = m s2 s , which
obviously cannot be SPR but only PR with a suitable choice of < 0 and
< 0; see Section 2.14. Thus more advanced tools will be needed to study
the asymptotic stability of (3.201); see Chapter 7, Section 7.2.5.
Dissipation inequality and storage function:
We consider the same inclusion as in (3.196) but with an input, i.e.
a.e.

x(t) = Ax(t) ByL (t) + Bu(t)



y(t) = Cx(t) (3.203)





yL (y)
It is then not dicult to calculate that

t t t
0 uT (s)y(s)ds = 0 uT (s)Cx(s)ds = 0 uT (s)B T P x(s)ds
t
= 0
(x(s) Ax(s) + ByL (s))T P x(s)ds

12 xT (t)P x(t) 12 xT (0)P x(0) = W (x(t)) W (x(0))

(3.204)
152 3 Kalman-Yakubovich-Popov Lemma

Therefore W () is a storage function for (3.203) that is smooth despite the

system is nonsmooth.We notice that if Bu(t) in (3.203) is replaced by Eu(t)
for some matrix E and with both (A, E, C) and (A, B, C) being PR, then the
above developments yield that W () is a storage function provided the two
triples have a set of KYP Lemma equations with the same solution P , so that
B T P = C.
Another kind of nonsmooth characteristic that does not t with the maxi-
mal monotone static nonlinearities can be found in [266] where the passivity of
an oscillator subject to a Preisach hysteresis is shown. The absolute stability
of systems with hysteresis non-linearities is also treated in [393].

3.9.5 Dissipative Evolution Variational Inequalities

Introduction

In this section we introduce a new formalism that is useful in many appli-

cations: evolution variational inequalities (in nite dimension). Let K IRn
be a nonempty closed convex set. Let F : IRn IRn be a nonlinear opera-
tor. For (t0 , x0 ) IR K, we consider the problem P (t0 , x0 ): Find a function
t x(t) (t t0 ) with x C 0 ([t0 , +); IRn ), dx
dt L,e ([t0 , +); IR ) and
n

such that:

x(t) K, t t0



 dx
dt (t) + F (x(t)), v x(t) 0, v K, a.e. t t0





x(t0 ) = x0
(3.205)

Here ., . denotes the euclidean scalar product in IRn . It follows from standard
convex analysis that (3.205) can be rewritten equivalently as the dierential
inclusion
dx
dt (t) + F (x(t)) NK (x(t))
(3.206)

x() K
where the denition of the normal cone to a set K is in (3.191). If K = {x |
Cx 0} the reader may use Proposition 3.73 together with (3.187), (3.190)
and (3.199) to deduce that (3.206) is the LCS
dx
dt (t) + F (x(t)) = C T (t)
(3.207)

0 Cx(t) (t) 0
 3.9 The Lure Problem (Absolute Stability) 153

Still, another formulation for (3.206) is as follows:

x(t)+ F (x(t), t), v x(t)+ (v) (x(t)) 0, v IRn , a.e. t 0 (3.208)

with (x) = K (x) and x(t) dom(), t 0, where dom() = {x

IRn | = } is the domain of the multivalued mapping . In general ()
is a proper convex and lower semi continuous function. One has dom()
dom() = {x IRn | (x) < +} and dom() = dom(): the two
domains dier only by the boundary. More on the equivalence between various
formalisms like the above ones can be found in [84]. The maximal monotone
property of operators is at the core of the equivalence. Let us give a well-
posedness result, which is one variant of the famous Katos Theorem [251].

Theorem 3.80. [167] Let K be a nonempty closed convex subset of IRn and
let A IRnn . Suppose that F : IRn IRn can be written as

F = F1 + 

where F1 is Lipschitz continuous, C 1 (IRn ; IR) is convex and  denotes

its derivative. Let t0 IR and x0 K be given. Then there exists a unique
x C 0 ([t0 , +); IRn ) such that
dx
L,e ([t0 , +); IRn ) (3.209)
dt
x is right-dierentiable on [t0 , +) (3.210)
x(t0 ) = x0 (3.211)
x(t) K, t t0 (3.212)
dx
 (t) + Ax(t) + F (x(t)), v x(t) 0, v K, a.e. t t0 (3.213)
dt

Suppose that the assumptions of Theorem 3.80 are satised and denote by
x(.; t0 , x0 ) the unique solution of Problem P (t0 , x0 ) in (3.205). Suppose now
in addition that
0K (3.214)
and
F (0) NK (0) (3.215)
that is
F (0), h 0, h K
Then
x(t; t0 , 0) = 0, t t0
i.e. the trivial solution 0 is the unique solution of problem P (t0 , 0).
154 3 Kalman-Yakubovich-Popov Lemma

Lyapunov Stability

Denition 3.81. The equilibrium point x = 0 is said to be stable in the sense

of Lyapunov if for every > 0 there exists = () > 0 such that for any
x0 K with x0  the solution x(; t0 , x0 ) of problem P (t0 , x0 ) satises
x(t; t0 , x0 ) , t t0 .

Denition 3.82. The equilibrium point x = 0 is asymptotically stable if it is

stable and there exists > 0 such that for any x0 K with x0  the
solution x(; t0 , x0 ) of problem P (t0 , x0 ) fullls limt+ x(t; t0 , x0 ) = 0.

We now give two Theorems inspired from [170] that guarantee that the
xed point of the systems is Lyapunov stable.

Theorem 3.83. [167] Suppose that the assumptions of Theorem 3.80 to-
gether with the condition (3.215) hold. Suppose that there exist > 0 and
V C 1 (IRn ; IR) such that
(1)
V (x) a(x), x K, x
with a : [0, ] IR satisfying a(t) > 0, t (0, )
(2) V (0) = 0
(3) x V  (x) K, x K, x
(4) Ax + F (x), V  (x) 0, x K, x
Then the trivial solution of (3.212) and (3.213) is stable.

Theorem 3.84. [167] Suppose that the assumptions of Theorem 3.80 to-
gether with the condition (3.215) hold. Suppose that there exist > 0, > 0
and V C 1 (IRn ; IR) such that
(1)
V (x) a(x), for all x K, x
with a : [0, ] IR satisfying a(t) ct , t [0, ], for some constants
c > 0, > 0
(2) V (0) = 0
(3) x V  (x) K, for all x K, x
(4) Ax + F (x), V  (x) V (x), for all x K, x
Then the trivial solution of (3.212) and (3.213) is asymptotically stable.

Copositive Matrices on a Set

We shall also need the denition of a number of sets of matrices.

Denition 3.85. [167] The matrix A IRnn is Lyapunov positive stable

on K if there exists a matrix P IRnn such that
 3.9 The Lure Problem (Absolute Stability) 155
P x,x
(1) inf xK\{0}
x
2 >0

(2) Ax, [P + P T ]x 0, x K
(3) x K [I [P + P T ]]x K

Denition 3.86. [167] The matrix A IRnn is Lyapunov positive strictly-

stable on K if there exists a matrix P IRnn such that
P x,x
(1) inf xK\{0}
x
2 >0
Ax,[P +P T ]x
(2) inf xK\{0}
x
2 >0

(3) x K [I [P + P T ]]x K

Remark 3.87. Condition (1) of Denitions 3.85 and 3.86 is equivalent to the
existence of a constant c > 0 such that

P x, x c  x 2 , x K (3.216)

Indeed, set
P x, x
C= inf
xK\{0} x2

If + > C > 0 then it is clear that (3.216) holds with c = C. If C = +

then necessarily K = {0} and the relation in (3.216) is trivial. On the other
hand, it is clear that if (3.216) holds then C c > 0.

Recall that a matrix P IRnn is said to be copositive on K if

P x, x 0, x K

A matrix P IRnn is said to be strictly copositive on K if

P x, x > 0, x K\{0}

These classes of matrices play an important role in complementarity theory

(see e.g. [137, 367]). The set of copositive matrices contains that of positive
semi denite (PSD) matrices [367, p.174]. Indeed a PSD matrix is necessarily
copositive on any set K. However it is easy to construct a matrix that is
copositive on a certain set K, but which is not PSD.
+
Let us here denote by PK (resp. PK ) the set of copositive (resp. strictly
++
copositive) matrices on K. Let us also denote by PK the set of matrices
satisfying condition (1) of Denition 3.85, that is

++
) Bx, x *
PK = B IRnn : inf 2
>0
xK\{0} x

It is clear that
156 3 Kalman-Yakubovich-Popov Lemma
++ +
PK PK PK
++ ++
K 1 K 2 PK 2
PK 1

Let us now denote by LK the set of Lyapunov positive stable matrices on

K and by L++K the set of Lyapunov positive strictly-stable matrices on K. We
see that
) ++
LK = A IRnn : P PK such that (I [P + P T ])(K) K
*
and P A + AT P PK
and
)
L++
K = A IR
nn ++
: P PK such that (I [P + P T ])(K) K
++
*
and P A + AT P PK
Let us note that P needs not be symmetric. In summary, the classical pos-
itive denite property of the solutions of the Lyapunov matrix inequality, is
replaced by the copositive denite property.

PR Evolution Variational Inequalities

To see how evolution variational inequalities are related to the systems in the
foregoing section, let us come back to the system in (3.196):
a.e.

x(t) = Ax(t) ByL (t)



y(t) = Cx(t) (3.217)





yL (y)

and let us assume that the convex function (y) is the indicator of a closed
convex set K IRn with 0 K. We therefore rewrite the problem as:
Find x C 0 ([0, ); IRn ) such that dx
dt L,e (0, +; IR ) and
n

dx
(t) = Ax(t) ByL (t), a.e. t 0 (3.218)
dt
y(t) = Cx(t) (3.219)
y(t) K (3.220)
yL (t) K (y(t)) (3.221)
x(0) = x0 (3.222)
Assume there exists a symmetric and invertible matrix R IR nn
such
that R2 C T = B. Suppose also that there exists
 3.9 The Lure Problem (Absolute Stability) 157

y0 = CR1 x0 Int(K).

(3.223)

Then using the change of state vector z = Rx and setting

K = {h IRn : CR1 h K} (3.224)

we see that problem (3.218) to (3.222) is equivalent to the following one: nd

z C 0 ([0, ); IRn ) such that dz
dt L,e ([0, ); IR ) and
n

dz
 (t) RAR1 z(t), v z(t) 0, v K, a.e. t 0 (3.225)
dt

z(t) K, t 0
z(0) = Rx0
Indeed, it suces to remark that

Cx K z K

x(0) = x0 z(0) = Rx0

and
dx dx
Ax BK (Cx) R RAR1 Rx RBK (CR1 Rx)
dt dt
dz
RAR1 z R1 R2 BK (CR1 z)
dt
dz
RAR1 z R1 C T K (CR1 z)
dt
dz
RAR1 z K (z)
dt
Indeed, K (z) = (K CR1 )(z) and thanks to (3.223) we obtain K (z) =
R1 C T K (CR1 z). We remark also that the set K is closed convex with
0 K. The variable change z = Rx is exactly the same as the variable change
used in Lemma 3.76. The following holds:

Lemma 3.88. [167] Let K IRn be a closed convex set containing x = 0, and
satisfying the condition (3.223). Dene K as in (3.224). Suppose that there
exists a symmetric and invertible matrix R IRnn such that R2 C T = B.
i) If RAR1 LK then the trivial equilibrium point of (3.218)(3.221) is
stable.
ii) If RAR1 L++ K
then then the trivial equilibrium point of (3.218)
(3.221) is asymptotically stable.
158 3 Kalman-Yakubovich-Popov Lemma

Example 3.89. Positive real evolution variational inequalities Assume

that G(s) = C(sIA)1 B, with (A, B, C) a minimal representation, is strictly
positive real. From the Kalman-Yakubovitch-Popov Lemma there exist P =
P T positive denite and Q = QT positive denite such that P A + AT P = Q
and B T P = C. Choosing R as the symmetric square root of P , i.e. R = RT , R
positive denite and R2 = P , we see that B T R2 = C and thus R2 C T = B.
Moreover
P Ax, x + AT P x, x = Qx, x, x IRn (3.226)
Thus
1
Ax, P x = Qx, x, x IRn (3.227)
2
It results that
RAx, Rx > 0, x IRn \{0} (3.228)
Setting z = Rx, we see that

RAR1 z, z > 0, z IRn \{0} (3.229)

So RAR1 PI++ ++ ++
Rn PK LK . All the conditions of Lemma 3.88 (part
ii)) are satised and the trivial solution of (3.218)(3.221) is asymptotically
stable. The results presented in the foregoing section are here recovered. In
case G(s) is positive real then Lemma 3.88 (part i)) applies. As shown above
(see Lemma 3.78) the equilibrium point is unique in this case.

Example 3.90. PR electrical circuit The following example is taken from

[82].

Fig. 3.9. A circuit with ideal diodes

 3.9 The Lure Problem (Absolute Stability) 159

Let us consider the circuit in Figure 3.9 (R1 , R2 , R3 0, L2 , L3 > 0). One
has 0 uD4 x2 0 and 0 uD1 x3 + x2 0, where uD4 and uD1
are the voltages of the diodes. The dynamical equations are

x1 (t) = x2 (t)



 


R1 +R3 1 1 1
x2 (t) + RL3 x3 (t) L3 C4 x1 (t) +
1
x2 (t) = L3 1 (t) + L3 2 (t)


L3

 
x3 (t) = R1L+R 1
x3 (t) + RL2 x2 (t) L2 1 (t)
2 1

2



 



1 (t) x3 (t) + x2 (t)
0 0
2 (t) x2 (t)
(3.230)
where x1 () is the time integral of the current across the capacitor, x2 () is
the current across the capacitor, and x3 () is the current across the inductor
L2 and resistor R2 , 1 is the voltage of the diode D1 and 2 is the voltage
of the diode D4 . The system in (3.230) can be written compactly as the LCS:
x(t) = Ax(t) + B(t), 0 (t) y(t) = Cx(t) 0, with

0 1 0
+R
A = L3 C4 1L3 3 L31
1
R R
R1 +R2
0 R1
L2 L2

0 0 
0 1 1
B = L13 L13 , C =
010
L12 0
The monotonicity (consequently the passivity) of the voltage-current relation
0 u i 0 at the poles of the diodes is certainly an essential property both
for existence and uniqueness of solutions, and for stability. We recall that this
relation is a multivalued mapping whose graph is as in Figure 3.8 (c). We set
1
C4 0 0
P = 0 L3 0
0 0 L2

It is clear that P is symmetric and positive denite. Moreover, we see that

AT P + P A = Q with

0 0 0
Q = 0 2(R1 + R3 ) 2R1
0 2R1 2(R1 + R2 )

The matrix Q IR33 is symmetric and positive semi-denite. Moreover,

P B = C T and the system in (3.230) is positive real, as expected from the
160 3 Kalman-Yakubovich-Popov Lemma

physics. We deduce that (3.230) can be rewritten as an evolution variational

inequality

dz
 dt (t) RAR1 z(t), v z(t) 0, v K, a.e. t 0
(3.231)

z(t) K, t 0

where z = Rx, R is a symmetric positive denite square root of P and K =

{h IRn : CR1 h K}. The change of state matrix R andthe new state
vector z are easily calculated (z1 = 1C x1 , z2 = L3 x2 , z3 = L2 x3 ).
4

It follows from the above that an extension of the KYP Lemma matrix
inequalities to linear evolution variational inequalities is possible at the price
of replacing positive deniteness by copositive deniteness of matrices. How-
ever what remains unclear is the link with frequency-domain conditions. In
other words, we have shown that if the triple (A, B, C) is PR (or SPR), then
it satises the requirements for the evolution variational inequality in (3.225)
to possess a Lyapunov stable equilibrium. Is the converse provable? Certainly
the answer is negative, as some examples show that the matrix A can be un-
stable (with eigenvalues with positive real parts) while A L++ K (thus the
corresponding evolution variational inequality has an asymptotically stable
xed point). Extension of the Krasovskii-LaSalle invariance principle to evo-
lution variational inequalities, has been considered in [82]. In Chapter 6, we
shall examine second order evolution variational inequalities, which arise in
some problems of mechanics with nonsmooth contact laws.

3.10 The Circle Criterion

Let us come back to the Lure problem with single-valued nonlinearities in the
feedback loop. Consider the observable and controllable system in (3.181). Its
transfer function H (s) is
1
H (s) = C (sIn A) B+D (3.232)
Assume that the transfer function H(s) is SPR and is connected in negative
feedback with a nonlinearity () as illustrated in Figure 3.10. The conditions
for stability of such a scheme are stated in the following Theorem.

Theorem 3.91. Consider the system in Figure 3.10. If H (s) in (3.232) is

SPR, the conditions of Lemma 3.59 are satised and if (t, y) is in the sector
[0, ), i.e.:
i) (t, 0) = 0 t0
ii) y T (t, y) 0 t 0, y IRm
then the origin is a globally exponentially stable equilibrium point.
 3.10 The Circle Criterion 161

Fig. 3.10. Linear system with a sector nonlinearity in negative feedback

Proof: Since H (s) = C (sI A)1 B + D is SPR, then there exist P > 0,
Q and W, > 0 such that
T
A P + P A = P QT Q
BT P + W T Q = C (3.233)
T
W W = D + DT
Dene the Lyapunov function candidate V (x) = xT P x. Then

V (x(t)) = xT (t)P x(t) + xT (t)P x(t)

T
= [Ax(t) B(t, y(t))] P x(t) + xT (t)P [Ax(t) B(t, y(t))]
 
= xT (t) AT P + P A x(t) T (t, y(t))B T P x(t) xT (t)P B(t, y(t))
(3.234)
Note that B T P = C W T Q. Hence, using the above, (3.181) and the
control u = (t, y), we get

xT (t)P B(t, y(t)) = T (t, y(t))B T P x(t)

= T (t, y(t))Cx(t) T (t, y(t))W T Qx(t)

= T (t, y(t)) [y(t) Du(t)] T (t, y(t))W T Qx(t)

= T (t, y(t)) [y(t) + D(t, y(t))] T (t, y(t))W T Qx(t).

Substituting the above into (3.234) we get
 
V (x(t)) = xT (t)P x(t) xT (t)QT Qx(t) T (t) D + DT (t)

T (t)W T Qx(t) xT (t)QT W (t) T (t)y(t) y T (t)(t)

162 3 Kalman-Yakubovich-Popov Lemma

Using (3.233) and the fact that y T 0 we have

V (x(t)) xT (t)P x(t) xT (t)QT Qx(t) T (t, y(t))W T W (t, y(t))

(t, y(t))T W T Qx xT QT W (t, y(t))

T
= xT (t)P x(t) [Qx(t) + W (t, y(t))] [Qx(t) + W (t, y(t))]

xT (t)P x(t)

T
Dene z(t) = [Qx(t) + W (t, y(t))] [Qx(t) + W (t, y(t))] which can
also be rewritten as V (x(t)) = V (xt)) + z(t)
Thus
t
V (x(t)) = et V (0) + e(t ) z ( ) d
0

et V (0)
Finally the xed point x = 0 is globally exponentially stable.

3.10.1 Loop Transformations

The above theorem applies when (, ) belongs to the sector [0, ). In order
to use the above result when (, ) belongs to the sector [a, b] we have to make
some loop transformations which are given next.

Fig. 3.11. Loop transformations

1) If (, ) belongs to the sector [a, b] then 1 = (t, y) a belongs to the
sector [0, b a] . This is illustrated in Figure 3.11.
 3.10 The Circle Criterion 163

2) If 1 (, ) belongs to the sector [0, c] with c = b a then we can make the

transformation indicated in Figure 3.12 where y = 2 (t, u) and > 0 is
arbitrarily small number. Therefore, as is shown next, 2 () belongs to the
sector [0, ).

Fig. 3.12. Loop transformations

Note that if 1 = c, then

c c (c )
y = c = u
1 c c c

Therefore:
1. if c = c, lim y
=
0 u
y
2. if c = 0, u = 0

Using the two transformations described above, the system in Figure 3.10
can be transformed into the system in Figure 3.13. We then have the following
corollary:
Corollary 3.92. If H2 in Figure 3.13 is SPR and the nonlinearity (, ) be-
longs to the sector [0, ) then the closed-loop system is globally exponentially
stable.
Note that H2 is SPR if and only if
2I
H1 (j) + H1 (j) + >0
ba
1
with H1 (s) = H (s) [I + aH (s)] and << 1. For m = 1 the above result
has a graphical interpretation which leads to the circle criterion. Suppose
164 3 Kalman-Yakubovich-Popov Lemma

Fig. 3.13. Loop transformations

z = x + jy is a complex number and a, b IR with a < b, a = 0. Consider the

condition

8
z 1
= Re + >0
1 + az ba
Now one has
z 1 x+jy 1
1+az + ba = 1+a(x+jy) + ba

x+jy[1+axjay] 1
= (1+ax)2 +y 2 a2
+ ba

Therefore
x(1+ax)+ay 2 1
= (1+ax)2 +y 2 a2
+ ba >0
or equivalently
) * 2
0 < (b a) x (1 + ax) + ay 2 + (1 + ax) + y 2 a2
) *
= (b a) x + ax2 + ay 2 + 1 + 2ax + a2 x2 + a2 y 2 (3.235)
) *
= ba x2 + y 2 + x (b + a) + 1
 3.10 The Circle Criterion 165

which implies
2
2 a+b (a + b)2
bay + ba x + +1 >0
2ab 4ab
Note that

(a + b)2 4ab a2 2ab b2 (a b)2

1 = =
4a2 b2 4ab 4ab
Introducing the above into (3.235) we get
2 2
2 a+b (a b)
bay + ba x + >
2ab 4ab
If ab > 0 this can be written as
2 2
2 a+b (a b)
y + ba x + >
2ab 4a2 b2
or
 
 
z + a + b  > |a b|
 2ab  2 |ab|
If ab < 0 then
 
 
z + a + b  < |a b|
 2ab  2 |ab|
a+b
Let D(a, b) denote the closed disc in the complex plane centered at 2ab
and with radius |ab|
2|ab| . Then

8
z 1
Re + >0
1 + az b a
if and only if  
 
z + a + b  > |a b| , ab > 0
 2ab  2 |ab|
In other words, the complex number z lies outside the disc D(a, b) in case
ab > 0 and lies in the interior of the disc D(a, b) in case ab < 0. We therefore
have the following important result.
166 3 Kalman-Yakubovich-Popov Lemma

Theorem 3.93 (Circle criterion). Consider again the system for

m=1 in Figure 3.13. The closed loop system is globally exponentially
stable if:
(i) 0 < a < b : The plot of h(j) lies outside and is bounded away
from the disc D(a, b). Moreover the plot encircles D(a, b) exactly
times in the counter-clockwise direction, where is the number
of eigenvalues of A with positive real part.
(ii) 0 = a < b : A is a Hurwitz matrix and

8
1
Re H(j) + >0 (3.236)
b

(iii) a < 0 < b : A is a Hurwitz matrix; the plot of h(j) lies in the
interior of the disc D(a, b) and is bounded away from the circum-
ference of D(a, b).
(iv) a < b 0 : Replace h(.) by h(.), a by b, b by a and apply (i)
or (ii) as appropriate.

Remark 3.94. If b a 0 the critical disc D(a, b) in case (i) shrinks to

the critical point0 1/a of the Nyquist criterion. The circle criterion is
applicable to time-varying and/or nonlinear systems, whereas the Nyquist
criterion is only applicable to linear time invariant systems.
A generalization of the circle criterion for the design of a nite-dimensional
controller for unstable innite-dimensional systems, has been proposed in
[509]. The case of an innite-dimensional linear system, illustrated by an elec-
trical transmission line, is considered in [172].

3.11 The Popov Criterion

Unlike the circle criterion, the Popov criterion [406408] is applicable only to
autonomous single input-single output (SISO) systems:


x(t) = Ax(t) + bu(t)

(t) = u(t)

y(t) = cx(t) + d(t)

u(t) = (y(t))
where u(t), y(t) IR, : IR IR is a time-invariant nonlinearity belonging
to the open sector (0, ), i.e.

(0) = 0, y(y) > 0, y = 0

 3.11 The Popov Criterion 167

The linear part can also be written as:

   
x(t) A0 x(t) b
= + u
(t) 0 0 (t) 1
 (3.237)
  x(t)
y(t) = cd
(t)

Hence the transfer function is

d
h(s) = + c(sI A)1 b
s
which has a pole at the origin. We can now state the following result:

Theorem 3.95 (Popovs criterion). Consider the system in

(3.237). Assume that
1. A is Hurwitz
2. (A, b) is controllable
3. (c, A) is observable
4. d>0
5. () belongs to the sector (0, )
Then the system is globally asymptotically stable if there exixts r > 0
such that Re[(1 + jr)h(j)] > 0, IR.

Remark 3.96. Contrary to Popovs criterion, the circle criterion does not apply
to systems with a pole at s = 0 and () belongs to the sector (0, ).

Proof of Popovs criterion: Note that

s(sI A)1 = (sI A + A)(sI A)1

= I + A(sI A)1
Hence
 
1
(1 + rs)h(s) = (1 + rs) d
s + c (sI A) b
= d
s+ rd + c(sI A)1 b
+rcb + rcA(sI A)1 b
d
Note that j is purely imaginary. From the above and by assumption we
have
168 3 Kalman-Yakubovich-Popov Lemma

 
Re [(1 + jr)h(j)] = Re r (d + cb) + c (I + rA) (j A)1 b > 0

Dene the transfer function

1
g (s) = r (d + cb) + c (I + rA) (sI A) b

i.e. {A, b, c (I + rA) , r (d + cb)} is a minimal realization of g(s). If Re [g()] >

0 then there exists P > 0, q and and > 0 such that
T
A P + P A = P q T q
bT P + q = c(I + rA)
2
= 2r(d + cb)
Choose the Lyapunov function candidate
y
V (x, ) = xT P x + d 2 + 2r 0 () d

Given that () belongs to the sector [0, ) it then follows that () d
0. Hence V (x, ) is positive denite and radially unbounded
.T
V (x, ) = x P x + xT P x + 2d + 2r(y)y

= (Ax b)T P x + xT P (Ax b)

2d + 2r [c (Ax b) d]
Note from (3.237) that d = y cx, thus

V (x(t), (t)) = xT (t)(AT P + P A)x(t) 2(y(t))bT P x(t)+

+2(y(t))c(I + rA)x(t) 2r (d + cb) 2 (y(t)) 2y(t)(y(t))

= xT (t)P x(t) (qx(t) (y(t)))2

r (d + cb) 2 (y(t)) 2y(t)(y(t))

Since g(j) r(d + cb) as it follows that r(d + cb) > 0. Hence

V (x(t), (t)) xT (t)P x(t) 2y(t)(y(t)) 0, x IRn , > 0

We now show that V (x, ) < 0 if (x, ) = (0, 0) . If x = 0 then V (x, ) < 0
since P > 0. If x = 0 but = 0, then y = d = 0, and y > 0 since () belongs
to the sector [0, ). Therefore the system (3.237) is globally asymptotically
stable.
 3.11 The Popov Criterion 169

Corollary 3.97. Suppose now that () belongs to the sector (0, k) , k > 0.
Then the system is globally asymptotically stable if there exists r > 0 such
that
1
Re [(1 + jr)h(j)] + >0 (3.238)
k

Proof: It follows from the loop transformation in Figure 3.14, where

Fig. 3.14. Loop transformations

 1

= I k1
1




1
g1 = g(s) + k

1

= (1 + jr)(hj) +


k


1
Re(g1 ) = Re[h(j)] + rIm[h(j)] + k > 0.

Remark 3.98. The circle and the Popovs criteria owe their great success to
the fact that they lend themselves to graphical interpretations as pointed out
170 3 Kalman-Yakubovich-Popov Lemma

above for the circle criterion. Consider for instance the inequality in (3.238).
Consider the function M (j) = Re[h(j)] + jIm[h(j)], > 0. Note that
Re[(1+jr)h(j)] = Re[h(j)]rIm[h(j)] = Re[M (j)]rIm[M (j))].
Then condition (3.238) means that there must exist a straight line with an
arbitrary, xed slope, passing through the point k1 , 0 in the complex plane,
such that the plot of M (j) lies to the right of this line. The slope of this
line which is tangent to the plot of M (j) is equal to 1r . The line is usually
called the Popovs line. In the multivariable case the graphical interpretation
becomes too complex to remain interesting; see [417].

Further reading: The circle criterion has been introduced in [431, 532,
533] and generalized after. Further results on the absolute stability problem
and Popovs criterion, can be found in [56, 102, 136, 166, 181, 182, 196, 200, 212,
217, 218, 220, 221, 258, 265, 293, 335, 336, 369, 382, 395, 456, 487, 503, 538]. These
references constitute only a few of all the works that have been published on
the topic. The reader is also referred to Section 5.10 on hyperstability. It is also
worth reading the European Journal of Control special issue dedicated to V.M.
Popov [134]. Generalization of the Popov criterion with Popov multipliers
can be found in [48, 190, 244]. An interesting comparative study between the
cicle criterion, the Popov criterion, and the small gain Theorem, has been led
in [193] on a 4th order spring-mass system with uncertain stiness. The result
in terms of conservativeness is that the Popov criterion design supersedes the
circle criterion design and that the small gain design is the most conservative
one.

3.12 Discrete-time Systems

3.12.1 The KYP Lemma

In this section we investigate how the KYP Lemma may be extended to

discrete-time systems of the following form:

x(k + 1) = Ax(k) + Bu(k)
(3.239)

y(k) = Cx(k) + Du(k)
with x(k) IRn , u(k) IRm , y(k) IRm . The KYP Lemma for systems as
(3.239) is due to [211, 483].

Denition 3.99. A discrete transfer matrix H(z) is positive real if

H(z) has analytic elements in |z| > 1, z C
H(z) + H
(z) 0 in |z| > 1
A discrete transfer matrix H(z) is strictly positive real if
H(z) has analytic elements in |z| > 1
 3.12 Discrete-time Systems 171

H(ej ) + H
(ej ) > 0 for [0, 2)

A discrete transfer matrix H(z) is strongly strictly positive real if it is SPR
and H() + GT () > 0.

It is noteworthy that the condition H(z) + H
(z) 0 in |z| > 1 implies

that H T (ej ) + H(ej ) 0 for all real such that no element of H(z) has a
pole at z = ej .

Lemma 3.100. Let H(z) = C(zIn A)1 B + D be a square matrix of real

rational functions of z, with no poles in |z| < 1. Let (A, B, C, D) be a minimal
realization of H(z). If for (A, B, C, D) there exist a real symmetric positive
denite matrix P and real matrices L and W such that
T
A P A P = L L
T



AT P B = C T LW (3.240)




T
W W = D + DT B T P B
then the transfer function H(z) is positive real.

Similarly to their continuous-time counterpart, the KYP Lemma con-

ditions can be written as an LMI, using for instance Proposition A.63.
One immediately notices from (3.240) that necessarily D = 0, otherwise
W T W = B T P B (and obviously we assume that B = 0). If B has full
rank m, then D must have full rank m so that D + DT > 0. Therefore a
positive real discrete time system with full rank input matrix has a relative
degree 0. Consequently in the monovariable case the relative degree is always
zero. However it is worth noting that this is true for passive systems only, i.e.
systems which are dissipative with respect to the supply rate w(u, y) = uT y.
If a more general supply rate is used, e.g. w(u, y) = uT Ru + 2uT Sy + y T Qy,
then the relative degree may not be zero.
When W = 0 and L = 0 in (3.240) the system is said lossless. Then
1 T 1
x (k + 1)P x(k + 1) xT (k)P x(k) = y T (k)u(k) (3.241)
2 2
for all u(k) and k 0, which in turn is equivalent to

1 T 1  k
x (k + 1)P x(k + 1) xT (0)P x(0) = y T (i)u(i) (3.242)
2 2 i=0

for all x(0) and k 0. Let us now formulate a KYP Lemma for SPR functions.

Lemma 3.101. [93, 250] Let (A, B, C, D) be a minimal realization of H(z).

The transfer matrix H(z) is SPR if and only if there exist matrices P = P T >
0, L and W such that
172 3 Kalman-Yakubovich-Popov Lemma

P = AT P A + LT L



0 = BT P A C + W T L (3.243)





0 = D + DT B T P B W T W

is satised, the pair (A, L) is observable, and rank(H(z)) = m for z = ej ,

IR, where (A, B, L, W ) is a minimal realization of H(z).

Similarly to the continuous time case, PR systems possess stable zeroes.

Let us assume that D is full rank. Then the zero dynamics is given by

A0 x(k) = (A BD1 C)x(k) (3.244)

which exactly is the dynamics on the subspace y(k) = 0. Then we have the
following result:

Proposition 3.102. [373] Let the system (3.239) be passive. Then the zero
dynamics exists and is passive.

Proof: Let us recall that passivity means that the system satises

V (x(k + 1)) V (x(k)) uT (k)y(k) (3.245)

along its trajectories, with V (x) = 12 xT P x and P
is the solution of the KYP
Lemma LMI in (3.240). One has V (A0 x) V (x)= xT M x, with M = (A
1 1
BD C) P (A BD C) P . If M 0 then the zero dynamics is stable.
T

Using the second equality of the KYP Lemma conditions, one obtains

M = (AT P A P ) C T [D1 + DT ]C + LW D1 C+
(3.246)
+(LW D1 C)T + C T DT B T P BD1 C

Using the equality C T DT (DT + D)D1 C = C T [D1 + DT ]C and using

the third equality of the KYP Lemma conditions(3.240), one gets

M = (AT P A P ) + LW D1 C + (LW D1 C)T (D1 C)T W T W (D1 C)

= (AT P A P ) [L (D1 C)T W T ][L (D1 C)T W T ]T + LLT

(3.247)
Injecting the rst matrix equality in (3.240) one concludes that M 0.
Therefore PR systems have a stable zero dynamics.
Positive real discrete-time transfer functions have proved to be quite use-
ful for identication; see [277, 278, 306]. In particular the so-called Landaus
scheme of recursive identication [278] is based on PRness. Further works can
 3.12 Discrete-time Systems 173

be found in [51,90,182,193,249,329,362,373,374,542]. Innite dimensional dis-

crete time systems and the KYP Lemma extension have been studied in [29].
The time-varying case received attention in [140, 141, 145]. In relation to the
relative degree zero property pointed out above, let us state the following
result:

Lemma 3.103. [326] Let H : IRn IRn be a linear operator (possibly time-
varying and unstable). Suppose that H is strictly causal, i.e.: if x(k) = 0 for
all 0 k n 1 then H(x(k)) = 0 for all 0 k n. Then H is passive if
and only if H = 0.

Passivity means here that nk=0 xT (k)H(x(k)) 0 for all n IN and
all real-valued sequences {x(k)}k0 . Applications of passivity in discrete-time
systems may be found in [112] for the design of repetitive controllers and
in [109] for haptic interfaces. The discrete passivity inequality has also been
used in the setting of time-discretised dierential inclusions where it proves
to be a crucial property for the behaviour of the numerical algorithms [2] (see
also [338] in the nonlinear framework of Lagrangian systems).

3.12.2 The Tsypkin Criterion

The Tsypkin criterion may be considered as the extension of Popovs and the
circle criteria, for discrete time systems. It was introduced in [492496]. For a
discrete-time system of the form

x(k + 1) = Ax(k) B(Cx, k) (3.248)

Tsypkin proved the absolute stability (i.e. the global asymptotic stability for
all (, ) in the sector (0, )) if the poles of the transfer function H(z) =
C(sIn A)1 B lie inside the unit disk and
1
Re[H(z)] + 0 for |z| = 1 (3.249)

This is the discrete-time analog of the circle criterion. When () is time
invariant and monotone, absolute stability holds if there exists a constant
0 such that
1
Re[(1 + (1 z 1 ))H(z)] + 0 for all |z| = 1 (3.250)

This is the discrete time analog of the Popovs criterion.

We present now the multivariable extension of Tsypkins result [250]. Let

us consider a minimal realization (A, B, C) of the transfer function H(z). The
discrete time system with a nonlinearity in the feedback is
174 3 Kalman-Yakubovich-Popov Lemma

x(k + 1) = Ax(k) B(y(k))
(3.251)

y(k) = Cx(k)
The nonlinearity is described as follows. Let M = M T > 0 be m m real
matrix. The set  () is

= { : IRm IRm such that T (y)[M 1 (y) y] < 0

for y IRm , y = 0, () is continuous

(3.252)
(y) = [1 (y1 ), 2 (y2 ), ...., m (ym )]T , and

i ()i ()
0< , IR, IR, = , i = 1, ..., m}

When m = 1 then we get the usual sector condition 0 < (y)y < M y 2 .
We also dene the matrices

A 0nm
Aa =
C 0m

B
Ba =
0m

Ca = [C Im ]

S = [C 0m ]
where Om denotes the zero m m matrix.

Theorem 3.104. [250] Let (A, B, C) be minimal, N =diag[N1 , ..., Nm ] be

positive denite, and assume that det(CA1 B) = 0, and that (A, C + N C
N CA1 ) is observable. Then

H(z) = M 1 + [Im + (1 z 1 )N ]H(z) (3.253)

T
is SPR if and only if there exist matrices P = P > 0, L and W such that


P = ATa P Aa + LT L



0 = BaT P Aa N Ca S + W T L (3.254)





0 = 2M 1 BaT P Ba W T W
Then the following function
 3.12 Discrete-time Systems 175
 m
yi

x
V (x) = [xT y T ]P +2 Ni i ()d (3.255)
y 0i=1

where yi = Ci x, Ci denotes the ith row of C, is a Lyapunov function for the

negative feedback interconnection of H(z) and the nonlinearity (), whose
exd point is globally asymptotically stable for all () .

Further details on the Tsypkin criterion can be found in [281] and in the
special issue [222]. See also [197, 198].

3.12.3 Discretization of PR Systems

In this section we are interested in a problem with a high practical interest:

given a PR or SPR continuous time system, is PRness preserved through time
discretization? The material is taken from De La Sen [447]. Let us start by
recalling some facts and denitions.
N (s)
Consider the transfer function H(s) = M(s) = H1 (s)+d, where the relative
1 (s)
degree of H(s) is 0, d IR and H1 (s) = N M(s) . H1 (s) is strictly proper. The
system is assumed to be stabilizable and detectable, i.e. N (s) = N1 (s)+dM (s)
and M (s) may possess common factors in the complex half plane Re[s] < 0.
Let (A, B, C, D) be a state representation of H(s). One has M (s) =det(sIn
A) and N (s) = CAdj(sIn A)B + D det(sIn A), where Adj() is the
adjoint matrix of the square matrix (). If M (s) and N (s) are coprime then
(A, B, C, D) is minimal (controllable and observable) but by assumption if
they are not coprime the uncontrollable or unobservable modes are stable.
We assume that the system is sampled with a zero-order hold device of
sampling period Ts , and we denote tk = kTs , xk = x(tk ) and so on. The
continuous time system (A, B, C, D) becomes when discretized a discrete time
system

xk+1 = xk + uk
(3.256)

yk+1 = Cxk+1 + Duk+1
 
T
for all k 0, k IN , = exp(ATs ), = 0 s exp(A(Ts ))d B. The
discrete transfer function from u(z) to y(z), z C, is given by

 
Nd (z) 1exp(Ts s) N1d (z)
G(z) = Md (z) =Z s H(s) = G1 (z) + D, G1 (z) = Md (z)
(3.257)
where G1 (z) has relative degree 1 and real coecients

N1d (z) = CAdj(zIn )
n (3.258)

Md (z) = det(zIn ) = z n + i=1 mi z
ni
,
176 3 Kalman-Yakubovich-Popov Lemma
n1 n1i 
where Adj(zIn ) = i=0 k=0 s k z nk1
i , n is the dimension of
the state vector x, Nd (z) = N1d (z) + DM (z), the degree of the polynomial
N1d is n 1 and the degree of Nd and Md is n. It is well-known that the poles
of G(z) and of G1 (z) are equal to exp(A Ts ) for each eigenvalue A of the
matrix A, so that the stability is preserved through discretization. However
such is not the same for the zeros of G1 (z) which depend on the zeros and the
poles of H1 (s), and on the sampling period Ts . It cannot be guaranteed that
these zeros are in |z| < 1. It is therefore clear that the preservation of PRness
imposes further conditions.
Let us denote H0 the set of stable transfer functions, possibly critically
stable (i.e. with pairs of purely imaginary conjugate poles). Let us denote G1
the set of discrete stable transfer functions, possible critically stable.

Theorem 3.105. Consider H1 (s) H0 with a numerator N1 (s) of degree

n 1, fullling the following conditions:

H1 (s) has a nonempty set of critically stable poles Ch with at most one
simple pole at s = 0, and any number N 0 of simple critically stable
complex conjugate poles s = jsi (i = 1, 2, ..., N0 , N = 2N0 ).
The residuals for all the critically stable poles are real and nonnegative.
Consider
  H(s) = H1 (s) + d , its discretized transfer function G(z) =
z1 H(s)
z Z = G1 (z) + D, and its transformed transfer function Gz (w) =
 s

G z = 1+w

1w . Then the following hold:

(i) G1 G1 (equivalently G1 z H0 ) for all suciently large absolute

Ts s1
values of D, provided that 2 < Arg(Gz (w)) < 2 for w = eeTs s+1 for all

s Ch .
(ii) If (i) holds then there is a constant D > 0 such that for all D D,
G(z) is (discrete) positive real and Gz (w) is (continuous) positive real.

It is interesting to note that (ii) is directly related to the comment made

right after the KYP Lemma 3.100. The homographic transformation w = z1 z+1
transforms the region |z| 1 into Re[w] 0, consequently the stability of
Gz (w) follows if all its poles are inside Re[w] 0.
4
Dissipative Systems

In this chapter we will further study the concept of dissipative systems which
is a very useful tool in the analysis and synthesis of control laws for linear
and nonlinear dynamical systems. One of the key properties of a dissipative
dynamical system is that the total energy stored in the system decreases with
time. Dissipativeness can be considered as an extension of PR systems to
the nonlinear case. Some relationships between Positive Real and Passive sys-
tems have been established in Chapter 2. There exist several important sub-
classes of dissipative nonlinear systems with slightly dierent properties which
are important in the analysis. Dissipativity is useful in stabilizing mechani-
cal systems like fully actuated robots manipulators [71], robots with exible
joints [6, 72, 78, 80, 318], underactuated robot manipulators, electric motors,
robotic manipulation [25], learning control of manipulators [26,27], fully actu-
ated and underactuated satellites [133], combustion engines [176], power con-
verters [18, 135, 234, 235, 458, 460], neural networks [122, 203, 528, 529], smart
actuators [171], piezo-electric structures [269], haptic environments and inter-
faces [109,128,284,285,289,309,333,422,423,454], particulate processes [131],
process and chemical systems [108, 152, 457, 459, 525], missile guidance [283],
model helicopters [332], magnetically levitated shafts [355,356], biological and
physiological systems [191, 192], at glass manufacture [526], visual feedback
control [252], etc. Some of these examples will be presented in the following
chapters.
Dissipative systems theory is intimately linked to Lyapunov stability the-
ory. There exists tools from the dissipativity approach that can be used to
generate Lyapunov functions. A dierence between the two approaches is that
the state of the system and the equilibrium point are notions that are required
in the Lyapunov approach while the dissipative approach is rather based on
input-output behavior of the plant. The input-output properties of a closed
loop system can be studied using Lp stability analysis. The properties of Lp
signals can then be used to analyze the stability of a closed loop control sys-
tem. Lp stability analysis has been studied by Desoer and Vidyasagar [125].
A clear presentation of this notions will also be given in this book since they
178 4 Dissipative Systems

are very useful in the stability analysis of control systems and in particular
in the control of robot manipulators. Popov introduced in 1964 the notion of
hyperstability which will be dened precisely in Section 5.10 and is in fact
quite close to dissipativity. This together with the celebrated Popovs crite-
rion for absolute stability, Popov multipliers [244], the Popov controllability
criterion, Popov parameters [246], certainly places V.M. Popov as one of the
major contributors in dissipative systems and modern control theories. As
quoted from [153]: V.M. Popov was the rst who studied passivity in detail
for linear control systems and gave its characterization in terms of frequency
domain inequality meaning positive realness of the system. Dissipativeness of
dynamical systems as it is known in the modern Systems and Control com-
munity has been introduced by Willems [510, 511]. Hill and Moylan [206, 207]
carried out an extension of the Kalman-Yakubovich-Popov (KYP) Lemma to
the case of nonlinear systems with state space representations that are ane
in the input. Byrnes et al. [89] further developed the concept of dissipative
systems and characterized the class of dissipative systems by obtaining some
necessary conditions for a nonlinear system to be dissipative and studied the
stabilization of dissipative systems.
Before presenting the denitions of dissipative systems we will study some
properties of Lp signals which will be useful in studying the stability of closed
loop control systems.

4.1 Normed Spaces

We will briey review next the notation and denitions of normed spaces,
Lp norms and properties of Lp signals. For a more complete presentation
the reader is referred to [125] or any monograph on mathematical analysis
[419421]. Let E be a linear space over the eld K (typically K is IR or the
complex eld C). The function (.), : E IR+ is a norm on E if and only
if:
1. x E and x = 0 (x) > 0, (0) = 0
2. (x) = ||(x), K, x E
3. (x + y) (x) + (y), x, y E (triangle inequality)

4.2 Lp Norms

Let x : IR IR be a function, and let || denote the absolute value. The most
common signal norms are the L1 , L2 , Lp and L norms which are respectively
dened as
 
|| x ||1 = |x(t)| dt
  1
|| x ||2 = |x(t)|2 dt 2
 4.2 Lp Norms 179
   p1
|| x ||p = |x(t)|p dt for 2 p < +

|| x || = ess sup |x(t)|dt
tIR

= inf{a| |x(t)| < a, a.e.}

= sup |x(t)|
t>0

 
where the integrals have to be understood on IR, i.e. = IR or, if the signals
 +
are dened on IR+ , as 0 . We say that a function f () belongs to Lp if
b
and only if f is locally Lebesgue integrable (i.e. | a f (t)dt| < + for any
IR  b a) and f p < +. To recapitulate:
For 1 p < +, Lp (I) = {f : I IR, f () is Lebesgue measurable and
  p1
I |f (t)| dt
p
< +}.
L (I) = {f : I IR, f () is Lebesgue measurable, dened and bounded
almost everywhere on I}.
Most of the time we shall write Lp instead of Lp (I), especially when I =
IR+ . In order to encompass multivariable systems, it is necessary to introduce
the norm for vector functions f : IR IRn , where fi Lp for each 1 i n
n 1
2 2
and ||f ||p = i=1 ||fi ||p .
9
Proposition 4.1. If f L1 L then f Lp for all 1 p +.

Proof: Since f L1 , the set A = {t| |f (t)| 1} has nite Lebesgue measure.
Therefore, since f L


|f (t)|p dt < , p [1, +)
A

Dene the set B = {t| |f (t)| < 1}. Then we have




|f (t)| dt
p
|f (t)|dt < |f (t)|dt < , p [1, +)
B B
Finally




|f (t)|p dt = |f (t)|p dt + |f (t)|p dt < +
A B
180 4 Dissipative Systems

4.2.1 Relationships Between L1 , L2 and L Spaces.

In order to understand the relationship between L1 , L2 and L spaces let us

consider the following examples that have been introduced in [125]:
f1 (t) = 1
1
f2 (t) = 1+t
1
1 1+t 4
f3 (t) = 1+t t 14
t
f4 (t) = e
1
1 1+t 4
f5 (t) = 1+t2 t 14
1
1 1+t 2
f6 (t) = 1+t2 t 12

It can be shown that (see Figure 4.1)

f1
/ L1 , f1
/ L2 and f1 L
f2
/ L1 , f2 L2 and f2 L
f3
/ L1 , f3 L2 and f3
/ L
f4 L1 , f4 L2 and f4 L
f5 L1 , f5 L2 and f5
/ L
f6 L1 , f6
/ L2 and f6
/ L

L
8

L2
f1

f3
f2
L1

f5
f4 f
6

Fig. 4.1. Relationships between L1 , L2 and L

4.3 Review of Some Properties of Lp Signals

The following facts are very useful to prove convergence of signals under dif-
ferent conditions.
 4.3 Review of Some Properties of Lp Signals 181

Fact 1: If V : IR IR is a non-decreasing function (see gure 4.2) and if

V (t) M for some M IR and all t IR, then V () converges.
Proof: Since V () is non-decreasing, then V () can only either increase or
remain constant. Assume that V () does not converge to a constant limit.
Then V () has to diverge to innity since it cannot oscillate. In other words
there exists a strictly increasing sequence of time instants t1 , t2 , t3 ... and a
> 0 such that V (ti ) + < V (ti+1 ). However this leads to a contradiction
since V has upper-bound M . Therefore, the sequence V (ti ) has a limit for any
sequence of time instants {ti }i1 so that V () converges.

0 t

Fig. 4.2. A nondecreasing function V ().

Examples:
t t
0 |s( )|d < 0 |s( )|d converges
Let V () be dierentiable. Then V () 0 and V () 0 = V () con-
verges.
t t
Fact 2: If 0 |f (t )|dt converges then 0 f (t )dt converges. Proof: In view
of the assumption we have

t


> |f (t )|dt = |f (t )|dt + |f (t )|dt
0 t|f (t)>0 t|f (t)0

Then both integrals in the right-hand side above converge. We also have

t


f (t )dt = |f (t )|dt |f (t )|dt
0 t|f (t)>0 t|f (t)0
t
Then 0 f ( )d converges too.
Fact 3: f L1 implies that f has a limit.
Proof: By assumption we have

t
t
|f (t) f (0)| = |
f (s)ds| |f(s)|ds <
0 0
182 4 Dissipative Systems
t
Using Fact 1 it follows that 0 |f(s)|ds converges. This implies that
t
0
f(s)ds converges which in turn implies that f () converges too.
Fact 4: If f L2 and f L2 then f (t) 0 as t + and f L .
Proof: Using the assumptions
t d 2
|f 2 (t) f 2 (0)| = | 0 ds [f (s)]ds|
t
0
| ds
d
[f 2 (s)]|ds
t
=2 0 |f (s)f(s)|ds (4.1)
t t
0
f 2 (s)ds + 0
f2 (s)ds

< +

In view of Fact 3 it follows that | dt d

[f 2 ]| L1 which implies that
t 2
d
0 ds [f (s)]ds converges which in turn implies that f 2 converges. But by as-
t 2
sumption 0 f (s)ds < , then f has to converge to zero. Clearly f L .
Fact 5: f L1 and f L1 f 0.
Proof: Using
t Fact 3 it follows that f L1 f has a limit. Since in addition
we have 0 |f (s)|ds < then f has to converge to zero.
Before presenting further results of Lp functions, some denitions are in
order.

Denition 4.2. The function (t, x)

f (t, x) is said to be globally Lipschitz
(with respect to x) if there exists a bounded k IR+ such that

|f (t, x) f (t, x )| k|x x |, x, x IRn , t IR+ (4.2)

Denition 4.3. The function (t, x)

f (t, x) is said to be locally Lipschitz
(with respect to x) if (4.2) holds for all x K, where K IRn is a compact
set. Then k may depend on K.

Example 4.4. Let f : x

x2 . Then f () is locally Lipschitz in [1, 1] since
|x2 y 2 | = |x y||x + y| 2|x y|, for all x, y [1, 1].

Denition 4.5. The function (t, x)

f (t, x) is said to be Lipschitz with
respect to time if there exists a bounded k such that

|f (t, x) f (t , x)| k|t t |, x IRn , t, t IR+

 4.3 Review of Some Properties of Lp Signals 183

Denition 4.6. The function f () is uniformly continuous in a set A if for

all > 0, there exists ( ) > 0:

|t t | < |f (t) f (t )| < , t, t A

Remark 4.7. Uniform continuity and Lipschitz continuity are two dierent no-
tions. Any Lipschitz function is uniformly continuous. However the inverse
implication is not true. For instance the function x
x is uniformly contin-
uous on [0, 1], but it is not Lipschitz on [0, 1]. This may be easily checked from
the denitions. The criterion in Fact 6 is clearly a sucient condition only
(very sucient, one should say!) to assure uniform continuity of a function.
Furthermore, uniform continuity has a meaning on a set. Asking whether a
function is uniformly continuous at a point is meaningless [420].

Fact 6: f L f is uniformly continuous.

Proof: f L implies that f is Lipschitz with respect to time t and that
f () is uniformly continuous.
Fact 7: If f L2 and is Lipschitz with respect to time then limt+ f (t) = 0.
t
Proof: By assumption: 0 f 2 (s)ds < and |f (t) f (t )| k|t t |, t, t .
Assume that
|f (t1 )| for some t1 , > 0
and
|f (t2 )| = 0 for some t2 t1
then
|f (t1 ) f (t2 )| k|t1 t2 |
i.e. |t1 t2 | k . We are now interested in computing the smallest upper-
t
bound for t12 f 2 (t)dt . We will therefore assume that in the interval of time
(t1 , t2 ) the function f () decreases at maximum rate which is given by k in
the equation above. We therefore have (see Figure 4.3):

t2
2 k 3
f 2 (s)ds =
t1 2 2k

Since f L2 , it is clear that the number of times |f (t)| can go from 0 to

is nite on IR. Since > 0 is arbitrary, we conclude that f (t) 0 as t .
Fact 8: If f Lp (1 p ) and if f is uniformly continuous, then f (t) 0
as t +.
Proof: This result can be proved by contradiction following the proof of Fact
7. Fact 9: If f1 L2 and f2 L2 , then f1 + f2 L2 .
Proof: The result follows from
 2  2
(f1 (t) + f2 (t)) dt = (f
 1 (t) + f22 (t) + 2f1 (t)f2 (t))dt
2 (f1 (t) + f22 (t))dt < +
2
184 4 Dissipative Systems

2
f (t)
2
H

t t t
1 2

_
H
k

Fig. 4.3. Proof of Fact 7

The following Lemma describes the behavior of an asymptotically stable

linear system when its input is L2 bounded.

Lemma 4.8. Consider the state space representation of a linear system

x(t) = Ax(t) + Bu(t) (4.3)

with u(t) IRm , x(t) IRn and A exponentially stable. If u L2 then x

L2 L , x L2 and limt+ x(t) = 0.

Remark 4.9. The system above with u L2 does not necessarily have an
equilibrium point. Therefore, we cannot use the Lyapunov approach to study
the stability of the system.

Proof of Lemma 4.8: Since A is exponentially stable then there exists

P = P T > 0, Q > 0 such that

P A + AT P = Q

which is the well known Lyapunov equation. Consider the following positive
denite function


V (x, t) = xT P x + k uT (s)u(s)ds
t

where k is a constant to be dened later. V (, ) is not a Lyapunov function

since the system may not have an equilibrium point. Note that since u L2 ,
there exists a constant k  such that

t

T
u (s)u(s)ds + uT (s)u(s)ds = k  <
0 t

Taking the derivative with respect to time we obtain

 4.3 Review of Some Properties of Lp Signals 185


d
uT (t)u(t) + uT (s)u(s)ds = 0
dt t

Using the above equations we get

V (x(t), t) = x(t)P x(t) + xT (t)P x(t) kuT (t)u(t)

= (xT (t)AT + uT (t)B T )P x(t) + xT (t)P (Ax(t) + Bu(t))

kuT (t)u(t)

= xT (t)(AT P + P A)x(t) + 2uT (t)B T P x(t) kuT (t)u(t)

= xT (t)Qx(t) + 2uT (t)B T P x(t) kuT (t)u(t)

(4.4)
Note that
2uT B T P x 2|uT B T P x|

2u B T P  x
  12   12 (4.5)
2
2u B T P  min Q
min Q
2 x

2
u2 B T P 2 min Q +
min Q
2 x2

where we have used the inequality 2ab a2 + b2 , for all a, b IR. Choosing
2
k = B T P 2 min Q we get

min Q
V (x(t), t) x(t)2
2
Therefore V (, ) is a non-increasing function and thus V L which implies
that x L . Integrating the above equation we conclude that x L2 . From
the system equation we conclude that x L2 (see also Fact 9). Finally x, x
L2 = limt+ x(t) = 0 (see Fact 4).
A more general result is stated in the following Theorem which can be
found in [125, p.59] where denotes the convolution product.

Theorem 4.10. Consider the exponentially stable and strictly proper system

x(t) = Ax(t) + Bu(t)
(4.6)

y(t) = Cx(t)
and its transfer function
186 4 Dissipative Systems

H(s) = C(sIn A)1 B

If u Lp , then y = h u Lp L , y Lp for p = 1, 2 and . For

p = 1, 2, then limt+ y(t) = 0.

The function h() in the Theorem is the inverse Laplace transform of H(s).
Theorem 4.10 is a consequence of the Datko-Pazy Theorem [123, 399] formu-
lated in an innite-dimensional framework.

4.3.1 Example of Applications of the Properties of Lp Functions in

Adaptive Control

Let us rst briey review the Gradient type Parameter Estimation Algorithm,
which is widely used in adaptive control and in parameter estimation. Let
y(t) IR, (t) IRn be measurable functions 1 which satisfy the following
linear relation:
y(t) = T (t)
where (t) IRn is an unknown constant vector. Dene y(t) = (t)T (t) and
e(t) = y(t) y(t); then
e(t) = (t)T (t) (4.7)
d d
where (t) = (t). Note that dt = dt . Dene the following positive function
1 2
V (, ) = e (4.8)
2
then

V d V d
V (, ) = + (4.9)
dt dt
Let us choose the following parameter adaptation algorithm:
T
d V
(t) = (4.10)
dt
Introducing (4.7) and (4.8) into (4.10) gives
T
d e
(t) = e = e
dt

The parameter adaptation law (4.10) is motivated by the fact that when = 0,
then introducing (4.10) into (4.9) leads to
1
Here measurable is to be taken in the physical sense, not in the mathematical
one. In other words we assume that the process is well-equipped with suitable
sensors.
 4.3 Review of Some Properties of Lp Signals 187

 T
V V
V (, ) = <0


Let W () = 12 T , then W () = T = T e. Integrating we obtain

t
(T )edt = W ((t)) W ((0)) W ((0))
0

We conclude that the operator H : e T is passive.

Example 4.11. (Adaptive control of a simple nonlinear system) Let

x(t) = f (x(t))T  + bu(t)

where u(t), x(t) IR. Dene



= b







(t) = (t)



(t) = f (x(t))x(t)






u(t) = T (t)f (x(t)) x(t) + v(t)
and
b T 1
+ x2
V (, x) =
2 2
Then along trajectories of the system we get

+ x(t)x(t)
V ((t), x(t)) = bT (t)(t)
 
= b(t)T f (x(t))x(t) + x(t) f (x(t))T  + bu(t)
(4.11)
= bx(t)[((t) )T f (x(t)) + T f (x(t)) + u(t)]

= bx2 (t) + bx(t)v(t)

From the last equation it follows that for v = 0, V () is a non-increasing

function and thus V, x, L . Integrating the last equation it follows that
x L2 L . Assume that f () has the required property so that x L
f (x) L . It follows that u L and also x L . x L2 and x L
implies limt+ x(t) = 0. Let us note from the last line of (4.11) that the
operator H : v
x is output strictly passive (OSP) as will be dened later.
188 4 Dissipative Systems

In order to present the Passivity theorem and the Small gain theorem
we will require the notion of extended spaces. We will next present a brief
introduction to extended spaces. For a more detailed presentation the reader
is referred to [125].

4.3.2 Linear Maps

Denition 4.12 (Linear maps). Let E be a linear space over K (IR orC).
Let L(E, E) be the class of all linear maps from E into E. L(E, E) is a linear
space satisfying the following properties x E, A, B L(E, E), K :


(A + B)x = Ax + Bx



(A)x = (Ax) (4.12)





(AB)x = A(Bx)

4.3.3 Induced Norms

Denition 4.13 (Induced Norms). Let |.| be a norm on E and A
L(E, E). The induced norm of the linear map A is dened as

A = sup |Ax|

|x|
x=0
(4.13)
= sup |Az|
|z|=1

4.3.4 Properties of Induced Norms

If A < and B < then the following properties hold for all x
E, K
1. |Ax| A|x|
2. A = ||A
3. A + B A + B
4. AB AB
Example 4.14. Let H be a linear map dened on E in terms of an integrable
function h : IR IR

H : u Hu = h u, u L
i.e.
t
(Hu)(t) = h(t )u( )d, t IR+
0
 4.3 Review of Some Properties of Lp Signals 189

Assume that h1 = 0
|h(t)|dt < .
Theorem 4.15. Under those conditions the following properties hold:
a) H : L L

b) H = h1 and h u h1 u, u L

and the right-hand side can be made arbitrarily close to the left-hand side of
the inequality by appropriate choice of u.
Proof: By denition and from (4.13) we obtain

H = sup||u|| =1 ||Hu||

= sup h u

u
=1
= sup sup |(h u) (t)|

u
=1 

t0

 t 
= sup sup  0 h(t )u( )d 

u
=1  t0
t
sup sup 0 |h(t )| |u( )| d

u
=1 t0

Since u = 1 we have

t
H sup 0 |h(t )| d
t0

t
= sup 0
|h(t )| d
t0

t
= sup 0
|h(t )| dt
t0


0
|h(t )| dt = h1
We can choose ut ( ) =sgn[h(t )], t IN . Thus

t
(h ut )(t) = |h(t )|d h ut 
0

Therefore t t
0
|h(  )|d  = 0
|h(t )|d

h ut 
(4.14)
H h1

= 0
|h(t )|dt
Letting t it follows that H = h1 .
190 4 Dissipative Systems

4.3.5 Extended Spaces

Consider a function f : IR+ IR and let 0 T < +. Dene the truncated

function
f (t) if t T
fT (t) = (4.15)

0 if t > T
The function fT is obtained by truncating f () at time T . Let us introduce
the following denitions:
T : subset of IR+ (typically, T = IR+ or IN ),
V: normed space with norm . (typically V = IR, IRn , C, Cn ),
F = {f | f : T V} the set of all functions mapping T into V.
The normed linear subspace L is given by

L = {f : T V | f  < }

Associated with L is the extended space Le dened by


Le = {f : T V | T T , fT  < }

In other words, the sets Lp,e or simply Le , consist of all Lebesgue measurable
functions f such that every truncation of f belongs to the set Lp (but f may
not belong to Lp itself, so that Lp Lp,e ). The following properties hold for
all f Lp,e :
1. The map t ft  is monotonically increasing
2. ft  f  as t +
 1
Remark 4.16. One sometimes speaks of Lp,loc , which means that I |f (t)|p dt p
< + for all compact intervals I IR. Obviously Lp,loc = Lp,e .

We can now introduce the notion of gain of an operator which will be used
in the small gain Theorem and the passivity Theorem.

4.3.6 Gain of an Operator

In the next Denition, we consider a general operator with state, input, and
output signal spaces. In particular, the input-output mapping is assumed to
be causal, invariant under time shifts, and it depends on the initial state x0 .
 4.3 Review of Some Properties of Lp Signals 191

Denition 4.17. [206] Consider an operator H : Le Le . H is

weakly nite-gain stable (WFGS) if there exist a function () and a
constant k such that

(Hu)T  kuT  + (x0 )

for all admissible u() and all x0 . If (x0 ) = 0, we call H nite-gain

stable (FGS).

In a more rigorous way, the input-output operator H should be denoted as

H(x0 ) or Hx0 as it may depend on x0 . This is a situation completely analogous
to that of passive operator as in Denition 2.1, where the constant may in
general depend on the initial state x0 . One may be surprized that the notion of
state intervenes in a denition that concern purely input-output operators (or
systems). Some denitions, indeed, do not mention such a dependence. This
is related to the very basic denition of what a system is, and well-posedness.
Then the notions of input, output and state can hardly be decoupled, in
general.
We call the gain of H the number k (or k(H)) dened by
k(H) = inf{k IR+ / : (Hu)T  kuT  + , u Le , T IR+ }
Let us recall the case of linear time invariant systems of the form

x(t) = Ax(t) + Bu(t)
(4.16)

y(t) = Cx(t) + Du(t), x(0) = x0 IRn .
Theorem 4.18. Suppose that the matrix A has all its eigenvalues with nega-
tive real parts ( x(t) = Ax(t) is asymptotically stable). Then the system
(4.16) is nite-gain stable where the norm can be any Lp with 1 p +.
In other words u Lp = y Lp and ||y||p kp ||u||p for some kp < +.
A rather complete exposition of input/output stability can be found in
[500, Chapter 6].

4.3.7 Small Gain Theorem

This Theorem gives sucient conditions under which a bounded input pro-
duces a bounded output (BIBO).
Theorem 4.19 (Small gain). Consider H1 : Le Le and H2 : Le Le .
Let e1 , e2 Le and dene (see Figure 4.4)

u 1 = e 1 + H2 e 2
(4.17)
u 2 = e 2 H1 e 1
192 4 Dissipative Systems

e1 y1
u1 + H1
-

+
y2 e2
H2 u2
+
Fig. 4.4. Closed-loop system with two external inputs

Suppose there are constants 1 , 2 , 1 , 2 0 such that for all t IR+ :

(H1 e1 )T  1 e1T  + 1
(4.18)
(H2 e2 )T  2 e2T  + 2

Under those conditions, if 1 2 < 1, then:

i)

e1t  (1 1 2 )1 (u1t  + 2 u2t  + 2 + 2 1 )

e2t  (1 1 2 )1 (u2t  + 1 u1t  + 1 + 1 2 )

ii) If in addition, u1 , u2  < +, then e1 , e2 , y1 , y2 have nite norms.

Proof: From (4.17) we have

e1t = u1t (H2 e2 )t

(4.19)
e2t = u2t + (H1 e1 )t
Then
e1t  u1t  + (H2 e2 )t  u1t  + 2 e2t  + 2
e2t  u2t  + (H1 e1 )t  u2t  + 1 e1t  + 1
Combining these two inequalities we get

e1t  u1t  + 2 + 2 (u2t  + 1 e1t  + 1 )

e2t  u2t  + 1 + 1 (u1t  + 2 e2t  + 2 )
Finally
 4.4 Dissipative Systems 193

e1t  (1 1 2 )1 [u1t  + 2 u2t  + 2 + 2 1 ]

e2t  (1 1 2 )1 [u2t  + 1 u1t  + 1 + 1 2 ]
The remainder of the proof follows immediately.
Clearly to be consistent with Denition 4.17, the constants 1 , 2 , 1
and 2 may also depend on initial states x1,0 and x2,0 . This obviously does
not modify the above calculations. A general notion of dissipativity will be
introduced, and some links with the gain theory will be established in Sections
4.4 and 5.1.

4.4 Dissipative Systems

4.4.1 Denitions

We will now review the denitions and properties of dissipative systems. Most
of the mathematical foundations on this subject are due to Willems [512], and
Hill and Moylan [206, 207]. One diculty when looking at the literature on
the subject, is that there are many dierent notions of dissipativity which are
introduced in many papers published here and there. One of the goals of this
chapter is to present all of them in one shot and also the existing relationships
between them. Consider a causal nonlinear system () : u(t)
y(t); u(t)
Lpe , y(t) Lpe , represented by the following state-space representation ane
in the input:


x(t) = f (x(t)) + g(x(t))u(t)



() y(t) = h(x(t)) + j(x(t))u(t) (4.20)





x(0) = x0
where x(t) IRn , u(t), y(t) IRm , f (), g(), h() and j() possess sucient
regularity so that the system with inputs in L2,e is well-posed (see Section
3.9.2), and f (0) = h(0) = 0. In other words the origin x = 0 is a xed point
for the uncontrolled (free) system, and there is no output bias at x = 0. The
state space is denoted as X IRn . Let us call w(t) = w(u(t), y(t)) the supply
rate and be such that for all admissible u() and x(0) and for all t IR+

t
|w(u(s), y(s))|ds < + (4.21)
0

i.e. we are assuming w() to be locally Lebesgue integrable independently

t
of the input and the initial conditions. In an electric circuit 0 w(s)ds can be
associated with the energy supplied to the circuit in the interval (0, t), i.e.
t
0
v(s)i(s) ds where v() is the voltage at the terminals and i() the current
194 4 Dissipative Systems

entering the circuit, see the example in Chapter 1. In the following, what we
will often call an admissible input, simply means that the ordinary dieren-
tial equation in (4.20) possesses a unique dierentiable solution. Hence it is
sucient that the vector eld f (x(t)) + g(x(t))u(t) satises the Caratheodory
conditions (see Theorem 3.55): u() may be a Lebesgue measurable function
of time.

Denition 4.20 (Dissipative System). The system () is said to

be dissipative if there exists a so-called storage function V (x) 0 such
that the following dissipation inequality holds:

t
V (x(t)) V (x(0)) + w(u(s), y(s))ds (4.22)
0

along all possible trajectories of () starting at x(0), for all x(0), t 0

(said dierently: for all admissible controllers u() that drive the state
from x(0) to x(t) on the interval [0, t]).

It follows from Lemma 3.1 and Corollary 3.3 that controllable and observ-
able LTI systems with a positive real transfer functions, are dissipative with
quadratic storage functions (see also [489] in the context of behavioural ap-
proach to linear dynamical systems). Two comments immediately arise from
Denition 4.20: rst storage functions are dened up to an additive constant;
second, if the system is dissipative with respect to supply rates wi (u, y),
1 i m, thenthe system is also dissipative with respect to any supply
m
rate of the form i=1 i wi (u, y), with i 0 for all 1 i m. One notices
that the Denition 4.20 (sometimes referred to as Willems dissipativity) does
not require any regularity on the storage functions: it is a very general deni-
tion. Actually, storage functions do possess some regularity properties under
suitable assumptions, see Section 4.4.5. When one imposes that the storage
functions be of class C r for some integer r 0, then one speaks of C r -
dissipativity. A third comment may be done: Willems denition postulates
that dissipativity holds whenever a storage function exists. Some other au-
thors like Hill and Moylan, start from a denition that is closer to Denition
2.1, and then prove the existence of storage functions.

Example4.21. Let us continue

t with Example 3.2. The input-output product
t
satises 0 u(t )y(t )dt = 0 u2 (t )dt 0 for any initial data x(0). Now choose
V (x) = 12 x2 . One has V (x(t)) V (x(0)) since solutions strictly decrease.
t
Thus V (x(t)) V (x(0)) 0 and V (x(t)) V (x(0)) 0 u(t )y(t )dt : the
system is dissipative, though neither observable nor controllable (but, it is
stable).
 4.4 Dissipative Systems 195

It is noterworthy that (4.22) is equivalent to the following: there exists

W () such that V (x1 ) V (x0 ) W (x1 , x0 ) with

t
W (x1 , x0 ) = inf w(u(s), y(s))ds (4.23)
u()U 0
along admissible controls which drive the state from x0 to x1 on the time
interval [0, t]. In the following we shall use either 0 or t0 to denote the initial
time for (4.20). Dissipativity is also introduced by Hill and Moylan [207] as
follows:
Denition 4.22. The system () is dissipative with respect to the supply rate
w(u, y) if for all admissible u() and all t1 t0 one has

t1
w(u(t), y(t))dt 0 (4.24)
t0

with x(t0 ) = 0 and along trajectories of ().

This corresponds to imposing that storage functions satisfy V (0) = 0. This
is justied by the fact that storage functions will often, if not always, be used
as Lyapunov functions for studying the stability of an equilibrium of () with
zero input u(). In a slightly more general setting, one may assume that the
controlled system has a xed point x
(corresponding to some input u
, and
with f (x
) + g(x
)u
= 0, y
= h(x
) + j(x
)u
, and w(u
, y
) = 0), and that
V (x
) < +. Then changing V () to V () V (x
) one obtains V (x
) = 0
(we could even have stated this as an assumption in Denition 4.20, as done
for instance in [510]). In the sequel of this chapter, we shall encounter some
results in which dissipativity is indeed assumed to hold with V (0) = 0. Such
results were originally settled to produce Lyapunov functions, precisely. Hill
and Moylan start from (4.24) and then prove the existence of storage functions,
adding properties to the system. The motivation for introducing Denition
4.22 is clear from Corollary 3.3, as it is always satised for linear invariant
positive real systems with minimal realizations.
Another denition [206] states that the system is weakly dissipative with
t
respect to the supply rate w(u, y) if t01 w(u(t), y(t))dt (x(t0 )) for some
() 0 with (0) = 0 [531] (we shall see later the relationship with Willems
Denition; it is clear at this stage that weak dissipativity implies dissipativity
in Denition 4.22, and that Willems dissipativity implies weak dissipativity
provided V (0) = 0). Still, another denition is as follows [232]:
Denition 4.23. The system () is said dissipative with respect to the supply
rate w(u, y) if there exists a locally bounded nonnegative function V : IRn
IR, such that

t 8
V (x) sup V (x(t)) w(u(s), y(s))ds : x(0) = x (4.25)
t0,uU 0
196 4 Dissipative Systems

where the supremum is therefore computed with respect to all trajectories of

the controlled system with initial condition x and admissible inputs.

This denition requires the local boundedness of storage functions (a real

valued function is locally bounded, if supxK | f (x) | C for some bounded
constant C > 0 and any compact set K of its domain). This additional prop-
erty happens to be important for further characterization of the storage func-
tions as solutions of partial dierential inequalities (see Section 4.6). Apart
from this additonal property, one sees that if V (x)(= V (x(0)) satises (4.25),
then it satises (4.22). Conversely since (4.22) is satised for all t 0 and for
all admissible u(), if V (x(0))(= V (x)) satises (4.22) then it satises (4.25).
Thus under the local boundedness assumption, Willems original denition
and the denition stemming from (4.25), are equivalent. The fact that Deni-
tion 4.20 implies Denition 4.22 provided that V (0) = 0 is clear. The converse
will be investigated in Section 4.5.2.
There is another denition of dissipativity that is sometimes used by Hill
and Moylan:

Denition 4.24. The system () is said to be cyclo-dissipative with respect

to the supply rate w(u, y) if

t1
w(u(s), y(s)ds 0 (4.26)
t0

for all t1 t0 , all admissible u(), whenever x(t0 ) = x(t1 ) = 0.

The dierence with Denition 4.20 is that the state boundary conditions
are forced to be the equilibrium of the uncontrolled system: trajectories start
and end at x = 0. A cyclo-dissipative system absorbs energy for any cyclic
motion passing through the origin. Cyclo-dissipativity and dissipativity are
related as follows:

Theorem 4.25. [209] Suppose that the system () denes a causal input-
output operator Hx(0) , and that the supply rate is of the form w(u, y) = y T Qy+
2y T Su + uT Ru, with Q non-positive denite. Suppose further that the system
is zero state detectable (i.e. u(t) = 0, y(t) = 0 t 0 = limt x(t) = 0).
Then dissipativity in the sense of Denition 4.22 and cyclo-dissipativity of
() are equivalent properties.

The proof of this Theorem relies on the denition of another concept known
as ultimate dissipativity, which we do not wish to introduce here for the sake
of briefness (roughly, this is dissipativity but only with t = + in (4.22)).
The reader is therefore referred to [209] for the proof of Theorem 4.25 (which
is a concatenation of Denitions 2, 3, 8 and Theorems 1 and 2 in [209]). Let
us recall that an operator H : u
y = H(u, t) is causal if the following
holds: for all admissible inputs u() and v() with u( ) = v( ) for all t,
then H(u(), t) = H(v(), t). In other words, the output depends only on the
 4.4 Dissipative Systems 197

past values of the input, and not on future values. It is noteworthy here that
causality may hold for a class of inputs and not for another class.
A local denition of dissipative systems is possible. Roughly, the dissi-
pativity inequality should be satised as long as the systems state remains
inside a closed domain of the state space [404].

Denition 4.26 (Locally dissipative system). Let X be the systems state

space. Let Ue = {u() | ut () U for all t IR}. The dynamical system is
locally dissipative with respect to the supply rate w(u, y) in a region X if

t
w(u(s), y(s))ds 0 (4.27)
0

for all u Ue , t 0, such that x(0) = 0 and x(s) for all 0 s t.

Still, another notion is known as the quasi-dissipativity:

Denition 4.27. [403] The system () is said quasi-dissipative with respect

to the supply rate w(u, y) if there exists a constant 0 such that it is
dissipative with respect to the supply rate w(u, y) + .

Actually, dissipativity is understood here as weak dissipativity, i.e.

t
w(u(s), y(s))ds
0

with 0 (see Denition 2.1). The interest of quasi-dissipativity is that a

quasi-dissipative system can contain a source of energy with nite power.
To conclude this subsection, we have at our disposal several notions of dissi-
pativity: Willems, Hill and Moylans, Denition 2.1, weak dissipativity (which
is intermediate between Denition 2.1 and Willems), cyclo-dissipativity,
quasi-dissipativity, ultimate dissipativity, local dissipativity, Denition 4.23.
There are more (like Jdissipativity [397], which is used in specic settings
like H control), exponential dissipativity (see Theorem 5.68), denitions
taylored to systems with time-varying parameters [302], and Popovs hyper-
stability.
t t
Remark 4.28. Some properties are stated as 0 for all t 0, and others as t01
for all t1 t0 . If trajectories (state) are independent of the initial time but
depend only on the elapsed time, clearly both ways of stating dissipativity are
equivalent.

4.4.2 The Signication of

The next result helps to understand the signication of the constant (apart
from the fact that, as we shall see later, one can exhibit examples which
prove that the value of (0) when is a function of the initial state, has a
198 4 Dissipative Systems

strong inuence on the stability of the origin x = 0). The supply rate that is
considered is the general supply rate w(u, y) = y T Qy + 2y T Su + uT Ru, where
Q = QT and R = RT (but no other assumptions are made, so that Q and R
may be zero). The denition of weak dissipativity is as seen above, but in a
local setting, i.e. an operator G : U Y which is denoted as Gx0 as it may
depend on the initial state. For a region IRn we denote G() the family
of operators Gx0 for all x0 . Considering such domain may be useful for
systems with multiple equilibria, see Example 4.34. Mimicking the denition
of weak nite gain (Denition 4.17):

Denition 4.29. [206] The operator G() is said weakly w(u, y)dissipative
if there exists a function : IR such that

t
w(u(s), y(s))ds (x0 ), (4.28)
0

for all admissible u(), all t 0, and all x0 . If (x0 ) = 0 in then

the operator is called w(u, y)dissipative.

This denition has some local taste as it involves possibly several equilibria
of the system (the set ). Therefore when time comes to settle some stability
of these equilibria, it may be that only local stability can be deduced. We also
need a reachability denition. The distance of x to is d(x, ) = inf x0 ||x
x0 ||.

Denition 4.30. [206] A region X1 IRn is uniformly reachable from

IRn if there exists a class K function (), and for every x1 X1 there
exists x0 , a nite t1 t0 and an admissible u() such that the trajectory
of the controlled system originating from x0 at t0 satises x(t1 ) = x1 and
 t1
0
u(s)T u(s)ds (d(x1 , )).

Uniform reachability means that a state x1 can be driven from some other
state x0 with an input that is small if the distance between the two states is
small. It is local in nature.

Theorem 4.31. [206] If G() is weakly w(u, y)dissipative, and X1 is uni-

formly reachable from , then G(X1 ) is weakly w(u, y)dissipative.

Proof: Take any x1 X1 and any t1 > t0 , any x0 , any u() U such
that the controlled trajectory emanating from x0 at t0 ends at x1 at t1 . The
value of u(t) for t > t1 is arbitrary. The inequality in (4.28) can be rewritten
as

t
w(u(s), y(s))ds new (x1 ) (4.29)
0
 4.4 Dissipative Systems 199
t
with new (x1 ) = (x0 ) 0 1 w(u(s), Gx0 (u(s)))ds, and we used the fact
that the operator is invariant under time shifts. The value new (x1 ) depends
only on x1 and not on u() because the control that intervenes in the denition
of new (x1 ) is the specic control which drives x0 to x1 . Thus Gx1 is weakly
dissipative.
If (x0 ) = 0 then the system is dissipative with respect to one initial state
(in the sense of Denition 4.22 if x0 = 0). But it is weakly dissipative with
respect to other reachable initial states. Therefore a way to interpret is that
it allows to take into account non-zero initial states. In Example 4.60 we will
see that weak nite-gain stability is not enough to assure that the uncontrolled
state space representation of the system has a Lyapunov  t stable xed point.
It follows from this analysis that dening passivity as 0 uT (s)y(s)ds 0 for
any initial state makes little sense if the initial state is not included in the
denition (or implicitly assumed to be zero).
The equivalence between Willems denition and weak dissipativity follows
from the following:

Theorem 4.32. [206] For some X1 X, G(X1 ) is weakly dissipative

if and only if there exists a function V : X1 IR, with V (x) 0 for
all x X, such that

t
V (x1 ) + w(u(s), y(s))ds V (x2 ) (4.30)
t0

for all x1 X1 , all admissible u() U, all t t0 , with y(t) =

Gx1 (u(t)) and x(t) = x2 is the state starting at x1 at t0 .

t
Proof: Let us denote V (u, y, t0 , t) = t0 w(u(s), y(s))ds. By time-invariance
V (u, y, t0 , t) depends only on t t0 but not on t and t0 separately. Let
V (x1 ) = inf V (u, Gx1 u, t1 , t). Because of t t1 , t may be chosen as
u()U ,tt1
t1 and consequently V (x1 ) 0. For any t2 t1 and t t2 one has V (x1 )
V (u, Gx1 u, t1 , t2 ) V (u, Gx2 u, t2 , t), where x(t2 ) = x2 is the state which
starts at x1 at time t1 and with the control u on [t1 , t2 ]. Since this inequality
holds for all u, it is true in particular that V (x1 ) V (u, Gx1 u, t1 , t2 )
inf V (u, Gx2 u, t2 , t) from which (4.30) follows. The inequality (4.28)
u()U ,tt2
implies that V (x1 ) (x1 ) so that 0 V (x) < + for all x X1 . Now
t
starting from (4.30) one sees that V (x1 ) + t0 w(u(s), y(s))ds 0 which shows
that the system is w(u, y)dissipative.
Moreover:
Theorem 4.33. [206] Assume that X1 X is uniformly reachable from
X. Then G() is w(u, y)dissipative if and only if there exists a function
200 4 Dissipative Systems

V : X1 IR satisfying the conditions of Theorem 4.32 and V (x) = 0 for all

x .

Proof: If G() is w(u, y)dissipative and X1 is reachable from , then

Theorem 4.31 shows that Gx1 is w(u, y)dissipative. Following the same steps
as in the proof of Theorem 4.32, the only thing that remains to be shown is
that V (x) = 0 for all x . The bounds 0 V (x) (x) for all x X1 and
Denition 4.29 imply it. The converse is a direct consequence of (4.30).
Thus summarizing Theorems 4.25, 4.32 and 4.33:

cyclo-dissipativity
Denition 4.24
 (if ZSD and Q 0)
Hill and Moylans dissipativity
Denition 4.22
(if reachability)
weak w(u, y)dissipativity [w(u, y)dissipativity + reachability]
Denition 4.29

Willems dissipativity [Willems dissipativity +V (0) = 0]
Denition 4.20
(if local boundedness of the storage function)


Denition 4.23

The link between w(u, y)dissipativity and dissipativity in Denition 4.22

can also be established from Theorem 4.33. The equivalence between weak
w(u, y)dissipativity and the other two, supposes that the required dynamical
system that is under study is as (4.20), so in particular 0 .

Example 4.34. [206] To illustrate Denition 4.29 consider the following sys-
tem:

x(t) = sin(x(t)) + 2u(t)




y(t) = sin(x(t)) + u(t) (4.31)





x(0) = x0
 4.4 Dissipative Systems 201

with > 0. Then V (x) = (1 cos(x)), V (x0 ) = 0 for all x0 = 2n, n IN .

Thus = {x0 | x0 = 2n}. This system is nite-gain stable, and the
equilibria are (locally) asymptotically stable.

4.4.3 Storage Functions (Available, Required Supply)

Having in mind this preliminary material, the next natural question is, given a
system, how can we nd V (x)? This question is closely related to the problem
of nding a suitable Lyapunov function in the Lyapunov second method. As
will be seen next, a storage function can be found by computing the maximum
amount of energy that can be extracted from the system.

Denition 4.35 (Available Storage). The available storage Va ()

of the system () is given by


t 8
0 Va (x) = sup w(u(s), y(s))ds (4.32)
x=x(0),u(),t0 0

where Va (x) is the maximum amount of energy which can be extracted

from the system with initial state x = x(0).

The supremum is taken over all admissible u(), all t 0, all signals with
initial value x(0) = x, and the terminal boundary condition x(t) is left free.
It is clear that 0 Va (x) (just take t = 0 to notice that the supremum cannot
be negative). When the nal state is not free but constrained to x(t) = 0
(the equilibrium of the uncontrolled system), then one speaks of the virtual
available storage, denoted as Va
() [209]. Another function plays an important
role in dissipative systems, called the required supply. We recall that the state
space of a system is said reachable from the state x
if, given any x and t there
exist a time t0 t and an admissible controller u() such that the state can be
driven from x(t0 ) = x
to x(t) = x. The state space X is connected provided
every state is reachable from every other state.

Denition 4.36 (Required Supply). The required supply Vr () of

the system () is given by


0 8
Vr (x) = inf w(u(s), y(s))ds (4.33)
u(),t0 t

with x(t) = x
, x(0) = x, and it is assumed that the system is

reachable from x
. The function Vr (x) is the required amount of energy
to be injected in the system to go from x(t) to x(0).
202 4 Dissipative Systems

The inmum is taken over all trajectories starting from x
at t and ending
at x(0) = x at time 0, and all t 0 (or, said dierently, over all admissible
controllers u() which drive the system from x
to x on the interval [t, 0]).
If the system is not reachable from x
, one may dene Vr (x) = +.
Remark 4.37. The optimal extraction control policy which allows one to
obtain the available storage in case of an LTI system as in (3.1) is given by
u = (D + DT )1 (B T P C)x, and the optimal supply control policy which
allows one to obtain the required supply is given by u = (D + DT )1 (B T P +
C)x, where P + and P are as in Theorem 3.44.
Remark 4.38. Contrary to the available storage, the required supply is not
necessarily positive, see however Lemma 4.45. When the system is reversible,
the required supply and the available storage coincide [512]. It is interesting to
dene accurately what is meant by reversibility of a dynamical system. This
is done thanks to the denition of a third energy function, the cycle energy:

t1
Vc (x) = inf u(t)T y(t)dt (4.34)
u(),t0 t1 ,x(t0 )=0 t0

where the inmum is taken over all admissible u() which drive the system
from x(t0 ) = 0 to x. The cycle energy is thus the minimum energy it takes
to cycle a system between the equilibrium x = 0 and a given state x. One
has Va () + Vc () = Vr () (assuming that the system is reachable so that the
required supply is not identically +). Then the following is in order:
Denition 4.39 (Reversibility). Let a dynamical system be passive in the
sense of Denition 2.1 with = 0, and let its state space representation be
reachable. The system is irreversible if Vc (x) = 0 only if x = 0. It is said
uniformly irreversible if there exists a class K function () such that for all
x IRn : Vc (x) (||x||). The system is said to be reversible if Vc (x) = 0 for
all x IRn , i.e. if Va () = Vr ().
The following is taken from [209].
Example 4.40. Let us consider the one-dimensional system


x(t) = x(t) + u(t)



y(t) = x(t) + 12 u(t) (4.35)





x(0) = x0 .
This system is dissipative with respect to the supply rate w(u, y) = uy. Indeed
t t  2 t t
1 2 x (s)
0 u(s)y(s)ds = 0 [(x(s) + x(s))x(s) + 2 u (s)]ds = 2 + 0 (x2 (s) +
0
 4.4 Dissipative Systems 203
2
1 2
2 u (s))ds x 2(0) . Then Va (x) = 2 3 2
2 x and Vr (x) = 2+2 3 x2 . Indeed the
available storage and required supply are the extrema solutions of the Riccati
equation AT P + AP + (P B C T )(D + DT )1 (B T P C) = 0, which is in
this case p2 4p + 1. Moreover the available storage and the virtual available
storage (where the terminal state is forced to be x = 0) are the same. One
sees that V (x) = 12 x2 is a storage function.
The following results link the boundedness of the functions introduced in
Denitions 4.35 and 4.36 to the dissipativeness of the system. As an example,
consider again an electrical circuit. If there is an ideal battery in the circuit,
the energy that can be extracted is not nite. Such a circuit is not dissipative.
The following results are due to Willems [510, 511].

Theorem 4.41. [510, 511] The available storage Va () in (4.32), is

nite for all x X if and only if () in (4.20) is dissipative in the
sense of Denition 4.20. Moreover, 0 Va (x) V (x) for all x X
for dissipative systems and Va () is itself a possible storage function.

Proof:
() In order to show that Va (x) < the system () in (4.20) is dissipa-
tive, it suces to show that the available storage Va in (4.32) is a storage
function i.e. it satises the dissipation inequality

t
Va (x(t)) Va (x(0)) + w(t)dt
0

But this is certainly the case because the available storage Va (x(t)) at
time t is not larger than the available storage Va (x(0)) at time 0 plus the
energy introduced into the system in the interval [0, t].
() Let us now prove that if the system () is dissipative then Va (x) < .
If () is dissipative then there exists V (x) 0 such that

t
V (x(0)) + w(t)dt V (x(t)) 0
0

From the above and (4.32) it follows that

t 8
V (x(0)) sup w(t)dt = Va (x)
x=x(0),t0,u 0

Since the initial storage function V (x(0)) is nite it follows that Va (x) <
+. The last part of the Theorem follows from the denitions of Va ()
and V () (see (4.25)).
204 4 Dissipative Systems

Therefore dissipativeness can be tested by attempting to compute Va (x):

if it is locally bounded, it is a strorage function and the system is dissipa-
tive with respect to the supply rate w(u, y). This is a variational approach.
Compare (4.32) with (4.25). It clearly appears why, among all possible stor-
age functions satisfying (4.25), the available storage is the smallest one.
Testing the dissipativity of the system () is by Theorem 4.41 equivalent
 +
to testing whether or not inf uU 0 w(u(t), y(t))dt under the constraints
x(t) = f (x(t)) + g(x(t))u(t), x(0) = x0 , is nite for all x0 IRn . As we saw
in Section 3.8.2 the value of this inmum yields the negative of the available
storage function. Similar results can be derived from the cyclo-dissipativity:

Lemma 4.42. [209] Let the system () be cyclo-dissipative. Then

(i) Vr (x(0)) < + for any reachable state x(0) and with x(t) = 0
(ii) Va
(x(0)) > for any controllable state x(0)
(iii) Va
(0) = Vr (0) = 0 if x(t) = 0
(iv) Vr (x) Va
(x) for any state x X

Controllability means in this context that there exists an admissible u()

that drives the state trajectory towards x = 0 at a time t 0. Proof:
(i) and (ii) are a direct consequence of reachability and controllability, and
the fact that w(u(s), y(s)) is integrable. Now let x(0) be both reachable
and controllable. Let us choose a state trajectory which  0passes through the
points x(t) = x(t) = 0, and with x(0) = x. Then t w(u(s), y(s))ds +
t
0 w(u(s), y(s))ds 0, from the denition of cyclo-dissipativity. From the def-
initions of Va
() (paragraph below Denition 4.35) and Vr (), (iv) follows using
0 t
that t w(u(s), y(s))ds 0 w(u(s), y(s))ds. (iv) remains true even in the
case of uncontrollability and unreachability, as in such a case Vr (x(0)) = +
and Va (x(0)) = .
Similarly to the above results concerning the available storage, we have
the following:

Theorem 4.43. [510,511] The system () in (4.20) is dissipative in

the sense of Denition 4.20 if and only if the required supply satises
Vr (x) K > for all x X and some K IR. Moreover,
0 Va (x) V (x) Vr (x) for all x X for dissipative systems.

Before presenting the next Lemma, let us introduce a notion of reachability.

Denition 4.44 (Locally wuniformly reachable). [209] The system ()
is said to be locally wuniformly reachable at the state x
if there exists a
neighborhood of x
and a class K function () such that for each x
 4.4 Dissipative Systems 205

there exist t 0 and an admissible u() driving the system from x
to x on

the interval [0, t) and

t
| w(u(s), y(s))ds| (||x x
||) (4.36)
0
The system is said to be locally uniformly wreachable in if it is locally
uniformly wreachable at all states x
.
A way to characterize such a property is indicated later; see Proposition
4.76. The following provides informations on whether or not the required
supply may serve as a storage function. It is extracted from [401, Theorem 2].

Lemma 4.45. Let the system () be dissipative in the sense of Denition 2.1
with respect to the supply rate w(u, y), and locally wuniformly reachable at
x
. Let V () be a storage function. Then the function Vr () + V (x(0)) is a
continuous storage function.

One sees that if the storage function satises V (0) = 0 and if x(0) = 0
then the required supply is a storage function. The value V (x(0)) plays the
role of the bias in Denition 2.1. When V (0) = 0 the system has zero bias
at the equilibrium x = 0. In fact a variant of Theorem 4.41 can be stated as
follows, where dissipativity is checked through Va () provided the system ()
is reachable from some state x
.

Lemma 4.46. [442] Assume that the state space of () is reachable from
x
X. Then () is dissipative in the sense of Denition 4.20 if and only if
Va (x
) < +.

The conditions of Theorem 4.41 are less stringent since reachability is not
assumed. However in practice, systems of interest are often reachable so that
Lemma 4.46 is important.
Notice that given two storage functions V1 () and V2 () for the same supply
rate, it is not dicult to see from the dissipation inequality that for any
constant [0, 1] then V1 () + (1 )V2 () is still a storage function. More
formally:

Lemma 4.47. The set of all possible storage functions of a dissipative system
is convex. In particular Va () + (1 )Vr () is a storage function provided
the required supply is itself a storage function.

Proof: Let V1 () and V2 () be two storage functions. Let 0 1 be a

constant. Then it is an easy computation to check that V1 () + (1 )V2 ()
is also a storage function. Since the available storage and the required supply
are storage functions, the last part follows.
The available storage and the required supply can be characterized as
follows:
206 4 Dissipative Systems

Proposition 4.48. Consider the system () in (4.20). Assume that it

is zero state observable (u(t) = 0 and y(t) = 0 for all t 0 imply that
x(t) = 0 for all t 0), with a reachable state space X, and that it is
dissipative with respect to w(u, y) = 2uT y. Let j(x) + j T (x) have full rank
for all x X. Then Va () and Vr () are solutions of the partial dierential
equality:

V T (x)f (x)+
   
+ hT (x) 12 V T (x)g(x) (j(x) + j T (x))1 h(x) 12 g T (x)V (x) = 0
(4.37)

Before presenting the proof we need an intermediate result:

Lemma 4.49. Let a function V () be dierentiable. Let W (x) =V T (x)f (x)

and S(x) = hT (x) 12 V T (x)g(x). Then along any trajectory of () in (4.20)
and for all t1 and t0 with t1 t0 , one has

 t1
t0 2u(t)T y(t)dt =

 t1 W (x(t)) S(x(t)) 1
= [V (x(t))]tt10 + t0 [1 uT (t)] dt
S T (x(t)) j(x(t)) + j T (x(t)) u
(4.38)

Proof: The proof is led by calculating the integral of the right-hand-side of

(4.38).

Proof of Proposition 4.48: Let us rewrite the available storage as

t 8
Va (x) = inf w(u(s), y(s))ds (4.39)
x=x(0),u(),t0 0

Using Lemma 4.49 one gets

 4.4 Dissipative Systems 207

Va (x) =

t 1
= inf [Va (x(t))]t0 + 0 [1 uT (t)]D(x(t)) dt
x=x(0),u(),t0
u (4.40)

t 1
= Va (x) inf Va (x(t)) + 0 [1 uT (t)]D(x(t)) dt
x=x(0),u(),t0
u

Wa (x(t)) Sa (x(t))
where we used that x(0) = x and D(x) = .
T T
Sa (x(t)) j(x(t)) + j (x(t))
Therefore


t 1
0= inf Va (x(t)) + [1 uT (t)]D(x(t)) dt (4.41)
x=x(0),u(),t0 0
u
If the inmum exists and since j(x(t))+j T (x(t)) is supposed to be full rank,
it follows that its Schur complement Wa (x) Sa (x)(j(x)+ j T (x))1 SaT (x) = 0
(see Lemma A.62), which exactly means that Va () satises the partial dier-
ential inequality (4.37). A similar proof may be made for the required supply.

In the linear time invariant case, and provided the system is observable and
controllable, then Va (x) = xT Pa x and Vr (x) = xT Pr x satisfy the above partial
dierential equality, which means that Pa and Pr are the extremal solutions of
the Riccati equation AT P +P A+(P B C T )(D+DT )1 (B T P C) = 0. Have
a look at Theorems 3.42, 3.43 and 3.44, and Theorem 4.43. One especially
deduces that the set of solutions P = P T > 0 of the KYP Lemma set of
equations in (3.2) has a maximum Pr and a minimum Pa elements, and that all
other solutions satisfy 0 < Pa P Pr . What is called G+ in Theorem 3.43
and is equal to Pa and what is called G is equal to Pr (it is worth recalling
that minimality of (A, B, C, D) is required in the KYP Lemma solvability with
positive denite symmetric solutions, and that the relaxation of the minimality
requires some care; see Section 3.3). Similarly P and P + in Theorem 3.44
are equal to Pa and Pr respectively.
The following is a consequence of Theorem 2.2 and relates to a notion
introduced at the beginning of this book for input-output systems, to the
notion of dissipativity introduced for state space systems.
Theorem 4.50 (Passive systems). Suppose that the system () in (4.20)
is dissipative with supply rate w(u, y) = uT y and storage function V () with
V (0) = 0, i.e. for all t 0:

t
V (x(t)) V (x(0)) + uT (s)y(s)ds (4.42)
0
208 4 Dissipative Systems

Then the system is passive.

Passivity is dened in Denition 2.1. Let us recall that a positive real (PR)
system is passive; see Corollary 2.35.
Denition 4.51 (Strictly state passive systems). A system () in
(4.20) is said to be strictly state passive if it dissipative with supply rate
w = uT y and the storage function V () with V (0) = 0, and there exists a
positive denite function S(x) such that for all t 0:

t
t
V (x(t)) V (x(0)) + uT (s)y(s)ds S(x(t))dt (4.43)
0 0

If the equality holds in the above and S(x) 0, then the system is said to be
lossless .
Some authors [228] also introduce a notion of weak strict passivity that is
more general than the strict state passivity: the function S(x) is replaced by
a dissipation function D(x, u) 0, D(0, 0) = 0. One gets a notion that is close
to (4.55). The notion of weak strict passivity is meant to generalize WSPR
functions to nonlinear systems.
Theorem 4.52. [510] Suppose that the system () in (4.20) is lossless with
a minimum value at x = x
such that V (x
) = 0. If the state space is reachable
from x
and controllable to x
, then Va () = Vr () and thus the storage function
0
is unique and given by V (x) = t1 w(u(t), y(t))dt with any t1 0 and u U
such that the state trajectory starting at x
at t1 is driven by u() to x = 0 at
t
t = 0. Equivalently V (x) = 0 1 w(u(t), y(t))dt with any t1 0 and u U
such that the state trajectory starting at x at t = 0 is driven by u() to x
at
t1 .
Remark 4.53. If the system () in (4.20) is dissipative with supply rate
w = uT y and the storage function V () satises V (0) = 0 with V () pos-
itive denite, then the system and its zero dynamics are Lyapunov stable.
This can be seen from the dissipativity inequality (4.22) by taking u or y
equal to zero.
Example 4.54 (passivity dissipativity). Consider H(s) = 1s 1+s . From Theo-
rem 4.18 this system has a nte Lp -gain for all 1 p + and it is dissipative
with respect to all supply rates w(u, y) = |u|p |y|p , 1 p +. However
H(s)  P R and it is not passive, i.e. it is not dissipative with respect to the
supply rate w(u, y) = uy.
A general supply rate has been introduced by [207] which is useful to
distinguish dierent types of strictly passive systems and will be useful in the
Passivity Theorems presented in the next section. Let us reformulate some
notions introduced in Denition 2.1 in terms of supply rate, where we recall
that 0.
 4.4 Dissipative Systems 209

Denition 4.55 (General Supply Rate). Let us consider a dissi-

pative system, with supply rate

w(u, y) = y T Qy + uT Ru + 2y T Su (4.44)

with Q = QT , R = RT . If Q = 0, R = Im , > 0, S = 12 Im , the

system is said to be input strictly passive (ISP), i.e.

t
t
y T (s)u(s)ds + uT (s)u(s)ds
0 0
1
If R = 0, Q = Im , > 0, S = 2 Im ,
the system is said to be output
strictly passive (OSP), i.e.

t
t
y T (s)u(s)ds + y T (s)y(s)ds
0 0

If Q = Im , > 0, R = Im , > 0, S = 12 Im , the system is said

to be very-strictly passive (VSP), i.e.

t
t
t
y T (s)u(s)ds + y T (s)y(s)ds + uT (s)u(s)ds
0 0 0

Note that Denitions 4.51 and 4.55 do not imply in general the asymptotic
2
stability of the considered system. For instance s+a
s is ISP as stated in Deni-
tion 4.55; see also Theorem 2.6. Though this will be examined at several places
of this book, let us explain at once the relationship between the nite-gain
property of an operator as in Denition 4.17, and dissipativity with respect to
a general supply rate. Assume that the system () is dissipative with respect
to the general supply rate, i.e.

t
V (x(t)) V (x(0)) [y T (s)Qy(s) + uT (s)Ru(s) + 2y T (s)Sy(s)]ds (4.45)
0

for some storage function V (). Let S = 0. Then it follows that

t
t
y T (s)Qy(s)ds uT (s)Ru(s)ds + V (x(0)) (4.46)
0 0
Let Q = Im and R = Im , > 0, > 0. Then we get

t

t T
y T (s)Qy(s)ds u (s)Ru(s)ds + V (x(0)) (4.47)
0 0
210 4 Dissipative Systems

so that the operator u

y has a nite L2 -gain with a bias equal to V (x(0)).
Dissipativity with supply rates w(u, y) = y T y + uT u will be commonly
met, and is sometimes called the H -behaviour supply rate of the system.
Therefore dissipativity with Q = Im and R = Im and S = 0 implies
nite-gain stability. What about the converse? The following is true:

Theorem 4.56. [206] The system is dissipative with respect to the

general supply rate in (4.44) with zero bias ( = 0) and with Q < 0,
if and only if it is nite-gain stable.

We note that the constant k in Denition 4.17 may be zero, so that no

condition on the matrix R is required in this Theorem. The = implication
has been shown just above. The  t = implication holds because of zero bias.
Then it can be shown that 0 0 [y T (s)Qy(s)+ uT (s)Ru(s)+ 2y T (s)Sy(s)]ds.
Dissipativity is here understood in the sense of Hill and Moylan in Denition
4.22.

Remark 4.57. A dynamical system may be dissipative with respect to several

supply rates, and with dierent storage functions corresponding to those sup-
ply rates. Consider for instance a linear time invariant system that is asymp-
totically stable: it may be SPR (thus passive) and it has a nite gain and is
thus dissipative with respect to a H supply rate.

Let us make an aside on linear invariant systems. A more general version

of Theorem 3.44 is as follows. We consider a general supply rate with Q 0
and R = R + SD + DT S + DT QD > 0. We denote S = S + DT Q. Then

Theorem 4.58. [531] Consider the system (A, B, C, D) with A asymptoti-

cally stable. Suppose that

t

t T
w(u(s), y(s))ds u (s)u(s)ds + (x0 ) (4.48)
0 2 0
where () 0 and (0) = 0. Then
There exists a solution P 0 to the ARE

AT P + P A + (P B C T S T )RT (B T P SC) C T QC = 0 (4.49)

such that A
= A + B R1 (B T P SC) is asymptotically stable, and

there exists a solution P > 0 to the ARI

AT P + P A + (P B C T S T )RT (B T P SC) C T QC < 0 (4.50)

 4.4 Dissipative Systems 211

Conversely, suppose that there exists a solution P 0 to the ARE (4.49)

such that the matrix A
= A + B R1 (B T P SC) is asymptotically stable.
Then the matrix A is asymptotically stable and the system (A, B, C, D)
satises (4.48) with the above supply rate.

We shall see in Section 4.5 and Chapter 5 that this can be generalized to a
class of nonlinear systems.

4.4.4 Examples

Example 4.59. At several places we have insisted on the essential role played
by the constant in Denition 2.1. Let us illustrate here how it may inuence
the Lyapunov stability of dissipative systems. For instance let us consider the
following example, brought to our attention by David Hill, where the open-
loop system is unstable:



x(t) = x(t) + u(t)



x(t)
y(t) = 1+x 4 (t) (4.51)





x(0) = x0
with > 0. Let us note that
 t1  t1 x(t)
t0
u(t)y(t)dt = t0
(x(t) x(t)) 1+x 4 (t) dt

(4.52)
2 [arctan(x2 (t1 )) arctan(x2 (t0 ))]
Thus the system is passive with respect to the storage function V (x) =

2(2 arctan(x2 )) and V (x) > 0 for all nite x IRn . Hence the system is
dissipative despite the fact that the open-loop is unstable. Note however that
V (0) = (0) < 0 and that the system loses its observability at x = .
We shall come back later on conditions that assure the stability of dissipative
systems.

Example 4.60. [206] The system is now given by

x(t) = x(t) + 1+xx(t)
4 (t) + 2u(t)




x(t)
y(t) = 1+x 4 (t) + u(t)
(4.53)





x(0) = x0
with > 0. Then we get that
212 4 Dissipative Systems

t
( 2 uT (s)u(s) y T (s)y(s))ds V (x(t)) V (x0 ) (4.54)
0

with the same V (x) as in the previous example. Thus the system is weakly
nite-gain stable, but the unique equilibrium of the uncontrolled system x = 0
is Lyapunov unstable. We notice that the system in (4.53) is not passive.
Therefore weak nite-gain stability is not sucient to guarantee the Lyapunov
stability.

In view of the above generalizations of the dissipativity and supply rate,

a dissipation equality that is more general than the one in Denition 4.51
can be introduced with a so-called dissipation function D(x, u, t) 0 for all
x X, admissible u, and t 0, such that along trajectories of the system
() one gets

t
V (x(t), t) = V (x(0), 0) + w(u(s), y(s))ds + D(x(0), u, t) (4.55)
0

Example 4.61. Let us continue with Example 4.40. Let us consider the storage
functions V (x) = 12 Cx2 , with 2 3 C 2 + 3. It is easily computed that
t
the dissipation function is D(x, u, t) = 0 [C(x(s)c u(s))2 +Rc u2 (s)]ds, with
1 2
c = C12C and Rc = 2 Cc . The choice for this notation stems from the
electrical circuit interpretation where C is a capacitor and Rc is a resistance.
It is worth noting that for each value of the coecient C, then there is a
dierent physical realization (dierent resistors, capacitors), but the state
equations (4.35) remain the same. Comparing with Denition 4.51, one has
S(x) = x2 when C = 1. Comparing with the ISP Denition 4.55 one has
= Rc , provided Rc > 0. An interesting interpretation
is in term
of phase
lag. Let us choose the two outputs as y1 = Rc u and y2 = C(x c u).
Then the transfer function between y2 (s) and u(s) (the Laplace transforms
of
both signals) is equal to C 11+s c c s
. As C varies from 2 3 to 2 + 3, c

varies monotonically from 12 ( 3 + 1) to 12 ( 3 1). Thus the phase lag of
y2 (s) with respect to u(s) increases monotonically with C. Let us now study
the variation of the dissipation function D(x, u, t) with C. For small C the
low-dissipation trajectories are those for which ||x|| is decreasing. For large C,
the low-dissipation trajectories are those for which ||x|| is increasing. There
are two extreme cases, as expected: when C = 2 3, then V (x) = Va (x) and
it is possible to drive the state to the origin with an arbitrarily small amount
of dissipation. In other words, the stored energy can be extracted from the
system. Doing the converse (driving the state from the origin to some other
state)
produces a large amount of dissipation. The other extreme is for C = 2+
3, then V (x) = Vr (x). In this case any state is reachable from the origin with
an arbitrarily small amount of dissipation. The converse (returning the state
to the origin) however dissipates signicantly. This illustrates that for small C
 4.4 Dissipative Systems 213

the dissipation seems to be concentrated at the beginning of a trajectory which

leaves the origin x = 0 and returns back to the origin, and that the opposite
behaviour occurs when C is large. This simple example therefore allows one
to exhibit the relationship between phase lag and dissipation delay.

Example 4.62. If a system (A, B, C, D) is SPR and the vector relative de-
gree r = (1 ... 1)T IRm (i.e. D = 0), then the system is OSP. Indeed
from the KYP Lemma 3.11, dening V (x) = xT P x one obtains V (x(t)) =
xT (t)(QQT + L)x(t) + 2y T (t)u(t) along the systems solutions. Integrating
and taking into account that L = 2P is full rank, the result follows. It is
noteworthy that the converse is not true. Any transfer function of the form
s+ s
2
s +as+b , b > 0, 0 < a < 2 b is SPR if and only if 0 < < a. However s2 +s+1
is not SPR (obvious!) but it denes an OSP system. One realization is given
by x1 (t) = x2 (t), x2 (t) = x1 (t) x2 (t) + u(t), y(t) = x2 (t). One checks
t t
that 0 u(s)y(s)ds 21 (x21 (0) + x22 (0)) + 0 y 2 (s)ds. Thus SPRness is only
sucient for OSPness, but it is not necessary.

Example 4.63. Consider the non-proper system y(t) = u(t) + au(t), a > 0,
with relative degree r = 1. This system is SSPR and ISP since Re [j + a] =
a and

t
t
u2 (t)
u(s)y(s)ds = + a u2 (s)ds
2
0 0

This plant belongs to the descriptor-variable systems (see Section 3.1.5),

with state space representation:


x1 (t) = x2 (t)



0 = x1 (t) + u(t)





y(t) = x2 (t) + au(t)
This can be rewritten as
    

10 x1 (t) 0 1 x1 (t) 0
= +
00 x2 (t) 1 0 x2 (t) u(t)
(4.56)



y(t) = (0 1)x(t) + au(t)
This system is regular since det(sE A) = 1. The conditions of Proposition
3.15 and of Theorem 3.16 can be checked on this example. PRness can be
checked with P = 0, while SSPRness amounts to nding p21 > 0, p11 = p22 ,
and w21 such that w21 + < 0, with = (p11 p22 )2 (p11 p22 )(p22 1),
= (p11 p22 )2 a + p21 (p11 p22 )(p22 1) + p21 (p22 1)2 .

Example 4.64. If a system (A, B, C, D) is SPR and if the matrix

214 4 Dissipative Systems

Q + LLT LW
Q =

W T LT D + D T
is positive denite with Q = AT P P A, then the system is 
VSP. This can be
x
proved by using again V (x) = xT P x. Let us denote x = . Dierentiating
u
and using the KYP Lemma 3.11, one gets V (x(t)) = xT (t)Qx(t)+2y T (t)u(t).
One deduces that

t1
t1
t1
uT (t)y(t)dt V (x(t0 )) + uT (s)u(s)ds + y T (s)y(s)ds
t0 t0 t0

for some > 0 and > 0 small enough 2 . Note that the condition Q > 0
implies that the vector relative degree of (A, B, C, D) is equal to (0 ... 0)T ,
which implies that the matrix D = 0. Indeed D + DT = W T W and W = 0
implies that Q does not have full rank. In the monovariable case m = 1, then
r = 0. In the multivariable case, Q > 0 implies that W has full rank m. Indeed
we can rewrite Q > 0 as xT (Q + LLT )x + uT W T W u 2xT LW u > 0. If W
has rank p < m, then we can nd a u = 0 such that W u = 0. Therefore for
the couple x = 0 and such a u, one has xT Qx = 0 which contradicts Q > 0.
We deduce that r = (0 ... 0)T IRm . VSP linear invariant systems possess a
uniform relative degree 0.
Example 4.65. If a system (A, B, C, D) is SPR, then it is strictly passive with
S(x) = xT Qx. This can be proved using the KYP Lemma.
1
Example 4.66. Consider the system H(s) = s+a , a > 0 . We will now prove
that the system is H(s) is OSP. The system is described by

y(t) = ay(t) + u(t)

Let us consider the positive denite function V (y) = 12 y 2 . Then

V (y(t)) = y(t)y(t) = ay 2 (t) + u(t)y(t)

Integrating we obtain

t
t
V (y(0)) V (y(t)) V (y(0)) = a y 2 (s)ds + u(s)y(s)ds
0 0

t
t
= u(s)y(s)ds + V (0) a y 2 (s)ds
0 0
Thus the system is OSP. Taking a = 0, we can see that the system whose
transfer function is 1s , denes a passive system (the transfer function being
PR).
2
Once again we see that the system has zero bias provided x(t0 ) = 0. But in
general (x(t0 )) = 0.
 4.4 Dissipative Systems 215

Remark 4.67. As we saw in Section 2.9 for linear systems, there exists a re-
lationship between passive systems and L2 gain [125]. Let : u y be a
passive system as in Denition 2.1. Dene the input-output transformation
u = w + z, y = w z, (compare with (2.85)) then

t
t
T
u (s)y(s)ds = ( 2 wT (s)w(s) z T (s)z(s))ds
0 0

which is equivalent to

t
t
z (s)z(s)ds
T
2 wT (s)w(s)ds
0 0

which means that the system : w
z has a nite L2 gain.

Example 4.68 (L2 -gain). Let us consider the system x(t) = x(t) + u(t),
y(t) = x(t). This system is dissipative with respect to the H supply rate
w(u, y) = 2 u2 y 2 if and only if there exists a storage function V (x) such
t
that 0 ( 2 u2 ( )y 2 ( ))d V (x(t))V (x(0)). Equivalently the innitesimal
dissipation inequality holds, i.e. 2 u2 (t) y 2 (t) V (x(t))(x(t) + u(t)) 0.
Consider V (x) = px2 . The innitesimal dissipation inequality then becomes
2 u2 (t) x2 (t) 2px(t)(x(t)
0. In a matrix form this is equivalent
+ u(t))
2p 1 p
to having the matrix 0. This holds if and only if
2
p
2 (2p 1) p2 0 (4.57)
2
This polynomial in p has a real solution if and only if 1. This polynomial
is a Riccati inequality whose solvability is equivalent to 2 1. The system
has an L2 gain equal to 1, and the condition that 2 1 agrees with this.
Indeed the fact that the system is dissipative with respect to the above H
supply rate implies that the H -norm of its transfer function is (this is
known as the Bounded Real Lemma; see Section 5.9).
This example together with Example 4.64 illustrates that the same system
can be dissipative with respect to several supply rates, and with dierent
storage functions.

Proposition 4.69. Consider the system represented in Figure 4.5, where ()

is a static nonlinearity, q 0 and () 0 for all IR. Then H : u
y
is passive.
t
Proof: Let us adopt the classical notation u|yt = 0 u(s)y(s)ds. Then
216 4 Dissipative Systems

Fig. 4.5. A linear system and a static nonlinearity in cascade

y | ut = ()|ut

= ()|q + t
t t
=q 0 ((s))(s)ds + 0 (s)[(s)]ds (4.58)
 (t) t
=q (0)
()d + 0
(s)((s))ds

 (t)  (0)
q 0 ()d q 0 []d

where we have used the fact that (t)((t)) 0 for all t 0. Note that

V () = 0 ()d 0 and is therefore qualied as a storage function, ()
being the state of this system.

Proposition 4.70. If a system is output-strictly passive, then it is also weakly

nite gain stable, i.e. OSP WFGS.

Proof: The following upperbound can be computed:

t t
0
y 2 (s)ds + 0
u(s)y(s)ds
t 1
t y(s) 2
+ 0 u(s)y(s)ds + 2 0 ( u(s)
) dt

(4.59)
t 1
t
=+
2 0 u2 (s)ds + 2 0 y 2 (s)ds
1
Choosing = one gets

t
t
1
y 2 (s)ds + u2 (s)dt
2 0 2 0

which ends the proof.

Several results are given in [512] which concern the Lyapunov stability of
systems which are nite-gain stable. They are not presented in this section
since they rather belong to the kind of results presented in Section 5.1.
 4.4 Dissipative Systems 217

Example 4.71. Let us consider two linear systems in parallel, i.e.

y1 (t) = k1 u(t)



y2 (t) = ay2 (t) + k2 u(t) (4.60)





y(t) = y1 (t) + y2 (t)
where a > 0. Thus, for some constants and k3

t t t
0
u(s)y(s)dt = 0
u(s)y1 (s)ds + 0
u(s)y2 (s)ds
t 2
k1 0
u2 (s)ds + + k3 0
y22 (s)ds
(4.61)
t t
k1
2 0
u2 (s)ds + + k  (y 2 (s) + y22 (s))ds
0 1
t k
t
k1
2 0 u2 (s)ds + + 2 0 (y1 (s) + y2 (s))2 ds
1
where k  k3 and k  2k1 . So the system () : u
y is VSP.

4.4.5 Regularity of the Storage Functions

Until now we have not said a lot on the properties of the storage functions:
are they dierentiable (in x)? Continuous? Discontinuous? We now state some
results which guarantee some regularity of storage functions. As we already
pointed out, storage functions are potential Lyapunov functions candidate. It
is well-known that Lyapunov functions need not be smooth, neither dieren-
tiable.

Continuous Storage Functions

Probably the rst result in this direction is the following Lemma, for which
we rst need a preliminary denition.

Denition 4.72. [209] A function V : X IR is called a virtual storage

function if it satises V (0) = 0 and

t1
V (x0 ) + w(u(s)y(s)ds V (x1 ) (4.62)
t0

for all t1 t0 and all admissible u(), with x(t0 ) = x0 and x(t1 ) = x1 .

Clearly if in addition one imposes that V (x) 0 then one gets storage
functions.
218 4 Dissipative Systems

Lemma 4.73. [209] Let the system () be locally w-uniformly reachable in

the sense of Denition 4.44. Then any virtual storage function which exists
for all x X is continuous.

Proof: Consider an arbitrary state x0 X, and let a virtual storage function

be V (). Then for any x1 in a neighborhood of x0 , it follows from (4.62)
that
t1
V (x0 ) + w(u(s), y(s)ds V (x1 ) (4.63)
t0

where the time t1 corresponds to t in (4.36) and the controller u() is the
one in Denition 4.44 (in other words, replace [0, t] in (4.36) by [t0 , t1 ]). From
(4.36) and (4.63) and considering transitions in each direction between x0 and
x1 , one deduces that | V (x1 ) V (x0 ) | ( x1 x0 ). Since x1 is arbitrary
in and since () is continuous, it follows that V () is continuous at x0 .

The next result concerns storage functions. Strong controllability means

local wuniform reachability in the sense of Denition 4.44, plus reachability,
plus controllability. We recall that a system is controllable if every state x X
is controllable, i.e. given x(t0 ), there exists t1 t0 and an admissible u()
on [t0 , t1 ] such that the solution of the controlled system satises x(t1 ) =
0 (sometimes this is named controllability to zero). Reachability is dened
before Denition 4.36. Dissipativity in the next Theorem, is to be understood
in Hill and Moylans way; see (4.24).

Theorem 4.74. [209] Let us assume that the system () in (4.20) is strongly
controllable. Then the system is cyclo-dissipative (resp. dissipative in the sense
of Denition 4.22) if and only if there exists a continuous function V : X IR
satisfying V (0) = 0 (resp. V (0) = 0 and V (x) 0 for all x X) and
V (x(t)) w(u(t), y(t)) for almost all t 0 along the systems trajectories.

A relaxed version of Theorem 4.74 is as follows:

Theorem 4.75. [401] Let the system x(t) = f (x(t), u(t)) be dissipative in
the sense of Denition 2.1 with supply rate w(x, u), and locally wuniformly
reachable at the state x
. Assume that for every xed u, the function f (, u) is
continuously dierentiable, and that both f (x, u) and fx (x, u) are continuous
in x and u. Then the set R(x
) of states reachable form x
is an open and
connected set of X, and there exists a continuous function V : R(x
) IR+
such that for every x0 R(x
) and every admissible u()

t
V (x(t)) V (x0 ) w(x(s), u(s))ds (4.64)
0

along the solution of the controlled system with x(0) = x0 . An example of

such a function is Vr (x) + , where is a suitable constant and Vr (x) is the
required supply as in Denition 4.36.
 4.4 Dissipative Systems 219

We have already stated the last part of the Theorem in Lemma 4.45.
The proof of Theorem 4.75 relies on an extended version of the continuous
dependence of solutions with respect to initial conditions, and we omit it
here. Let us now state a result that is more constructive, in the sense that it
relies on veriable properties of the system. Before this, we need the following
intermediate Proposition.

Proposition 4.76. [401] If the linearization of the vector eld f (x) + g(x)u
around x = 0, given by z(t) = Az(t) + Bu(t) with A = f g
x (0) and B = x (0),
is controllable, then the system () in (4.20) is locally wuniformly reachable
at x = 0.

Of course, controllability of the tangent linearization is here equivalent

to having the Kalman matrix of rank n. This sucient condition for local
wuniform reachability is easy to check, and one sees in passing that all time-
invariant linear systems which are controllable, also are local wuniformly
reachable. Then the following is true, where dissipativity is understood in Hill
and Moylans sense; see (4.24):

Corollary 4.77. [401] Let the system () be dissipative and suppose its tan-
gent linearization at x = 0 is controllable. Then there exists a continuous
storage function dened on the reachable set R(x
).

Renements and generalizations can be found in [402]. In Section 4.5 gener-

alizations of the Kalman-Yakubovich-Popov Lemma will be stated which hold
under the restriction that the storage functions (see then as the solutions of
partial dierential inequalities) are continuously dierentiable (of class C 1 on
the state space X). It is easy to exhibit systems for which no C 1 storage func-
tion exists. This will pose a diculty in the extension of the KYP Lemma,
which relies on some sort of innitesimal version of the dissipation inequality.
Indeed the PDIs will have then to be interpreted in a more general sense.
More will be said in section 4.6. Results on dissipative systems depending
on time-varying parameters, with continuous storage functions may be found
in [302].

Dierentiable Storage Functions

Let us end this section on regularity with a result that shows that in the one-
dimensional case, the existence of locally Lipschitz storage functions implies
the existence of continuous storage functions whose restriction to IRn \{x = 0}
is continuously dierentiable. Such a set of functions is denoted as C01 . We
specialize here to systems which are dissipative with respect to the supply rate
w(u, y) = 2 uT uy T y. This is a particular choice of the general supply rate in
(4.44). In the dierentiable case, the dissipation inequality in its innitesimal
form is
220 4 Dissipative Systems

V T (x(t))[f (x(t) + g(x(t))u] 2 uT (t)u(t) y T (t)y(t) (4.65)

Let us dene the following generalized derivative of the (non-dierentiable)

function V () at x
1
V (x) = lim inf [V (x + h) V (x) T h] (4.66)
h0 |h|
where IRn . When V (x) 0, one calls a viscosity subgradient of V () at
x. The set of all such vectors , i.e. D V (x), is possibly empty, but can also
be non-single-valued (in other words: multivalued!). The viscosity subgradient
is also sometimes called a regular subgradient [415, Equation 8(4)]. In case the
function V () is proper convex, then the viscosity subgradient is the same as
the subgradient from convex analysis dened in (3.188) [415, Proposition 8.12],
and if V () is dierentiable it is the same as the usual Euclidean gradient. An
introduction to viscosity solutions is given in Section A.3 in the Appendix.
With this machinery in mind, one may naturally rewrite (4.65) as

T [f (x(t) + g(x(t))u] 2 uT (t)u(t) y T (t)y(t), V (x) (4.67)

for all x X \ {0} and all admissible u() (see Proposition A.52 in the Ap-
pendix). If the function V () is dierentiable, then (4.67) becomes the usual
innitesimal dissipation inequality V T (x)[f (x(t)+g(x(t))u] 2 uT (t)u(t)
y T (t)y(t). As we saw in Section 3.9.4, it is customary in nonsmooth and con-
vex analysis, to replace the usual gradient by a set of subgradients. The set
of all continuous functions V : IRn IR+ that satisfy (4.67) is denoted as
W(, 2 ). The set of all functions in W(, 2 ) which are proper (radially
unbounded) and positive denite, is denoted as W (, 2 ).

Theorem 4.78. [418] Let n = m = 1 in (4.74) and assume that the vector
elds f (x) and g(x) are locally Lipschitz. Assume that for some > 0 there
exists a locally Lipschitz V W (, 2 ). Then W (, 2 ) C01 = .

The proof is rather long and technical so we omit it here. This result means
that for scalar systems, there is no gap between locally Lipschitz and C01 cases.
When n 2 the result is no longer true as the following examples prove [418].

Example 4.79. [418] Consider the system (1 ) with n = m = 2:

x1 (t) =| x1 (t) | (x1 (t)+ | x2 (t) | +u1 (t))
(4.68)

x2 (t) = x2 (t)(x1 (t) | x2 (t) | +u2 (t))
Let us dene V1 (x) = 2 | x1 | +2 | x2 |, which is a proper, positive denite
and globally Lipschitz function. Moreover V1 W (1 , 1). However it is not
C01 and any function that is C01 and which belongs to W(1 , 1), is neither
positive denite nor proper [418, Proposition 2.2].
 4.4 Dissipative Systems 221

Example 4.80. [418] Consider the system (2 ) with n = m = 2:

x1 (t) = x1 (t) + x2 (t) + u1 (t)
(4.69)
4
x2 (t) = 3x2 (t)(x1 (t) x2 (t) + u2 (t))
3

2
Let us consider V2 (x1 , x2 ) = x21 +x23 . This function is proper, positive denite,
and continuous. Moreover V2 W (2 , 1). However any locally Lipschitz
function in W(2 , 1) is neither positive denite nor proper.

Things are however not so dramatic as the next Theorem shows:

Theorem 4.81. [418] For any system () with locally Lipschitz vector elds
f (x) and g(x),

inf { | W (, 2 ) = } = inf { | W (, 2 ) C01 = } (4.70)

In other words, Theorem 4.81 says that, given a , if one is able to exhibit
at least one function in W (, 2 ), then increasing slightly allows one to
get the existence of a function that is both in W (, 2 ) and in C01 . This is
a sort of regularization of the storage function of a system that is dissipative
with respect to the supply rate w(u, y) = 2 uT u y T y.

Remark 4.82. The results hold for systems which are ane in the input, as in
(4.20). For more general systems they may not remain true.

Example 4.83. Let us lead some calculations for the system and the Lyapunov
function of Example 4.79. We get

2 or 2 or [2, 2]
V1 (x) =
2 or 2 or [2, 2]
(4.71)

xi > 0 xi < 0 xi = 0
Thus the left hand side of (4.67) is

1 |x1 |(x1 + |x2 | + u1 )
(4.72)

2 x2 (x1 |x2 | + u2 )
Thus we may write the rst line, taking (4.71) into account, as
222 4 Dissipative Systems


2(x21 + x1 |x2 | + x1 u1 ) if x1 > 0



2(x21 x1 |x2 | x1 u1 ) if x1 < 0 (4.73)





[2|x1 |(x1 + |x2 | + u1 ); 2|x1 |(x1 + |x2 | + u1 )] = {0} if x1 = 0

and similarly for the second line. It happens that V () is not dierentiable at
x = 0, and that f (0)+g(0)u = 0. Let y1 = x1 , y2 = x2 . Consider the case x1 >
0, x2 > 0. We obtain 2y T y + 2y T u 2y T y + y T y + uT u = y + y T y + uT u.
For x2 > 0 and x1 = 0 we obtain 2y22 +2y2 u2 y+y T y+uT u = y22 +uT u.

4.5 Nonlinear KYP Lemma

4.5.1 A Particular Case

The KYP Lemma for linear systems can be extended for nonlinear systems
having state-space representations ane in the input. In this section we will
consider the case when the plant output y is not a function of the input u. A
more general case will be studied in the next section. Consider the following
nonlinear system


x(t) = f (x(t)) + g(x(t))u(t)



() y(t) = h(x(t)) (4.74)





x(0) = x0
where x(t) IRn , u(t) IRm , y(t) IRm , f : IRn IRn with f (0) = 0,
h(0) = 0, g : IRn IRnm , h : IRn IRm , are smooth functions of x. We
then have the following result.
Lemma 4.84 (KYP Lemma for nonlinear systems). Consider the non-
linear system (4.74). The following statements are equivalent.
(1) There exists a C 1 storage function V (x) 0, V (0) = 0 and a function
S(x) 0 such that for all t 0:

t
t
V (x(t)) V (x(0)) = y T (s)u(s)ds S(x(s))ds (4.75)
0 0

The system is strictly passive for S(x) > 0, passive for S(x) 0 and
lossless for S(x) = 0.
(2) There exists a C 1 non-negative function V : X IR with V (0) = 0, such
that
Lf V (x) = S(x)
(4.76)

Lg V (x) = hT (x)
 4.5 Nonlinear KYP Lemma 223
V (x)
where Lg V (x) = x g(x).

Remark 4.85. Note that if V (x) is a positive denite function (i.e. V (x) > 0),
then the system x(t) = f (x(t)) has a stable equilibrium point at x = 0. If in
addition S(x) > 0 then x = 0 is an asymptotically stable equilibrium point.

Proof of Lemma 4.84:

(1) (2). By assumption we have

t
t
V (x(t)) V (x(0)) = y (s)u(s)ds
T
S(x(s))ds (4.77)
0 0

Taking the derivative with respect to t and using (4.74)

d(V x) V (x)
dt (t) = x x(t)

V (x)
= x (f (x(t)) + g(x(t))u(t))
(4.78)

= Lf V (x(t)) + Lg V (x(t))u(t)

= y T (t)u(t) S(x(t)) (see (4.74))

Taking the partial derivative with respect to u, we get Lf V (x) = S(x)
and therefore Lg V (x) = hT (x).
(2)(1). From (4.74) and (4.76) we obtain

d(V x)
(t) = Lf V (x(t)) + Lg V (x(t))u(t) = S(x(t)) + hT (x(t))u(t)
dt
Integrating the above we obtain (4.74).

Remark 4.86. From these developments, the dissipativity equality in (4.75) is

equivalent to its innitesimal version V = Lf V + Lg V u = hT (x)u(t) S(x) =
u, y S(x). Obviously this holds under the assumption that V () is suf-
ciently regular (dierentiable). No dierentiability is required in the gen-
eral Willems Denition of dissipativity, however. Some authors [228] system-
atically dene dissipativity with C 1 storage functions satisfying (||x||)
V (x) (||x||) for some class-K functions, and innitesimal dissipation
equalities or inequalities. Such a denition of dissipativity is therefore much
more stringent than the basic denitions of Section 4.4.

4.5.2 Nonlinear KYP Lemma in the General Case

We will now consider the more general case in which the system is described
by the following state-space representation ane in the input:
224 4 Dissipative Systems

x(t) = f (x(t)) + g(x(t))u(t)
() (4.79)

y(t) = h(x(t)) + j(x(t))u(t),
where x(t) IRn , u(t) IRm , y(t) IRm , and f : IRn IRn , g : IRn
IRnm , h : IRn IRm , j : IRn IRmm , are smooth functions of x with
f (0) = 0, h(0) = 0. What follows may be seen as settling the material of
Deniton 2.1, Theorem 2.2 and Corollary 2.3 in the context of dissipative
systems.
Assumption 3 The state space of the system at (4.79) is reachable from
the origin. More precisely given any x1 and t1 , there exists t0 t1 and an
admissible control u() such that the state can be driven from x(t0 ) = 0 to
x(t1 ) = x1 .
Assumption 4 The available storage Va (), when it exists, is a dierentiable
function of x.
These two assumptions are assumed to hold throughout this section. Con-
sider the general supply rate:

w(u, y) = y T Qy + 2y T Su + uT Ru
  (4.80)
  Q S y
= y T uT
ST R u
with Q = QT , R = RT . We then have the following Theorem which is due to
Hill and Moylan [207], and concerns the dissipativity as in Denition 4.22.
 4.5 Nonlinear KYP Lemma 225

Lemma 4.87 (NL KYP Lemma: general case). The nonlinear

system (4.79) is dissipative in the sense of Denition 4.22 with respect
to the supply rate w(u, y) in (4.80) if and only if there exists functions
V : IRn IR, L : IRn IRq , W : IRn IRqm (for some integer q),
with V () dierentiable, such that:

V (x) 0

V (0) = 0

V T (x)f (x) = hT (x)Qh(x) LT (x)L(x) (4.81)

1 T
2 g (x)V (x) = S T (x)h(x) W T (x)L(x)

R(x) = W T (x)W (x)

where


S(x) = Qj(x) + S
(4.82)

T T T
R(x) = R + j (x)S + S j(x) + j (x)Qj(x)

Proof:
Suciency. From (4.80), (4.79), (4.81) and (4.82) we obtain

w(u, y) = y T Qy + 2y T Su + uT Ru

= (h(x) + j(x)u)T Q(h(x) + j(x)u) + 2(h(x) + j(x)u)T Su + uT Ru

= hT (x)Qh(x) + 2uT j T (x)Qh(x) + uT j T (x)Qj(x)u + uT Ru+

+2uT j T (x)Su + 2hT (x)Su

= hT (x)Qh(x) + 2uT j T (x)Qh(x) + uT R(x)u + 2hT (x)Su

(4.83)
so that
226 4 Dissipative Systems

w(u, y) = V T (x)f (x) + LT (x)L(x) + uT R(x)u + 2uT [S T + j T (x)Q]h(x)

= V T (x)f (x) + LT (x)L(x) + uT R(x)u + 2uT S T (x)h(x)

= V T (x)f (x) + LT (x)L(x) + uT W T (x)W (x)u + uT g T (x)V (x)+

+2uT W T (x)L(x)

= V T (x)x + (L(x) + W (x)u)T (L(x) + W (x)u)

V T (x)x = V (x)
(4.84)
Integrating the above we get

t
w(s)ds V (x(t)) V (x(0)) (4.85)
0

Necessity. We will show that the available storage function Va (x) is a solu-
tion to the set of equations (4.81) for some L() and W (). Since the system
is reachable from the origin, there exists u(.) dened on [t1 , 0] such that
x(t1 ) = 0 and x(0) = x0 . Since the system (4.79) is dissipative it satises
(4.24), then there exists V (x) 0, V (0) = 0 such that:
t 0 t
t1
w(s)ds = t1 w(t)dt + 0 w(s)ds

V (x(t)) V (x(t1 ))

0
t
Remember that t1 w(s)ds is the energy introduced into the system. From
the above we have
t
0
w(s)ds w(t)dt
0 t1

The right-hand side of the above depends only on x0 . Hence, there exists a
bounded function C() IR such that

t
w(s)ds C(x0 ) >
0

Therefore the available storage is bounded:

t 8
0 Va (x) = sup w(s)ds < +.
x=x(0),t1 0,u 0

Dissipativeness in the sense of Denition 4.22 implies that Va (0) = 0 and the
available storage Va (x) is itself a storage function, i.e.
 4.5 Nonlinear KYP Lemma 227

t
Va (x(t)) Va (x(0)) w(s)ds t 0
0

or
t
dVa
0 (w(s) (s))ds t 0
0 dt
Since the above inequality holds for all t 0, taking the derivative in the
above it follows that
d(Va x)
0 w(u, y) = d(x, u)
dt
Introducing (4.79)
d(Va x)
d(x, u) = w(u, y) dt

= w[u, h(x) + j(x)u] Va

x (x) [f (x) + g(x)u] (4.86)

Since d(x, u) 0 and since w(u, y) = y T Qy + 2y T Su + uT Ru, it follows that

d(x, u) is quadratic in u and may be factored as

d(x, u) = [L(x) + W (x)u]T [L(x) + W (x)u]

for some L(x) IRq , W (x) IRqm and some integer q. Therefore from the
two previous equations and the system (4.79) and the Denitions in (4.82) we
obtain

d(x, u) = V T
x (x) [f (x) + g(x)u] + (h(x) + j(x)u) Q(h(x) + j(x)u)+
a

+2(h(x) + j(x)u)T Su + uT Ru

= VaT (x)f (x) VaT (x)g(x)u + hT (x)Qh(x)+

 
+2hT (x) [Qj(x) + S] u + uT R + j T (x)S + S T j(x) + j T (x)Qj(x) u

= VaT (x)f (x) VaT (x)g(x)u + hT (x)Qh(x)+

+2hT (x)S(x)u + uT R(x)u

= LT (x)L(x) + 2LT (x)W (x)u + uT W T (x)W (x)u

(4.87)
which holds for all x, u. Equating coecients of like powers of u we get:
228 4 Dissipative Systems

VaT (x)f (x) = hT (x)Qh(x) LT (x)L(x)

1 T
2 g (x)Va (x) = S T (x)h(x) W T (x)L(x) (4.88)

R(x) = W T (x)W (x)

which concludes the proof.

If cyclo-dissipativity is used instead of dissipativity, then the rst two
conditions on the storage function V () can be replaced by the single condition
that V (0) = 0 [209]. Consequently, Lemma 4.87 proves that:

Hill-Moylans dissipativity + reachability from x = 0 + C 1 available

storage

Willems dissipativity with one C 1 storage function V () with
V (0) = 0.

Actually the Lemma proves the sense, and the sense is obvious. Using
the suciency part of the proof of the above Theorem we have the following
Corollary, which holds under Assumptions 3 and 4:

Corollary 4.88. [207] If the system (4.79) is dissipative with respect to the
supply rate w(u, y) in (4.80), then there exists V (x) 0, V (0) = 0 and some
L : X IRq , W : X IRqm such that
d(V x) T
= [L(x) + W (x)u] [L(x) + W (x)u] + w(u, y).
dt

Under the conditions of Corollary 4.88, the dissipation function in (4.55) is

t T
equal to D(x(0), u, t) = 0 [L(x(s)) + W (x(s))u(s)] [L(x(s)) + W (x(s))u(s)]
ds. What about generalizations of the KYP Lemma when storage functions
may not be dierentiable (even possibly discontinuous)? The extension passes
through the fact that the conditions (4.81) and (4.82) can be rewritten as a
partial dierential inequality which is a generalization of a Riccati inequation
(exactly as in Section 3.1.2 for the linear time invariant case). Then relax the
notion of solution to this PDI to admit continuous (or discontinuous) storage
functions. See Section 4.6.

Remark 4.89. The Lemma 4.84 is a special case of Lemma 4.87 for
1
Q = 0, R = 0, S = I, j = 0
2
 4.5 Nonlinear KYP Lemma 229

In that case (4.81) reduces to

V T (x)f (x) = LT (x)L(x) = S(x)
(4.89)

g T (x)V (x) = h(x)
Remark 4.90. If j(x) 0, then the system in (4.79) cannot be ISP (that
corresponds to having R = I in (4.80) for some > 0). Indeed if (4.79) is
dissipative with respect to (4.80) we obtain along the systems trajectories:
d(V x)
dt (t) = w(u(t), y(t))

= hT (x(t))Qh(x(t)) L(x(t))LT (x(t)) + 2hT (x(t))S(x(t))u(t)

LT (x(t))W (x(t))u(t)

= (y(t) j(x(t))u(t))T Q(y(t) j(x(t))u(t)) L(x(t))LT (x(t))

+2(y(t) j(x(t))u(t))T [Qj(x(t)) + S]u(t) LT (x(t))W (x(t))

= y T (t)Qy(t) 2y T (x(t))Qj(x(t))u(t) + uT (t)j T (x(t))Qj(x(t))u(t)

L(x(t))LT (x(t))

+2y T (t)Qj(x(t))u(t) + 2y T (t)Su(t) 2uT (t)j T (x(t))Qj(x(t))u(t)

2uT (t)j T (x(t))Su(t)

= y T (t)Qy(t) + 2y T (t)Su(t) uT (t)u(t)

(4.90)
If j(x) = 0 we get L(x)LT (x) = uT u which obviously cannot be satised
with x and u considered as independent variables (except if both sides are
constant and identical). This result is consistent with the linear case (a PR
or SPR function has to have relative degree 0 to be ISP).

4.5.3 Time-varying Systems

All the results presented until now deal with time-invariant systems. This
is partly due to the fact that dissipativity is a tool that is used to study
and design stable closed-loop systems, and the Krasovskii-LaSalle invariance
principle is at the core of stability proofs (this will be seen in Chapter 5). As
far as only dissipativity is in question, one can say that most of the tools we
have presented in the foregoing sections, extend to the case:

x(t) = f (x(t, t)) + g(x(t), t)u(t)
(t ) (4.91)

y(t) = h(x(t), t) + j(x(t), t)u(t)
230 4 Dissipative Systems

where the well-posedness conditions are assumed to be fullled (see section

3.9.2). The available storage and required supply are now dened as

t1
Va (t0 , x) = sup w(u(t), y(t))dt (4.92)
x=x(t0 ),u(),t1 t0 t0

and

t0
Vr (t0 , x) = inf w(u(t), y(t))dt (4.93)
u(),tt0 t

Then one has:

Lemma 4.91. Let Assumptions 3 and 4 hold for (4.91). Suppose moreover
that the required supply Vr (t, x) is continuously dierentiable on IRn IR.
The system (4.91) is dissipative in the sense of Denition 2.1 with = 0 if
and only if there exists a continuous almost everywhere dierentiable function
V : IR IRn IR, V (t, x) 0 for all (t, x) IR IRn , V (t, 0) = 0 for all
t IR, and such that

V T (x)f (x) V
t hT (x) 12 V T (x)g(x, t)
0 (4.94)
h(x) 12 g T (x, t)V (x) j(x, t) + j T (x, t)

4.5.4 Nonlinear-in-the-input Systems

So far only nonlinear systems which are linear in the input have been consid-
ered in this book. It seems that there is no KYP Lemma extension for systems
of the form


x(t) = f (x(t), u(t))



y(t) = h(x(t), u(t)) (4.95)





x(0) = x0
with f (0, 0) = 0 and h(0, 0) = 0. It is assumed that f (, ) and h(, ) are
smooth functions (innitely dierentiable).

Proposition 4.92. [303] Let = {x IRn | V x f (x, 0) = 0}. Necessary

conditions for the system in (4.95) to be passive with a C 2 storage function
V () are that
x f (x, 0) 0
(a) V
x g(x, 0) = h (x, 0) for all x
(b) V T
 4.6 Dissipative Systems and Partial Dierential Inequalities 231
n 2 fi h T
(c) V
i=1 u2 (x, 0). xi u (x, 0) + h
u (x, 0) for all x

where fi (x, u) is the ith component of the vector function f (x, u).

Proof: [303] Consider an auxiliary function F : IRn IRm IR dened as

x f (x, u) h (x, u)u. Since the system in (4.79) is passive, it is
F (x, u) = V T

clear that F (x, u) 0 for all u IRm . Therefore (a) follows by setting u = 0.
For all x , one has F (x, 0) = V x f (x, 0) = 0. Thus F (x, u) F (x, 0) = 0
for all x and for all u IRm . In other words F (x, u) attains its maximum
at u = 0 on the set . Let us now dene g0 (x) = f u (x, 0). We obtain for all
x

0= F
u (x, 0) = V f
x u (x, 0) hT (x, 0)



 
2F ((V /x)(f /u)) h T
0 u2 (x, 0) = u |u=0 h
u (x, 0) + u (x, 0) (4.96)



 

n
= 2 fi V
h h T
i=1 u2 (x, 0). xi u (x, 0) + u (x, 0)

from which (b) and (c) follow.

4.6 Dissipative Systems and Partial Dierential

Inequalities

As we have seen in Section 4.4.5, storage functions are continuous under some
reasonable controllability assumptions. However it is a much stronger assump-
tion to suppose that they are dierentiable, or of class C 1 . The versions of
the KYP Lemma that have been presented above, rely on the property that
V () is C 1 . Here we show how to relax this property, by considering the in-
nitesimal version of the dissipation inequality: this is a partial dierential
inequality which represents the extension of the KYP Lemma to the case of
continuous, non-dierentiable storage functions.

4.6.1 The linear invariant case

First of all and before going on with the nonlinear ane-in-the-input case,
let us investigate a novel path to reach the conclusions of Section 3.1.2. We
consider the linear time-invariant system

x(t) = Ax(t) + Bu(t)
(4.97)

y(t) = Cx(t) + Dx(t)
Let us dene the Hamiltonian function
232 4 Dissipative Systems

H(x, p) = sup [pT (Ax + Bu) w(u, y)] (4.98)

m
uIR

where the supply rate is chosen as w(u, y) = uT y. By rearranging terms one

gets

H(x, p) = pT Ax + sup [(pT B xT C T )u uT Du] (4.99)

m
uIR

D>0
Let us assume that D > 0 ( D + DT > 0), so that the maximizing u is
given by

u
= (D + DT )1 (B T p Cx) (4.100)
and the matrix D + D arises from the derivation of u Du. Injecting u
into
T T

H(x, p) and rewriting uT Du as 12 uT (D + DT )u, one obtains

1
H(x, p) = pT Ax + (B T p Cx)T (D + DT )1 (B T p Cx) (4.101)
2
Let us now consider the quadratic function V (x) = 12 xT P x, P = P T , and

H(x, P ) = H(x, V
x ). We obtain

1
H(x, P ) = xT P Ax + (B T P x Cx)T (D + DT )1 (B T P x Cx) (4.102)
2
Now imposing that H(x, P ) 0 for all x IRn and using xT P Ax =
1 T T
2 x (A P + P A)x we get

AT P + P A + (P B C T )(D + DT )1 (B T P C) 0, (4.103)
which is the Riccati inequality in (3.17). We have therefore shown that under
the condition D > 0 the inequality H(x, V x ) 0 is equivalent to the Riccati
inequality in (4.103), thus to the matrix inequality in (3.3).

D=0
Let us now investigate what happens when D = 0. Following the same rea-
soning one nds that the maximizing input does not exist (the function to
maximize is (pT B xT C T )u) so that it is necessary for the supremum to have
a meaning (to be dierent from +) that pT B xT C T = 0 for all x IRn .
Choosing the same storage function as above it follows that H(x, V x ) 0
yields P A + AT P 0 and P B = C T .
 4.6 Dissipative Systems and Partial Dierential Inequalities 233

D0

Let f : IRn IR {+} be a function not identically +, minorized by

an ane function. Then the conjugate function of f () is dened by [210,
Denition 1.1.1]

f
(z) = sup [z T u f (u)] (4.104)
udom(f )

Doing the analogy with (4.98) one nds f (u) = uT Du, z = B T p Cx, and
H(z) is the sum of the conjugate of f (u) and pT Ax. It is a basic result from
convex analysis that if D + DT > 0 then

f
(z) = z T (D + DT )1 z, (4.105)
from which one straightforwardly recovers the previous results and the Riccati
inequality. We also saw what happens when D = 0. Let us now investigate
the case D + DT 0. We get [210, Example 1.1.4]:

+ if z  Im(D + DT )


f (z) = (4.106)
T
z (D + DT ) z if z Im(D + DT )
where (D + DT ) is the Moore-Penrose pseudo-inverse of (D + DT ). Replacing
z by its value we obtain

H(x, p) = pT Ax+

+ if B T p Cx  Im(D + DT )
+

(B T p Cx)T (D + DT ) (B T p Cx) if B T p Cx Im(D + DT )
(4.107)
1 T
Setting p = V x and V = 2 x P x with P = P T
it follows from H(x, p) 0
for all x IRn , that P is the solution of a degenerate Riccati inequality (DRI):

(i) Im(B T P C) Im(D + DT )
(4.108)

(ii) P A + AT P + (B T P C)T (D + DT ) (B T P C) 0

Is (4.108) equivalent to the KYP Lemma conditions? The following can be

proved:
(3.2) = (4.108) (i),
The conditions in (3.2) are equivalent to
234 4 Dissipative Systems

(i) LLT LW (W T W ) W T L 0
(4.109)

(ii) LW [Im W T W (W T W ) ] = 0
whose proof can be deduced almost directly from Lemma A.65 noticing
that W T W 0.
Notice that (4.109) (ii) is equivalently rewritten as

P B C T = P B C T (D + DT )(D + DT ) (4.110)
It follows from (4.110) and standard matrix algebra [272, p.78,p.433] that
Im(B T P C) =Im[(D + DT ) (D + DT )(B T P C)] Im[(D + DT ) (D +
DT )] Im((D + DT ) ) = Im(D + DT ). Thus (4.110) (4.109) (ii)
(4.108) (i). Now obviously (4.109) (i) is nothing else but (4.108) (ii). We
therefore conclude that the conditions of the KYP Lemma in (3.2) are equiv-
alent to the degenerate Riccati inequality (4.108).
To summarize:

(ARI) in (4.103) [ KYP conditions (3.2) ]

(D > 0)

Hamiltonian function in (4.98)

D=0
= LMI in (3.2) with W = 0
(D 0)
DRI in (4.108) or RORE in (A.40)

It is worth noting that there is no minimality assumption in (4.97).

Remark 4.93. In the degenerate case D+DT 0 with rank(D+DT ) = r < m,

there exists an orthogonal transformation = [1 2 ] such that
T
1 R1 0
(D + DT )[1 2 ] = (4.111)
T
2 0 0
with R1 > 0. When H(s) is PR the transfer function T H(s) = T C(sIn
A)1 B + T D is PR [506].

Remark 4.94 (Singular optimal control). As we saw in Section 3.1.2 and Sec-
tion 3.8, the link between passivity (the KYP Lemma) and optimal con-
trol exists when R = D + DT > 0. The optimal control problem is then
regular. There must exist a link between the KYP Lemma conditions with
D + DT 0 and singular optimal control problems. We consider the optimal
control with cost function w(u, x) = uT y = uT (Cx + Du) = 12 uT Ru + xT Cu.
 4.6 Dissipative Systems and Partial Dierential Inequalities 235

Let rank(D + DT ) = r < m, and s = m r be the dimension of the singular


C1
control. Let n s and partition B and C as B = [B1 B2 ] and C = ,
C2
with B1 IRnr , B2 IRns , C1 IRrn , C2 IRsn . Then (A, B, C, D) is
PR if and only if D + DT 0 and there exists P = CB(BB) > 0 satisfying
P B = C T and
P A AT P P B1 + C1T
0 (4.112)
B1T P + C1 R1
The proof can be found in [506]. It is based on the fact that when D + DT is
not full rank, then (3.3) can be rewritten as P B2 + C2T = 0 and (4.112).
Remark 4.95. In [213] an algorithm is proposed which allows one to construct
a reduced Riccati equation for the case D + DT 0. The authors start from
the KYP Lemma LMI for the WSPR case (then indeed D is not full rank
otherwise the transfer would be SSPR). We recall this algorithm and this
important result on a degenerate Riccati equation in Appendix A.4.

4.6.2 The Nonlinear Case y = h(x)

We consider in this section the system () in (4.74). Let us rst state the
following Theorem, which shows what kind of partial dierential inequality,
the storage functions of dissipative systems (i.e. systems satisfying (4.25)) are
solutions of. Let us dene the Hamiltonian function

H(x, p) = pT f (x) + sup [pT g(x)u w(u, y)] (4.113)

u()U

Also let V
(x) = limzx inf V (z) be the lower semi-continuous envelope
of V (). A locally bounded function V : X IR is a weak or a viscosity
solution to the partial dierential inequality H(x, V ) 0 for all x X,
if for every C 1 function : X IR and every local minimum x0 IRn of
V
, one has H(x0 , x
(x0 )) 0. The PDI H(x, V ) 0 for all x X
is also called a Hamilton-Jacobi inequality. The set U plays an important
role in the study of the HJI, and also for practical reasons (for instance,
if u is to be considered as a disturbance, then it may be assumed to take
values in some compact set, but not in the whole of IRm ). Let us present
the following theorem, whose proof is inspired by [304]. Only those readers
familiar with partial dierential inequalities and viscosity solutions should
read it. The others can safely skip the proof. The next Theorem concerns the
system in (4.74), where f (), g() and h() are supposed to be continuously
dierentiable, with f (0) = 0, h(0) = 0 (thus x = 0 is a xed point of the
uncontrolled system), and f g h
x , x and x are globally bounded.

Theorem 4.96. [232] (i) If the system () in (4.74) is dissipative in the

sense of Denition 4.23 with storage function V (), then V () satises the
partial dierential inequality
236 4 Dissipative Systems

H(x, V (x)) = V T (x)f (x) + sup [V T (x)g(x)u w(u, y)] 0 in IRn

u()U
(4.114)
(ii) Conversely, if a nonnegative locally bounded function V () satises (4.114),
then () is dissipative and V
(x) is a lower semi-continuous storage function.

The suprema in (4.113) and (4.114) are computed over all admissible u().
It is noteworthy that the PDI in (4.114) is to be understood in a weak sense
(V () is a viscosity solution), which means that V () needs not be continuously
dierentiable to be a solution. The derivative is understood as the viscosity
derivative, see (4.66) and Appendix A.3.

In short, Theorem 4.96 says that a dissipative system as () in (4.74)

possesses a storage function that is at least lower semi-continuous.

Proof of Theorem 4.96:

(i) Let () C 1 (IRn ) and suppose that V
attains a local minimum at
the point x0 IRn . Let us consider a constant input u (u(t) = u for all t 0),
and let x(t) be the corresponding trajectory with initial condition x(0) = x0 .
For suciently small t 0 we get

V
(x0 ) V
(x(t)) (x0 ) (x(t)) (4.115)

since V
attains a local minimum at the point x0 IR . Since the system
n

() is dissipative in the sense of Denition 4.23 with storage function V (),

and since V
() satises the dissipation inequality each time its associated
storage V () does, it follows that

t
V
(x0 ) V
(x(t)) w(u, y(s))ds (4.116)
0
Combining (4.115) and (4.116) one obtains


(x(t)) (x0 1 t
w(u, y(s)ds 0 (4.117)
t t 0
By letting t 0, t > 0, one gets

T (x0 ) + T (x0 )g(x0 )u w(u, h(x0 )) 0 (4.118)

Since this inequality holds for all u, it follows that

H(x0 , (x0 )) = T (x0 )f (x0 ) + sup [T (x0 )g(x0 )u w(u, h(x0 ))] 0
uU
(4.119)
 4.6 Dissipative Systems and Partial Dierential Inequalities 237

holds for all u U. We have therefore proved that V is a viscosity solution

of (4.114).

(ii) Let us dene UR = {u U |  u  R}, R > 0. Let UR denote the set

of controllers with values in UR . Since V
() is lower semi continuous, there
exists a sequence {i }
i=1 of locally bounded functions such that i V
and
i V
as i +, i V
. Let > 0 and dene

8
i
ZR (x, s) = sup i (x( )) w(u(r)y(r))dr | x(s) = x (4.120)
uUR s

i
Then ZR () is continuous and is the unique solution of

Z i

t + (ZR ) (x, s)f (x) + supuUR [(ZR ) (x, s)g(x)u w(u, y)] = 0
R i T i T



in IRn (0, )





ZRi
(x, ) = i (x) in IRn
(4.121)
Compare (4.120) and (4.121) with (4.25) and (4.113) respectively. By def-
inition of a so-called viscosity supersolution, it follows that precisely V
()
is a viscosity supersolution of this partial dierential equality (roughly, be-
cause V
() upperbounds i () and is a viscosity solution of (4.114)). By the
comparison Theorem it follows for all integer i 1 that

V
(x) ZR
i
(x, s) (x, s) IRn [0, ] (4.122)
Setting s = 0 yields

8
V
(x) sup i (x( )) w(u(r), y(r))dr | x(0) = x (4.123)
uUR 0

Letting i + we obtain

8
V
(x) sup V
(x( )) w(u(r), y(r))dr | x(0) = x (4.124)
uUR 0

Letting R +

8
V
(x) sup V
(x( )) w(u(r), y(r))dr | x(0) = x (4.125)
uU 0

where we recall that U is just the set of admissible inputs, i.e. locally square
Lebesgue integrable functions of time (locally L2 ) such that (4.21) is satised.
238 4 Dissipative Systems

This last inequality holds for all 0, so that (4.25) holds. Consequently ()
is dissipative and V
() is a storage function.

When specializing to passive systems then the following holds:

Corollary 4.97. [232] The system () in (4.74) is passive if and only if

there exists a locally bounded non-negative function V () such that V (0) = 0
and

V T (x)f (x) + sup [V T (x)g(x)u uT y] 0 in IRn (4.126)

u()U

In case U = IRm then (4.126) reads

V T (x)f (x) 0
(4.127)

V T (x)g(x) = h(x)
for all x IRn .

In (4.127), solutions are supposed to be weak, i.e.: if () C 1 (IRn ) and

V
attains a local minimum at x0 IRn , then

T (x0 )f (x0 ) 0
(4.128)

T (x0 )g(x0 ) = h(x0 ).

One sees that the set of conditions in (4.128) is nothing else but (4.76) ex-
pressed in a weak (or viscosity) sense.

4.6.3 The Nonlinear Case y = h(x) + j(x)u

We now consider systems as in (4.79), and the supply rate is w(u, y) =

2 uT u y T y (Q = Im , R = 2 Im , S = 0 in Denition 4.55). The dissi-
pation inequality then reads

t
V (x(t)) V (x(0)) [ 2 uT (s)u(s) y T (s)y(s)]ds (4.129)
0

If one supposes that V (0) = 0 and x(0) = 0 then it follows from (4.129)
that

t
0 V (x(t)) [ 2 uT (s)u(s) y T (s)y(s)]ds (4.130)
0
from which one deduces that

t
t
2
y (s)y(s)ds
T
uT (s)u(s)ds (4.131)
0 0
 4.6 Dissipative Systems and Partial Dierential Inequalities 239

which simply means that the system denes an input-output operator Hx

which has a nite L2 gain at most (see Denition 4.17), and Hx=0 has
zero bias. An argument of local wuniform reachability assures that storage
functions are continuous. Let us assume that V () is a smooth storage function.
Then the dissipation inequality (4.129) is equivalent to its innitesimal form
V T (x)[f (x) + g(x)u] + (h(x) + j(x)u)T (h(x) + j(x)u) 2 uT u 0. (4.132)
Since the dissipation inequality is required to hold for a certain set U of ad-
missible inputs, the innitesimal form (4.132) is a Hamilton-Jacobi inequality
H(x, V (x)) 0, with Hamiltonian function

H(x, p) = sup [pT (f (x) + g(x)u) + (h(x) + j(x)u)T (h(x) + j(x)u) 2 uT u]

uU
(4.133)
If in addition the term (x) = 2 Im j(x)T j(x) > 0 for all x X, then
the Hamiltonian can be written in a explicit way as

H(x, p) = pT [f (x) + g(x)1 (x)j(x)T h(x)] + 14 pT g(x)1 (x)g(x)T p+

+h(x)T [Im + j(x)1 (x)j(x)T ]h(x)

(4.134)
Let us note once again that if u() is considered as a disturbance, and not
a control input, then it makes perfect sense to consider the set U in which
the disturbance is supposed to live. This is also the case if the admissible
inputs are bounded because of physical saturations. Those developments are
then at the core of the H theory for nonlinear systems [442]. Similarly to the
above, the obstacle in studying such PDIs is that storage functions may not be
dierentiable: in general they are only continuous. How does this machinery
extends to such a case? Once again weak (or viscosity) solutions are the key.
Theorem 4.98. [33] Suppose that V : X IR+ is continuous. Then V () is
a storage function for the system () in (4.79) if and only if it is a viscosity
solution of the Hamilton-Jacobi inequality H(x, V (x)) 0 for all x X,
with H(, ) given in (4.133).
Under some conditions, the available storage and required supply are
proved to be the viscosity solutions of Hamilton-Jacobi equalities, thereby
extending (4.37).
Assumption 5 Given x0 IRn and t1 < t2 with t2 t1 suciently small,
there exists a bounded set Bx0 IRm such that

t2 8
sup Va (x(t2 )) Va (x0 ) ( 2 uT (t)u(t) y T (t)y(t))dt =0
u L2 ([t1 , t2 ]) t1
u(t) Bx0
(4.135)
240 4 Dissipative Systems

where x(t) and y(t) correspond to the solution initialized at x0 and controlled
by u() on [t1 , t].
Assumption 6 Given x0 IRn and t0 < t1 with t1 t0 suciently small,
there exists a bounded set Bx0 IRm such that

t1 8
sup Vr (x0 ) Va (x(t0 )) ( 2 uT (t)u(t) y T (t)y(t))dt =0
u L2 ([t0 , t1 ]) t0
u(t) Bx0
(4.136)
where x(t) and y(t) correspond to the solution initialized at x0 and controlled
by u() on [t1 , t].

Theorem 4.99. [33] Assume that the system in (4.79) has nite-gain at most
and is uniformly controllable, so that Va () and Vr () are both well-dened
continuous storage functions. Then
Va () is a viscosity solution of H(x, V (x)) = 0 if Assumption 6 is
satised.
Vr () is a viscosity solution of H(x, V (x)) = 0 if Assumption 5 is satis-
ed.

Remark 4.100. Storage functions that satisfy (4.81) can also be shown to
be the solutions of the following partial dierential inequation:
1 1
V T (x)f (x) + (hT (x) V T (x)g(x))R1 (x)(h(x) g T (x)V (x)) 0
2 2
(4.137)
when R = j(x) + j T (x) is full-rank, R = 0, Q = 0, S = 12 I. The proof
is exactly the same as in the linear time invariant case (Section 3.1.2).
The available storage and the required supply satisfy this formula (that is
similar to a Riccati equation) as an equality (Proposition 4.48).
In the linear invariant case, the equivalent to Hamilton-Jacobi inequalities
are Riccati equations, see Section 3.1.2. This also shows the link with
optimal control. Hamilton-Jacobi equalities also arise in the problem of
inverse optimal control, see section 4.6.5.
In the time varying case (4.91), the PDI in (4.137) becomes

V
t (x, t) + V T (x, t)f (x, t)+

+(hT (x, t) 12 V T (x, t)g(x, t))R1 (x, t)(h(x, t) 12 g T (x, t)V (x, t)) 0
(4.138)
 4.6 Dissipative Systems and Partial Dierential Inequalities 241

In order to illustrate the above developments let us present an example,

taken from [116].

Example 4.101. Consider the following system

x1 (t) = x1 (t)[(r2 (t) 1)(r2 (t) 4) + r(t)(r2 (t) 4)u(t)] x2 (t)



x2 (t) = x2 (t)[(r2 (t) 1)(r2 (t) 4) + r(t)(r2 (t) 4)u(t)] + x1 (t) (4.139)





y(t) = r2 (t) 1, r = x21 + x22

In polar coordinates one gets

r(t) = r(t)(r2 (t) 1)(r2 (t) 4) + r(t)(r2 (t) 4)u(t)



(t) = 1 mod [2] (4.140)





y(t) = r2 (t) 1

The set S = {x IR2 | r = 1} is invariant under the uncontrolled

dynamics (u = 0), and is asymptotically stable. The open set R = {x IR2 |
0 < r < 2} is the largest basin of attraction of S (still with u = 0). Moreover
all points in R are reached from S in nite time by suitable control. Invariance
of S is easy to check as f (x) = x21 + x22 1 is a rst integral of the uncontrolled
system. The objective is to prove that the system in (4.139) is dissipative with
respect to the supply rate w(u, y) = 2 uT u y T y, for all 1. Let us look
for a storage function of the form V (r2 ). Thus V dV
x (x) = (2x1 2x2 ) d(r 2 ) . The
pre- Hamiltonian function P H() (that is the function to be supremized in
(4.113)) is equal to

dV 2 2
P H(r, u) = 2 r [(r 1)(r2 4)+ r(r2 4)u] 2uT u + (r2 1)2 (4.141)
d(r2 )

and the maximizing controller is

1 2 2 dV
u= 2
r (r 4)2 (4.142)
d(r2 )
So the Hamilton-Jacobi inequality in (4.114) reads on 0 < r < 2:
 2
H(r, V (r)) = r2 (r2 4) d(r
dV 2
2 ) + (r 1)
(4.143)
   2
1
1 2 r4 (r2 4)2 dV
d(r 2 ) 0
242 4 Dissipative Systems

Obviously this PDI has a solution if and only if 1. By inspection one

sees that any solution to the ordinary dierential equation r2 (r2 4) d(r
dV
2) +
2
(r 1) = 0 with minimal set condition V (1) = 0 solves this HJI. One such
solution is given by
1 3 3
V (r) = ln(r2 ) ln(4 r2 ) + ln(3) (4.144)
4 4 4
This V (r) is locally bounded on the set R, V (r) 0, it is radially un-
bounded for all x R (all states approaching the boundary of R, in par-
ticular the origin), and V (r) = 0 on the circle S. Therefore the system in
(4.139) is dissipative with respect to supply rates w(u, y) = 2 uT u y T y, for
all 1. The exhibited storage function is dierentiable. One can check by
calculation that V (r) = 1r (r2 1)2 0 along trajectories of the uncontrolled
system and for all x R. One has V (r) = 0 for all x S .
Let us summarize the developments in this section and the foregoing ones,
on the characterization of dissipative systems.

H(x, V (x)) 0 with Hamiltonian function in (4.113) or (4.133) or

(4.134)

PDI in (4.114) or (4.126) or in Theorem 4.98, general lsc storage
functions (viscosity solutions)

PDI in (4.137) or (4.76), C 1 storage functions

nonlinear KYP Lemma 4.84 or 4.87 with C 1 storage functions

Riccati inequality (3.17) for LTI systems

KYP Lemma for LTI systems

PR transfer functions

where the implications just mean that the problems are decreasing in math-
ematical complexity.
 4.6 Dissipative Systems and Partial Dierential Inequalities 243

4.6.4 Recapitulation

Let us take advantage of the presentation of this section, to recapitulate some

tools that have been introduced throughout the foregoing: Riccati inequalities,
Hamiltonian function, Popovs functions, and Hermitian forms. A Hermitian
form has the general expression

x
H(x, y) = [xT y T ] (4.145)
y

Q YT
with x IRn , y IRn , = , Q IRnn , Y IRnn , R IRnn ,
Y R
Q = QT , R = RT . Let y = P x for some P = P T IRnn . Then

H(x, P x) = 0 for all x IRn

if and only if

Q + P Y + Y T P + P RP = 0 (P = P T ).
The proof is done by calculating explicitly H(x, P x). The analogy with
(4.102) and (4.103) is straightforward (with equalities instead of inequalities).
A solution to the ARE is stabilizing if the ODE x(t) = dH dy |y=P x = 2(Y +
RP )x(t) is globally asymptotically stable. The results of Theorems 3.42, 3.43,
3.44 and 4.58 allow us to assert that stabilizing solutions exist in important
cases.
Linking this with the spectral (or Popovs) function (s) in Theorems
2.30 and 3.46, or (3.141) (3.142), we see that taking x = (jIn A)1 B and
y = Im in (4.145) (with appropriate dimensions of the matrices Y IRmn
and R IRmm ) yields that (j) is a rational Hermitian matrix valued
function dened on the imaginary axis. The positivity of (j) is equivalent
to the passivity of the system with realization (A, B, Y ), which in turn can
be characterized by a LMI (the KYP Lemma set of equations) which in turn
is equivalent to an ARI: the loop is closed!

4.6.5 Inverse Optimal Control

A particular optimal control problem

 is to nd the control input u() that
minimizes the integral action 0 [q(x(t)) + uT (t)u(t)]dt under the dynamics
in (4.74), where q(x) is continuously dierentiable and positive denite. From
standard dynamic programming arguments it is known that the optimal in-
T
put is u
(x) = 12 g T (x) V
x (x), where V
() is the solution of the partial
dierential equation, called a Hamilton-Jacobi-Bellman equation:
244 4 Dissipative Systems
. /
V
1 V
V
T
(x)f (x) T
(x)g(x)g (x) (x) + q(x) = 0 (4.146)
x 4 x x

Moreover V
(x(t)) = inf u() t [q(x( )) + uT ( )u( )]d , V
(0) = 0. One rec-
ognizes that u
(x) is nothing else but a static feedback of the passive output
of the system (4.74) with storage function V
(). Applying some of the results
in this section and in Section 5.4 one may additionally study the stability
of the closed-loop system with the optimal input (see in particular Theo-
rem 5.24). Let us consider the linear time-invariant case with quadratic cost
q(x) = xT Qx. Then one looks for storage functions of the form V (x) = xT P x.
The Hamilton-Jacobi-Bellman equation in (4.146) then becomes the Riccati
equation

P A + AT P P BB T P + Q = 0 (4.147)
The optimal controller is classically given by u
(x) = B T P x (recall that
T
V (x) = V
x (x) = 2P x). It is worth comparing (4.147) with (3.17) (take
D + DT = Im , C = 0, and the cost is P A + AT P + Q). See also (3.138).

Let us now describe the so-called inverse optimal control problem [363,365].
We are given the system

x(t) = f (x(t)) + Bu(t), x(0) = x0 (4.148)

where f () is smooth, f (0) = 0, and B is a constant matrix. We are also given
a performance index


t
T T
V = lim (x(t)) + (L (x(s))L(x(s)) + u (s)u(s))ds (4.149)
t+ 0

with (x) 0 for all x X, (0) = 0, L(0) = 0, and a feedback controller

u
(x) = k(x). (4.150)

Let us assume that u
(x) is optimal with respect to the performance index
(4.149), and let us denote the minimum value of V as (x0 ). In general,
there is not a unique L(x) and (x) for which the same controller is optimal.
In other words there may exist many dierent L(x), to which correspond
dierent (x), for which the same controller is optimal. The inverse optimal
control problem is as follows: given the system (4.148) and the controller
(4.150), a pair ((), L()) is a solution of the inverse optimal control problem
if the performance index (4.149) is minimized by (4.150), with minimum value
(x0 ). In other words, the inverse approach consists of designing a stabilizing
feedback control law, and then to show that it is optimal with respect to a
meaningful and well dened cost functional.
 4.6 Dissipative Systems and Partial Dierential Inequalities 245

Lemma 4.102. [365] Suppose that the system in (4.148) and the controller in
(4.150) are given. Then a pair ((), L()) is a solution of the inverse optimal
control problem if and only if (x) and L(x) satisfy the equations


T (x)[f (x) 12 Bk(x)] = LT (x)L(x)






12 B T (x) = k(x)
(4.151)



(0) = 0





(x) 0 for all x X

The following should not be surprising to the reader who has followed the
previous developments.

Lemma 4.103. [365] A necessary and sucient condition for the existence
of a solution to the inverse optimal control problem, is that the system

x(t) = f (x(t)) 12 Bk(x(t)) + Bu
(4.152)

y(t) = k(x(t))

be passive. If this is the case, then there exists two solutions (a (), La ()) and
(r (), Lr ()) of (4.151) such that all other solutions satisfy a (x) (x)
r (x) for all x X.

Indeed the equations in Lemma 4.102 are nothing else but the KYP Lemma
conditions for the system (4.152). The interpretation of a (x) and r (x) as
the available storage and required supply, respectively, is obvious as well. One
recovers the HJB equation (4.146) replacing g(x) by B and q(x) by LT (x)L(x).

Remark 4.104. The inverse optimal control problem was rst solved by Kalman
[248] in the case of linear systems with linear state feedback. Other works can
be found in [142].

Let us end this section with a result that completes the above ones. We
consider the system


x(t) = f (x(t)) + g(x(t))u(t)



y(t) = h(x(t)) + j(x(t))u(t) (4.153)





x(0) = x0
where all the mappings are continuously dierentiable and f (0) = 0, h(0) = 0.
Let us dene the set of stabilizing controllers:
246 4 Dissipative Systems

S(x0 ) = {u() | u U and solution of (4.153) satises x(t) 0 as t +}

We also consider a nonlinear nonquadratic performance criterion

J(x0 , u()) = [L(x(t)) + uT (t)Ru(t)]dt (4.154)
0
+
with L : IR IR , 0 < R IR
n mm
.

Theorem 4.105. [363, 502] Consider the system in (4.153) with the per-
formance index in (4.154). Let us assume that there exists a continuously
dierentiable and radially unbounded function V : IRn IR with V (0) = 0
and V (x) > 0 for all x = 0, satisfying

1
L(x) + T V (x)f (x) T V (x)g(x)R1 g T (x)V (x) = 0 (4.155)
4
Moreover let h(x) = L(x) and suppose that the new system in (4.153) is
zero-state observable. Then the origin x = 0 of the closed-loop system

x(t) = f (x(t)) g(x(t))(x(t)), x(0) = x0 , t 0 (4.156)

is globallly asymptotically stable with the feedback control input
1
u(x) = (x) = R1 g T (x)V (x) (4.157)
2
The action in (4.154) is minimized in the sense that

J(x0 , (x())) = min J(x0 , u()), x0 IRn (4.158)

u()S(x0

and we have J(x0 , (x())) = V (x0 ), x0 IRn

The extension of Theorem 4.105 towards the output feedback case is given
in [99, Theorem 6.2]. The equation in (4.155) is a Hamilton-Jacobi-Bellman
equation. Consider the Hamiltonian function

H(x, p, u) = L(x) + uT Ru + pT (f (x) + g(x)u) (4.159)

One may calculate that the HJB equation in (4.155) is in fact

min H(x, u, V (x)) = 0,

uU

using the strict convexity of the integrand in (4.154) (since R > 0), so that the
minimizing input is u(x) = 21 R1 g T (x)p. Various application examples may
be found in [502], like the stabilization of the controlled Lorenz equations,
the stabilization of the angular velocity with two actuators, and with one
actuator.
 4.7 Nonlinear Discrete-time Systems 247

4.7 Nonlinear Discrete-time Systems

The material of this section is taken mainly from [90]. The following class of
systems is considered:

x(k + 1) = f (x(k)) + g(x(k))u(k)
(4.160)

y(k) = h(x(k)) + j(x(k))u(k)
where x(k) IRn , u(k) IRm , y(k) IRm , and the functions f (), g(), h()
and j() are smooth mappings. It is assumed that f (0) = 0 and h(0) = 0.

Denition 4.106. The dynamical system in (4.160) is said dissipative with

respect to the supply rate w(u, y) if there exists a nonnegative function V :
IRn IR with V (0) = 0 called a storage function, such that for alll u IRm
and all k IN one has

V (x(k + 1)) V (x(k)) w(u(k), y(k)), (4.161)

or equivalently


k
V (x(k + 1)) V (x(0)) w(u(i), y(i)) (4.162)
i=0

for all k, u(k) and x(0). The inequality (4.162) is called the dissipation in-
equality in the discrete-time setting.

Similarly to the continuous-time case we have

Denition 4.107. The dynamical system in (4.160) is said passive if it is

dissipative with respect to the supply rate w(u, y) = uT y. It is said strictly
passive if V (x(k + 1)) V (x(k)) < uT (k)y(k) for all u(k) unless x(k) is
identically zero. Equivalently the system is strictly passive if there exists a
positive denite function S : IRn IR such that V (x(k + 1)) V (x(k))
uT (k)y(k) S(x(k)) for all u(k) and all k. It is said lossless if V (x(k +
1)) V (x(k)) = uT (k)y(k) for all u(k) and all k, equivalently V (x(k + 1))
k
V (x(0)) = i=0 uT (i)y(i) for all u(k) and all k.

It is of interest to present the extension of the KYP Lemma for such

nonlinear discrete-time systems, that is the nonlinear counterpart to Lemma
3.100.

Lemma 4.108 (KYP Lemma). [90] The system (4.160) is lossless with a
C 2 storage function if and only if
248 4 Dissipative Systems


V (f (x)) = V (x)





V
z (z)|z=f (x) g(x) = hT (x)
(4.163)

2

g T (x) zV2 (z)|z=f (x) g(x) = j T (x) + j(x)





V (f (x)) + g(x)u) is quadratic in u

Proof: Necessity: If the system is lossless there exists a nonnegative storage

function V (x) such that

V (f (x(k) + g(x(k))u(k)) V (x(k)) = hT (x(k))u(k)+

+ 12 uT (k)[j(x(k)) + j T (x(k))]u(k)
(4.164)
for all u(k) IRm and all k IN . Setting u(k) = 0 one gets the rst equality
in (4.108). Now one may calculate that (from now on we drop the k argument
in the functions)

V (f (x) + g(x)u) V
= |z=f (x)+g(x)u = hT (x) + uT [j T (x) + j(x)] (4.165)
u z
and
2 V (f (x)+g(x)u) 2

u2 = g T (x) zV2 |z=f (x)+g(x)u g(x)

(4.166)
= j(x) + j T (x)
Equations (4.165) and (4.166) imply the second and third equations in
(4.108). The last condition in (4.108) follows easily from (4.164).

Suciency: Suppose that the last condition in (4.108) is satised. One deduces
that

V (f (x)) + g(x)u) = A(x) + B(x)u + uT C(x)u (4.167)

for all u IRm and some functions A(x), B(x), C(x). From the Taylor expan-
sion of V (f (x)) + g(x)u) at u = 0 we obtain

A(x) = V (f (x))




B(x) = V (f (x)+g(x)u) |u=0 = V
z |z=f (x) g(x) (4.168)


u


2 2
C(x) = V (f (x)+g(x)u)
u2 |u=0 = 12 g T (x) zV2 |z=f (x) g(x)
 4.8 PR tangent system and dissipativity 249

From the rst three equations of (4.108) it follows that

V (f (x) + g(x)u) V (x) = y T u (4.169)

for all u IR , which concludes the proof.
m

Further results on nonlinear dissipative discrete-time systems may be

found in [185, 371, 373, 374].

4.8 PR tangent system and dissipativity

The topic of this section is the following: consider a nonlinear system with suf-
ciently regular vector eld, and its tangent linearization about some point
(x
, u
). Suppose that the tangent linearization is positive real, or strictly pos-
itive real. Then, is the nonlinear system locally dissipative? Or the converse?
Let us consider the following nonlinear system:


x(t) = f (x(t)) + g(x(t))u(t)



() y(t) = h(x(t)) (4.170)





x(0) = x0
where f (), g(), h() are continuously dierentiable functions of x, f (0) = 0,
g(x)u
h(0) = 0. Let us denote A = f x (0), B = u (x = 0, u = 0) = g(0),
C = hx (0). The tangent linearization of the system in (4.170) is the linear
time-invariant system


z(t) = Az(t) + Bu(t)



(t ) (t) = Cz(t) (4.171)





z(0) = x0
The problem is as follows: under which conditions are the following equiv-
alences true?
?
(t ) PR () is locally passive

?
(t ) SPR () is locally strictly dissipative
It also has to be said whether dissipativity is understood in Willems sense
(existence of a storage function), or in Hill and Moylans sense. Clearly one will
also be interested in knowing whether or not the quadratic storage functions
for (t ) are local storage functions for (). Important tools to study the
above two equivalences, will be the local stability, the local controllability, and
250 4 Dissipative Systems

the local observability properties of () when (A, B) is controllable, (A, C)

is observable, and A has only eigenvalues with nonpositive real parts. For
instance local wuniform reachability of () (Denition 4.44) is implied by
the controllability of (t ) (Proposition 4.76). One can thus already state that
if A has eigenvalues with negative real parts, and if (A, B) is controllable and
(A, C) is observable, then () has properties that make it a good candidate
for local dissipativity with positive denite storage functions and a Lyapunov
asymptoticaly stable xed point of x(t) = f (x(t)) (see Lemmas 5.18 and 5.20
in the next chapter).

Example 4.109. Let us consider the scalar system

x(t) = 12 x2 (t) + (x(t) + 1)u(t)



() y(t) = x(t) (4.172)





x(0) = x0

Then its tangent linearization around x = 0 is

z(t) = u(t)



(t ) (t) = z(t) (4.173)





z(0) = x0

The tangent system (t ) is an integrator H(s) = 1s . It is PR, though the

uncontrolled () is unstable (it may even have nite escape times).

Example 4.110. Let us consider the scalar system

x(t) = x2 (t) x(t) + (x3 (t) + x(t) + 1)u(t)



() y(t) = x2 (t) + x(t) (4.174)





x(0) = x0

Then the tangent linearization around z = 0 is

z(t) = z(t) + u(t)



(t ) (t) = z(t) (4.175)





z(0) = x0
1
The tangent system has transfer function H(s) = s+1 SPR. The un-
2
x
controlled () is locally stable (take V (x) = 2 ). However () in (4.174)
 4.8 PR tangent system and dissipativity 251

is not dissipative with this storage function and the supply rate uy since
y = g T (x) V
x (x). Consider now


x(t) = x2 (t) x(t) + u(t)



() y(t) = x(t) (4.176)





x(0) = x0
whose tangent linearization is in (4.175). This system is locally stable with
2
Lyapunov function V (x) = x2 , and y = g T (x) V
x (x). Easy computation yields
t
that 0 u(s)y(s)ds V (x(t))V (x(0)) for x (1, 1). Hence V (x) is a storage
function for (4.176), which is locally dissipative in (1, 1)  x.

Let us present a result which states under which conditions the tangent
linearization of a dissipative system, is a SPR system. Consider the system


x(t) = f (x(t)) + g(x(t))u(t)



() y(t) = h(x(t)) + j(x(t))u(t) (4.177)





x(0) = x0
with the dimensions for signals used throughout this book, f (0) = 0 and
h(0) = 0. The notion of dissipativity that is used is that of exponential dissi-
pativity,i.e. dissipativity with respect to exp( t)w(u(t), y(t)) for some > 0.
Assumption 7 There exists a function : IRm IRm , (0) = 0, such that
w((y), y) < 0, y = 0.
Assumption 8 The available storage function Va () is of class C 3 .
Assumption 9 The system is completely reachable if for all x0 IRn there
exists a nite t0 0, and an admissible input dened on [t0 , 0] which can
drive the state x() from the origin x(t0 ) = 0 to x(0) = x0 .

Theorem 4.111. [99] Let Q = QT IRmm , S = S T IRmm , R = RT

IR=m , and assume that Assumptions 7, 8 and 9 hold, and that the system
in (4.177) is exponentially dissipative with respect to the general supply rate
w(u, y) = y T Qy + 2y T Su + uT Ru. Then there exists matrices P IRnn ,
L IRpn , W IRpm , P = P T 0, and a scalar > 0 such that
T

A P + P A + P C T QC + LT L = 0



P B C T (QD + S) + LT W = 0 (4.178)





R + S T D + DT S + DT QD W T W = 0
252 4 Dissipative Systems

with A = fx (0), B = g(0), C =

h
x (0), D = j(0). If in addition the pair
(A, C) is observable then P > 0.

A similar result was proved in [187]. Theorem 4.111 proves that under some
conditions a dissipative system possesses a positive real tangent linearization.
What about the converse,i.e. if the tangent linearization is positive real, is
the system (locally) dissipative? The following brings an answer.

Theorem 4.112. [442] Consider the system in (4.177) and suppose that
j(0) = 0. Suppose that the tangent linearization is dissipative with respect to
the supply rate w(u, y) = y T Qy + 2y T Su + uT Ru, with R > 0, and w(0, y) 0
for all y. Suppose that the matrix

A BR1 SC BR1 B T
(4.179)
C T QC (A BR1 SC)T

has no purely imaginary eigenvalues, and that A is asymptotically stable. Then

there exists a neighborhood N IRn of x = 0 and V : N IR with v(0) = 0,
x (0) = 0 such that x [f (x) + g(x)u] w(u, h(x) + j(x)u) for all x N and
V V

all u U IR , V (x) 0 for all x N. Consequently the system in (4.177)

m

is locally dissipative in N with respect to w(u, y).

One remarks that the matrix (4.179) corresponds to the transition ma-
trix of the Hamiltonian system of the rst order necessary condition of the
Pontryagin principle for the Bolza problem, with a cost function equal to
uT Ru + xT C T QCx, under the constraint x(t) = (A BR1 SC)x(t) + Bu(t).
The two above examples do not t within the framework of Theorem 4.112 as
the dissipativity of the tangent linearizations holds with respect to the supply
rate w(u, y) = uT y, and thus R = 0.

4.9 Innite-dimensional Systems

4.9.1 An Extension of the KYP Lemma

The rst extensions of the KYP Lemma to the innite-dimensional case have
been achieved by Yakubovich et al. [300, 301, 520, 521]. Let us briey report
an extension of the KYP Lemma. We consider a system

x(t) = Ax(t) + Bu(t)
(4.180)

y(t) = Cx(t) + Du(t), x(0) = x0 X
where X is a real Hilbert space. The operator A : dom(A) X X is the
innitesimal generator of a C0 -semigroup U (t). The operators B : IRm X,
 4.9 Innite-dimensional Systems 253

C : X IRm , D : IRm IRm , are assumed to be bounded 3 . The solution of

(4.180) is

t
x(t) = U (t)x0 + U (t s)Bu(s)ds (4.181)
0

Denition 4.113. The operator H : L2,e L2,e is said (, )-passive if

t
t
es (Hu)T (s)u(s)ds es ||u(s)||2 ds (4.182)
0 0
for all u L2,e .

We have the following:

Lemma 4.114. [509] Let H : L2,e L2,e be dened by y = H(u) and

(4.180). Suppose that the C0 semigroup associated with H satises ||U (t)||
M et for some M 1 and > 0. Then for < 2, < min (D), H
is (, )passive if and only if for each 0 < , there exist bounded linear
operators 0 < P = P T : X X, L # 0 : X X, Q : X IRm , and a
matrix W IRmm , such that

(A P + P A + 2P + L + Q Q)x = 0 for all x dom(A)



BP = C W Q (4.183)





W W = D + D 20 Im

dom(A) is the domain of the operator A. A semigroup that satises the

condition of the lemma is said exponentially stable. The notation L() # 0
means that L() is a positive operator that is bounded invertible (or coercive).

4.9.2 The Wave Equation

This section presents an example of an innite-dimensional system which is

dissipative: the wave equation. Let IRn be an open set with boundary
. Let us denote Q = (0, +) and = (0, +). The problem is to
nd a function u(x, t) : [0, +) IR such that

3
An operator may here be much more general than a linear operator represented
by a constant matrix A IRmn : x  Ax IRm . For instance the Laplacian
P 2 2
= n i=1 x2 , or the DAlembertian t2 are operators.
i
254 4 Dissipative Systems
2

t2 u = 0
u
on Q






u = 0 on
(4.184)



u(x, 0) = u0 (x) on




u
t (x, 0) = v0 (x) on
 2
where = ni=1 x 2 is the Laplacian with respect to state variables, u0 ()

and v0 () are data. The system in (4.184) is called the wave equation: this
is an hyperbolic equation. When n = 1 and = (0, 1), (4.184) models the
small vibrations of a free rope. For each t 0, the graph of the function
x
u(x, t) coincides with the rope conguration at time t. When n = 2,
it models the small vibrations of an elastic shell. From a general point of view,
(4.184) models wave propagation in an elastic homogeneous medium IRn .
The second condition in (4.184) is the Dirichlet boundary condition. It means
that the rope is xed on the boundary . The third and fourth conditions in
(4.184) are the Cauchy initial data for the system (initial position and initial
velocity). It is assumed that the boundary data and satisfy some regularity
conditions, so that the solution of (4.184) exists and is unique as a C 2 (IR+ )
and L2 () function (we do not present here the rigorous denition of the
functional spaces which are needed to correctly dene the solution, because
this would bring us much too far in such a brief presentation). The interesting
part for us is:

Lemma 4.115. Along the solutions of (4.184) one has

+ +
+ u +2
+ (t)+ + ||u(t)||22, = ||v0 ||22, + ||u0 ||22, (4.185)
+ x +
2,

for all t 0.
+ +2  + u +  + + u
+2
+
One has is + u
x (t)+
2,
= + (t)+ dx and ||u(t)||2 =
x 2, +
xi (x, t)+ dx.
The equality in (4.185) means that the system is lossless (energy is conserved).
Notice that the wave equation may be rewritten as a rst order system
u
t v = 0 on Q
(4.186)
v
t u = 0 on Q
 
u 0n In
If X = then (4.186) becomes dX +AX = 0 with A = X.
v dt 0n
It happens that the operator A + I2n is maximal monotone. We retrieve here
this notion that we used also in the case of nite-dimensional nonsmooth
systems in Section 3.9.4.
 4.10 Further Results 255

4.9.3 The Heat Equation

The notation is kept form the foregoing subsection. The heat equation is given
as
u


t u = 0 on Q


u=0 on (4.187)





u(x, 0) = u0 (x) on
The variable u may be the temperature in the domain . Under the as-
sumption that u0 L2 (), there exists a unique solution u(x, t) for (4.187)
in C 1 (IR+ ) which is itself L2 (). Moreover:

Lemma 4.116. Along the solutions of (4.187) one has

t
1
||u(x, t)||2 dx + u(t)22, dt = ||u0 ||22, (4.188)
0 2
+
n + u + 2
2 +
for all t 0, where u(t)2, = i=1 + x i
(x, t)+ dx.

The operator A : u
u is maximal monotone. The equality in (4.188)
means that the temperature decreases on Q at a xed position x.

Let us mention more work on innite dimensional systems that may be

found in [30,39,40,5759,117120,172,196,224,307,391,507,509]. The case of
a parabolic equation describing the temperature control problem for a homo-
geneous rod of unit length is provided in [57, 4].

4.10 Further Results

Nonnegative systems: the theory of dissipative systems and the KYP Lemma
have also been applied to nonnegative systems [191,192]. Nonnegative dynam-
ical systems are derived from mass and energy balance considerations that in-
volve states whose values are nonnegative. For instance in ecological models,
the quantity of shes in a lake cannot be negative (if the mathematical model
allows for such negative values then surely it is not a good model). A matrix
A IRnm is nonnegative if Aij 0 for all 1 i n and all 1 j m.
It is positive if the strict inequality > 0 holds. A matrix A IRnn is called
essentially nonnegative (positive) if A is a Z-matrix, i.e. if Aij 0 (> 0) for
all 1 i n and all 1 j n with i = j. A matrix A IRnn is essentially
nonnegative if and only if exp(At) is nonnegative for all t 0. A sucient
condition for the solutions of the system x(t) = Ax(t), x(0) = x0 0, t 0,
to satisfy x(t) 0 for all t 0, is that A be essentially nonnegative.
256 4 Dissipative Systems

Let us now consider a system whose realization is the quadruple (A, B, C, D),
with A IRnn being essentially nonnegative, B IRnm , C IRmn and
D IRlm being nonnegative matrices. Suppose also that the inputs are re-
stricted to nonnegative values, i.e. u(t) 0 for all t 0. Then the system is
nonnegative in the sense that x(t) 0 and y(t) 0 for all t 0 [191, Lemma
2.2].

Theorem 4.117 (KYP Lemma for nonnegative systems). [191] Let q

IRl and r IRm . Consider the nonnegative dynamical system with realization
(A, B, C, D) where A is essentially nonnegative, B, C and D are nonnegative.
Then the system is exponentially dissipative with respect to the supply rate
w(u, y) = q T y + rT u if and only if there exist nonnegative vectors p IRn ,
l IRn , and w IRm , and a scalar 0 such that
T
A p + p C T q + l = 0
(4.189)
T
B p DT q r + w = 0

Clearly when = 0 the system is simply dissipative and no longer expo-

nentially dissipative. This result extends to positive nonlinear systems.

The word dissipative is sometimes used in a dierent context in the theory

of dynamical systems, see e.g. [98].
5
Stability of Dissipative Systems

In this chapter, various results concerning the stability of dissipative systems

are presented. First, the input/output properties of several feedback intercon-
nections of passive systems are reviewed. Then the conditions under which
storage functions are Lyapunov functions are given in detail. The chapter ends
with an introduction to H theory for nonlinear systems that is related to a
specic dissipativity property, and with a section on Popovs hyperstability.

5.1 Passivity Theorems

In this section we will study the stability of the interconnection in negative

feedback of dierent types of passive systems. We will rst study closed-loop
interconnections with one external input (one-channel results) and then in-
terconnections with two external inputs (two-channel results). The implicit
assumption in the passivity theorems is that the problem is well-posed, i.e.
that all the signals belong to L2e .

Remark 5.1. Dierent versions of passivity theorems can be obtained depend-

ing on the properties of the subsystems in the interconnections. We will only
consider here the most classical versions.

5.1.1 One-channel Results

Theorem 5.2 (Passivity (one-channel) [207]). Assume that both H1 , H2

are pseudo-VSP, i.e.

t
t
t
y1T (s)u1 (s)ds + 1 1 y1T (s)y1 (s)ds + 1 uT1 (s)u1 (s)ds
0 0 0
258 5 Stability of Dissipative Systems

t
t
t
y2T (s)u2 (s)ds + 2 2 y2T (s)y2 (s)ds + 2 uT2 (s)u2 (s)ds
0 0 0

with

1 + 1 > 0, 2 + 2 > 0
The feedback closed-loop system (see Figure 5.1) is nite gain stable if
2 0, 1 0, 2 + 1 > 0,
where 2 or 1 may be negative.

Fig. 5.1. Closed-loop system with one external input

Corollary 5.3. The feedback system in Figure 5.1 is L2 nite-gain stable if

1. H1 is passive and H2 is ISP i.e. 1 0, 2 > 0, 1 0, 2 0
2. H1 is OSP and H2 is passive i.e. 1 0, 2 0, 1 > 0, 2 0

t
Proof: Let r|yt = 0 r(s)y(s)ds. Then

r|yt = u1 + y2 |yt

= u1 |y1 t + y2 |u2 t
(5.1)
1 + 1 u1 2 + 1 y1 2t + 2 + 2 u2 2 + 2 y2 2t

1 + 2 + (1 + 2 )y2t ,
 5.1 Passivity Theorems 259

where y2t = y|yt . Using the Schwartz inequality we have

t 
t 12 
t 12
2 2
r|yt = r(s)y(s)ds r (s)ds y (s)ds = rt yt
0 0 0

Then
rt yt r|yt 1 + 2 + (1 + 2 )y2t
For any IR the following holds
 2
1 2
2 rt + 2 y2t = 12 1 rt yt + rt yt (5.2)
1 + 2 + (1 + 2 )y2t
Choosing = 1 + 2 we get

r2t (1 + 2 )
1 + 2 + y2t
2(1 + 2 ) 2
which concludes the proof.

5.1.2 Two-channel Results

Consider now the system depicted in Figure 5.2 where r1 , r2 can represent dis-
turbances, initial condition responses or commands. Assume well-posedness.
Theorem 5.4. (Passivity (two-channel) [500]) Assume H1 , H2 are pseudo
VSP. The feedback system is L2 nite-gain stable if

1 + 2 > 0

2 + 1 > 0
where i , i may be negative.
Corollary 5.5. The feedback system is L2 nite-gain stable if
1. H1 , H2 are ISP ( 1 > 0, 2 > 0, 1 = 2 = 0)
2. H1 , H2 are OSP (1 > 0, 2 > 0, 1 = 2 = 0)
3. H1 is VSP, H2 is passive ( 1 > 0, 1 > 0, 2 = 2 = 0)
4. H1 is passive, H2 is VSP ( 2 > 0, 2 > 0, 1 = 1 = 0)
Proof:

u1 |y1 t + y2 |u2 t = r1 y2 |y1 t + y2 |y1 + r2 t

= r1 |y1 t + y2 |r2 t
(5.3)
1 + 1 u1 2 + 1 y1 2t + 2

+ 2 u2 2t + 2 y2 2t

260 5 Stability of Dissipative Systems

Fig. 5.2. Closed-loop system with two external inputs

Note that t
u1 2t = 0 uT1 (s)u1 (s)ds
t
= 0 (r1 (s) y2 (s))T (r1 (s) y2 (s))ds (5.4)

2r1 |y2 t + y2 2t

and similarly
u2 2t 2r2 |y1 t + y1 2t
Then

r1 |y1 t + y2 |r2 t +2 1 r1 |y2 t 2 2 r2 |y1 t

(5.5)
1 + 2 + ( 1 + 2 )y2 2t + ( 2 + 1 )y1 2t

Note that for any IR, for i = 1, 2 we have

ri |yi t yi t ri t + 12 ( 1 ri t i yi t )2
i
(5.6)
1 2 2i 2
2i ri t + 2 yi t

2 +1 1 +2
We choose 1 = 2 and 2 = 2 :

If 1 = 0 then 2 1 r1 |y2 t 0
 5.1 Passivity Theorems 261

If 1 > 0 then for any 1 IR

1
2 1 r1 |y2 t r1 2t + 1 1 y2 2t
1
 1 +2
Let us choose 1 = 1
1 and 1 = 4 . Therefore
(1 +2 ) (2 +1 )
1 + 2 + 4 y2 2t + 4 y1 2t
(5.7)
1 421 1 421
r1 2t ( 2 +1
+ 1 +2 ) + r2 2t ( 1 +2
+ 1 +2 )

which concludes the proof.

Boundedness of the closed-loop signals can be ensured if H1 and H2 have
nite gain as can be seen from the following Lemma, which is no longer a
purely input/output result but involves the state of the system.

Lemma 5.6. Consider again the negative feedback interconnection of H1 and

H2 as in Figure 5.2. Assume that the operators H1 and H2 are pseudo VSP
i.e.

t
t
t
uTi (s)yi (s)ds = Vi (xi )Vi (xi (0))+ i uTi (s)ui (s)ds+i yiT (s)yi (s)ds
0 0 0

with V1 () and V2 () positive denite functions. Then the origin is an asymp-

totically stable equilibrium point if:

1 + 2 > 0

and
2 + 1 > 0
and both H1 and H2 are zero-state observable (i.e. ui 0, yi 0 xi = 0).

Proof: Consider the positive denite function which is the sum of the two
storage functions for H1 and H2 , i.e.:

V (x) = V1 (x1 ) + V2 (x2 )

Then using the dissipativity inequalities in their innitesimal form we get

along the trajectories of the system
2 
 
V (x(t)) = uTi (t)yi (t) i uTi (t)ui (t) i yiT (t)yi (t)
i=1 (5.8)
= ( 1 + 2 )uT1 (t)u1 (t) ( 2 + 1 )y1T (t)y1 (t)
The result follows from the Krasovskii-LaSalle Theorem and the assump-
tion guaranteeing that yi 0, ui 0 xi = 0. If in addition V1 () and V2 ()
are radially unbounded, then one gets global stability.
262 5 Stability of Dissipative Systems

Roughly speaking, the foregoing lemma says that the feedback interconnec-
tion of two dissipative systems is asymptotically stable provided an observabil-
ity property holds. Let us now state a result which uses the quasi-dissipativity
property as dened in Denition 4.27. Each subsystem H1 and H2 of the in-
terconnection is supposed to be dissipative with respect to a general supply
rate of the form wi (ui , yi ) = yiT Qi yi + 2yiT Si ui + uTi Ri ui , with QTi = Qi
and RiT = Ri . Before stating the next Proposition, we need a preliminary
denition:

Denition 5.7. A system x(t) = f (x(t), u(t)), y(t) = h(x(t)) has uniform
nite power gain 0 if it is quasi-dissipative with supply rate w(u, y) =
2 uT u y T y.

The following holds:

Proposition 5.8. [403] Suppose that the systems H1 and H2 are quasi-
dissipative with respect to supply rates w1 (u1 , y1 ) and w2 (u2 , y2 ), respectively.
Suppose there exists > 0 such that the matrix

Q1 + R2 S1 + S2T
Q = (5.9)
S1T + S2 R1 + Q2
is negative denite. Then the feedback system in Figure 5.2 has uniform nite
power gain.

Proof: taking into account the interconnections u1 = r1 y2 and u2 = r2 +y1 ,

it follows that

w1 (u1 , y1 ) + w2 (u2 , y2 ) =

  y1   y1
= y1T y2 Q
T + y1T y2T S +
y2 y2 (5.10)

  r1
+ r1T r2T R
r2
for some matrices S and R . Since Q < 0 it follows that there exists > 0
and > 0 such that

(y1T y1 + y2T y2 ) + (r1T r1 + r2T r2 ) w1 (u1 , y1 ) + w2 (u2 , y2 ) (5.11)

Integrating from t = 0 to t = 0 and using the fact that H1 and H2

are quasi-dissipative with constants 1 0 and 2 0, we obtain
 5.1 Passivity Theorems 263

0 [(y1T (t)y1 (t) + y2T (t)y2 (t)) + (r1T (t)r1 (t) + r2T (t)r2 (t))]dt+
(5.12)
+(1 + 2 ) + 1 + 2 0

where 1 0 and 2 0 are the bias for H1 and H2 .

This proof is really an input/output system stability result as it does not
mention the state. Let us mention a result in [466] that contains a version of
the passivity Theorem, using the so-called secant condition for the stability of
polynomials of the form p(s) = (s + a1 )(s + a2 )...(s + an ) + b1 b2 ...bn , with all
ai >
0 and all bi > 0. This p(s)
is the characteristic polynomial of the matrix
a1 0 .... 0 b1
b2 a2 .... 0 0

A= ... ... ...
. The secant condition states that A is Hurwitz

0 0 .... bn an
 n
provided that ab11 ...b
...an < sec n
n
= (cos(1 ))n .
n

5.1.3 Lossless and WSPR Blocks Interconnection

It is known that the feedback interconnection of a PR and a SPR blocks

yields an asymptotically stable system; see Lemma 3.37. In the case of non-
linear systems, and using a pure input/output denition of passivity (as in
Denition 2.1 where is not assumed to depend on the initial state value)
the passivity Theorem provides L2 -stability results for the interconnection of
a passive block with an ISP, OSP or a VSP block (see e.g. [500]). Lyapunov
stability can be obtained when the blocks are passive in the sense of Willems
(i.e. the state intervenes in the denition). The goal of the following lemma is
to present stability results with slightly relaxed requirements on the feedback
block. More precisely, we will deal with the interconnection of lossless blocks
with WSPR blocks. The results presented in this section relax the conditions
of the passivity Theorem as was conjectured in [310].
We now consider the negative feedback interconnection of a lossless (pos-
sibly nonlinear) system with a linear WSPR system and prove the stability of
the closed-loop system.

Lemma 5.9. Assume that H1 in Figure 5.2 is zero-state observable and

lossless with a radially unbounded positive denite storage function V1 (x1 ),
whereas H2 is WSPR. Then the feedback interconnection of H1 and H2 is
Lyapunov globally asymptotically stable.

Proof: Consider V (x1 , x2 ) = xT2 P2 x2 + 2V1 (x1 ), where V1 () is a radially un-

bounded positive denite storage function for H1 . In view of the assumptions
and of the KYP Lemma, there exists matrices P2 , L2 , W2 such that Equations
(3.2) are satised for H2 . Then
264 5 Stability of Dissipative Systems

V (x1 , x2 ) = xT2 LT2 L2 x2 + 2xT2 P2 B2 u2 + 2uT1 y1

= x2 LT2 L2 x2 + 2uT2 (C2 W2T LT2 )x2 + 2uT1 y1

= xT2 LT2 L2 x2 2uT2 (W2T LT2 x2 + D2 u2 )

(5.13)
= xT2 LT2 L2 x2 2uT2 W2T LT2 x2 uT2 (D2 + D2T )u2

= (uT2 W2T + xT2 LT2 )(W2 u2 + L2 x2 )

= y2T y2
 
The above ensures that xT = xT1 xT2 = 0 is a stable equilibrium point,
which implies that the state x is bounded. Moreover the transfer function

H2 (s) = W2 + L2 (sI A2 )1 B2
has no zeros on the imaginary axis (see Lemma 3.18). Note that Y2 (s) =
H2 (s)U2 (s). Therefore, when y2 (t) 0, u2 (t) can only either exponentially
diverge or exponentially converge to zero. However, if u2 (t) diverges, it fol-
lows from y2 (t) = W2 u2 + L2 x2 0 that x2 should also diverge which is a
contradiction. It then follows that u2 should converge to zero. Note that for
u2 = 0 the H2 system reduces to x2 = A2 x2 with A2 Hurwitz. Therefore
if y2 (t) 0, then x2 0. On the other hand u2 = y1 and so we also have
y1 0. In view of the zero-state observability of H1, we conclude that x1 0.
Hence, from the Krasovskii-La Salle invariance set Theorem, the largest in-
variant set S inside the set y2 0 is reduced to x = 0 plus all the trajectories
such that x tends to the origin. Therefore, the origin x = 0 is asymptoti-
cally stable. Moreover, when V1 (x1 ) is radially unbounded any trajectory is
bounded, and the equilibrium is globally asymptotically stable.
Another proof can be found in [213]. It makes use of the material in Ap-
pendix A.4.1 which possesses its own interest.

5.1.4 Large-scale Systems

Large-scale systems consist of an interconnection of N subsystems Hi , which

are all dissipative. It is assumed here that the subsystems are dissipative
in the sense of Denition 4.22 and with respect to a general supply rate
wi (ui , yi ) = yiT Qi yi + 2yiT Si ui + uTi Ri ui . The interconnection relationship is


N
ui = ue,i Hij yj (5.14)
j=1

where ui is the input of subsystem Hi , yi is its ouput, ue,i is an external

input, and all the Hij are constant matrices. Grouping the inputs, outputs
 5.1 Passivity Theorems 265

and external inputs as N -vectors u, y and ue respectively, one may rewrite

(5.14) as

u = ue Hy (5.15)
N N
where H IR . Let us dene Q =diag(Qi ), S =diag(Si ) and R =diag(Ri ),
and the matrix

Q = SH + H T S T H T RH Q (5.16)

Theorem 5.10. [364] The overall system with input ue () and output y()
and the interconnection in (5.15) is L2 nite-gain stable if Q > 0 in (5.16).

Proof: For each subsystem Hi we have by assumption

t1
wi (ui (t), yi (t))dt 0 (5.17)
t0

for all t1 t0 . By summation over all i one obtains

t1
w(u(t), y(t))dt 0 (5.18)
t0

Using (5.15) and (5.16) one obtains

t1
t1
1
[y T (t)Qy(t) 2y T (t)Q 2 Sue (t)]dt uTe (t)Rue (t)dt (5.19)
t0 t0

1
with S = Q 2 (S H T R). Let > 0 be a nite real such that R + S T S
2 IN . Clearly one can always nd such a scalar. Then one nds after some
manipulation

t1
t1
1 1
[Q 2 y(t) Sue (t)]T [Q 2 y(t) Sue (t)]dt 2 uTe (t)ue (t)dt, (5.20)
t0 t0

so that

t1
t1
y T (t)y(t)dt k 2 uTe (t)ue (t)dt (5.21)
t0 t0
1
with k = ||Q 2 ||( + ||S||).
Let us recall that we assumed at the beginning of this section that all
signals belong to the extended space L2,e (more rigorously: the inputs are
in L2,e and we assume that the systems are well-posed in the sense that the
outputs also belong to L2,e ). Under such an assumption, one sees that stating
(5.21) for all t1 t0 0 is equivalent to stating ||y||2,t k||ue ||2,t for all
266 5 Stability of Dissipative Systems

t 0, where || ||2,e is the extended L2 norm. One notes that Theorem 5.10
is constructive in the sense that the interconnections Hi,j may be chosen
or designed so that the Riccati inequality Q > 0 in (5.16) is satised. The
literature on large-scale systems stability is abundant, and an early reference
to be read for more informations and results is [499]. Let us end this subsection
with a result which will allow us to make a link between the interconnection
strucuture, and so-called Mmatrices.
Theorem 5.11. [364] Let the subsystem Hi have a L2 nite-gain i and
suppose that all subsystems are single input single output (SISO). Let =diag(i ),
and A = H. Then if there exists a diagonal positive denite matrix P such
that

P AT P A > 0 (5.22)
the interconnected system is L2 nite-gain stable.
A sucient condition for the existence of a matrix P as in the theorem is
that the matrix B made of the entries bii = 1 |aii |, bij = |aij | for i = j, has
all its leading principal minors positive. Such a matrix is called an Mmatrix.

Further works on large-scale systems may be found in [184, 185].

5.2 Positive Deniteness of Storage Functions

In this section we will study the relationship between dissipativeness and
stability of dynamical systems. Let us rst recall that in the case of linear
systems, the plant is required to be asymptotically stable to be WSPR, SPR or
SSPR. For a PR system it is required that its poles be in the left-half plane and
the poles in the jaxis be simple and have non-negative associated residues.
Consider a dissipative system as in Denition 4.20. It can be seen that if
u = 0 or y = 0, then V (x(t)) V (x(0)). If in addition the storage function
is positive denite, then we can conclude that the system x(t) = f (x(t))
has a Lyapunov stable xed point x = 0, and the systems zero dynamics is
stable. Furthermore, if the system is strictly passive (i.e. S(x) > 0 in (4.51))
then the system x(t) = f (x(t)), and the systems zero dynamics are both
asymptotically stable (see Theorem 4.10).
Let us now consider passive systems as given by Denition 2.1. The two
following Lemmae will be used to establish the conditions under which a
passive system is asymptotically stable.
Denition 5.12 (locally ZSD). A nonlinear system (4.79) is locally zero-
state detectable (ZSD)[locally Zero state observable (ZSO)] if there exists a
neighborhood N of 0 such that for all x(t) N

u(t) = 0, h(x(t)) = 0, t 0 lim x(t) 0 [x(t) = 0 for all t 0]

t+
 5.2 Positive Deniteness of Storage Functions 267

If N = IRn the system is ZSD [ZSO].

Lemma 5.13. [207] Consider a dissipative system with a general sup-

ply rate w(u, y), and let Assumptions 3 and 4 of Section 4.5.2 hold.
Assume that:
1. The system is zero-state observable
2. For any y = 0 there exists some u such that w(u, y) < 0
Then all the solutions to the NL-KYP set of equations (4.81) are pos-
itive denite.

Proof: We have already seen that the available storage

t 8
Va (x) = sup w(s)ds
x=x(0),t0,u 0

is a (minimum) solution of the KYP-NL set of equations (4.81), see the

necessity part of the proof of Lemma 4.87 and Theorem 4.41. Recall that
0 Va (x) V (x). If we choose u such that w(u, y) 0 on [t0 , ), with strict
inequality on a subset of positive measure, then Va (x) > 0, y = 0. Note from
the equation above that the available storage Va (x) does not depend on u(t)
for t [t0 , ). When y = 0 we can choose u = 0 and therefore x = 0 in view
of the zero-state observability assumption. We conclude that Va (x) is positive
denite and that V (x) is also positive denite (see Denition A.9).

Lemma 5.14. Under the same conditions of the previous lemma, the free
system x = f (x) is (Lyapunov) stable if Q 0 and asymptotically stable if
Q < 0, where Q is the weighting matrix in the general supply rate (4.80).

Proof: From Corollary 4.88 and Lemma 4.87 there exists V (x) > 0 for all
x = 0, V (0) = 0, such that (using (4.81) and (4.82))
d(V x) T
dt (t) = [L(x(t)) + W (x(t))u(t)] [L(x(t)) + W (x(t))u(t)] +

+y T (t)Qy(t) + 2y T (t)Su(t) + uT (t)Ru(t)

= LT (x(t))L(x(t)) 2LT (x(t))W (x(t))u(t) uT (t)W T (x(t))

W (x(t))u(t) + (h(x(t)) + j(x(t))u)T Q(h(x(t)) + j(x(t))u(t))+

+2(h(x(t)) + j(x(t))u(t))T Su(t) + uT (t) Ru(t)

(5.23)
268 5 Stability of Dissipative Systems

so that

d(V x)
dt (t) = LT (x(t))L(x(t)) uT (t)W T (x(t))W (x(t))u(t)+

+uT (t)[R + j T (x(t))Qj(x(t)) + j T (x(t))S + S T j(x(t))]u(t)+

+2[LT (x(t))W (x(t)) + hT (x(t))(Qj(x(t)) + S)]u(t)+

+hT (x(t))Qh(x(t))

= LT (x(t))L(x(t)) uT (t)R(x(t))u(t) + uT (t)R(x(t))u(t)+

+2[LT W (x(t)) + hT (x(t))S(x(t))]u(t) + hT (x(t))Qh(x(t))

= LT (x(t))L(x(t)) + V T (x(t))g(x(t))u(t) + hT (x(t))Qh(x(t))

(5.24)
For the free system x(t) = f (x(t)) we have

d(V x)
(t) = LT (x(t))L(x(t)) + hT (x(t))Qh(x(t)) hT (x(t))Qh(x(t)) 0
dt

If Q < 0 then d(Vdtx) (t) 0 which implies stability of the system. If Q 0

we use Krasovskii-LaSalle Invariance Principle. The invariant set is given by
: {|h() = y = 0} and therefore x() converges to the set . In view of the
zero-state observability we conclude that x(t) 0 asymptotically. One sees
that under the conditions of Lemma 5.13 and with Q < 0, then necessarily
x = 0 is an isolated xed point of x(t) = f (x(t)).

Example 5.15. Let us come back to Example 4.59. The system in (4.51) is not
zero state detectable, since u 0 and y 0 do not imply x 0 as t +.
And the uncontrolled (or free) system is exponentially unstable (x(t) = x(t)).
This shows the necessity of the ZSD condition.

Corollary 5.16. [207] Consider a dissipative system with a general supply

rate w(u, y). Assume that:
1. The system is zero-state observable(i.e. u(t) 0 and y(t) 0 x(t) = 0)
2. For any y = 0 there exists some u such that w(u, y) < 0
Then passive systems (i.e. Q = R = 0, S = I) and input strictly passive
systems (ISP) (i.e. Q = 0, 2S = I, R = ) are stable, while output passive
systems (OSP) (i.e. Q = , 2S = I, R = 0) and very strictly passive systems
(VSP) (i.e. Q = , 2S = I, R = ) are asymptotically stable.

Before stating the next lemmas let us introduce another notion of zero
state detectability.
 5.2 Positive Deniteness of Storage Functions 269

Denition 5.17. A dynamical system is said to be locally zero state detectable

in a region z if for any x0 z , x0 = 0, such that the solution x(t) ,
for all 0 t for some > 0, with u() = 0, there exists a continuous
function : IR IR+ , (0) = 0, (w) > 0 for all w = 0, such that

t
y T (t )y(t )dt (||x0 ||) (5.25)
0

for some t < + such that 0 t . If in addition for any sequence

{wn } , one has (wn ) + as ||wn || +, the system is said to be
locally uniformly zero state detectable in z with respect to .

Clearly a system that is ZSD according to this denition is also ZSD in

the sense of Denition 5.12. Sometimes a system that satises the rst part of
Denition 5.17 is called uniformly observable. The local versions of Lemmas
5.13 and 5.14 are as follows:

Lemma 5.18. [404] Let the dynamical system in (4.79) be

Locally dissipative with respect to a general supply rate (4.80) in a region
IRn ,
Locally wuniformly reachable in a region c with respect to
Locally uniformly zero state detectable in z with respect to
Suppose that z c = . Then the dynamical system has all its storage
functions V : z c IR continuous, V (0) = 0, and V (x) > 0 for all
x z c . Moreover for any sequence {xn } z c , V (xn ) + as
||xn || +.

We will also say that a system is said to be locally reachable with respect
to in a region r , if every state x1 r is locally reachable with
respect to from the origin x = 0 and for all t0 IR, with an input that
keeps the state trajectory inside .

Denition 5.19. A system is said locally connected with respect to in a

region con , if any x1 con is locally reachable with respect to from
any x0 con , and for all t0 IR.

Now we are ready to state the main result which concerns the local stability
deduced from local dissipativity.

Lemma 5.20. [404] Let the dynamical system in (4.79) be

Locally dissipative with respect to a general supply rate (4.80) in a region
IRn
Locally wuniformly reachable in a region c with respect to
Locally uniformly zero state detectable in z with respect to
Locally connected in a region con with respect to
270 5 Stability of Dissipative Systems

Locally Lipschitz continuous in

and be such that there exists a feedback controller u
(x) such that w(u
, y) < 0
for all y = 0, u
(0) = 0 and u
() drives the system from x0 to x1
while keeping the trajectory inside . Suppose that the region c z con
contains an open neighborhood of x = 0. Then if Q < 0 the origin x = 0 is
asymptotically stable.

The above conditions imply that all the dened regions contain x = 0. We
now state a result which is based on the notion of weak w(u, y)dissipativity
(Denition 4.29) and is interesting as it applies to systems with multiple
equilibria, and makes no assumption on the dierentiability of the storage
functions. This theorem is linked to Theorems 4.31, 4.32 and 4.33. d(x, ) =
inf y ||x y|| denotes the distance from x to .

Theorem 5.21. [206] Suppose that G() is w(u, y)dissipative for some
Q < 0. Let X1 = {x | d(x, ) d1 } for some d1 > 0, be uniformly reachable
from and zero state observable with respect to . Then there exists some
d2 > 0 (dependent on d1 ) such that, with input u() 0, all state trajectories
starting in X2 = {x | d(x, ) d2 } remain in X1 , and asymptotically
approach .

As an illustration one may consider Example 4.34. Let us now introduce

the following denition:

Denition 5.22 (Proper function). A function V : x IR is said to be

proper if for each a > 0, the set V 1 [0, a] = {x : 0 V (x) a} is compact
(closed 1 and bounded).

A variant of Lemma 5.14 is as follows:

Lemma 5.23. [442] Let V () 0 be a solution of (4.76), with S(x) =

hT (x)h(x), > 0, V (0) = 0 and V (x) > 0, x = 0, and suppose that the
system in (4.74) is zero-state detectable. Then x = 0 is a locally asymptoti-
cally stable equilibrium of x(t) = f (x(t)). If additionally V () is proper then
x = 0 is globally asymptotically stable.

5.3 WSPR Does not Imply OSP

In this subsection we prove that if a system is WSPR (Weakly Strictly Positive
Real), it does not necessarily imply that the system is OSP (Output Strictly
Passive). The proof is established by presenting a counterexample. The pas-
sivity Theorems concern interconnections of two blocks, where the feedback
block must be either ISP, OSP or VSP. The interest of the results in Section
1
A set is closed if it contains its limit points.
 5.3 WSPR Does not Imply OSP 271

5.1.3 is that the conditions on the feedback block are relaxed to WSPR. We
prove now that the following transfer function (which is WSPR; see Example
2.59)
s+a+b
H(s) = (5.26)
(s + a)(s + b)
is not OSP. This proves that in general WSPR  OSP. A minimal state

0 1
space representation (A, B, C) for H(s) is given by A = ,
ab a b

1
B = , C = (1, 0). Let us choose a = 1, b = 2, x(0) = 0, u = sin(t).
0
Then
t
y(t) = [2 exp( t) exp(2 2t)] sin( )d (5.27)
0
It can be shown that

y(t) = f1 () cos(t) + f2 () sin(t) (5.28)

3
7 6
with f1 () = (1+ 2 )(4 2 ) , and f2 () = (1+ 2 )(4 2 ) . It can also be proved

that

t
f1 () f2 () sin(2t)
u( )y( )d = [cos(2t) 1] + [t ] (5.29)
0 4 2 2

and that
t    
sin(2t) sin(2t)
0 y 2 ( )d = f12 () t
2 + 4 + f22 () t
2

  (5.30)
cos(2t)
f1 ()f2 () 2 1

Let us choose tn = 2n for some integer n > 0. When +, then

 tn
f2 ()2n
0 u( )y( )d = 4 , whereas

tn 
2 2n(f12 () + f22 ()) 1
y ( )d = + f1 ()f2 () 1
0 4 2
 tn
 tn 2
It follows that 0 u( )y( )d
5 while 0 y ( )d 3 for
some positive real and . Therefore we have found an input u(t) = sin(t)
t t
and a time t such that the inequality 0 u( )y( )d 0 y 2 ( )d cannot be
satised for any > 0, as +.
272 5 Stability of Dissipative Systems

5.4 Stabilization by Output Feedback

5.4.1 Autonomous Systems
Consider a causal nonlinear system () : u(t) y(t); u(t) Lpe , y(t) Lpe
represented by the following state-space representation ane in the input:

x(t) = f (x(t)) + g(x(t))u(t)
() (5.31)

y(t) = h(x(t)) + j(x(t))u(t)
where x(t) IRn , u(t), y(t) IRm , f (), g(), h(), and j() are smooth func-
tions of x and f (0) = h(0) = 0. We can now state the following result:
Theorem 5.24 (Global asymptotic stabilization [89]). Suppose (5.31) is
passive and locally ZSD. Let (y) be any smooth function such that (0) = 0
and y T (y) > 0, y = 0. Assume that the storage function V (x) > 0 is proper.
Then, the control law u = (y) asymptotically stabilizes the equilibrium point
x = 0. If in addition (5.31) is ZSD then x = 0 is globally asymptotically stable.

Proof: By assumption, V (x) > 0 for all x = 0. Replacing u = (y) in

(4.42) we obtain

t
V (x(t)) V (x(0)) y T (s)(y(s))ds 0
0
It follows that V (x(t)) V (x(0)) < , which implies that x(t) <
for all t 0, and thus y(t) < . Therefore V (x()) is non-increasing and
thus
 t T converges. In the limit, the left hand side of the inequality is 0 , i.e.
0 y (s)(y(s))ds 0 as t . Thus y(t) 0 as t + and u also
converges to 0. Since the system is locally ZSD, then x(t) 0 as t +. If
in addition the system is globally ZSD, then x = 0 is globally asymptotically
stable.
Lemma 5.25. Suppose the system (5.31) is passive and zero state observable,
with feedback control law u = (y), (0) = 0. Then the storage function of
the closed-loop system is positive denite, i.e. V (x) > 0, for all x = 0.
Proof: Recall that the available storage satises 0 Va (x) V (x) and
2  3
t
Va (x) = sup 0 y T (s)u(s)ds
x=x(0),t0,u 2 3
t T (5.32)
= sup 0
y (s)(y(s))ds
x=x(0),t0,u

If Va (x) = 0, then necessarily y(t) = 0. In view of zero state observability,

y = 0 x = 0. Thus Va (x) vanishes only at x = 0 and so does V (x).
The asymptotic stabilization by output feedback of nonlinear systems non-
linear in the input as in (4.95) continues to hold [303, Theorem 4.4]. Fur-
ther results on the asymptotic stabilization by output feedback may be found
in [430, Theorem 6].
 5.4 Stabilization by Output Feedback 273

5.4.2 Time-varying Nonlinear Systems

In this section we consider systems of the form

x(t) = f (t, x(t)) + g(t, x(t))u(t)
(5.33)

y(t) = h(t, x(t))
where f (, ), g(, ) and h(, ) are continuous functions IR+ IRn IRn ,
f (t, 0) = 0 and h(t, 0) = 0 for all t 0. It is further supposed that f (, ),
g(, ) and h(; ) are uniformly bounded functions. Since the system is not
autonomous, it is no longer possible to apply the arguments based on the
Krasovskii-LaSalle invariance principle. An extension is proposed in [286]
which we summarize here. Before stating the main result, some denitions
are needed.

Denition 5.26. [286] Let g : IR+ X IRm be a continuous function.

An unbounded sequence = {tn } in IR+ is said to be an admissible sequence
associated with g() if there exists a continuous function g : IR+ X
IRm such that the associated sequence {gn | (t, x)
g(t + tn , x)} converges
uniformly to g () on every compact subset of IR+ X. The function g () is
uniquely determined and called the limiting function of g() associated with .

Denition 5.27. [286] Let g : IR+ X IRm be a continuous function.

It is said to be an asymptotically almost periodic (AAP) function if, for any
unbounded sequence {tn } in IR+ there exists a subsequence of {tn } so that
is an admissible sequence associated with g().

The set of all admissible sequences associated with an AAP function g()
is denoted as (g). As an example, any continuous function g : X IRm ,
x
g(x), has all its limiting functions equal to itself. A function g : IR+
IRp IRm that is continuous and such that g(, x) is periodic for each xed
x, has limiting functions which can be written as time-shifting functions gt0 :
(t, x)
g(t + t0 , x) of g(, ) for some t0 > 0.

Lemma 5.28. [286] Suppose that g : IR+ X IRm is uniformly continuous

and bounded on IR+ for every compact X. Then g(, ) is an AAP
function.

Let f (, ) and h(, ) be AAP functions. To the system in (5.33) one asso-
ciates its reduced limiting system

z(t) = f (t, z(t))
(5.34)

(t) = h (t, z(t))
The following assumption is made, which is a simplied zero state de-
tectability hypothesis:
274 5 Stability of Dissipative Systems

Assumption 10 For any admissible sequence (f ) (h) and any

bounded solution z : IR+ X of the reduced limiting system in (5.34) satis-
fying the equation h (t, z(t)) = 0 for all t 0, it holds that either the origin
is a limit point of z() or z(t0 ) = 0 for some t0 0.
Let us now recall the KYP property for time-varying systems (this is (4.76)
with the explicit dependence on time):
Assumption 11 There exists a continuously dierentiable, positive denite
and proper storage function V : IRn IR+ such that:
V
x (x)f (t, x) 0 t 0, x IR
n

(5.35)
 V T
h(t, x) = x (x)g(t, x) , t 0, x IR n

We are now ready for the following proposition:

Proposition 5.29. [286] Consider a system of the form (5.33), with the
output feedback law u = ky, k > 0. Let Assumption 11 hold, and Assump-
 T
tion 10 hold with the output function h(t, x) = Vx (x)[f (t, x), g(t, x)] . Let
f (, ) and g(, ) be both AAP functions. Then the origin of the closed-loop
system is uniformly globally asymptotically stable. Conversely, the uniform
global asymptotic stability implies Assumption 10 when f (, ) and h(, ) are
locally Lipschitz continuous, uniformly in t.

There is in fact a strong link between AAP functions and the condition
of persistency of excitation of a bounded matrix-valued function : IR+
 t+
IR , which states that t (s) T (s)ds Ip for some > 0, some > 0
pq

and all t 0. The persistency of excitation is a well-known condition which

guarantees the convergence of parameters in adaptive control of linear time
invariant systems, and is consequently a tool which allows to prove the asymp-
totic convergence towards the equilibrium. When h : (t, x) T (t)x is an
AAP function, the persistency of excitation can be interpreted as a nonzero
property of limiting functions.

5.4.3 Evolution Variational Inequalities

Let us come back on the evolution variational inequalities as in Section 3.9.5.

We consider the linear case, that is a system similar to the system in (3.218)
(3.222) and its transformed form (3.225). We however consider now the con-
trolled case, i.e.:

dz
 (t) RAR1 z(t) RF u(t), v z(t) 0, v Ku , a.e. t 0 (5.36)
dt
 5.4 Stabilization by Output Feedback 275

with an output y = Cx + Du = CR1 z + Du, Ku = {h IRn | (CR1 h +

Du K}. Remember that R2 B = C T . The input matrix B is hidden in this
formulation, but we recall that the variational inequality (5.36) is equivalent
to the inclusion


x(t) Ax(t) F u BNK (y(t))



y(t) = Cx(t) + Du (5.37)





y(t) K, t 0
via the state transformation z = Rx. We consider a static state feedback
u = Gx. We are therefore back to the case of an output with no feedthrough
term y = (C + DG)x K and K = {h IRn | CR1 h + DF Gh K}. The
closed-loop system thus becomes

dz
 (t) R(A + F G)R1 z(t), v z(t) 0, v K, a.e. t 0 (5.38)
dt
There are two steps in the stabilization design:
To render the triple (A + F G, B, C + DG) positive real
To study the asymptotic stability
One notes that we could also consider an output feedback u = Gy, in which
case the rst step would be to test the PRness of the triple (A + F GC, B, C +
DGC). If F = B and D = 0 the rst step can be solved using Theorem 2.64
and involves conditions on the triple (A, B, C).
Lemma 5.30. [82] Let (A+F GC, B, C +DGC) be positive real. If ker[R(A+
F GC)R1 + R1 (A + F GC)T R] K = {0}, then the trivial solution of the
system
dz

 dt (t) R(A + F GC)R1 z(t), v z(t) 0, v K, a.e. t 0







z(t) K, t 0



z(0) = Rx0 , R2 B = C T (5.39)







K = {h IRn | CR1 h K}





y(t) K t 0
is asymptotically stable.
One sees that the output feedback stabilization problem for evolution vari-
ational inequalities of the form (5.36) is consequently more complex than the
usual problem of rendering a system SPR by static output feedback, as it
involves two input matrices: F which is the controller matrix, B which
characterizes the convex set K in which the state z() lives.
276 5 Stability of Dissipative Systems

5.5 Equivalence to a Passive System

Byrnes, Isidori and Willems [89] have found sucient conditions for a non-
linear system to be feedback equivalent to a passive system with a positive
denite storage function. See Chapter A for a short review on dierential ge-
ometry tools for nonlinear systems. Consider a nonlinear system described
by

x(t) = f (x(t)) + g(x(t))u(t), x(0) = x0
(5.40)

y(t) = h(x(t))

Denition 5.31. The system (5.40) is feedback equivalent to a passive system

if there exists a feedback u(x, t) = (x) + (x)v(t) such that the closed-loop
system (f (x) + g(x)(x), g(x)(x), h(x)) is passive.

This is an extension to the nonlinear case of what is reported in Sections

3.5 and 2.15.3 (2 ). This is often referred to as the problem of passication of
nonlinear systems [400]. The system has relative degree {1, . . . , 1} at x = 0 if
Lg h(0) = h(x)
x g(x)|x=0 is a non singular (m m) matrix. If in addition the
vector eld g1 (x), . . . , gm (x) is involutive then the system can be written in
the normal form

z(t) = q(z(t), y(t))
(5.41)

y(t) = b(z(t), y(t)) + a(z(t), y(t))u(t)
where
b(z, y) = Lf h(x)
(5.42)

a(z, y) = Lg h(x)
The normal form is globally dened if and only if
H1: Lg h(x) is non singular for all x
H2: The columns of g(x)[Lg h(x)]1 form a complete vector eld
H3: The vector eld formed by the columns of g(x)[Lg h(x)]1 commutes
The zero dynamics describes the internal dynamics of the system when
y 0 and is characterized by

z(t) = q(z(t), 0)

Dene the manifold Z = {x : h(x) = 0} and

f(x) = f (x) + g(x)u (x) (5.43)

2
The problem of rendering the quadruple (A, B, C, D) passive by pole shifting is
to nd IR such that (A + In , B, C, D) is PR.
 5.5 Equivalence to a Passive System 277

with

u (x) = [Lg h(x)]1 Lf h(x) (5.44)

Let f (x) be the restriction to Z of f(x). Then the zero dynamics is also
described by

x(t) = f (x(t)) for all x Z (5.45)

Denition 5.32. Assume that the matrix Lg h(0) is nonsingular. Then the
system (5.40) is said to be
1. Minimum phase if z = 0 is an asymptotically stable equilibrium of (5.45),

2. Weakly minimum phase if W (z) C r , r 2, with W (z) positive

denite, proper and such that Lf W (z) 0 locally around z = 0

These denitions become global if they hold for all z and H1-H3 above
hold.

Denition 5.33. x0 is a regular point of (5.40) if rank{Lg h(0)} is constant

in a neighborhood of x0 .

Recall that a necessary condition for a strictly proper transfer to be PR

is to have relative degree equal to 1. The next theorem extends this fact for
multivariable nonlinear systems. We will assume in the sequel that rank g(0) =
rank dh(0) = m.

Theorem 5.34. [89] Assume that the system (5.40) is passive with a C 2
positive denite storage function V (x). Suppose x = 0 is a regular point.
Then Lg h(0) is nonsingular and the system has a relative degree {1, . . . , 1} at
x = 0.

Proof: If Lg h(0) is singular, there exists u(x) = 0 for x in the neighborhood

N(0) of x = 0 such that
Lg h(x)u(x) = 0
Since rank{dh(x)} = m, for all x N (0), we have

(x) = g(x)u(x) = 0

for all x N (0). Given that the system (5.40) is passive it follows that
Lg V (x) = hT (x) so that

L2 V (x) = L [Lg V (x)u(x)] = L [uT (x)h(x)]

where
278 5 Stability of Dissipative Systems

(uT h)
L [uT h] =
x
and

 
(uT h) (uT h) (uT h)
x (x) = x1 , . . . , xn

T T
= [ u T h u T h
x1 h(x) + u (x) x1 ; . . . ; xn h(x) + u (x) xn ]
(5.46)
T u u T h h
=h (x)[ x 1
, . . . , x 2
] +u (x)[ x 1
, . . . , x n
]

= hT (x) u T h
x + u (x) x

Then

L2 V (x) = L [uT (x)h(x)]

= hT (x)L u(x) + uT (x)Lh(x)

(5.47)
= (L u(x))T h(x) + uT (x)L h(x)

= v T (x)h(x)
with
v T (x) = (L u(x))T + uT (x)L
=
Let t (xt0 ) denote the ow of the vector eld (), i.e. the solution of (t)
((t)) for 0 = x(t0 ). Dene f (t) = V (t (0)). Using Taylors Theorem for
n = 2 we have
1
f (t) = f (0) + f (1) (0)t + f (2) (s) t2
2
where 0 s t. Note that


f (t) = V (t (0))



V (t (0)) V (
t (0))
f (1) (t) = = ((t)) = L V (t (0)) (5.48)





(2) (1)
f (t) = f (t) = L f (1) (t) = L2 V (t (0))

Therefore
1
V (t (0)) = V (0) + L V (0)t + L2 V (s (0)) t2
2
Given that V (0) = 0 we have
 5.5 Equivalence to a Passive System 279
V (x)
L V (0) = x g(x)u(x)|x=0

= Lg V (x)u(x)|x=0 (5.49)

= hT (0)u(0) = 0
Thus
1
V (t (0)) = v T (s (0))h(s (0)) t2
2
Recall that h(x) h()
x g(x)u(x) = 0, for all x and in particular we have g()u() =
0 which implies that h() = 0 h() = 0 h() = where IR is a

constant. Thus h(t (0)) = h(0) = 0 and then V (t (0)) = 0 t (0) = 0
(0)) = 0 which is a contradiction. Therefore Lg h(0) must be nonsingular.
Recall that a necessary condition for a strictly proper transfer to be PR is
to be minimum phase. The next theorem extends this fact to general nonlinear
systems. A function V : IRn IR+ is non degenerate in a neighborhood of
2
x = 0 if its Hessian matrix xV2 (x) has full rank n in this neighborhood.
Theorem 5.35. [89] Assume that the system (5.40) is passive with a C 2
positive denite storage function V (). Suppose that either x = 0 is a regular
point or that V () is non degenerate. Then the system zero-dynamics locally
exists at x = 0 and the system is weakly minimum phase.
Proof: In view of Theorem 5.34 the system has relative degree {1 . . . 1} at
x = 0 and therefore its zero-dynamics locally exists at x = 0. Dene the
positive denite function W = V |Z with Z = {x : h(x) = 0}. Since
the system is passive we have 0 Lf V (x) and Lg V (x) = hT (x). Dene
f (x) = f (x) + g(x)u (x) and u (x) = [Lg h(x)]1 Lf h(x). Thus:
0 Lf V (x)
= Lf V (x) Lg V (x)u (x)
(5.50)
= Lf V (x) hT (x)u (x)
= Lf V (x)
along any trajectory of the zero dynamics (h(x) = 0).

The two theorems above show essentially that any passive system with a
positive denite storage function, under mild regularity assumptions, neces-
sarily has relative degree {1 . . . 1} at x = 0 and is weakly minimum phase.
These two conditions are shown to be sucient for a system to be feedback
equivalent to a passive system as stated in the following theorem.

Theorem 5.36. [89] Suppose x = 0 is a regular point. Then the

system (5.40) is locally feedback equivalent to a passive system with C 2
storage function V () which is positive denite, if and only if (5.40)
has relative degree {1 . . . 1} at x = 0 and is weakly minimum phase.
280 5 Stability of Dissipative Systems

This has been extended to the relative degree zero case in [435]. Specically
one considers systems of the classical form

x(t) = f (x(t)) + g(x(t))u(t), x(0) = x0
(5.51)

y(t) = h(x(t)) + j(x(t))u(t)
with x(t) IRn , u(t) IRm , y(t) IRm , f () and g() are smooth vector elds,
f (0) = 0, h(0) = 0, rank[g(0)] = m. The notion of invertibilty will play a role
in the next result, and is therefore introduced now.

Denition 5.37. The system in (5.51) is invertible at x = 0 with relative

order 1 if
(i) The matrix j(x) has constant rank m p in a neighborhood N of x = 0
(ii) If D(x) is a p m matrix of smooth functions such that, for all x N:
rank[D(x)] = p and D(x)j(x) = 0, then the (m + p) m matrix

j(x)
H(x) =
Lg [D(x)h(x)]
has constant rank m for all x N.
If this property holds for N = IRn then the system is said uniformly invertible
with relative order 1.

The following links invertibilty with passivity.

Proposition 5.38. [435] Consider the system in (5.51), let N be a neigh-

borhood of x = 0 and assume that
(i) j(x) has constant rank m p for all x N.
Let D(x) be a p m matrix the rows of which are linearly independent for
all x N. Let

j(x)
H(x) =
Lg [D(x)h(x)]
and assume that

(ii) H(x) has constant rank for all x N.

Suppose that the system is passive with a C 2 positive denite storage func-
tion V (). Then there is a neighborhood N N such that the system is in-
vertible with relative order 1 for all x N.
 5.6 Cascaded Systems 281

We then have the following proposition on feedback equivalence to a pas-

sive system

Proposition 5.39. [435] Consider a system as in (5.51) and assume that

it satises the regularity hypotheses of Proposition 5.38. Then there exists a
regular static state feedback which locally transforms the system into a passive
system having a C 2 positive denite storage function, if and only if the system
is invertible with relative order 1 and is weakly minimum phase.

The notion of weak minimum phase for (5.51) is similar to that for sys-
tems as in (5.40), except that the input u (x) is changed, since the output is
changed. The zero-dynamics input is calculated as the unique solution of

j(x)
H(x)u (x) + =0
Lf [D(x)h(x)]
and is such that the vector eld f (x) = f (x) + g(x)u (x) is tangent to the
submanifold Z = {x N | D(x)h(x) = 0}. The proof of Proposition
5.39 relies on the cross-term cancellation procedure and a two-term Lyapunov
function, so that the results of Section 7.3.3 may be applied to interpret the
obtained closed-loop as the negative feedback interconnection of two dissipa-
tive blocks. Further works on feedback equivalence to a passive system can be
found in [56, 99, 130, 156, 161, 236, 303, 414]. The adaptive feedback passivity
problem has been analyzed in [453].

Remark 5.40. Most of the results on feedback equivalence to a passive sys-

tem, relative degree, zero dynamics, are extended to nonlinear discrete-time
systems in [372, 374].

5.6 Cascaded Systems

Cascaded systems are important systems that appear in many dierent prac-
tical cases. We will state here some results concerning this type of systems
which will be used later in the book. Consider a cascaded system of the fol-
lowing form

(t) = f0 ((t)) + f1 ((t), y(t))y(t)



x(t) = f (x(t)) + g(x(t))u(t) (5.52)





y(t) = h(x(t))
The rst equation above is called the driven system while the second and
third equations are called the driving system.
282 5 Stability of Dissipative Systems

Theorem 5.41. [89] Consider the cascaded system (5.52). Suppose that
the driven system is globally asymptotically stable and the driven system is
(strictly) passive with a C r , r 2, storage function V () which is positive
denite. The system (5.52) is feedback equivalent to a (strictly) passive system
with a C r storage function which is positive denite.

The cascaded system in (5.52) can also be globally stabilized using a

smooth control law as is stated in the following Theorem for which we need
the following denitions. Concerning the driving system in (5.52) we dene
the associate distribution [227, 381]

D = span{adkf gi : 0 k n 1, 1 i m} (5.53)
and the following set

f L V (x) = 0, for all D, 0 m < r}

S ={x X : Lm (5.54)

Theorem 5.42. [89] Consider the cascaded system (5.52). Suppose that the
driven system is globally asymptotically stable and the driven system is passive
with a C r , r 1, storage function V () which is positive denite and proper.
Suppose that S ={0}. Then the system (5.52) is globally asymptotically stabi-
lizable by the smooth feedback

uT (, x) = Lf1 (,h(x)) U () (5.55)

where U () is a Lyapunov function for the driven system part of the cascaded
system (5.52).

Some additional comments on the choice of u in (5.55) are given in Sub-

section 7.3.3, where the role of cross-terms cancellation is highlighted. Further
work on the stabilization of cascaded systems using dissipativity may be found
in [101].

5.7 Input-to-State Stability (ISS) and Dissipativity

Close links between passive systems and Lyapunov stability have been shown
to exist in the foregoing sections. This section demonstrates that more can
be said. E.D. Sontag has introduced the following notion of input-to-state
stability (ISS): given a system

x(t) = f (x(t), u(t)), x(0) = x0 (5.56)

where f (, ) is locally Lipschitz, f (0, 0) = 0, and U is a set of measurable lo-
cally essentially bounded functions from IR+ into IRm , one studies the input-
to-state mapping (x0 , u())
x() and its stability (a notion that will be
 5.7 Input-to-State Stability (ISS) and Dissipativity 283

dened next). The material in this section is to be considered as a brief intro-

duction to the eld of ISS and is taken from [468, 469]. We shall be especially
interested by the relationships with dissipativity, as the reader may expect.
The problem is the following: assume that the equilibrium x = 0 of the free
system x(t) = f (x(t), 0) is globally asymptotically stable. The question is to
determine if this property implies, or is equivalent to: [limt+ u(t) 0 =
limt+ x(t) 0], or to: [u() bounded = x() bounded ]. Equivalence is
known to be true for linear time invariant systems. This is not the case for
nonlinear systems, as the following example proves: x(t) = x(t) + (x2 (t) +
1
1)u(t) with u(t) = (2t + 2) 2 . The trajectory which starts at x0 = 2 is
1
given by x(t) = (2t + 2) 2 which is unbounded, though limt+ u(t) 0 and
x(t) = x(t) is globally asymptotically stable.

Denition 5.43. The system (5.56) is ISS if:

For each x0 there is a unique solution in C 0 (IR+ , IRn )
The map IRn U C 0 (IR+ , IRn ), (x0 , u)
x() is continuous at
(0, 0)
There exists a nonlinear asymptotic gain () of class K so that

lim sup ||x(t, x0 , u)|| (|| u || )

t+

uniformly on x0 in any compact set and all u U.

Continuity in the second item means that for any sequence {x0,n } such
that limn+ x0,n = x0 and any sequence {un } such that limn+ un = u,
then the solution x(t; x0,n , un ) x(t; x0 , u) as n +. Then the following
holds:

Theorem 5.44. [468,469] The system (5.56) is ISS if and only if there exists
a class KL-function (, ), and two functions 0 (), 1 () of class K such that

t 
||x(t, x0 , u)|| (||x0 ||, t) + 0 est 1 (||u(s)||)ds (5.57)
0

for all t 0. Equivalently, the system is ISS if

||x(t, x0 , u)|| (||x0 ||, t) + (||u|| ) (5.58)

for some class KL function (, ) and some class K function ().

Let us now dene an ISS-Lyapunov function.

284 5 Stability of Dissipative Systems

Denition 5.45. A dierentiable storage function V () is an ISS-Lyapunov

function if there exists two functions 1 () and 2 () of class K such that

V T (x)f (x, u) 1 (||u||) 2 (||x||) (5.59)

for all x IR , u IR . Equivalently, a storage function with the property
n m

that there exist two class-K functions () and () such that

||x|| (||u||) = V T (x)f (x, u) (||x||) (5.60)

holds for all x IR and all u IR , is an ISS-Lyapunov function.
n m

One notices that (5.59) means that along trajectories of the system
dV x
dt (t) 1 (||u(t)||) 2 (||x(t)||). One also immediately realizes that (5.59)
is a dissipation inequality (in its innitesimal form, so that indeed some dif-
ferentiability of the storage function is required). Integrating on any interval
[t0 , t1 ] we get that along the systems trajectories

t
V (x(t1 )) V (x(t0 )) w(u(s), x(s))ds, (5.61)
0

where the supply rate w(u, x) = 1 (||u||)2 (||x||). The dissipation inequality
(5.61) might be written even if V () is not dierentiable, using the notion of
viscosity solutions. However, as far as ISS is concerned, the following holds:

Theorem 5.46. [469] The system in (5.56) is ISS if and only if it admits a
smooth ISS-Lyapunov function.

This strong result shows that ISS is more stringent that dissipativity. We
recall that smooth means innitely dierentiable.

Example 5.47. [468] Consider x(t) = x3 (t) + x2 (t)u1 (t) x(t)u2 (t) +
u1 (t)u2 (t). When u1 and u2 are zero, the origin x = 0 is globally asymp-
totically stable. This can be easily checked with the Lyapunov function
2
V (x) = x2 . One also has V T (x)(x3 + x2 u1 xu2 + u1 u2 ) 29 x4 , pro-
vided that 3|u1 | |x| and 3|u2 | x2 . This is the case if ||u|| (||x||),
2 2
with (r) = min( r3 , r3 ). Thus V (x) = x2 is an ISS-Lyapunov function with
(r) = 29 r4 and = 1 in (5.60).

Let us now introduce a slightly dierent property known as the integral

ISS (in short iISS):

Denition 5.48. The system in (5.56) is iISS provided that there exist func-
tions () and () of class K , and a function (, ) of class KL such that

t
(||x(t)||) (||x0 ||, t) + (||u(s)||)ds (5.62)
0
holds along solutions of the system.
 5.7 Input-to-State Stability (ISS) and Dissipativity 285

An iISS-Lyapunov function is dened as follows:

Denition 5.49. A smooth storage function V () is an iISS-Lyapunov func-
tion for the system in (5.56) if there is a function () of class K and a
positive denite function : IR+ IR+ , such that

V (x(t), u(t)) (||x(t)||) + (||u||) (5.63)

for all x(t) IR and all u(t) IR .
n m

Notice that V (x(t), u(t)) = V x (f (x(t), u(t)). The fact that (5.63) is a
dissipation inequality (in its innitesimal form) with supply rate w(x, u) =
(||x(t)||) + (||u||) is obvious. Since every class K function is also positive
denite, an ISS-Lyapunov function is also an iISS-Lyapunov function. But the
converse is not true. Similarly to Theorem 5.46 one has the following:
Theorem 5.50. [469] The system in (5.56) is iISS if and only if it admits
a smooth iISS-Lyapunov function.
Example 5.51. Let us present an example of a scalar system that is not ISS
but is iISS. Consider

x(t) = tan1 (x(t)) + u(t) (5.64)

This system is not ISS because the input u(t) 2 gives unbounded trajecto-
1
ries. But it is iISS. Indeed choose V (x) = x tan (x). Then
 2
V (x(t), u(t)) tan1 (|x(t)| + 2|u(t)| (5.65)
and consequently this storage function V () is an iISS-Lyapunov function.
More can be said about iISS stability, as the following shows:
Theorem 5.52. The system in (5.56) is iISS if and only if the uncontrolled
system x(t) = f (x(t), 0) has a globally asymptotically stable xed point x = 0
and there is a smooth storage function V () such that for some function ()
of class K

V (x(t), u(t)) (||u(t)||) (5.66)

for all x(t) IR and all u(t) IR .
n m

Let us now state a result on ISS in which zero-state detectability (Deni-

tion 5.12) intervenes:
Theorem 5.53. [469] A system is iISS if and only if there exists a continuous
output function y = h(x), h(0) = 0, which provides zero-state detectability and
dissipativity in the following sense: there exists a storage function V () and a
function () of class K and a positive denite function () so that
V (x(t), u(t)) (||u(t)||) (h(x(t))) (5.67)
for all x(t) IRn and all u IRm .
286 5 Stability of Dissipative Systems

The next results may be seen as a mixture of results between the stability
of feedback interconnections as in Figure 5.2, the ISS property, and quasi-
dissipative systems. Two denitions are needed before stating the results.
Denition 5.54. A dynamical system x(t) = f (x(t), u(t)), y(t) = h(x(t)),
with f (, ) and h() locally Lipschitz functions, is strongly nite time de-
tectable if there exists a time t > 0 and a function () of class K such
that for any x0 IRn and for any u U the following holds:

t
(uT (s)u(s) + y T (s)y(s))ds (||x0 ||) (5.68)
0

This property is to be compared to the zero state detectability in Denition

5.17. Roughly, a system that is strongly nite time detectable and starts with
a large initial state, must have either a large input or a large output, or both.
A system that is ZSD in the sense of Denition 5.17 must have a large output
when the initial state is large.

Denition 5.12 = Denition 5.17 = Denition 5.54 = Denition 5.12

Denition 5.55. [403] The system x(t) = f (x(t), u(t)), y(t) = h(x(t))
is input-to-state ultimately bounded (ISUB), or has input-to-state ultimately
bounded trajectories, if for given a 0 and r 0, one has

||x(0)|| r and supt0 ||u(t)|| a

there exist Ca,r r such that supt0 ||x(t)|| Ca,r

(5.69)
and

there exist D 0 (independent of r) and tr 0 (independent of a)

such that supttr ||x(t)|| D

This denition is closely related to the ISS with respect to a compact

invariant set. However, ISUB implies only boundedness, not stability, and is
therefore a weaker property. The next proposition is an intermediate result
which we give without proof.
Proposition 5.56. [403] Suppose that the system x(t) = f (x(t), u(t)),
y(t) = h(x(t)) has uniform nite power gain, with a locallly bounded radi-
ally unbounded storage function, and is strongly nite time detectable. Then
it is ISUB.
 5.7 Input-to-State Stability (ISS) and Dissipativity 287

The denition of a nite power gain is in Denition 5.7. Then we have the
following.

Theorem 5.57. [403] Suppose that each of the subsystem H1 and H2 in

Figure 5.2 has the dynamics xi (t) = fi (xi (t), ui (t)), yi (t) = hi (xi (t)), i = 1, 2,
and is
Quasi-dissipative with general supply rate wi (ui , yi ), with a locally bounded
radially unbounded storage function
Strongly nite time detectable
Suppose that there exists > 0 such that the matrix Q in (5.9) is negative
denite. Then the feedback system is ISUB.

Proof: From Proposition 5.8 one sees that the feedback system has uniform
nite power gain. Suppose that V1 () and V2 () are locallly bounded radially
unbounded storage functions for H1 and H2 respectively. Then V1 ()+V2 () is
a locally bounded radially unbounded storage function of the feedback system.
Let us now show that the feedback system is strongly nite-time detectable.
We have
 t1 T T T
0 [r1 (s)r1 (s) + y2 (s)y2 (s) + y1 (s)y1 (s)]ds
 t1
0 [uT1 (s)u1 (s) + y1T (s)y1 (s)]ds (5.70)

1 (||x1 (0)||)
and
 t2
0 [r2T (s)r2 (s) + y2T (s)y2 (s) + y1T (s)y1 (s)]ds
 t1
0
[uT2 (s)u2 (s) + y2T (s)y2 (s)]ds (5.71)

2 (||x2 (0)||)
for some t1 > 0, t2 > 0, 1 () and 2 () K . Combining (5.70) and (5.71)
we obtain

 t
0 [r1T (s)r1 (s) + r2T (s)r2 (s) + y2T (s)y2 (s) + y1T (s)y1 (s)]ds

12 [1 (||x1 (0)||) + 2 (||x2 (0)||)] (5.72)

(max{||x1 (0)||, ||x2 (0)||})

1
where t
= max(t1 , t2 ), and
() = 2 min{1 (), 1 ()} K . Using Proposi-
tion 5.56, the result follows.
288 5 Stability of Dissipative Systems

The literature on ISS stability is abundant and our objective in this section
was just to mention the connections with dissipativity. The interested reader
should have a look at [469] and the bibliography therein to realize the richness
of this eld.

5.8 Passivity of Linear Delay Systems

The above developments focus on particular classes of smooth nite dimen-
sional dynamical systems. Let us investigate another type of systems that
does not t within these classes, namely time-delayed systems. Stability and
control of linear systems with delayed state are problems of recurring interest
since the existence of a delay in a system representation may induce instabil-
ity, oscillations or bad performances for the closed-loop scheme. In this section
we shall consider the passivity problem of a linear system described by dier-
ential equations with delayed state. The interconnection schemes with passive
systems will be also treated. The proposed approach is based on an appropri-
ate Lyapunov-Krasovskii functional construction. The material presented in
this section follows the analysis given in [377]; see also [291, 380]. The corre-
sponding results may include or not delay information and are expressed in
terms of solutions of some algebraic Riccati equations. The results presented
here can be extended to the multiple delays case by an appropriate choice of
the Lyapunov functional.

5.8.1 Systems with State Delay

Consider the following system

x(t) = Ax(t) + A1 x(t ) + Bu(t)
(5.73)

y(t) = Cx(t)
where x(t) IRn , y(t) IRp , u(t) IRp are the state, the output and the input
of the system and denotes the delay. The matrices A IRnn , A1 IRnn
and B IRnm are constant. Time-delay systems may be seen as innite-
dimensional systems. In particular the initial data for (5.73) is a function
: [, 0] IRn that is continuous in the uniform convergence topology
(i.e. |||| = sup tt ||(t)||). The initial condition is then denoted as x(t0 +
) = () for all [, 0]. There exists a unique continuous solution [201,
Theorem 2.1] which depends continuously on the initial data (x(0), ) in the
following sense: the solution of (5.73) is denoted as

x(t + ) if t + 0
xt () = (5.74)

(t + ) if t + 0
 5.8 Passivity of Linear Delay Systems 289

with [, 0]. Let {n ()}n0 be a sequence of functions that converge

uniformly towards (). Then xn (0) x(0), and xt,n () converges uniformly
towards xt (). The transfer function of the system in (5.73) is given by G() =
C( A A1 e )1 B, with (A + A1 e ) C where (M ) = { C |
(In M ) is full rank} for M IRnn [178].
The main result of this section can be stated as follows:

Lemma 5.58. If there exists positive denite matrices P > 0 and S > 0 and
a scalar 0 such that

= AT P + P A + P A1 S 1 AT1 P + S < C T C
(5.75)
T
C=B P

then the system (5.73) satises the following inequality:

t 1
t
0
uT (s)y(s)ds 2 [V (x(t), t) V (x(0), 0)] 12 0
y T (s)y(s)ds (5.76)

where
t
V (x(t), t) = x(t)T P x(t) + t x(s)T Sx(s)ds (5.77)

Remark 5.59. Note that the system (5.73) is passive only if = 0. Roughly
speaking for > 0 we may say system (5.73) is less than output strictly
passive. This gives us an extra degree of freedom for choosing P and S in
(5.75) since inequality in (5.75) becomes more restrictive for = 0. We can
expect to be able to stabilize the system (5.73) using an appropriate passive
controller as will be seen in the next section. Note that for < 0 the system
is output strictly passive but this imposes stronger restrictions on the system
(see (5.75)).

Proof: From (5.73) and the above conditions we have

290 5 Stability of Dissipative Systems
t t
2 0
uT (s)y(s)ds = 2 0
uT (s)Cx(s)ds
t
=2 0 uT (s)B T P x(s)ds
t t
= 0 uT (s)B T P x(s)ds + 0 x(s)T P Bu(s)ds
 t 2 dx T
= 0 ds Ax(s) A1 x(s ) P x(s) +
 dx * (5.78)
+ xT (s)P ds Ax(s) A1 x(s ) ds
 t 2 d(xT (s)P x(s))
= 0 ds x(s)T (AT P + P A)x(s)
*
x(s )T AT1 P x(s) x(s)T P A1 x(s ) ds
 t 2 dV (s) 3
= 0 ds x(s)T x(s) + I(x(s), x(s )) ds

where is given by (5.75) and

 T
I(x(t), x(t )) = S 1 AT1 P x(t) x(t ) S
 
S 1 AT1 P x(t) x(t )
Note that V (x, t) is a positive denite function and I(x(t), x(t )) 0
for all the trajectories of the system. Thus from (5.76) and (5.78) it follows
that:

t 1 1
t
0
uT (s)y(s)ds 2 [V (x(t), t) V (x(0), 0)] 2 0
xT (s) x(s)ds

1
t
2 [V (x(t), t) V (x(0), 0)] 12 0 xT (s)C T Cx(s)ds
t
21 V (x(0), 0) 12 0
y T (s)y(s)ds t > 0
(5.79)
Therefore if = 0 then the system is passive.

5.8.2 Interconnection of Passive Systems

Let us consider the block interconnection depicted in Figure 5.1 where H1

represents the system (5.73) and H2 is the controller which is input strictly
passive as dened above, i.e. for some > 0

t
t
uT2 (s)y2 (s)ds 22 + uT2 (s)u2 (s)ds (5.80)
0 0
 5.8 Passivity of Linear Delay Systems 291

for some IR and for all t 0. H2 can be a nite dimensional linear

system for example. For the sake of simplicity we will consider H2 to be an
asymptotically stable linear system. We will show next that the controller
satisfying the above property will stabilize the system (5.73). From Lemma
5.58, the interconnection scheme and (5.80) we have


u1 = u

y1 = y
(5.81)

u 2 = y1

y2 = u1 .
Therefore from (5.76) and (5.80) we have

t t
0= 0 uT1 (s)y1 (s)ds + 0 uT2 (s)y2 (s)ds
t t
12 V (x(0), 0) 12 0 y1T (s)y1 (s)ds 22 + 0 uT2 (s)u2 (s)ds
t
2 + ( 12 ) 0
y1T (s)y1 (s)ds

where 2 = 12 V (x(0), 0) + 22 . If 12 > 0 then y1 is L2 . Since H2 is

an asymptotically stable linear system with an L2 input, it follows that the
corresponding output y2 is also L2 . Given that the closed-loop system is com-
posed of two linear systems, the signals in the closed-loop cannot have peaks.
Therefore all the signals converge to zero which means the stability of the
closed loop system.

5.8.3 Extension to a System with Distributed State Delay

Let us consider the following class of distributional convolution systems:

x(t) = Ax(t) + Bu(t)
(5.82)

y(t) = Cx(t)
where A denotes a distribution of order 0 on some compact support [, 0].
Let us choose

A = A() + A1 ( 1 ) + A2 () (5.83)
where () represents the Dirac delta functional and A2 () is a piece-wise
continuous function. Due to the term A2 () the system has a distributed
delay. For the sake of simplicity we shall consider A2 () constant. The system
(5.82) becomes
0
x(t) = Ax(t) + A1 x(t 1 ) + A2 x(t + )d + Bu(t)
(5.84)

y(t) = Cx(t)
292 5 Stability of Dissipative Systems

Some details on the well-posedness of such state delay systems are provided
in Section A.6 in the Appendix.

Lemma 5.60. If there exists positive denite matrices P > 0, S1 > 0 and
S2 > 0 and a scalar 0 such that

AT P + P A + P A1 S11 AT1 P + S1 + (P A2 S21 AT2 P + S2 )

( ) =
< C T C (5.85)

C = BT P

then the system (5.84) veries the following inequality:

t t
0
u(s)T y(s)ds 12 [V (t) V (0)] 12 0 y(s)T y(s)ds (5.86)

where t
V (x(t), t) = x(t)T P x(t) + t1
x(s)T S1 x(s)ds+
(5.87)
0 t
+ ( t+ x(s)T S2 x(s)ds)d.

Proof: We shall use the same steps as in the proof of Lemma 5.58. Thus
from (5.84) and the above conditions we have
t t t
2 0
uT (s)y(s)ds = 2 0
uT (s)Cx(s)ds = 2 0
uT (s)B T P x(s)ds
t t
= 0 uT (s)B T P x(s)ds + 0 xT (s)P Bu(s)ds
 t ) dx
= 0 ds Ax(s) A1 x(s 1 )

0 3T (5.88)
A2 x(s + )d P x(s)ds+

t ) dx
+ 0
x(s)T P ds Ax(s) A1 x(s 1 )
0 3

A2 x(s + )d ds

We also have
 5.8 Passivity of Linear Delay Systems 293
t  t 2 d(x(s)T P x(s))
2 0 uT (s)y(s)ds = 0 ds x(s)T (AT P + P A)x(s)
*
x(s 1 )T AT1 P x(s) xT (s) P A1 x(s 1 ) ds
t2 0
0
xT (s)P
A2 x(s + )d +
  3
0 T T
+
x (s + )A2 d P x(s) ds

 t 2 dV (s)
= 0 ds x(s)T ( )x(s) + I1 (x(s), x(s 1 )) +

+ I2 (x(s), x(s + ))} ds

(5.89)
where ( ) is given by (5.85) and
 T
I1 (x(t), x(t 1 )) = S11 AT1 P x(t) x(t 1 ) S1
(5.90)
 
S11 AT1 P x(t) x(t 1 )
 0  1 T T
I2 (x(t), x(t + )) =
S2 A2 P x(t) x(t + ) S2
(5.91)
 
S21 AT2 P x(t) x(t + ) d
Note that V (t) is a positive denite function and I1 (x(t), x(t 1 )) 0
and I2 (x(t), x(t + )) 0 for all the trajectories of the system. Thus from
(5.85) and (5.87) it follows tha

t 1 1
t
0 uT (s)y(s)ds 2 [V (x(t), t) V (x(0), 0)] 2 0 xT (s) ( )x(s)ds

1
t
2 [V (x(t), t) V (x(0), 0)] 12 0
xT (s)C T Cx(s)ds
t
21 V (x(0), 0) 12 0
y T (s)y(s)ds t > 0
(5.92)
Therefore if = 0 then the system is passive.
Remark 5.61. The presence of a distributed delay term in the system (5.84)
imposes extra constraints in the solution of inequality (5.85). Note that for
= 0 we recover the previous case having only a point state delay. Extensions
of the result presented in this section can be found in [413]. Other work may
be found in [15,100,158,192,328,331]. The passication of time-delay systems
with an observer-based dynamic output feedback is considered in [173]. Re-
sults for systems with delay both in the state and the input may be found
in [452]. The stability and L2 -gain of a class of switching systems with delay
with time-continuous solutions have been analysed in [481].
294 5 Stability of Dissipative Systems

Remark 5.62. Note also that given that the system (5.84) satises the inequal-
ity (5.92), it can be stabilized by an input strictly passive system as described
in the previous section. Furthermore due to the form of the Riccati equation
the upper bound for the (sucient) distributed delay (seen as a parameter)
may be improved by feedback interconnection for the same Lyapunov-based
construction. Such result does not contradict the theory since the derived
condition is only sucient, and not necessary and sucient.

5.8.4 Absolute Stability

Let us end this section on time-delay systems by noting that the absolute
stability problem for systems of the form


x(t) = Ax(t) + Bx(t ) + Dw(t)



y(t) = M x(t) + N x(t ) (5.93)





w(t) = (t, y(t))

has been studied in [202], with x() = () for all [, 0], > 0 is the
constant delay and : IR+ IRm IRm is a static, piecewise continuous in t
and globally Lipschitz continuous in y nonlinearity. This nonlinearity satises
the sector condition [(t, y(t)) K1 y(t)]T [(t, y(t)) K2 y(t)] 0 where K1
and K2 are constant matrices of appropriate dimensions and K = K1 K2
is symmetric positive denite. One says that the nonlinearity belongs to the
sector [K1 , K2 ]. The following result holds:

Proposition 5.63. [202] For a given scalar > 0, the system (5.93) is
globally uniformly asymptotically stable for any nonlinear connection in the
sector [0, K] if there exists a scalar 0, real matrices P > 0, Q > 0, R > 0
such that

AT P + P A + Q R PB + R P D M T K T AT R


(P B + R)T Q R N KT T
B R
T

<0 (5.94)

(P D M T K T )T ( N T K T )T 2 Im D R
T


( AT R)T ( B T R)T ( DT R)T R

Other works on absolute stability of time-delay systems can be found in

[53, 159, 204, 295, 298, 299, 411, 464, 537].
 5.9 Nonlinear H Control 295

5.9 Nonlinear H Control

In this section we rst briey recall basic results on H control of linear time
invariant systems, then a brief review of the nonlinear case is done. We nish
with an extension of the nite power gain notion. It has already been seen in
the case of linear time invariant systems that there exists a close relationship
between bounded realness and positive realness; see e.g. Theorem 2.23. Here
we investigate similar properties starting from the so-called Bounded Real
Lemma.

5.9.1 Introduction

Let us recall that the input/output mapping u

y = H(u) of a linear time
invariant system (A, B, C) with stable transfer function H(s) has the H
norm equal to

||y(s)||2 ||y(t)||2
||H|| = sup = sup max (H(j)) = sup (5.95)
u(s)H2 ||u(s)||2 IR u(t)=0 ||u(t)||2

where H2 is :the Hardy space of functions : C Cn analytic in Re(s) > 0,

 +  +
||f (s)||2 = 12 ||f (j)||2 d = ||f ||2 = 0 ||f (t)||2 dt < +, by Par-
sevals equality, provided f L2 (IR+ ). Thus the H norm exactly corre-
sponds to the L2 -gain of the said operator, and its nonlinear extension corre-
sponds to having

t
t
y T ( )y( )d 2 uT ( )u( )d (5.96)
0 0
for all t 0. Let us recall the following, known as the Bounded Real Lemma:

Lemma 5.64 (Bounded Real Lemma). Consider the system x(t) = Ax(t)+
Bu(t), y(t) = Cx(t). Let (A, B) be controllable and (A, C) be observable. The
following statements are equivalent:
||H|| 1
The Riccati equation AT P + P A + P BB T P + C T C = 0, has a solution
P >0

The Strict Bounded Real Lemma is as follows:

Lemma 5.65 (Strict Bounded Real Lemma). Consider the system x(t) =
Ax(t) + Bu(t), y(t) = Cx(t). The following statements are equivalent:
A is asymptotically stable and ||H|| < 1
296 5 Stability of Dissipative Systems

There exists a matrix P > 0 such that AT P + P A + P BB T P + C T C < 0

The Riccati equation AT P + P A + P BB T P + C T C = 0 has a solution
P 0 with A + BB T P asymptotically stable

The Strict Bounded Real Lemma therefore makes no controllability nor

observability assumptions. In order to make the link with the bounded realness
of rational functions as introduced in Denition 2.24, let us recall that a
transfer function H(s) Cmm is bounded real if and only if all the elements
of H(s) are analytic in Re(s) 0 and ||H|| , or equivalently 2 Im
H
(j)H(j) 0 for all Re(s) 0, > 0. We have only replaced the
upperbound 1 in Denition 2.24 by . The transfer function H(s) is said to
be strictly bounded real if there exists > 0 such that H(s ) is bounded
real. It is strongly bounded real if it is bounded real and 2 Im DDT > 0,
where D = G().
The extension of the above lemmas to the relative degree 0 case where a
direct feedthrough matrix D = 0 exists is as follows:

Lemma 5.66. The transfer matrix H(s) = C(sIn A)1 B + D of the system
(A, B, C, D) is stable and has an H -norm ||H|| < if and only if there
exists a matrix P = P T > 0 such that
T
A P + PA PB CT

T
B P T
Im D < 0 (5.97)


C D Im

From Theorem A.61 and the LMI (5.97) one is able to recover a Riccati
inequality.
We rst
notice that the inverse of the negative denite matrix
Im DT
D = is equal to
D Im

(Im + 1 DT D)1 DT (DDT 2 Im )1
D1 = .
2
D(D D Im )
T 1
( 1 DDT 1
Im )

Notice that in particular one has 2 Im + DT D < 0 ( 2 Im DT D > 0)

and 2 Im + DDT < 0, still using Theorem A.61 applied to D < 0. This
indeed secures that the terms in D1 are dened. One then calculates the
following identities:
 5.9 Nonlinear H Control 297
T

D (DDT 2 Im )1 = (DT D 2 Im )1 DT



( 2 Im + DDT )1 D = D(DT D 2 Im )1 (5.98)




2
(DDT 2 Im )1 = D(DT D 2 Im )1 Im
From Theorem A.61 we can rewrite the LMI as
T 
B P
AT P + P A (P B C T )D1 < 0,
C

P = P T > 0. After some manipulations and using the above identities one
gets

AT P + P A (P B + C T D)(DT D 2 Im )1 (B T P + DT C) + C T C < 0

This Riccati inequality tells us that the system is dissipative with respect
to the H supply rate w(u, y) = 2 uT u y T y. This can be checked using for
instance the KYP Lemma 4.87 with the right choice of the matrices Q, R and
S. Using Theorem A.61 one can further deduce that the Riccati inequality is
equivalent to the LMI: nd P = P T > 0 such that
T
A P + P A + CT C P B + CT D
<0 (5.99)
B T P + DT C DT D Im
The equivalence between the LMI in (5.97) and the LMI in (5.99) can be
shown using once again Theorem A.61, considering this time the Schur comple-
ment of the matrix Im in (5.97). We therefore have shown the equivalence
between two LMIs and one Riccati inequality which all express the Bounded
Real Lemma. We once again stress the fundamental role played by Theorem
A.61. The main result of this part is summarized as follows.

Let 2 Im DT D > 0. The existence of a positive denite solution

P = P T to the ARI

AT P +P A+(B T P +DT C)T ( 2 Im DT D)1 (B T P +DT C)+C T C < 0

implies that the system (A, B, C, D) is strictly dissipative with respect

to the supply rate 2 uT uy T y, which in turn implies that ||H|| < .

Letting D 0 and = 1 one recovers the Riccati equation for (A, B, C)

in Lemma 5.65. The following results hold also true and somewhat extend the
above:
298 5 Stability of Dissipative Systems

Theorem 5.67. [99] Let (A, B, C, D) be a minimal realization of the transfer

function H(s) Cmm , with input y() and output y(). Then the following
statements are equivalent:
H(s) is strictly bounded real
H(s) is exponentially nite gain, i.e.

t
t
2
exp( )y ( )y( )d
T
exp( )uT ( )u( )d
0 0

for all t 0 and some > 0

There exists matrices P = P T > 0, L IRnp , W IRpm and a scalar
> 0 such that
T
T T
A P + P A + P + C C + LL = 0



P B + C T D + LW = 0 (5.100)




2
Im DT D W T W = 0

Furthermore H(s) is strongly bounded real if and only if there exists P =

P T > 0 and R = RT > 0 such that

AT P +P A+(B T P +DT C)T ( 2 Im DT D)1 (B T P +DT C)+R = 0 (5.101)

From Proposition A.63 the set of equations in (5.100) is equivalent to the

LMI
T
A P + P A + P + C T C P B + C T D
0
DT C + B T P DT D 2 Im
Similarly the next theorem holds that concerns positive realness.

Theorem 5.68. [99] Let (A, B, C, D) be a minimal realization of the transfer

function H(s) Cmm , with input y() and output y(). Then the following
statements are equivalent:
H(s) is strictly positive real
t
H(s) is exponentially passive, i.e. 0 exp( )uT ( )y( )d 0, for all t 0

The conditions of the Lefschtez-Kalman-Yakubovich Lemma 3.8 in (3.23)

are satised
 5.9 Nonlinear H Control 299

Furthermore H(s) is SSPR if and only if there exists P = P T > 0 and

R = RT > 0 such that

AT P + P A + (B T P C)T (DT + D)1 (B T P C) + R = 0 (5.102)

The Riccati equations in Theorems 5.67 and 5.68 can be deduced from
Lemma A.62. Notice that the Riccati equations are not identical from one
theorem to the other, since the considered supply rates dier: the rst
one concerns the H supply rate, while the second one concerns the pas-
sivity supply rate. The exponential dissipativity can also be expressed via
the existence of a storage function and the dissipation inequality is then
t
exp( t)V (x(t))V (x(0)) 0 exp( )uT ( )y( )d , for all t 0. If V () is con-
tinuously dierentiable, then the innitesimal form of the dissipation inequal-
ity is V (x(t)) + V (x(t)) uT (t)y(t) for all t 0. Another denition of expo-
nential dissipativity has been introduced in [156], which is strict passivity (Def-
inition 4.51) with the storage functions that satisfy 1 ||x||2 V (x) 2 ||x||2
and 3 ||x||2 S(x) for some 1 > 0, 2 > 0, 3 > 0. Such a denition
was motivated by a result of Krasovskii [263]. If a system is exponentially
dissipative in this sense then the uncontrolled system is exponentially sta-
ble. The denition in Theorem 5.68 is more general since the exponential
dissipativity implies the strict dissipativity: in case the storage function satis-
es 1 ||x||2 V (x) 2 ||x||2 then the second condition is also satised with
S(x) = V (x). The exponential nite gain property has been used in [481,539]to
study the stability of switched systems with delay and time-continuous solu-
tions.
Notice that the material presented in Section 3.8.3 nds application in the
H problem, via the so-called four-block Nehari problem. This may be seen
as an extension of the Bounded Real Lemma; see [224, Lemma 2, Theorem
3]. Further results on H control in the so-called behavorial framework, may
be found in [491].

Remark 5.69 (Finite Lp -gain). A system has a nite Lp -gain if it is dissipative

with respect to a supply rate of the form

w(u, y) = ||u||p ||y||p (5.103)

for some > 0, > 0. It is noteworthy that such supply rates satisfy the
condition 2 in Lemma 5.13 in a strong sense since w(0, y) < 0 for all y = 0.

The paper [305] concerns the standard H problem and relationships be-
tween LMI, ARE, ARI, and is worth reading.
300 5 Stability of Dissipative Systems

5.9.2 Closed-loop Synthesis: Static State Feedback

Let us make an aside on the problem of designing a feedback u(t) = v(t) +

Kx(t) applied to the linear time invariant system


x(t) = Ax(t) + Bu(t)



y(t) = Cx(t) + Du(t) (5.104)





x(0) = x0
so that the closed-loop system is dissipative with respect to the supply rate
w(v, y) = uT Ru y T Jy. Such systems, when they possess a storage function
xT P x, are named (R, P, J)dissipative [478]. The feedback gain K has to be
chosen in such a way that the closed-loop system (A + BK, B, C + DK, D) is
(R, P, J)dissipative. This gives rise to the following set of matrix equations:
T

A P + P A + C T JC = K T RK



P B + C T JD = K T R (5.105)




T
D JD = R
A suitable choice of the matrices P , R and J allows one to obtain several
standard one-block or two-block results, to which Riccati equalities corre-
spond. This is summarized as follows, where the dimensions  are not given
y
explicitly but follow from the context. The notation y = means that the
u
signal y is split into two subsignals, one still called the output y, the other
one being the input u. The following ingredients (LMI and Riccati equalities)
have already been seen in this book, under slightly dierent forms. This is once
again the opportunity to realize how the supply rate modications inuence
the type of problem one is solving.
   
y C 0 Im 0
Let y = ,C = ,D= ,J = . The matrix R in J and
u 0 Im 0 R
R in (5.105) are the same matrix. With this choice of input and matrices
one obtains from (5.105) the standard LQR Riccati equation. Indeed one
gets
T
A P + P A + C T C = K T RK
(5.106)
T
B P = RK
with R 0, J 0, P 0. If R > 0 then one can eliminate K to get the
Riccati equation

AT P + P A + C T C P BR1 B T P = 0 (5.107)
 5.9 Nonlinear H Control 301
   
y C D Im 0
Let y = ,C = ,D= ,J = . This time one gets
u 0 Im 0 Im
the normalized coprime factorization problem, still with J 0, P 0,
R 0. From (5.105) it follows that
T

A P + P A + C T C = K T RK



P B + C T D = K T R (5.108)




T
D D + Im = R
If R > 0 then both R and K can be eliminated and we obtain the normal-
ized coprime factorization Riccati equation

AT P +P A+C T C(P B+C T D)(Im +DT D)1 (B T P +DT C) = 0 (5.109)

   
y C D Im 0
Let y = ,C = ,D= ,J = . We obtain the
u 0 Im 0 2 Im
Bounded Real Lemma, and (5.105) becomes
T

A P + P A + C T C = K T RK



C T D + P B = K T R (5.110)





R = DT D 2 Im
If is such that R < 0 and P 0, one can eliminate R and K from the
above and obtain the Bounded Real Lemma Riccati equality

AT P +P A+C T C+(P B+C T D)( 2 IDT D)1 (B T P +DT C) = 0 (5.111)

   
y C D 0 Im
Let y = , C = , D = , J = . We obtain the
u 0 Im Im 0
Positive Real Lemma, and (5.105) becomes the set of equations of the
KYP Lemma, i.e.
T

A P + P A = K T RK



CT P B = KT R (5.112)





R = (D + DT )
One has R 0 and it is required that P 0. If the matrix D + DT is
invertible, then one can eliminate both R andf K to obtain the Positive
Real Lemma Riccati equation

AT P + P A + C T C + (P B C)(DT D)1 (B T P C) = 0 (5.113)

302 5 Stability of Dissipative Systems
    
y C D11 D12 Im 0 w
Let y = ,C = ,D= ,J = ,u = ,
u 0 0 Im 0 2 Im u
B = [B1 B2 ]. With such a choice we obtain the H full information
problem. In this problem P 0. If D12 = 0 then (5.105) becomes:


K1

AT
P + P A + C T
C = [K T
1 K T
2 ]R

K2



  T  T 


K1 B1 D11
R = P + C
K2 B2T 0 (5.114)








T
D11 D11 0



R =

0 2 Im
A system that is dissipative with respect to this choice of the supply rate
is called Jdissipative. For more details on the Jdissipative approach
and its application in H control, one is referred to [397].

5.9.3 Closed-loop Synthesis: PR Dynamic Feedback

The problem that is of interest here, and which is in a sense of the same
nature as the problem treated in Section 3.8.5, is about the design of a robust
controller that is also PR. Let us consider the dynamical system


x(t) = Ax(t) + B1 w(t) + B2 u(t)



z(t) = C1 x(t) + D12 u(t) (5.115)





y(t) = C2 x(t) + D21 w(t)
The signal u() is the controller, w() is a disturbance. Let us denote
Hij (s) = Ci (sIn A)1 Bj + Dij , s C. Since D11 = 0 and D22 = 0,
the transfer matrices H11 (s) and H22 (s) are strictly proper. In a compact
notation one has
 
z(s) w(s)
= H(s) (5.116)
y(s) u(s)
 
C1 0 D12
with H(s) = (sIn A)1 (B1 B2 ) + . The objective of
C2 D21 0
the control task is to construct a positive real controller with transfer matrix
K(s) such that

||Tzw (s)|| = ||H11 (s) + H12 (s)K(s)(Im H22 (s)K(s))1 H21 (s)|| <
(5.117)
for some > 0. Some assumptions are in order:
 5.9 Nonlinear H Control 303

Assumption 12 (i) D11 = D22 = 0.

(ii) The pair (A, B1 ) is stabilizable, the pair (A, C1 ) is detectable.
(iii) The pair (A, B2 ) is stabilizable, the pair (A, C2 ) is detectable.
(iv) D12
T
(C1 D12 ) =
(0 R) with R invertible.
B1 0
(v) T
D21 = with N invertible.
D21 N
Assumptions (ii) and (iii) will guarantee that some Riccati equations in
(5.120) and (5.121) possess a solution, respectively. Assumptions (iv) and (v)
concern the exogeneous signal w() and how it enters the transfer H(s): w()
includes both plant disturbances and sensor noise, which are orthogonal, and
the sensor noise weighting matrix is nonsingular. Assumption (iv) means that
C1 x and D12 u are orthogonal so that the penalty on z = C1 x + D12 u includes
a nonsingular penalty on the control u.
Let us disregard for the moment that the controller be PR. We obtain the
so-called central controller

K(s) = Fc (sIn Ac )1 Zc Lc (5.118)

where the various vectors and matrices satisfy

(i) Ac = A + 2 B1 B1T Pc + B2 Fc + Zc Lc C2

(ii) Fc = R1 B2T Pc
(5.119)
(iii) Lc = Yc C2T N 1

(iv) Zc = (Im 2 Yc Pc )1
with Pc = PcT 0, Yc = YcT 0, (Yc Pc ) < 2 , and these matrices are
solutions of the Riccati equations

AT Pc + Pc A + Pc [ 2 B1 B1T B2 R1 B2T ]Pc + C1T C1 = 0 (5.120)

and

AT Yc + Yc A + Yc [ 2 C1 C1T C2 N 1 C2T ]Yc + B1T B1 = 0. (5.121)

The next step is to guarantee that the controller is PR. To this end an
additional assumption is made:
Assumption 13 The triple (A, B2 , C2 ) satises the assumptions of The-
orem 3.29.
The transfer matrix H22 (s) is PR, equivalently there exists P = P T > 0
and Q = QT 0 such that AT P + P A + Q = 0 and B2T P = C2 .
304 5 Stability of Dissipative Systems

Proposition 5.70. [238] Let B1 B1T = P 1 QP 1 2 P 1 C1T C1 P 1 +

C2 N 1 C2T , and N = R. Then the controller transfer matrix K(s) given
in (5.119) through (5.121) is PR if

Qr = C1T C1 (Yc1 2 Pc Yc Pc )B2 R1 B2T (Yc1 2 Pc Yc Pc )

2 AT Pc Pc A 2 Yc Pc Pc Yc

2 Pc (In 2 Yc Pc )B1 B1T (In 2 Pc Yc )Pc + (5.122)

+(ZcT Pc + Yc1 )T B2 R1 B2T (ZcT Pc + Yc1 )+

+(APc + Yc Pc )T (APc + Yc Pc ) + 6 Pc Yc Pc B1 B1T Pc Yc Pc

is positive denite.

Proof: The proof consists of showing that with the above choices of B1 and
of the matrix Qr > 0, then there exists Pr = PrT > 0 and Qc = QTc 0 such
that

ATc Pr + Pr Ac + Qc = 0 (5.123)
and

C2 Yc ZcT Pr = B2T Pc (5.124)

where in fact Qc = Qr . The fact that C2 = B2T P implies that

B2T P Yc (In 2 Pc Yc )1 Pr = B2T Pc (5.125)

A solution to this equation is given by

P Yc ZcT Pr = Pc (5.126)
Now let us consider

(i) Yc = P 1
(5.127)

(ii) Pr = ZcT Pc
We can remark that

Pr = ZcT Pc = (In 2 Pc Yc )Pc

(5.128)
= Pc 2 Pc Yc Pc = PrT
and that Yc = P 1 is a solution of equation (5.121), i.e.
 5.9 Nonlinear H Control 305

AP 1 + P 1 AT + P 1 ( 2 C1T C1 C2T N 1 C2 )P 1 + B1 B1T = 0 (5.129)

Indeed let us pre- and postmultiply (5.129) with P . This gives

P A + AT P + 2 C1T C1 C2T N 1 C2 + P B1 B1T P = 0 (5.130)

The choice made for B1 B1T reduces (5.130) to the KYP Lemma Equation
A P + P A + Q = 0. This shows that Yc = P 1 is a solution of equation
T

(5.121). Now inserting (5.119)(i), (5.122) and (5.127)(ii) into (5.123) reduces
this equation to (5.120). This proves that the above choices for Ac , Pr , Qr
guarantee that (5.123) is true with Qc = Qr . In other words we have shown
that with the choices for the matrices Ac , Pr and Qr the KYP Lemma rst
Equation (5.123) is satised as it reduces to the KYP Lemma equation AT P +
P A + Q = 0 which is supposed to be satised. The second equation is also
satied because B2T P = C2 is supposed to hold.
Let us end these two subsections by mentioning the work in [20, 21] in
which the H problem is solved with a nonsmooth quadratic optimization
problem, making use of the same tools from nonsmooth analysis that we saw
in various places of this book (subderivatives, subgradients). The Bounded
Real Lemma has been extended to a class of nonlinear time-delayed systems
in [15]; see also [378, 379] for details on the H control of delayed systems.
Other, related results, may be found in [430] using the PRness property
(see Denition 2.61). A discrete-time version of the Bounder Real Lemma is
presented in [470].

5.9.4 Nonlinear H

A nonlinear version of the Bounded Real Lemma is obtained from (4.81) (4.82)
setting Q = Im , S = 0, R = 2 Im . One obtains


S(x) = j(x)







R(x) = 2 Im j T (x)j(x) = W T (x)W (x)



1 T
2 g (x)V (x) = j (x)j(x) W (x)L(x)
T T (5.131)






V T (x)f (x) = hT (x)h(x) LT (x)L(x)






V (x) 0, V (0) = 0
which we can rewrite as the LMI
306 5 Stability of Dissipative Systems
1 T
V T (x)f (x) + hT (x)h(x) 2g
T
(x)V (x) + j T (x)j(x)
=
1 T
2 g (x)V (x) + j T (x)j(x) 2 Im + j T (x)j(x)
(5.132)

LT (x)
= [L(x) W (x)] 0
T
W (x)

From (5.132) one easily gets a Hamilton-Jacobi inequality that is a general-

ization of the above ones, using Proposition A.63.
Let us now pass to the main subject of this subsection. Given a plant of
the form


x(t) = A(x(t)) + B1 (x(t))w(t) + B2 (x(t))u(t)






z(t) = C1 (x(t)) + D12 (x(t))u(t)
(5.133)


y(t) = C2 (x(t)) + D21 (x(t))w(t)






x(0) = x0
with A(0) = 0, C1 (0) = 0, C2 (0) = 0, C2 () and D21 () are continuously dif-
ferentiable, the nonlinear H control problem is to construct a state feedback

(t) = a((t)) + b((t))y(t)
(5.134)

u(t) = c((t))
with continuously dierentiable a(), b(), c(), a(0) = 0, c(0) = 0, dim((t)) =
l, such that there exists a storage function V : IRn IRl IR+ such that

t1
V (x(t1 ), (t1 )) V (x(t0 ), (t0 )) { 2 wT (t)w(t) z T (t)z(t)}dt (5.135)
t0

for any t1 t0 , along the closed-loop trajectories. The controller u() may be
static, i.e. u = u(x). One may also formulate (5.135) as

t1
t1
2
z (t)z(t)dt
T
wT (t)w(t)dt + (x(t0 )) (5.136)
t0 t0

for some non negative function () with (0) = 0.

Theorem 5.71. Let B1 () and B2 () be bounded, all data in (5.133) have
T T
bounded rst derivatives, D12 D12 = Im , D21 D21 = Iq , D21 and D12 be con-
stant. Consider the state feedback u(x). If the closed-loop system satises
(5.136) there exists a storage function V (x) 0, V (0) = 0, such that the
Hamilton-Jacobi equality
 5.9 Nonlinear H Control 307

V (x)(A(x) B2 (x)C1 (x)) 12 V (x)[B2 (x)B T (x) 2 B1 (x)B1T (x)]V T (x)

+ 12 C1T (I D12 D12

T
)C1 (x) = 0
(5.137)
is satised, where the function V () may be continuous but not dierentiable
so that the PDE (5.137) has to be interpreted in the viscosity sense. Con-
versely, if (5.137) has a smooth solution V (x) > 0 for x = 0, V (0) = 0, then
the state feedback controller u(x) = (D12 T
C1 (x) + B2T (x)V T (x)) makes the
closed-loop system satisfy (5.136). The stability of the closed-loop system is
guaranteed provided that the system

x(t) = A(x(t)) + B2 (x(t))u(x(t)) + B1 (x(t))w(t)
(5.138)

z(t) = C1 (x(t)) + D12 (x(t))u(x(t))
is zero state detectable.
Much more material can be found in [33, 129, 288, 397] and the books
[257,442]. Extensions of the strict Bounded Real Lemma 5.65 to the nonlinear
ane in the input case, where storage functions are allowed to be lower semi
continuous only, has been proposed in [233].

5.9.5 More on Finite-power-gain Systems

We have already introduced the notion of nite power gain in Denition 5.7.
Here we rene it a little bit, which gives rise to the characterization of a
new quantity (the power bias) with a partial dierential equality involving a
storage function. The material is taken from [129]. In particular an example
will show that storage functions are not always dierentiable and that tools
based on viscosity solutions may be needed. We consider systems of the form

x(t) = f (x(t)) + g(x(t))u(t)
(5.139)

y(t) = h(x(t)) + j(x(t))u(t)
with the usual dimensions of vectors, and all vector elds are continuously
dierentiable on IRn . It is further assumed that ||g(x)|| < +, ||j(x)|| <
+, and that f g h j
x (x), x (x), x (x), x (x) are (globally) bounded.

Denition 5.72. The system (5.139) has nite power gain if there exists
nite non-negative functions : IRn IR (the power bias) and : IRn IR
(the energy bias) such that

t
t
y T (s)y(s)ds 2 uT (s)u(s)ds + (x)t + (x) (5.140)
0 0
for all admissible u() (here u L2,e ), all t 0 and all x IRn .
308 5 Stability of Dissipative Systems

The presence of the term (x)t may be explained as follows: dening the
norm
;

t 8
1
||y||f p = lim sup T
y (s)y(s)ds (5.141)
t+ t 0
and dividing both sides of (5.140) by t and letting t + one obtains

||y||f p 2 ||u||f p + (x) (5.142)

It is noteworthy that (5.140) implies (5.142) but not the contrary. Moreover
the link between (5.140) and dissipativity is not dicult to make, whereas it
is not clear with (5.142). Since (5.142) is obtained in the limit as t +,
possibly the concept of ultimate dissipativity could be suitable. This is why
nite power gain is dened as in Denition 5.72.

Proposition 5.73. [129] Any system with nite power gain and zero
power bias has an L2 gain . Conversely, any system with L2 gain
has a nite power gain with zero power bias.

From (5.140) let us dene the quantity

t 8
(t, x) = sup (y T (s)y(s) 2 uT (s)u(s))ds | x(0) = x (5.143)
uL2,e 0

This represents the energy that can be extracted from the system on [0, t].
It is non decreasing in t and one has for all t 0 and all x IRn :

(t, x) (x)t + (x) (5.144)

Denition 5.74. The available power a (x) is the most average power that
can be extracted from the system over an innite time when initialized at x,
i.e.

8
(t, x)
a (x) = lim sup (5.145)
t+ t

Proposition 5.75. [129] Suppose that the system has nite power gain
with power bias and energy pair (, ). Then the available power is nite, with
a (x) (x) for all x IRn .

One realizes that the framework of nite power gain systems tends to
generalize that of dissipative systems.
 5.9 Nonlinear H Control 309

Example 5.76. [129] Consider the scalar system x(t) = ax(t) + bu(t), y(t) =
c(x(t)), where c() is a saturation


c if x <



c(x) = cx if |x| (5.146)





c if x >
For this system one has
22 2 2 

ab2 bac2 2 if < | bc
a|
a = (5.147)


0 if | bc
a|

The power gain
= inf { 0 | (5.140) holds} thus depends on the

power bias:
b
| a | c2 2 if [0, c2 2 )


= (5.148)

0 if [c2 2 , +)

We are now going to characterize the property of nite power gain through
a partial dierential equation, similarly to what has been developed in Section
4.6.

Theorem 5.77. [129] Let the system in (5.139) satisfy

j T (x)j(x) 2 Im < 0 (5.149)

Suppose that the system has nite power gain . Then there exists a
nite viscosity solution pair (, V ) of the partial dierential inequality

H(x, V (x)) (5.150)

where H(x, p) = maxvIRm H(x, p, v) and

H(x, p, v) = pT (f (x)+g(x)v)+(h(x)+j(x)v)T (h(x)+j(x)v) 2 v T v (5.151)

Conversely, if there is a viscosity solution pair (, V ) to the partial dif-

ferential inequality in (5.150), then the system has nite power gain .
If V () is continuously dierentiable, the worst case disturbance is given by
v
= argmaxvIRm H(x, V (x), v).

The following may be useful for calculations:

310 5 Stability of Dissipative Systems

Theorem 5.78. [129] Suppose there exists a Lipschitz continuous solution

pair (, V ) of the partial dierential equality

H(x, V (x)) = (5.152)

Then the power bias is minimal, i.e. a = and is consequently unique.

Example 5.79. Let us continue with the above example. The system is scalar,
so that the partial dierential equality (5.152) reduces to a quadratic in
V (x). One may compute that for | bca|

2 2 

bax (1 1 2 ) if |x| <
2

V (x) = 2 2 2

2
bax
2
+ 2 a|x|
x2 2 2 2 a2
log
|x|+
x 2 if |x|
2 b2 b2 + 1
(5.153)
where = | a
bc
|, and for < | bc
a |:

2 2    
2a


bax
2 b2 2 1 |x| 2 x2 + 2 arcsin |x| if |x| <




V (x) = 2 2 2  
2 a2 |x|+ x2 2

bax
2 + ba|x|
2 x2 2 b2 log

2 a2  
2b2 2 1 if |x|
(5.154)
It is expected from these expressions that the function V (x) may not be
dierentiable everywhere, so that viscosity solutions have to be considered.

Let us end with a generalized version of the small gain theorem.

Theorem 5.80. [129] Consider a feedback interconnection as in Figure 5.2.

Suppose that the subsystems H1 and H2 are both causal and with nite power
gain 1 and 2 , respectively, and power bias 1 and 2 , respectively. If
1 2 < 1 then for all inputs ||r1 ||f p < + and ||r2 ||f p < +, the closed-
loop interconnection is stable in the sense that ||u1 ||f p < +, ||u2 ||f p < +,
||y1 ||f p < +, ||y2 ||f p < +, where the norm || ||f p is dened in (5.141).

5.10 Popovs Hyperstability

The notion of hyperstable system has been introduced by Popov in 1964
[405, 409]. It grew out of the concept of absolute stability which was reviewed
in Section 3.9. Let us consider the system
 5.10 Popovs Hyperstability 311

x(t) = Ax(t) + Bu(t)
(5.155)

y(t) = Cx(t) + Du(t)
and the quadratic functional

Q S x(s)
 t t
(0, t) = xT (s)Ax(s) 0 + [xT (s) uT (s)] ds (5.156)
0 ST R u(s)

for all t 0. It is assumed that (A, B) is controllable.

Denition 5.81. The pair (5.155) (5.156) is hyperstable if for any constant
0, 0, and for every input u() such that

(0, t) 2 + sup ||x( )||, t 0 (5.157)

0 t

there exists a k IR+ such that

||x(t)|| k( + + ||x(0)||), t 0 (5.158)

Moreover if limt+ ||x(t)|| = 0 the pair (5.155) and (5.156) is asymp-
totically hyperstable.

Denition 5.82. The pair (5.155) and (5.156) has the minimal stability
property if for any initial condition x(0) there exists a control input um ()
such that the trajectory of (5.155) satises
limt+ ||x(t)|| = 0
(0, t) 0, for all t 0

The following theorem is taken from [145], and generalizes the results
in [13, 223, 276, 277].

Theorem 5.83. [145] Suppose that the pair (5.155) and (5.156) has the
minimal stability property. Then the pair (5.155) and (5.156) is
Hyperstable if and only if the spectral function

 (sIn A)1 B
Q S
(s) = [B T (sIn AT )1 Im ] (5.159)
ST R
Im

is nonnegative
Asymptotically hyperstable if this spectral function is nonnegative and
(j) > 0 for all IR
312 5 Stability of Dissipative Systems

It is worth recalling here Proposition 2.31, Theorem 3.46, as well as the

equivalence at the end of Section 3.8.6 between the spectral function positivity
and the KYP Lemma set of equations solvability.

Proof: Let us prove the rst item of the Theorem.

Hyperstability implies positivity:

Let us consider the Hermitian matrix

Q S (sIn A)1 B
(s) = [B T (sIn AT )1 Im ]
S T R + (s + s)A Im
(5.160)
and let us prove that 0 for all Re[s] > 0 is implied by the hyperstability.
Indeed suppose that for some s0 with Re[s0 ] > 0, (s0 ) < 0. Then there
exists a nonzero vector u0 Cm such that u
0 (s0 )u0 0. For the input
u(t) = u0 exp(s0 t) with the initial data x(0) = (s0 In A)1 Bu0 , one has
x(t) = (s0 In A)1 Bu0 exp(s0 t). Clearly ||x(t)|| is increasing with the same
rate as exp(Re[s0 ]t), and it cannot satisfy an inequality as (5.158). On the
other hand the constraint (5.157) is satised since for all t 0 one has (0, t) =
t
u
0 (s0 )u0 0 exp(rRe[s0 ] )d 0. Consequently (s) is Hermitian positive
for all s with Re[s] > 0. By continuity one concludes that (j) = (j) 0
for all IR.

Positivity implies hyperstability:

Take any symmetric matrix G and notice that the functional in (5.156) can
be rewritten as

t Q AT G GA S GB 
(0, t) = [xT (A + G)x]t0 + [xT uT ] x d
0 u
ST BT G R
(5.161)
If one considers the matrix G = Pr that is the maximal solution of the
KYP Lemma set of equations (see e.g. the arguments after Proposition 4.48),
then

t
T t
(0, t) = [x (A + Pr )x]0 + ||
x( ) +
u( )||d (5.162)
0

for some
and
. Let um () be an input which renders (0, t) 0, in-

troduced via the minimal stability assumption. If xm () is the corresponding
state trajectory with initial condition xm (0) = x0 , then xT0 (A + Pr )x0
xT (t)(A + Pr )x(t) for all t 0, which implies, since limt+ x(t) = 0
 5.10 Popovs Hyperstability 313

for u() = um ()), that xT0 (A + Pr )x0 0 for all x0 . Thus the matrix
A + Pr is semi positive denite. Suppose that there exists x0 such that
xT0 (A + Pr )x0 = 0. The condition that (0, t) 0 for the input um () im-
plies that
x( ) +
um ( ) = 0. In other words the state trajectory xm () of
the dynamical system

x(t) = Ax(t) + Bu(t)
(5.163)

y(t) =
x(t) +
um (t)
with initial state xm (0) = x0 and the input um (), results in an identically
zero output y(). The inverse system of the system in (5.163), which is given
by

x(t) = (A B(
)1
)x(t) + B(
)1 y(t)
(5.164)

u(t) = (
)1
x(t) + (
)1 y(t)
has an unstable transfer function. It is deduced that one has lim ||xm (t)|| = 0
t+
when applying an identically zero input y() to (5.164). The assumption is
contradicted. Thus A + Pr is positive denite. There exists two scalars > 0
and > 0 such that

0 < 2 ||x||2 xT (A + Pr )x 2 ||x||2 (5.165)

If the input u() satises the constraint (5.157), one has

2 ||x(t)||2 sup ||x( )|| + 2 ||x(0)||2 + 2 (5.166)

0 t

and

2 ||x(t)||2 sup ||x( )|| + (||x(0)|| + )2 (5.167)

0 t

from which it follows that


+ + ||x(0)|| 1
||x(t)|| sup , [ + + ||x(0)||] (5.168)

and the proof is done.

Further work on hyperstability may be found in [97,354,446,449451]. The
name hyperstability is used in a dierent context than Popovs one in other
elds of science; see e.g. [262].
6
Dissipative Physical Systems

In this chapter we shall present a class of dissipative systems which corre-

spond to models of physical systems and hence embed in their structure the
conservation of energy (rst principle of thermodynamics) and the interaction
with their environment through pairs of conjugated variables with respect to
the power. First, we shall recall three dierent denitions of systems obtained
by an energy based modeling: controlled Lagrangian, input-output Hamilto-
nian systems and port controlled Hamiltonian systems. We shall illustrate
and compare these denitions on some very simple examples. Second we shall
treat a class of systems which gave rise to numerous stabilizing control us-
ing passivity theory and corresponds to models of robotic manipulators. In
each worked case we show how the main functions associated to a dissipative
system (the available storage, the required supply, storage functions) can be
computed analytically and related to the energy of the physical system.

6.1 Lagrangian Control Systems

Lagrangian systems arise from variational calculus and gave a rst general
analytical denition of physical dynamical systems in analytical mechanics
[1, 271, 294]. They also allow to describe the dynamics of various engineering
systems as electromechanical systems or electrical circuits. They also gave rise
to intensive work in control in order to derive dierent control laws by taking
into account the structure of the systems dynamics derived from energy based
modeling [437,439]. In this section we shall present the denition of controlled
Lagrangian systems and particular attention will be given to the expression
of the interaction of a system with its environment.
316 6 Dissipative Physical Systems

6.1.1 Denition and Properties

In this section we shall briey recall the denition of Lagrangian systems

with external forces on IRn and the denition of Lagrangian control systems
derived from it.
Denition 6.1 (Lagrangian systems with external forces). Consider a
conguration manifold Q = IRn whose points are denoted by q IRn and are
called generalized coordinates. Denote by T Q = IR2n its tangent space and its
elements by (q, q) IR2n where q is called generalized velocity. A Lagrangian
system with external forces on the conguration space Q = IRn is dened by
a real function L(q, q), from the tangent space T Q to IR called Lagrangian
function and the Lagrangian equations:

d L L
(q, q) (q, q) = F (6.1)
dt q q

where F (t) IRn is the vector of generalized forces acting on the system and
F (x)
x denotes the gradient of the function F (x) with respect to x.

Remark 6.2. In this denition the conguration space is the real vector space
IRn to which we shall restrict ourselves hereafter, but in general one may
consider a dierentiable manifold as conguration space [294]. Considering
real vector spaces as conguration spaces corresponds actually to considering
a local denition of a Lagrangian system.

If the vector of external forces F () is the vector of control inputs, then the
Lagrangian control system is fully actuated. Such models arise for instance
for fully actuated kinematic chains [366] .

Example 6.3 (Harmonic oscillator with external force). Let us consider the
very simple example of the linear mass-spring system consisting in a mass
attached to a xed frame through a spring and subject to a force F . The
coordinate q of the system is the position of the mass with respect to the
xed frame and the Lagrangian function is given by L(q, q) = K(q) U (q)
where K(q) = 12 mq 2 is the kinetic co-energy of the mass and U (q) = 12 kq 2 is
the potential energy of the spring. Then the Lagrangian system with external
force is
mq(t) + kq(t) = F (t) (6.2)

Lagrangian systems with external forces satisfy, by construction, a power

balance equation that leads to some passivity property.

Lemma 6.4 (Lossless Lagrangian systems with external forces). A

Lagrangian system with external forces (6.1) satises the following power bal-
ance equation:
 6.1 Lagrangian Control Systems 317

dH
F T q = (6.3)
dt
where the real function H() is obtained by the Legendre transformation of
the Lagrangian function L(q, q) with respect to the generalized velocity q and
is dened by
H(q, p) = q T p L(q, q) (6.4)
where p is the vector of generalized momenta:

L
p(q, q) = (q, q) (6.5)
q

and the Lagrangian function is assumed to be hyperregular [294] in such a

way that the map from the generalized velocities q to the generalized momenta
p is bijective. If moreover the function H() is bounded from below, then the
Lagrangian system with external forces is lossless with respect to the supply
rate: F T q with storage function H().

Proof: let us rst compute the power balance equation by computing F T q

using the Lagrangian equation (6.1) and the denition of the generalized mo-
mentum (6.5). We get
   
q (q, q) q (q, q)
d L L
q T F = q T dt

d
= q T dt p q T L
q

 T 
q p q T p + q T L (6.6)
q
d d
= dt dt L(q, q)
 T 
= d
dt q p L(q, q)

dH
= dt

Then, using as outputs the generalized velocities and assuming that the func-
tion H() is bounded from below, the Lagrangian system with external forces
is passive and lossless with storage function H().

Remark 6.5. The name power balance equation for (6.3) comes from the fact
that for physical systems, the supply rate is the power ingoing the system due
to the external force F and that the function H() is equal to the total energy
of the system.

Example 6.6. Consider again Example 6.3 of the harmonic oscillator. In this
case the supply rate is the mechanical power ingoing the system and the
storage function is H(p, q) = K(p) + U (q) and is the total energy of the
system, i.e. the sum of the elastic potential and kinetic energy.
318 6 Dissipative Physical Systems

Actually the denition of Lagrangian systems with external forces may be

too restrictive, as, for instance, the external forces F may not correspond to
actual inputs. For example they may be linear functions of the inputs u:

F = J T (q)u (6.7)

where J(q) is a p n matrix depending on the generalized coordinates q. This

is the case when for instance the dynamics of a robot is described in general-
ized coordinates for which the generalized velocities are not colocated to the
actuators forces and torques. Then the matrix J(q) is the Jacobian of the
geometric relations between the actuators displacement and the generalized
coordinates [366]. This system remains lossless with storage function H(q, p)
dened in (6.4) by choosing the outputs: y = J(q)q.
In order to cope with such situations, a more general denition of La-
grangian systems with external controls is given and consists in considering
that the input is directly modifying the Lagrangian function [437, 439].
Denition 6.7 (Lagrangian control system). Consider a conguration
space Q = IRn and its tangent space T Q = IR2n , an input vector space U =
IRp . A Lagrangian control systems is dened by a real function L(q, q, u) from
T Q U to IR, and the equations

d L L
(q, q, u) (q, q, u) = 0 (6.8)
dt q q

This denition includes the Lagrangian systems with external forces (6.1)
by choosing the Lagrangian function to be

L1 (q, q, F ) = L(q, q) + q T F (6.9)

It includes as well the case when the the external forces are given by (6.7)
as a linear function of the inputs where the matrix J(q) is the Jacobian of
some geometric function C(q) from IRn to IRp :
C
J(q) = (q) (6.10)
q
Then the Lagrangian function is given by

L1 (q, q, F ) = L(q, q) + C(q)T u (6.11)

However it also encompasses Lagrangian systems where the inputs do not

appear as forces as may be seen on the following example.
Example 6.8. Consider the harmonic oscillator, but assume now that the
spring is no longer attached to a xed basis but to a moving basis with its
position u considered as an input. Let us choose as coordinate q, the position
 6.1 Lagrangian Control Systems 319

of the mass with respect to the xed frame. The displacement of the spring
2
then becomes q u and the potential energy becomes: U (q, u) = 12 k(q u)
and the Lagrangian becomes
1 2 1
L(q, q, u) = mq k(q u)2 (6.12)
2 2
The Lagrangian control systems then becomes

mq(t) + kq(t) = ku(t) (6.13)

Lagrangian control systems also allow one to consider more inputs that
the number of generalized velocities as may be seen on the next example.
Example 6.9. Consider again the harmonic oscillator and assume that the ba-
sis of the spring is moving with controlled position u1 and that there is a
force u2 exerted on the mass. Consider a gain as generalized coordinate, the
position q IR of the mass with respect to an inertial frame. Then considering
the Lagrangian function
1 1
L(q, q, u) = m(q)2 k(q u1 )2 + qu2 (6.14)
2 2
one obtains the Lagrangian control system

mq(t) + k(q(t) u1 (t)) u2 (t) = 0 (6.15)

This system has two inputs and one generalized coordinate.

Lagrangian control systems were derived rst to treat mechanical control
systems, as robots for instance, but they may also be derived for other types
of systems like electrical circuits or electromechanical systems [243]. However
for such systems the denition of the conguration space is no more based on
some geometric conguration like for mechanical systems. The choice of the
conguration variables is based on the denition of some potential functions
associated with the dierent energies involved in the physical system. In par-
ticular for electrical circuits, the denition of Lagrangian systems describing
their dynamical behaviour has led to numerous dierent denitions [47, 103].
Furthermore the Lagrangian formulation is in competition with two other for-
mulations: the Brayton-Moser formulation and the Hamiltonian formulation
which will be treated in the next section. Therefore we shall not present the
dierent formulations of the dynamics of LC-circuits, but only present one of
them as an example.

Example 6.10 (An LC circuit of order 3). Consider the LC circuit depicted in
Figure 6.1.
We shall follow the procedure proposed by Chua and McPherson [103],
in order to establish a Lagrangian formulation of its dynamical behaviour.
The rst step consists in dening the space of generalized velocities. One
320 6 Dissipative Physical Systems

Fig. 6.1. LC circuit

considers a maximal tree in the circuit graph (called spanning tree) that is a
maximal set of edges without loops, and that furthermore contains a maximal
number of capacitors. The generalized velocities are then dened as the vector
of voltages of the capacitors in the tree and currents in the inductors in the
co-tree. Denoting the edges by the element which they connect, the circuit
may be partitioned into the spanning tree: = 1 2 = {C} {Su } and
its cotree: = 1 2 = {L1 } {L2 }. Hence one may choose as vector of
generalized velocities the voltages of the capacitors in the tree 1 and the
currents of the inductors in the cotree 2 :

vC
q = (6.16)
iL2
where vC denotes the voltage at the port of the capacitor and iL2 denotes the
current in the inductor labeled L2 . The vector of generalized coordinates is
hence obtained by integration of the vector of generalized velocities:

C
q= (6.17)
QL 2
Note that this denition of the variables is somewhat unnatural as it
amounts to associate ux-type variables with capacitors and charge-like vari-
ables with inductors (see the discussions in [344, 482]). The second step
consists in the denition of the Lagrangian function which describes both
 6.1 Lagrangian Control Systems 321

the electro-magnetic energy of the circuit and the Kirchhos laws. The La-
grangian function is constructed as the sum of four terms:

L(q, q, u) = E (q) E (q) + C (q, q) + I (q, u) (6.18)

The function E (q) is the sum of the electric coenergy of the capacitors in
the tree 1 and the magnetic coenergy of the inductors in the cotree 2 which
is, in this example, in the case of linear elements:
1 1 1 1
E (q) = CvC 2 + L2 iL2 2 = C q12 + L2 q22 (6.19)
2 2 2 2
The function E (q) is the sum of the magnetic energy of the inductors in
the cotree 1 and the electric energy of the capacitors in the tree 2 which is
1 1 2
E (q) = L1 2 = (q1 + q10 ) (6.20)
2L1 2L1
where the relation between the ux L1 of the inductor L1 was obtained by
integrating the Kirchhos mesh law on the mesh consisting of the capacitor
C and the inductor L1 yielding L1 = (q1 + q10 ) and q10 denotes some real
constant which may be chosen to be null. The function C (q, q) accounts for
the coupling between the capacitors in the tree 1 and inductors in the cotree
2 depending on the topological interconnection between them and is

C (q, q) = iL2 C = q2 q1 (6.21)

The function I (q, u) is an interaction potential function describing the

action of the source element and is:

I (q, u) = qL2 u = q2 u (6.22)

The Lagrangian control system is then :

1
C q1 (t) q2 (t) + (q1 (t) + q10 ) = 0 (6.23)
L1
L2 q2 (t) + q1 (t) u(t) = 0 (6.24)

Note that this system is of order 4 (it has 2 generalized coordinates) which
does not correspond to the order of the electrical circuit which, by topological
inspection, would be 3; indeed one may choose a maximal tree containing
the capacitor and having a cotree containing the 2 inductors. We shall come
back to this remark and expand it in the sequel when we shall treat the same
example as a port controlled Hamiltonian system.

This example illustrates that, although the derivation of Lagrangian sys-

tem is based on the determination of some energy functions and other physical
properties of the system, its structure may not agree with the physical insight.
Indeed the Lagrangian control systems are dened on the state space T Q, the
322 6 Dissipative Physical Systems

tangent space to the conguration space. This state space has a very special
structure; it is endowed with a symplectic form which is used to give an intrin-
sic denition of Lagrangian systems [294]. A very simple property of this state
space is that its dimension is even (there are as many generalized coordinates
as generalized velocities). Already this property may be in contradiction with
the physical structure of the system.
Lagrangian control systems, in the same way as the Lagrangian systems
with external forces, satisfy, by construction, a power balance equation and
losslessness passivity property [68].
Lemma 6.11 (Lossless Lagrangian control systems). A Lagrangian con-
trol system, (Denition 6.7), satises the following power balance equation
dE
uT z = (6.25)
dt
where
n
2 H H  2 H H
n
zi = + (6.26)
i=1
qj ui pj i=1
pj ui qj
and the real function E is obtained by the Legendre transformation of the
Lagrangian function L(q, q) with respect to the generalized velocity q and the
inputs and is dened by:
H
E(q, p, u) = H(q, p, u) uT (6.27)
u
with
H(q, p, u) = q T p L(q, q, u) (6.28)
where p is the vector of generalized momenta

L
p(q, q, u) = (q, q) (6.29)
q
and the Lagrangian function is assumed to be hyperregular [294] in such a
way that the map from the generalized velocities q to the generalized momenta
p is bijective for any u.
If moreover the Hamiltonian (6.28) is ane in the inputs (hence the func-
tion E is independent of the inputs), the controlled Lagrangian system will be
called ane Lagrangian control system. And assuming that E(q, p) is bounded
from below, then the Lagrangian system with external forces is lossless with
respect to the supply rate uT z with storage function E(q, p).
As we have seen above, the ane Lagrangian control systems are lossless
with respect to the storage function E(q, p) which in physical systems may be
chosen to be equal to the internal energy of the system. However, in numerous
systems, dissipation has to be included. For instance for robotic manipulator,
the dissipation will be due to the friction at the joints and in the actuators.
This may be done by modifying the denition of Lagrangian control systems
and including dissipating forces as follows:
 6.1 Lagrangian Control Systems 323

Denition 6.12 (Lagrangian control system with dissipation). Con-

sider a conguration space Q = IRn and its tangent space T Q = IR2n , an
input vector space U = IRp . A Lagrangian control systems with dissipation
is dened by a Lagrangian function L(q, q, u) from T Q U to IR, a function
R(q) from T Q to IR, called Rayleigh dissipation function and which satises
R
q T (q) 0 (6.30)
q
and the equations

d L L R
(q, q, u) (q, q, u) + =u (6.31)
dt q q q

Example 6.13. Consider the example of the vertical motion of a magnetically

levitated ball. There are three types of energy involved: the magnetic energy,
the kinetic energy of the ball and its potential energy. The vector of generalized
coordinates may be chosen as a vector in IR2 where q1 denotes a primitive of
the current in the inductor (according to the procedure described in Example
6.10); q2 = z is the altitude of the sphere. The Lagrangian function may then
be chosen as the sum of three terms:

L(q, q, u) = Em (q, q) + Ek (q) U (q) + I (q, u) (6.32)

The function Em (q, q) is the magnetic coenergy of the inductor and de-
pends on the currents in the coil as well on the altitude of the sphere:
1
Em (q, q) = L (q2 ) q12 (6.33)
2
where
k
L (q2 ) = L0 + (6.34)
q2 z0
The function Ek (q) is the kinetic coenergy of the ball
1 2
Ek (q) = mq (6.35)
2 2
The function U (q) denotes the potential energy due to the gravity

U (q) = gq2 (6.36)

The interaction potential is

I (q, u) = q1 u (6.37)

In order to take into account the dissipation represented by the resistor

R, one also dene the following Rayleigh potential function:
324 6 Dissipative Physical Systems

1 2
R (q) = Rq (6.38)
2 1
This leads to the following Lagrangian control system with dissipation:
L
L (q2 (t)) q1 (t) + (q2 (t)) q2 (t)q1 (t) + Rq1 (t) u(t) = 0 (6.39)
q2
1 L
mq2 (t) (q2 (t)) q12 (t) + g = 0 (6.40)
2 q2

6.1.2 Simple Mechanical Systems

An important subclass of Lagrangian contol systems is given by the so-called

simple mechanical systems where the Lagrangian function takes a particular
form.

Denition 6.14 (Simple mechanical systems with external forces).

The Lagrangian system for a simple mechanical system is a Lagrangian system
with external forces according to Denition 6.1 with Lagrangian function:

L(q, q) = T (q, q) U (q) (6.41)

where U (q) is a real function from the conguration space Q on IR and is

called potential energy and T (q, q) is a real function from T Q on IR, called
kinetic energy and is dened by
1 T
T (q, q) = q M (q)q (6.42)
2
where the matrix M (q) IRnn is symmetric positive denite and is called
the inertia matrix.

Considering the special form of the Lagrangian function, the Lagrangian

equations (6.1) may be written in some special form which is particularly
useful for deriving stabilizing controllers as will be presented in the subsequent
chapters.

Lemma 6.15 (Lagrangian equations for simple mechanical systems).

The Lagrangian equations (6.1) for a simple mechanical system may be written

M (q(t))q(t) + C(q(t), q(t))q(t) + g(q(t)) = F (t) (6.43)

where g(q) = dU
dq (q) IRn ,


n
C(q, q) = ijk qk (6.44)
k=1
 6.1 Lagrangian Control Systems 325

and ijk are called the Christoels symbols associated with the inertia matrix
M (q) and are dened by

1 Mij Mik Mkj
ijk = + (6.45)
2 qk qj qi

A property of Christoels symbols which is easily derived but is of great

importance for the derivation of stabilizing control laws, is given below. What
is denoted as M (q) is the time derivative of the time function M (q(t)).
Lemma 6.16. The Christoels symbols (6.45) satisfy the following property:
the matrix M (q) 2C(q, q) is skew-symmetric. Equivalently M (q) = C(q, q) +
C T (q, q).
The notation M (q) is, without any dependence
 on
 time in M (q), a lit-
Mij
tle meaningless. Its meaning is that M (q) = q q , so that M (q(t)) =
i,j
d
dt M (q(t)). Let us note that an immediate consequence is that M (q) =
C(q, q) + C T (q, q).
Remark 6.17. A consequence of the Lemma is that:
 
q T M (q) 2C(q, q) q = 0 (6.46)
 
and hence reects that the generalized inertial forces M (q) 2C(q, q) q do
not work. This may be seen as follows:
T q = dH
dt (q, p) = q T M (q)q + 12 q T M (q)q + g(q)

= q T [C(q, q)q g(q) + ] + 12 q T M (q)q + g(q) (6.47)

 
= q T + 12 q T M (q) 2C(q, q) q

from which (6.46) follows. Such forces are sometimes called gyroscopic [351].
It is noteworthy that (6.46) does not mean that the matrix M (q) 2C(q, q)
is skew-symmetric. Skew-symmetry is true only for the particular denition
of the matrix C(q, q) using Christoels symbols.
Remark 6.18. The denition of a positive denite symmetric inertia matrix for
simple mechanical systems, may be expressed in some coordinate independent
way by using so-called Riemannian manifolds [1]. In [442, Chapter 4] the
properties of the Christoells symbols, that shall be used in the sequel for
the synthesis of stabilizing controllers, may also be related to properties of
Riemannian manifolds.
A class of systems which typically may be represented in this formula-
tion is the dynamics of multibody systems, for which systematic derivation
procedures were obtained (see [366] and the references herein).
326 6 Dissipative Physical Systems

6.2 Hamiltonian Control Systems

There is an alternative to the Lagrangian formulation of the dynamics of

physical controlled systems that is the Hamiltonian formalism. This formal-
ism has been derived from the Lagrangian one in the end of the nineteenth
century and has now become the fundamental structure of the mathematical
description of physical systems [1, 294]. In particular it allowed one to deal
with symmetry and reduction and also to describe the extension of classical
mechanics to quantum mechanics.

6.2.1 Input-output Hamiltonian Systems

Lagrangian systems may be transformed to standard Hamiltonian systems by

using the Legendre transformation [1, 294].

Lemma 6.19 (Legendre transformation of a Lagrangian system).

Consider a Lagrangian system with external forces and dene the vector of
generalized momenta:

L
p(q, q) = (q, q) IRn (6.48)
q

Assume that the map from generalized velocities to generalized momenta

is invertible, and consider the Legendre transformation with respect to q of
the Lagrangian function, called Hamiltonian function:

H0 (q, p) = q T p L(q, q) (6.49)

Then the Lagrangian system with external forces is equivalent to the following
standard Hamiltonian system:
H0
q(t) = p (q(t), p(t))
(6.50)
p(t) = H 0
q (q(t), p(t)) + F (t)

There is an alternative way of writing these equations as follows (we drop

the arguments):
H0
q q 0n
= Js + F (6.51)
H0
p p
In

where Js is the following matrix, called symplectic matrix:


0n In
Js = (6.52)
In 0n
 6.2 Hamiltonian Control Systems 327

This symplectic matrix is the local representation, in canonical coordi-

nates, of the symplectic Poisson tensor eld which denes the geometric struc-
ture of the state space of standard Hamiltonian systems (the interested reader
may nd an precise exposition to symplectic geometry in [294].)
In the same way as a Lagrangian system with external forces may be ex-
pressed as a control Lagrangian system (for which the inputs are an argument
of the Lagrangian function), the standard Hamiltonian system with external
forces (6.51) may be expressed as Hamiltonian system where the Hamiltonian
function depends on the inputs H(q, p, u) = H0 (q, p) q T F which yields
H0 H0
q q In q 0n
= Js + F = Js + F (6.53)
H0 H0
p p
0 n p
In

As the simplest example let us consider the harmonic oscillator with an

external force.
Example 6.20 (Harmonic oscillator with external force). First let us recall
that in its Lagrangian representation (see Example 6.3), the state space is
given by the position of the mass (with respect to the xed frame) and
its velocity. Its Lagrangian is L(q, q, F ) = 12 mq 2 12 kq 2 + q T F . Hence the
(generalized) momentum is p = L q = mq. The Hamiltonian function, ob-
tained through the Legendre transformation is H(q, p, F ) = H0 (q, p) q T F
where the Hamiltonian function H0 (q, p) represents the total internal energy
2
H0 (q, p) = K(p) + U (q), the sum of the kinetic energy K(p) = 12 pm and the
potential energy U (q). The Hamiltonian system becomes
   
q(t) 0 1 kq(t) 0
= + F (t) (6.54)
p(t) 1 0 p(t)
m
1

Hamiltonian systems with external forces may be generalized to so-called

input-output Hamiltonian systems [68] for which the Hamiltonian function
depends on the inputs. In the sequel we shall restrict ourselves to systems for
which the Hamiltonian function depends linearly on the inputs, which actually
constitute the basis of the major part of the work dedicated to the system
theoretic analysis and the control of Hamiltonian systems [68, 381, 439].
Denition 6.21 (Input-output Hamiltonian systems). An input-output
Hamiltonian system on IR2n is dened by a Hamiltonian function

m
H(x) = H0 (x) Hi (x) ui (6.55)
i=1

composed of the sum of the internal Hamiltonian H0 (x) and a linear combina-
tion of m interaction Hamiltonian functions Hi (x) and the dynamic equations
328 6 Dissipative Physical Systems

m
x = Js dH0 (x) + i=1 Js dHi (x)ui
(6.56)
yi = Hi (x), i = 1, .., m

denoting the state by xT = (q T , pT ) IR2n and the gradient of a function H

2n
by dH = dHdx IR .

One may note that an input-output Hamiltonian system (6.56) is a non-

linear system ane in the inputs in the sense of [227, 381]. It is composed
of a Hamiltonian drift vector eld Js dH0 (q, p) and the input vector elds
Js dHi (q, p) are also Hamiltonian and generated by the interaction Hamilto-
nian functions.
The outputs are the Hamiltonian interaction functions and are called natu-
ral outputs [68]. We may already note here that these outputs, although called
natural, are not the outputs conjugated to the inputs for which the system
is passive as will be shown in the sequel.

Example 6.22. Consider again Example 6.9. The state space is given by the
displacement of the spring and its velocity. Its Lagrangian is
1 2 1
L(q, q, F ) = m(q + u1 ) kq 2 + qu2 (6.57)
2 2
Hence the generalized momentum is: p = L q = m(q + u1 ) The Hamiltonian
function, obtained through the Legendre transformation with respect to q is

H(q, p, u1 , u2 ) = q T p L(q, q, u1 , u2 ) = H0 (q, p) pu1 qu2 (6.58)

2
where the Hamiltonian function H0 (q, p) = 12 pm + 12 kq 2 represents, as in the
preceding example, the sum of the kinetic and the elastic potential energy.
The interaction potentials are the momentum of the mass H1 (q, p) = p, for
the input u1 which represents the controlled velocity of the basis and the the
displacement of the spring H2 (q, p) = q for the input u2 which is the external
force exerted on the mass. The dynamics is now described by the following
input-output Hamiltonian system:
    
q(t) 0 1 kq(t) 1 0
= + u 1 (t) + u2 (t) (6.59)
p(t) 1 0 p(t)
m
0 1

Note that the denition of the generalized momentum p corresponds

to a generalized state space transformation involving the input u1 . Conse-
quently in the Hamiltonian formulation (6.59) the derivative of the input no
longer appears, contrary to the Lagrangian dynamics in (6.15). Moreover, like
ane Lagrangian control systems, input-output Hamiltonian systems satisfy
a power balance equation, however considering, instead of the natural outputs
yi (6.56), their derivatives.
 6.2 Hamiltonian Control Systems 329

Lemma 6.23 (Lossless input-output Hamiltonian systems). An input-

output Hamiltonian system (according to Denition 6.21), satises the follow-
ing power balance equation:
dH0
uT y = (6.60)
dt
If, moreover, the Hamiltonian function H0 (x) is bounded from below, then
the input-output Hamiltonian system is lossless with respect to the supply rate:
uT y with storage function H0 (q, p).

Let us comment on this power balance equation using the example of the
harmonic oscillator with moving frame and continue Example 6.22.

Example 6.24. The natural outputs are then the momentum of the system:
y1 = H1 (q, p) = p which is conjugated to the input u1 (the velocity of the
basis of the spring) and the displacement of the spring y2 = H2 (q, p) = q
which is conjugated to the input u2 (the external force exerted on the mass).
The passive outputs dening the supply rate are then

y 1 = p = kq + u2 (6.61)

and
p
y 2 = q = u1 (6.62)
m
Computing the supply rate, the terms in the inputs cancel each other and one
obtains
p
y 1 u1 + y 2 u2 = kqu1 + u2 (6.63)
m
This is precisely the sum of the mechanical power supplied to the mechanical
system by the source of displacement at the basis of the spring and the source
of force at the mass. This indeed is equal to the variation of the total energy
of the mechanical system. However it may be noticed that the natural outputs
as well as their derivatives are not the variables which one uses in order to
dene the interconnection of this system with some other mechanical system:
the force at the basis of the spring which should be used to write a force
balance equation at that point and the velocity of the mass m which should
be used in order to write the kinematic interconnection of the mass (their
dual variables are the input variables). In general, input- output Hamiltonian
systems (or their Lagrangian counterpart) are not well suited for expressing
their interconnection.

Example 6.25. Consider the LC circuit of order 3 in Example 6.10. In the

Lagrangian formulation, the generalized velocities were q1 = VC the voltage
of the capacitor, q2 = iL2 the current of the inductor L2 and the generalized
coordinates were some primitives denoted by q1 = C and q2 = QL2 . The
Lagrangian function was given by L(q, q, u) = E (q) E (q) + C (q, q) + I (q, u)
where E (q) is the sum of the electric coenergy of the capacitor and of the
330 6 Dissipative Physical Systems

inductor L2 , E (q) is the magnetic energy of the inductor L1 , C (q, q) is a

coupling function between the capacitor and the inductor L2 and I (q, u) is
the interaction potential function.
Let us now dene the generalized momenta. The rst momentum variable
is
L E C E
p1 = = + = = C q1 = QC (6.64)
q1 q1 q1 q1
and is the electrical charge of the capacitor, i.e. its energy variable. The second
momentum variable is

L E C
p2 = = + = L2 q2 + q1 = L2 + C (6.65)
q2 q2 q2

and is the sum of the the total magnetic ux of the inductor L2 (its energy
variable) and of the ctitious ux at the capacitor C . The Hamiltonian func-
tion is obtained as the Legendre transformation of L(q, q, u) with respect to
q:
H(q, p, u) = q1 p1 + q2 p2 L(q, q, u) = H0 (q, p) Hi (q)u (6.66)
where Hi = q2 and H0 is
1 2 1 2 1
H0 (q, p) = q1 + p1 + (p2 q1 )2 (6.67)
2L1 2C 2L2
Note that the function H0 (p, q) is the total electromagnetic energy of the cir-
cuit as the state variables are equal to the energy variables of the capacitors
and inductors. Indeed using Kirchhos law on the mesh containing the induc-
tor L1 and the capacitor C, up to a constant q1 = C = L1 is the magnetic
ux in the inductor, by denition of the momenta p1 = QC is the charge of
the capacitor and p2 q1 = L2 is the magnetic ux of the inductor L2 . This
input-output Hamiltonian system again has order 4 (and not the order of the
circuit). But one may note that the Hamiltonian function H0 does not depend
on q2 . Hence it has a symmetry and the drift dynamics may be reduced to
a third order system (the order of the circuit) and in a second step to a sec-
ond order system [294]. However the interaction Hamiltonian depends on the
symmetry variable q2 , so the controlled system may not be reduced to a lower
order input-output Hamiltonian system. The power balance equation (6.60)
becomes dH 0
dt = uq2 = iL2 u which is exactly the power delivered by the source
as the current iL2 is also the current owing in the voltage source.

The preceding input-output Hamiltonian systems may be extended by

considering more general structure matrices than the symplectic structure
matrix Js which appear in the reduction of Hamiltonian systems with sym-
metries [294]. Indeed one may consider so-called Poisson structure matrices
that are matrices J(x) depending on x(t) IR2n , skew-symmetric and satis-
fying the Jacobi identities:
 6.2 Hamiltonian Control Systems 331
n
 
Jik Jkj Jji
Jlj (x) + Jli (x) (x) + Jlk (x) = 0 (6.68)
xl xl xl
k,l=1

Remark 6.26. These structure matrices are the local denition of Poisson
brackets dening the geometrical structure of the state-space [1,294] of Hamil-
tonian systems dened on dierentiable manifold endowed with a Poisson
bracket. Such systems appear for instance in the Hamiltonian formulation
of a rigid body spinning around its center of mass (the Euler-Poinsot prob-
lem) [294].

Remark 6.27. Poisson structure matrices may be related to symplectic struc-

ture matrices as follows. Note rst that, by its skew-symmetry, the rank of
the structure matrix of a Poisson bracket at any point is even, say 2n (then
one says also that the Poisson bracket has the rank 2n). Suppose moreover
that the structure matrix has constant rank 2n in a neighborhood of a point
x0 M . Then the Jacobi identities (6.68) ensure the existence of canonical
coordinates (q, p, r) = (q1 , .., qn , p1 , .., pn , r1 , .., rl ) where (2n + l) = m, such
that the m m structure matrix J(q, p, r) is given as follows:

0n In 0nl
J(q, p, r) = In 0n 0nl (6.69)
0ln 0ln 0ll
One may hence see a symplectic matrix appear associated with the rst 2n
coordinates. The remaining coordinates correspond to so-called distinguished
functions or Casimir functions which dene an important class of dynamical
invariants of the Hamiltonian system [294].

With such structure matrices, the input-output Hamiltonian systems may

be generalized to Poisson control systems as follows [381].
Denition 6.28 (Poisson control systems). A Poisson control system on
IRn is dened bya Poisson structure matrix J(x), a Hamiltonian function
m
H(x) = H0 (x) i=1 Hi (x) ui composed of the sum of the internal Hamilto-
nian H0 (x) and a linear combination of m interaction Hamiltonian functions
Hi (x) and the dynamic equations:


m
x = J(x)dH0 (x) J(x)dHi (x) ui (6.70)
i=1

6.2.2 Port Controlled Hamiltonian Systems

As the examples of the LC circuit and of the levitated ball have shown,
although the input-output Hamiltonian systems represent the dynamics of
332 6 Dissipative Physical Systems

physical systems in a way that the conservation of energy is embedded in the

model, they fail to represent accurately some other of their structural prop-
erties. Therefore another type of Hamiltonian systems, called port controlled
Hamiltonian systems was introduced which allow to represent both the energy
conservation as well as some other structural properties of physical systems,
mainly related to their internal interconnection structure [342, 442].

Denition 6.29 (Port controlled Hamiltonian system). A port con-

trolled Hamiltonian system on IRn is dened by a skew-symmetric structure
matrix J(x), a real-valued Hamiltonian function H0 (x), m input vector elds
gi (x) and the dynamic equations
m
x = J(x)dH0 (x) + i=1 gi (x)Ui
(6.71)

yi = giT (x)dH0 (x)

One may note that port controlled Hamiltonian system, as the input out-
put Hamiltonian systems, are ane with respect to the inputs [227, 381].

Remark 6.30. The system-theoretic properties of port controlled Hamiltonian

systems were investigated in particular concerning the external equivalence,
but as this subject goes beyond the scope of this book, the reader is referred
to [441] [442, Chapter 4].

The systems (6.71) have been called port controlled Hamiltonian system in
allusion to the network concept of the interaction through ports [342,441,442].
In this case the Hamiltonian function corresponds to the internal energy of
the system, the structure matrix corresponds to the interconnection structure
associated with the energy ows in the system [343345] and the interaction
with the environment of the network is dened through pairs of port vari-
ables [342, 441]. Moreover, the underlying modeling formalism is a network
formalism which provides a practical frame to construct models of physical
systems and roots on a rmly established tradition in engineering [62] which
found its achievement in the bond graph formalism [63, 342, 398].
Port controlled Hamiltonian systems dier from input-output Hamilto-
nian systems in three ways which we shall illustrate below on some examples.
First, the structure matrix J(x) does not have to satisfy the Jacobi identi-
ties (6.68); such structure matrices indeed arise in the reduction of simple
mechanical systems with non-holonomic constraints [440]. Second the input
vector elds are no more necessarily Hamiltonian, that is they may not derive
from an interaction potential function. Third, the denition of the output is
changed. The most simple examples of port controlled Hamiltonian system
consist in elementary energy storing systems, corresponding for instance to a
linear spring or a capacitor.
 6.2 Hamiltonian Control Systems 333

Example 6.31 (Elementary energy storing systems). Consider the following

rst order port controlled Hamiltonian system:

x(t) = u(t)
(6.72)

y = dH 0
dx (x)

where x(t) IRn is the state variable, H0 (x) is the Hamiltonian function and
the structure matrix is equal to 0. In the scalar case, this system represents
the integrator which is obtained by choosing the Hamiltonian function to be:
H0 = 12 x2 . This system represents also a linear spring , where the state
variable x() is the displacement of the spring and the energy function is the
elastic potential energy of the spring (for instance H(x) = 12 k q 2 where k is
the stiness of the spring). In the same way (6.72) represents a capacitor with
x being the charge and H0 the electrical energy stored in the capacitor, or an
inductance where x is the total magnetic ux and H0 is the magnetic energy
stored in the inductance.
In IR3 such a system represents the point mass in the three-dimensional
Euclidean space with mass m where the state variable x(t) IR3 is the mo-
mentum vector, the input u IR3 is the vector of forces applied on the mass,
the output vector y(t) IR3 is the velocity vector and the Hamiltonian func-
1 T
tion is the kinetic energy H0 (x) = 2m x x.
It may be noted that such elementary systems may take more involved
forms when the state variable belongs to some manifold dierent from IRn , as
it is the case for instance for spatial springs which deform according to rigid
body displacements [143, 144, 308, 345].
Like ane Lagrangian control systems and input-output Hamiltonian sys-
tems, port controlled Hamiltonian systems satisfy a power balance equation
and under some assumption on the Hamiltonian function are lossless.
Lemma 6.32 (Losslessness of port controlled Hamiltonian systems).
A port controlled Hamiltonian system (according to Denition 6.29), satises
the following power balance equation:
dH0
uT y = (6.73)
dt
If moreover the Hamiltonian function H0 (x) is bounded from below, then
the port controlled Hamiltonian system is lossless with respect to the supply
rate uT y with storage function H0 (x).
Again in the case when the Hamiltonian function is the energy, the balance
equation corresponds to a power balance expressing the conservation of energy.
Let us now consider a slightly more involved example, the LC circuit of order
3 treated here above, in order to comment on the structure of port controlled
Hamiltonian sytems as well as to compare it to the structure of input output
and Poisson control systems.
334 6 Dissipative Physical Systems

Example 6.33 (LC circuit of order 3). Consider again the circuit of Example
6.10. According to the partition of the interconnection graph into the span-
ning tree: = {C} {Su } and its cotree: = {L1 } {L2 }, one may write
Kirchhos mesh law for the meshes dened by the edges in and the node
law corresponding to the edges in as follows:

iC 0 1 1 0 vC
vL1 1 0 0 0 iL1

vL2 = 1 0 0 1 iL2 (6.74)
iS 0 0 1 0 vS

Now, taking as state variables the energy variables of the capacitor (the
charge QC , the total magnetic uxes L1 and L2 in the two inductors) one
identies immediately the rst three components of the left hand side in (6.74)
T
as the time derivative of the state vector x = (QC , L1 , L2 ) . Denoting by
HC (QC ), HL1 (L1 ) and HL2 (L2 ) the electric and magnetic energies stored in
the elements, one may identify the coenergy variables as follows: vC = H QC ,
C

H H
iL1 = LL1 and iL2 = LL2 . Hence the rst three components of the vector on
1 2
the right hand side of Equation (6.74) may be interpreted as the components
of the gradient of the total electromagnetic energy of the LC circuit H0 (x) =
HC (QC ) + HL1 (L1 ) + HL2 (L2 ). Hence the dynamics of the LC circuit may
be written as the following port controlled Hamiltonian system:

x(t) = JdH0 (x(t)) + gu(t)
(6.75)

y = g T dH0 (x)

where the structure matrix J and the input vector g are part of the matrix
describing Kirchhos laws in (6.74) (i.e. part of the fundamental loop matrix
associated with the tree ):

0 1 1 0
J = 1 0 0 and g = 0 (6.76)
1 0 0 1

The input is u = vS and the output is the current with generator sign con-
vention y = iS . In this example the power balance equation (6.73) is simply
interpreted as the time derivative of the total electromagnetic energy being
the power supplied by the source. Actually this formulation is completely
general to LC circuits and it may be found in [344] as well as the compari-
son with the formulation in terms of Lagrangian or input-output Hamiltonian
systems [47, 344].
The port controlled Hamiltonian formulation of the dynamics of the LC
circuit may be compared with the input-output formulation derived in the
Example 6.25. First, one may notice that in the port controlled Hamiltonian
formulation, the information on the topology of the circuit and the information
 6.2 Hamiltonian Control Systems 335

about the elements (i.e. the energy) is represented in two dierent objects: the
structure matrix and the input vector on the one side and the Hamiltonian
function on the other side. In the input-output Hamiltonian formulation this
information is captured solely in the Hamiltonian function (with interaction
potential), in the same way as in the Lagrangian formulation in Example
6.10. Second, the port controlled Hamiltonian system is dened with respect
to a non-symplectic structure matrix and its order coincides with the order
of the circuit, whereas the input-output system is given (by denition) with
respect to a symplectic (even order) structure matrix of order larger than the
order of the circuit. Third, the denition of the state variables in the port
controlled system corresponds simply to the energy variables of the dierent
elements of the circuit whereas in the input-output Hamiltonian system, they
are dened for the total circuit and for instance the ux of capacitor L2 does
not appear as one of them. Finally, although the two structure matrices of
the port controlled and the input output Hamiltonian systems may be related
by projection of the dynamics using the symmetry in q2 of the input output
Hamiltonian system, the controlled systems remain distinct. Indeed, consider
the input vector g; it is clear that it is not in the image of the structure matrix
J. Hence there exist no interaction potential function which generates this
vector and the port controlled Hamiltonian formulation cannot be formulated
as an input output Hamiltonian system or Poisson control system.
In order to illustrate a case where the energy function denes some in-
terdomain coupling, let us consider the example of the iron ball in magnetic
levitation. This example may be seen as the one-dimensional case of general
electromechanical coupling arising in electrical motors or actuated multibody
systems.
Example 6.34. Consider again the example of the vertical motion of a mag-
netically levitated ball as treated in Example 6.13. Following a bond graph
modeling approach, one denes the state space as being the variables dening
T
the energy of the system. Here the state vector is then x = (, z, pb ) where
is the magnetic ux in the coil, z is the altitude of the sphere and pb is the
kinetic momentum of the ball. The total energy of the system is composed of
three terms: H0 (x) = Hmg (, z) + U(z) + Hkin (pb ) where Hmg (, z) denotes
the magnetic energy of the coil and is
1 1
Hmg (, z) = 2 (6.77)
2 L (z)
where L(z) is given in (6.34), U(z) = gz is the gravitational potential energy
1 2
and Hkin (pb ) = 2m p is the kinetic energy of the ball. Hence the gradient
of the energy function H0 is the vector of the coenergy variables: H 0
x =
(vL , f, vb ) where vL is the voltage at the coil:
Hmg
vL = = (6.78)
L(z)
336 6 Dissipative Physical Systems

The sum of the gravity force and the electromagnetic force is given by f =
g fmg :
1 2 L
fmg = (z) (6.79)
2 L2 (z) z
and vb = pmb is the velocity of the ball. Then from Kirchhos laws and the
kinematic and static relations in the system, it follows that the dynamics
may be expressed as a port controlled Hamiltonian system (6.71) where the
structure matrix is constant:

0 0 0
J = 0 0 1 (6.80)
0 1 0
and the input vector is constant:

1
g = 0 (6.81)
0
Note that the structure matrix is already in canonical form. In order to take
into account the dissipation represented by the resistor R, one also denes the
following dissipating force vR = RiR = RiL which may be expressed in a
Hamiltonian-like format as a Hamiltonian-system with dissipation [121].
Let us compare now the port controlled Hamiltonian formulation with the
Lagrangian or input output Hamiltonian formulation. Therefore recall rst the
input output Hamiltonian system obtained by the Legendre transformation
of the Lagrangian system of Example 6.13. The vector of the momenta is

L
p= (q, q) = (6.82)
q pb
and the Hamiltonian function obtained by Legendre transformation of the
Lagrangian function, dened in Example 6.13, is
H(q, p) = H0 (x) q1 u (6.83)
Hence the state space of the input-output representation is the state space of
the port controlled system augmented with the variable q1 (the primitive if
the current in the inductor). Hence the order of the input output Hamiltonian
system is 4 and larger than 3, the natural order of the system (a second
order mechanical system coupled with a rst order electrical circuit), which is
precisely the order of the port controlled Hamiltonian system. Moreover the
state variable in excess is q1 and is precisely the symmetry variable of the
internal Hamiltonian function H0 (x) in H(q, p). In an analogous way as in
the LC circuit example above, this symmetry variable denes the interaction
Hamiltonian, hence the controlled input-output Hamiltonian system may not
be reduced. And again one may notice that the input vector g does not belong
to the image of the structure matrix J, hence cannot be generated by any
interaction potential function.
 6.2 Hamiltonian Control Systems 337

Now we shall compare the denitions of the outputs for input-output

Hamiltonian or Poisson control systems and port controlled Hamiltonian sys-
tems. Consider the port controlled system (6.71) and assume that the input
vector elds are Hamiltonian, i.e. there exists interaction Hamiltonian func-
tions such that gi (x) = J(x)dHi (x). The port conjugated outputs are then
yi = dH0T (x)gi (x) = dH0T (x)J(x)dHi (x). The natural outputs are yi = Hi (x).
Using the drift dynamics in (6.71), their derivatives are computed as

m
y i = dHiT (x)x = yi + uj dHiT (x)J(x)dHj (x) (6.84)
j=1,j=i

Hence the passive outputs of both systems dier, in general, by some skew
symmetric terms in the inputs. This is related to the two versions of the
Kalman-Yakubovich-Popov Lemma where the output includes or not a skew
symmetric feedthrough term.

Example 6.35 (Mass-spring system with moving basis). Consider again the
mass-spring system with moving basis and its input-output model treated
in Examples 6.22 and 6.24. The input vector elds are Hamiltonian, hence
we may compare the denition of the passive outputs in the input-output
Hamiltonian formalism and in the port controlled Hamiltonian formalism.
The derivatives of the natural outputs derived in Example 6.24 are y 2 =
p
q = m y 1 u1 
u1 and + y 2 u2 = u1 (kq) + u2 m
p
. The
port
 conjugated outputs
kq kq
are y1 = (1, 0) p = kq and y2 = (0, 1) p p
= m . These outputs,
m m
contrary to the natural outputs and their derivatives, are precisely the in-
terconnection variables needed to write the kinematic and static relation for
interconnecting this mass-spring system to some other mechanical systems.

The mass-spring example shows how the dierent denitions of the pairs
of input-output variables for input-output and port controlled Hamiltonian
systems, although both dening a supply rate for the energy function as stor-
age function, are fundamentally dierent with respect to the interconnection
of the system with its environment. One may step further and investigate the
interconnection of Hamiltonian and Lagrangian systems which preserve their
structure. It was shown that the port controlled Hamiltonian systems may
be interconnected in a structure preserving way by so-called power continu-
ous interconnections [121, 346]. Therefore a generalization of port controlled
Hamiltonian systems to implicit port controlled Hamiltonian systems (en-
compassing constrained systems) was used in [121,346,441,442]. However this
topic is beyond the scope of this section and we shall only discuss the inter-
connection of Lagrangian and Hamiltonian systems on the example of the ball
in magnetic levitation.

Example 6.36 (Levitated ball as the interconnection of two subsystems). We

have seen that the dynamics of the levitated ball may be formulated as a
338 6 Dissipative Physical Systems

third order port controlled Hamiltonian system where the coupling between
the potential and kinetic energy is expressed in the structure matrix (the
symplectic coupling) and the coupling through the electromagnetic energy in
the Hamiltonian function. However it also allows one to express this system
as the coupling, through a passivity preserving interconnection, of two port
controlled Hamiltonian systems. Therefore one may conceptually split the
physical properties of the iron ball into purely electric and purely mechanical
ones. Then the electromechanical energy transduction is represented by a
second order port controlled Hamiltonian system:
Hmg
(t) 0 1 ((t), z(t)) 1 0
=

+ u(t) + u1 (t) (6.85)
z(t) 1 0 H mg 0 1
z ((t), z(t))

with output equations

Hmg


iS = (1, 0) (6.86)
Hmg
z
(6.87)

Hmg


y1 = fmg = (0, 1) (6.88)
Hmg
z

The second subsystem simply represents the dynamics of a ball in vertical

translation submitted to the action of an external force u2 :
H2
q(t) 0 1 q (q(t), p(t)) 0
= + u2 (t) (6.89)
p(t) 1 0 H2
p (q(t), p(t)) 1

where the Hamiltonian H2 is the sum of the kinetic and the potential energy
1 2
of the ball: H2 (q, p) = 2m p + gq and the conjugated output is the velocity of
the ball:
H2
q
y2 = (0, 1) (6.90)
H2
p

Consider the interconnection dened by:

u 1 = y2 (6.91)
u2 = y1 (6.92)

It is clear that this interconnection satises a power balance: u1 y1 + u2 y2 = 0.

Hence it may be proved [121, 346, 442] that the interconnection of the two
 6.2 Hamiltonian Control Systems 339

port controlled Hamiltonian systems leads to a port controlled Hamiltonian

system (actually much more general interconnection relations may be consid-
ered, involving also constraints). In this example a simple elimination of the
variables involved in the interconnection leads to the port controlled Hamil-
tonian system with Hamiltonian function Htot = Hmg + H2 and structure
matrix
0 0 0 0
0 0 0 1
Jtot = 0 0 0 1
(6.93)
0 1 1 0
Considering lines 2 and 3 of the structure matrix, one deduces that the vari-
ations of z and q satisfy
z q = 0 (6.94)
This is precisely a Casimir function, i.e. a dynamical invariant of any Hamil-
tonian system dened with respect to the structure matrix Jtot . Hence it is
possible to identify (up to an arbitrary constant) the two positions z and q,
thus to reduce this system to the three-dimensional port controlled Hamilto-
nian system presented here above. It is clear that this splitting is not possible
using the input output Hamiltonian system or Poisson control systems as the
subsystem 1 in (6.85) has a non-symplectic (null) structure matrix and the
input vector hence are not Hamiltonian (else they would be null too).

As a conclusion to this section we shall present an extension of lossless

port control Hamiltonian systems to dissipative system, called port controlled
Hamiltonian systems with dissipation introduced in [121]. The main dierence
is that the skew-symmetry of the structure matrix J is no more required, hence
the structure matrix is in general an addition of a skew-symmetric matrix and
a symetric positive matrix.

Denition 6.37 (Port Controlled Hamiltonian system with dissipa-

tion). A port controlled Hamiltonian system on IRn is dened by a skew-
symmetric structure matrix J(x), a symmetric positive matrix R(x), a real-
valued Hamiltonian function H0 (x), m input vector elds gi (x) and the dy-
namic equations
m
x = (J(x), R(x)) H 0
x (x) + i=1 gi (x)Ui
(6.95)
T H0
yi = gi (x) x (x)

Of course such a system is no more lossless, but it still satises a power

balance equation and under some assumption on the Hamiltonian system, a
passivity property.
340 6 Dissipative Physical Systems

Lemma 6.38 (Dissipativity of Port Controlled Hamiltonian systems).

A port controlled Hamiltonian system with dissipation (according to Denition
6.37) satises the following power balance equation:

dH0 H0 T H0
uT y = + (x)R(x) (x). (6.96)
dt x x
If, moreover, the Hamiltonian function H0 (x) is bounded from below, then
the port controlled Hamiltonian system with dissipation is dissipative with re-
spect to the supply rate uT y with storage function H0 (x).

As an example recall the levitated ball as the interconnection of two sub-

systems.
Example 6.39. Consider rst the magnetic part. Considering the losses in the
coil amounts to add to the skew symmetric structure matrix dened in (6.85)
the symmetric positive matrix:

R 0
R= (6.97)
0 0
Then the total system also becomes a port controlled Hamiltonian system
with a symmetric matrix Rtot = diag(R 03 ).

6.3 Rigid JointRigid Link Manipulators

In this section and in the next ones we shall recall the simple models corre-
sponding to electromechanical systems, which motivated numerous results on
passivity-based control. We shall recall and derive their passivity properties,
and we illustrate some concepts introduced in the previous sections and chap-
ters. Actually the results in the next sections of the present chapter will serve
as a basis for introducing the control problem in Chapter 7. Our aim now is
to show how one can use the passivity properties of the analyzed processes,
to construct globally stable control laws. We shall insist on the calculation of
storage functions, and it will be shown at some places (see for instance Section
7.3) that this can be quite useful to derive Lyapunov functions for closed-loop
systems.
The dynamics of the mechanism constituting the mechanical part of a
robotic manipulator is given by a simple mechanical system according to Def-
inition 6.14 and Lemma 6.15:

M (q(t))q(t) + C(q(t), q(t))q(t) + g(q(t)) = (t) (6.98)

From Lemma 6.11, it follows that they are lossless systems with respect
to the supply rate T q with storage function E(q, q) = 12 q T M (q)q + V (q) and
g(q) = V
q is the gradient of the gravitation potential energy V (q).
 6.3 Rigid JointRigid Link Manipulators 341

6.3.1 The Available Storage

We have seen that storage functions play an important role in the dissipativity
theory. In particular the dissipativity of a system can be characterized by the
available storage Va (q, q) and the required supply Vr (q, q) functions. Let us
focus now on the calculation of the available storage function (see Denition
4.35), which represents the maximum internal energy contained in the system
that can be extracted from it. More formally recall that we have

t
Va (q0 , q0 ) = inf T (s)q(s)ds
:(0,q0 ,q0 ) 0

(6.99)
t
= sup T
(s)q(s)ds
:(0,q0 ,q0 ) 0

The notation inf means that one performs the innimization over all
:(0,q0 ,q0 )
trajectories of the system on intervals [0, t], t 0, starting from the extended
state (0, q0 , q0 ), with (q0 , q0 ) = (q(0), q(0)), with admissible inputs (at least
the closed-loop system must be shown to be well-posed). In other words the
innimization is done over all trajectories (t; 0, q0 , q0 , ), t 0. From (6.99)
one obtains
, t -
1 T
Va (q0 , q0 ) = sup q M (q)q + Ug (q(t)) Ug (q(0))
:(0,q0 ,q0 ) 2 0

(6.100)
= 12 q(0)T M (q(0))q(0) + Ug (q(0))

= E(q0 , q0 )

where we have to assume that Ug (q) K > for some K < +, so

that we may assume that the potential energy has been normalized to secure
that Ug (q) 0 for all q IRn . It is not surprizing that the available storage
is just the total initial mechanical energy of the system (but we shall see in a
moment that for certain systems this is not so evident).

Remark 6.40. We might have deduced that the system is dissipative since
Va (q, q) < + for any bounded state; see Theorem 4.41. On the other hand,
Va (q, q) must be bounded since we already know that the system is dissipative
with respect to the chosen supply rate.

Remark 6.41. In Section 6.1 we saw that the addition of Rayleigh dissipation
enforces the dissipativity property of the system. Let us recalculate the avail-
able storage of a rigid joint-rigid link manipulator when the dynamics is given
by
R
M (q(t))q(t) + C(q(t), q(t))q(t) + g(q(t)) + (t) = (t) (6.101)
q
342 6 Dissipative Physical Systems

One has:

t
Va (q0 , q0 ) = sup T qds
:(0,q0 ,q0 ) 0

,  t
t -
1 T t T R
= sup q M (q)q [Ug (q)]0 q ds
:(0,q0 ,q0 ) 2 0 0 q

= 12 q(0)T M (q(0))q(0) + Ug (q(0))

= E(q0 , q0 )
(6.102)
q q q for some > 0. One therefore concludes that the dissipa-
since q T R T

tion does not modify the available storage function, which is a logical feature
from the intuitive physical point of view (the dissipation and the storage are
dened independently).

6.3.2 The Required Supply

Let us now compute the required supply Vr (q, q) as in Denition 4.36, with
the same assumption on Ug (q). Recall that it is given in a variational form
by:
0
Vr (q0 , q0 ) = inf T (s)q(s)ds (6.103)
:(t,qt ,qt )(0,q0 ,q0 ) t

where (qt , qt ) = (q(t), q(t)), (q0 , q0 ) = (q(0), q(0)), t 0. Thus this time
the minimization process is taken over all trajectories of the system, joining
the extended states (t, qt , qt ) and (0, q0 , q0 ) (i.e. (q0 , q0 ) = (0; t, qt , qt , )).
For the rigid manipulator case one nds

Vr (q0 , q0 ) = inf [E(q0 , q0 ) E(q(t), q(t))]

:(t,qt ,qt )(0,q0 ,q0 )
(6.104)
= E(0) E(t)

Note that Vr () hence dened is not necessarily positive. However if we

compute it from (t, qt , qt ) = (t, 0, 0) then indeed Vr () 0 is a storage
function. Here one trivially nds that Vr (q0 , q0 ) = E(q0 , q0 ) (= Va (q0 , q0 )).

Remark 6.42. The system is reachable from any state (q0 , q0 ) (actually, this
system is globally controllable). Similarly to the available storage function
property, the system is dissipative with respect to a supply rate if and only if
the required supply Vr K for some K > ; see Theorem 4.41. Here we
can take K = E(t).
 6.4 Flexible JointRigid Link Manipulators 343

6.4 Flexible JointRigid Link Manipulators

In this section we consider another class of systems which corresponds to mod-
els of manipulators whose joints are no longer assumed to be perfectly rigid,
but can be fairly modelled by a linear elasticity. Their simplied dynamics
can be written as

M (q1 (t))q1 (t) + C(q1 (t), q1 (t))q1 (t) + g(q1 (t)) = K(q2 (t) q1 (t))
(6.105)

J q2 (t) = K(q1 (t) q2 (t)) + u(t)

where q1 (t) IRn is the vector of rigid links angles, q2 (t) IRn is the vector
of motor shaft angles, K IRnn is the joint stiness matrix and J IRnn
is the motor shaft inertia matrix (both assumed here to be constant and
diagonal). It is a simple
mechanical
 system in Lagrangian  form (6.43),
 we can
M (q1 ) 0 C(q1 , q1 ) 0 0
say that M (q) = , C(q, q) = , = , g(q) =
0 J 0 0 u
 
g(q1 ) K(q2 q1 )
+ . Actually the potential energy is given by the sum of
0 K(q1 q2 )
the gravity and the elasticity terms, Ug (q1 ) and Ue (q1 , q2 ) = 12 (q2 q1 )T K(q2
q1 ) respectively. The dynamics of exible joint-rigid link manipulators can
be seen as the interconnection of the simple mechanical system representing
the dynamics of the rigid joint-rigid link manipulators with a set of linear
Lagrangian systems with external forces representing the inertial dynamics of
the rotor, interconnected by the rotational spring representing the compliance
of the joints. It may be seen as the power continuous interconnection of the
corresponding three port controlled Hamiltonian systems in a way completely
similar to the example of the levitated ball (Example 6.36). We shall not
detail the procedure here but summarize it on Figure 6.2. As a result it follows
that the system is passive, lossless with respect to the supply rate uT q2 with
storage function being the sum of the kinetic energies and potential energies
of the dierent elements. We shall see in Section 6.6 that including actuator
dynamics produces similar interconnected systems, but with quite dierent
interconnection terms. These terms will be shown to play a crucial role in the
stabilizability properties of the overall system.

Remark 6.43. The model in (6.105) was proposed by Spong [471] and is based
on the assumption that the rotation of the motor shafts due to the link angular
motion does not play any role in the kinetic energy of the system, compared
to the kinetic energy of the rigid links. In other words the angular part of the
kinetic energy of each motor shaft rotor is considered to be due to its own ro-
tation only. This is why the inertia matrix is diagonal. This assumption seems
satised in practice for most of the manipulators. It is also satised (mathe-
matically speaking) for those manipulators whose actuators are all mounted
at the base, known as parallel-drive manipulators (the Capri robot presented
344 6 Dissipative Physical Systems

Fig. 6.2. Flexible jointrigid link: interconnection as two passive blocks

in chapter 9 is a parallel-drive manipulator). If this

assumption is not satis-

M (q1 ) M12 (q1 )
ed [488], the inertia matrix takes the form M (q) = T .
M12 (q1 ) J

The particular feature of the model in (6.105) is that it is static feedback

linearizable and possesses a triangular structure [325] that will be very useful
when we deal with control.
Let us now prove in some other way that the system is passive (i.e. dissi-
pative with respect to the supply rate T q = uT q2 ). We get for all t 0:
t T t T
0 u (s)q2 (s)ds = 0 [J q2 (s) + K(q2 (s) q1 (s))] q2 (s)ds
t
0
(q2 (s) q1(s) )T K q1 (s)ds
(6.106)
1 t 1 t
2 (q2 q1 ) K(q2 q1 ) 0 +
T T
= 2 q2 J q2 0
+
t
+ 0 (q2 (s) q1 (s))T K q1 (s)ds
The last integral term can be rewritten as

t
t
(q2 q1 )T K q1 ds = q1T [M (q1 )q1 + C(q1 , q1 )q1 + g(q1 )] ds (6.107)
0 0

Looking at the rigid jointrigid link case, one sees that

t  t
1 T
(q2 q1 )T K q1 ds = q1 M (q1 )q1 + Ug (q1 ) (6.108)
0 2 0
 6.4 Flexible JointRigid Link Manipulators 345

Therefore grouping (6.106) and (6.108) one obtains

t T
0
u q2 ds 12 q2 (0)T J q2 (0)

12 q1 (0)T M (q1 (0)q1 (0) (6.109)

12 [q2 (0) q1 (0)]T K[q2 (0) q1 (0)] Ug (q1 (0))

The result is therefore true whenever Ug (q1 ) is bounded from below.

Remark 6.44. One could have thought of another decomposition of the system
as depicted in Figure 6.3. In this case the total system is broken down into two
Lagrangian control systems with input being the free end of the springs with
respect to each submodel. The subsystem with generalized coordinate q1 (i.e.
representing the dynamics of the multibody system of the robot) is analogous
to the harmonic oscillator of Example 6.12 and with input q2 . The dynamics of
the rotors (with generalized coordinates q2 ) is again analogous to an additional
external force u. But the interconnection of these two subsystems is dened
by : u1 = q2 and u2 = q1 involving the generalized coordinates which are not
passive outputs of the subsystems.

Fig. 6.3. Flexible jointrigid link manipulator

Remark 6.45. Let us point out that manipulators with prismatic joints cannot
be passive, except if those joints are horizontal. Hence all those results on
open-loop dissipativity hold for revolute joint manipulators only. This will
not at all preclude the application of passivity tools for any sort of joints
when we deal with feedback control for instance it suces to compensate
for gravity to avoid this problem.
346 6 Dissipative Physical Systems

6.4.1 The Available Storage

Mimicking the rigid joint-rigid link case, one nds that

Va (q, q) = E(q, q) = 12 q1T M (q1 )q1 + 12 q2T J q2

(6.110)
+ 12 [q1 q2 ]T K[q1 q2 ] + Ug (q1 )

6.4.2 The Required Supply

From Subsection 6.3.2 one nds that the energy required from an external
source to transfer the system from the extended state

(t, q1 (t), q2 (t), q1 (t), q2 (t)) = (t, q1t , q2t , q1t , q2t )

to
(0, q1 (0), q2 (0), q1 (0), q2 (0)) = (0, q10 , q20 , q10 , q20 ),
is given by

Vr (q1 (0), q2 (0), q1 (0), q2 (0)) = E(q1 (0), q2 (0), q1 (0), q2 (0))
(6.111)
E(q1 (t), q2 (t), q1 (t), q2 (t))

The KYP Lemma Conditions

Recall from the Positive Real (or Kalman-Yacubovich-Popov) Lemma 4.84

that a system of the form

x = f (x) + g(x)u
(6.112)

y = h(x)

is passive (dissipative with respect to the supply rate uT y) if and only if

there exists at least one function V (t, x) 0 such the following conditions are
satised: T
h (x) = V x (x)g(x)
(6.113)
V
x (x)f (x) 0
The if part of this Lemma tells us that an unforced system that is Lyapunov
stable with Lyapunov function V () is passive when the output has the partic-
ular form in (6.113). The only if part tells us that given an output function,
then passivity holds only if the searched V () does exist.
Now let us assume that the potential function Ug (q1 ) is nite for all q C.
Then it follows that the available storage calculated in (6.110) is a storage
function, hence it satises the conditions in (6.113) when y = JJ 1 q2 = q2 and
 6.5 A Bouncing System 347

u is dened in (6.105). More explicitly the function E(q, q) in (6.110) satises

the partial dierential equations (in (6.105) one has g T (x) = (0, 0, 0, J 1 ))


E T 1 T
q2 J = q2




E T E T 1
q1 q1 + q1 M (q1 ) [C(q1 , q1 )q1 g(q1 ) + K(q2 q1 )] + (6.114)






+ E T q + E J 1 [K(q q )] = 0
q2 2 q2 1 2

6.5 A Bouncing System

We may conclude from the preceding examples that in general, for mechanical
systems, the total mechanical energy is a storage function. However the cal-
culation of the available storage may not always be so straightforward as the
following example shows. Let us consider a one degree-of-freedom system com-
posed of a mass striking a compliant obstacle modelled as a spring-dashpot
system. The dynamical equations for contact and non-contact phase are given
by

f q(t) kq(t) if q(t) > 0
mq(t) = (t) + (6.115)

0 if q(t) 0
It is noteworthy that the system in (6.115) is nonlinear since the switch-
ing condition depends on the state. Moreover existence of a solution with q
continuously dierentiable is proved in [392] when is a Lipschitz continuous
function of time, q and q. The control objective is to stabilize the system at
rest in contact with the obstacle. To this aim let us choose the input

= 2 q 1 (q qd ) + v (6.116)
with qd > 0 constant, 1 > 0, 2 > 0 and v is an auxiliary signal. The
input in (6.116) is a PD controller but can also be interpreted as an input
transformation. Let us now consider the equivalent closed-loop system with
input v and output q, and supply rate w = v q. The available storage function
is given by
t
Va (x0 , x0 ) = sup v(s)q(s)ds (6.117)
:(0,q0 ,q0 ) t0

Due to the systems dynamics in (6.115) we have to consider two cases:

Case q0 0: Let us denote 2i = [t2i , t2i+1 ] the time intervals such that
q(t) 0, and 2i+1 = [t2i+1 , t2i+2 ] the intervals such that q(t) > 0, i IN .
From (6.116) and (6.115) one has
348 6 Dissipative Physical Systems

Va (q0 , q0 ) =



8
= sup (mq(s) + 2 q(s) + 1 q(s) 1 qd )q(s)ds
:(0,q0 ,q0 ) i0 2i

, -


(mq(s) + 2 q(s) + (1 + k)q(s) 1 qd )q(s)ds
i0 2i+1

,  t  t2i+1
-
 q 2 2i+1 1 2 2
= sup m (x xd ) 2 q (t)dt
:(0,q0 ,q0 ) i0 2 t2i 2 t2i 2i

4
 2 5t2i+2

q 2 1 + k 1 qd
+ m q (2 + f ) q 2 (t)dt
2 2 1 + k 2i+1
i0 t2i+1
(6.118)
In order to maximize the terms between brackets it is necessary that the
integrals i q 2 (t)dt be zero and that q(t2i+1 ) = 0. In view of the sys-
tems controllability, there exists an impulsive input v that fulls these
requirements [246] (let us recall that this impulsive input is applied while
the system evolves in a free-motion phase, hence has linear dynamics). In
 t1
order to maximize the second term 21 (q qd )2 t0 it is also necessary
that q(t1 ) = 0. Using similar arguments, it follows that q(t2i+2 ) = 0 and
that q(t2 ) = 11+k
qd
. This reasoning can be iterated to obtain the optimal
path which is (q0 , q0 ) (0, 0) ( 11+k
qd
, 0) where all the transitions are
instantaneous. This leads us to the following available storage function:

q02 1 q02 21 qd2

Va (q0 , q0 ) = m + 1 qd q0 + (6.119)
2 2 2(1 + k)

Case q0 > 0: Using a similar reasoning one obtains

2
q 2 (1 + k) 1 qd
Va (q0 , q0 ) = m 0 + q0 (6.120)
2 2 1 + k
Notice that the two functions in (6.119) and (6.120) are not equal. Their
concatenation yields a positive denite function of (q, q) = (0, 0) with q = q
1 qd
1 +k , that is continuous at q = 0, but not dierentiable (this is in accordance
with [33, Proposition]).

Remark 6.46. Let us now consider the following systems

mq(t) + 2 q(t) + 1 (q(t) qd ) = v(t) (6.121)

and
 6.6 Including Actuator Dynamics 349

mq(t) + (2 + f )q(t) + 1 (q(t) qd ) + kq(t) = v(t) (6.122)

that represent the persistent free motion and the persistent contact motion
dynamics respectively. The available storage function for the system in (6.121)
is given by (see Remark 6.41)
1 2 1
Va (q, q) = mq + 1 (q qd )2 (6.123)
2 2
whereas it is given for the system in (6.122) by
1 2 1 1
Va (q, q) = mq + 1 (q qd )2 + kq 2 (6.124)
2 2 2
It is clear that the functions in (6.119) and (6.123), (6.120) and (6.124), are
respectively not equal. Notice that this does not preclude that the concate-
nation of the functions in (6.123) and (6.124) yield a storage function for the
system (in which case it must be larger than the concatenation of the func-
tions in (6.119) and (6.120) for all (q, q)). In fact an easy inspection shows that
kq2
the functions in (6.123) and (6.124) are obtained by adding 12 11 +kd to those
in (6.119) and (6.120) respectively. Thus their concatenation indeed yields a
storage function for the system in (6.115) with input (6.116).
An open issue is to study the conditions under which the available storage
function of the piecewise continuous system


x(t) = fi (x(t), u(t)) if Ci x(t) 0



= gi (x(t), u(t)) if Ci x(t) < 0
x(t) (6.125)





i {1, , m}
can be deduced as a concatenation of the available storages of the independent
systems x = fi (x, u) and x = gi (x, u). More generally one should study the
dissipativity of switching systems. An important contribution can be found
in [540], where a suitable denition of dissipativity is proposed that involves
several supply rates and storage functions 1 . Other contributions are in [534,
535].

6.6 Including Actuator Dynamics

6.6.1 Armature-controlled DC Motors
In all the foregoing examples it has been assumed that the control is directly
provided by the generalized torque . In reality the actuators possess their
1
It is noteworthy that the class of systems considered in [534, 535, 540] does not
encompass the nonsmooth systems which are examined elsewhere in this book,
like unbounded dierential inclusions, variational inequalities. They are dierent
types of dynamical systems.
350 6 Dissipative Physical Systems

own dynamics, and the torque is just the output of a dynamical system. In
practice the eect of neglecting those dynamics may deteriorate the closed-
loop performance [79]. In other words, the dynamics in (6.43) are replaced by
a more accurate armature-controlled DC motor model as:

M (q(t))q(t) + C(q(t), q(t))q(t) + g(q(t)) = = Kt I(t)
(6.126)

RI(t) + L dI
dt (t) + Kt q(t) = u(t)

u
.
state (I)
Kt I . q
state (q, q)
_

Kt

Fig. 6.4. Manipulator + armature-controlled DC motor

where R, L, Kt are diagonal constant matrices with strictly positive entries,

R IRnn is a matrix whose jth entry is the resistance of the jth motor
armature circuit, L IRnn has entries which represent the inductances of
the armature, Kt IRnn represents the torque constants of each motor,
u IRn is the vector of armature voltage, I IRn is the vector of armature
currents. For the sake of simplicity we have assumed that all the gear ratios
that might relate the various velocities are equal to one. Moreover the inertia
matrix M (q) is the sum of the manipulator and the motorschaft inertias.
The new control input is therefore u, see Figure 6.4. For the moment we are
interested in deriving the passivity properties of this augmented model. We
shall see further that the (disconnected) dynamics of the motor are strictly
output passive with respect to the supply rate uT I.
Remark 6.47. One may consider this system as the interconnection of two
subsystems as in Figure 6.5. One notes at once a strong similarity between the
model in (6.126) and the example of the magnetic ball in Example 6.34. The
dierence is that there is no coupling through the energy function (no state
variable in common) but that the simple mechanical system, representing the
dynamics of the mechanical part, is non-linear. The interconnection structure
 6.6 Including Actuator Dynamics 351

Kt-1u Kt I
motor
_ dynamics

manipulator
dynamics
_

Rayleigh
dissipation

(to be added if dissipation in the

manipulator)

Fig. 6.5. Negative feedback interconnection in two dissipative blocks

is best seen on the formulation using port controlled Hamiltonian systems

as follows and illustrated in Figure 6.5. The Legendre transformation of the
simple mechanical system leads to the denition of the momentum vector
1 T 1
p = L q = M (q)q, the Hamiltonian function H(q, p) = 2 p M (q)p + U (q)
and the following port controlled Hamiltonian system:
  . H / 
q 0n In q 0n
= + (6.127)
p In 0n H
p
In
(6.128)
. /
H
q
ymech = (0n , In ) H = q (6.129)
p

where the input represents the electromechanical forces. The dynamics of

the motors is described by the following port controlled Hamiltonian system
with dissipation with state variable being the total magnetic ux = LI and
1 2
the magnetic energy being Hmg = 2L :

Hmg
= R + u + umg (6.130)

Hmg
ymg = =I (6.131)

352 6 Dissipative Physical Systems

where umg represents the electromotive forces. Note that the structure matrix
consists only of a negative denite part, thus it is purely an energy dissipating
system. The interconnection between the two subsystems is dened by the
following power continuous interconnection:

= Kt ymg (6.132)
umg = Kt ymech (6.133)
A simple elimination leads to the following port controlled Hamiltonian
system with dissipation
H

q

q 0n In 0n 02n 02nn
p = In 0n Kt +

H

p

0n Kt 0n 0n2n R
Hmg (6.134)


0n
+ 0n u
In
H

q


H
y = (0n , 0n , In ) =I (6.135)
p


Hmg

From this formulation of the system as interconnected port controlled
Hamiltonian with dissipation, the interconnected system is seen to be pas-
sive with supply rate uT I and storage function H(q, p) + Hmg ().

Passivity with Respect to the Supply Rate uT I

Let us calculate directly the value of u, It , where the choice of this supply
rate is motivated by an (electrical) energy expression:
t  
u, It = 0 I T RI + L dIdt + Kv q

t 1
 t
= 0 I(s)T RI(s)ds + 2 I(s)T LI(s) 0
 t
+ 12 q(s)T M (q(s))q(s) 0 + [Ug (q(s))]t0 (6.136)
t
0
I(s)T RI(s)ds 12 I(0)T LI(0)

12 q(0)T M (q(0))q(0) Ug (q(0))

 6.6 Including Actuator Dynamics 353

where we used the fact that R > 0, L > 0. One sees that the system in
(6.126) is even strictly output passive when the output is y = Kt I. Indeed
I T RI min (R)y T y where min (R) denotes the minimum eigenvalue of R.

Available Storage and Required Supply

Using the same supply rate as in Subsection 6.6.1, one gets

Va (q, q, I) = 12 I T LI + 12 q T M (q)q + Ug (q)
(6.137)
= Vr (q, q, I)

Necessity and Suciency for the Supply Rate to be uT I

The supply rate uT I has been chosen according to the denition of conjugated
port variables of port controlled Hamiltonian systems. In the sequel, we shall
prove that no other form on the port variables may be chosen to dene a
supply rate for another storage function. Therefore let us introduce a more
general supply rate of the form uT AT BI for some constant matrices A and
B of suitable dimensions. Our goal is to show that if the system is dissipative
with respect to this new supply rate, then necessarily (and suciently) A =
1 1 1
U Kt and B = Kt U , where = 0 and U is a full-rank symmetric
matrix. Let us compute the available storage associated to this supply rate,
i.e.

t
Va (q0 , q0 , I0 ) = sup uT (s)AT BI(s)ds
u2 :(0,q0 ,q0 ,I0 ) 0

t
1 T (6.138)
= sup [I LAT BI]t0 + I T RAT BIds
u2 :(0,q0 ,q0 ,I0 ) 2 0

t 3
+ 0
q T Kt AT BKt1 [M (q)q + C(q, q)q + g(q)] ds

It follows that the necessary conditions for Va (q, q, I) to be bounded are that
LAT B 0 and RAT B 0. Moreover the last integral concerns the dissi-
pativity of the rigid joint-rigid link manipulator dynamics. We know storage
functions for this dynamics, from which it follows that an output of the form
Kt1 B T AKt q does not satisfy the (necessary) Kalman-Yakubovic-Popov prop-
erty, except if Kt1 B T AKt = In . One concludes that the only supply rate with
respect to which the system is dissipative must satisfy
1 T

Kt B AKt = In



LAT B 0 (6.139)





RAT B 0
354 6 Dissipative Physical Systems

Hence A = 1 U 1 Kt1 and B = Kt U for some = 0 and some full-rank

matrix U = U T .

6.6.2 Field-controlled DC Motors

Now consider the model of rigid joint-rigid link manipulators actuated by
eld-controlled DC motors:
dI

L1 dt1 + R1 I1 = u1



L2 dI2
dt + R2 I2 + Kt (I1 )q = u2
(6.140)





M (q)q + C(q, q)q + g(q) + Kv q = = Kt (I1 )I2
where I1 , I2 are the vectors of currents in the coils of the n motors actuating
the manipulator, L1 and L2 denote their inductances, R1 and R2 are the
resistors representing the losses in the coils. The matrix Kt (I1 ) represent the
electromechanical coupling and is dened by a constant diagonal matrix Kt
as follows:

Kt (I1 ) = diag(kt1 I11 , , ktn I1n ) = Kt I1 (6.141)

with kti > 0. The last equation is the Lagrangian control system represent-
ing the dynamics of the manipulator with n degrees of freedom dened in
(6.98) where the diagonal matrix Kv is positive denite and represents the
mechanical losses in the manipulator.
In order to reveal the passive structure of the system, we shall again,
like in the preceding case, assemble it as the interconnection of two passive
port controlled Hamiltonian systems. Therefore let us split this system in
two parts: the magnetic part and the mechanical part and interconnect them
through a power continuous interconnection. The rst port controlled Hamil-
tonian system with dissipation represents the magnetic energy storage and
the electromechanical energy transduction. The state variables are the to-
T
tal magnetic
 uxes in the coils = (1 , 2 ) dening the magnetic energy
1 1 2 1 2
Hmg = 2 L1 + L1 and becomes

 Hmg  
R1 0n 1 1 0
= + u1 + u2
0n R2 Hmg 0 1
(6.142)
 2
0
+ 1 umg
Kt L 1

with the conjugated outputs associated to the voltages u1 and u2 :

Hmg Hmg
1
1

y1 = (1, 0) = I1 and y2 = (0, 1) = I2 (6.143)
Hmg Hmg
2 2
 6.6 Including Actuator Dynamics 355

and the output conjuguated to the electromotive force umg is

 Hmg
1 1

ymg = 0, Kt (6.144)
L1 Hmg
2

where the two conjugated port variables umg and ymg dene the intercon-
nection with the mechanical system. The second port controlled Hamiltonian
system with dissipation represents the dynamics of the manipulator and was
presented above:
H
q 0n In q

= + 0n umech (6.145)
In
p In Kv H
p
H
q
ymech = (0n , In ) = q (6.146)
H
p

where one notes that the dissipation dened by the matrix Kt was included
in the structure matrix. The interconnection of the two subsystems is dened
as an elementary negative feedback interconnection:

umech = ymg (6.147)

umg = ymech (6.148)

Again a simple elimination of the interconnection variables leads to the

port controlled Hamiltonian system with dissipation, with Hamiltonian being
the sum of the Hamiltonian of the subsystems: Htot (, q, p) = Hmg () +
H(q, p) and structure matrice with skew-symmetric part

0n 0n 0n


=
0n 0n Kt L1

1
Jtot (6.149)

1
0n Kt L 1
0n

and symmetric positive structure matrix:

Rtot = diag(R1 , R2 , Kv ) (6.150)

Hence the complete system is passive with respect to the supply rate of
the remaining port variables: u1 y1 + y2 u2 and with storage function being the
total energy Htot .
356 6 Dissipative Physical Systems

Passivity of the Manipulator Plus Field-controlled DC Motor

Motivated by the preceding physical analysis of the eld-controlled DC motor,

using the integral formulation of the passivity, let us prove the dissipativity
with respect to the supply rate uT1 I1 + uT2 I2 :
t
u1 , I1 t + u2 , I2 t 12 I1 (0)T L1 I1 (0) + 0
I1T (s)R1 I1 (s)ds
t
21 I2 (0)T L2 I2 (0) + 0 I2T (s)R2 I2 (s)ds+
t
+ q T (s)Kt (I1 (s))I2 (s)ds (6.151)
0

12 I1 (0)T L1 I1 (0) 12 I2 (0)T L2 I2 (0)

21 q(0)T M (q(0))q(0) Ug (q(0))

which proves the statement.

Remark 6.48 (Passivity of the motors alone). The dynamics of a eld-controlled

DC motor is given by
dI

L1 dt1 (t) + R1 I1 (t) = u1 (t)



L2 dI 2
dt (t) + R2 I2 (t) + Kv (I1 (t))q(t) = u2 (t)
(6.152)





J q(t) = Kt (I1 (t))I2 (t) Kvt q(t)

where J IRnn is the rotor inertia matrix. It follows that the (disconnected)
actuator is passive with respect to the supply rate uT1 I1 + uT2 I2 . Actually
we could have started by showing the passivity of the system in (6.152) and
then proceeded to showing the dissipativity properties of the overall system in
(6.140) using a procedure analog to the interconnection of subsystems. Similar
conclusions hold for the armature-controlled DC motor whose dynamics is
given by
J q(t) = Kt I(t)
(6.153)

RI(t) + L dI
dt (t) + K t q(t) = u(t)
and which is dissipative with respect to uT I. This
 dynamics is even output
I1
strictly passive (the output is y = I or y = ) due to the resistance.
I2

The Available Storage

The available storage function of the system in (6.140) with respect to the
supply rate uT1 I1 + uT2 I2 is found to be, after some calculations:
 6.6 Including Actuator Dynamics 357

1 T 1 1
Va (I1 , I2 , q, q) = I L1 I1 + I2T L2 I2 + q T M (q)q + Ug (q) (6.154)
2 1 2 2
This is a storage function and a Lyapunov function of the unforced system in
(6.140).

Remark 6.49. Storage functions for the disconnected DC motors are given by
Vadc (I, q, q) = 12 q T J q + 12 I T LI and Vf dc (I1 , I2 , q, q) = 12 q T J q + 12 I1T L1 I1 +
1 T
2 I2 L2 I2 . Notice that they are not positive denite functions of the state
(q, q, I) but they are positive denite functions of the partial state (q, I).
Hence the xed point (q, I) = (0, 0) (or (q, I1 , I2 ) = (0, 0, 0)) is asymptotically
stable.
Notice that the actuator dynamics in (6.152) with input (u1 , u2 ) and out-
put (I1 , I2 ) (which are the signals from which the supply rate is calculated,
hence the storage functions) is zero-state detectable: ((u1 , u2 ) (0, 0) and
I1 = I2 = 0) = q = 0 (but nothing can be concluded on q), and is strictly
output passive. From Lemma 5.13 one may conclude at once that any func-
tion satisfying the Kalman-Yacubovich-Popov conditions is indeed positive
denite.

Remark 6.50. The model of eld-controlled DC motors in (6.152) is similar to

that of induction motors, that may be given in some reference frame by (here
we show the model for one motor whereas in (6.152) the dynamics represent
a system composed of n motors):



u1 (t)
Lz(t) + C(z(t), u3 (t))z(t) + Rq(t) = E + d(t)
u2 (t)
(6.155)



y(t) = Lsr (I2 (t)I3 (t) I1 (t)I4 (t))

where z T = [I1 , I2 , I3 , I4 , q] IR5 , uT = [u1 , u2 , u3 ] IR3 , dT

= [0, 0, 0, 0, d5],
Ce (u3 , q) c(q)
L = diag(Le , vJ) IR55 , C(z, u3 ) = IR55 , E =
T
c (q) 0

I2 52 55 44
IR , R = diag(Re , vb) IR . Le IR is a matrix of in-
032
ductance, v IR is the number of pole pairs, J IR is the rotor inertia,
Re IR44 is the matrix of resistance, b IR is the coecient of motor
damping, u1 and u2 are stator voltages, u3 is the primary frequency, I1 and
I2 are stator currents, I3 and I4 are rotor currents, q is the rotor angular veloc-
ity, d5 = vyl where yL is the load torque. Finally y(t) IR is the generated
torque, where Lsr IR is the mutual inductance.
It can be shown that this model shares properties with the Euler-Lagrange
dynamics. In particular [385] the matrix C(z, u3 ) satises the skew-symmetry
requirement for a certain choice of its denition (which is not unique), and
z T C(z, u3 )z = 0 (similarly to workless forces). Also this system is strictly
358 6 Dissipative Physical Systems

passive with respect to the supply rate I1 u1 + I2 u2 , with storage function

H(z) = 12 z T Lz and function S(z) = z T Rz (see Denition 4.51).

6.7 Passive Environment

In this section we shall briey treat systems which may be considered as
models of manipulators in contact with their environment through their end-
eector or some other body (for instance in assembly tasks or in cooperation
with other robots). These systems are part of a more general class of con-
strained dynamical systems or implicit dynamical systems which constitute
still an open problem for their simulation and control. More precisely we
shall consider simple mechanical systems which are subject to two types of
constraints. First, we shall consider ideal, i.e. workless, constraints on the
generalized coordinates or velocities which again may be split into integrable
constraints which may be expressed on the generalized coordinates and non-
holonomic constraints which may solely be expressed in terms of the gen-
eralized velocities. Second we shall consider the case when the environment
itself is a simple mechanical system and hence consider two simple mechanical
systems related by some constraints on their generalized coordinates.

6.7.1 Systems with Holonomic Constraints

Let us consider rst a robotic manipulator whose motion is constrained by

some m bilateral kinematic constraints, for instance following a smooth surface
while keeping in contact. Its model may be expressed as a simple mechanical
system (6.43) of order 2n with m < n kinematic constraints of order zero, and
dened by some real function from the space of generalized coordinates IRn
in IRm :

(q) = 0 (6.156)
Let us assume moreover that the Jacobian J(q) = q is of rank m everywhere
and the kinematic constraints (6.156) dene a smooth submanifold Qc of
IRn . Then by dierentiating the constraints (6.156) one obtains kinematic
constraints of order 1, dened on the velocities:

J(q)q = 0 (6.157)

The two sets of constraints (6.156) and (6.157) dene now a submanifold
S on the state space T IRn = IR2n of the simple mechanical system (6.43):
) *
S = (q, q) IR2n : (q) = 0, J(q)q = 0 (6.158)

The dynamics of the constrained simple mechanical system is then de-

scribed by the following system:
 6.7 Passive Environment 359

M (q(t))q(t) + C(q(t), q(t))q(t) + g(q(t)) = (t) + J T (q(t))(t)
(6.159)

J(q(t))q(t) = 0

where IRm is the m dimensional vector of the Lagrangian multipliers

associated with the constraint (6.156). They dene the reaction forces Fr =
J T (q) associated with the constraint which enforce the simple mechanical
system to remain on the constraint submanifold S dened in (6.158).

Remark 6.51. Note that the constrained system (6.159) may be viewed as
a port controlled Hamiltonian system with conjugated port variables and
y = J(q)q interconnected to a power continous constraint relation dened by
y = 0 and IRm . It may then be shown that this denes an implicit port
controlled Hamiltonian system [121,346]. More general denition of kinematic
constraints were considered in [345, 347].

Remark 6.52. Constrained dynamical systems are the subject of numerous

works which are impossible to present here in any detail, and we refer the
interested reader to [337] for a brief historical presentation and presentation
of related Hamiltonian and Lagrangian formulation as well as to [438] for a
Hamiltonian formulation in some more system theoretic setting.

Remark 6.53. Note that the kinematic constraint of order zero (6.156) is not
included in the denition of the dynamics (6.159). Indeed it is not relevant to
it, in the sense that this dynamics is valid for any constraint (q) = c where
c is a constant vector and may be xed to zero by the appropriate initial
conditions.

One may reduce the constrained system to a simple mechanical system

of order 2(n m) by using an adapted set of coordinates as proposed by
McClamroch and Wang [324]. Using the Theorem of implicit functions, one
may nd, locally, a function from IRnm to IRm such that

((q2 ), q2 ) = 0 (6.160)

Then dene the change of coordinates:


q1 (q2 )
z = Q(q) = (6.161)
q2

Its inverse is then simply


z1 (z2 )
q = Q(z) = (6.162)
z2

In the new coordinates (6.161), the constrained simple mechanical system

becomes
360 6 Dissipative Physical Systems

Q T Im
M (z(t))z(t) + C(z(t), z(t))z(t) + g(z(t)) = q (t) (t) + (t)
0nm

z1 (t) = (Im 0nm ) z(t) = 0
(6.163)
where the inertia matrix is dened by

Q T Q
M (z) = (Q(z))M (Q(q)) (Q(z)) (6.164)
q q

and g(z) is the gradient of the potential function U (Q(z)). The kinematic
constraint is now expressed in a canonical form in (6.163) or in its integral
form z1 = 0. The equations in (6.163) may be interpreted as follows: the sec-
ond equation corresponds to the motion along the tangential direction to the
constraints. It is not aected by the interaction force since the constraints are
assumed to be frictionless. It is exactly the reduced-order dynamics that one
obtains after having eliminated m coordinates, so that the nm remaining co-
ordinates z2 are independent. Therefore the rst equation must be considered
as an algebraic relationship that provides the value of the Lagrange multiplier
as a function of the systems state and external forces.
Taking into account the canonical expression of the kinematic constraints,
the constrained system may then be reduced to the simple mechanical system
of order 2(nm) with generalized coordinates z2 , and inertia matrix (dening
the kinetic energy) being the submatrix Mr (z2 ) obtained by extracting the
last n m columns and rows from M (z) and setting z1 = 0. The input term
is obtained by taking into account the expression of Q and computing its
Jacobian: 
Q T Im 0m(nm)
= (6.165)
z q

2
(Q(z) Inm
The reduced dynamics is then a simple mechanical system with inertia
matrix Mr (z) and is expressed by


Mr (z(t))z(t) + Cr (z(t), z(t))z(t) + gr (z(t)) = (z2 (t)), Inm (t)
q2
(6.166)
The port conjuguated output to is then

q2 (q2 (t))
yr (t) = z2 (t) (6.167)
Inm

Hence the restricted system is passive and lossless with respect to the sup-
ply rate T yr and storage function being the sum of the kinetic and potential
energy of the constrained system.

Remark 6.54. We have considered the case of simple mechanical systems sub-
ject to holonomic kinematic constraints, that means kinematic constraints of
order 1 in (6.157), that full some integrability conditions which guarantee
 6.7 Passive Environment 361

the existence of kinematic constraints of order 0 (6.156). If this is not the

case, the constraints are said to be non-holonomic. This means that the sys-
tem may no more be reduced to a lower order simple mechanical system. As
we shall not treat them in the sequel, we do not give a detailed presentation
and give a sketch of the results indicating only some references. These sys-
tems may still be reduced by choosing an adapted set of velocities (in the
case of a Lagrangian formulation) or momenta in the case of a Hamiltonian
formulation) and then projecting the dynamics along a subspace of velocities
or momenta [95, 337, 440]. This dynamics cannot be expressed as a controlled
Lagrangian systems, however it has been proved that it may still be expressed
as a port controlled Hamiltonian system for which the structure matrix does
not satisfy the Jacobi identities (6.68) [264, 440].

6.7.2 Compliant Environment

The General Dynamics

The general dynamical equations of a rigid joint-rigid link manipulator in

permanent contact with an obstacle (that is also a Euler-Lagrange system
and can be for instance another uncontrolled kinematic chain) are given by

M (q(t))q(t) + C(q(t), q(t))q(t) + g(q(t)) = (t) + Fq (t)
dRe
Me (x(t))x(t) + Ce (x(t), x(t))x(t) + dx (t)
+ ge (x(t)) + Ke x(t) = Fx (t)
(6.168)
where q(t) IRn , x(t) IRm , m < n, Fq (t) IRn and Fx (t) IRm rep-
resent the generalized interaction force in coordinates q and in coordinates
x respectively. In other words, if x = (q) for some function (), then
T
x = d T
dx (q)q = J(q)q, and Fq = J (q)Fx . If we view the system in (6.168) as
a whole, then the interaction force becomes an internal force. The virtual work
principle (for the moment let us assume that all contacts are frictionless) tells
us that for any virtual displacements q and x, one has xT Fx = q T Fq .
This can also be seen as a form of the principle of mutual actions. Let us fur-
ther assume that rank() = m and that Ke > 0. Let us note that the relation
x = (q) relates the generalized displacements of the controlled subsystem to
those of the uncontrolled one, i.e. to the deexion of the environment. With
this in mind, one can dene following
 McClamroch
 a nonlinear
 transforma-
z 1 K (q1 , q2 ) q1 (z1 , z2 )
tion q = Q(z), z = Q1 (q) = = e
, = ,
4 z2 5 q2 q2 z2
T T
q = T (z)z, with T (z) = z1 z2 , where z1 (t) IRm , z2 (t) IRnm ,
0 Inm
and ((z1 , z2 ), z2 ) = z1 for all z in the conguration space. Notice that from
the rank assumption on (q) and due to the procedure to split z into z1 and z2
362 6 Dissipative Physical Systems

(using the implicit function Theorem), the Jacobian T (z) is full-rank. More-
over z2 = q2 where q2 are the n m last components of q. In new coordinates
z one has z1 = x and


z1 (t)

M (z(t))z(t) + C(z(t), z(t)) z(t) + g(z(t)) = (t) +

0

(6.169)

Me (z1 (t))z1 (t) + Ce (z1 (t), z1 (t))z1 (t) + dR
dz1 (t) + ge (z1 (t))+
e





+Ke z1 (t) = z1 (t)
where z1 (t) IRm , M (z) = T (z)T M (q)T (z), and = T T (z) . In a sense this
coordinate change splits the generalized coordinates into normal direction
z1 and tangential direction z2 , similarly as in Subsection 6.7.1. The virtual
work principle tells us that z T Fz = z1 z1 for all virtual displacement z,
hence the form of Fz in (6.169) where the principle of mutual actions clearly
appears. The original system may appear as having n + m degrees of freedom.
However since the two subsystems are assumed to be bilaterally coupled, the
number of degrees of freedom is n. This is clear once the coordinate change in
(6.169) has been applied. The system in (6.168) once again has a cascade form
where the interconnection between both subsystems is the contact interaction
force.
Remark 6.55. An equivalent representation as two passive blocks is shown in
Figure 6.6. As an exercise one may consider the calculation of the storage
functions associated to each block.

Fig. 6.6. Manipulator in bilateral contact with a dynamical passive obstacle

 6.8 Nonsmooth Lagrangian Systems 363

Dissipativity Properties

Let us assume that the potential energy terms Ug (z) and Uge (z1 ) are bounded
from below. This assumption is clearly justied by the foregoing developments
on passivity properties of Euler-Lagrange systems. Now it is an evident choice
that the suitable supply rate is given by ( T + FzT )z Tz1 z1 . Notice that
although one might be tempted to reduce this expression to T z since FzT z =
Tz1 z1 , it is important to keep it since they do represent the outputs and
inputs of dierent subsystems: one refers to the controlled system while the
other refers to the uncontrolled obstacle. Let us calculate the available storage
of the total system in (6.169):

t ) T *
Va (z, z) = sup ( + FzT )z Tz1 z1 ds
:(0,z(0),z(0)) 0

(6.170)
= 12 z T (0)M (z(0))z(0) + 12 z1T (0)Me (z1 (0))z1 (0)

+ 12 z1T (0)Ke z1 (0) + Ug (z(0)) + Uge (z1 (0))

Hence the system is dissipative since Va () is bounded for bounded state.

Since we introduced some Rayleigh dissipation in the environment dynamics,
the system has some strict passivity property.

6.8 Nonsmooth Lagrangian Systems

The material in this section may be seen as the continuation of what we
exposed in Chapter 3, Sections 3.9.4 and 3.9.5. The notation is the same.

6.8.1 Systems with C 0 Solutions

Let us introduce a class of nonsmooth Lagrangian systems, which are me-

chanical systems subject to some nonsmooth friction forces. Let : IRl
IR {+} be a convex proper and lower semicontinuous function. Let
M = M T > 0 IRnn , C IRnn , K RRnn , H1 IRnl , H2 IRln
be constant matrices. For (t0 , q0 , q0 ) IR IRn IRn , with H2 q0 D(),
we consider the problem [3]: Find a function q : t
q(t) (t t0 ) with
q C 1 ([t0 , +); IRm ) such that
(i) q() L,e ([t0 , +); IRm )
(ii) q() is right-dierentiable on [t0 , +)
(iii) q(t0 ) = q0 , q(t0 ) = q0

H2 q(t) dom() (6.171)

for all t t0
364 6 Dissipative Physical Systems

M q(t) + C q(t) + Kq(t) H1 (H2 q(t)) (6.172)

a.e. on [t0 , +).
We recall that dom() denotes the domain of the subdierential of the
convex function (). The term H1 (H2 ) is supposed to model the unilat-
erality of the contact induced by friction forces (for instance the Coulomb
friction model). Unilaterality is not at the position level as it is in the next
section, but at the velocity level. This is important because it means that the
solutions are much more gentle. Notice that if the system is considered as
a rst order dierential system, then as the section title indicates solutions
(q(), q()) are time continuous.
Theorem 6.56. [3] Suppose that
(a) There exists a matrix R = RT IRnn , nonsingular, such that

R2 H2T = M 1 H1 (6.173)
(b) There exists y0 = H2 R1 (x0 IRn ) at which () is nite and con-
tinuous.
Let t0 IR, q0 , q0 IRn with H2 q0 dom(). Then there exists a
unique q C 1 ([t0 , +); IRm ) satisfying conditions (i) (ii) (iii) and (6.171)
and (6.172).

We do not go into the details of the proof of this well-posedness result.

Let us just mention that thanks to the existence of the matrix R one can
perform a variable change z = Rq which allows one to rewrite the system as
a rst-order system

x(t) + Ax(t) (x)
(6.174)

x(t0 ) = x0

0nn In 
with A = , x = z . The function : IRn
z
RM 1 KR1 RM 1 CR1
IR {+} is proper, convex and lower semicontinuous and is dened as
(x) = (z), with (z) = ( H2 R1 )(z). The well-posedness of the system
in (6.174) can be shown relying on a Theorem quite similar to Theorem 3.80,
with a slight dierence as the variational inequality that concerns (6.174) is
of the form

x(t) + Ax(t), v x(t) + (v) (x(t)) 0, v IRn , a.e. in [t0 , +)

(6.175)
 6.8 Nonsmooth Lagrangian Systems 365

This reduces to (3.213) if one chooses () as the indicator function of a

convex set K, and is in turn equivalent to an unbounded dierential inclusion.
Indeed one has
M u + q, v u + (v) (u) 0, v IRn

 (6.176)

M u + q (u)
for any proper, convex, lower semicontinuous function with closed domain,
M IRnn , q IRn . The stability analysis of these mechanical systems will
be led in Section 7.2.5.

6.8.2 Systems with BV Solutions

We deal in this section with mechanical Lagrangian systems, subject to fric-
tionless unilateral constraints on the position and impacts. This material is
necessary to study the stability issues as will be done in Section 7.2.4. There
are some tools from convex analysis which have already been introduced in
Section 3.9.4, and which are useful in this setting as well. More precisely, let
us consider the following class of unilaterally constrained mechanical systems


M (q(t))q(t) + F (q(t), q(t)) = h(q(t))(t)






q(0) = q0 , q(0 ) = q0
(6.177)



0 h(q(t)) (t) 0




+
q(tk ) = eq(t
k ) + (1 + e)proxM(q(tk )) [q(tk ), V (q(tk ))]

In (6.177) M (q) = M T (q) > 0 is the n n inertia matrix, F (q, q) =

C(q, q)q + U
q (q) where C(q, q)q denotes centripetal and Coriolis generalized
forces, whereas U (q) is a smooth potential energy from which conservative
forces derive, and h() : IRn
IRm . We assume that h(q0 ) 0. The set
V (q) is the tangent cone to the set = {q IRn | h(q) 0}; see (3.193):
V (q) = T (q). The impact times are generically denoted as tk , the left-limit
q(t +
k ) V (q(tk )) whereas the right-limit q(tk ) V (q(tk )). The third line
in (6.177) is a collision mapping that relates pre- and post-impact general-
ized velocities, and e [0, 1] is a restitution coecient [327]. The notation
proxM(q) means the proximation in the kinetic metric, i.e. the metric dened
q(t+ )+eq(t )
as xT M (q)y for x, y IRn : the vector k 1+e k is the closest vector to the
pre-impact velocity, inside V (q(tk )) (it can therefore be computed through
a quadratic programme) [358]. In particular the impact law in (6.177) im-
plies that the kinetic energy loss at time tk satises (see [327], [69, p.199,
p.489], [34])
366 6 Dissipative Physical Systems

11e + T  
TL (tk ) = q(tk ) q(t
k) M (q(tk )) q(t+
k ) q(tk ) 0 (6.178)
21+e

Remark 6.57. The formulation of the unilateral constraints in (6.177) does

not encompass all closed domains = {q | h(q) 0}, as simple non-convex
cases with so-called reentrant corners prove [77]. It can be used to describe
admissible domains which are dened either by a single constraint (i.e.
m = 1), or with m < + where convexity holds at nondierentiable points
of the boundary (such sets are called regular [105]). It is easy to imagine
physical examples that do not t within this framework, e.g. a ladder.

Let us note that the tangent cone V (q(t)) is assumed to have its origin
at q(t) so that 0 V (q(t)) to allow for post-impact velocities tangential
to the admissible set boundary . The second line in (6.177) is a set of
complementarity conditions between h(q) and , stating that both these terms
have to remain non-negative and orthogonal one to each other. Before passing
to the well-posedness results for (6.177), let us dene a function of bounded
variation.

Denition 6.58. Let f : [a, b] IR be a function, and let the total variation
of f () be dened as


N
V(x) = sup |f (ti ) f (ti1 )|, (a x b) (6.179)
i=1

where the supremum is taken along all integers N , and all possible choices of
the sequence {ti } such that

a = t0 < t1 < .... < tN = x

The function f () is said of bounded variation on [a, b] if V(b) < +.

One should not confuse BV functions with piecewise continuous functions.

We say that a function f : I J is piecewise continuous if there exists a con-
stant > 0 and a nite partition of I into intervals (ai , bi ), with I = i [ai , bi ],
and bi ai for all i, and f () is continuous on each (ai , bi ) with left limit
at ai and right-limit at bi . Thus piecewise continuous functions are a dier-
ent class of functions. There are well-known examples of continuous functions
which are not BV, like f : x
x sin( x1 ) dened on [0, 1]. Clearly f (0) = 0
but the innite oscillations of f () as x approaches 0 hamper the bounded
variation. In addition, piecewise continuity precludes nite accumulations of
discontinuities. BV functions are such that given any t there exists > 0 such
that the function is continuous on (t, t + ). But this may not be uniform
with respect to t. Denition 6.58 holds whatever the function f (), even if f ()
 6.8 Nonsmooth Lagrangian Systems 367

is not AC. One may consult [357] for more informations on BV functions. One
speaks of local bounded variation (LBV) functions when f : IR IR and f ()
is BV on all compact intervals [a, b]. LBV functions possess very interesting
properties, some of which are recalled below.
T
Assumption 14 The gradients hi (q) = h q (q) are not zero at the contact
congurations hi (q) = 0, and the vectors hi , 1 i m, are independent.
Furthermore the functions h(), F (q, q), M (q) and the systems conguration
manifold are real analytic, and ||F (q, q)||q d(q, q(0)) + ||q||q , where d(, ) is
the Riemannian distance and || ||q is the norm induced by the kinetic metric.

Then the following results hold, which are essentially a compilation of

Proposition 32, Theorems 8 and 10, and Corollary 9 of [34]:
i) Solutions of (6.177) exist on [0, +) such that q() is absolutely contin-
uous (AC), whereas q() is right-continuous of local bounded variation
(RCLBV). In particular the left and right-limits of these functions exist
everywhere.
ii) The function q() cannot be supposed to be everywhere dierentiable.
t a.e.
One has q(t) = q(0) + 0 v(s)ds for some function v() = q(). More-
+ +
over q(t ) = v(t ) and q(t ) = v(t ) [270].
iii) Solutions are unique (however in general they do not depend continuously
on the initial conditions).
iv) The acceleration q is a measure dv, which is the sum of two measures: an
atomic measure da , and a Lebesgue integrable function which we denote
q(), i.e. dv = da +q(t)dt. The atoms correspond to the impact times [358].
See Remark 6.59 for some comments on this decomposition.
v) The setof impact times is countable. In many applications one has
+
da = k0 [q(tk ) q(tk )]tk , where t is the Dirac measure and the
sequence {tk }k0 can be ordered, i.e. tk+1 > tk . However phenomena like
accumulations of left-accumulations of impacts may exist (at least bounded
variation does not preclude them). In any case the ordering may not be
possible. This is a sort of complex Zeno behaviour (2 ). In the case of elastic
impacts (e = 1) it follows from [34, Prop.4.11] that tk+1 tk > 0 for
some > 0. Hence solutions are piecewise continuous in this case.
vi) Any quadratic function W () of q is itself RCLBV, hence its derivative is a
measure dW [358]. Consequently dW 0 has a meaning and implies that
the function W () does not increase [127, p.101].
These results enable one to lead a stability analysis safely. Let us now intro-
duce a new formulation of the dynamics in (3.205), which can be written as
the following Measure Dierential Inclusion (MDI): [358]

2
I.e. all phenomena involving an innity of events in a nite time interval, and
which occur in various types of hybrid systems like Filippovs inclusions, etc.
368 6 Dissipative Physical Systems

M (q(t))dv F (q(t), v(t+ ))dt V (q(t)) (w(t)) (q(t)) (6.180)

v(t+ )+ev(t )
where w(t) = 1+e V (q(t)) from (6.177). If e = 0 then w(t) = v(t+ ),
v(t+ )+v(t )
if e = 1 then w(t) = 2 . Moreover when
v() is continuous then
w(t) = v(t). The term MDI has been coined by Moreau, and (6.180) may
also be called Moreaus second order sweeping process. The inclusion in the
right-hand-side of (6.180) is proved as follows: for convenience let us rewrite
the following denitions for a convex set :

N (q) = {z | z T 0, V (q)} (6.181)

which precisely means that the normal cone is the polar cone of the tangent
cone (see Denition 3.67), and

V (q) (w) = {z | z T ( q) 0, V (q)} (6.182)

Since V (q) as it is dened in (3.193) (replace K by ) is a cone and since
q V (q), one can choose = + q with V (q) as a particular value for
. Thus if z V (q) (w) one gets z T z T q(t+ ) and introducing = + q,
one gets z T 0 so that z N (q). Therefore Moreaus inclusion in (6.180)
is proved.

Let us note that the cones are to be understood as being attached to the
same origin in the inclusion. Moreover some natural identications between
spaces (the dual Tq Tq Q at q of the tangent space Tq Q at q to the conguration
space Q, and the cotangent space Tq Q) have been made, thanks to the linear
structure of these spaces in which the cones V (q) () and N (q) are dened.
This allows one to give a meaning to the inclusion in (6.180). This is just a
generalization of the well-known identication between the space of velocities
and that of forces acting on a particle in a three-dimensional space, which
both are identied with IR3 . More details are in [34] and [358].

What happens at impact times? It is well-known in Mechanics that the

dynamics become algebraic at an impact time [69]. Such is the case for the
measure dierential inclusion in (6.180). Let x and z be two vectors of a linear
Euclidean space E, V be a closed convex cone of E, and N be the polar cone
to V . Then from Moreaus Lemma of the two cones [210, p.51] [69, lemma
D1], one has (x z) V (x) x = prox[V, z] z x = prox[N, z].
Times tk are atoms of the measure dv in (6.180). Via a suitable base change,
the kinetic metric at an impact time can be considered as a Euclidean metric
since q() is continuous at tk , and in particular all the identications between
various dual spaces can be done. One gets from (6.180): M (q(tk ))[q(t+ k)
q(t
k )] V (q(t k )) (w(t +
k )) q(t +
k ) + e q(t
k ) = prox M(q(tk )) [V (q(t k )), (1 +
e)q(t
k )] q(t +
k ) + e q(t
k ) = (1 + e)prox M(q(tk )) [V (q(t k )), q(t
k ), where the
second equivalence is proved in [358].
 6.8 Nonsmooth Lagrangian Systems 369

When q(t) is discontinuous, (6.180) implies that Moreaus collision rule

in (6.177) is satised. The term V (q(t)) (w(t)) can be interpreted as a ve-
locity potential. The MDI in (6.180), whose left-hand-side is a measure and
whose right-hand-side is a cone, has the following meaning [327, 357]: there
exists a positive measure d such that both dt and dv possess densities
dt dv
with respect to d, denoted respectively as d () and d (). One also has
dt dt([t,t+])
d (t) = lim0,>0 d([t,t+]) [360], [357, p.9], which shows the link with the
usual notion of a derivative. The choice of d is not unique because the right-
hand-side is a cone [358]. However by the Lebesgue-Radon-Nikodym Theo-
dt dv
rem [419], the densities d () and d () are unique functions for a given d. To
shed some light on this, let us consider for instance d = dt + k0 tk , which
corresponds to applications where the system is subject to impacts at times tk
dt
and otherwise evolves freely. Then d (tk ) = 0 (the Lebesgue measure dt and
the Dirac measure t are mutually singular) whereas d dv
(tk ) = v(t+
k )v(tk ) (tk
is an atom of the measure dv). When t = tk then d dt
(t) = 1 and d dv
(t) = v(t).
Therefore the meaning of (6.180) is that there exists a positive measure
d with respect to which both dt and dv possess densities, and

dv dt
M (q(t)) (t) F (q(t), v(t+ )) (t) V (q(t)) (w(t)) (q(t)) (6.183)
d d
holds dalmost everywhere. In a sense, densities replace derivatives, for mea-
sures. When dealing with measure dierential equations or inclusions, it is
then natural to manipulate densities instead of derivatives. In general one can
choose d = |dv| + dt [357, p.90], where |dv| is the absolute value of dv, or
d = ||v(t)||dt+da , or d = dt+da . It is fundamental to recall at this stage,
that the solution of (6.183) does not depend on this choice. For instance, if
1 q(t)
d = ||v(t)||dt + da then for all t = tk , d
dt
(t) = ||v(t)|| dv
and d (t) = ||v(t)|| .
Whereas if d = dt + da then for all t = tk , d (t) = 1 and d (t) = q(t).
dt dv

Remark 6.59. The above mathematical framework is more than just a mathe-
matical fuss. Indeed as noted in [358], introducing the velocity into the right-
hand-side of the dynamics as done in (6.180), not only allows one to get a
compact formulation of the nonsmooth dynamics (see Figure 6.7 in this re-
spect), but it also paves the way towards the consideration of friction in the
model. In turn it is clear that introducing friction, is likely to complicate the
dynamics. Especially the above framework paves the way towards more com-
plex cases where the measure dv may contain a third term dna which is a
nonatomic measure singular with respect to the Lebesgue measure dt (assump-
tion 14 implies that dna = 0 [34]). In summary the dynamics in (6.183) is rich
enough to encompass complex behaviours involving solutions which may be
far from merely piecewise continuous. This is a consequence of replacing func-
tions by the more general notion of measure, at the price of a more involved
model. In fact using measures allows one to encompass somewhat complex
370 6 Dissipative Physical Systems

Zeno behaviours occurring in unilaterally constrained mechanical systems in

a rigorous manner.

Dissipative Systems Interpretation

Let us end this section with a dissipativity interpretation of Moreaus inclu-
sion in (6.180). The dynamics in (6.180) has the interpretation as in Figure
6.7, where V (q(t)) (w(t)). Since V (q(t)) (w(t)) N (q) = V
(q) (the
cone polar to V (q)), the feedback loop in Figure 6.7 contains the cone com-
plementarity problem (or complementarity problem over a cone [113, p.31])

N (q) V (q(t)) (w(t))  w(t) V (q) (6.184)

When m = 1 and q , one has V (q) = IR and N (q) = IR in a
+

suitable frame attached to q, and the graph of the multivalued mapping is the
so-called corner law. In general this is an example of an m-dimensional mono-
tone multivalued mapping w(t)
. It is noteworthy that the feedback loop
in Figure 6.7 contains both the complementarity conditions and the collision
mapping in (6.177). A quite similar structure can be found for the dynamics
in (3.198).

Fig. 6.7. Unilaterally constrained Lagrangian system

This interpretation of the dynamics motivates us to search for a dissipa-

tion equality applying to (6.177), with a supply rate handling both continuous
 6.8 Nonsmooth Lagrangian Systems 371

and discontinuous motions. This will be done in Section 7.2.4 when the ma-
nipulations leading the stability analysis have been presented.
7
Passivity-based Control

This chapter is devoted to investigating how the dissipativity properties of

the various systems examined in the foregoing chapter can be used to de-
sign stable and robust feedback controllers. We start with a classical result
of mechanics, which actually is the basis of Lyapunov stability and Lyapunov
functions theory. The interest of this result is that its proof hinges on im-
portant stability analysis tools, and allows one to make a clear connection
between Lyapunov stability and dissipativity theory. The next section is a
brief survey on passivity-based control methods, a topic that has been the
object of numerous publications.

7.1 Brief Historical Survey

The fundamental works on dissipative systems and positive real transfer func-
tion which are exposed in the foregoing chapters have been mainly motivated
by the stability and stabilization of electrical networks. It is only at the begin-
ning of the 1980s that work on mechanical systems and the use of dissipativity
in their control started to appear, with the seminal paper by Takegaki and
Arimoto [484]. Roughly speaking, two classes of feedback controllers have
emerged:
Passivity-based controllers: the control input is such that the closed-
loop system can be interpreted as the negative interconnection of two
dissipative subsystems. The Lyapunov function of the total system is close
to the process total energy, in the sense that it is the sum of a quadratic
function 12 T M (q) for some depending on time, generalized positions
q and velocities q, and a term looking like a potential energy. Sometimes
additional terms come into play, like in adaptive control where the on-line
estimation algorithm provides supplementary state variables. Such algo-
rithms have been motivated originally by trajectory tracking and adaptive
374 7 Passivity-based Control

motion control of fully actuated robot manipulators. The machinery be-

hind this is dissipative systems and Lyapunov stability theory. This chapter
will describe some of these schemes in great detail, consequently we do not
insist on passivity-based controllers in this short introduction.
Controlled Lagrangian (or Hamiltonian): the objective is not only
to get a two-block dissipative interconnection, but also to preserve a La-
grangian (or a Hamiltonian) structure in closed-loop. In other words, the
closed-loop system is itself a Lagrangian (or a Hamiltonian) systems with
a Lagrangian (or Hamiltonian) function, and its dynamics can be derived
from a variational principle such as Hamiltons principle. In essence, one
introduces a feedback that changes the kinetic energy tensor M (q). Dif-
ferential geometry machinery is the underlying tool. The same applies to
port-Hamiltonian systems which we saw in Chapter 6. Regulation tasks for
various kind of systems (mechanical, electromechanical, underactuated)
have been the original motivations of such schemes. The method is de-
scribed in Section 7.9.
Related terms are potential energy shaping, energy shaping, damping in-
jection or assigment, energy balancing. The very starting point for all those
methods, is the Lagrange-Dirichlet (or Lejeune-Dirichlet) Theorem which is
described in Section 7.2. It is dicult to make a classication of the numerous
schemes that have been developed along the above two main lines. Indeed this
would imply to highlight the discrepancies between:
Trajectory tracking vs regulation
Full actuation vs underactuation
Fixed parameters vs adaptive control
Static feedback vs dynamic feedback
Smooth systems vs nonsmooth systems
Constrained systems vs unconstrained systems
Rigid systems vs exible systems
etc.
We will therefore rather present the contributions in a chronological order,
as they appeared in the literature. As said above, the starting point may be
situated in 1981 with [484]. The challenge then in the Systems and Control
and the Robotics communities was about nonlinear control of fully actuated
manipulators for trajectory tracking purpose, and especially the design of
a scheme allowing for parameter adaptive control. The rst robot adaptive
control algorithms were based on tangent linearization techniques [216]. Then
two classes of schemes emerged: those requiring an inversion of the dynamics
and acceleration measurement or inversion of the inertia matrix M (q) [7, 114,
115, 477], and those avoiding such drawbacks [215, 255, 352, 425427, 461463].
Despite the fact that they were not originally designed with dissipativity in
mind, the schemes of the second class were all proved to belong to passivity-
based schemes in [71] (the schemes in [425,461] were proved to be hyperstable
 7.2 The Lagrange-Dirichlet Theorem 375

in [275], while the term passivity-based was introduced in [384]). Then many
schemes have been designed, which more or less are extensions of the previous
ones but adapted to constrained systems, systems in contact with a exible
environment, etc.
The next step, as advocated in [384], was to solve the trajectory tracking
problem in the adaptive control context, for exible joint robots. This was
done in [70, 72, 316, 318], using what has been called afterwards backstepping,
together with a specic parametrization to guarantee the linearity in the un-
known parameters, and a dierentiable parameter projection. The adaptive
control of exible joint manipulators is a non-trivial problem combining these
three ingredients. See [78] for further comparisons between this scheme and
schemes designed with the backstepping approach, in the xed parameters
case. Almost at the same time the regulation problem with passivity-based
control of induction motors was considered in [385, 386], using a key idea
of [316,318]. The control of induction motors then was a subject of excitation
for several years.
Later came controlled Lagrangian and Hamiltonian methods as developed
by Bloch, Leonard, Mardsen [54, 55] and in [52, 387], to cite a few.

7.2 The Lagrange-Dirichlet Theorem

In this section we present a stability result that was rst stated by Lagrange
in 1788 and subsequently proved rigorously by Dirichlet. It provides sucient
conditions for a conservative mechanical system to possess a Lyapunov stable
xed point. The case of Rayleigh dissipation is also presented. The develop-
ments base on the dissipativity results of Chapter 4.

7.2.1 Lyapunov Stability

Let us consider the Euler-Lagrange dynamics in (6.1) or that in (6.43). Let
us further make the following:
Assumption 15 The potential energy U (q) is such that i) dU
dq (q) =0q=
2
d U
q0 and ii) dq2 (q0 ) > 0.
In other words U (q) is locally convex around q = q0 and q0 is a critical point
of the potential energy. Hence the point (q0 , 0) is a xed point of the dynamics
in (6.1). Then it follows that the considered system with input , output q
and state (q q0 , q), is zero-state observable (see Denition 5.12). Indeed if
0 and q 0, it follows from (6.43) that g(q) = dU dq = 0, hence q = q0 . The
following is then true:
Theorem 7.1 (Lagrange-Dirichlet). Let Assumption 15 hold. Then the
xed point (q, q) = (q0 , 0) of the unforced system in (6.1) is Lyapunov sta-
ble.
376 7 Passivity-based Control

Proof: First of all notice that the local (strict) convexity of U (q) around
q0 precludes the existence of other q arbitrarily close to q0 and such that
dU
dq (q1 ) = 0. This means that the point (q0 , 0) is a strict local minimum for the
total energy E(q, q). We have seen that E(q, q) is a storage function provided
that U (q) remains positive. Now it suces to dene a new potential energy
as U (q) U (q0 ) to full this requirement, and at the same time to guarantee
that the new E(q, q) satises E(0, 0) = 0, and is a positive denite function
(locally at least) of (0, 0). Since this is a storage function, we deduce from the
dissipation inequality (which is actually here an equality) that for all 0
one gets
t
E(0) = E(t) T (s)q(s)ds = E(t) (7.1)
0
Therefore the xed point of the unforced system is locally Lyapunov stable.
Actually we have just proved that the system evolves on a constant energy
level (what we already knew) and that the special form of the potential energy
implies that the state remains close enough to the xed point when initialized
close enough to it. Notice that (7.1) is of the type (4.75) with S(x) = 0: the
system is lossless. All in all, we did not make an extraordinary progress. Before
going ahead with asymptotic stability, let us give an illustration of Theorem
7.1.

Example 7.2. Let us consider the dynamics of planar two-link revolute joint
manipulator with generalized coordinates the link angles (q1 , q2 ) (this nota-
tion is not to be confused with that employed for the exible joint-rigid link
manipulators). We do not need here to develop the whole stu. Only the
potential energy is of interest to us. It is given by

U (q) = a1 sin(q1 ) + a2 sin(q1 + q2 ) (7.2)

where a1 > 0 and a2 > 0 are constant depending on masses, dimensions

dU a 1 cos(q1 ) + a2 cos(q1 + q2 )
and gravity. It is easy to see that dq = =
 a2 cos(q1 + q2 )
0
implies that q1 + q2 = (2n + 1) 2 and q1 = (2m + 1) 2 for n, m
0
IN . In particular q1 = 2 and q2 = 0 (i.e. n = m = 1) is a point that
satises the requirements
 of assumption 15. One computes that at this point
2
d U a 1 + a 2 a 2
dq2 = a2 a2
that is positive denite since it is symmetric and its
determinant is a1 a2 > 0. Intuitively one notices that global stability is not
possible for this example since the unforced system possesses a second xed
point when q1 = 2 , q2 = 0, which is not stable.

7.2.2 Asymptotic Lyapunov Stability

Let us now consider the dynamics in (6.31). The following is true:

 7.2 The Lagrange-Dirichlet Theorem 377

Lemma 7.3. Suppose that Assumption 15 holds. The unforced Euler-Lagrange

dynamics with Rayleigh dissipation satisfying q t R q q q for some > 0,
T

possesses a xed point (q, q) = (q0 , 0) that is locally asymptotically stable.

Proof: It is not dicult to prove that the dynamics in Denition 6.12 de-
nes an OSP system (with the velocity q as the output signal). Therefore
the system now denes as well an output strictly passive operator
q.
We could directly compute the derivative of E(q, q) along the trajectories of
the unforced system to attain our target. Let us however use passivity. We
know (see Remark 4.86) that the dissipation inequality is equivalent to its
innitesimal form, i.e.

dV T R
f (x, ) = T x2 xT2 (7.3)
dx x2
 
x1 q
where x = = , f (x, u) denotes the system vector eld in state
x2 q
space notations, and V (x) is any storage function. Let us take V (x) = E(q, q).
We deduce that
dE T
E = f (x, 0) = q T q (7.4)
dx
The only invariant set inside the set {(q, q) : q 0} is the xed point (q0 , 0).
Resorting to Krasovskii-La Salle Invariance Theorem one deduces that the
trajectories converge asymptotically to this point, provided that the initial
conditions are chosen in a suciently small neighborhood of it. Notice that
we could have used Corollary 5.16 to prove the asymptotic stability.

Remark 7.4. Convexity: Convex properties at the core of stability in me-

chanics: in statics the equilibrium positions of a solid body lying on a
horizontal plane, submitted to gravity, are characterized by the condition
that the vertical line that passes by its center of mass crosses the convex
hull of the contact points of support. In dynamics Assumption 15 shows
that without a convexity property (maybe local), the stability of the xed
point is generically impossible to obtain.
The Lagrange-Dirichlet Theorem also applies to constrained Euler-Lagrange
systems as in (6.163). If Rayleigh dissipation is added and if the poten-
tial energy satises the required asumptions, then the (z2 , z2 ) dynamics
are asymptotically stable. Thus z2 (t) tends towards zero as well so that
z1 (t) = g1 (z2 (t)) as t +.
It is clear that if Assumption 15 is strengthened to having a potential
energy U (q) that is globally convex, then its minimum point is globally
Lyapunov stable.
Other results generalizing the Lagrange-Dirichlet Theorem for systems
T
with cyclic coordinates (i.e. coordinates such that q i
(q) = 0) were given
by Routh and Lyapunov; see [351].
378 7 Passivity-based Control

Remark 7.5. It is a general result that output strict passivity together with
zero-state detectability yields under certain conditions asymptotic stability;
see Corollary 5.16. One basic idea for feedback control may then be to nd a
control law that renders the closed-loop system strictly output passive with
respect to some supply rate, and such that the closed-loop operator is zero-
state detectable with respect to the considered output.
Example 7.6. Let us come back on the example in section 6.5. As we noted
the concatenation of the two functions in (6.119) and (6.120) yields a positive
denite function of (q, q) = (0, 0) with q = q 11+k
qd
, that is continuous at
q = 0. The only invariant set for the system in (6.115) with the input in (6.116)
is (q, q) = ( 11+k
qd
, 0). Using the Krasovskii-La Salle invariance Theorem one
concludes that the point q = 0, q = 0 is globally asymptotically uniformly
Lyapunov stable.

7.2.3 Invertibility of the Lagrange-Dirichlet Theorem

One question that comes to ones mind is that, since the strong assumption
on which the Lagrange-Dirichlet Theorem relies is the existence of a minimum
point for the potential energy, what happens if U (q) does not possess a min-
imum point? Is the equilibrium point of the dynamics unstable in this case?
Lyapunov and Chetaev stated the following:
Theorem 7.7. (a) If at a position of isolated equlibrium (q, q) = (q0 , 0) the
potential energy does not have a minimum, and, neglecting higher order terms,
it can be expressed as a second order polynomial, then the equilibirum is un-
stable. (b) If at a position of isolated equilibrium (q, q) = (q0 , 0) the potential
energy has a maximum with respect to the variables of smallest order that oc-
cur in the expansion of this function, then the equilibrium is unstable. (c) If
at a position of isolated equilibrium (q, q) = (q0 , 0) the potential energy, which
is an analytical function, has no minimum, then this xed point is unstable.
2
1
Since U (q) = U (q0 ) + dU dq (q0 )(q q0 ) + 2 (q q0 ) dq2 (q q0 ) + o[(q
T d U

q0 )T (q q0 )], and since q0 is a critical point of U (q), the rst item tells us
2
that the Hessian matrix ddqU2 is not positive denite, otherwise the potential
energy would be convex and hence the xed point would be a minimum. With-
out going into the details of the proof since we are interested in dissipative
systems, not unstable systems, let us note that the trick consisting of reden-
ing the potential energy as U (q) U (q0 ) in order to get a positive storage
function no longer works. Moreover, assume there is only one xed point for
the dynamical equations. It is clear at least in the one degree-of-freedom case
2
that if ddqU2 (q0 ) < 0 then U (q) for some q. Hence the available stor-
age function that contains a term equal to sup [U (q(t))]t0 cannot be
:(0,q(0),q(0))
bounded, assuming that the state space is reachable. Thus the system cannot
be dissipative, see Theorem 4.41.
 7.2 The Lagrange-Dirichlet Theorem 379

7.2.4 The Lagrange-Dirichlet Theorem for Nonsmooth Lagrangian

Systems (BV Solutions)

Let us consider the class of Lagrangian systems as in Section 6.8.2, i.e.

fully actuated Lagrangian systems with complementarity conditions and im-
pacts. The constraints are supposed to be frictionless. First notice that since
q and 0 V (q), xed points of (6.180) satisfy the generalized
F (q, 0) = U
equation 0 (q
) + U
q (q ) which in particular implies q . Conditions

under which such a generalized equation possess at least one solution, and
numerical algorithms to compute one solution, exist [169]. In the following we
shall assume for simplicity that the solutions are isolated, or even more: that
it is unique.
Lemma 7.8. Consider a mechanical system as in (6.177). Assume that the
potential function U (q) is radially unbounded. Then if (q) + U (q) has a
strict global minimum at q , the equilibrium point (q , 0) is globally Lyapunov
stable.
Let us note that needs not be convex in general (for instance the equi-
librium may exist in Int(), or it may belong to but be forced by the con-
tinuous dynamics; see Figure 7.1 for planar examples with both convex and
non-convex ; it is obvious that in the depicted non-convex case all points
(q , 0) with q are xed points of the dynamics).

Fig. 7.1. Convex and nonconvex admissible sets

Proof: The proof may be led as follows. Let us consider the nonsmooth Lya-
punov candidate function
1 T
W (q, q) = q M (q)q + (q) + U (q) U (q ) (7.5)
2
380 7 Passivity-based Control

Since the potential (q) + U (q) has a strict global minimum at q equal to
U (q ) and is radially unbounded, this function W () is positive denite on
the whole state space and is radially unbounded. Also W (q, q) (||q||, ||q||)
for some class K function () is satised on ( q(t) for all t 0). The
potential function (q) + U (q) is continuous on . Thus W (q, q) in (7.5)
satises the requirements of a Lyapunov function candidate on , despite the
indicator function has a discontinuity on (but is continuous on the closed
set ; see (3.187)). Moreover since (6.180) secures that q(t) for all t 0,
it follows that (q(t)) = 0 for all t 0. In view of this one can safely discard
the indicator function in the subsequent stability analysis. Let us examine
the variation of W (q, q) along trajectories of (6.183). In view of the above
discussion, one can characterize the measure dW by its density with respect
to d and the function W decreases if its density dW d (t) 0 for all t 0.
We recall Moreaus rule for dierentiation of quadratic functions of RCLVB
functions [357, pp.8-9]: let u() be RCLBV, then d(u2 ) = (u+ + u )du where
u+ and u are the right-limit and left-limit functions of u(). Let us now
compute the density of the measure dW with respect to d:
1
[q(t+ ) + q(t )] M (q(t)) d
dW T dv U dq
d (t) = 2 (t) + q d (t)
(7.6)
 + T  dq
+ 12 q

q(t ) M (q(t))q(t+ ) d (t)
where dq = v(t)dt since the function v() is Lebesgue integrable. Let us now
dt dq dv
choose d = dt + da . Since d (tk ) = 0 and d (tk ) = 0 whereas d (tk ) =
+ +
v(tk ) v(tk ) = q(tk ) q(tk ), it follows from (7.6) that at impact times one
gets

dW 1 + T  
(tk ) = q(tk ) + q(t
k) M (q(t)) q(t+
k ) q(tk ) = TL (tk ) 0 (7.7)
d 2
where TL (tk ) is in (6.178). Let the matrix function M (q, q) be dened by
d
M (q(t), q(t)) = dt M (q(t)). Let us use the expression of F (q, q) given after
(6.177), and let us assume that Christoels symbols of the rst kind are used
  T
to express the vector C(q, q)q = M (q, q) 12 q

q T M (q(t))q . Then the
matrix M (q, q) 2C(q, q) is skew-symmetric; see Lemma 6.16. Now if t = tk ,
dv dt
one gets d (t) = v(t) = q(t) and d (t) = 1 [357, p.76] and one can calculate
from (7.6), using the dynamics and the skew-symmetry property (see Lemma
6.16):
dW
d = dW
dt = q T C(q, q)q + 12 q T M (q, q)q q T z1
(7.8)
= q T z1
where z1 V (q(t)) (w(t)) and W () is dened in (7.5). To simplify the
notation we have dropped arguments in (7.8), however q is to be understood
 7.2 The Lagrange-Dirichlet Theorem 381

as q(t) = q(t+ ) since t = tk . Now since for all t 0 one has q(t+ ) V (q) [358]
which is polar to (q(t)), and from Moreaus inclusion in (6.180), it follows
that z1T q(t+ ) 0. Therefore the measure dW is non-positive. Consequently
the function W () is non-increasing [127, p.101]. We nally notice that the
velocity jump mapping in (6.177) is a projection and is therefore Lipschtiz
continuous as a mapping q(t +
k )
q(tk ), for xed q(tk ). In particular it is
continuous at (q
, 0), so that a small pre-impact velocity gives a small post-
impact velocity. All the conditions for Lyapunov stability of (q
, 0) are fullled
and Lemma 7.8 is proved.

The main feature of the proof is that one works with densities (which
are functions of time) and not with the measures themselves, in order
to characterize the variations of the Lyapunov function.

Remark 7.9. The above result holds also locally thanks to the continuity
property of the impact mapping in (6.177).
The inclusion of the indicator function (q(t)) in the Lyapunov func-
tion not only guarantees its positive deniteness (which anyway is assured
along solutions of (6.183) which remain in ), but it also allows one to
consider cases where the smooth potential has a minimum that is outside
. Saying (q) + U (q) has a strict minimum at q is the same as saying
U (q) has a strict minimum at q inside . Since the indicator function
has originally been introduced by Moreau as a potential associated to uni-
lateral constraints, it nds here its natural use. In fact we could have kept
the indicator function in the stability analysis. This would just add a null
term q(t+ )T z2 d
dt
(t) in the right-hand-side of (7.6), with z2 (q(t)).
As alluded to above, taking e = 1 in (6.177) ensures that there is no
accumulation of impacts, thus the sequence of impact times {tk }k0 can be
ordered, da = k0 tk , and velocities are piecewise continuous. Then a
much simpler formulation can be adopted by separating continuous motion
phases occurring on intervals (tk , tk+1 ) from impact times. The system is
therefore non-Zeno for e = 1 and if Assumption 14 holds.
One doesnt need to make further assumptions on the measure da to
conclude, and one sees that this conclusion is obtained directly applying
general dierentiation rules of RCLBV functions. The dynamics might
even contain dense sets of velocity discontinuities, (7.6) and (7.7) would
continue to hold. This shows that using the MDI formalism in (6.180) or
(6.183) places the stability analysis in a much more general perspective
than, say, restricting q() to be piecewise continuous.
Other work on energy-based control of a class of nonsmooth systems may
be found in [188, 195].
382 7 Passivity-based Control

A Dissipation Inequality

Let us now derive a dissipation inequality for the dynamical system (6.177). To
that end let us take advantage of the compact formalism (6.183). We consider
a Lebesgue measurable input () so that (6.183) becomes

dv dt dt
M (q(t)) (t) F (q(t), v(t+ )) (t) (t) V (q(t)) (w(t)) (7.9)
d d d

Following (6.184) let denote a measure that belongs to the normal cone
to the tangent cone V (q(t)) (w(t)), and let us denote dR
d () its density with
respect to . The system in (7.9) is dissipative with respect to the generalized
supply rate
1 dt dR
 (v(t+ ) + v(t )), (t) + (t) (7.10)
2 d d
Noting that = h(q) for some measure we obtain
1 dt d
 (v(t+ ) + v(t )), (t) + h(q) (t) (7.11)
2 d d
where we recall that v() satises the properties in item ii) in Section 6.8.2 and
dt
that outside impacts (i.e. outside atoms of the measure dR) one has d =0
because the Lebesgue measure has no atom. It is noteworthy that (7.11) is
a generalization of the Thomson-Taits Formula of Mechanics [69, 4.2.12],
which expresses the work performed by the contact forces during an impact.
The supply rate in (7.11) may be split into two parts: a function part and
a measure part. The function part describes what happens outside impacts,
and one has 12 (v(t+ ) + v(t ) = v(t) = q(t). The measure part describes what
happens at impacts tk . Then one gets

(v(t+ + +
k ) + v(tk )), h(q) d (tk ) = (v(tk ) + v(tk )), M (q(tk )(v(tk ) v(tk ))
d

= v T (t+ + T
k )M (q(tk ))v(tk ) v (tk )M (q(tk ))v(tk ) = 2TL (tk ) 0
(7.12)
where we used the fact that the dynamics at an impact time is algebraic:
M (q(tk ))(v(t+
k ) v(tk )) = h(q) d (tk ) with a suitable choice of the basis
d

measure . The storage function of the system is nothing else but its total
energy. It may be viewed as the usual smooth energy 12 q T M (q)q + U (q), or as
the unilateral energy 12 q T M (q)q + U (q) + (q), which is nonsmooth on IRn
IRn . It is worth remarking, however, that the nonsmoothness of the storage
function is not a consequence of the impacts, but of the complementarity
condition 0 h(q) 0.
 7.2 The Lagrange-Dirichlet Theorem 383

Further Reading and Discussion

The foregoing developments concern a specic class of nonsmooth dynamical

systems involving state jumps and measures. Other classes of systems with
impulsive terms exist, which can be written as


x(t) = F (x(t), t) if t = tk



x(t+ ) x(t ) = S(x(t )) if t = tk (7.13)





x(0 ) = x0
where some assumptions are made on the set of times tk , see for instance
[183,184,186,189]. Such assumptions always make the set of state jump times,
a very particular case of the set of discontinuities of a LBV function. It is
noteworthy that the systems in (7.13) and in (6.177) are dierent dynam-
ical systems. Most importantly the complementarity conditions are absent
from (7.13). Another class of impulsive systems is that of measure dierential
equations (MDE), or impulsive ODEs. Let us consider one example:

 
5 3
x(t) = sin x(t) + + cos x(t) + u(t), x(0 ) = x0 , x(t) IR
4 4
(7.14)
where u() is of bounded variation. Applying [65, Theorem 2.1], this MDE has
a unique global generalized solution. Consider now

 5
  3

x(t) = sin x(t) + 4 + cos x(t) + 4 (t), x(0 ) = x0 , x(t) IR

0 x(t) (t) 0
(7.15)
Suppose that x0 = 0. Then if (0) = 0 one gets x(0) = sin( 5 4 ) < 0. It
is necessary that there exists a (0) > 0 such that x(0) 0. However since
cos( 3
4 ) < 0, this is not possible and necessarily x(0) < 0. If x0 < 0, then an
initial jump must occur and x(0+ ) 0. If x(0+ ) = 0 the previous analysis
applies. One sees that dening generalized solutions as in [65, Denition 2.1]
is not sucient. Therefore the complementarity system in (7.15) is not well-
posed, despite its resemblance with the MDE in (7.14). One notices that
the class of nonsmooth Euler-Lagrange systems considered for instance in
[194, 195] and in (6.177) are, in the same way, dierent classes of nonsmooth
dynamical systems (the discrepancy being the same as the one between (7.13)
and (7.15)). In other words, the considered models are not the same, since the
models in [194, 195] do not incorporate the complementarity conditions.
384 7 Passivity-based Control

7.2.5 The Lagrange-Dirichlet Theorem for Nonsmooth Lagrangian

Systems (C 0 Solutions)

Let us now pass to the stability analysis of the systems presented in Section
6.8.1. The set of stationary solutions of (6.171) and (6.172) is given by

W = {q IRm | K q H1 (0)} (7.16)

Denition 7.10. A stationary solution q W is stable provided that for

 ( ) > 0 such that for any q0 IR , q0 IR ,
n n
any > 0 there exists
2 2
H2 q0 D(), with ||q0 q|| + ||q0 || , the solution q(, t0 , q0 , q0 ) of
the problem (i) (ii) (iii) (6.171) (6.172) satises


||q(t, t0 , q0 , q0 ) q||2 + ||q(t, t0 , q0 , q0 )||2 , t t0 (7.17)

We then have the following theorems which we give without proofs.

Theorem 7.11. [3] Let the assumptions of Theorem 6.56 hold, and 0
D(). Suppose in addition that
RM 1 CR1 0
RM 1 KR1 > 0 and is symmetric
Then the set W = and any stationary solution q W of (6.171) and
(6.172) is stable.

A variant is as follows:

Theorem 7.12. [3] Let the assumptions of Theorem 6.56 hold, and 0
D(). Let q W be a stationary solution of (6.171) and (6.172). Suppose
that
RM 1 CR1 z + RM 1 K q, z + (H2 R1 z) (0) 0, z IRn
RM 1 KR1 > 0 and is symmetric
Then q is stable.

The next theorem concerns the attractivity of the stationary solutions,

and may be seen as an extension of the LaSalle invariance principle. Let
d[s, M] = inf mM ||s m|| be the distance from a point s IRn to a set
M IRn .

Theorem 7.13. [3] Let the assumptions of Theorem 6.56 hold, and 0
D(). Suppose that:
RM 1 KR1 > 0 and is symmetric
RM 1 CR1 z + RM 1 K q, z + (H2 R1 z) (0) > 0, z IRn \ {0}
 7.2 The Lagrange-Dirichlet Theorem 385

D() is closed
Then for any q0 IRn , q0 IRn , H2 q0 D(), the orbit

(q0 , q0 ) = {(q(, t0 , q0 , q0 ), q(, t0 , q0 , q0 )) | t0 } (7.18)

is bounded and

lim d[q(, t0 , q0 , q0 ), W] = 0, lim q(, t0 , q0 , q0 ) = 0 (7.19)

+ +

The proof is led with the help of the quadratic function V (x) = 12 (q
q) R2 M 1 K(q q) + 12 q T R2 q. Notice that (q q)T R2 M 1 K(q q) = (q
T

q)T R(RM 1 KR1 )R(q q). More on the attractivity properties of similar
evolution problems can be found in [4].

Example 7.14. We are given the dynamics

mq(t) + cq(t) + kq(t) (q(t)) (7.20)

of a one-degree-of-freedom system acted upon by a spring with stiness k > 0
and with viscous friction c > 0. Coulombs friction is obtained by setting
(z) = |z|. Then W = k , k , and lim + d[q(, t0 , q0 , q0 ), W] = 0,
lim + q(, t0 , q0 , q0 ) = 0. The mass stops somewhere within W, as ex-
pected. Actually one may even expect convergence in nite time. Finite-time
convergence properties for a class of dierential inclusions have been shown
in [5, 91].

7.2.6 Conclusion

These theorems generalize the Lagrange-Dirichlet Theorem for a class of non-

smooth systems. It is worth recalling that the subdierential of a proper con-
vex lower semicontinuous mapping, denes a maximal monotone mapping (see
Section 3.9.4 where some basic results of convex analysis are recalled). The
system in (6.171) and (6.172) can consequently be seen as the feedback inter-
connection of a Lagrangian system and a monotone mapping. Both subsystems
can be described as follows:

M q(t) + C q(t) + Kq(t) = u1 , y1 (t) = q(t)
(7.21)

u2 (t) = q(t), y2 (t) H1 (H2 u1 (t)), y2 (t) = u1 (t)

More precisely the variable change dened in (6.173) allows one to rewrite
the dynamics (6.172) as
386 7 Passivity-based Control

z(t) + RM 1 CR1 z(t) + RM 1 KR1 z(t) (z(t)) (7.22)

1
with (w) = ( H2 R )(w) for all w IR , and
n

(w) = R1 H2T (H2 R1 w)

for all w IRn . It is clear that () is proper convex lower semicontinuous so
that its subdierential denes a maximal monotone mapping. Let the assump-
tions of Theorem 7.11 be fullled. The feedback interconnection is described
as

z(t) + RM 1 CR1 z(t) + RM 1 KR1 z(t) = u1 (t), y1 (t) = z(t)
(7.23)

u2 (t) = z(t), y2 (t) (z(t)), y2 (t) = u1 (t)
and both subsystems are passive. This interpretation together with the one
at the end of section 6.8.2 allow us to conclude that maximal monotone
dierential inclusions permit to nicely recast such nonsmooth systems into a
sound and established framework which extends the usual passivity theorems.

7.3 Rigid JointRigid Link Systems: State Feedback

In this subsection we shall present various feedback controllers that assure
some stability properties for the rigid joint-rigid link model in (6.98). We
start with the regulation problem and then generalize to the tracking case.
In each case we emphasize how the dissipativity properties of the closed-loop
systems constitute the basis of the stability properties.

7.3.1 PD Control
Let us consider the following input:
= 1 q 2 q (7.24)
where 1 > 0 and 2 > 0 are the constant feedback gains (for simplicity we
consider them as being scalars instead of positive denite n n matrices, this
is not very important for what follows), q = q qd , qd IRn is a constant
desired position. The closed-loop system is given by

M (q(t))q(t) + C(q(t), q(t))q(t) + g(q(t)) + 1 q(t) + 2 q(t) = 0 (7.25)

Two paths are possible: we can search for the available storage function of
the closed-loop system in (7.25) which is likely to provide us with a Lyapunov
function, or we can try to interpret this dynamics as the negative interconnec-
tion of two passive blocks and then use the passivity theorem (more exactly
one of its numerous versions) to conclude on stability. To x the ideas we
develop both paths in detail.
 7.3 Rigid JointRigid Link Systems: State Feedback 387

The Closed-loop Available Storage

First of all notice that in order to calculate an available storage we need a

supply rate, consequently we need an input (that will be just an auxiliary
signal with no signicance). Let us therefore just add a term u in the right-
hand-side of (7.25) instead of zero. In other words we proceed as we did for the
example in Section 6.5: we make an input transformation and the new system
is controllable. Let us compute the available storage along the trajectories of
this new input-output system, assuming that U (q) is bounded from below,
i.e. U (q) Umin > for all q Q:

t
Va (q0 , q0 ) = sup uT (s)q(s)ds
u:(0,q0 ,q0 ) 0

t
= sup q T (s) {M (q(s))q(s) + C(q(s), q(s))q(s) + g(q(s))
u:(0,q0 ,q0 ) 0

+1 q(s) + 2 q(s)} ds
,  t
1 t
= sup q T (s)M (q(s))q(s) [U (q(t))]0
u:(0,q0 ,q0 ) 2 0

1 t t T 3
2 2 q T
(s)q(s) 0
1 0 q (s)q(s)ds

= 12 q(0)T M (q(0))q(0) + U (q(0)) + 12 2 q(0)T q(0)

(7.26)
where we used the fact that q T [M (q, q) 2C(q, q)]q = 0 for all q Q and all
q Tq Q; see Lemma 6.16 1 . Let us now make a little stop: we want to show
some stability property for the unforced system in (7.25), so what is the xed
point of this system? Letting q 0 in (7.25) one nds

g(q) + 2 q = 0. (7.27)

Let us state the following:

Assumption 16 The equations in (7.27) possess a nite number of isolated
roots q = qi . Moreover the qi s are strict local minima of U (q).
Then we have the following:
Lemma 7.15. Assume that Assumption 16 is true. The rigid joint-rigid link
manipulator dynamics in (6.98) with PD controller in (7.24) has locally
asymptotically stable xed points (q, q) = (qi , 0).
1
Actually this equality is is always true, even if the matrix M (q, q) 2C(q, q) is
not skew-symmetric.
388 7 Passivity-based Control

Proof: From the second part of Assumption 16 it follows that the available
storage Va in (7.26) is a storage function for the closed-loop system with input
u (ctitious) and output q. Next this also allows us to state that Vpd (q

qi , q) = Va (q, q) U (qi ), is a Lyapunov function for the unforced system in
(7.25): indeed this is a storage function and the conditions of Lemma 5.13
are satised. Now let us calculate the derivative of this function along the
trajectories of (7.25):
 
Vpd (q(t) qi , q(t)) = 1 q T (t)q(t) + q T (t) g(q(t)) + dU
dq (t)
(7.28)
= 1 q T (t)q(t)

One therefore just has to apply the Krasovskii-La Salle Lemma to deduce that
the xed points (qi , 0) are locally asymptotically Lyapunov stable. Lyapunov
second method guarantees that the basin of attraction Bri of each xed point
has a strictly positive measure.

Remark 7.16 (Potential energy shaping). One remarks that asymptotic sta-
bility has been obtained in part because the PD control injects some strict
output passivity inside the closed-loop system. This may be seen as a forced
damping. On the other hand the position feedback may be interpreted as a
modication of the potential energy so as to shape it adequately for control
purposes. It seems that this technique was rst advocated by Takegaki and
Arimoto in [484].

Remark 7.17. The PD control alone cannot compensate for gravity. Hence the
system will converge to a conguration that is not the desired one. Clearly
increasing 2 reduces the steady-state error. But increasing gains is not always
desirable in practice, due to measurement noise in the sensors.

Equivalent Closed-loop Interconnections

Since the closed-loop system possesses several equilibrium points, the under-
lying passivity properties of the complete closed-loop system must be local in
nature, i.e. they hold whenever the state remains inside the balls Bri [404]. It
is however possible that each block of the interconnection, when considered
separately, possesses global dissipativity properties. But the interconnection
does not.

A rst interconnection:

Looking at (7.25) one is tempted to interpret those dynamics as the intercon-

nection of two subsystems with respective inputs u1 , u2 and outputs y1 and
y2 , with y1 = u2 and y2 = u1 , and
 7.3 Rigid JointRigid Link Systems: State Feedback 389

u1 = 1 q 2 q
(7.29)

y1 = q

Evidently this is motivated by the fact that the rigid joint-rigid link manip-
ulator dynamics
in (7.25) denes a passive operator between u1 and y1 , with
q
state vector and dynamics
q

M (q(t))q(t) + C(q(t), q(t))q(t) + g(q(t)) = u1 (t) (7.30)

Let us write this second subsystem in state space form as

z1 = u2
(7.31)

y2 = 2 z1 + 1 u2

with z1 (0) = q(0)qd . Its transfer matrix is given by Hpd (s) = 2 +s

1s
In where
In is the n n identity matrix. Thus Hpd (s) is PR; see Denition 2.28. From
Theorem 5.2 and Corollary 5.3 it follows that q L2 (IR+ ). Notice that this is a
consequence of the fact that Hpd (s) denes an input strictly passive operator;
see Theorem 2.6 2). We cannot say much more if we do not pick up the
storage functions of each subsystem. Now the second subsystem has dynamics
such that the associated operator u2
y2 is input strictly passive (hence
necessarily of relative degree zero) and with storage function 22 z1T z1 . From the
fact that z1 = q and due to the choice of the initial data, one has for all t 0:
z1 (t) = q(t). It is easy to see then that the rst subsystem (the rigid joint-rigid
links dynamics) has a storage function equal to 12 q T M (q)q + U (q) U (qi ).
The sum of both storage functions yields the desired Lyapunov function for
the whole system. The interconnection is depicted in Figure 7.2.

Remark 7.18. Looking at the dynamics of both subsystems it seems that the
total system order has been augmented. But the interconnection equation y1 =
z1 may be rewriten as z1 = q. This denes a dynamical invariant z1 q = q0 ,
where q0 IRn is xed by the initial condition z1 (0) = q(0) qd . Hence the
system (7.30) and (7.31) may be reduced to the subspace z1 q = qd and
one recovers a system of dimension 2n (in other words the space (q, q, z1 ) is
foliated by invariant manifolds z1 q = qd ).

Remark 7.19. In connection with the remarks at the beginning of this sub-
subsection, let us note that the xed points of the rst unforced (i.e. u1 0)
subsystem are given by {(q, q) : g(q) = 0, q = 0}, while those of the unforced
second subsystem are given by {z1 : z1 = 0 q = q(0)}. Thus the rst
subsystem has Lyapunov stable xed points which correspond to its static
equilibrium, while the xed point of the second subsystem corresponds to the
desired static position qd . The xed points of the interconnected blocks are
390 7 Passivity-based Control

Fig. 7.2. The rst equivalent representation

given by the roots of (7.27). If one looks at the system from a pure input-
output point of view, such a xed points problem does not appear. However
if one looks at it from a dissipativity point of view, which necessarily implies
that the input-output properties are related to the state space properties, then
it becomes a necessary step.

Remark 7.20. Hpd (s) provides us with an example of a passive system that is
ISP but obviously not asymptotically stable, only stable (see Corollary 5.16).

A second interconnection:

A second possible interpretation of the closed-loop system in (7.25) is made

of the interconnection of the two blocks:

u1 = 2 q, y1 = q
(7.32)

u 2 = y1 , y2 = 2 q

The rst subsystem then has the dynamics

M (q(t))q(t) + C(q(t), q(t))q(t) + g(q(t)) + 1 q(t) = u1 (t) (7.33)

from which one recognizes an output strictly passive system, while the second
one has the dynamics
z1 = u2
(7.34)

y2 = 2 z1
 7.3 Rigid JointRigid Link Systems: State Feedback 391

with z1 (0) = q(0) qd . One can check that it is a passive lossless system
t
since u2 , y2 t = 0 2 q T (s)q(s)ds = 22 [q T q(t) q T q(0)], with storage func-
tion 22 q T q. Therefore applying the passivity Theorem (see theorem 5.2 and
Corollary 5.3), one still concludes that q L2 (IR+ ). We however may go a
little further with this decomposition. Indeed consider the system with input
u = u1 + y2 and output y = y1 . This denes an output strictly passive op-
erator u
y. Setting u y 0 one obtains that (q qi , q) = (0, 0). Hence
this closed-loop system is zero-state observable. Since the storage function
(the sum of both storage functions) we have exhibited is positive denite with
respect to this error equation xed point, and since it is proper, it follows
that the equilibrium point of the unforced system (i.e. u 0) is globally
asymptotically stable. This second interconnection is depicted in Figure 7.3.

Fig. 7.3. The second equivalent representation

In conclusion it is not very important whether we associate the strict pas-

sivity property to one block or the other. What is important is that we can
systematically associate to these dissipative subsystems some Lyapunov func-
tions that are systematically deduced from their passivity property. This is
a fundamental property of dissipative systems that one can calculate Lya-
punov functions for them. It has even been originally the main motivation for
studying passivity, at least in the eld of control and stabilization of dynamic
systems.

7.3.2 PID Control

The PID control is also a feedback controller that is widely used in prac-
tice. Let us investigate whether we can redo the above analysis for the PD
392 7 Passivity-based Control

controller. If we proceed in the same manner, we decompose the closed-loop

dynamics

t
M (q(t))q(t) + C(q(t), q(t))q(t) + g(q(t)) + 1 q(t) + 2 q(t) + 3 q(s)ds = 0
0
(7.35)
into two subsystems, one of which corresponds to the rigid joint-rigid link
dynamics, and the other one to the PID controller itself. The input and output
signals of this interconnection are this time chosen to be
t
u1 = 1 q 2 q 3 0 q(s)ds = y2
(7.36)

y1 = q = u2

The dynamics of the PID block is given by (compare with (7.31)):

z1 (t) = z2 (t)



z2 (t) = u2 (t) (7.37)





y2 = 1 u2 + 2 z2 + 3 z1

The transfer matrix of this linear operator is given by (compare with (2.64)
and (2.65))
1 s2 + 2 s + 3
Hpid (s) = In (7.38)
s2
Thus it has a a double pole with zero real part and it cannot be a PR transfer
matrix, see Theorem 2.38. This can also be checked by calculating u2 , y2 t
t
that contains a term 0 u2 (s)z1 (s)ds which cannot be lower bounded.
If one chooses u2 = q then the PID block transfer matrix becomes

1 s2 + 2 s + 3
Hpid (s) = In (7.39)
s
which this time is a PR transfer function for a suitable choice of the gains,
and one can check that

t 
t t
1 T 3
u2 , y2 t = [q (s)q(s)]t0 + 2 q T (s)q(s)ds + q T (s)q(s)ds
2 0 2 0 0
(7.40)
which shows that the system is even input strict passive (but the transfer
function is not SPR otherwise this system would be strictly passive (in the
state space sense); see Example 4.65, which it is not from inspection of (7.40)).
However this change of input is suitable for the PID block, but not for the
rigid joint-rigid link block, that we know is not passive with respect to the
supply rate uT1 q because of the relative degree of this output. As a consequence
the dynamics in (7.35) cannot be analyzed through the passivity theorem.
 7.3 Rigid JointRigid Link Systems: State Feedback 393

Remark 7.21. Notice however that the system in (7.35) can be shown to be
locally Lyapunov
stable [17] with a Lyapunov function V (z), where z(t) =
t
0
q(s)ds
q(t) . Let us add a ctitious input in the right-hand-side of (7.35)
q(t))
instead of zero. From the KYP lemma we know that there exists an output y
(another ctitious signal) such that this closed-loop is passive with respect to
the supply rate T y. One has y = (0, 0, 1) V
z .

7.3.3 More about Lyapunov Functions and the Passivity Theorem

Before going on with controllers that assure tracking of arbitrary (smooth

enough) desired trajectories, let us investigate in more detail the relation-
ships between Lyapunov stable systems and the passivity theorem (which has
numerous versions, but is always based on the study of the interconnection
of two dissipative blocks). From the study we made about the closed-loop
dynamics of the PD and PID controllers, it follows that if one has been able
to transform a system (should it be open or closed-loop) as in Figure 3.2 and
such that both blocks are dissipative, then the sum of the respective storage
functions of each block is a suitable Lyapunov function candidate. Now one
might like to know whether a Lyapunov stable system possesses some dissi-
pativity properties. More precisely, we would like to know whether a system
that possesses a Lyapunov function, can be interpreted as the interconnection
of two dissipative subsystems. Let us state the following [74, 317]:

Lemma 7.22. Let L denote a set of Lyapunov stable systems with equilibrium
point (x1 , x2 ) = (0, 0), where (x1 , x2 ) generically denotes the state of systems
in L. Suppose the Lyapunov function V (x1 , x2 , t) satises
1.
V (x1 , x2 , t) = V1 (x1 , t) + V2 (x2 , t) (7.41)
where V1 (), V2 () are positive denite radially unbounded functions
2.
V (x1 , x2 , t) 1 1 ( x1 ) 2 2 ( x2 ) (7.42)
along trajectories of systems in L, where 1 () and 2 () are class K func-
tions, and 1 0, 2 0.
Suppose there exist functions F1 () and F2 () such that for all x1 , x2 and t t0

V1 V1 T
+ F1 (x1 , t) 1 1 ( x1 ) (7.43)
t x1

V2 V2 T
+ F2 (x1 , t) 2 1 ( x2 ) (7.44)
t x2
394 7 Passivity-based Control

and Fi (0, t) = 0, dim xi =dim xi for i = 1, 2, for all t t0 . Then there
exists a set P of Lyapunov stable systems, with the same Lyapunov function
V (x1 , x2 , t), that can be represented as the feedback interconnection of two
(strictly) passive subsystems with states x1 and x2 respectively. These systems
are dened as follows:
   
x1 (t) = F1 (x1 (t), t) + G1 (x1 (t), x2 (t), t)u1
(7.45)
y = GT (x (t), x (t), t) V1 (x , t)
1 1 1 2 x  1 1

   
x2 (t) = F2 (x2 (t), t) + G2 (x1 (t), x2 (t), t)y1
(7.46)
y = GT (x (t), x (t), t) V2 (x , t) = u
2 2 1 2 x 2 1
2

where G1 () and G2 () are arbitrary smooth nonzero functions 2 , which can be

shown to dene the inputs and the outputs of the interconnected systems.

The proof of Lemma 7.22 is straightforward from the KYP property of

the outputs of passive systems. Note that Lemma 7.22 does not imply any
relationship between the system in L and the system in P other than the
fact that they both have the same Lyapunov function structure. That is why
we used dierent notations for their states (x1 , x2 ) and (x1 , x2 ). We are now
interested in establishing sucient conditions allowing us to transform a sys-
tem in L into a system in P having the particular form given in (7.45) and
(7.46). These conditions are discussed next. Suppose (L ) has the following
form (notice that this is a closed-loop form):

x1 (t) = F1 (x1 (t), t) + G1 (x1 (t), x2 (t), t)u1
(7.47)

y1 = h1 (x1 (t), t) = u2

x2 (t) = F2 (x2 (t), t) + G2 (x1 (t), x2 (t), t)u2
(7.48)

y2 = h2 (x2 (t), t) = u1
From (7.42) we thus have

V1 V1 T V2 V2 T
V (x1 , x2 , t) = t + x1 F1 (x1 , t) + t + x2 F2 (x2 , t)

T V2 T
+ V1
x1 G1 (x1 , x2 , t)h2 (x2 , t) + x2 G2 (x1 , x2 , t)
(7.49)
h1 (x1 , t)

1 1 ( x1 ) 2 2 ( x2 )
2
We assume that the considered systems have 0 as a unique equilibrium point.
 7.3 Rigid JointRigid Link Systems: State Feedback 395

with inequalities (7.43) and (7.44) satised for both systems in (7.47) and
(7.48). Now let us rewrite (L ) in (7.47) (7.48) as follows (we drop the argu-
ments for convenience; u1 = h2 (x2 ), u2 = h1 (x1 )):

x1 = (F1 + G1 u1 g1 u1 ) + g1 u1
(7.50)

y1 = g1T V1
x1 = u 2

x2 = (F2 + G2 u2 g2 u2 ) + g2 u2
(7.51)

y2 = g2T V2
x2 = u 1

Notice that (L ) in (7.50) and (7.51) and (L ) in (7.47) and (7.48) strictly
represent the same system. We have simply changed the denition of the
inputs and of the outputs of both subsystems in (7.47) and (7.48). Then the
following Lemma is true:
Lemma 7.23. Consider the closed-loop Lyapunov stable system (L ) in (7.47)
and (7.48), satisfying (7.49), with F1 and F2 satisfying (7.43) and (7.44). A
sucient condition for (L ) to be able to be transformed into a system in P
is that the following two inequalities are satised:
1. 
V1 T T V2
G1 h2 + g1 g2 0 (7.52)
x1 x2
2. 
V2 T T V1
G2 h1 g2 g1 0 (7.53)
x2 x1
for some non-zero, smooth matrices g1 , g2 of appropriate dimensions, and
with
F1 (0, t) + G1 (0, x2 , t)u1 (0, x2 , t) g1 (0, x2 , t)u1 (0, x2 , t) = 0

x2 , t 0
(7.54)
F2 (0, t) + g(x1 , 0, t)u2 (x1 , 0, t) g2 (x1 , 0, t)u2 (x1 , 0, t) = 0

x1 , t 0

Notice that these conditions are sucient only for transforming the system
in P; see Remark 7.29.

Proof: The proof of Lemma 7.23 is straightforward. Inequalities (7.52) and

(7.53) simply guarantee that Vxi (fi + gi ui gi ui ) i i ( xi ), and (7.54)
i

guarantees that xi = fi + gi ui gi ui has xi = 0 as equilibrium point. Thus

(L ) is in P.
396 7 Passivity-based Control

Example 7.24. Consider the following system:

x1 (t) = F1 (x1 (t), t) + G1 (x1 (t), x2 (t), t)u1
(7.55)
V1
y1 (t) = x1 (x1 , t) = u2

x2 (t) = F2 (x2 (t), t) GT1 (x1 (t), x2 (t), t)u2
(7.56)
V2
y2 (t) = x2 (x2 , t) = u1
T
t + xi fi i i ( xi ), i 0, fi (0, t) = 0 for all t t0 , and V
with V i Vi

satises (7.41) and (7.42). Then we get along trajectories of (7.55) and (7.56):
V = V1 + V2 1 1 ( x1 ) 2 2 ( x2 ). However the subsystems in
(7.55) and (7.56) are not passive, as they do not verify the KYP property.
The conditions (7.52) and (7.53) reduce to

V1 T V2 V1 T V2
G1 + g1 g2T =0 (7.57)
x1 x2 x1 x2
T T
as in this case V x1 G1 h2 = x2 G2 h1 . Now choose g1 = G1 , g2 = 1,
1 V2

u1 = x2 , u2 = G1 x1 : (7.57) is veried.
V2 T V1

In conclusion, the system in (7.55) and (7.56) is not convenient because its
outputs and inputs have not been properly chosen. By changing the denitions
of the inputs and outputs of the subsystems in (7.55) and (7.56), leaving the
closed-loop system unchanged, we transform the system such that it belongs
to P. In most of the cases, the functions gi , hi and fi are such that the only
possibility for the equivalent systems in (7.50) and (7.51) to be Lyapunov
stable with Lyapunov functions V1 () and V2 () respectively is that gi ui gi ui ,
i.e. we only have to rearrange the inputs and the outputs to prove passivity.
From Lemma 7.23 we can deduce the following result:

Corollary 7.25. Consider the system in (7.47) and (7.48). Assume (7.49) is
T V2 T
x1 G1 h2 = x2 G2 h1 (let us denote this equality as the
satised, and that V1

Cross Terms Cancellation Equality CTCE). Then i) If one of the subsystems

in (7.47) or (7.48) is passive, the system in (7.47) and (7.48) can be trans-
formed into a system that belongs to P. ii) If the system in (7.47) and (7.48)
is autonomous, it belongs to P.

Proof: Using the CTCE, one sees that inequalities in (7.52) and (7.53) reduce
either to:
V1 T V1 T V2
G1 h2 + g1 g2T =0 (7.58)
x1 x1 x2
or to
V2 T V2 T V1
G2 h1 + g2 g1T =0 (7.59)
x2 x2 x1
 7.3 Rigid JointRigid Link Systems: State Feedback 397

Suppose that the system in (7.48) is passive. Then h2 = GT2 V 2

x2 , thus it
suces to choose g2 = G2 , g1 = G1 . If the system in (7.47) is passive, then
h1 = GT1 V 1
x1 , and we can take g2 = G2 , g1 = G1 . The second part of the
corollary follows from the fact that one has for all x1 and x2 :

V1 T V2 T
G1 (x1 )h2 (x2 ) = G2 (x2 )h1 (x1 ) (7.60)
x1 x2
Then (7.47) (7.48) can be transformed into a system that belongs to P. Nec-
x2 and h1 (x1 ) = G1 x1 , or h2 (x2 ) = G2 x2 and
essarily h2 (x2 ) = GT2 V 2 T V1 T V2

h1 (x1 = GT1 V 1
x1 , which correspond to solutions of (7.58) or (7.59) respectively.

In the case of linear time invariant systems, one gets G1 C2 P21 + G1 GT2 = 0
or G2 C1 P11 + G2 GT1 = 0 instead of (7.58) and (7.59) respectively. Suppos-
ing either C2 = GT2 P2 or C1 = GT1 P1 the result follows and the passive
interconnection is found.

Example 7.26. Throughout this chapter and Chapter 8 we shall see several
applications of Lemmas 7.22 and 7.23. In particular it happens that the can-
cellation of cross terms in Lyapunov functions derivatives has been widely
used for stabilization and almost systematically yields an interpretation via
the passivity theorem. To illustrate those results let us reconsider the PD
controller closed-loop dynamics in (7.25). Let us start from the knowledge of
the Lyapunov function deduced from the storage function in (7.26). Letting
x1 = (q, q) be the state of the rigid joint-rigid link dynamics and x2 = z1 be
the state of the second subsystem in (7.31), one sees that the sum of the stor-
age functions associated to each of these blocks forms a Lyapunov function
that satises the conditions of Lemma 7.22. Moreover the conditions of Corol-
lary 7.25 are satised as well, in particular the CTCE. Indeed from (7.32) we
get (but the same could be done with the interconnection in (7.29))


V1 T
 T  0

G (x )h
1 1 2 2(x ) = g (q), q T
M (q) (2 q)

x1 M 1 (q)


= 2 q T q (7.61)






V2 T
x2 G2 (x2 )h1 (x1 ) = 2 q T q

Hence the dynamics in (7.25) can indeed be interpreted as the negative

feedback interconnection of two dissipative blocks. As another example, con-
sider Theorem 5.42: notice that choosing the controller u of the driving system
as uT = (Lf1 U ()) exactly corresponds to a CTCE. Hence the closed-loop
system thereby constructed can be analyzed through the passivity theorem.
This is the mechanism used in [317].
398 7 Passivity-based Control

Such closed-loop interpretations of Lyapunov stable systems are not funda-

mental from a stability point of view, since the system is already known to be
stable. However they have been widely used in the Systems and Control liter-
ature since they provide an elegant manner to analyze the closed-loop system.
Moreover they may provide the designer with ideas linked to the properties of
interconnections of passive systems. We shall illustrate again the application
of Lemmas 7.22 and 7.23 and Corollary 7.25 in the sequel; see in particular
Sections 7.4, 7.6 and Chapter 8.

7.3.4 Extensions of the PD Controller for the Tracking Case

The tracking problem for the model in (6.98) can be easily solved using a
linearizing feedback that renders the closed-loop system equivalent to a double
integrator. Then all the classical machinery for linear systems can be applied.
However we are not interested here in following this path. We would rather
like to see how the PD control may be extended to the tracking case, i.e.
how we can preserve and use the system dissipativity to derive a globally
stable controller guaranteeing tracking of any suciently dierentiable desired
trajectory.

A First Extension of the PD Controller: the Paden and Panja

Scheme

The rst idea is a direct extension of the PD structure, applying the control
[389]:
= M (q(t))qd (t) + C(q(t), q(t))qd (t) + g(q(t)) 1 q(t)
2 q(t) (7.62)

with qd () C 2 (IR+ ). Setting qd constant one retrieves a PD controller with

gravity compensation. The closed-loop system is given by:

+ C(q(t), q(t))q(t)
M (q(t))q(t) + 2 q(t) = 0
+ 1 q(t) (7.63)
This closed-loop dynamics resembles the one in (7.25). This motivates us
to study its stability properties by splitting it into two subsystems as

+ C(q(t), q(t))q(t)
M (q(t))q(t) = u1 (t) = y2 (t)
(7.64)
= u2 (t)
y1 (t) = q(t)
and

z1 (t) = u2 (t)



y2 (t) = 1 u2 (t) + 2 z1 (t) (7.65)





z1 (0) = q(0) qd (0)
Let us make the following Assumption (see Lemma 6.16):
 7.3 Rigid JointRigid Link Systems: State Feedback 399

Assumption 17 The matrix C(q, q) is written in such a way that M (q, q)

2C(q, q) is skew-symmetric.
Then one computes that
t  
u1 , y1 t = 0 qT ( ) M (q( ))q(
) + C(q( ), q( ))q(
) d

 
= 1 ) t
qT ( )M (q( ))q( (7.66)
2 0

12 q(0)
T M (q(0))q(0)

and that t  t
1
u2 , y2 t = 1 0
)d +
qT ( )q( 2 q(s)T q(s) 0
(7.67)
21 q(0)T q(0)
Notice that the second block is input strictly passive. Similarly to the
PD controller analysis, one concludes that the dynamics in (7.63) can indeed
be transformed into the interconnection of two passive blocks. We could also
have deduced from Lemma 7.23 that such an interconnection exists, checking
= 1 qM
that V (q, q) (q)q + 1 2 q T q is a Lyapunov function for this system,
2 2
whose derivative along the trajectories of (7.63) is semi-negative denite (i.e.
1 = 0 in Lemma 7.22) (we let the reader do the calculations by him/herself).
However one cannot apply the Krasovskii-La Salle Theorem to this system
because it is not autonomous (the inertia and Coriolis matrices depend ex-
One has to resort to
plicitly on time when the state is considered to be (q, q)).
Matrosovs Theorem to prove the asymptotic stability (see Theorem A.35 and
Lemma A.36 in the Appendix) [389]. Equivalent representations (that are to
be compared to the ones constructed for the PD control in Subsection 7.3.1)
are depicted in Figures 7.4 and 7.5.

The Slotine and Li Controller (Passivity Interpretation)

The above scheme has the advantage of being quite simple. However its ex-
tension to the adaptive case (when the inertia parameters are supposed to
be unknown, one needs to introduce some on-line adaptation) is really not
straightforward. One big challenge in the Robotics and Systems and Control
elds during the 1980s was to propose a feedback controller that guarantees
tracking and which extends also to an adaptive version (which will be pre-
sented in Section 8.1.1). Let us consider the following input [425, 461] 3 :

(q(t), q(t), t) = M (q(t))qr (t) + C(q(t), q(t))qr (t) + g(q(t)) 1 s(t) (7.68)
3
It seems that what is now widely known as the Slotine and Li scheme, was also
designed in [425] at the same time so that the Slotine and Li scheme could be
named the Slotine-Li-Sadegh-Horowitz scheme.
400 7 Passivity-based Control

Fig. 7.4. First interconnection: lossless manipulator dynamics

Fig. 7.5. Second interconnection: OSP manipulator dynamics

where qr (t) = qd (t) q(t), s(t) = q(t) qr (t) = q(t)

+ q(t), and we recall
2 +
that qd () is supposed to be in C (IR ). Introducing (7.68) into (6.98) one
obtains
M (q(t))s(t) + C(q(t), q(t))s(t) + 1 s(t) = 0 (7.69)
Notice that contrary to the scheme in (7.62), setting qd constant in (7.68)
does not yield the PD controller. However the controller in (7.68) can be seen
as a PD action (1 s) with additional nonlinear terms whose role is to assure
some tracking properties. Before going on let us note that the whole closed-
loop dynamics is not in (7.69) since this is an nth order system with state s,
 7.3 Rigid JointRigid Link Systems: State Feedback 401

whereas the whole system is 2n-th order. To complete it one needs to add to
(7.69):

= q(t) + s(t)
q(t) (7.70)
Therefore the complete closed-loop dynamical system is given by

M (q(t))s(t) + C(q(t), q(t))s(t) + 1 s(t) = 0

= q(t) + s(t)
q(t)

q(0) = q0 , q(0) = q0

It should be clear from now all the foregoing developments that the sub-
system in (7.69) denes a passive operator between u1 = 1 s = y2 and
y1 = s = u2 , with storage function V1 (s, t) = 12 sT M (q)s (which is a Lya-
punov function for this subsystem which is zero-state observable). This is
strongly based on Assumption 17. The equivalent feedback interconnection of
the closed-loop is shown in Figure 7.6.

Fig. 7.6. Closed-loop equivalent representation

Remark 7.27. The subsystem in (7.69) can be at once proved to dene an

asymptotically stable system since one can view it as the interconnection of
402 7 Passivity-based Control

a passive mapping u
y = q with zero-state detectable dynamics M (q)s +
C(q, q)s = u, with a static output feedback u = 1 y. Hence Theorem 5.24
applies and one concludes that s(t) 0 as t +.

The second subsystem obtained from (7.70) can be rewritten as

z1 (t) = z1 (t) + u2 (t)
(7.71)

y2 = 1 u2

It therefore has a relative degree r2 = 0, and the state is not observable

from the output y2 . However it is zero-state detectable since {y2 = u2 = 0}
lim z1 (t) = 0. We also notice that this system is very strictly passive since
t+

t
u2 , y2 t = 1 0
uT2 (s)u2 (s)ds

1
t
= 1 0 y2T (s)y2 (s)ds (7.72)

1
t 1
t
= 2 0
uT2 (s)u2 (s)ds + 21 0
y2T (s)y2 (s)ds

Let us compute storage functions for this system. Let us recall from (4.137)
that for systems of the form x = f (x, t) + g(x, t)u, y = h(x, t) + j(x, t)u with
j(x, t)+j T (x, t) = R full-rank, the storage functions are solutions of the partial
dierential inequality (that reduces to a Riccati inequation in the linear case)

. / 
V T V 1 V T 1 V
f (x, t) + + h T
g R1 h g 0 (7.73)
x t 2 x 2 x

and that the available storage Va () and the required supply Vr () (with
x(t) = 0) satisfy (7.73) as an equality. Thus the storage functions V (z1 )
for the system in (7.71) are solutions of

dV T 1 dV T dV
z1 + 0 (7.74)
dz1 41 dz1 dz1
If we set the equality it follows that the two solutions satisfy
dV
dz1 (t) = 0
(7.75)
dV
dz1 (t) = 41 z1 (t)
for all t 0, from which one deduces that Va (z1 ) = 0 and Vr (z1 ) = 21 z1T z1 ,
whereas any other storage function satises 0 = Va (z1 ) V (z1 ) Vr (z1 ).
 7.3 Rigid JointRigid Link Systems: State Feedback 403

Remark 7.28. Let us retrieve the available storage and the required supply
from their variational formulations (notice that the system in (7.71) is con-
trollable so that the required supply can be dened):

t
Va (z1 (0)) = sup 1 uT2 u2 ds = 0 (7.76)
u2 :(0,z1 (0)) 0

which means that the best strategy to recover energy from this system through
the output y2 is to leave it at rest (so as to recover nothing, actually!), and

0
Vr (z1 (0)) = inf uT2 y2 ds
u2 :(t,0)(0,z1 (0)) t

0 ) T * (7.77)
= inf 1 (z1 + z1T )(z1 + z1 ) ds
u2 :(t,0)(0,z1 (0)) t

= 1 z1T (0)z1 (0)

where the last step is performed by simple integration of the cross term and
dropping the other two terms which are always positive, for any control strat-
egy.
We conclude that a suitable Lyapunov function for the closed-loop system
in (7.69) (7.70) is given by the sum
1 T
V (s, q, t) = s M (q, t)s + 21 q T q (7.78)
2

It is noteworthy that we have really deduced a Lyapunov function from the

knowledge of some passivity properties of the equivalent interconnection form
of the closed-loop system. Historically, the closed-loop system in (7.69) and
(7.70) has been studied rst using the storage function of the rst subsystem
in (7.69) only, and then using additional arguments to prove the asymptotic
convergence of the whole state towards zero [461]. It is only afterwards that
the Lyapunov function for the whole closed-loop system has been proposed
[472]. We have shown here that it is possible to construct it directly from
passivity arguments. It must therefore be concluded on this example that
the dissipativity properties allow one to directly nd out the right Lyapunov
function for this system.
Remark 7.29. Lemmas 7.22 and 7.23 can in general be used if one starts from
the knowledge of the Lyapunov function. However the cross-term-cancellation-
equality (CTCE) is not satised since

V T T 1 T
x11 G1 (x1 )h2 (x2 ) = s M (q)M (q)1 s = 1 s s
(7.79)

V2 T
x2 G2 (x2 )h1 (x1 ) = 1 q s
T
404 7 Passivity-based Control

V1 T
This comes from the fact that this time one has to add x1 G1 (x1 )h2 (x2 ) +
V2 T T
= 1 sT s + 1 q T s to V
x2 G2 (x2 )h1 (x1 )
2
x2 F2 (x2 ) = 22 1 q T q in or-
der to get the inequality in (7.49). One may also check that the inequalities
in (7.52) and (7.53) can hardly be satised by any g1 and g2 . Actually the
conditions stated in Lemma 7.23 and Corollary 7.25 are sucient only. For
instance from (7.49) one can change the inequalities in (7.52) and (7.53) to
T V2 T
incorporate the terms V 1
x1 F1 (x1 , t) and x2 F2 (x2 , t) in the conditions re-
quired for the matrices g1 and g2 . Actually Lemmae 7.22 and 7.23 will be
useful when we deal with adaptive control; see Chapter 8, in which case the
CTCE is generally satised.

The Slotine and Li Controller (Stability Analysis)

There are two ways to prove the stability for the closed-loop system in (7.69)
and (7.70). The rst proof is based on the positive function V (s, q, t) =
1 T
2 s M (q)s (which we denoted as V1 (s, t) above), where one notices that
q(t) = q(t) + qd (t). Hence the explicit time-dependency in V (s, q, t). This
proof makes use of Lemma 4.8. This proof does not show Lyapunov stability
but merely shows the boundedness of all signals as well as the asymptotic con-
vergence of the tracking error and its derivative towards zero. The second prof
is based on the Lyapunov function (candidate) in (7.78). Lyapunov stability
of the error (closed-loop) system equilibrium point is then concluded.

First stability proof:

Let us consider
1 T
s M (q)s,
V (s, q, t) = (7.80)
2
and let us calculate its derivative along the solutions of (7.69):

V (s, q, t) = sT (t)M (q(t))s(t) + 12 sT (t)M (q(t), q(t))s(t)

= sT (t)(C(q(t), q(t)) 1 s(t)) + 12 sT (t)M (q(t), q(t))s(t)

= 1 sT (t)s(t) + sT (t)[C(q(t), q(t)) + 12 M (q(t), q(t))]s(t)

= 1 sT (t)s(t) 0
(7.81)
where the last equality is obtained thanks to the skew-symmetry property
(Lemma 6.16). Let us now integrate both sides of (7.81):

t
V (s(t), q(t)) V (s(0), q(0)) sT ( )s( )d (7.82)
0
 7.3 Rigid JointRigid Link Systems: State Feedback 405

which implies that

t
sT ( )s( )d V (s(0), q(0)) (7.83)
0
since V (, ) 0. Therefore s() is in L2 . Let us now consider the system in
(7.70). This is an asymptotically stable system whose state is q() and whose
input is s(). Applying Lemma 4.8 we deduce that q L2 L , q L2 ,
and limt+ q(t) = 0. Furthermore since V (s(t), q(t), t) V (s(0), q(0), 0), it
follows that for bounded initial data, ||s(t)|| < +, i.e. s L . Therefore
q L as well, and from Fact 6 (Section 4.1) the function q() is uniformly
continuous. Using (7.69) it follows that s L , so using Fact 6 and then
Fact 8 we conclude that s(t) 0 as t +. Thus q(t) 0 as t +.
converges
All the closed-loop signals are bounded and the tracking error (q, q)
globally asymptotically to zero. However we have not proved the Lyapunov
stability of the equilibrium point of the closed-loop error system (7.69) and
(7.70).

Lyapunov stability proof:

Let us now consider the positive denite function in (7.78). Computing its
derivative along the closed-loop system (7.69) and (7.70) trajectories yields

V (q(t), q(t)) = 1 qT (t)qT (t) 2 1 q T (t)q(t) 0
(7.84)
=
from which the global asymptotic Lyapunov stability of the xed point (q, q)
(0, 0) follows. The skew-symmetry property is used once again to compute
the derivative. It was further shown in [472] that when the system has only
revolute joints then the stability is uniform. This comes from the fact that in
such a case, the inertia matrix M (q) contains only bounded (smooth) functions
like cos() and sin() and is thus bounded, consequently the Lyapunov function
is also upperbounded by some class K function. It is interesting to see how
the technology inuences the stability.

In both stability proofs, one can conclude about exponential convergence.

Indeed for the rst proof one has V (s, q, t) 1 sT (t)s(t) minM(q)
1
V (s, q, t).
Therefore s() converges to zero exponentially fast, and so do q() and q().
The interest of the above proof is that when we deal with the adaptive case,
then exponential stability will be lost, and the stability proof is then identical
to the above one.

7.3.5 Other Types of State Feedback Controllers

The use of the property in Assumption 17 is not mandatory. Let us describe

now a control scheme proposed in [239], that can be classied in the set of
passivity-based control schemes, as will become clear after the analysis. Let
us consider the following control input:
406 7 Passivity-based Control

= 12 M (q(t), q(t))[q(t)
+ q(t)] + M (q(t))[qr (t) q(t)]+

  (7.85)
+C(q(t), q(t))q(t) + g(q(t)) d +
1
d q(t)
q(t)

Introducing (7.85) into the dynamics (6.98) one obtains:


1
M (q(t))s(t) + M (q(t), q(t))s(t) + d + q(t) + d q(t) = 0 (7.86)
2 1
which we can rewrite equivalently as
 
21 + d 2
M (q(t))s(t) + C(q(t), q(t))s(t) + d + 2
q(t) 1 q(t) =
(7.87)
= 12 M (q(t), q(t))s(t) + C(q(t), q(t))s(t)
These two representations of the same closed-loop system are now ana-
lyzed from a passivity theorem point of view. Let us consider the following
negative feedback interconnection:

u1 = y2 = 21 M (q, q)s + C(q, q)s
(7.88)

u 2 = y1 = s
where the
  rst subsystem has dynamics M (q(t))s(t) + C(q(t), q(t))s(t) +
+ d q(t) = u1 (t) while the second one is a static operator
d + 1 q(t)
between u2 = s and y2 given by u2 (t) = 12 M (q(t), q(t))s(t) C(q(t), q(t))s(t).
It is easily checked that if Assumption 17 is satised then

t
1
u2 , y2 t = sT ( )[M (q( ), q( )) 2C(q( ), q( ))]s( )d = 0 (7.89)
2 0
and that the available storage of the second block is the zero function as well.
Concerning the rst subsystem one has

t    
u1 , y1 t = sT ( ) M (q( ))s + C(q( ), q( ))s( ) + d + ) + d q( ) d
q(
0 1

 
2
= 12 [sT ( ) M (q( ))s( )]t0 + 1
2 2d + 1 [q T ( )q( )]t0

 t 2  3
+ d + ) + 2 d q T ( )q( ) d
qT ( )q(
0 1

 
2
12 s(0)T M (q(0))s(0) 1
2 2d + 1 q(0)T q(0)
(7.90)
 7.3 Rigid JointRigid Link Systems: State Feedback 407

which proves that it is passive with respect to the supply rate uT1 y1 . It can
also be calculated that the available storage function of this subsystem is given
by:

t
Va (q(0), s(0)) = sup sT ( ) {M (q( ))s( ) + C(q( ), q( ))s( )
u1 :[q(0),s(0)] 0

  3
+ d + ) + d q( ) d
q(
1

 
2
= 12 s(0)T M (q(0))s(0) + d + 21 q T (0)q(0)
(7.91)
Since this subsystem is zero-state detectable (u1 s 0 q 0 as
t +) one concludes that the available storage in (7.91) is actually a
Lyapunov function for the corresponding unforced system, whose xed point
= (0, 0)) is asymptotically stable. This also holds for
(q, s) = (0, 0) (or (q, q)
the complete closed-loop system since the second block has storage functions
equal to zero and the dynamics in (7.86) is zero-state detectable when one
considers the input to be u in the left-hand-side of (7.86) and y = y1 = s (set
u 0 and s 0 and it follows from (7.86) that q 0 exponentially). Actually,
the derivative of Va (q, s) in (7.91) along trajectories of the rst subsystem is
given by:

T
Va (q(t), s(t)) = d + q (t)q(t) 2 d q T (t)q(t) 0 (7.92)
1
It is noteworthy that the result in (7.92) can be obtained without using the
skew-symmetry property in assumption 17 at all. But skew-symmetry was
used to prove the dissipativity of each block in (7.88).
Remark 7.30. Originally the closed-loop system in (7.86) has been proven to
be Lyapunov stable using the Lyapunov function
1 T 1
q) =
V (q, q M (q)q + qT M (q)q + q T [2 M (q) + 1 In ]q (7.93)
2 2
which can be rearranged as
1 T 1
V (s, q) = s M (q)s + 1 q T q (7.94)
2 2
The derivative of V () in (7.93) or (7.94) along closed-loop trajectories is
given by:


1

V (q(t), q(t)) = qT (t) d + q(t) 2d qT (t)q(t) 2 d q T (t)q(t)

(7.95)
408 7 Passivity-based Control

Notice that Va () in (7.91) and V () in (7.94) are not equal one to each
other. One concludes that the passivity analysis of the closed-loop permits to
discover a (simpler) Lyapunov function.

Remark 7.31. The foregoing stability analysis does not use the CTCE of
Lemma 7.23. One concludes that the schemes that are not based on the skew-
symmetry property in Assumption 17 do not lend themselves very well to an
analysis through the passivity Theorem. We may however consider the con-
troller in (7.85) to be passivity-based since it does not attempt at linearizing
the system, similarly to the Slotine and Li scheme.

7.4 Rigid JointRigid Link: Position Feedback

Usually most manipulators are equipped with position and velocity sensors,
and controlled point-to-point with a PD. The tracking case requires more, as
we saw. However the controllers structure becomes more complicated, hence
less robust. It is of some interest to try to extend the separation principle for
linear systems (a stable observer can be connected to a stabilizing controller
without destroying the closed-loop stability), towards some classes of nonlin-
ear systems. The rigid joint-rigid link manipulator case seems to constitute
a good candidate, due to its nice properties. At the same time such systems
are nonlinear enough, so that the extension is not trivial. In the continuity
of what has been done in the preceding sections, we shall investigate how
the dissipativity properties of the Slotine and Li and of the Paden and Panja
schemes can be used to derive (locally) stable controllers not using velocity
feedback.
In the following we shall start by the regulation case (see Section 7.4.1),
and then analyze the tracking of trajectories (see Sections 7.4.2 and 7.4.3).

7.4.1 P + Observer Control

In this subsection we present the extension of the PD controller when the

velocity is not available as done in [44, 488]. Basically the structure of output
(position) feedback controllers is that of the original input where the velocity
q is replaced by some estimated value. Let us consider the dynamics in (6.98)
with the controller:

= g(qd ) 1 q 12 (q z)
(7.96)

z = 3 (q z)

so that the closed-loop dynamics is given by

 7.4 Rigid JointRigid Link: Position Feedback 409
1
M (q(t))q(t) + C(q(t), q(t))q(t) + g(q(t)) g(qd ) + 2 (q(t) z(t)) = 1 q(t)

z(t) q(t) = 3 (q(t) z(t)) q(t)

(7.97)
Let us now make a direct application
of Corollary
  7.25. Let us rst rewrite
x11 q
(7.97) in a state-space form, with x1 = = and x2 = q z. We
x12 q
obtain

x11 (t) = x12 (t)

x12 (t) = M 1 (x11 (t) + qd )[C(x11 (t) + qd , x12 (t))x12 (t) + g(x11 (t) + qd )

g(qd ) + 1 x11 (t)] + M 1 (x11 (t) + qd )h2 (x2 (t))

x2 (t) = 3 x2 (t) + h1 (x1 (t))

h2 (x2 ) = 12 x2

h1 (x1 ) = x12
(7.98)
where h1 () and h2 () are as in (7.47) and (7.48). The closed-loop scheme can
be shown to be globally asymptotically Lyapunov stable with the Lyapunov
function V (x11 , x12 , x2 ) = V1 (x11 , x12 ) + V2 (x2 ) dened as

V1 (x11 , x12 ) = 2 12 xT12 M (x11 + qd )x12 + 21 xT11 x11 + Ug (x11 + qd )
 (7.99)
Ug (qd ) xT11 g(qd )

and
1 T
V2 (x2 ) =
x x2 (7.100)
2 2
It can be shown that V1 () is positive denite and has a global minimum
at (x11 , x12 ) = (0, 0) provided 1 where is a Lipschitz constant for g().
Dierentiating V () along the trajectories of (7.97) or equivalently (7.98) one
nds

V (x2 ) = 3 xT2 x2 (7.101)

V1 T T
where the CTCE is satised since = xT12 x2 =
x1 G1 h2 V2
x2 G2 h1 .
Since
the system is autonomous, Corollary 7.25 ii) applies. Now it is easy to see
that the second subsystem with state vector x2 , input u2 = h1 (x1 ) and ouput
y2 = h2 (x2 ) is passive:
410 7 Passivity-based Control
t1 T
u2 , y2 t = x (s)u2 (s)ds
0 2 2

t1 T
= x (s)(x2 (s)
0 2 2
+ 3 x2 (s))ds (7.102)

1 T t 3
t
= 22 [x2 (s)x2 (s)]0 + 2 0 xT2 (s)x2 (s)ds
and one recognizes a storage function S2 (x2 ) equal to 12 V2 with V2 in (7.100).
Notice that the second subsystem (with state x2 ) is strictly passive in the sense
of Lemma 4.84, but it is also output strictly passive. The other subsystem is
dened with input u1 = y2 = h2 (x2 ) and output y1 = u2 = h1 (x1 ) and is
passive as one can check:

u1 , y1 t = x12 , h2 t =
t
= 0 xT12 (s)[M (x11 (s) + qd )x12 (s) + C(x11 (s) + qd , x12 (s))x12 (s)
(7.103)
+g(x11 (s) + qd )x12 (s) g(qd )x12 (s) + 1 x11 (s)x12 (s)]ds

= S1 (t) S1 (0),

where we used x11 (t) = x12 (t) in the calculation.

Remark 7.32. In connection with Remark 7.18, let us note that this time
the closed-loop scheme has an order strictly larger than the open-loop one.
One has V (x1 , x2 ) = V1 (x1 ) + V2 (x2 ) = 2 S1 (x1 ) + 2 S2 (x2 ). This is due
to the particular choice of h1 (x1 ) and h2 (x2 ).
The output strict passivity plus zero state detectability properties of the
second block is important because it is precisely these properties that
allow one to use the Krasovskii-La Salle Theorem to prove the asymptotic
stability.

7.4.2 The Paden and Panja + Observer Controller

The material that follows is mainly taken from [45]. In fact it is to be expected
that the separation principle does not extend completely to the nonlinear sys-
tems we deal with. Indeed the presented schemes assure local stability only
(more exactly they assure semi-global stability, i.e. the region of attraction of
the closed-loop xed point can be arbitrarily increased by increasing some
feedback gains). In what follows we shall not develop the whole stability
proofs. We shall just focus on the passivity interpretation of the obtained
closed-loop system, and in particular on the local stability that results from
the fact that the storage function satises the dissipation inequality locally
only.
 7.4 Rigid JointRigid Link: Position Feedback 411

The foregoing subsection was devoted to an extension of PD controllers

and concerns global regulation around a xed position only. It is of interest to
consider the tracking case which is, as one expects, much more involved due
to the non-autonomy of the closed-loop scheme. Let us consider the following
xed parameter scheme (compare with the expression in (7.62)):

= M (q)qd + C(q, q0 )qd + g(q) 1 (q0 qr )



Controller qr (t) = qd (t) 2 e(t)




3 q(t)
q0 (t) = q(t) (7.104)

= z(t) + 4 q(t) = z(t) + (6 + 3 )q(t)
q(t)
Observer

z(t) = qd (t) + 5 q(t) = qd (t) + 6 3 q(t)

where e = qqd (t) is the tracking error, q = qq is the estimation error, i > 0
3 q,
for all i = 1, , 6. Let us denote s1 = qqr = e+2 e and s2 = qq0 = q+
so that (q0 qr ) = s1 s2 . Introducing (7.104) into (6.98) and using some
properties of the matrix C(q, q) (like the fact that C(q, y)x = C(q, x)y and
C(q, z + x)y = C(q, z)y + C(q, x)y for all x, y IRn and IR) one gets
the following closed-loop error equation:

M (q(t))e(t) + C(q(t), q(t))s1 (t) + 1 s1 (t) = 1 s2 (t) + C(q(t), q(t))2 e(t)

C(q(t), s2 (t))qd (t)

e(t) = 2 e(t) + s1 (t)

M (q(t))s2 (t) + C(q(t), q(t))s2 (t) + [6 M (q(t)) 1 In ]s2 (t) = 1 s1 (t)+

+ C(q(t), s2 (t) q(t))e(t)

q(t) = 3 q(t) + s2 (t)

(7.105)
Dene K1 (q, e) = 22 [2 21 M (q)] and K2 (q, q) = 23 1 . It can be shown
using the positive denite function

V (e, s1 , q, s2 ) = 12 sT1 M (q)s1 + 12 eT K1 (q, e)e + 12 sT2 M (q)s2

(7.106)
+ 21 q T K2 (q, q)q

that for a suitable choice of the initial data within a ball Br whose radius r is
directly related to the control gains, the closed-loop xed point (e, s1 , q, s2 ) =
412 7 Passivity-based Control

(0, 0, 0, 0) is (locally) exponentially stable. As pointed out above r can actually

be varied by varying 6 or 1 , making the scheme semi-global. An intuitive
decomposition of the closed-loop system in (7.105) is as follows, noting that
M (q)e = M (q)s1 2 M (q)e:

M (q)s + C(q, q)s = u1 , q = 2 q + s1 , e = 3 e + s2
(7.107)

y1 = s, u2 = y1 , y2 = T (q, q, s) = u1

where

s
s= 1 (7.108)
s2

1 s2 + 2 C(q, q)e C(q, qd )s2 + 2 M (q)e
T (q, q, s) = (7.109)
1 s1 + C(q, s2 q)e

M (q) = diag[M (q), M (q)] (7.110)

C(q, q) = diag[C(q, q), C(q, q)] (7.111)
The rst subsystem is clearly passive with respect to the supply rate uT1 y1 .
The second subsystem is a memoryless operator u2
T (q, q, u2 ). If it can
be shown that locally uT2 T (q, q, u2 ) uT2 u2 , then the system with input
u = u1 + y2 and output y = y1 is output strictly passive. Indeed

u, yt = u1 + y2 , yt = u1 , y1 t + y2 , u2 t
t (7.112)
12 s(0)T M (q(0))s(0) + 0
uT2 (s)u2 (s)ds

for some > 0. In other words the function in (7.106) satises the dissipation
T
dx [f (x) + g(x)u] u h(x)
inequality along the closed-loop trajectories: dV T

hT (x)h(x) for all u and x locally only, where xT = (eT , sT1 , q T , sT2 ) and
y = h(x). Then under suitable zero-state detectability properties, any storage
function which is positive denite with respect to the closed-loop xed point
is a strict (local) Lyapunov function. Notice that the total closed-loop system
is zero-state detectable since y1 = s 0 and u 0 implies that y2 0, hence
u1 0 and e 0 and q 0 as t +.

7.4.3 The Slotine and Li + Observer Controller

Let us consider the following xed parameter scheme:

 7.4 Rigid JointRigid Link: Position Feedback 413


= M (q)qr + C(q, q0 )qr + g(q) 1 (q0 qr ) 2 e



Controller qr (t) = qd (t) (q(t) qd (t))




(q(t) q(t))
q0 (t) = q(t)

= z(t) + 3 (q(t) q(t))
q(t)
Observer

z(t) = qr (t) + 4 (q(t) q(t)) + 2 M 1 (q(t))[qd (t) q(t)]
(7.113)
Introducing (7.113) into (6.98) one obtains the closed-loop error equation

M (q(t))s1 (t) + C(q(t), q(t))s1 (t) + 1 s1 (t) + 2 e(t) =







= 1 s2 (t) C(q(t), s2 (t))qr (t)



= q(t) + s2 (t)
e(t) = (e(t) q(t)) + s1 (t), q(t)







M (q(t))s2 (t) + C(q(t), q(t))s2 (t) + (6 M (q(t)) 1 In )s2 (t) + 2 q(t) =





= 2 s1 (t) + C(q(t), s1 (t))[s2 (t) q(t)]
(7.114)
with 3 = 6 + , 4 = 6 . Again a natural decomposition of the closed-loop
scheme is similarly done as in the previous case, i.e.

M (q(t))s(t) + C(q(t), q(t))s
= u1 (t), e(t) = (e(t) q(t)) + s1 (t), q(t)

= q(t) + s2 (t)

y1 (t) = s(t), u2 (t) = y1 (t), y2 (t) = T (q(t), q(t), s(t)) = u1 (t)

(7.115)
where this time

1 s1 [1 + C(q, s1 q)]s2
T (q, q, s) = (7.116)
1 C(q, s2 q)s1 + [6 M (q) 1 In ]s2
t
It can be shown that locally T (q, q, s) > 0 so that u2 , y2 t 0 uT2 (s)u2 (s)ds
for some > 0. The same conclusions as above follow about semi-global
asymptotic Lyapunov stability of the closed-loop xed point.
414 7 Passivity-based Control

7.5 Flexible JointRigid Link: State Feedback

7.5.1 Passivity-based Controller: The Lozano and Brogliato
Scheme

In Section 6.4 we saw how the dissipativity properties derived for the rigid
joint-rigid link manipulator case extend to the exible joint-rigid link case,
and we presented what we called passivity-based schemes. Considering the
Lyapunov function in (7.78) let us try the following [70, 72, 316, 318]:
V (q1 , q2 , s1 , s2 ) = 12 sT1 M (q1 )s1 + 12 sT2 Js2 + 1 q1T q1 + 1 q2 q2
(7.117)
+ 21 (q1 q2 )T K (q1 q2 )
The various signals have the same denition as in the rigid case. One sees
that similarly to (7.78) this positive denite function mimics the total energy
function of the open-loop unforced system. In order to make it a Lyapunov
function for the closed-loop system, one can classically compute its derivative
along the trajectories of (6.105) and try to nd out a u that makes its deriva-
tive negative denite. Since we already have analyzed the rigid joint-rigid link
case, we can intuitively guess that one goal is to get a closed-loop system of
the form

M (q1 (t))s1 (t) + C(q1 (t), q1 (t))s1 (t) + 1 s1 (t) = f1 (s1 (t), s2 (t), q1 (t), q2 (t))

J s2 (t) + 1 s2 (t) = f2 (s1 (t), s2 (t), q1 (t), q2 (t))

(7.118)
For the moment we do not x the functions f1 () and f2 (). Since the Lyapunov
function candidate preserves the form of the systems total energy, it is also
to be strongly expected that the potential energy terms appear in the closed-
loop dynamics. Moreover we desire that the closed-loop system consists of
two passive blocks in negative feedback. Obviously V () in (7.117) contains
the ingredients for Lemmas 7.22
and 7.23 to apply. The rst block may be
q1
s1
chosen with state vector x1 =
q2 . We know it is passive with respect to the
s2  
s1 K(q1 q2 )
supply rate uT1 y1 with input u1 = and output y2 = .
s2 K(q1 q2 )
One storage function for this subsystem is

1 T 1
V1 (x1 , t) = s1 M (q1 )s1 + sT2 Js2 + 1 q1T q1 + 1 q2 q2 (7.119)
2 2
However notice that we have not xed the input and output of this subsystem,
since we leave for the moment f1 () and f2 () free. Now the second subsystem
must have a storage function equal to:
 7.5 Flexible JointRigid Link: State Feedback 415

1
V2 (x2 , t) = (q1 q2 )T K (q1 q2 ) (7.120)
2
and we know it is passive with respect to the supply rate uT2 y2 , with an
input u2 = y1 and an output y2 = u1 , and from (7.120) with a state vector
x1 = K (q1 q2 ). Its dynamics is consequently given by

x2 = x2 + K(s2 s1 ). (7.121)
In order for Lemmas 7.22 and 7.23 to apply we also require the CTCE to be
T V2 T
x1 G1 h2 = x2 G2 h1 , where we get from (7.118)
satised, i.e. V1

T
sT1 f1 + sT2 f2 = (q2 q1 ) K(s2 s1 ) (7.122)

from which one deduces that f2 (s1 , s2 , q1 , q2 ) = K(q1 q2 ) and f1 (s1 , s2 , q1 , q2 )

= K(q2 q1 ). Thus since we have xed the input and output of the second
subsystem so as to make it a passive block, we can deduce from Lemma 7.23
that the closed-loop system that consists of the feedback interconnection of
the dynamics in (7.118) and (7.121) can be analyzed through the passivity
theorem.
Notice however that we have not yet checked whether a state feedback
exists that assures this closed-loop form. This is what we develop now. Let us
consider the following controller:

u = J q2r + K(q2d q1d ) 1 s2
(7.123)

q2d = K 1 ur + q1d

where q2r = q2d q2 and ur is given by the rigid joint-rigid link controller
in (7.68), i.e.

ur = M (q1 )qr + C(q1 , q1 )qr + g(q1 ) 1 s1 (7.124)

It is noteworthy that the controller is thus formed of two controllers similar to

the one in (7.68): one for the rst rigid link subsystem and the other for the
motorschaft dynamics. The particular form of the interconnection between
them makes it possible to pass from the rst dynamics to the second one
easily. It should be noted that the form in (7.123) and (7.124) depends on the
state (q1 , s1 , q2 , s2 ) only, and not on any acceleration nor jerk terms.
To recapitulate, the closed-loop error dynamics is given by
416 7 Passivity-based Control

M (q1 (t))s1 (t) + C(q1 (t), q1 (t))s1 (t) + 1 s1 (t) = K(q2 (t) q1 (t))

J s2 (t) + 1 s2 (t) = K(q1 (t) q2 (t))

q1 (t) = q1 (t) + s1 (t)

q2 (t) = q2 (t) + s2 (t)

(7.125)

It is possible to replace the potential energy terms in (7.117) by

t T
t 
[s1 s2 ]d K [s1 s2 ]d (7.126)
0 0

This does not modify signicantly the structure of the scheme, apart from the
fact that this introduces a dynamic state feedback term in the control loop.
Actually as shown in [72] the static state feedback scheme has the advantage
over the dynamic one of not constraining the initial conditions on the open-
loop state vector and on q1d (0), q1d (0) and q1d (0). The stability of the scheme
with the integral terms as in (7.126) may be shown using the function
1 T 1 1
V (s1 , s2 , z) = s M (q1 )s1 + sT2 Js2 + z T Kz (7.127)
2 1 2 2
with

q2d = q1d x + K 1 (s1 + M (q1 )q1r + C(q1 , q1 )q1r + g(q1 ))







q1r (t) = q1d (t) q1 (t)



x(t) = q1 (t) q2 (t)







z(t) = x(t) + (q1 (t) q2(t) ) (z(t) = s1 (t) s2 (t))





u = s2 J[q2d + q2 ] K[q1d q2d x]

Then one gets along closed-loop trajectories V (s1 , s2 , z) = sT1 s1 sT2 s2 .

See [72] for more details.

Remark 7.33. A strong property of the controller in (7.123) and (7.124) in

closed-loop with the dynamics in (6.105), with the Lyapunov function in
(7.117), is that they converge towards the closed-loop system in (7.69) and
(7.70) when K + (all the entries diverge). Indeed one notices that
 7.5 Flexible JointRigid Link: State Feedback 417

K(q2d q1d ) = ur for all K and that q2d q1d as K . Noting that
all the closed-loop signals remain uniformly bounded for any K and introduc-
ing these results into u in (7.123) one sees that u = J qr + ur 1 s1 which is
exactly the controller in (7.68) applied to the system in (6.105), letting q1 q2
and adding both subsystems. We therefore have constructed a real family of
controllers that share some fundamental features of the plant dynamics.

A Recursive Algorithm Construction

A close look at the above developments, shows that the control scheme in
(7.123) and (7.124) is based on a two-step procedure:
The control of the rst equation in (6.105) using q2d as a ctitious input.
Since q2d is not the input, this results in an error term K(q2 q1 ).
A specic transformation of the second equation in (6.105) that makes the
control input u explicitly appear. The controller is then designed in such
a way that the closed-loop dynamics possesses a Lyapunov function as in
(7.117).

This is typically an instance of what has been called afterwards the back-
stepping design method and passivity-based controllers. It is the rst time
these two techniques have been applied simultaneously for tracking control
of Lagrangian systems.

Stability Proof

The stability proof for the xed parameters Lozano and Brogliato scheme,
very much mimics that of the Slotine and Li scheme. One may for instance
choose as a quadratic function

1 T 1 1 T
V (q1 , q2 , s1 , s2 ) = s1 M (q1 )s1 + sT2 Js2 + (q1 q2 ) K (q1 q2 ) (7.128)
2 2 2
instead of the Lyapunov function candidate in (7.117). The function in (7.128)
is the counterpart for exible joint systems, of the function in (7.80). Let us
compute the derivative of (7.128) along the trajectories of the error system
(7.125):
418 7 Passivity-based Control

V (q1 (t), q2 (t), s1 (t), s2 (t)) = sT1 (t)M (q1 (t))s1 (t) + sT2 (t)J s2 (t)+

+ 12 sT1 (t)M (q1 (t))s1 (t)

T  
+ (q1 (t) q2 (t)) K q1 (t) q2 (t)

= sT1 (t)[ 12 M (q1 (t)) C(q1 (t), q1 (t))s1 (t) 1 s1 (t)

+K(q2 (t) q1 (t))]+

T
sT2 [1 s2 (t) + K(q1 (t) q2 (t))] + (q1 (t) q2 (t))

K (1 q1 (t) + s1 (t) + 1 q2 (t) s2 (t))

= 1 sT1 (t)s1 (t) 1 sT2 (t)s2 (t)

1 (q1 (t) q2 (t))T K(q1 (t) q2 (t)) 0

(7.129)
It follows from (7.129) that all closed-loop signals are bounded on [0, +),
and that s1 L2 , s2 L2 . Using similar arguments as for the rst stability
proof of the Slotine and Li controller in Section 7.3.4, one concludes that
q1 (t), q2 (t), q1 (t) and q2 (t) all tend towards zero as t +. One may again
also conclude on the exponential convergence of these functions towards zero
noticing that V (q1 , q2 , s1 , s2 ) V (q1 , q2 , s1 , s2 ) for some > 0.
It is also possible to lead a stability analysis using the Lyapunov func-
tion candidate in (7.117). We reiterate that the quadratic function in (7.128)
cannot be named a Lyapunov function candidate for the closed-loop system
(7.125), since it is not a radially unbounded nor positive denite function of
the state (q1 , q2 , q1 , q2 ).

7.5.2 Other Globally Tracking Feedback Controllers

A Recursive Method for Control Design

As pointed out one may also view the passivity-based controller in (7.123) as
the result of a procedure that consists of stabilizing rst the rigid part of the
dynamics, using the signal q2d (t) as a ctitious intermediate input, and then
looking at the rest of the dynamics. However instead of looking at the rest
as a whole and considering it as a passive second order subsystem, one may
treat it step by step: this is the core of a popular method known under the
name of backstepping. Let us develop it now for the exible joint-rigid link
manipulators.
Step 1: Any type of globally stabilizing controller can be used. Let us still
use ur in (7.124), i.e. let us set
 7.5 Flexible JointRigid Link: State Feedback 419

q2d = K 1 ur + q1 (7.130)
so that we get

M (q1 (t))s1 (t) + C(q1 (t), q1 (t))s1 (t) + 1 s1 (t) = K q2 (t) (7.131)
The system in (7.131) with q2 0 thus denes a globally uniformly asymp-
totically stable system with Lyapunov function V1 (q1 , s1 ) = 12 sT1 M (q1 )s1 +
1 q1T q1 . The interconnection term is therefore quite simple (as long as the
stiness matrix is known!). Let us take its derivative to obtain
q2 (t) = q2 (t) q2d (t) = q2 (t) + f1 (q1 (t), q1 (t)q2 (t)) (7.132)
where f1 () can be computed using the dynamics (actually q2d is a function
of the acceleration q1 which can be expressed in terms of q1 , q1 and q2 by
simply inverting the rst dynamical equation in (6.105)).
Step 2: Now if q2 was the input we would set q2 = f1 (q1 , q1 q2 ) 2 q2
Ks1 so that the function V2 = V1 + 12 q2T q2 has a negative denite derivative
along the partial closed-loop system in (7.131) and
q2 (t) = 2 q2 (t) Ks1 (t) (7.133)
However q2 is not an input, so that we shall rather dene a new error
signal as e2 = q2 e2d , with e2d = f1 (q1 , q1 q2 ) 2 q2 Ks1 . One obtains

e2 (t) = q2 (t) e2d (t) = q2 (t) + f2 (q1 (t), q1 (t), q2 (t), q2 (t))

= J 1 [K(q1 (t) q2 (t)) + u(t)] + f2 (q1 (t), q1 (t), q2 (t), q2 (t))

(7.134)
Step 3: Since the real control input appears in (7.134) this is the last step.
Let us choose
u = K(q2 q1 ) + J[f2 (q1 , q1 , q2 , q2 ) e2 q2 ] (7.135)
so that we get:
e2 (t) = 3 e2 (t) q2 (t) (7.136)
where the term q2 has been chosen to satisfy the CTCE (see Lemma
7.23) when the function V2 is augmented to
1
V3 (q1 , s1 , q2 , e2 ) = V2 + eT2 e2 (7.137)
2
Then along the closed-loop trajectories of the system in (7.131) (7.118)
(7.136) one gets
V3 (q1 (t), s1 (t), q2 (t), e2 (t)) = 1 q1T (t)q1 (t) 2 1 q1T (t)q1 (t)
(7.138)
q2T (t)q2(t) eT2 (t)e2 (t)
which shows that this closed-loop system is globally uniformly exponen-
tially stable.
420 7 Passivity-based Control

It is noteworthy that e2 is not the time derivative of q2 . Therefore the back-

stepping method hinges upon a state variable transformation which actually
depends on the system dynamics in the preceding steps.

Remark 7.34. The control law in (7.135) can be computed from the deni-
tion of q2d in (7.130) and q2d as well as q2d are to be calculated using the
(3)
dynamics to express the acceleration q1 and the jerk q1 as functions of
positions and velocities only (take the rst dynamical equation in (6.105)
and invert it to get the acceleration. Dierentiate it again and introduce
the expression obtained for the acceleration to express the jerk). Clearly
u is a complicated nonlinear function of the state, but it is a static state
feedback. This apparent complexity is shared by all the nonlinear con-
trollers described in Section 7.5. Notice however that it is only a matter
of additions and multiplications, nothing else!
We noticed in Remark 7.33 that the passivity-based controller tends to-
wards the Slotine and Li input when the joint stiness tends to innity.
This is no longer the case with the backstepping controller derived here.
Even more, after some manipulations, it can be shown [78] that the con-
troller in (7.135) can be equivalently rewritten as

u = J[q2d (2 + 3 )q2 (1 + 2 3 )q2 K(s1 + s1 )]
(7.139)

q2d = K 1 ur + q1

where it immediately appears that the term K(s1 + s1 ) is not bounded

as K grows without bound. Here comes into play the exibility of the
backstepping method: let us modify the function V2 above to V2 = V1 +
1 T
2 q2 K q2 . Then in step 2 it is sucient to choose q2 = f1 (q1 , q1 q2 )
2 q2 s1 , so that the nal input becomes

u = J[q2d (2 + 3 )q2 (1 + 2 3 )q2 (s1 + s1 )]
(7.140)

q2d = K 1 ur + q1

Such a modication may appear at rst sight quite innocent, easy to do,
and very slight: it is not! The experimental results presented in Chapter
9 demonstrate it. Actually the term K(s1 + s1 ) introduces a high-gain
in the loop that may have disastrous eects. This may be seen through
simulations, see [78]. It is noteworthy that even with quite exible systems
(some of the reported experiments were led with a system whose stiness
is k = 3.5 Nm/rad) this term makes the control law in (7.135) behave
less satisfactorily than the one in (7.140). More details can be found in
Chapter 9.
This recursive design method applies to all systems that possess a trian-
gular structure [325]. See [67] for a survey of backstepping methods for
exible joint manipulators.
 7.5 Flexible JointRigid Link: State Feedback 421

Compare (7.139) and (7.140) to (7.123). Although these controllers have

the same degree of complexity and can be considered as similar, they have
signicant discrepencies as explained above. For instance in (7.123) one
has K(q2d q1d ) = ur while in (7.139) and (7.140), K(q2d q1d ) = ur + q1 .

A Passivity Theorem Interpretation

As we pointed out the procedure relies on the CTCE at each step. Since the
rst subsystem in (7.131) is output strictly passive with respect to the supply
rate uT1 y1 with u1 = K q2 and y1 = s1 , we are tempted to apply the result of
Lemmas 7.22 and 7.23 to interpret the closed-loop scheme in (7.131), (7.118)
and (7.136) as an interconnection of passive blocks. From the developments
concerning the rigid joint-rigid link case we know that the rst subsystem
can be seen as the interconnection of two passive blocks in (7.69) and (7.71).
However, now the rst subsystem is passive when the input is changed to
u1 = K q2 1 s1 . We shall therefore dene four subsystems as follows:

(H11) : M (q1 (t))s1 (t) + C(q1 (t), q1 (t))s1 (t) = K q2 (t) 1 s1 (t)






u11 (t) = K q2 (t) 1 s1 (t), y11 (t) = s1 (t), state s1
(H1)



(H12) : q1 (t) = 1 q1 (t) + s1 (t)





u12 (t) = s1 (t), y12 (t) = 1 s1 (t), state q1


(H21) : q2 (t) = 2 q2 (t) + e2 (t) Ks1 (t)






u21 (t) = e2 (t) Ks1 (t), y21 (t) = q2 (t), state q2
(H2)



(H22) : e2 (t) = 3 e2 (t) q2 (t)





u22(t) = q2 (t), y22 (t) = e2 (t), state e2
(7.141)
Then the closed-loop system can be viewed as the negative feedback inter-
connection of the block (H1) with u1 = u11 + y12 = K q2 , y1 = y11 , with the
block (H2) with input K 1 u2 = s1 = y1 and output Ky2 = K q2 = u1 .
This is depicted in Figure 7.7.
Remark 7.35. The backstepping procedure also yields a closed-loop system
that can be analyzed through the passivity Theorem. However the major
dierence with the passivity-based method is that the block (H2) is not related
to any physical relevant energetical term. In a sense this is similar to what
one would get by linearizing the rigid jointrigid link dynamics, applying a
new linear feedback so as to impose some second order linear dynamics which
may dene an articial passive system.
422 7 Passivity-based Control

u1 u11 y11
y1
_ H11
_
y12
H12
u12

-K -K

u21
y2 u2
H21
y21 _

u22
H22 y22

Fig. 7.7. Flexible joint-rigid link (equivalent interpretation)

7.6 Flexible JointRigid Link: Output Feedback

7.6.1 PD Control

We have seen in Section 7.3.1 that a PD controller stabilizes globally and

asymptotically rigid joint-rigid link manipulators. It is a combination of pas-
sivity and detectability properties that makes such a result hold: the former
is a guide for the choice of a Lyapunov function, while the latter allows the
Krasovskii-La Salle invariance principle to apply. More precisely, the output
strict passivity property is crucial, because output strict passivity together
with zero-state detectability of a system, imply its asymptotic stability in
the sense of Lyapunov (see Corollary 5.16). Let us consider the dynamics in
(6.105) and the following controller:

u = 1 q2 2 (q2 qd ) (7.142)

with qd a constant signal, so that the closed-loop system is given by

 7.6 Flexible JointRigid Link: Output Feedback 423

M (q1 (t))q1 (t) + C(q1 (t), q1 (t))q1 (t) + g(q1 (t)) = K(q2 (t) q1 (t))
(7.143)

J q2 (t) + 1 q2 (t) + 2 (q2 (t) qd ) = K(q1 (t) q2 (t))

Let us proceed as for the rigid joint-rigid link case, i.e. let us rst guess
a Lyapunov function candidate from the available storage function, and then
show how the application of the passivity Theorem applies equally well.

The Closed-loop Available Storage

Similarly as for the rigid joint-rigid link case, one may guess that a PD con-
troller alone will not enable one to stabilize any xed point. The closed-loop
xed point is given by

g(q1 ) = K(q2 q1 )
(7.144)

2 (q2 qd ) = K(q1 q2 )
and we may assume for simplicity that this set of nonlinear equations (which
are not in general algebraic but transcendental) possesses a unique root
(q1 , q2 ) = (q10 , q20 ). We aim at showing the stability of this point. To com-
pute the available storage of the closed-loop system in (7.143) we consider a
ctitious input u in the second dynamical equation, while the output is taken
as q2 . Then we obtain the following:

Z t
Va (q1 , q1 , q2 , q2 ) = sup q2T (s)u(s)ds
u:(0,q1 (0),q1 (0),q2 (0),q2 (0)) 0

Z t
= sup uT [J q2 + K(q2 q1 ) + 1 q2 + 2 q2 ]ds
u:(0,q1 (0),q1 (0),q2 (0),q2 (0)) 0 (7.145)

= 12 q1 (0)T M (q1 (0))q1 (0) + Ug (q1 (0)) + 12 q2 (0)T J q2 (0)+

+ 12 (q2 (0) q1 (0))T K(q2 q1 ) + 12 2 q2T (0)q2 (0)

where qi = qi qi0 , i = 1, 2. Now the supply rate satises w(0, q2 ) 0 for

all q2 , and obviously (q1 , q1 , q2 , q2 ) = (0, 0, 0, 0) is a strict (global) minimum
of Va in (7.145) provided Ug (q1 ) has a strict minimum at q10 . Notice that
q2 = 0 (q1 q2 ) = 0 g(q1 ) = 0 so that q1 = q10 is a critical point for
Ug (q1 ) (that we might assume to be strictly globally convex, but this is only
sucient). Hence from Lemmae 5.13 and 4.8 one deduces that the closed-loop
system in (7.143) is Lyapunov stable. To show asymptotic stability, one has
to resort to the Krasovskii-La Salle invariance principle.
424 7 Passivity-based Control

Closed-loop Feedback Interconnections

Motivated by the rigid joint-rigid link case let us look for an equivalent feed-
back interconnection such that the overall system is strictly output passive
and zero-state detectable. To this end let us consider the following two blocks:

u1 = K(q1 q2 ), y1 = q2
(7.146)

u 2 = y1 , y2 = u1
where the rst block has the dynamics J q2 (t) + 1 q2 (t) + 2 (q2 (t) qd ) =
K(q1 (t) q2 (t)), while the second one has the dynamics M (q1 (t))q1 (t) +
C(q1 (t), q1 (t))q1 (t) + g(q1 (t)) = K(q2 (t) q1 (t)). It is easy to calculate the
following:
u1 , y1 t 21 q2 (0)T J q2 (0) 2 (q2 (0) qd )T (q2 (0) qd )
t
+1 0 q2T (s)q2 (s)ds
(7.147)
u2 , y2 t 21 [q1 (0) q2 (0)]T K[q1 (0) q2 (0)]

12 q1 (0)T M (q1 (0))q1 (0) Ug (q1 (0))

from which one deduces that the rst block is output strictly passive (actually
if we added Rayleigh dissipation in the rst dynamics, the second block would
not be output strictly passive with the proposed decomposition). Each block
possesses its own storage functions which are Lyapunov functions for them.
The concatenation of these two Lyapunov functions forms the available storage
in (7.145). Let us now consider the overall system with input u = u1 + y2 and
output y = y1 . Setting u y 0 implies q2 0 and q1 0, q1 0
asymptotically. The system is zero-state detectable. Hence by Lemmae 5.13
and 4.8 its xed point is globally asymptotically Lyapunov stable.
Remark 7.36 (Collocation). The collocation of the sensors and the actuators
is an important feature for closed-loop stability. It is clear here that if the PD
control is changed to
u(t) = 1 q1 (t) 2 q1 (t) (7.148)
then the above analysis no longer holds. It can even be shown that there
are some gains for which the closed-loop system is unstable [473]. One choice
for the location of the sensors may be guided by the passivity property be-
tween their output and the actuators torque (in case the actuator dynamics
is neglected).

7.6.2 Motor Position Feedback

A position feedback controller similar to the one in Section 7.4.1 can be derived
for exible joint-rigid link manipulators [43]. It may be seen as a PD controller
with the velocity feedback replaced by an observer feedback. It is given by
 7.6 Flexible JointRigid Link: Output Feedback 425
1
u(t) = g(qd ) 1 q2 (t) 2 (q2 (t) z(t))
(7.149)

z(t) = 3 (q2 (t) z(t))

with q2 = q2 qd +K 1 g(qd ), and qd is the desired position for q1 . The analysis

is quite close to the one done for the rigid joint-rigid link case. Due to the
autonomy of the closed-loop (qd is constant) Corollary 7.25 is likely to apply.
The stability proof bases on the following global Lyapunov function:

V (q1 , q1 q2 , q2 ) = 2 12 q1T M (q1 )q1 + 12 q2T J q2 + 12 q1T K q1
 (7.150)
+ 21 q2T (K + 1 In )q2 22 q1T K q2 + 12 (q2 z)T (q2 z)
Compare with V () = V1 () + V2 () in (7.99) and (7.100): the structure of
V () in (7.150) is quite similar. It is a positive denite function provided
 1
K + dg(q)
dq (qd ) > 0 and 1 In + K K K + dq (qd )
dg(q)
> 0, for all qd .
This implies that K and 1 are suciently large. The decomposition into
two subsystems as in (7.98) can be performed, choosing x2 = q2 z and
xT1 = (q1T , q1T , q2T , q2T ) = (xT11 , xT12 , xT13 , xT14 ). The closed-loop scheme is given
by

x11 (t) = x12 (t)

x12 (t) = M (x11 (t) + qd )[C(x11 (t) + qd , x12 (t))x12 (t) + K(x11 (t) x12 (t))

+g(x11 (t) + qd ) g(qd )]

x13 (t) = x14 (t)

x14 (t) = J 1 [K(x11 (t) x13 (t)) g(qd ) 1 x13 (t) 1

2 x2 (t)]

x2(t) = 3 x2(t) + x14 (t)

(7.151)
1
Dene h2 (x2 ) = 2 x2 and h1 (x1 ) = x14 . It follows that the CTCE is satis-
T T
x1 G1 h2 = x14 x2 = x2 G2 h1 . Indeed one may calculate that
ed since V 1 T V2

GT1 = (0, 0, 0, J 1 ) IRn4n whereas G2 = In . Hence once again Corollary

7.25 applies and the closed-loop system can be interpreted via the passivity
theorem.
Remark 7.37. Battilotti et al [44] have presented a result that allows one
to recast the dynamic position feedback controllers presented in this sub-
section and in Section 7.4 into the same general framework. It is based
on passiability and detectability properties. The interpretation of the P
+ observer schemes in Subsections 7.4.1 and 7.6.2 via Corollary 7.25 is
however original.
426 7 Passivity-based Control

It is also possible to derive a globally stable P + observer controller using

only the measurement of q1 [44]. Its structure is however more complex
than the above one. This is easily understandable since in this case the ac-
tuators and sensors are non-collocated. Energy shaping is used in [254] to
globally stabilize exible jointrigid link manipulators. PD control for ex-
ible jointrigid link manipulators with disturbances and actuatir dynamics
is analysed in [315].

7.7 Including Actuator Dynamics

7.7.1 Armature-controlled DC Motors

We have seen in Section 6.6 that the available storage of the interconnec-
tion between the rigid joint-rigid link manipulator model and the armature-
controlled DC motor is given by
1 T 1
Va (q, q, I) = I LI + q T M (q)q + Ug (q) (7.152)
2 2
Motivated by the method employed for the design of stable controllers for
rigid joint-rigid link and exible joint-rigid link manipulators, let us consider
the following positive denite function:

= 1 IT LI + 1 sT M (q)s + +21 q T q
V (q, s, I) (7.153)
2 2

where s = q + q. Let us consider the dynamics in (6.126) which we recall

here for convenience:

M (q(t))q(t) + C(q(t), q(t))q(t) + g(q(t)) = (t) = Kt I(t)
(7.154)

RI(t) + L dI
dt (t) + Kt qdt(t) = u(t)

Let us set

Id = Kt1 [M (q)qr + C(q, q)qr + g(q) 1 s] (7.155)

where s = q qr , so that the manipulator dynamics in (6.126) becomes

M (q(t))s(t) + C(q(t), q(t))s(t) + 1 s(t) = Kt I(t) (7.156)

where I = I Id . Then it is easy to see that the control input

u = RI kv q + L1 Id L1 Kt s I (7.157)

(which is a state feedback) leads to

 7.7 Including Actuator Dynamics 427

= I(t)
I(t) + L1 Kt s(t) (7.158)
in (7.153) along closed-loop trajectories in
Taking the derivative of V (q, s, I)
(7.156) and (7.158) one gets:

V (q(t), s(t), I(t)) = IT (t)LI(t) 2 1 q T (t)q(t)
1 qT (t)q(t) (7.159)
= (0, 0, 0) is globally
showing that the closed-loop xed point (q, s, I)
asymptotically uniformly stable in the sense of Lyapunov.
Remark 7.38 (Regulation of cascade systems). Consider the system in (7.154)
with Rayleigh dissipation in the manipulator dynamics. Let us write the sec-
ond subsystem in (7.154) as
= L1 RI(t) L1 Kt q(t) + L1 u(t)
I(t) (7.160)

Let L1 u = L1 Kv q + u so that we obtain the cascade system

dR

M (q(t))q(t) + C(q(t), q(t))q(t) + g(q(t)) + dq (t) = Kt y(t)



= L1 RI(t) + u(t)
I(t) (7.161)





y(t) = I(t)

The terms corresponding to (5.52) can be easily identied by inspection.

One sees that the conditions of Theorem 5.42 are satised (provided the po-
tential energy U (q) satises the requirements of Assumption 15), so that
this (partially) closed-loop system is feedback equivalent to a strictly pas-
sive system. In other words there exists a feedback input u = (I, q, q) + v
such that there exists a positive denite function V (I, q, q) of the xed point
(I, q, q) = (0, 0, 0) and a positive denite function S(I, q, q) such that

t
t
V (t) V (0) = vT (s)y(s)ds S(I(s), q(s), q(s))ds (7.162)
0 0

Thus the unforced system (i.e. take v = 0) has a globally asymptotically

stable xed point (in the sense of Lyapunov).
A similar analysis for the eld-controlled DC motor case can be led. The
dissipativity properties of the driven and the driving subsystems allow the
designer to construct a globally stabilizing feedback law.
Remark 7.39 (Nested passive structure). The computation of V () relies on a
CTCE as required in Lemma 7.23 and Corollary 7.25. Thus if we had started
from the a priori knowledge of the function V () we could have deduced that
the closed-loop system can be analyzed as the negative feedback interconnec-
tion of two passive blocks, one with input u1 = Kt I and output y1 = s and
428 7 Passivity-based Control

dynamics in (7.156), the second one with dynamics in (7.158) and u2 = y1 ,

y2 = u1 . Recall from Section 7.3.4 that the rst subsystem can be in turn de-
composed as a negative feedback interconnection of two passive blocks given in
(7.69) and (7.70): the overall system therefore possesses a structure of nested
negative feedback interconnections of passive systems.

7.7.2 Field-controlled DC Motors

Let us recall the model of rigid joint-rigid link manipulators in cascade with
a eld-controlled DC motor:

dI1

L1 dt (t) + R1 I1 (t) = u1 (t)



L2 dI2
dt (t) + R2 I2 (t) + Kt (I1 (t))q(t) = u2 (t)




q(t) + g(q(t)) + Kvt q(t) = = Kt (I1 (t))I2 (t)
M (q(t))q(t) + C(q(t), q(t))
(7.163)
The regulation problem around the constant xed points (q, q, I1 , I2 ) =
(q0 , 0, I1d , 0) or (q0 , 0, 0, I2d ) is solvable, where q0 is as in assumption 15. In-
deed the subsystem can be seen as a cascade system as in (5.52) that satises
the requirements of Theorem 5.42. Hence it is feedback equivalent to a strictly
passive system (in the sense of Theorem 5.36), whose unforced version is Lya-
punov globally asymptotically stable. One remarks that the tracking control
problem is quite similar to that of the exible joint-rigid link manipulators
with torque input. However this time the matrix that premultiplies I2 is no
longer constant invertible. Actually Kt (I1 ) may pass through singular values
each time I1i = 0 for some i {1, , n}. The extension of the regulation
case is therefore not trivial. Nevertheless if the goal is to track a reference
trajectory for (q, q) only, then one may keep I1 constant such that Kt (I1 )
remains full-rank, through a suitable u21 , so that the armature-controlled DC
motor case is recovered.

Remark 7.40. All the preceeding developments apply to exible joint-rigid link
manipulators. Notice also that induction motors have the same complexity as
eld-controlled DC motors for control since the generated torque for each
motor is given by = Lsr (I2 I3 I1 I4 ); see Remark 6.50 for details.

7.8 Constrained Mechanical Systems

In real robotic tasks, the manipulators seldom evolve in a space free of ob-
stacles. A general task may be thought as involving free-motion as well as
constrained motion phases. In this section we shall focus on the case when
the system is assumed to be in a permanent contact with some environment.
 7.8 Constrained Mechanical Systems 429

In other words the constraint between the controlled system and the obstacle
is supposed to be bilateral. In all the sequel we assume that the potential
energy of the controlled system Ug (z) and of the passive environment Uge (z1 )
each have a unique strict minimum, and to simplify further that they are
positive (they have been chosen so).

7.8.1 Regulation with a Position PD Controller

Before going on with particular environment dynamics, let us analyze the

regulation problem for the system in (6.169). To this end let us dene the PD
control
= 2 z 1 z (7.164)
where z = z(t) zd , zd a constant signal. Since we have assumed that the
constraints are bilateral, we do not have to restrict zd to a particular domain
of the state space (i.e. we do not care about the sign of the interaction force).
Let us invent a Lyapunov function candidate by mimicking the available
storage in (6.170), i.e.

V (z, z, z1 ) = 12 z T M (z)z + 12 z1 Me (z1 )z1

(7.165)
+ 12 2 z T z + Ug (z) + Uge (z1 ) + 12 z1T Ke z1

Instead of computing the derivative of this function along the closed-loop

system (6.169) and (7.164), let us decompose the overall system into two
blocks. The rst blockcontains the controlled subsystem dynamics, and has
z
input u1 = Fz = , output y1 = z. The second block has the dynamics of
0
the environment, output u2 = z and input u2 = z. These two subsystems
are passive since
t  
u1 , y1 t = 0 z T M (z)z + C(z, z)z + g(z) + 2 z + 2 z ds
(7.166)
12 z(0)T M (z(0))z(0) Ug (z(0)) 12 2 z(0)T z(0)

and
u2 , y2 t =
t  
dRe
= 0
z1T Me (z1 )z1 + Ce (z1 , z1 )z1 + dz1 + Ke z1 + ge (z1 ) ds (7.167)

12 z1 (0)T Me (z1 (0))z1 (0) 12 z1 (0)T Ke z1 (0) Uge (z1 (0))

Now the inputs and outputs have been properly chosen so that the two sub-
systems are already in the required form for the application of the passivity
theorem. Notice that they are both controllable and zero-state detectable from
430 7 Passivity-based Control

the chosen inputs and outputs. Therefore the storage functions that appear
in the right-hand-sides of (7.166) and (7.167) are Lyapunov functions (see
Lemmae 5.13 and 4.8) and their concatenation is the Lyapunov function can-
didate in (7.165) which is a Lyapunov function. The asymptotic stability of
the closed-loop system xed point can be shown using the Krasovskii-La Salle
Theorem, similarly to the case of rigid joint-rigid link manipulators controlled
by a PD feedback. Notice that similarly to (7.27) the xed points are given
as solutions of the following equation (obtained by summing the dynamics of
the two subsystems)
 
Ke z1 + g1 (z) + ge (z1 ) + 2 z1 0m1
= (7.168)
2 z2 + g2 (z) 0(nm)1

We may assume that this equation has only one root z = zi so that the xed
point (z, z) = (zi , 0) is globally asymptotically stable.

Remark 7.41. It is noteworthy that this interpretation works well because the
interconnection between the two subsystems satises Newtons principle of
mutual actions. The open-loop system is therefore ready for a decomposition
through the passivity theorem.

Remark 7.42. Let us note that there is no measurement of the environment

state in (7.164). The coordinate change presented in Section 6.7.2 just allows
one to express the generalized coordinates for the controlled subsystem in a
frame that coincides with a natural frame associated to the obstacle. It is
clear however that the transformation relies on the exact knowledge of the
obstacle geometry.

The next step, that consists of designing a passivity-based nonlinear con-

troller guaranteeing some tracking properties in closed-loop, has been per-
formed in [320]. It has been extended in [368] when the geometry of the
obstacle surface is unknown (it depends on some unknown parameters) and
has to be identied (then an adaptive version is needed). Further works using
closed-loop passivity may be found in [296, 297].

7.8.2 Holonomic Constraints

Let us now analyze the case when Me (z1 )z1 = 0 and the contact stiness Ke
and damping Re (z1 ) tend to innity, in which case the controlled subsystem
is subject to a bilateral holonomic constraint (q) = 0 4 . In the transformed
coordinates (z1 , z2 ) the dynamics is given in (6.163); see Subsection 6.7.1. We
saw that the open-loop properties of the unforced system transport from the
4
Actually the way these coecients tend to innity is important to pass from
the compliant case to the rigid body limit. This is analyzed for instance in [323]
through a singular perturbation approach.
 7.8 Constrained Mechanical Systems 431

free-motion to the reduced constrained motion systems. Similarly, it is clear

that any feedback controller that applies to the dynamics in (6.98) applies
equally well to the reduced order dynamics (z2 , z2 ) in (6.163). The real prob-
lem now (which has important practical consequences) is to design a controller
such that the contact force tracks some desired signal. Let us investigate the
extension of the Slotine and Li scheme in this framework. The controller in
(7.68) is slightly transformed into

1 = M12 z2r + C12 (z2 , z2 )z2r + g1 2 zd
(7.169)

2 = M22 z2r + C22 (z2 , z2 )z2r + g2 2 s2
where all the terms keep the same denition as for (7.68). d is some
desired value for the contact force z1 . The closed-loop system is therefore
given by

M12 (z2 (t))s2 (t) + C(z2 (t), z2 (t))s2 (t) = 2 (z1 (t) d )




M22 (z2 (t))s2 (t) + C(z2 (t), z2 (t))s2 (t) + 1 s2 (t) = 0 (7.170)





z2 (t) = z2 (t) + s2 (t)

The dissipativity properties of the free-motion closed-loop system are sim-

ilar to those of (7.69) and (7.70). Notice that due to the asymptotic stability
properties of the xed point (z2 , s2 ) one gets z1 (t) d (t) as t +.

7.8.3 Nonsmooth Lagrangian Systems

In practice one often has to face unilateral or inequality constraints where

(6.162) is replaced by (q) 0, which models the fact that contact may be
lost or established with obstacles. Let us just point out that this yields to
nonsmooth systems containing impact rules (or state reinitializations) and
so-called complementarity relationships between z1 and z1 , of the form

z1 0, z1 0, Tz1 z1 = 0 (7.171)

The inclusion of such complementarity conditions into the dynamics, yields

a Lagrangian complementarity system as (6.177) or a measure dierential in-
clusion as in (6.180) . See Section 6.8.2 for more developments on nonsmooth
systems. The trajectory tracking problem for such systems has been studied
in [60, 75, 76]. Specic stability notions are developed that take into account
the subtleties of this problem. For instance the times of rst impacts when one
wants to stabilize the system on a surface z1 = 0 is ususally unknown, as well
as the time of detachment from this surface. It is shown in [60] that the Slotine
and Li controller is a suitable basic nonlinear controller to achieve the sta-
bility requirements developed therein, because of its exponential convergence
432 7 Passivity-based Control

property, and also because the quadratic Lyapunov function (7.80) is close
to the kinetic energy of the open-loop system (consequently it should possess
nice properties at impacts, following the kinetic energy variation in (6.178)).
A switching or hybrid Slotine and Li controller is designed in [60].
More details on nonsmooth mechanical systems dynamics and control can
be found in [69, 75, 76, 164].

7.9 Controlled Lagrangians

Until now we have focussed in this chapter on passivity-based controllers, de-
signed for trajectory tracking and adaptive control. Let us briey introduce
the method of controlled Lagrangians. As said in the introduction of the chap-
ter, the objective is to shape both the kinetic and potential energies, with a
suitable feedback. Let us describe the method in the simplest case, i.e. a fully
actuated Lagrangian system

M (q(t))q(t) + C(q(t), q(t))q(t) + g(q) = (7.172)

The objective is to design in such a way that the closed-loop system
becomes

Mc (q(t))q(t) + Cc (q(t), q(t))q(t) + gc (q) = 0 (7.173)

where Mc (q) is a desired kinetic tensor, and gc (q) = Uc (q) where Uc (q) is a
desired potential energy. Let us propose

= M (q)Mc1 (q)[Cc (q, q)q gc (q)] + C(q, q)q + g(q) (7.174)

Injecting (7.174) into (7.172) one obtains

M (q(t))q(t) = M (q(t))Mc1 (q(t))[Cc (q(t), q(t))q(t) gc (q(t))] (7.175)

Since M (q) is full rank one can rewrite (7.175) as (7.173). The fully actu-
ated case is therefore quite trivial, and the methods owns its interest to the
underactuated case. Let us therefore consider

M (q(t))q(t) + C(q(t), q(t))q(t) + g(q) = G(q(t) (7.176)

for some n m matrice G(q) with rank(G(q)) = m for all q IRn .
There exists a matrix G (q) such that G (q)G(q) = 0 for all q. Also
Im(G (q))+Im(GT (q)) = IR2n , and both subspaces are orthogonal. It is thus
equivalent to rewrite (7.176) as
 7.9 Controlled Lagrangians 433

G (q){M (q(t))q(t) + C(q(t), q(t))q(t) + g(q)} = 0
(7.177)

GT (q){M (q(t))q(t) + C(q(t), q(t))q(t) + g(q)} = GT (q)G(q)

where one notices that GT (q)G(q) is a m m invertible matrix. Obviously

the same operation may be applied to the objective system, i.e.

G (q){Mc (q(t))q(t) + Cc (q(t), q(t))q(t) + gc (q)} = 0
(7.178)

GT (q){Mc (q(t))q(t) + Cc (q(t), q(t))q(t) + gc (q)} = 0
One says that the two systems (7.177) and (7.178) match if they possess
the same solutions for any initial data (q(0), q(0)). It is easy to see that by
choosing

) *
= (GT (q)G(q))1 GT (q) M (q)Mc1 (q)[Cc (q, q)q gc (q)] + C(q, q)q + g(q)
(7.179)
one obtains

GT (q){Mc (q(t))q(t) + Cc (q(t), q(t))q(t) + gc (q)} = 0. (7.180)

It then remains to examine what happens with the rest of the closed-
loop dynamics. Matching between (7.173) and (7.176) occurs if and only if
G (q){Mc (q(t))q(t) + Cc (q(t), q(t))q(t) + gc (q)} = 0 holds along the solutions
of the closed-loop system (7.176) and (7.179). In other words matching occurs
if and only if

M (q(t))q(t) + C(q(t), q(t))q(t) + g(q) G(q(t) =

(7.181)
= Mc (q(t))q(t) + Cc (q(t), q(t))q(t) + gc (q)
Note that if there is matching then we can also express the acceleration as

q = M 1 (q)G(q) M 1 (q)[C(q, q)q + g(q)]

(7.182)
Mc1 (q)[Cc (q, q)q + gc (q)]
so that

G(q) = M (q)Mc (q)[Cc (q, q)q + gc (q)] + C(q, q)q + g(q) (7.183)
and premultiplying by G (q) one gets

G (q) {M (q)Mc (q)[Cc (q, q)q + gc (q)] + C(q, q)q + g(q)} = 0 (7.184)
Consequently matching between (7.173) and (7.176) occurs if and only if
(7.184) holds and is as in (7.179).
434 7 Passivity-based Control

Remark 7.43. All these developments may be led within a dierential geome-
try context [55]. This does not help in understanding the underlying simplicity
of the method (on the contrary it may obscure it). However it highlights the
fact that the equality in (7.184) is in fact a partial dierential equation for
Mc (q) and Uc (q). Consequently the controlled Lagrangian method boils down
to solving a PDE.
8
Adaptive Control

This chapter is dedicated to present so-called direct adaptive controllers ap-

plied to mechanical and to linear invariant systems. We have already studied
some applications of dissipativity theory in the stability of adaptive schemes
in Chapters 1, 2, 3 and 4. Direct adaptation means that one has been able
to rewrite the xed parameter input u in a form that is linear with re-
spect to some unknown parameters, usually written as a vector IRp ,
i.e. u = (x, t), where (x, t) is a known matrix (called the regressor) func-
tion of measurable 1 terms. The parameters i , i {1, , p}, are generally
nonlinear combinations of the physical parameters (for instance in the case
of mechanical systems, they will be nonlinear combinations of moments of
inertia, masses). When the parameters are unknown, one cannot use them in
the input. Therefore one replaces in u by an estimate, that we shall denote
in the sequel. In other words, u = (x, t) is replaced by u = (x, t) at the
input of the system, and is estimated on-line with a suitable identication
algorithm. As a consequence, one easily imagines that the closed-loop system
stability analysis will become more complex. However through the passivity
theorem (or the application of Lemma 7.23) the complexity reduces to adding
a passive block to the closed-loop system that corresponds to the estimation
algorithm dynamics. The rest of the chapter is composed of several examples
that show how this analysis mechanism work. It is always assumed that the
parameter vector is constant: the case of time-varying parameters, although
closer to the reality, is not treated here due to the diculties in deriving stable
adaptive controllers in this case. This is a topic in itself in adaptive control
theory and is clearly outside the scope of this book.

1
In the technological sense, not in the mathematical one.
436 8 Adaptive Control

8.1 Lagrangian Systems

8.1.1 Rigid JointRigid Link Manipulators

In this subsection we rst examine the case of a PD controller with an adaptive

gravity compensation. Indeed it has been proved in Section 7.3.1 that gravity
hampers asymptotic stability of the desired xed point, since the closed-loop
system possesses an equilibrium that is dierent from the desired one. Then
we pass to the case of tracking control of n degree-of-freedom manipulators.

PD + Adaptive Gravity Compensation

A First Simple Extension

Let us consider the following controller + estimation algorithm:

(t) = 1 q(t) 2 q(t) + Yg (q(t))g (t)
(8.1)

(t) = Y T q(t)
g 3 g

where we suppose that the gravity generalized torque g(q) = Yg (q)g for some
known matrix Yg (q) IRnp and unknown vector g , and g = g g . The
estimation algorithm is of the gradient type, and we know from Subsection
4.3.1 that such an estimation law denes a passive operator q
gT Yg (q),
with storage function V2 (g ) = 12 T . This strongly suggests one should de-
compose the closed-loop system obtained by introducing (8.1) into (6.98) into
two blocks as follows:


M (q(t))q(t) + C(q(t), q(t))q(t) + 1 q(t) + 2 q(t) = Yg (q(t))(t)
(8.2)

(t) = Y T (t)q(t)
g 3 g

Obviously the rst block with the rigid joint-rigid link dynamics and input
u1 = Yg (q)(= y2 ) and output y1 = q(= u2 ) denes an output strictly
passive operator with storage function V1 (q, q) = 12 q T M (q)q + 22 q T q; see
Subsection 7.3.1. One is tempted to conclude about the asymptotic stability
with a Lyapunov function V (q, q, ) = V1 (q, q) + V2 (g ). However notice that
the overall system with input u = u1 + y2 and output y = y1 , although
output strictly passive, is not zero-state detectable. Indeed u y 0 implies
2 q = Yg (q)g and g = 0, nothing more. Hence very little has been gained
by adding an estimation of the gravity, despite the passivity theorem applies
well.
 8.1 Lagrangian Systems 437

How to Get Asymptotic Stability?

The lack of zero-state detectability of the system in (8.2) is an obstacle to the
asymptotic stability of the closed-loop scheme. The problem is therefore to
keep the negative feedback interconnection structure of the two highlighted
blocks, while introducing some detectability property in the loop. However
the whole state is now (q, q, g ) and it is known in identication and adaptive
control theory that the estimated parameters converge to the real ones (i.e.
g (t) 0) only if some persistent excitation conditions are fulllled. Those
conditions are related to the spectrum of the signals entering the regressor ma-
trix Yg (q). Such a result is hopeless here since we are dealing with regulation.
Hence the best one may expect to obtain is convergence of (q, q) towards zero.
We may however hope that there exists a feedback adaptive controller that
can be analyzed through the passivity theorem and such that the underlying
storage function can be used as a Lyapunov function with Krasovskii-La Salle
Theorem to prove asymptotic convergence. Let us consider the estimation
algorithm proposed in [488]:

2q(t)
g (t) = 3 YgT (t) 4 q(t) + (8.3)
1 + 2q T (t)q(t)
Note that this is still a gradient update law. It denes a passive operator
2q
4 q + 1+2q T q
Yg (q)g , not q
Yg (q)g . We therefore have to look at
the dissipativity properties of the subsystem with dynamics  M (q(t))q(t)  +
2q(t)
C(q(t), q(t))q(t) + 1 q(t) + 2 q(t) = u1 (t), y1 (t) = 4 q(t) + 1+2qT (t)q(t) :
this is new compared to what we have seen until now in this book. Let us
analyze it in detail:
u1 , y1 t =

t 2q
T
= 0 4 q + 1+2qT q [M (q)q + C(q, q)q + 1 q + 2 q]ds

t2 T d

4 T 2qT M(q)q
3
=q 0 4 q (2 q + 1 q) + ds 2 q M (q)q + 1+2qT q ds+

 t 2 2qT M(q)q+2qT C(q,q)q 8qT M(q)q qT q

3
0
q
+ 1+2qT q + 1+2qT q 2 1+2q T q (2 q + 1 q) ds
(8.4)
 t  t t
2qT M(q)q
4 2
2
T
q q 0
+ 4 T
2 q M (q)q + 1+2qT q 0 + 4 1 0
T
q qds+

t2 qT q
 
||q||.||q||
3
+ 0
22 1+2q q
T 4M + kc

2
q T q 2 11+2q T q ds

2q(0)T M(q(0))q(0)
422 q(0)T q(0) 4 T
2 q(0) M (q(0))q(0) + 1+2q(0)T q(0) +
t
+4 1 0
q T (s)q(s)ds
438 8 Adaptive Control

where we have used the fact that due to the skew-symmetry of M (q)2C(q, q)
we have M (q) = C(q, q) + C T (q, q), and where

2  8
1 1 kc 2M
4 > max + 4M + ,
1 22 2 m 2
with m In M (q) M In , ||C(q, q)|| kc ||q|| for any compatible matrix
and vector norms. Under these gain conditions, one sees from (8.4) that the
rst subsystem is passive with respect to the supply rate uT1 y1 , and a storage
function is given by

4 2 T 4 2q T M (q)q
V1 (q, q) = q q + q T M (q)q + (8.5)
2 2 1 + 2q T q
The rst subsystem even possesses some strict passivity property; see (8.4).
Finally a complete storage function is provided by the sum V (q, q, g ) =
V1 (q, q) + V2 (g ), and it can be shown that its derivative is semi-negative
denite and that the largest invariant set contained in the set V 0 is con-
tained in the set (q, q) = (0, 0) which ends the proof.

Remark 8.1. The storage function associated to the rst subsystem is quite
original. It looks like the available storage of the closed-loop system when
a PD controller is applied, but the added term comes from nowhere! Our
analysis has been done easily because we knew beforehand that such a storage
function was a good one. The intuition behind it is not evident. It was rst
discovered in [261] and then used in [488].

The Adaptive Slotine and Li Controller

Let us now pass to the controller presented in Subsection 7.3.4 in (7.68). It

turns out that this scheme yields a much more simple stability analysis than
the PD with adaptive gravity compensation: this is due to the fact that as
pointed out earlier, it uses the inertia matrix explicitly even for regulation.

Gradient Estimation Law

Consider the following controller:

(t) = M (q(t))qr (t) + C(q(t), q(t))qr (t) + g(q(t)) 1 s(t)

= Y (q(t), q(t), t)(t)

t)

= M (q(t))qr (t) + C(q(t), q(t))qr (t) + g(q(t)) 1 s(t) Y (q(t), q(t),

(t) = 2 Y T (t)(q(t), q(t), t)s(t), 2 > 0

(8.6)
 8.1 Lagrangian Systems 439

where we used the fact that the xed parameter controller can be rewritten
under the required linear form Y (q, q, t), where is a vector of unknown
inertia parameters. Actually one has M (q(t))q(t)+C(q(t), q(t))q(t)+g(q(t)) =
Y (q(t), q(t), q(t)). The closed-loop system is therefore given by

M (q(t))s(t) + C(q(t), q(t))s(t) + 1 s(t) = Y (q(t), q(t), t)(t)



= q(t) + s(t)
q(t)





(t) = Y T (t)(q(t), q(t), t)s(t)
2
(8.7)

The interpretation through the passivity theorem is obvious: the update

law in (8.6) is a gradient that denes a passive operator s
Y (q, q, t) and
the rst subsystem has state (q, s). From the developments in Subsection 7.3.4
one therefore sees that the adaptive version of the Slotine and Li controller
just yields a closed-loop system that is identical to the one in (7.69) (7.70) with
an additional passive block interconnected to the two previous ones in (7.69)
and (7.71); see Figure 8.1 and compare with Figure 7.6. The storage function
follows immediately. Similarly to the PD with adaptive compensation scheme,
one cannot expect to get asymptotic stability of the whole state because of
the parameter estimates that generally do not converge towards the real ones.
Let us consider the quadratic function
1 T 1
V (s, t) = s M (q)s + T (8.8)
2 2
Computing its derivative along the closed-loop trajectories and u sing the
same arguments as for the rst stability proof of the xed parameters Slotine
and Li controller in Section 7.3.4, one easily concludes on the global conver-
gence of all signals but (t) to zero as t +, and on the boundedness of
all signals on [0, +).

Remark 8.2. Historically the passivity interpretation of the Slotine and Li

scheme has been deduced from Lemma 7.23, see [71, 74], where most of the
adaptive schemes (including e.g. [426]) designed for rigid manipulators have
been analyzed through the passivity theorem. Indeed this is based on a CTCE
as dened in Lemma 7.23. Actually the rst subsystem in (8.7) with state s has
relative degree one between its input u1 = Y (q, q, t) and its output y1 = s.
As we shall remark when we have presented the adaptive control of linear
invariant systems with relative degree one, the CTCE is ubiquitous in direct
adaptive control. The extension of the Slotine and Li scheme to the case of
force-position control when the system is in permanent contact with a exible
440 8 Adaptive Control

Fig. 8.1. Closed-loop equivalent representation (adaptive case)

sti environment, has been done in [320] (see also [368] for an extension of
the scheme in [320]).

Least-squares Estimation Law

Until now we have presented only gradient-type update laws. It is clear that
the estimation block can be replaced by any system that has inside the state
and is passive with respect to the same supply rate. The classical recursive
least-squares estimation algorithm does not satisfy such requirements. How-
ever it can be passied as explained now. First of all let us recall the form
of the classical least-squares algorithm:

ls (t) = P (t)Y T (q(t), q(t), t)s(t)
(8.9)

P (t) = P Y (q(t), q(t), t)Y T (q(t), q(t), t)P (t), P (0) > 0

The required passivity property is between s and Y (q, q, t) (recall we

have dened = ). Let us compute the available storage of this system:
 8.1 Lagrangian Systems 441

t
Va (, P ) = sup sT Y ds
s:(0,(0),P (0)) 0

1  T 1 t 1 t
= sup P + T P 1 ds (8.10)
s:(0,(0),P (0)) 2 0 2 0

1  T 1 t 1 t
= sup P + T Y Y T ds
s:(0,(0),P (0)) 2 0 2 0

where we used the fact that P 1 = Y Y T . One remarks that the available
storage in (8.10) is not far from being bounded: it would suce that Y T
be L2 -bounded. However it seems dicult to prove this. Consequently let us
propose the following modied least-squares estimation algorithm 2 :

(t) = ls (t) + S(t)



   



P Y T (t)Y (t)


P (t) = (t) (t) 1+tr(Y T (t)Y (t))
+ R P (t) + P (t)







sT (t)Y (t)Y T (t)s(t)
(t) = (1+sT (t)s(t))(1+

tr(Y T (t)Y (t)))






A = Y T (t)Y (t)
+ R
1+tr(Y T (t) Y (t))

 



S(t) = Y T (t) s(t)
T
(t)A (t) + M (1 + (R))

1+tr(Y T (t)Y (t)) 1+sT (t)s(t) ls ls max







0, R > 0



 



min (R)In P 1 (0) max (R) + 1 In





M T
(8.11)
Then the following is true [73, 319]:
Lemma 8.3. (a) min (R) i (P 1 ) max (R) + 1 , where i (P 1 ) denotes
t  t  t T 1
the eigenvalues of P 1 . (b) 0 sT Y ds = 12 ls P ls 21 0 ls
T 1
P ls d
t T 0

0
s Y Sd , where ls = ls , where ls is the classical least-squares estimate
 t T 1 t
ls = P Y T s. (c) 12 0 ls P ls d 0 sT Y Sd 0.

2
Let us note that the denomination least-squares somewhat loses its original
meaning here, since it is not clear that the proposed scheme minimizes any
quadratic criterion. However the name least-squares is kept for obvious reasons.
442 8 Adaptive Control

It follows that the mapping s

Y is passive with storage function
1 T 1
2 ls P ls .
The proof of Lemma 8.3 is not given here for the sake of brief-
ness and also because despite its originality, it has not been proved that such
passied least-square yields better closed-loop performance than the simple
gradient update law (for instance in terms of parameter convergence speed
and of robustness). It is therefore to be seen more like a theoretical exercise
(nd out how to passify the classical least-squares) rather than something mo-
tivated by applications. The interest for us here is to illustrate the modularity
provided by passivity-based controllers. As we shall see further, it applies
equally well to adaptive control of relative degree one and two linear invariant
systems.

8.1.2 Flexible JointRigid Link Manipulators: The Adaptive

Lozano and Brogliato Algorithm

In this section we provide the complete exposition of the adaptive version

of the scheme of Section 7.5.1, which is the only adaptive scheme proposed
in the literature solving both the linearity-in-the-paramaters and the a priori
knowledge of the stiness matrix K issues, and at the same time guaranteeing
the global convergence of the tracking errors, the boundedness of all the closed-
loop signals, with only positions and velocity measurements (no acceleration
feedback). It has been published in [72, 318]. This scheme uses ingredients
from [461] and from [426] in the stability proof.
The starting point for the stability analysis of the adaptive version is the
quadratic function

V (s1 , s2 , q1 , , q2 ) = 12 sT1 M (q1 )s1 + 12 det(M (q1 ))sT2 Js2 +

+ 21 (q1 q2 )T K(q1 q2 ) + + 21 p q1T q1 + 12 T

(8.12)

where p > 0 is a feedback gain and (t) = (t) is the parameter error
vector. We do not dene what is at this stage, because this vector of unknown
parameters will be constructed in proportion as the stability proof progresses.
Actually it will be proved in Lemma 8.6 below that the nonadaptive control
law may be written as

5T h(q1 )u + Y6 (q1 , q1 , q2 , q2 )6 = 0 (8.13)

where 5T h(q1 )
=det(M (q1 ). Thus a nice property that will be used is that
M (q1 ) > 0 so that det(M (q1 )) > 0: the controller hence dened is not singular.
This is used when the parameter vector 5 is replaced by its estimate 5 (t), by
 8.1 Lagrangian Systems 443

dening a suitable projection algorithm. Another issue is that of the a priori

knowledge of the diagonal matrix K =diag(kii ) which has to be replaced
by an estimate K in the controller. Since the ctitious input q2d is dened
with K 1 , its adaptive counterpart will be dened with K 1 (t), so that K(t)
has to be nonsingular. Moreover the signal q2d has to be twice dierentiable.
This implies that K(t) will have in addition to be twice dierentiable as
well. The two parameter projection algorithms are given as follows. We dene
K = (k11 , k22 , ...., knn )T , and we assume that a lower bound In on M (q1 )
is known.

The Parameter-adaptation Algorithms

It is possible to dene a subspace spanned by h(q1 ) as S = {v | v =

h(q1 ) for some q1 }, and a set = {v | v T h n for all h S}. The set is
convex, and 5 . The rst parameter adaptation law is as follows:

h(q1 (t))uT (t)s2 (t) if 5 (t) Int()




5 (t) = Pr [h(q1 (t))uT (t)s2 (t)] if 5 (t) () (8.14)





and [h(q1 (t))uT (t)s2 (t)]T 5 > 0

where Pr [] denotes the orthogonal projection onto , () is the boundary

of , and 5 is the vector normal to () at 5 (t), and

6 (t) = Y6T (q1 (t), q1 (t), q2 (t), q2 (t))s2 (t) (8.15)
The gradient update laws in (8.14) and (8.15) will then be used to dene
the adaptive controller as

5T (t)h(q1 (t))u(t) + Y6 (q1 (t), q1 (t), q2 (t), q2 (t))6 (t) = 0 (8.16)

The second projection algorithm is as follows, and concerns the estimate
of the stiness matrix K:

xi (t) if ki (t) k




i
i
k (t) = x (t) if ki (t) 2k and xi (t) 0 (8.17)



 xi (t)


f (i (t)) xi (t) if k ki (t) 2k and xi (t) 0
k

where xi (t) = Y2d i

(q1 (t), q1 (t), q1d (t), q2d (t)) and 0 < k min ki . The row
vector Y2d () is dened as
444 8 Adaptive Control
T
s1 K[q2d q1d ] = kT diag(si1 )[q2d q1d ] = Y2d (q1 , q1 , q1d , q2d )k
(8.18)

Y2d (q1 , q1 , q1d , q2d ) = [q2d q1d ]T diag(si1 )

The function f () has to be chosen as a smooth function 0 f (ki ) 1

with f ( 2k ) = 0 and f (k ) = 1. This implies that the parameter projection in
(8.17) is twice dierentiable and that ki (t) 2k for all t 0 and all 1 i n
3
.

The rational behind the choice for the various functions appearing in these
update laws, will be claried. We now introduce a Lemma that will be useful
in constructing a function q2d () whose second derivative q2d () depends only
on position and velocity.
Lemma 8.4. [72] One has M (q1 (t))s1 (t) = Y4f (t)4 , where Y4f (t)+Y4f (t) =
Y4 (q1 (t), q1 (t), q2 (t)) for some Y4 (q1 (t), q1 (t), q2 (t)).
Proof: Let us lter the rst dynamical equation in (6.105) as

1
[M (q1 (t))q1 (t) + C(q1 (t), q1 (t))q1 (t) + g(q1 (t)) K(q2 (t) q1 (t))] = 0
1+s
(8.19)
1
where we implicitly mean that 1+s [f (t)] is the Laplace transform of f (t). Now
we have (we drop the time argument for simplicity)
1
1+s [M (q1 )q1 ] = M (q1 )q1 M (q1 (0))q1 (0)
(8.20)
1 1
1+s [M (q1 )q1 M (q1 (0)q1 (0)] 1+s [M (q1 , q1 )q1 ]

which follows from M (q1 )q1 = d

dt (M (q1 )q1 ) M (q1 , q1 )q1 . Now

t
1
[M (q1 )q1 ] = exp(t + )M (q1 ( ))d (8.21)
1+s 0
Then using integration by parts one gets

1
1+s [M (q1 )q1 ] =
  t

= exp(t) exp( ) M (q1 ) M (q1 (0)q1 (0) 0 M (q1 (y), q1 (y))dy
0

t    
0
exp( ) M (q1 ( ))q1 ( ) M (q1 (0)q1 (0) 0 M (q1 (y), q1 (y))dy d
(8.22)
3
Another type of C n projections is presented in [92], whose motivation is quite in
the spirit of this one, see e.g. [92, III].
 8.1 Lagrangian Systems 445

which nally yields

1
1+s [M (q1 )q1 ] =
t
= M (q1 )q1 M (q1 (0))q1 (0) 0
M (q1 (y), q1 (y))dy
t   
0
exp(t + ) M (q1 )q1 M (q1 (0))q1 (0) 0 M (q1 (y), q1 (y))dy d
(8.23)
Still integrating by parts we get

t
0
exp(t + )M (q1 ( ), q1 ( ))q1 ( )d =
t t  

= 0
M (q1 ( ), q1 ( ))q1 ( )d 0
exp(t + ) 0
M (q1 (y), q1 (y))q1 (y)dy d
(8.24)
from which we can deduce (8.20) combining (8.23) and (8.24). Now using
(8.19) and (8.20) we obtain

1
M (q1 )q1 = M (q1 (0))q1 (0) + s+1 [M (q1 )q1 M (q1 (0))q1 (0)] +

1 1
+ s+1 [M (q1 , q1 )q1 ] s+1 [C(q1 q1 )q1 + g(q1 ) + Kq1 ]+ (8.25)

1
+ s+1 [Kq2 ]

The terms between brackets can be written as Yi (q1 , q1 )i for some con-
1 1
stant vector i . Therefore s+1 [Yi (q1 , q1 )i ] = s+1 [Yi (q1 , q1 )]i = Yif (t)i with
Yif (t) + Yif (t) = [Yi (q1 (t), q1 (t)). It follows that (8.20) can be written as
M (q1 )s1 = Y4f (t)4 with Y4f (t) + Y4f (t) = [Y4 (q1 (t), q1 (t), q2 (t)).

Let us now proceed with the stability proof, which we start by dierentiat-
ing the function (8.12) along the systems trajectories. The controller u() will
then be constructed step by step within the proof. Afterwards we shall reca-
pitulate and present compact forms of the input and the closed-loop system.
We obtain

V (s1 , s2 , q1 , , q2 ) = sT1 [M (q1 )s1 + C(q1 , q1 )]s1 + det(M (q1 ))sT2 J s2 +

+(q1 q2 )T K(q1 q2 ) + p q1T q1 + T + (8.26)

+ 21 dt
d
[det(M (q1 ))]sT2 Js2

Notice that
446 8 Adaptive Control

(q1 q2 )T K(q1 q2 ) = (q1 q2 )T K(q1 + s1 + q2 s2 )

= (q1 q2 )T K(q1 q2 ) + (s1 s2 )T K(q1 q2 )

(8.27)
Introducing this in (8.26) we obtain

V (s1 , s2 , q1 , , q2 ) sT1 [M (q1 )s1 + C(q1 , q1 )s1 + K(q1 q2 )] +

 1 d

+ sT2 det(M (q1 ))J s2 + K(q1 q2 )] + p q1T q1 + T
2 dt [det(M (q1 ))Js2
(8.28)
where the skew-symmetry property of Lemma 6.16 has been used to introduce
the term C(q1 , q1 )s1 . Let us manipulate the rst term between brackets in the
right-hand-side of (8.28):

T1 = sT1 [M (q1 )s1 + C(q1 , q1 )s1 + K(q1 q2 )]

) *
= sT1 M (q1 )(q1 q1d + q1 ) + C(q1 , q1 )(q1 + q1 ) + K(q2d q1d )

= sT1 [1 + 2 + K(q2d q1d )]

(8.29)
where we dene

1 = (M (q1d ) M (q1 )q1d + (C(q1d , q1d ) C(q1 , q1 ))q1d + g(q1d )

(8.30)
g(q1 ) + (M (q1 )q1 + C(q1 , q1 )q1

2 = M (q1d )q1d C(q1d , q1d )q1d g(q1d ) (8.31)

We now need a technical result from [426].
Lemma 8.5. [426, Lemma 1] The following inequality holds:

sT1 [M (q1d )q1d + C(q1d , q1d )q1d + g(q1d ) M (q1 )(q1d q1 )

C(q1 , q1 )(q1d q1 ) g(q1 )]s1

(8.32)
sT (M (q1 ) + b1 In )s + sT (2 M (q1 ) + b2 In )q1 + b3 (sT s ||q1 ||+

+ ||s|| q1T q1 )

for some positive bounded functions b1 (), b2 (), b3 () of q1d (), q1d (), and
q1d ().
 8.1 Lagrangian Systems 447

This allows us to upperbound the term sT1 1 as follows:

sT1 1 sT1 (M (q1 ) + b1 In ) + sT1 (2 M (q1 ) + b2 In )q1 +

(8.33)
+b3 (sT1 s1 ||q1 || + ||s1 || q1T q1 )

Now notice that

sT
1 s1 q1T q1 1 2
sT1 s1 ||q1 || + ||s1 || q1T q1 = 4 + 4 sT1 s1 2 q1T q1
1 2
q1T q1 2 sT1 s1 + (1 + )sT1 s1 q1T q1 (8.34)

sT
1 s1 q1T q1
4 + 4 + (1 + )sT1 s1 q1T q1

Introducing (8.34) into (8.33) we get

sT1 1 a1 sT1 s1 + a2 q1T q1 + a3 sT1 s1 q1T q1 (8.35)

where a1 (), a2 () and a3 () are positive bounded functions of q1d , q1d , q1d ,
and of the dynamic model parameters. Now from (8.31) and the fact that the
various terms of the dynamical model are linear in the parameters, we can
write

2 = Yd (q1d , q1d , q1d )1 (8.36)

where the matrix Yd (q1d , q1d , q1d ) is of appropriate dimensions and 1 is a
vector of constant parameters. Now since K is a diagonal matrix we can write

sT1 K(q2d q1d ) = kT diag(si1 )(q2d q1d ) (8.37)

with k = (k11 , k22 , ...., knn )T . From (8.37) we have

sT1 K(q2d q1d ) = Y2d (q1 , q1 , q1d , q2d )k (8.38)

where

Y2d (q1 , q1 , q1d , q2d ) = (q2d q1d )T diag(si1 ) (8.39)

(we recall that si1 denotes the ith element of the vector s1 IR ). Now n

injecting (8.38) into (8.29) we obtain

T1 = sT1 (1 + 2 ) Y2d (q1 , q1 , q1d , q2d )k + Y2d (q1 , q1 , q1d , q2d )k

(8.40)
(v + n q1T q1 )sT1 M (q1 )s1
448 8 Adaptive Control

where k (t) = k (t) k , v > 0, n > 0. The last term in (8.40) will be used
to compensate the term sT1 1 . Now from Lemma 8.4 we have M (q1 (t))s1 (t) =
Y4f (t)4 . Introducing this into (8.40) we obtain

T1 = sT1 (1 + 2 ) Y2d (q1 , q1 , q1d , q2d )k + Y2d (q1 , q1 , q1d , q2d )k +

(8.41)
+(v + n q1T q1 )sT1 Y4f (t)4 (v + n q1T q1 )sT1 M (q1 )s1

Provided kii > 0 for all 1 i n, we can safely dene the function q2d ()
as follows:

K(q2d q1d ) = (v + n q1T q1 )Y4f (t)4 Yd (q1d , q1d , q1d )1 p q1 (8.42)

where K =diag(kii ) and k = (k11 , k22 , ..., knn )T . Introducing (8.42) into
(8.39) we obtain

Y2d (q1 , q1 , q1d , q2d )k = k diag(si1 )(q2d q1d ) = sT1 K(q2d q1d )

= (v + n q1T q1 )sT1 Y4f (t)4 sT1 Yd (q1d , q1d , q1d )1

p sT1 q1
(8.43)
where p > 0. Introducing (8.43) and (8.36) into (8.41) we obtain

T1 = sT1 1 sT1 Yd 1 Y2d k (v + n q1T q1 )(sT1 Y4f 4 + sT1 M (q1 )s1 )

p sT1 q1
(8.44)
Furthermore from (8.35) we have that

sT1 (v + n q1T q1 )sT1 M (q1 )s1 p q1T q1

sT1 s1 (min (M (q1 )) v a1 ) q1T q1 (p a2 ) (8.45)

sT1 s1 q1T q1 (min (M (q1 )) n a3 )

If v , p , n are chosen large enough so that


min (M (q1 )) v a1 0 > 0



p a2 1 > 0 (8.46)





(min (M (q1 )) n a3 0
 8.1 Lagrangian Systems 449

we obtain

T1 0 sT1 s1 1 q1T q1 p q1T q1 sT1 Yd (q1d , q1d , q1d )1

(8.47)
Y2d (q1 , q1 , q1d , q2d )k (v + n q1T q1 )sT1 Y4f (t)4

Combining (8.28), (8.29) and (8.47) we obtain

V (s1 , s2 , q1 , , q2 ) 0 sT1 s1 1 q1T q1 sT1 Yd (q1d , q1d , q1d )1

(8.48)

Y2d (q1 , q1 , q1d , q2d )k (v + n q1T q1 )sT1 Y4f (t)4 + T + sT2 T2

with
J d
T2 = det(M (q1 ))s2 + [det(M (q1 ))] K(q1 q2 ) (8.49)
2 dt
Let us dene

= [kT 1T 4T 5T 6T ]T (8.50)
where the precise denition of 5 and 6 will be given later. Let us introduce
the parameter update laws:


1 (t) = YdT (q1d (t), q1d (t), q1d (t))s1 (t)
(8.51)

T
4 (t) = (v + n q1 q1 )Y4f (t)s1 (t)
where we recall that M (q1 )s1 = Y4f (t)4 with

Y4f (t) + Y4f (t) = Y4 (q1 (t), q1 (t), q2 (t))

from Lemma 8.4. Now let us introduce (8.50), (8.51) and (8.17) into (8.48),
in order to obtain

V (s1 , s2 , q1 , , q2 ) 0 sT1 s1 1 q1T q1 + 5T 5 + 6T 6 + sT2 T2 (8.52)

where the equality


(k Y2d )T k = (ki Y2d
i
)(ki ki ) (8.53)
i

has been used. The expression for the controller is obtained from the fol-
lowing lemma:
450 8 Adaptive Control

Lemma 8.6. [318] The term T2 in (8.49) can be expressed as

T2 = 5T h(q1 )u + Y6 (q1 , q1 , q2 , q2 )6 (8.54)

with det(M (q1 )) = > for some > 0 and all q1 IR . The vectors
5T h(q1 ) n n

5 and 6 are unknown parameters and h(q1 ) and Y6 (q1 , q1 , q2 , q2 ) are known
functions.

Proof: From (6.105) and (8.49) we can deduce that

T2 = det(M (q1 ))[u + K(q1 q2 )] + Jdet(M (q1 ))(q2d + q2 )

(8.55)
+ J2 d
dt det(M (q1 ))s2 K(q2 q1 )

Since M (q1 ) > 0 then det(M (q1 )) > 0 and the linearity-in-the-parameters
property of the dynamical equations allows one to write det(M (q1 )) =
5T h(q1 ). Considering the second order time-derivative of (8.42), it can be
proved that det(M (q1 ))q2d is a linear-in-the-parameters function of positions
and velocities (notice that the way Y4f is dened plays a crucial role here)
and of the acceleration q1 . Similarly q2d is a measurable signal (i.e. a function
of positions and velocities); see (8.39), Lemma 8.4, (8.42) and (8.17). However
notice that det(M (q1 ))q1 is a function of q1 , q1 , and q2 . Thus q2d is a function
of positions and velocities only. We conclude that T2 can indeed be written in
a compact form as in (8.54).
In view of Lemma 8.6 we obtain

V (s1 , s2 , q1 , , q2 ) 0 sT1 s1 1 q1T q1 + 5T 5 + 6T 6 + sT2 5T h(q1 )u+

+sT2 Y6 (q1 , q1 , q2 , q2 )6

= 0 sT1 s1 1 q1T q1 + 5T 5 sT2 hT (q1 )5T u + sT2 h(q1 )5T u

+sT2 Y6 (q1 , q1 , q2 , q2 )6
(8.56)
Introducing the parameters adaptation laws in (8.14) and (8.15) and the
adaptive control law in (8.16), into (8.56), we get

V (s1 , s2 , q1 , , q2 ) 0 sT1 s1 1 q1T q1 + 5T [5 h(q1 )uT s2 ]. (8.57)

The term h(q1 )uT s2 can be broken down as

h(q1 )uT s2 = Pr (h(q1 )uT s2 ) + Pr (h(q1 )uT s2 ), (8.58)

 8.1 Lagrangian Systems 451

where we recall that Pr (z) denotes the orthogonal projection on the hyper-
plane tangent to () at z and Pr (z) is the component of z that is perpen-
dicular to this hyperplane at z. Then using (8.15) we obtain

0 if 5 Int()




5T [5 h(q1 )uT s2 ] = 5T (h(q1 )uT s2 ) 0 if 5 ()





and (h(q1 )uT s2 )T 5 > 0
(8.59)
Consequently we nally obtain

V (s1 , s2 , q1 , , q2 ) 0 sT1 s1 1 q1T q1 (8.60)

It immediately follows from (8.12), (8.61), Lemma 4.8 and Theorem 4.10
that (), s1 (), s2 (), q1 (), q1 (), q2 () and q2 () are bounded functions of time
on [0, +) 4 . Moreover s1 L2 . Using the same reasonning as in the proof
of the xed parameters Slotine and Li or Lozano and Brogliato schemes, we
deduce that q1 (t) 0 as t +. It is deduced from (8.25) that the term
1
s+1 [q2 ] is bounded, so that q2d is bounded also, and consequently both q2 ()
and q1 () are bounded. The boundedness of q2 () follows from dierentiat-
ing (8.42), which proves that q2d () is bounded. Thus q2 () is bounded. The
boundedness of the controller u can be inferred from (8.16). One deduces that
q2 () is bounded on [0, +).

Recapitulation

The closed-loop system that results from the controller dened in (8.16),
(8.14), (8.15), (8.51) and (8.17) does not have a form as simple and intu-
itive as the closed-loop system of the Slotine and Li adaptive controller, or
of the closed-loop system of the Lozano and Brogliato xed parameters con-
troller. This seems however to be an intrinsic property of the adaptive scheme
for (6.105), because one needs to invert the rst dynamical equation to avoid
the acceleration q1 (t) measurement. Consequently the matrix M 1 (q1 ) nec-
essarily appears in the xed parameters scheme, and it is a nonlinear-in-the-
parameters function. The adaptation for the matrix K may be avoided in
practice if one is able to estimate it accurately enough. But the linearity-in-
the-parameters issue is unavoidable and intrinsic to such controlled dynamics.
After a certain number of manipulations based on the above developments
we may write the closed-loop dynamics as follows:

4
It is clear that the desired trajectory q1d (t) and its rst and second derivatives,
are chosen as bounded functions of time. Any other choice would be silly.
452 8 Adaptive Control

M (q1 (t))s1 (t) + C(q1 (t), q1 (t))s1 (t) = K(q2 q1 ) + K(q1d (t) q2d (t))

(v + n q1T q1 )Y4f (t)4 (t) Yd (q1d (t), q1d (t), q1d (t))1 (t) p q1

(v + n q1T q1 )M (q1 (t))s1 (t) + W (q1 (t), q1 (t), q1d (t), q1d (t), q1d (t))

with W (q1 (t), q1 (t), q1d (t), q1d (t), q1d (t)) = M (q1 (t))[q1d (t) q1 (t)]+

+C(q1 (t), q1 (t))[q1d (t) q1 (t)] g(q1 (t)) + Yd (q1d (t), q1d (t), q1d (t))1

5T (t)h(q1 (t))u(t) + Y6 (q1 (t), q1 (t), q2 (t), q2 (t))6 (t) = 0

Updates laws in (8.14), (8.15), (8.17) and (8.51)

qi (t) = qi (t) + si (t), i = 1, 2

(8.61)

where we recall that Yd (q1d (t), q1d (t), q1d (t))1 = M (q1d )q1d C(q1d , q1d )q1d
g(q1d ); see (8.31). It is worth comparing (8.61) with (7.125) to measure the
gap between adaptive control and xed-parameter control, and comparing
(8.61) with (8.7) to measure the gap between the exible-joint case and the
rigid-joint case.

Remark 8.7. As we saw in Section 7.5.1, the xed parameters Lozano and
Brogliato scheme is a passivity-based controller using a backstepping design
method. The adaptive scheme is a highly non-trivial extension, where the
linearity-in-the-parameters and the unknown stiness matrix issues imply the
use of very specic update laws, and hampers the direct application of back-
stepping methods designed elsewhere for some classes of nonlinear systems.

8.1.3 Flexible JointRigid Link Manipulators: The Backstepping

Algorithm

Let us now investigate how the backstepping approach may be used to solve
the adaptive control problem for exible joint manipulators. We will assume
that K is a known matrix. We have to solve two main problems in order to
extend the xed parameter scheme presented in Subsection 7.5.2 towards an
adaptive version:
1) The input u in must be LP (Linear in some set of Parameters).
 8.1 Lagrangian Systems 453

2) The signals q2 and e2 have to be available on line because they will be

used in the update laws.
To solve 1), we can use the idea introduced in [318] which consists of adding
the determinant of the inertia matrix det(M (q1 )) in the Lyapunov function
V1 () (see the foregoing subsection on the adaptive passivity-based scheme).
As we explained the nonlinearity in the unknown parameters comes from the
terms containing the inverse of the inertia matrix M 1 (q1 ). Premultiplying by
det(M (q1 )) allows us to retrieve LP terms, as det(M (q1 ))M 1 (q1 ) is indeed
LP (the price to pay is an overparametrization of the controller). Moreover 2)
implies that q2d (see (7.130)) and e2d (see after (7.133)) are available on line,
and thus do not depend on unknown parameters. We can proceed as follows:
Step 1: The right-hand-side of (6.105) can be written as Y1 (q1 , q1 , t)1 .
Thus we choose q2d in (7.130) as

Kq2d = Y1 (q1 , q1 , t)1 (8.62)

where 1 stands for an estimate of 1 . Thus

q2 = q2 K 1 Y1 (q1 , q1 , t)1 (8.63)

Adding Y1 ()1
to the right-hand-side of the rst equation in (6.105) and
dierentiating (8.63), one obtains:

M (q1 (t))s1 (t) + C(q1 (t), q1 (t))s1 (t) + 1 s1 (t) = K q2 (t) Y1 (t)1

q2 (t) = q2 (t) K 1 dt
d
(Y1 (t)1 )
(8.64)
Step 2: Now consider e2d dened after (7.133). The rst two terms are
available but the third term is a function of unknown parameters and it
is not LP (it contains M 1 ). Assume now that V2 is replaced by
1 1
V2a = Vr (q1 , s1 , t) + 1T 1 + det(M (q1 ))q2T q2 (8.65)
2 2
Setting q2 = e2d + e2 , i.e. q2 = e2d + e2 K 1 dt d
(Y1 1 ), we get along
trajectories of (8.64):

V2a 1 q1T q1 2 1 q1T q1 sT1 Y1 1 + 1T 1 + q2T Ks1 +

+q2T det(M (q1 ))e2 + q2T det(M (q1 ))[e2d q2d ]+ (8.66)

d (det(M(q1 ))
+q2T dt { 2 }q2
Let us denote det(M ) = Y2 (q1 )2 , and choose
454 8 Adaptive Control

Y2 2 e2d = Y3 (q1 , q1 , q2 , t)3 q2 (8.67)

where

8
d det(M (q1 ))
Y3 (q1 , q1 , q2 , t)3 = q2 det(M (q1 ))q2d + Ks1 (8.68)
dt 2
Choose also

1 (t) = Y1T (q1 (t), q1 (t), t)s1 (t) (8.69)
Thus we obtain

V2a 1 q1T q1 2 1 q1T q1 + q2T det(M (q1 ))e2 + q2T [Y2 2
T e2d + Y3 3
T ]
(8.70)
(we drop the arguments for convenience). Introducing q2 Y2 2 e2d we ob- T

tain

V2a 1 q1T q1 2 1 q1T q1 + q2T det(M (q1 ))e2

(8.71)
q2T e2d Y2 2 + q2T Y3 3 q2T q2
Dene V3a = V2a + 12 2T 2 + 12 3T 3 and set

3 (t) = Y3T (q1 (t), q1 (t), q2 (t), t)q2 (t) (8.72)

2 (t) = Y2T (q1 (t))eT2d (t)q2 (t) (8.73)
We therefore obtain

V3a 1 q1T q1 2 1 q1T q1 + q2T det(M (q1 ))e2 q2T q2 (8.74)

Remark 8.8. In order to avoid any singularity in the control input, the
update law in (8.73) has to be modied using a projection algorithm,
assuming that 2 belongs to a known convex domain. We refer the reader
to the foregoing subsection for details about how this domain may be
calculated, and the stability analysis related to the projection. For the sake
of clarity of this presentation, we do not introduce this modication here,
although we know it is necessary for the implementaton of the algorithm.
Step 3: At this stage our goal is partially reached, as we have dened
signals q2 and e2 available on line. Now consider the function
1
V4a = V3a + det(M (q1 ))eT2 e2 (8.75)
2
We obtain
V4a 1 q1T q1 2 1 q1T q1 + q2T det(M (q1 ))e2
2 3 (8.76)
q2T q2 + eT2 [v e2d ] + eT2 dt
d det(M(q)) e
2 2
 8.1 Lagrangian Systems 455

Notice that

8
d det(M (q1 ))
det(M (q1 ))e2d + e2 = Y4 (q1 , q1 , q2 , q2 )4 (8.77)
dt 2

for some Y4 and 4 matrices of suitable dimensions. Let us denote this

time det(M ) = Y2 (q1 )5 (this is strictly equal to Y2 (q1 )2 dened above,
but we choose a dierent notation because the estimate of 5 is going to
be chosen dierently). Let us choose v = q2 + w and

Y2 5 w = Y4 4 e2 (8.78)

We obtain

V4a 1 q1T q1 2 1 q1T q1 q2T q2 eT2 wY2 5 + eT2 Y4 4 eT2 e2 (8.79)

Finally we choose as a Lyapunov function for the whole closed-loop system

1 1
V = V4a + 4T 4 + 5T 5 (8.80)
2 2
and the following update laws:

4 (t) = Y4T (q1 (t), q1 (t), q2 (t), q2 (t))e2 (t) (8.81)

5 (t) = Y2T (q1 (t))wT (t)e2 (t) (8.82)
(a projection algorithm has to be applied to 5 ; see Remark 8.8 above).
We obtain

V 1 q1T q1 2 1 q1T q1 q2T q2 eT2 e2 . (8.83)

We therefore conclude that L , q2 , e2 , q1 , s1 L2 L , q2 L ,
(see (8.63)) , q2 L . Finally from the denition of s1 and Theorem 4.10
we conclude that q1 L2 L , q1 L2 and q1 0 as t +.
To conclude this subsection, one may say that the backstepping procedure
does not bring much more than the passivity-based one to the adaptive control
problem for exible joint Lagrangian systems. The fact that the ctitious
input q2d is premultiplied by an unknown term K, creates a diculty that
has been solved in [318] but has never been tackled in the backstepping
literature. The linearity-in-the-parameters problem solution also is an original
one, motivated by the physics of the process, and whose solution also was
proposed in [318] and nowehere else, to the best of the authors knowledge.
456 8 Adaptive Control

8.2 Linear Invariant Systems

The problem of adaptive control of linear invariant systems has been a very
active eld of research since the beginning of the 1960s. Two paths have
been followed: the indirect approach which consists of estimating the process
parameters, and using those estimated values into the control input, and the
direct approach that we described in the introduction of this chapter. The
direct approach has many attractive features, among them the nice passivity
properties of the closed-loop system, which actually is a direct consequence
of Lemma 7.23. This is what we develop now.

8.2.1 A Scalar Example

Before passing to more general classes of systems, let us reconsider the fol-
lowing rst order system similar to the one presented in Subsection 1.4:

x(t) = a x(t) + b u(t) (8.84)

where x(t) IR, a and b are constant parameters, and u(t) IR is the input
signal. The control objective is to make the state x() track some desired signal
xd () IR dened as follows:

xd (t) = xd (t) + r(t) (8.85)

where r() is some time function. Let us assume rst that a and b are known
to the designer and dene the tracking error as e = x xd . Then it is easy to
see that the input
1
u = (r (a + 1)x) (8.86)
b
forces the closed-loop to behave like e(t) = e(t) so that e(t) 0 as t +.
Let us assume now that a and b are unknown to the designer, but that it
is known that b > 0. Let us rewrite the input in (8.86) as u = T , where

T = ( a b+1 1
, b ) and
T
= (x, r) are the vector of unknown parameters and
the regressor, respectively. Clearly it is possible to rewrite the error dynamics
as  
e(t) = e(t) + b T (t)(t) + u(t) (8.87)
Since the parameters are unknown, let us choose (following the so-called
certainty equivalence principle, which is not a principle but mainly a heuristic
method) the control as

u(t) = T (t)(t) (8.88)

where T = (1 , 2 ) is a vector of control parameters to be estimated online.
Notice that we intentionally do not impose any structure on these parameters,
since they are not meant to represent the plant parameters, but the control
parameters: this is what is called direct adaptive control. An indirect adaptive
 8.2 Linear Invariant Systems 457

scheme would aim at estimating the plant parameters and then introducing
these estimates in the control input: this is not the case in what we shall
describe in this part of the book. Introducing (8.88) into (8.87) we obtain

e(t) = e(t) + b T (t)(t) (8.89)

where = . The reader may have a look now at (8.2) and (8.7)
to guess what will follow. The dynamics in (8.89) may be rewritten as
1 T
[e](s) = 1+s b [ ](s), where [](s) denotes the Laplace transform and s C.
Consequently a gradient estimation algorithm should suce to enable one to
b
analyze the closed-loop scheme with the passivity theorem, since 1+s is SPR.
Let us choose

(t) = (t)e(t) (8.90)
As shown in Subsection 4.3.1, this denes a passive operator e
T . The
rest of the stability analysis follows as usual (except that since we deal here
with a time-varying system, one has to resort to Barbalats Lemma to prove
the asymptotic convergence of e() towards 0. The zero state detectability
property plus Krasovskii-La Salle invariance Lemma do not suce so that the
various results exposed in Section 5.1 cannot be directly applied).

Remark 8.9. The system in (8.85) is called a model of reference, and this
adaptive technique approach is called the Model Reference Adaptive Con-
trol MRAC, a term coined by Landau [274].
One can easily deduce the storage functions associated to each subsystem
and form a Lyapunov candidate function for the overall closed-loop scheme.
One may also proceed with a Lyapunov function analysis, and then retrieve
the passivity interpretation using the results in Subsection 7.3.3.
We have supposed that b > 0. Clearly we could have supposed b < 0.
However when the sign of b is not known, then the design becomes much
more involved. A solution consists of an indirect adaptive scheme with a
modied estimation algorithm [311]. The above passivity design is lost in
such schemes.

8.2.2 Systems with Relative Degree r = 1

Let us consider the following controllable and observable system:

x(t) = Ax(t) + Bu(t)
(8.91)

y(t) = C T x(t)

with u(t) IR, y(t) IR, x(t) IRn , whose transfer function is given by

B(s)
H(s) = k = C T (sIn A)1 B (8.92)
A(s)
458 8 Adaptive Control

where s is the Laplace variable. The constant k is the high-frequency gain of

the system, and we assume in the following that
k>0
A(s) and B(s) are monic polynomials, and B(s) is Hurwitz (the system
has strictly stable zero dynamics), with known order m = n 1
The problem is basically that of cancelling the dynamics of the process with a
suitable dynamic output feedback in order to get a closed-loop system whose
dynamics matches that of a given reference model with input r(t) and output
ym (t). The reference model transfer function is given by

Bm (s)
Hm (s) = km (8.93)
Am (s)

where Hm (s) is chosen as a SPR transfer function.

The control problem is that of output tracking, i.e. one desires to nd
out a dierentiator-free dynamic output feedback such that all closed-loop
signals remain bounded, and such that lim |y(t) ym (t)| = 0. It is clear
t+
that one chooses r(t) bounded so that ym (t) is. Due to the fact that the
parameters of the polynomials A(s) and B(s) as well as k are unknown, the
exact cancellation procedure cannot be achieved. Actually the problem can
be seen as follows: in the ideal case when the process parameters are known,
one is able to nd out a dynamic output controller of the following form

T T T
u(t) = ((t)r, 1 (t), y(t), 2 (t))



1 (t) = 1 (t) + bu(t), 2 (t) = 2 (t) + by(t) (8.94)




T
= [r, 1T , y, 2T ], T = [kc , 1 , 0 , 2 ]

with 1 (t), 1 , 2 and 2 (t) IRn1 , 0 IR, and (, b) is controllable. One

sees immediately that u in (8.94) is a dynamic output feedback controller with
a feedforward term. The set of gains [k, 1 , 0 , 2 ] can be properly chosen such
that the closed-loop transfer function is

kc kB(s)(s)
H0 (s) = = Hm (s) (8.95)
((s) C(s))A(s) kB(s)D(s)
(s)
where the transfer function of the feedforward term is given by (s)C(s) while
D(s)
that of the feedback term is given by (s) .C(s) has order n 2 and D(s)
has order n 1. Notice that (s) is just the characteristic polynomial of the
matrix , i.e. (s) = (sIn1 )1 and is therefore Hurwitz. We do not
develop further the model matching equations here (see e.g. [370] or [436] for
details). Let us just denote the set of ideal controller parameters such that
(8.95) holds as . In general those gains will be combinations of the process
 8.2 Linear Invariant Systems 459

parameters. Let us now write down the state space equations of the whole
system. Notice that we have

x(t) A 0 0 B


z(t) = 0

1 (t) = 0 z(t) + b u(t) (8.96)

2 (t) bC T x(t) 0 0

from which one deduces using (8.94) that

A + B0T C T B1T B2T Bk


z(t) =
b0 C T + b1T b2T z(t) + bk r(t)
(8.97)

bC T 0 0

Now since the process parameters are unknown, so is . The controller in

(8.94) is thus replaced by its estimated counterpart, i.e. u = . This gives
rise to exactly the same closed-loop structure as in (8.97), except that
is replaced by . Notice that the system in (8.97) is not controllable nor
observable, but it is stabilizable and detectable. Also its transfer function is
exactly equal to H0 (s) when the input is r(t) and the output is y. This is
therefore a SPR transfer function.
Now we have seen in the manipulator adaptive control case that the clas-
sical way to proceed is to add and substract T in the right-hand-side of
(8.97) in order to get (see (8.96) and (8.97)) a system of the form

z(t) = Am z(t) + Bm T (t)(t) + Bm k r(t) (8.98)

where Am is given in the right-hand-side of (8.97) while Bm is in the right-

hand-side of (8.96) (actually in (8.97) the input matrix is given by Bm k ).
We are now ready to set the last step of the analysis: to this end notice that
we can dene the same type of dynamical structure for the reference model
as the one that has been developed for the process. One can dene lters of
the input r(t) and of the output ym (t) similarly to the ones in (8.94). Let us
denote their state as 1m () and 2m (), whereas the total reference model
state will be denoted as zm (). In other words one is able to write

zm (t) = Am zm (t) + Bm k r(t) (8.99)

Dening e(t) = z(t) zm (t) and introducing (8.99) into (8.98) one gets the
following error equation:

e(t) = Am e(t) + Bm T (t)(t) (8.100)

This needs to be compared with (8.7) and (8.2). Let us dene the signal
T
e1 = Cm e = C T (x xm ): clearly the transfer function Cm
T
(sI3n2 Am )1 Bm
460 8 Adaptive Control

is equal to Hm (s) which is SPR by denition. Hence the subsystem in (8.100)

is strictly passive with input T and output e1 (in the sense of Lemma 4.84)
and is also output strictly passive since it has relative degree r = 1 (see
Example 4.62). A gradient estimation algorithm of the form

= (t)e (t)
(t) (8.101)
1 1

where 1 > 0, is passive with respect to the supply rate u2 y2 with y2 =

T and u2 = e1 . Due to the stabilizability properties of the rst block in
(8.100), it follows from the Meyer-Kalman-Yakubovich Lemma that the overall
system is asymptotically stable. Indeed there exists a storage function V1 (e) =
eT P e associated to the rst block, and such that V (e, ) = V1 (x) + 12 T
is a Lyapunov function for the system in (8.100) and (8.101), i.e. one gets
V = eT qq T e 0. Notice that in general the closed-loop system is not
autonomous, hence the Krasovskii-La Salle Theorem does not apply. One has
to resort to Barbalats Lemma (see the Appendix) to prove the asymptotic
convergence of the tracking error e towards 0. Notice also that the form of V
follows from a CTCE so that Lemma 7.23 directly applies.

8.2.3 Systems with Relative Degree r = 2

Let us now concentrate on the case when the plant in (8.91) and (8.92) has
relative degree two. Let us pass over the algebraic developments that allow
one to show that there is a controller such that when the process parameters
are known, then the closed-loop system has the same transfer function as the
model reference. Such a controller is a dynamic output feedback of the form
u = T . It is clear that one can repeat exactly the above relative degree
one procedure to get a system as in (8.100) and (8.101). However this time
Hm (s) cannot be chosen as a SPR transfer function, since it has relative degree
two! Thus the interconnection interpretation through the passivity theorem
no longer works. The basic idea is to modify the input u so that the transfer
function between the estimator output and the rst block output e1 is no
longer Hm (s) but (s+ a)Hm (s) for some a > 0 such that (s+ a)Hm (s) is SPR.
1
To this end let us dene a ltered regressor = s+a [], i.e. + a = . Since
we aim at obtaining a closed-loop system such that e1 = Hm (s)(s + a)T ,
let us look for an input that realizes this goal:
 8.2 Linear Invariant Systems 461

e1 = Hm (s)(s + a)T

= Hm (s)[T + T + aT ]

= Hm (s)[T + T ( a) + aT ] (8.102)

= Hm (s)[T + T ]

= Hm (s)[u T ]

It follows that a controller of the form

u(t) = T (t)(t) + T (t)(t) (8.103)

will be suitable. Indeed one can proceed as for the relative degree one case, i.e.

add and substract T to u in order to get z(t) = Am z(t) + Bm (T (t)(t) +
T T
(t)(t)) such that the transfer function between and e1 is Hm (s)(s+a).
Then the update law can be logically chosen as

(t) = 1 (t)e1 (t) (8.104)

(compare with (8.101)), and the rest of the proof follows.

8.2.4 Systems with Relative Degree r 3

The controller in (8.103) is implementable without dierentiation of the plant

output y because the derivative is available. The extension of the underly-
ing idea towards the case r 3 would imply it is possible to have at ones
disposal an estimation algorithm that provides the higher order derivatives
of the estimates: this is not the case of a simple gradient update law. The
relative degree problem has been for a long time a major obstacle in direct
adaptive control theory. The next two paragraphs briey present two solu-
tions: the rst one uses the backstepping method that we already used in
Subsection 7.5.2 to derive a globally stable tracking controller for the exible
joint-rigid link manipulators. It was presented in [267]. The second method
is due to Morse [361]. It can be considered as an extension of the controllers
in Subsections 8.2.2 and 8.2.3. It is based on the design of update laws which
provide as well as its derivatives up to the order r 1. In the following we
shall restrict ourselves to the presentation of the closed-loop error equations:
the whole developments would take us too far.

The Backstepping Approach

Given a plant dened as in (8.91), r = n m, it is possible to design u(t) such

that the closed-loop system becomes
462 8 Adaptive Control


z(t) = A(z(t), t, )z(t) + b(z(t), t, )( T (t)(t) + 2 )





= bT (t)(z(t), t, )z(t)

(t)




(t) = A0 (t) (8.105)








= A0 (t) + en z1 (t)
(t)






(t) = Ab (t) + bz1 (t)

where (t) IR(m+n)1 , (t) IR(m+n)1 , b IRm1 , b IRr1 , z(t)

IRr1 , en IRn1 and is the nth coordinate vector in IRn , , (t) IRn1 ,
(t) IRm1 . z1 (t) is the rst component of z(t) and z1 (t) = y(t)yr (t) is the
tracking error, yr (t) is the reference signal; all other terms come from ltered
values of the input u(t) and the output y(t). Ab and A0 are stable matrices.
What is important in the context of our study is that the closed-loop system
in (8.105) can be shown to be stable using the function

V (z, , , , ) = Vz (z) + V () + V () + V () + V () (8.106)

whose time derivative along trajectories of (8.105) is


n
V i zi2 T T T (8.107)
i=1

with Vz (),V (),V (),V (),V () positive denite functions, i > 0, 1 i ,

> 0, > 0, > 0.
Now let us have a look at the equations in (8.105): note that we can rewrite
the closed-loop system similarly as in (7.47) and (7.48) as follows (e1 is the
rst component vector in IRr ):

z(t) A 0 0 z(t) b T (t) b2 (t)


(t) = en eT1 A0 0 (t) + 0n + 0n (8.108)


T
be1 0 Ab (t) 0m 0m
(t)

(t) = (t)bT z(t). (8.109)
We can thus directly conclude from Lemma 7.23 that the closed-loop system
can be transformed into a system in P 5 . With the notations of the preceding
section, we get V1 =V = 12 T 1 , V2 =Vz +V +V , y2 =u1=z, y1 =u2 =b T ,
5
2 can be seen as a L2 -bounded disturbance and is therefore not important in our
study.
 8.2 Linear Invariant Systems 463
Vz

z

Ir V

= ), g2 = 0nr , ( x22 =
1
V
g1 = b T
( V
). The CTCE is veried

0mr
V

V1 T V2 T T T
as x1 g1 u1 = x2 g2 u2 =z b .

Morses High Order Tuners

Similarly to the preceding case, we only present here the closed-loop equations
without entering into the details on how the dierent terms are obtained. The
interested reader can consult the original paper [361,383] for a comprehensive
study of high order tuners. The closed-loop equations are the following:

m
e(t) = e(t) + q0 T (t)(t) + q0 i (t)czi (t) + (8.110)
i=1

zi (t) = Azi (t)(1 + i2 (t)) sign(q0 )A1 bi (t)e(t), i m (8.111)

ki (t) hi (t) = czi (t), i m (8.112)
= sign(q )(t)e(t)
(t) (8.113)
0

where m = {1, ..., m}, e is the scalar tracking error, > 0, q0 is the high
frequency gain of the open-loop system, |q0 | q, k IRm is the vector of
estimated parameters to be tuned, h() is an internal signal of the high order
tuner, = h qP , qP IRm is a vector of unknown parameters, (c, A, b) is
the minimal realization of a stable transfer function, IRm is a regressor,
and is an exponentially decaying term due to non-zero initial conditions.
ki and hi denote the ith component of k and h respectively, whereas is a
A1 b

constant satisfying > 2mq

c
.
P
T

. It is proved in [361] that the system

in (8.110) through (8.113) is stable using the function

m
V (e, , z) = e2 + |q0 |T + ziT P zi (8.114)
i=1

T
where AT P + P A = Im , =
Pq
c
A1 b

. The time derivative of V () along

trajectories of (8.110) through (8.113) is given by
1 2
V (e, , z)
e2 + (8.115)


1
with
= 2mq
c
.
P
T
A b

. Now let us rewrite the system in (8.110)

through (8.113) as follows:
464 8 Adaptive Control
0 1
e(t)
B z1 (t) C
B C
B z2 (t) C =
B C
@ ... A
zm (t)
0 1
q0 1 c ... ... ... q0 m c
B sgn(q0 )A1 b1 A(1 + 12 ) 0 ... ... 0 C
B C
=B
B sgn(q0 ) A 1
b 2 0 A(1 + 22 ) 0 ... 0 C
C (8.116)
@ ... ... ... ... ... ... A
sgn(q0 )A1 bm 0 ... ... 0 A(1 + m )2

0 1 0 1 0 1
e(t) q0 T (t)(t) (t)
B z1 (t) C B 0 C B 0 C
B C B C B C
B z
B 2 (t) C +
C B
B 0 C+B 0 C
C B C
@ ... A @ ... A @ ... A
zm (t) 0 0

= sgn(q )(t)e(t)
(t) (8.117)
0

We conclude from Corollary 4 that the system in (8.116) (8.117) belongs

1
0
m
to P, with V1 = |q0 |T , V2 = e2 + i=1 ziT P zi , g1 = q0 , g2 =
0 ,

...
0
u1 = y2 = e, u2 = y1 = q0 T . (We can neglect in the analysis or
consider it as a L2 -bounded disturbance).
Comparing Euqtaions (8.108) and (8.109) and Equations (8.116) and
(8.117) we conclude that the closed-loop error equations in both cases are
very much similar. However, this similarity is limited to the closed-loop sys-
tem stability analysis. First, the basic philosophies of each scheme are very
dierent: Roughly speaking, the high order tuners philosophy aims at render-
ing the operator between the tracking error and the estimates strictly passive
(using a control input that is the extension of classical certainty equivalent
control laws), while preserving stability of the overall system with an appro-
priate update law. On the contrary, the backstepping method is based on the
use of a very simple classical update law (a passive gradient), and the di-
culty is to design a control input (quite dierent in essence from the certainty
equivalent control laws) which guarantees stability. Second, notice that in
(8.109) truly represents the unknown parameters estimates, while in (8.117)
is the dierence between the vector of unknown plant parameters and a signal
h() internal to the high order update law (the control input being computed
with the estimates k and their derivatives up to the plant relative degree mi-
nus one). Third, the tracking error in the backstepping scheme is part of a
r-dimensional dierential equation (see the rst equation in (8.105)), while it
 8.2 Linear Invariant Systems 465

is the solution of a rst order equation in the high order tuner method (see
(8.110)).
In [383], it is proved that the high order tuner that leads to the error
equations in (8.110) through (8.113) denes a passive operator between the
tracking error e and (k qP )T , and that this leads to nice properties of
the closed-loop system, such as guaranteed speed of convergence of the track-
ing error towards zero. In [268], it has been shown that the backstepping
method also possesses interesting transient performances. Such results tend
to prove that the schemes that base on passivity properties possess nice closed-
loop properties. Other types of adaptive controllers using passivity have been
studied in [388].
9
Experimental Results

In this chapter we present experimental results on three experimental mechan-

ical systems. They illustrate the applicability of the methodologies exposed in
the foregoing chapters. The rst set of experiments concerns exible-joint ma-
nipulators, whose dynamics and control have been thoroughly explained. The
second focuses on an underactuated system, the inverted pendulum, which
does not fall into the classes of mechanical systems presented so far. The
reader is referred to the introduction of Chapter 4 where a list of applications
of passivity to control design is given.

9.1 Flexible Joint Manipulators

9.1.1 Introduction

The state feedback control problem of exible joint manipulators has consti-
tuted an interesting challenge in the Systems and Control and in the Robotics
scientic communities. It was motivated by practical problems encountered
for instance in industrial robots equipped with harmonic drives, that may
decrease the tracking performances, or even sometimes destabilize the closed-
loop system. Moreover as we pointed out in the previous chapter, it repre-
sented at the end of the 1980s (twentieth century) a pure academic problem,
due to the particular structure of the model. From a historical point of view,
the main directions that have been followed to solve the tracking and adaptive
control problems have been: singular perturbation techniques (the stability
results then require a high enough stiness value at the joints so that the
stability theoretical results make sense in practice) [475, 476], and nonlinear
global tracking controllers derived from design tools such as the backstepping
or the passivity-based techniques. We have described these last two families
of schemes in the previous chapter; see Sections 7.5 and 7.5.2. In this section
we aim at illustrating on two laboratory processes how these schemes work
in practice and whether they bring signicant performance improvement with
468 9 Experimental Results

respect to PD and the Slotine and Li controllers (which can both be cast into
the passivity-based schemes, but do not a priori incorporate exibility eects
in their design). What follows is taken from [79,80]. More generally the goal of
this section is to present experimental results for passivity-based controllers
with increasing complexity, starting from the PD input. Let us stress that
the reliability of the presented experimental works is increased by the fact
that theoretical and numerical investigations predicted reasonably well the
obtained behaviours of the real closed-loop plants; see [78]. The experimen-
tal results that follow should not be considered as a denitive answer to the
question: What is the best controller?. Indeed the answer to such a question
may be very dicult, possibly impossible to give. Our goal is only to show
that the concepts that were presented in the previous chapters may provide
good results in practice.

9.1.2 Controller Design

In this work the model as introduced in [471] is used; see (6.105). As we saw in
Section 6.4 this model possesses nice passivity properties as well as a triangular
structure that make it quite attractive for control design; see Sections 7.5, 7.5.2
and 7.6.1. Only xed parameter controllers are considered here. As shown
in [78] (see (7.123) and (7.140)), the three nonlinear controllers for exible
joint manipulators which are tested can be written shortly as follows:

Controller 1

u = J[q2d 2q2 2q2 K(s1 + s1 )] + K(q2 q1 )
(9.1)

q2d = K 1 uR + q1

Controller 2

u = J[q2d 2q2 2q2 (s1 + s1 )] + K(q2 q1 )
(9.2)

q2d = K 1 uR + q1

Controller 3

u = J q2r + K(q2d q1d ) B2 s2
(9.3)

q2d = K 1 uR + q1d

where uR = M (q1 )q1r + C(q1 , q1 )q1r + g(q1 ) 1 s1 is as in (7.124). The

signals q1r = q1d q1 , s1 = q1 + q1 are the classical signals used in the
design of this controller (the same denitions apply with subscript 2). Let us
reiterate that the expressions in (9.1), (9.2) and (9.3) are equivalent closed-
 9.1 Flexible Joint Manipulators 469

loop representations. In particular no acceleration measurement is needed for

the implementation, despite the fact that s1 may appear in the equivalent
form of u.
As pointed out in Remark 7.33, the last controller is in fact an improved
version (in the sense that it is a static state feedback) of the dynamic state
feedback proposed in [72, 318], that can be written as

t

u = J q2r K[q1d q2d 0
(1 q1 2 q2 )d ] 2 s2
2 3 (9.4)
q2d = p[pI + 2 ]1 K 1 uR + q1d  t (1 q1 2 q2 )d

0

with p C. This controller has not been considered in the experiments,

because it is logically expected not to provide better results than its simplied
counterpart: it is more complex, but based on the same idea. Controllers 1
and 2 are designed following a backstepping approach. The two backstepping
controllers dier from the fact that in Controller 2, the joint stiness K no
longer appears before s1 + s1 in the right-hand-side of the u-equation. This
modication is expected to decrease signicantly the input magnitude when
K is large. This will indeed be conrmed experimentally.
In [78] these controllers have been commented and discussed from several
points of views. Most importantly it was shown that when the joint sti-
ness grows unbounded (i.e. the rigid manipulator model is retrieved), then
the only controller that converges to the rigid Slotine and Li control law is
the passivity-based Controller 3 in (9.3). In this sense, it can be concluded
that Controller 3 is the extension of the rigid case to the exible joint case,
which cannot be stated for the other two control laws. We believe that this
elegant physical property plays a major role in the closed-loop behaviour of
the plant. As shown in Section 7.5.2 the backstepping schemes presented here
do possess some closed-loop passivity properties. However they are related
to transformed coordinates, as the reader may see in Section 7.5.2. On the
contrary, the passivity-based schemes possess this property in the original
generalized coordinates q: consequently they are closer to the physical system
than the other schemes. This is to be considered as an intuitive explanation
of the good experimental results obtained with passivity-based schemes (PD,
Slotine and Li, and Controller 3).

9.1.3 The Experimental Devices

This subsection is devoted to present the two experimental devices in detail:

a planar two degree-of-freedom (dof) manipulator, and a planar system of
two pulleys with one actuator. They are shown in photographs 9.34 and 9.35
respectively. We shall concentrate on two points: the mechanical structure and
the real time computer connected to the process. Actually we focus essentially
in this description on the rst plant, that was a two dof planar manipulator
470 9 Experimental Results

of the Laboratoire dAutomatique de Grenoble, France, named Capri. The

second process is much more simple and is depicted in Figure 9.1. It can be
considered as an equivalent one dof exible joint manipulator. Its dynamics is
linear. Its physical parameters are given by: I1 = 0.0085 kg.m2 , I2 = 0.0078
kg.m2 , K = 3.4 Nm/rad.

I2 I1

q2 q1
Fig. 9.1. A one dof exible joint manipulator.

Mechanical Structure of the Capri Robot

The Capri robot is a planar mechanism constituted by two links, of respective

lengths 0.16 and 0.27 m, connected by two hubs. The rst link is an aluminium
AU4G, U-frame to improve stiness, with respect to the forearm which can
be designed less rigid. The second link has a more peculiar structure because
it supports the applied forces: It is designed as a pipe of diameter 0.05 m, and
it is equipped with force piezo-electric sensors. The force magnitude, point
of application and orientation can be measured and calculated. The sides
of the forearm with Kistler quartz load washers can measure extension and
compression forces, and the half-spherical extremity possesses a Kistler three
components force transducer (only two of them are used) from which it is
possible to calculate the magnitude and the orientation of the applied force.
In this work these force measurement devices are not needed, since we are
concerned by motion control only.
The robot arm is actuated by two DC motors located at the underside of
the basement table (therefore the Capri robot is a parallel-drive manipulator
for which the model in (6.105) is the exact one; see Remark 6.43). They are
coupled to the links by reducers (gears and notched belts), each of them with
ratio 1/50. The rst motor (Infranor MX 10) delivers a continuous torque of
30 N.cm and a peak torque of 220 N.cm for a total weight of 0.85 kg. The
second motor (Movinor MR 08) provides a continuous torque of 19 N.cm and
a peak torque of 200 N.cm, for a weight of 0.65 kg. The drive arrangement
is such that the weight is not boarded on the links, to increase speed. Both
motors are equipped with a 500 pulses/turn incremental encoder and a DC
tachometer making joint position q2 and velocity q2 available for feedback. The
 9.1 Flexible Joint Manipulators 471

position q1 is measured by a potentiometer mounted on the last link. In the

experiments the velocity q1 has been obtained by dierentiating the position
signal (a ltering action has been incorporated by calculating the derivative
from one measurement every four only, i.e. every four sampling times).
The eective working area of the robot arm is bounded by sensors: an in-
ductive sensor prevents the rst arm from doing more than one turn, i.e.
q11 [ 2 , 2 ] (see Figure 9.2 for the denition of the angles). Two mi-
croswitches prevents the second arm from overlapping on the rst one. They
both inhibit the inverters (Infranor MSM 1207) controlling the DC motors.

Remark 9.1. The Capri robot has been modeled as a parallel-drive rigid-link
robot, with the second joint elastic. It is clear that such a model is only a
crude approximation of the real device. Some approximations may be quite
justied, like the rigidity of the rst joint and of the links. Some others are
much more inaccurate.
i) The belt that couples the second actuator and the second joint is modeled
as a spring with constant stiness, which means that only the rst mode
of its dynamic response is considered.
ii) There is some clearance in the mechanical transmission (especially at the
joints, due to the belts and the pulleys), and a serious amount of dry
friction.
iii)The frequency inverters that deliver the current to the motors possess
a nonsymmetric dead zone. Therefore, dierent amounts of current are
necessary to start motion in one direction or the other.
iv) The value of q1 used in the algorithm and obtained by dierentiating a
potentiometer signal is noisy, despite a ltering action.
v) The inertial parameters have been calculated by simply measuring and
weighting the mechanical elements of the arms. The second joint stiness
has been measured statically o-line. It has been found to be 50 Nm/rad.
This value has been used in the experiments without any further identi-
cation procedure.
vi) Some saturation on the actuators currents has been imposed by software,
for obvious safety reasons. Since nothing a priori guarantees stability when
the inputs are saturated, the feedback gains have to be chosen so that the
control input remains inside these limits.
Some of these approximations stem from the process to be controlled, and
cannot be avoided (points i, ii, iii): this would imply modifying the mechanical
structure. The measurement noise eects in iv could perhaps be avoided via
the use of observers or of position dynamic feedbacks. However on one hand the
robustness improvement is not guaranteed and would deserve a deep analytical
study. On the other hand the structure of the obtained schemes would be
signicantly modied (compare for instance the schemes in Sections 7.3.4
and 7.4 respectively). A much more simple solution consists of replacing the
potentiometer by an optical encoder. The saturation in vi is necessary to
472 9 Experimental Results

protect the motors, and has been chosen in accordance with the manufacturer
recommendations and our own experience on their natural robustness. The
crude identication procedure in v has been judged sucient, because the
aim of the work was not to make a controller perform as well as possible in
view of an industrial application, but rather to compare several controllers
and to show that nonlinear control schemes behave well. In view of this the
most important fact is that they be all tested with the same (acceptable)
parameters values, i.e. if one controller proves to behave correctly with these
set of parameters, do the others behave as well or not? Another problem is
that of the choice of the control parameters, i.e. feedback gains. We will come
back on this important point later.

Real-time Computer

A real-time computer was connected to both processes in the workshop of

the Laboratoire dAutomatique de Grenoble. It consisted of a set of DSpace
boards and a host PC. The PC is a HP Vectra running at 66 MHz with 8 Mo
of RAM and a hard disk of 240 Mo. The DSpace system is made of:
A DS 1002 oating-point processor board built around the Texas Instru-
ments TMS/320C30 digital signal processor. This processor allows 32 bits
oating point computation at 33 MFlops. A static memory of 128 K words
of 32 bits is available on this board. A 2 K words dual-port RAM is used
simultaneously by the host PC and the DSP.
A DS 2002 multi-channel ADC board with 2 A/D 16 bits resolution con-
verters (5 s conversion time) and a 16 channel multiplexer for each con-
verter.
A DS 2001 D/A converter board comprising 5 parallel analog output chan-
nels with 12 bits DAC (3 s conversion time)
A DS 3001 incremental encoder board with 5 parallel input channels. A
4-fold pulse multiplication, a digital noise lter and a 24 bits width counter
are used for each channel.
A DS 4001 digitak I/O and timer board with 32 digital I/O lines cong-
urable as inputs or outputs in groups of 8 lines.
All these boards are attached together by the 32 bits PHS-Bus at a 16
MB/sec transfer speed. They are located in a separate rack connected to the
host PC by a cable between two adaptation boards.
The PC is used for developments and supervision of the application. Sev-
eral softwares are available for the DSpace system:
SED30 and MON30 are used to congure the hardware.
C30 is the Texas Instruments Compiler for the TMS320C30.
TRACE30W is a graphical real-time software which permits to display the
selected variables of the application.
 9.1 Flexible Joint Manipulators 473

The application itself was made of two parts: The control algorithm runing
on the DSP, sampled at 1 ms in our case, and the dialogue interface running
on the PC which allows the operator to supervise the execution of the control
through the dual port memory. To guarantee repeatability of the experiments,
there was an initialization procedure that was to be activated each time the
origins have been lost, or at the beginning of the experiments.

9.1.4 Experimental Results

In this section we present the experimental results obtained by implement-

ing the three controllers described above on each plant. A PD controller as
in (7.142), and the scheme in (7.68) have also been implemented, as if the
manipulator had both joints rigid (i.e. one replaces q in (7.68) by q2 ). This
allows to dissociate clearly the eects of the nonlinearities (the reference tra-
jectories have been chosen fast enough so that Coriolis and centrifugal eects
are eective), from the eects of the exibility (once the rigid controllers
are implemented, one can see how the exible ones improve the closed-loop
behaviour, if they do). In the case of the linear system in Figure 9.1, the
scheme in (7.68) reduces to a PD control.

q12

q11

Fig. 9.2. Joint angles on the Capri robot

In order to perform the experiments, three dierent desired trajectories

have been implemented for the Capri robot (see Figure 9.2 for the deni-
tion of the angles, due to the fact that the Capri robot is a parallel-drive
manipulator):
 
q11d 0.8 sin(f t)
Desired trajectory 1: q1d = =
q 0.8 sin(f t)
12d 
0.4 sin(2f t)
Desired trajectory 2: q1d =
0.8 sin(f t)
474 9 Experimental Results

b5
(s+b)5 [g(t)]

Desired trajectory 3: q1d =
b5
(s+b) 5 [g(t)]

with f = (1 exp(at))4 , a = 14, = 9 rad/s, g(t) is a square function

with magnitude 0.8 rad, period 5 s, and b = 30. The variable s is the Laplace
transform variable. The choice for f allows one to smooth suciently the de-
sired orbit to be tracked, as required by the theoretical developments. The
other parameters values have been chosen so that the nonlinearities and the
exibilities eects are signicant. Concerning the system in Figure 9.1, two
b5
desired trajectories have been chosen: q1d = sin(t) and q1d = (s+b) 5 [g(t)].

The parameters and b have been varied as indicated in the gures cap-
tions. These time functions, which are suciently dierent to one another,
have been chosen to permit to conclude about the capability of adaptation
of the controllers to a modication of the desired motion. This is believed to
constitute an important property in applications, since it dispenses the user
from retuning the control gains between two dierent tasks. As a matter of
fact, the following criteria have been retained to evaluate the performance of
the controllers:
The tracking error during the steady-state regime is an important parame-
 20
ter for performance evaluation. The quadratic errors sums ei = 10 qi2 (t)dt
for each joint (i = 1, 2 for the Capri robot and i = 3 for the pulleys) and
the maximum tracking error (pulleys) have been computed on-line.
The shape and magnitude of the input signal.
The capabilities of the various control schemes to provide an acceptable
performance for any of the above desired motions, without having to retune
the feedback gains.
The transient behaviour has not been included in this list. This will be
explained from the observation of the experimental results. Let us emphasize
that the presented results therefore concern two quite dierent plants (one
nonlinear with high stiness, the other one linear and with high exibility),
and with signicantly dierent motions. They are consequently expected to
provide an objective view of the capabilities of the various controllers.

Remark 9.2 (Feedback gains tuning method). Two methods have been em-
ployed to tune the gains. From a general point of view, one has to confess
that one of the main drawbacks of nonlinear controllers such as backstepping
and passivity-based ones, is that Lyapunov-like analysis does not provide the
designer or the user with any acceptable way to tune the gains. The fact that
increasing the gains accelerates the convergence of the Lyapunov function to-
wards zero, is a nice theoretical result, that happens to be somewhat limited
in practice.
Concerning the Capri robot, experiments were started with the rst link
xed with respect to the base, i.e. with only the second link to be controlled.
 9.1 Flexible Joint Manipulators 475

The gains of the PD input were chosen from the second-order approximation
obtained by assuming an innite joint stiness. From the fact that the Slotine
and Li scheme in (7.68) mainly consists of a PD action plus a nonlinear part,
these values have been used as a basis for the tuning of the gains and 1
in (7.68). The full-order system is linear of order 4 (a one degree-of-freedom
exible joint manipulator). The gains were tuned by essentially placing the
closed-loop poles according to simple criteria like an optimal response time,
nonoscillatory modes. In all cases, the desired trajectory 1 was used to de-
termine a rst set of gains. This provided a basis to choose the gains for the
complete robot. Experiments were started with trajectory 1, and the gains
were modied in real-time (essentially by increasing them in a heuristic man-
ner) until the performance observed through the TRACE30W could no more
be improved. Then trajectories 2 and 3 were tested, and the gains modied
again if needed.
It has to be stressed that even in the linear case (like for the pulley-
system), tuning the gains of such nonlinear controls is not evident. Indeed the
gains appear quite nonlinearly in the state feedback, and their inuence on
the closed-loop dynamics is not obvious. For instance it is dicult to nd a
region in the gain space of the passivity-based controller in (9.3), such that
the gains can be modied and at the same time the poles remain real.
In view of these limitations and of the lack of a systematic manner to
calculate optimal feedback gains, advantage has been taken in [79] of the
pulley-system linearity. Since this system is linear, the controllers in (9.1),
(9.2) and (9.3) reduce to linear feedbacks of the form u = Gx + h(t), where
h(t) accounts for the tracking terms. De Larminat [279] has proposed a sys-
tematic (and more or less heuristic) method to calculate the matrix G for LQ
controllers. Actually one should notice that despite the fact that the nonlin-
ear backstepping and passivity-based controllers have a linear structure when
applied to a linear system, their gains appear in a very nonlinear way in the
state feedback matrix G. As an example, the term multiplying q1 for the
scheme in (9.3) is equal to (2 + k) k1 + (2 + I2 ) I1I+
1
1
+ I2 I11 (the
gains 1 and 2 can be introduced in (7.123) and (7.124) respectively instead
of using only one gain in both expressions, so that the passivity-based con-
troller has three gains). The tuning method proposed in [279] that applies to
LQ controllers allows one to choose the weighting matrices of the quadratic
form to be minimized, in accordance with the desired closed-loop bandwidth
(or cut-o frequency c (CL)). The advantages of this method are that the
user focuses on one closed-loop parameter only to tune the gains, which is
quite appreciable in practice. Therefore one gets an optimal state feedback
matrix GLQ , with a controller u = GLQ x in the case of regulation. Since the
various controllers used in the experiments yield some state feedback matrices
GPD , GBACK1 , GBACK2 and GMES respectively, which are (highly) nonlinear
functions of the gains as shown above, we choose to calculate their gains so
that the norms ||GLQ GCONT || are minimum. This amounts to solving a
nonlinear set of equations f (Z) = 0 where Z is the vector of gains. This is in
476 9 Experimental Results

general a hard task, since we do not know a priori any root (otherwise the job
would be done!). This has been done numerically by constructing a grid in the
gain space of each scheme and minimizing the above norm with a standard
optimization routine. The experimental results prove that the method may
work well, despite possible improvements (especially in the numerical way to
solve f (Z) = 0). Its extension towards the nonlinear case remains an open
problem.

The quadratic error sums e1 , e2 are reported in Tables 9.1 and 9.2. The
error e3 is in Table 9.3. The maximum tracking errors |q1 qd |max for the
pulley-system are reported in Table 9.4. All the results for the pulley-system
in Tables 9.3 and 9.4 concern the desired motion q1d = sin(t). In each case
the presented gures represent an average of several experiments. Concerning
trajectories 2 and 3 in Tables 9.1 and 9.2, the results outside brackets have
been obtained after having retuned the feedback gains. The ones in brackets
have been obtained using the same gains as for trajectory 1. When they are
not modied, it means that we have not been able to improve the results.
A cross x indicates that no feedback gains have been found to stabilize the
system.
The next results that concern the Capri robot are reported in Figures 9.3
9.21 and 9.33. The tracking errors q11 , q12 and the inputs (currents) Ic1 and
Ic2 at each motor, are depicted in Figures 9.39.13. Figures 9.149.21 con-
tain results concerning the transient behaviour when the second link position
tracking errors are initially of 0.4 rad. The inputs Ic1 and Ic2 are the calcu-
lated ones, not the true input of the actuators (they coincide as long as there
is no saturation, i.e. Ic1 2 A and Ic2 2 A). The results concerning the
pulley-system are in Figures 9.229.32. The signals qd (t) and q1 (t) are shown
in the upper boxes, and the torque input u is depicted in the lower boxes.
The following comments can be made:

Adaptation to the Desired Motion

The gains of the PD controller that correspond to the tests on the Capri
robot, reported in Tables 9.1 and 9.2, are given in Table 9.5. They show that
signicant changes have been necessary from one desired motion to the next.
One sees that the PD gains have had to be modied drastically to maintain
a reasonable performance level. On the contrary it is observable from Tables
9.1 and 9.2 that even without any gain modication, the other controllers still
perform well in general. In any case the modications have seldom exceeded
50 % and concerned very few gains [80]. Since this is also true for the Slotine
and Li controller, we conclude that the insensitivity of the performance with
respect to desired motion changes is essentially due to the compensation of
the nonlinearities.
The Slotine and Li controller seems to provide the most invariant per-
formance with respect to the desired motion. This is especially apparent for
 9.1 Flexible Joint Manipulators 477

0.015 0.3

0.01 0.2

0.1
0.005
q11t [rad]

q12t [rad]
0
0
0.1
0.005
0.2
0.01 0.3

0.015 0.4
0 5 10 15 0 5 10 15

4
2

2
1
Ic1 [A]

Ic2 [A]
0 0

1
2

2
4
0 5 10 15 0 5 10 15

Fig. 9.3. PD controller, desired trajectory 1

0.08 0.5

0.06

0.04
q11t [rad]

q12t [rad]

0.02 0

0.02

0.04 0.5
0 5 10 15 0 5 10 15

4
2

2
1
Ic1 [A]

Ic2 [A]

0 0

1
2

2
4
0 5 10 15 0 5 10 15

Fig. 9.4. PD controller, desired trajectory 2

478 9 Experimental Results

trajectory 2 on the Capri experiments. In this case it provides the best error
e2 , even after having retuned the gains for Controllers 2 and 3. This may
be explained by the fact that the input in (7.68) is much smoother than the
others (see Figure 9.7). This in turn may be a consequence of its simplicity,
and from the fact that it does not use the noisy potentiometer signal.

0.6 1

0.4 0.8

0.2 0.6
q11t [rad]

q12t [rad]
0 0.4

0.2 0.2

0.4 0

0.6 0.2

0.8 0.4
0 5 10 15 0 5 10 15

4
2

2
1
Ic1 [A]

Ic2 [A]

0 0

1
2

2
4
0 5 10 15 0 5 10 15

Fig. 9.5. PD controller, desired trajectory 3

Backstepping Controllers
For the Capri experiments, it has not been possible to nd feedback gains
that stabilize controller 1. On the contrary this has been possible for the
pulley-system, see Figures 9.30, 9.23 and 9.26. This conrms the fact that the
modication of the intermediate Lyapunov function (see (7.139) and (7.140))
may play a signicant role in practice, and that the term K(s1 + s1 ) is a
high-gain in the loop if K is large.

Compensation of Nonlinearities
Although the PD algorithm provides a stable closed-loop behaviour in all cases
(for the Capri experiments and at the price of very large gain modications as
we pointed out above), its performance is poor for trajectories 1 and 2. The
behaviour is much better for trajectory 3. This can be explained since this is
almost a regulation task. The improvements obtained with the Slotine and Li
scheme show that the Coriolis and centrifugal terms may play an important
role depending on the desired motion.
 9.1 Flexible Joint Manipulators 479

0.015 0.3

0.01
0.2
0.005
0.1
q11t [rad]

q12t [rad]
0

0.005 0
0.01
0.1
0.015

0.02 0.2
0 5 10 15 0 5 10 15

4
2

2
1
Ic1 [A]

Ic2 [A]
0 0

1
2

2
4
0 5 10 15 0 5 10 15

Fig. 9.6. SLI controller, desired trajectory 1

3
x 10
6 0.4

4 0.2
2
0
q11t [rad]

q12t [rad]

0
0.2
2

4 0.4

6 0.6
0 5 10 15 0 5 10 15

4
2

2
1
Ic1 [A]

Ic2 [A]

0 0

1
2

2
4
0 5 10 15 0 5 10 15

Fig. 9.7. SLI controller, desired trajectory 2

480 9 Experimental Results

0.2 1.2

1
0
0.8
0.2
q11t [rad]

q12t [rad]
0.6

0.4 0.4

0.2
0.6
0

0.8 0.2
0 5 10 15 0 5 10 15

4
2

2
1
Ic1 [A]

Ic2 [A]
0 0

1
2

2
4
0 5 10 15 0 5 10 15

Fig. 9.8. SLI controller, desired trajectory 3

Compensation of Flexibilities

The PD and the Slotine and Li controls behave well for the Capri robot
because the joint stiness is large. The results obtained for the pulley-system
show that the behaviour deteriorates a lot if K is small; see Tables 9.3 and
9.4.

Controller Complexity

The rather complex structure of the nonlinear Controllers 1, 2 and 3 is not

an obstacle to their implementation with the available real-time computer de-
scribed above. In particular recall that the acceleraton and jerk are estimated
by inverting the dynamics (see Section (7.5)). Such terms have a complicated
structure and depend on the systems physical parameters in a nonlinear way.
Some experiments have shown that the sampling period (1 ms) could have
been decreased to 0.5 ms.

Torque Input

The major problem that prevents certain controllers from behaving correctly is
the input magnitude and shape. This has been noted above. The performance
of Controllers 2 and 3 may be less good than that of the Slotine and Li
algorithm, mainly because of the chattering in the input, inducing vibrations
in the mechanical structure. Chattering is particularly present during the
 9.1 Flexible Joint Manipulators 481

0.015 0.4

0.2
0.01
0
0.005
q11t [rad]

q12t [rad]
0.2

0 0.4

0.6
0.005
0.8

0.01 1
0 5 10 15 0 5 10 15

4
2

2
1
Ic1 [A]

Ic2 [A]
0 0

1
2

2
4
0 5 10 15 0 5 10 15

Fig. 9.9. Controller 2, desired trajectory 1

0.015 0.2

0.01 0.1

0.005 0
q11t [rad]

q12t [rad]

0 0.1

0.005 0.2

0.01 0.3
0 5 10 15 0 5 10 15

4
2

2
1
Ic1 [A]

Ic2 [A]

0 0

1
2

2
4
0 5 10 15 0 5 10 15

Fig. 9.10. Controller 2, desired trajectory 2

482 9 Experimental Results

regulation phases in Ic2 for trajectory 3 and Controllers 2 and 3; see Figures
9.11 and 9.13. On the contrary Figures 9.5 and 9.8 show smooth inputs. It
may be expected from Figures 9.189.21 that a less noisy velocity q1 obtained
from a better position measurement would bring the shape of Ic2 close to the
input in gures 9.16 and 9.17. Indeed they dier only in terms of chatter. One
concludes that an optical encoder to measure q1 would be a better solution.

Backstepping vs Passivity-based Controls

It is noteworthy that Controllers 2 and 3 possess quite similar closed-loop

behaviours; see Figures 9.31 and 9.32, 9.24 and 9.25, 9.27 and 9.28 for the
pulley-system, 9.9 and 9.33, 9.10 and 9.12, 9.11 and 9.13 for the Capri robot
(although Ic2 chatters slightly less for Controller 3, see Figures 9.9 and 9.33,
and 9.11 and 9.13). The advantage of passivity-based methods is that the
controllers are obtained in one shot, whereas the backstepping approach a
priori leads to various algorithms. This can be an advantage (more degrees
of freedom), but also a drawback as Controller 1 behaviour proves. Notice on
gures 9.29, 9.30, 9.31 and 9.32 that Controllers 2 and 3 allow one to damp
the oscillations much better than Controller 1 and the PD (it is possible that
the PD gains could have been tuned in a better way for these experiments;
see however the paragraph below on gain tuning for the pulley-system).

Transient Behaviour

The transient behaviour for the tracking error q12 can be improved slightly
when the exibilities are taken into account in the controller design. This can
be seen by comparing gures 9.6 and 9.7 with gures 9.9 and 9.10, 9.33 and
9.12. The tracking error tends to oscillate more for the Slotine and Li scheme
than for the others. Notice that these results have been obtained with initial
tracking errors close to zero. However the results in Figures 9.149.21 prove
that the controllers respond quite well to initial state deviation. The transient
duration is around 0.5 s for all the controllers. The tracking errors have a
similar shape once the transient has vanished. The only signicant dierence
is in the initial input Ic2 . The torque is initially much higher for nonzero initial
conditions.

Feedback Gains Tuning

The method described in remark 9.2 for tuning the gains in the case of the
pulley-system provides good preliminary results. The gains that have been
used in all the experiments for the pulley-system have not been modied
during the tests on the real device to tentatively improve the results. They
have been kept constant. This tends to prove that such a method is quite
promising since it relies on the choice of a single parameter (the closed loop
 9.1 Flexible Joint Manipulators 483

bandwidth, chosen as c (CL) = 11 rad/s in the experiments) and is therefore

quite attractive for potential users.
The actuators and current drivers neglected dynamics may have a signif-
icant inuence on the closed-loop behaviour. A close look at Tables 9.3 and
9.4 shows the existence of a resonance phenomenon in the closed-loop. This
u
can be conrmed numerically by replacing u with uf = 1+ s which allows one
to suspect that this actuator neglected dynamics may play a crucial role in
the loop. It might be then argued that developing velocity observers for such
systems may not be so important, whereas some neglected dynamics, whose
inuence has received less attention in the literature, have a signicant eect.

Remark 9.3. The peaks in the input Ic2 for trajectory 1 are due to the satu-
ration of the DC tachometers when the trajectory is at its maximum speed.
When the saturation stops, the velocity signal delivered by the tachometers
has a short noisy transient that results in such peaks in the input. However
this has not had any signicant inuence on the performance, since such peaks
are naturally ltered by the actuators (let us recall that the calculated inputs
are depicted).

9.1.5 Conclusions

In this section we have presented experimental results that concern the ap-
plication of passivity-based (PD, Slotine and Li, the controller in Subsection
7.5.1) and backstepping controllers, to two quite dierent laboratory plants
which serve as exible joint-rigid link manipulators. The major conclusion is
that passivity-based controllers provide generally very good results. In partic-
ular the PD and Slotine and Li algorithms show quite good robustness and
provide a high level of performance when the exibility remains small enough.
Tracking with high exibility implies the choice of controllers which are de-
signed from a model that incorporates the joint compliance. These experimen-
tal results illustrate nicely the developments of the foregoing chapter: one goes
from the PD scheme to the one in Subsection 7.5.1 by adding more complexity,
but always through the addition of new dissipative modules to the controller,
and consequently to the closed-loop system. These three schemes can really
be considered to belong to the same family, namely passivity-based con-
trollers. It is therefore not surprizing that their closed-loop behaviour when
applied to real plants reproduces this dissipative modularity: the PD works
well when nonlinearities and exibilities remain small enough, the Slotine and
Li algorithm improves the robustness with respect to nonlinearities, and the
scheme in Subsection 7.5.1 provides a signicant advantage over the other two
only if these two dynamical eects are large enough. Finally it is noteworthy
that all controllers present a good robustness with respect to the uncertainties
listed in Subsection 9.1.3.
484 9 Experimental Results

0.2 0.8

0 0.6

0.2 0.4
q11t [rad]

q12t [rad]
0.4 0.2

0.6 0

0.8 0.2
0 5 10 15 0 5 10 15

4
2

2
1
Ic1 [A]

Ic2 [A]
0 0

1
2

2
4
0 5 10 15 0 5 10 15

Fig. 9.11. Controller 2, desired trajectory 3

0.02 0.5

0.015
0
0.01
q11t [rad]

q12t [rad]

0.005
0.5
0

0.005
1
0.01

0.015 1.5
0 5 10 15 0 5 10 15

4
2

2
1
Ic1 [A]

Ic2 [A]

0 0

1
2

2
4
0 5 10 15 0 5 10 15

Fig. 9.12. Controller 3, desired trajectory 2

 9.1 Flexible Joint Manipulators 485

0.2 1.2

1
0
0.8
0.2
q11t [rad]

q12t [rad]
0.6

0.4 0.4

0.2
0.6
0

0.8 0.2
0 5 10 15 0 5 10 15

4
2

2
1
Ic1 [A]

Ic2 [A]
0 0

1
2

2
4
0 5 10 15 0 5 10 15

Fig. 9.13. Controller 3, desired trajectory 3

0.2

0.1

0
q12t [rad]

0.1

0.2

0.3
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

1
Ic2 [A]

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Fig. 9.14. PD controller, desired trajectory 1, zero initial conditions

486 9 Experimental Results

0.4

0.3

0.2
q12t [rad]

0.1

0.1

0.2

0.3
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

1
Ic2 [A]

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Fig. 9.15. PD controller, desired trajectory 1, nonzero initial conditions

0.4

0.2

0
q12t [rad]

0.2

0.4

0.6

0.8

1
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

1
Ic2 [A]

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Fig. 9.16. SLI controller, desired trajectory 1, zero initial conditions

 9.1 Flexible Joint Manipulators 487

0.4

0.2
q12t [rad]

0.2

0.4
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

1
Ic2 [A]

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Fig. 9.17. SLI controller, desired trajectory 1, nonzero initial conditions

0.2

0.15

0.1
q12t [rad]

0.05

0.05

0.1

0.15
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

1
Ic2 [A]

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Fig. 9.18. Controller 2, desired trajectory 1, zero initial conditions

488 9 Experimental Results

0.5

0.4

0.3
q12t [rad]

0.2

0.1

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

1
Ic2 [A]

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Fig. 9.19. Controller 2, desired trajectory 1, nonzero initial conditions

0.2

0.1

0
q12t [rad]

0.1

0.2

0.3

0.4

0.5
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

1
Ic2 [A]

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Fig. 9.20. Controller 3, desired trajectory 1, zero initial conditions

 9.1 Flexible Joint Manipulators 489

0.6

0.4

0.2
q12t [rad]

0.2

0.4
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

1
Ic2 [A]

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Fig. 9.21. Controller 3, desired trajectory 1, nonzero initial conditions

3.5
qd,q1 (rd)

2.5
0 0.5 1 1.5 2 2.5
time(s)

5
u (Nm)

5
0 0.5 1 1.5 2 2.5
time(s)

Fig. 9.22. PD controller, = 7.5 rad/s.

490 9 Experimental Results

3.5
qd,q1 (rd)

2.5
0 0.5 1 1.5 2 2.5
time(s)

5
u (Nm)

5
0 0.5 1 1.5 2 2.5
time(s)

Fig. 9.23. Controller 1, = 7.5 rad/s

3.5
qd,q1 (rd)

2.5
0 0.5 1 1.5 2 2.5
time(s)

5
u (Nm)

5
0 0.5 1 1.5 2 2.5
time(s)

Fig. 9.24. Controller 2, = 7.5 rad/s

 9.1 Flexible Joint Manipulators 491

3.5
qd,q1 (rd)

2.5
0 0.5 1 1.5 2 2.5
time(s)

5
u (Nm)

5
0 0.5 1 1.5 2 2.5
time(s)

Fig. 9.25. Controller 3, = 7.5 rad/s

3.5
qd,q1 (rd)

2.5
0 0.5 1 1.5 2 2.5
time(s)

5
u (Nm)

5
0 0.5 1 1.5 2 2.5
time(s)

Fig. 9.26. Controller 1, = 20 rad/s

492 9 Experimental Results

3.5
qd,q1 (rd)

2.5
0 0.5 1 1.5 2 2.5
time(s)

5
u (Nm)

5
0 0.5 1 1.5 2 2.5
time(s)

Fig. 9.27. Controller 2, = 20 rad/s

3.5
qd,q1 (rd)

2.5
0 0.5 1 1.5 2 2.5
time(s)

5
u (Nm)

5
0 0.5 1 1.5 2 2.5
time(s)

Fig. 9.28. Controller 3, = 20 rad/s

 9.1 Flexible Joint Manipulators 493

3.5
qd,q1 (rd)

2.5
5 6 7 8 9 10 11 12 13 14 15
time(s)

5
u (Nm)

5
5 6 7 8 9 10 11 12 13 14 15
time(s)

Fig. 9.29. PD controller, b = 40

3.5
qd,q1 (rd)

2.5
5 6 7 8 9 10 11 12 13 14 15
time(s)

5
u (Nm)

5
5 6 7 8 9 10 11 12 13 14 15
time(s)

Fig. 9.30. Controller 1, b = 40

494 9 Experimental Results

3.5
qd,q1 (rd)

2.5
5 6 7 8 9 10 11 12 13 14 15
time(s)

5
u (Nm)

5
5 6 7 8 9 10 11 12 13 14 15
time(s)

Fig. 9.31. Controller 2, b = 40

3.5
qd,q1 (rd)

2.5
5 6 7 8 9 10 11 12 13 14 15
time(s)

5
u (Nm)

5
5 6 7 8 9 10 11 12 13 14 15
time(s)

Fig. 9.32. Controller 3, b = 40

 9.1 Flexible Joint Manipulators 495

0.015 0.1

0.01 0.05
0.005
0
q11t [rad]

q12t [rad]
0
0.05
0.005

0.01 0.1

0.015 0.15
0 5 10 15 0 5 10 15

4
2

2
1
Ic1 [A]

Ic2 [A]
0 0

1
2

2
4
0 5 10 15 0 5 10 15

Fig. 9.33. Controller 3, desired trajectory 1

Table 9.1. Quadratic error sums e1 and e2 (Capri robot)

Controller e1 (traj. 1) e2 (traj. 1) e1 (traj. 2) e2 (traj. 2)
PD 0.346 84.5 1.4 (1.6) 360 (1000)
SLI 0.11 37.9 0.02 (0.034) 40 (51)
Controller 1 x x x x
Controller 2 0.34 12 0.3 75 (173)
Controller 3 0.64 9 0.224 (0.6) 70 (150)

Table 9.2. Quadratic error sums e1 and e2 (Capri robot)

Controller e1 (traj. 3) e2 (traj. 3)
PD 0.3 (0.3) 50 (50)
SLI 0.055 (0.055) 30 (30)
Controller 1 x x
Controller 2 0.135 (0.135) 30 (30
Controller 3 0.19 (0.19) 15 (15)
496 9 Experimental Results

Table 9.3. Quadratic error sum e3 (pulley system)

(rad/s) PD Control. 1 Control. 2 Control. 3
2.5 0.70 0.21 0.25 0.33
5 3.54 2.57 1.54 2.78
7.5 20.86 8.53 4.17 7.92
10 x 20.60 13.00 19.03
12.5 x 48.07 35.15 36.05
15 x 63.44 53.33 31.03
20 x 37.70 2.97 8.58

Table 9.4. Maximum tracking error (pulley system)

(rad/s) PD Controller 1 Controller 2 Controller 3
2.5 0.0630 0.0293 0.0374 0.0386
5 0.0943 0.1138 0.0840 0.0983
7.5 0.1946 0.1501 0.1040 0.1472
10 x 0.2428 0.1823 0.2150
12.5 x 0.4138 0.2965 0.2910
15 x 0.4494 0.3418 0.2581
20 x 0.2842 0.0842 0.1364

Table 9.5. Feedback gains (Capri robot)

PD Controller traj. 1 traj. 2 traj. 3
21 1500 650 1500
22 250 10 250
11 30 4 30
12 5 3.5 5

9.2 Stabilization of the Inverted Pendulum

9.2.1 Introduction

The inverted pendulum is a very popular experiment used for educational

purposes in modern control theory. It is basically a pole which has a pivot
on a cart that can be moved horizontally. The pole moves freely around the
cart and the control objective is to bring the pole to the upper unstable
equilibrium position by moving the cart on the horizontal plane. Since the
angular acceleration of the pole cannot be controlled directly, the inverted
pendulum is an underactuated mechanical system. Therefore, the techniques
developed for fully-actuated mechanical robot manipulators cannot be used
to control the inverted pendulum.
The cart and pole system is also known because the standard nonlinear
control techniques are ineective to control it. Indeed the relative degree of
 9.2 Stabilization of the Inverted Pendulum 497

Fig. 9.34. The Capri robot of the Laboratoire dAutomatique de Grenoble

the system is not constant (when the output is chosen to be the swinging en-
ergy of the pendulum), the system is not input-output linearizable. Jakubczyk
and Respondek [231] have shown that the inverted pendulum is not feedback
linearizable. An additional diculty comes from the fact that when the pen-
dulum swings past the horizontal the controllability distribution does not have
a constant rank.

9.2.2 Systems Dynamics

Consider the cart and pendulum system as shown in Figure 9.36. We will con-
sider the standard assumptions, i.e. massless rod, point masses, no exibilities
and no friction. M is the mass of the cart, m the mass of the pendulum, con-
centrated in the bob, the angle that the pendulum makes with the vertical
and l the length of the rod. The equations of motion can be obtained either
by applying Newtons second law or by the Euler-Lagrange formulation.
The system can be written as

M (q(t))q(t) + C(q(t), q(t))q(t) + g(q(t)) = (t) (9.5)

where:
 
x M + m ml cos
q= , M (q) = (9.6)
ml cos ml2
498 9 Experimental Results

Fig. 9.35. The pulley system of the Laboratoire dAutomatique de Grenoble

y -l sin 
6 x
6 
 m

?
 mg
l cos

? 

f - M

m m
-x

Fig. 9.36. The cart pendulum system

 9.2 Stabilization of the Inverted Pendulum 499


0 ml sin
C(q, q) = (9.7)
0 0
 
0 f
g(q) = and = (9.8)
mgl sin 0
Note that M (q) is symmetric and

det(M (q)) = (M + m)ml2 m2 l2 cos2

(9.9)
= M ml2 + m2 l2 sin2 > 0
Therefore, M (q) is positive denite for all q. From (9.6) and (9.7) it follows
that

0 ml sin
M (q, q) 2C(q, q) = (9.10)
ml sin 0
which is a skew-symmetric matrix (see Lemma 6.16). The potential energy
of the pendulum can be dened as U () = mgl(cos 1). Note that U () is
related to g(q) as follows:

U 0
g(q) = = (9.11)
q mgl sin

Passivity of the Inverted Pendulum

The total energy of the cart and pole system is given by

E(q, q) = K(q, q) + U (q)

(9.12)
= 12 q T M (q)q + mgl(cos 1)
Therefore from (9.5), (9.6), (9.7), (9.8), (9.10) and (9.11) we obtain:

d
dt E(q(t), q(t)) = q T (t)M (q(t))q(t) + 12 q T (t)M (q(t))q(t) + q T (t)g(q(t))

= q T (t)(C(q(t), q(t))q(t) g(q(t)) + (t) + 12 M (q(t))q(t)) + q T (t)g(q(t))

= q T (t) (t) = x(t)f (t)

(9.13)
Integrating both sides of the above equation we obtain
t   
0 x(t )f (t )dt = E(t) E(0)
(9.14)
2mgl E(0)
500 9 Experimental Results

Therefore, the system having f as input and x as output is passive. Note

that for f = 0 and [0, 2[ the system (9.5) has a subset of two equi-
librium points. (x, x, , ) = (, 0, 0, 0) is an unstable equilibrium point and
(x, x, , ) = (, 0, , 0) is a stable equilibrium point. The total energy E(q, q)
is equal to 0 for the unstable equilibrium point and to 2mgl for the stable
equilibrium point. The control objective is to stabilize the system around its
unstable equilibrium point, i.e. to bring the pendulum to its upper position
and the cart displacement to zero simultaneously.

9.2.3 Stabilizing Control Law

Let us rst note that in view of (9.12) and (9.6), if x = 0 and E(q, q) = 0 then
1 2 2
ml = mgl(1 cos ) (9.15)
2
The above equation denes a very particular trajectory which corresponds
to a homoclinic orbit. Note that = 0 only when = 0. This means that
the pendulum angular position moves clockwise or counter-clockwise until
it reaches the equilibrium point (, ) = (0, 0). Thus our objective can be
reached if the system can be brought to the orbit (9.15) for x = 0, x = 0 and
E = 0. Bringing the system to this homoclinic orbit solves the problem of
swinging up the pendulum. In order to balance the pendulum at the upper
equilibrium position the control must eventually be switched to a controller
which guarantees (local) asymptotic stability of this equilibrium [474]. By
guaranteeing convergence to the above homoclinic orbit, we guarantee that the
trajectory will enter the basin of attraction of any (local) balancing controller.
We do not consider in this book the design of the balancing controller.
The passivity property of the system suggests us to use the total energy
E(q, q) in (9.12) in the controller design. Since we wish to bring to zero x, x
and E we propose the following Lyapunov function candidate:
kE 2 kv kx
V (q, q) =E (q, q) + x2 + x2 (9.16)
2 2 2
where kE , kv and kx are strictly positive constants. Note that V (q, q) is a
positive semi-denite function. Dierentiating V (q, q) and using (9.13) we
obtain

V (q, q) = kE E E + kv xx + kx xx

= kE E xf + kv xx + kx xx (9.17)

= x(kE Ef + kv x + kx x)
Let us now compute x from (9.5). The inverse of M (q) can be obtained
from (9.6), (9.7) and (9.9) and is given by:
 9.2 Stabilization of the Inverted Pendulum 501


1 ml2 ml cos
M 1 = (9.18)
det(M ) ml cos M + m
with det(M ) = ml2 (M + m sin2 ). Therefore we have
  
x 1 0 m2 l3 sin x
= [det(M (q))] +
0 m2 l2 sin cos
  
m2 l2 g sin cos ml2 f
+ +
(M + m)mgl sin mlf cos

Thus x() can be written as

1  
x(t) = 2 m sin (t)(l2 (t) g cos (t)) + f (t) (9.19)
M + msin (t)

Introducing the above in (9.17) one has

   2 
kv kv m sin (l g cos )
V (q, q) = x f kE E + M+msin2 + M+msin2 + kx x (9.20)

For simplicity and without loss of generality we will consider M = m =

l = 1, thus

   
kv kv sin ( 2 g cos )
V (q, q) = x f kE E + 1+sin2 + 1+sin2 + kx x (9.21)

We propose a control law such that


kv kv sin (2 g cos )
f kE E + + + kx x = kdx x (9.22)
1 + sin2 1 + sin2
which will lead to

V (q, q) = kdx x2 (9.23)

Note that other functions f (x) such that xf (x) > 0 are also possible. The
control law in (9.22) will have no singularities provided that

kv
kE E + = 0 (9.24)
1 + sin2
The above condition will be satised if for some > 0
kv
kv
|E| kE
<  kE 2  (9.25)
2 1 + sin
502 9 Experimental Results

Note that when using the control law (9.22), the pendulum can get stuck
at the (lower) stable equilibrium point, (x, x, , ) = (0, 0, , 0). In order to
avoid this singular point, which occurs when E = 2mgl (see (9.12)), we
require |E| < 2mgl i.e. |E| < 2g (for m = 1, l = 1). Taking also (9.25) into
account, we require
. /
kE
kv
|E| < c = min 2g, (9.26)
2

Since V () is a non-increasing function (see (9.23)), (9.26) will hold if the

initial conditions are such that

c2
V (0) < (9.27)
2
The above denes the region of attraction as will be shown in the next
section.

Domain of Attraction

The condition (9.27) imposes bounds on the initial energy of the system.
Note that the potential energy U = mgl(cos 1) lies between 2g and 0,
for m = l = 1. This means that the initial kinetic energy should belong to
[0, c + 2g). Note also that the initial position of the cart x(0) is arbitrary since
we can always choose an appropiate value for kx in V () in (9.16). If x(0)
is large we should choose kx small. The convergence rate of the algorithm
may however decrease when kx is small. Note that when the initial kinetic
energy K(q(0), q(0)) is zero, the initial angular position (0) should belong
to (, ). This means that the only forbidden point is (0) = . When the
initial kinetic energy K(q(0), q(0)) is dierent from zero, i.e. K(q(0), q(0))
belongs to (0, c + 2g) (see (9.26) and (9.27)), then there are less restrictions
on the initial angular position (0). In particular, (0) can even be pointing
downwards, i.e. = provided that K(q(0), q(0)) is not zero. Despite the
fact that our controller is local, its basin of attraction is far from being small.
The simulation example and the real-time experiments will show this feature.
For future use we will rewrite the control law f from (9.22) as
   
kv sin g cos 2 1 + sin2 (kx x + kdx x)
f=   (9.28)
kv + 1 + sin2 kE E
The stability analysis can be obtained by using the Krasovskii-LaSalles
invariance Theorem. The stability properties are summarized in the following
lemma.

Lemma 9.4. Consider the inverted pendulum system (9.5) and the controller
in (9.28) with strictly positive constants kE , kv , kx and kdx . Provided that the
 9.2 Stabilization of the Inverted Pendulum 503

state initial conditions satisfy the inequalities at Equations (9.26) and (9.27),
then the solution of the closed-loop system converges to the invariant set M
given by the homoclinic orbit (9.15) with (x, x) = (0, 0). Note that f () does
not necessarily converge to zero.

Proof: The proof can be found in [314].

9.2.4 Simulation Results

In order to observe the performance of the proposed control law based on an

c
energy approach of the system, we have performed simulations on MATLAB
c
using Simulink .
We have considered the real system parameters M = M + m = 1.2, ml2 =
0.0097 and ml = 0.04, and g = 9.804 ms2 of the inverted pendulum at
the University of Illinois at Urbana-Champaign. Recall that the control law
requires initial conditions such that (9.27) is satised. We have chosen the
gains kE = 1, kv = 1, kx = 102 and kdx = 1. These gains have been chosen
to increase the convergence rate in order to switch to a linear stabilizing
controller in a reasonable time. The algorithm brings the inverted pendulum
close to the homoclinic orbit but the inverted pendulum will remain swinging
while getting closer and closer to the origin. Once the system is close enough
to the origin, i.e. (|x| 0.1, |x| 0.2, || 0.3, || 0.3), we switch to the
linear LQR controller f = K[x x ]T where K = [44 23 74 11].
Figure 9.37 shows the results for an initial position:

x = 0.1, x = 0
(9.29)

= 2
3 = 0
Simulations showed that the nonlinear control law brings the system to
the homoclinic orbit (see the phase plot in gure 9.37). Switching to the
linear controller occurs at time t = 120 s. Note that before the switching the
energy E goes to zero and that the Lyapunov function V () is decreasing and
converges to zero.

9.2.5 Experimental Results

We have performed experiments on the inverted pendulum setting at the

University of Illinois at Urbana-Champaign. The parameters of the model
used for the controller design and the linear controller gains K are the same as
in the previous section. For this experiment we have chosen the gains kE = 1,
kv = 1.15, kx = 20 and kdx = 0.001. Figure 9.38 shows the results for an
initial position:
504 9 Experimental Results

Distance: x Phase plot

0.2 15

Angular velocity [rad/s]

0.15 10
0.1
Displacement

5
0.05
0
0
5
0.05

0.1 10

0.15 15
0 100 200 300 0 2 4 6
Time [s] Angle [rad]

Regulated Pendulum Energy: E Lyapunov Function: V

0.5 0.8
0.7
0.6
0
0.5
Energy

0.4
V

0.3
0.5
0.2
0.1
1 0
0 100 200 300 0 100 200 300
Time [s] Time [s]

Fig. 9.37. Simulation results for the inverted pendulum

x = 0, x = 0
(9.30)

= + 0.1, = 0.1
Real-time experiments showed that the nonlinear control law brings the
system to the homoclinic orbit (see the phase plot in Figure 9.38). Switching
to the linear controller occurs at time t = 27 s. Note that the control input
lies in an acceptable range. Note that in both simulation and experimental
results, the initial conditions lie slightly outside the domain of attraction. This
proves that the domain of attraction in (9.26) and (9.27) is conservative.

9.3 Conclusions
In the rst part of this chapter dedicated to experimental validations of
passivity-based control schemes, we have presented a set of experiments on
 9.3 Conclusions 505

Distance: x Angle: Theta

0.1 7
0.08 6
0.06 5
Displacement

Angle [rad]
0.04 4
0.02 3
0 2
0.02 1
0.04 0
0.06 1
0 10 20 30 40 0 10 20 30 40
Time [s] Time [s]

Control force Phase plot

0.5 15
Angular velocity [rad/s]
0.4 10
0.3
5
Control

0.2
0
0.1
5
0

0.1 10

0.2 15
0 10 20 30 40 0 2 4 6
Time [s] Angle [rad]

Fig. 9.38. Experimental results for the inverted pendulum

two types of manipulators with exible joints and rigid links: the rst setup is
nonlinear, with low exibility. The second setup is linear but with high exibil-
ity. Various passivity-based controllers, with increasing complexity, have been
tested on the two devices. The results are quite encouraging and show that
this design concept yields very nice results for robust tracking control. Then
we have presented a control strategy for the inverted pendulum that brings
the pendulum to a homoclinic orbit, while the cart displacement converges
to zero. Therefore the state will enter the basin of attraction of any locally
convergent controller. The control strategy is based on the total energy of the
system, using its passivity properties. A Lyapunov function is obtained using
the total energy of the system. The convergence analysis is carried out using
the Krasovskii-LaSalles invariance principle. The system nonlinearities have
not been compensated which has enabled us to exploit the physical proper-
ties of the system in the stability analysis. The proposed control strategy is
506 9 Experimental Results

proved to be applicable to a wider class of underactuated mechanical systems

(see [138, 139]).
As recalled in the introduction of Chapter 4, there are many other elds of
applications to which the passivity-based approach applies and provides good
results. Experimental results with passivity-based controllers have been pre-
sented in many other papers which are impossible to describe comprehensively
in this chapter.
A
Background Material

In this Appendix we present the background for the main tools used through-
out the book; namely, Lyapunov stability, dierential geometry for nonlinear
systems, Riccati equations, viscosity solutions of PDEs, some useful matrix al-
gebra results, and some results that are used in the proof of the KYP Lemma.

A.1 Lyapunov Stability

Let us consider a nonlinear system represented as

x(t) = f (x(t), t), x(0) = x0 (A.1)

where f () is a nonlinear vector function, and x(t) IRn is the state vec-
tor. We suppose that the system is well-posed, i.e. a unique solution exists
globally (see Section 3.9.2 for details on existence, uniqueness and continuous
dependence on parameters). We may for instance assume that the conditions
of Theorem 3.55 are satised. We refer the reader to Theorems 3.83 and 3.84
for extensions of Lyapunov stability to more general systems like evolution
variational inequalities. In this Appendix we focus on ODEs.

A.1.1 Autonomous systems

The nonlinear system (A.1) is said to be autonomous (or time-invariant) if

f () does not depend explicitly on time, i.e.,

x(t) = f (x(t)) (A.2)

Otherwise the system is called non-autonomous (or time-varying). In this

section, we briey review the Lyapunov theory results for autonomous systems
while non-autonomous systems will be reviewed in the next section. Lyapunov
theory is the fundamental tool for stability analysis of dynamic systems. The
basic stability concepts are summarized in the following denitions.
508 A Background Material

Denition A.1 (Equilibrium). A state x
is an equilibrium point of (A.2)

if f (x
) = 0.

Denition A.2 (Stability). The equilibrium point x = 0 is said to be stable

if, for any > 0, there exists r > 0 such that if x(0) < r, then x(t) <
t 0. Otherwise the equilibrium point is unstable.

Denition A.3 (Asymptotic stability). An equilibrium point x = 0 is

asymptotically stable if it is stable, and if in addition there exists some r > 0
such that x(0) < r implies that x(t) 0 as t .

Denition A.4 (Marginal stability). An equilibrium point that is Lya-

punov stable but not asymptotically stable is called marginally stable.

Denition A.5 (Exponential stability). An equilibrium point is exponen-

tially stable if there exist two strictly positive numbers and independent
of time and initial conditions such that

x(t) x(0) exp(t), t > 0 (A.3)

in some ball around the origin.

The above denitions correspond to local properties of the system around

the equilibrium point. The above stability concepts become global when their
corresponding conditions are satised for any initial state.

Lyapunov Linearization Method

Assume that f (x) in (A.2) is continuously dierentiable and that x = 0 is an

equilibrium point. Then, using Taylor expansion, the system dynamics can be
written as 
f 
x(t) = x(t) + o(x) (A.4)
x x=0
where o stands for higher-order terms in x. Linearization of the original non-
linear system at the equilibrium point is given by

x(t) = Ax(t) (A.5)

where A denotes the Jacobian matrix of f with respect to x at x = 0, i.e.,


f 
A=
x x=0

A linear time-invariant system of the form (A.5) is (asymptotically) stable

if A is a (strictly) stable matrix, i.e., if all the eigenvalues of A have (negative)
nonpositive real parts. The stability of linear time-invariant systems can be
determined according to the following theorem.
 A.1 Lyapunov Stability 509

Theorem A.6. The equilibrium state x = 0 of the the system (A.5) is asymp-
totically stable if and only if, given any matrix Q > 0, the solution P to the
Lyapunov equation

AT P + P A = Q (A.6)
is positive denite. If Q is only positive semi-denite (Q 0), then only
stability is concluded.

The following theorem somewhat claries some points:

Theorem A.7. [500] Given a matrix A IRnn , the following statements

are equivalent:
A is a Hurwitz matrix.
There exists some positive denite matris Q IRnn such that AT P +
P A = Q has a corresponding unique solution for P , and this P is positive
denite.
For every positive denite matrix Q IRnn , AT P + P A = Q has a
unique solution for P , and this solution is positive denite.

The term corresponding unique solution means the matrix

P = exp(AT t)Q exp(At)dt
0
Local stability of the original nonlinear system can be inferred from sta-
bility of the linearized system as stated in the following theorem.

Theorem A.8. If the linearized system is strictly stable (unstable), then the
equilibrium point of the nonlinear system is locally asymptotically stable (un-
stable).

The above theorem does not allow us to conclude anything when the lin-
earized system is marginally stable. Then one has to rely on more sophisticated
tools like the invariant manifold theory [256].

Lyapunovs Direct Method

Let us consider the following denitions.

Denition A.9 ((Semi-)deniteness). A scalar continuous function V :

IR+ IRn is said to be locally positive (semi-)denite if V (0) = 0 and V (x) >
0 (V (x) 0) for x = 0. Similarly, V () is said to be negative (semi-)denite
if V () is positive (semi-)denite.

Another denition of positive deniteness can be given:

510 A Background Material

Denition A.10. A function V : IR+ IRn is said to be locally positive

denite if it is continuous, V (0) = 0, and there exists a constant r > 0 and a
function () of class K such that

(||x||) V (x) (A.7)

for all ||x|| r.

It happens that both characterizations are equivalent [500, Lemma 5.2.6].

In fact if V (0) = 0 and V (x) > 0 when x = 0, one can always nd a class-K
function which locally lowerbounds V () in a neighborhood of x = 0.

Denition A.11 (Lyapunov function). V (x) is called a Lyapunov func-

tion for the system (A.2) if, in a ball B containing the origin, V (x) is positive
denite and has continuous partial derivatives, and if its time derivative along
the solutions of (A.2) is negative semi-denite, i.e., V (x) = (V /x)f (x) 0.

The following Theorems can be used for local and global analysis of sta-
bility, respectively. Assume that f (0) = 0 and that x = 0 is an isolated xed
point of (A.2).

Theorem A.12. [Local stability] The equilibrium point x = 0 of the sys-

tem (A.2) is (asymptotically) stable in a ball B if there exists a scalar function
V (x) with continuous derivatives such that V (x) is positive denite and V (x)
is negative semi-denite (negative denite) in the ball B.

Theorem A.13. [Global stability] The equilibrium point of system (A.2) is

globally asymptotically stable if there exists a scalar function V (x) with con-
tinuous rst order derivatives such that V (x) is positive denite, V (x) is neg-
ative denite and V (x) is radially unbounded, i.e., V (x) as x .

Clearly the global asymptotic stability implies that 0 is the unique xed
point of (A.2) in the whole state space IRn .

Krasovskii-LaSalles Invariant Set Theorem

Krasovskii-LaSalles results extend the stability analysis of the previous The-

orems when V () is only negative semi-denite. They are stated as follows.

Denition A.14 (Invariant set). A set S is an invariant set for a dynamic

system if every trajectory starting in S remains in S.

Invariant sets include equilibrium points, limit cycles, as well as any tra-
jectory of an autonomous system.
 A.1 Lyapunov Stability 511

Theorem A.15. [Krasovskii-LaSalle] Consider the system (A.2) with f ()

continuous, and let V (x) be a scalar function with continuous rst partial
derivatives. Consider a region dened by V (x) < for some > 0. As-
sume that the region is bounded and V (x) 0 x . Let be the set
of all points in where V (x) = 0, and M be the largest invariant set in .
Then, every solution x(t) originating in tends to M as t . On the
other hand, if V (x) 0 x and V (x) as x , then all solutions
globally asymptotically converge to M as t .

Some crucial properties for the invariance principle to hold, are that state
trajectories are continuous with respect to initial data, and that the limit
sets are compact invariant sets. Not all the systems examined in this book
possess those properties (for instance the nonsmooth Lagrangian systems of
Section 6.8.2 do not necessarily enjoy the continuity-in-the-initial-data prop-
erty). Another formulation of this result is as follows [351].

Theorem A.16. Under the same assumptions of Theorem A.15, let K be the
set of points not containing whole trajectories of the system for t . Then
if V (x) 0 outside of K and V (x) = 0 inside K, the system is asymptotically
stable.

Notice in particular that {x = 0}  K. K can be a surface, a line, etc. In

Theorem A.6, notice that if Q = C T C with (A, C) being an observable pair,
then asymptotic stability is obtained again. More formally:

Corollary A.17. If C IRmn and the pair (A, C) is observable, then the
matrix A is asymptotically stable if and only if there exists a matrix P =
P T > 0 that is the unique solution of AT P + P A + C T C = 0.

The proof of this corollary is based on the quadratic function V (x) =

xT P x, whose derivative is computed along the solutions of x(t) = Ax(t). Then
use the Krasovskii-LaSalle Theorem to conclude on the asymptotic stability,
using that the Kalman observability matrix is full-rank.

A.1.2 Non-autonomous Systems

In this section we consider non-autonomous nonlinear systems represented

by (A.1). The stability concepts are characterized by the following denitions.

Denition A.18 (Equilibrium). A state x
is an equilibrium point of (A.1)

if f (x
, t) = 0 t t0 .

Denition A.19 (Stability). The equilibrium point x = 0 is stable at t = t0

if for any > 0 there exists an r(, t0 ) > 0 such that x(t0 ) < r x(t) <
, t t0 . Otherwise the equilibrium point x = 0 is unstable.
512 A Background Material

Denition A.20 (Asymptotic stability). The equilibrium point x = 0 is

asymptotically stable at t = t0 if it is stable and if it exists r(t0 ) > 0 such
that x(t0 ) < r(t0 ) x(t) 0 as t .
Denition A.21 (Exponential stability). The equilibrium point x = 0 is
exponentially stable if there exist two positive numbers and such that
x(t) x(t0 ) exp((t t0 )) t t0 , for x(t0 ) suciently small.
Denition A.22 (Global asymptotic stability). The equilibrium point
x = 0 is globally asymptotically stable if it is stable and x(t) 0 as t
x(t0 ) IRn .
The stability properties are called uniform when they hold independently
of the initial time t0 as in the following denitions.
Denition A.23 (Uniform stability). The equilibrium point x = 0 is uni-
formly stable if it is stable with r = r() that can be chosen independently of
t0 .
Denition A.24 (Uniform asymptotic stability). The equilibrium point
x = 0 is uniformly asymptotically stable if it is uniformly stable and there
exists a ball of attraction B, independent of t0 , such that x(t0 ) B x(t) 0
as t .

Lyapunovs Linearization Method

Using Taylor expansion, the system (A.1) can be rewritten as

x(t) = A(t)x(t) + o(x, t) (A.8)
where 
f 
A(t) = (t)
x x=0
A linear approximation of (A.1) is given by
x(t) = A(t)x(t) (A.9)
The result of Theorem A.6 can be extended to linear time-varying systems
of the form (A.9) as follows.
Theorem A.25. A necessary and sucient condition for the uniform asymp-
totic stability of the origin of the system (A.9) is that a matrix P (t) exists such
that
V (t, x) = xT P (t)x > 0
and
V (t, x(t)) = xT (t)(AT P (t) + P (t)A + P (t))x(t) k(t)V (t, x(t))
t
where limt t0 k( )d = uniformly with respect to t0 .
 A.1 Lyapunov Stability 513

We can now state the following result.

Theorem A.26. If the linearized system (A.9) is uniformly asymptotically

stable, then the equilibrium point x
= 0 of the original system (A.1) is also
uniformly asymptotically stable.

Lyapunovs Direct Method

We present now the Lyapunov stability theorems for non-autonomous systems.

The following Denitions are required.

Denition A.27. [Function of class K] A continuous function : [0, k)

IR+ is said to be of class K if
(i) (0) = 0
(ii)() > 0 > 0
(iii)() is nondecreasing
Statements (ii) and (iii) can also be replaced with
(ii) is strictly increasing
so that the inverse function 1 () is dened. The function is said to be of
class K if k = and () as .

Denition A.28. A class KLfunction is a function : IR+ IR+ IR+

such that (, t) is of class K for each t and limt+,t0 (r, t) = 0.

Based on the denition of function of class K, a modied denition of

exponential stability can be given.

Denition A.29 (K-exponential stability). The equilibrium point x = 0 is

K-exponentially stable if there exist a function () of class K and a positive
number such that x(t) (x(t0 )) exp((t t0 )) t t0 , for x(t0 )
suciently small.

Denition A.30 (Positive denite function). A function V (x, t) is said

to be locally (globally) positive denite if and only if there exists a function
() of class K such that V (0, t) = 0 and V (x, t) (x) t 0 and x in
a ball B.

Denition A.31 (Decrescent function). A function V (x, t) is locally (glob-

ally) decrescent if and only if there exists a function () of class K such that
V (0, t) = 0 and V (x, t) (x), t > 0 and x in a ball B.

The main Lyapunov stability theorem can now be stated as follows.

514 A Background Material

Theorem A.32. Assume that V (x, t) has continuous rst derivatives around
the equilibrium point x
= 0. Consider the following conditions on V () and
V () where (), () and () denote functions of class K,and let Br be the
closed ball with radius r > 0 and center x
= 0:

(i) V (x, t) (x) > 0, x Br , t t0

(ii) V (x, t) 0

(iii) V (x, t) (x), x Br , t t0 (A.10)

(iv) V (x, t) (x) < 0, x Br , t t0

(v) lim (x) = .

x

Then the equilibrium point x
= 0 is:

Stable if conditions (i) and (ii) hold
Uniformly stable if conditions (i)(iii) hold
Uniformly asymptotically stable if conditions (i)(iv) hold
Globally uniformly asymptotically stable if conditions (i)(iv) hold globally,
i.e. Br = IRn and (v) holds

Barbalats Lemma

Krasovskii-LaSalles results are only applicable to autonomous systems. On

the other hand, Barbalats Lemma can be used to obtain stability results when
the Lyapunov function derivative is negative semi-denite.

Lemma A.33. [Barbalat] If the dierentiable function f () has a nite limit

as t , and if f() is uniformly continuous, then f(t) 0 as t .

This lemma can be applied for studying stability of non-autonomous sys-

tems with Lyapunov Theorem, as stated by the following result.

Lemma A.34. If a scalar function V (x, t) is lower bounded and V (x, t) is

negative semi-denite, then V (x, t) 0 as t if V (x, t) is uniformly
continuous in time.

Matrosovs Theorem

Theorem A.35 (Matrosovs Theorem). Let IRn be an open con-

nected domain containing the origin x = 0. If there exists two continuously
dierentiable functions V : [t0 , +) IR and W : [t0 , +) IR, a
continuous function V
: IR, three functions (), (), () of class K,
such that for every (x, t) [t0 , +) one has
 A.2 Dierential Geometry Theory 515

(||x||) V (t, x) (||x||)

V (t, x) V
(x) 0
|W (t, x)| is bounded
max(d(x, E)), |W (t, x)|) (||x||), where E = {x | V
(x) = 0}
||f (t, x)|| is bounded
Choosing a > 0 such that the closed ball Ba , dene for all t
1
[t0 , +): Vt,a = {x | V (t, x) (a)}. Then

For all x0 Vt1

0 ,a
, x(t) tends to zero asymptotically uniformly in t0 , x0
The origin is uniformly asymptotically stable in the sense of Lyapunov

The following may help in checking the Theorems conditions.

Lemma A.36. [389] The fourth condition in Matrosovs Theorem is satised

if:
W (x, t) is continuous in both arguments and depends on time in the follow-
ing way: W (x, t) = g(x, (t)) where g() is continuous in both arguments,
() is continuous and its image lies in a bounded set K.
There exists a class K function () such that |W (x, t)| (||x||) for all
x E and alll t t0 .

A.2 Dierential Geometry Theory

Consider a nonlinear ane single-input/single-output system of the form

x(t) = f (x(t)) + g(x(t))u(t)
(A.11)

y(t) = h(x(t))

where h : IRn IR and f, g : IRn IRn are smooth functions. For ease of
presentation we assume that the system (A.11) has an equilibrium at x = 0.

Denition A.37 (Lie derivative). The Lie derivative of h with respect to

f is the scalar
h
Lf h = f
x
and the higher derivatives satisfy the recursion

Lif h = Lf (Li1
f h)

with L0f h = h.
516 A Background Material

Denition A.38 (Lie bracket). The Lie bracket of f and g is the vector

g f
[f, g] = f g,
x x
and the recursive operation is established by

adif g = [f, adi1

f g]

Some properties of Lie brackets are:

[1 f1 + 2 f2 , g] = 1 [f1 , g] + 2 [f2 , g]

[f, g] = [g, f ]
and the Jacobi identity

Ladg h = Lf (Lg h) Lg (Lf h)

To dene nonlinear changes of coordinates we need the following concept.

Denition A.39 (Dieomorphism). A function (x) : IRn IRn is said

to be a dieomorphism in a region IRn if it is smooth, and 1 (x) exists
and is also smooth.

A sucient condition for a smooth function (x) to be a dieomorphism

in a neighbourhood of the origin is that the Jacobian /x be nonsingular
at zero. The conditions for feedback linearizability of a nonlinear system are
strongly related with the following theorem.

Theorem A.40. [Frobenius] Consider a set of linearly independent vectors

{f1 (x), . . . , fm (x)} with fi (x) : IRn IRn . Then, the following statements
are equivalent:
(i) (Complete integrability) there exist nm scalar functions hi (x) : IRn IR
such that
Lfj hi = 0 1i j nm
where hi /x are linearly independent
(ii)(Involutivity) there exist scalar functions ijk (x) : IRn IR such that


m
[fi , fj ] = ijk (x)fk (x)
k=1
 A.2 Dierential Geometry Theory 517

A.2.1 Normal Form

In this section we present the normal form of a nonlinear system which has
been instrumental for the development of the feedback linearizing technique.
For this, it is convenient to dene the notion of relative degree of a nonlinear
system.

Denition A.41 (Relative degree). The single input-single output sys-

tem (A.11) has relative degree r at x = 0 if
(i) Lg Lkf h(x) = 0, for all x in a neighbourhood of the origin and for all k <
r1
(ii)Lg Lr1
f h(x) = 0

It is worth noticing that in the case of linear systems, e.g., f (x) = Ax,
g(x) = Bx, h(x) = Cx, the integer r is characterized by the conditions
CAk B = 0 k < r 1 and CAr1 B = 0. It is well known that these are
exactly the conditions that dene the relative degree of a linear system. An-
other interesting interpretation of the relative degree is that r is exactly the
number of times we have to dierentiate the output to obtain the input ex-
plicitly appearing. Let us now asumme that u and y both have dimension m
in (A.11).

Denition A.42 (Vector relative degree). The multi input-multi output

system (A.11) has vector relative degree [r1 , r2 , ..., rm ]T IRm at x = 0 if
(i) Lgj Lkf h(x) = 0, x in a neighborhood of the origin and k < rj 1

Lg1 Lrf1 1 h1 . . Lgm Lrf1 1 h1
. .. .
(ii)The matrix
is nonsingular in a neigh-

. .. .
Lg1 Lrfm 1 hm . . Lgm Lrfm 1 hm
borhood of the origin

Denition A.43 (Uniform vector relative degree). Let u(t) IRm and
y(t) IRm in (A.11). The system is said to have a uniform relative degree r
if ri = r for all 1 i m in the previous denition.

We note that this denition is dierent from the denition of the uniform
relative degree in [227, p.427] where uniformity refers to the fact that the
system (single input-single output) has a (scalar) relative degree r at each
x(t) IRn . Here we rather employ uniformity in the sense that the vector
relative degree has equal elements. In the linear invariant multivariable case,
such a property has favourable consequences as recalled a few lines below.
The functions Lif h for i = 0, 1, . . . , r 1 have a special signicance as
demonstrated in the following theorem.
518 A Background Material

Theorem A.44. [Normal form] If the single input-single output system (A.11)
has relative degree r n, then it is possible to nd nr functions r+1 (x), . . . , n (x)
so that

h(x)


Lf h(x)

..
.
(x) = Lr1 h(x)
(A.12)
f
r+1 (x)

..
.
n (x)
is a dieomorphism z = (x) that transforms the system into the following
normal form


z1 = z2



z2 = z3

.

..



zr1 = zr
(A.13)

zr = b(z) + a(z)u



zr+1 = qr+1 (z)



..

.

zn = qn (z)
Moreover, a(z) = 0 in a neighborhood of z0 = (0).

A similar canonical form can be derived for the multivariable case, however
more it is more involved [227]. In the case of a linear time invariant system
(A, B, C), a similar canonical state space realization has been shown to exist
in [432], provided CAi B = 0 for all i = 0, 1, ..., r 2, and the matrix CAr1 B
is nonsingular. This Sannutis canonical form is quite interesting as the zero
dynamics taks the form (t) = A0 (t) + B0 z1 (t): it involves only the output
z1 of the system. The conditions on the Markov parameters are sucient
conditions for the invertibility of the system [429]. Other such canonical state
space representations have been derived by Sannuti and co-workers [424, 433,
434], which are usually not mentioned in textbooks.

A.2.2 Feedback Linearization

From the above theorem we see that the state feedback control law
1
u= (b(z) + v) (A.14)
a(z)
 A.2 Dierential Geometry Theory 519

yields a closed-loop system consisting of a chain of r integrators and an (nr)-

dimensional autonomous system. In the particular case of r = n we fully
linearize the system. The rst set of conditions for the triple {f (x), g(x), h(x)}
to have relative degree n is given by the partial dierential equation
h  
g(x), adf g(x), . . . , adn2
f g(x) =0
x
The Frobenius Theorem shows that the existence of solutions to this equation
is equivalent to the involutivity of {g(x), adf g(x), . . . , adn2
f g(x)}. It can be
shown that the second condition, i.e. Lg Lf h(x) = 0 is ensured by the linear
n1

independence of {g(x), adf g(x), . . . , adn1

f g(x)}.
The preceding discussion is summarized by the following key Theorem.

Theorem A.45. For the system (A.11) there exists an output function h(x)
such that the triple {f (x), g(x), h(x)} has relative degree n at x = 0 if and
only if:
(i) The matrix {g(0), adf g(0), . . . , adn1
f g(0)} is full rank
(ii)The set {g(x), adf g(x), . . . , adf g(x)} is involutive around the origin
n2

The importance of the preceding theorem can hardly be overestimated.

It gives (a priori veriable) necessary and sucient conditions for full lin-
earization of a nonlinear ane system. However, it should be pointed out
that this control design approach requires on one hand the solution of a set of
partial dierential equations. On the other hand, it is intrinsically nonrobust
since it relies on exact cancellation of nonlinearities. In the linear case this is
tantamount to pole-zero cancellation.

A.2.3 Stabilization of Feedback Linearizable Systems

If the relative degree of the system r < n then, under the action of the feedback
linearizing controller (A.14), there remains an (n r)-dimensional subsystem.
The importance of this subsystem is underscored in the proposition below.

Theorem A.46. Consider the system (A.11) assumed to have relative de-
gree r. Further, assume that the trivial equilibrium of the following (n r)-
dimensional dynamical system is locally asymptotically stable:


zr+1 = qr+1 (0, . . . , 0, zr+1 , . . . , zn )
..
.

zn = qn (0, . . . , 0, zr+1, . . . , zn )
where qr+1 , . . . , qn are given by the normal form. Under these conditions, the
control law (A.14) yields a locally asymptotically stable closed-loop system.
520 A Background Material

The (n r)-dimensional system (A.15) is known as the zero dynamics. It

represents the dynamics of the unobservable part of the system when the input
is set equal to zero and the output is constrained to be identically zero. It is
worth highlighting the qualier local in the above theorem; in other words,
it can be shown that the conditions above are not enough to ensure global
asymptotic stability.

A.2.4 Further Reading

The original Lyapunov Theorem is contained in [322], while stability of nonlin-

ear dynamic systems is widely covered in [282,292]. The proofs of the theorems
concerning Lyapunov stability theorem can be found in [199,256,500]. An ex-
tensive presentation of dierential geometry methods can be found in [227]
and the references therein. For the extension to the multivariable case and
further details we refer the reader again to [227, 381].

A.3 Viscosity Solutions

This section intends to briey describe what viscosity solutions of rst order
nonlinear partial dierential equations of the form

F (x, V (x), V (x)) = 0 (A.15)

are, where x IR , V : IR IR, is the dierential operator (the Euclidean
n n

gradient), and F : IRn IR IRn IR is continuous. A function V () is

dierentiable at x and with derivative if

V (z) V (x) T (z x)
lim =0 (A.16)
zx |z x|
and this equality can equivalently be stated with the two inequalities supposed
to hold simulatenaously

V (z) V (x) T (z x)
lim sup 0 (A.17)
zx |z x|
(in other words satises (A.17) if and only if the plane z
V (z) + T (z x)
is tangent from above to the graph of V () at x), and

V (z) V (x) T (z x)
lim inf 0, (A.18)
zx |z x|
(in other words satises (A.18) if and only if the plane z
V (z) + T (z x)
is tangent from below to the graph of V () at x). The superdierential of V ()
at x is then dened as the set
 A.3 Viscosity Solutions 521

D+ V (x) = { IRn | (A.17)holds}

and the subdierential of V () at x is then dened as the set

D V (x) = { IRn | (A.18)holds}

It is noteworthy that such sets may be empty, see the examples below.
Sometimes these sets are named one-sided dierentials. The function V () is
said to be a viscosity subsolution of the partial dierential equation (A.15) if
for each x IRn one has

F (x, V (x), ) 0
+
for all D V (x). The function V () is said to be a viscosity supersolution
of the partial dierential equation (A.15) if for aeach x IRn one has

F (x, V (x), ) 0
for all D V (x). The function V () is said to be a viscosity solution of
the partial dierential equation (A.15) if it is both a viscosity subsolution and
a viscosity supersolution of this partial dierential equation. As we already
pointed out in section 4.4.5, in case of proper 1 convex functions the viscosity
subdierential (or subgradient) and the convex analysis subgradient, are the
same [415, Proposition 8.12]. We now consider two illustrating examples taken
from [64].

Example A.47. Consider the function

0 if x < 0



V (x) = x if x [0, 1] (A.19)





1 if x > 1

Then D+ V (0) = , D V (0) = [0, +), D+ V (x) = D V (x) = { 12 x} if
x (0, 1), D+ V (1) = [0, 12 ], D V (1) = .
+
Example A.48. Consider F (x, V (x), V (x)) = 1 | V x |. Then V : IR IR
x
|x| is a viscosity solution of 1 | x | = 0. Indeed V () is dierentiable at
V

all x = 0 and one has D+ V (0) = , and D V (0) = [1, 1]. V () is indeed a
supersolution since 1 || 0 for all D u(0) = [1, 1].

Example A.49. The same function V (x) = |x| is not a viscosity solution of
x | = 0. At x = 0 and choosing = 0 one obtains 1 + |0| = 1 < 0
1 + | V
so the function is not a supersolution, though it is a viscosity subsolution.

1
Proper in this context means that V (x) < + for at least one x IRn , and
V (x) > for all x IRn .
522 A Background Material

It is a fact that if V () is convex and not dierentiable at x then D+ V (x) =

. The following Lemma says a bit more.

Lemma A.50. Let V () be continuous on some interval I  x. Then:

If V () is dierentiable at x: D+ V (x) = D V (x) = {V (x)}
If the sets D+ V (x) and D V (x) are both nonempty, then V () is dier-
entiable at x and the rst item holds
The sets of points where a one-sided dierential exists:
) *
I + = x I | D+ V (x) =

and ) *
I = x I | D V (x) =
are both nonempty. Both I + and I are dense in I.

The second item says that if a function is not dierentiable at x then

necessarily one of the two sets must be empty. This conrms the above ex-
amples. The third item says that the points x where the continuous function
V () admits a superdierential and a subdierential, exist in I and even are
numerous in I: they form dense subsets of I (take any point y I and any
neighborhood of y: there is an x in such a neighborhood at which V () has a
one-sided dierential). There is another way to dene a viscosity solution.

Lemma A.51. Let V () be continuous on some interval I. Then

D+ V (x) if and only if there exists a function C 1 (I) such that
(x) = and V has a local maximum at x.
D V (x) if and only if there exists a function C 1 (I) such that
(x) = and V has a local minimum at x.

From the rst item it becomes clear why a convex function that is not
dierentiable at x has D+ V (x) = . Then a continuous function V () is a
viscosity subsolution of F (x, V (x), V (x)) = 0 if for every C 1 function ()
such that V has a local maximum at x one has F (x, V (x), (x)) 0. It
is a viscosity supersolution of F (x, V (x), V (x)) = 0 if for every C 1 function
() such that V has a local minimum at x one has F (x, V (x), (x)) 0.
The following result is interesting:
 A.4 Algebraic Riccati Equations 523

Proposition A.52. [418] Given a system x(t) = f (x(t), u(t)) whose

solution on [t0 , t1 ] is an absolutely continuous function such that
x(t) = f (x(t), u(t)) for almost all t [t0 , t1 ], a supply rate w(x, u)
such that w(0, u) 0, and a continuous function V : IRn IR such
that V (0) = 0, then:

 t1
V (x(t1 )) V (x(t0 ) t0
w(x(t), u(t))dt

holds for every solution [t0 , t1 ] IRn

T f (x, u) w(x, u) for every x IRn , u U, and D V (x)

(A.20)

In other words one may write the innitesimal version of the dissipation
inequality when the storage function is not dierentiable, by replacing its
gradient by a viscosity subgradient.

A.4 Algebraic Riccati Equations

The topic of studying and solving Riccati equations is a wide topic and we
do not pretend to cover it in this small appendix. The results we present
only aim at showing that under some conditions which are dierent from the
conditions stated in the foregoing chapters, existence of solutions to algebraic
Riccati equations can be guaranteed. Let us consider the following algebraic
Riccati equation:

P DP + P A + AT P C = 0, (A.21)
where A IR nn
, C IR nn
and D IR nn
. P is the unknown ma-
trix. Before going on we need a number of denitions. A subspace IR2n
is called N neutral if xT N y = 0 for all x, y ( may be Ker(N ),
or Ker(N T )). The neutrality index (M, N ) of a pair of matrices (M, N )
is the maximal dimension of a real M invariant N neutral subspace in
IR2n . A pair of matrices (A, D) is sign controllable if every 0 IR at least
one of the subspaces Ker(0 In A)n and Ker(0 In A)n is contained in
Range[D, AD, ..., An1 D], and for every + j C, IR and IR,
= 0, at least one of the two subspaces Ker[(2 + 2 )In 2A + A2 ]n is
contained in Range[D, AD, ..., An1 D]. Another way to characterize the sign-
controllability of the pair (A, D) is: for any C, at least one of the two
524 A Background Material

matrices (In A D) and (In A D) is full-rank [273]. Sign-controllability

of (A, D) implies that there exists a matrix K such that F = A + DK is un-
mixed,i.e. (F ) (F T ) = .

We now dene the two matrices in IR2n2n

A D
M =
C AT

0 In
H=
In 0
Theorem A.53. [416] Let D 0 and (A, D) be sign controllable. Suppose
that the matrix M is invertible. Then the following statements are equivalent:
The ARE (A.21) has a real solution.
The ARE (A.21) has a real solution P for which rank(P P T ) 2(n
(M, H)).
The matrix M has a real ndimensional invariant subspace.
Either n is even, or n is odd and M has a real eigenvalue.
If (M, N ) = n there exists a real symmetric solution.
We recall that an M invariant subspace is a subspace such that for all
v , M v . Comparing (A.21) to (3.140) one sees that A in (A.21) is
replaced by A+ BR1 C in (3.140), whereas BR1 B T in (3.140) plays the role
of D in (A.21), and C T R1 C + Q in (3.140) plays the role of C in (A.21).
Theorems 3.42, 3.43 and 3.44 state stronger results than Theorem A.53 since
the negative deniteness (resp. positivite deniteness) of the solution is crucial
in the framework of optimal control (resp. dissipative systems). On the other
hand the conditions of Theorems 3.42, 3.43 and 3.44 look much simpler than
those of Theorem A.53.
Let us now consider the following Riccati inequality:

P A + AT P P BB T P + Q > 0 (A.22)
Lemma A.54. [259] Suppose that the pair (A, B) is stabilizable. The follow-
ing three statements are equivalent:
There exists a symmetric matrix P solving (A.22).
There exists a symmetric matrix P such that
P A + AT P P BB T P + Q = 0, (A BB T P ) C .

A BB T
The Hamiltonian matrix H = has no eigenvalues on the
Q AT
imaginary axis.
 A.4 Algebraic Riccati Equations 525

Suppose that one of these conditions hold. Then any solution P of (A.22)
satises P < P .

The notation (A) C means that all the eigenvalues of A have negative
real parts. In case the pair (A, B) is not stabilizable, things are more complex
and one has rst to perform a decomposition of A and B before proposing a
test; see [443].
Further study on Riccati equations, their solvability and their link with
the KYP Lemma set of equations solvability, may be found in [37, 38]. A
special type of Riccati equations that correspond to the KYP Lemma set of
equations for descriptor systems may be found in [501]. See also [290] for upper
bounds estimation of solutions to AREs. The problem of the existence of a real
symmetric negative semi-denite solution to AREs is a tricky problem [490].

A.4.1 Reduced Riccati Equation for WSPR Systems

The following is taken from Hodaka et al. [213]. As we have explained the
KYP Lemma set of equations form a LMI which is in turn equivalent to a
Riccati equation. When the transfer function H(s) Cmm is SSPR, then
D + DT > 0 and this Riccati equation has a positive denite symmetric
solution P . The point now is: what happens when H(s) is not SSPR and
when D = 0? The algorithm that is proposed next, allows one to character-
ize WSPRness, SPRness and PRness in terms of Riccati equations and the
inherent integration of the system. The developments are rather lengthy and
need some preliminary results.

Denition A.55. [429] The transfer function H(s) = C(sIn A)1 B +

D Cmm is invertible if there exists a proper transfer function H(s) and a
nonnegative integer l such that
1
H(s)H(s) = Im (A.23)
sl
The least integer l satisfying (A.23) is called the inherent integration of H(s).

A m m transfer function is invertible if and only if it has rank m over

the eld of proper transfer functions.

Lemma A.56. [429] The inherent integration of H(s) is k if and only if for
l = 0, 1, 2, ..., k 1,

rank(Ql [A, B, L, W ]) rank(Ql1 [A, B, L, W ]) < m (A.24)

and rank(Qk [A, B, L, W ])rank(Qk1 [A, B, L, W ]) = m, where Q0 (A, B, L, W )

= W , for l = 1, 2, ...
526 A Background Material

W 0 0 .... 0


LB W 0 ... 0



Ql (A, B, L, W ) =

LAB LB W ... 0


...


LAl1 B LAl2 B LAl3 B ... W

and rank(Q1 [A, B, L, W ]) = 0.

We now proceed with the main developments. Let us assume that H(s)

is PR and that rank(H(s) + H T (s)) = m. Let H0 (s) = H(s), and 0 =
(A0 , B0 , C0 , D0 ) is a minimal realization of H0 with A0 IRn0 n0 , B0
IRn0 m , C0 IRmn0 , D0 IRmm , n0 1. The rest of the sequence {i }i0
is constructed as follows: i = (Ai , Bi , Ci , Di ), Ai IRni ni , Bi IRni m ,
Ci IRmni , Di IRmm , Hi (s) = Ci (sIn Ai )1 Bi + Di , and we dene
i (s) = Hi (s) + HiT (s). Let us suppose that Hi (s) is PR. The next three
conditions are denoted as i :
(a) (Ai , Bi , Ci , Di ) is minimal.
(b) The KYP Lemma set of equations in (3.2) is satised with (Ai , Bi , Ci ,Di ),
Pi = PiT > 0 and Li , Wi , and we denote Ri = WiT Wi = Di + DiT .
(c) rank(i (s)) = m.
Taking Gi (s) = Wi + LTi (sIni Ai )1 Bi gives i (s) = GTi (s)Gi (s), so
that (c) can be replaced by rank(Gi (s)) = m or

Ai sIni Bi
det  0 (A.25)
T
Li Wi
Let ri =rank(Ri ). If ri = m then Ri > 0 and the algorithm is terminated.
In such a case the transfer function H(s) is SSPR. If ri < m, the algorithm
proceeds as follows. Since Ri = RiT 0, there exists an orthogonal matrix
Si IRmm such that

Ri 0
SiT Ri Si = (A.26)
0 0
where Ri IRri ri is positive denite. Partition Si such that Si = [Si1 Si2 ]
with Si1 IRmri , Si2 IRm(mri ) . Using the nonsingular matrix Si , let us
introduce the matrices

[Bi1 Bi2 ] = [Bi Si1 Bi Si2 ] (A.27)

 A.4 Algebraic Riccati Equations 527

[Wi1 Wi2 ] = [Wi Si1 Wi Si2 ] (A.28)

T

Ci1 Si1 Ci
= (A.29)
T
Ci2 Si2 Ci
T T

Di1 Di2 Si1 Di Si1 Si1 Di Si2
= (A.30)
T T
Di3 Di4 Si2 Di Si1 Si2 Di Si2

Lemma A.57. [213] If the property i is satised, then:

T
W12 = 0 and Pi Bi2 = Ci2 .
Bi2 has full column rank, Ci2 has full row rank and ni m ri .

Proof: Let us pre and post-multiply the KYP Lemma set of equations of prop-
erty i (b) above, by diag(Ini , SiT ) and diag(Ini , Si ), respectively. The rst
condition follows. Now postmultiplying the matrix in (A.25) by diag(Ini , Si )
gives

Ai sIni Bi1 Bi2
det  0 (A.31)
LTi Wi1 0
This secures that Bi2 has full column rank and consequently ni m ri .
T
Finally it follows from Pi Bi2 = Ci2 and P i > 0 that Ci2 has full row rank.
T
Let Ei = Ci2 Bi2 = Bi2 Pi Bi2 = EiT > 0 by Lemma A.57. A square root
1
i = Ei2 = Ti > 0 exists. Moreover there exist matrices Ni IRni+1 ni and
Mi IRni ni+1 such that

Ni Bi2 = 0, Ci2 Mi = 0, Ni Mi = Ini+1 , (A.32)

where ni+1 = ni m + ri . Let us dene a nonsingular matrix Ti IRni ni as

Ni
Ti = , Ti1 = [Mi Bi2 1
i ] (A.33)
1
i Ci2
Using the full-rank matrices Ti and Si we can consider transformations of
input, output and state variables as ui = Si ui , yi Si yi , xi = Ti xi , where
T
Si1 ui ui1
ui = =

(A.34)
T
Si2 ui ui2
528 A Background Material
T

Si1 yi yi1
yi = =


(A.35)
T
Si2 yiyi2

Ni xi xi+1
xi = =

(A.36)
1
i Ci2 xi xi
Let us now introduce matrices with subscripts (i + 1) dened by

Ai+1 = Ni Ai Mi , Bi+1 = [Ni Ai Bi2 1

i Ni Bi1 ]

1
i Ci2 Ai Mi
Ci+1 =
Ci1 Mi

1 1
i Ci2 Ai Bi2 i 1
i Ci2 Bi1 (A.37)
Di+1 =
Ci1 Bi2 1
i Di1

Pi+1 = MiT Pi Mi , Li+1 = MiT Li

Wi+1 = [LTi Bi2 1

i
T
Wi1 ], Ri+1 = Di+1 + Di+1
and the i + 1-th system is dened as i+1 = (Ai+1 , Bi+1 , Ci+1 , Di+1 ) with
ui+1 = (xTi uTi1 ).
Lemma A.58. [213] If the property i holds for the system i , then i+1
holds for the system i+1 .
Proof: Dene the matrices Ai = Ti Ai Ti1 , Bi = Ti Bi Si , Ci = SiT Ci TiT ,
Di = SiT Di Si , Ri = SiT RiSi , Pi = TiT Pi Ti1 , LTi = LTi Ti1 , Wi = Wi Si .
(a): the controllability of (Ai , Bi ) is equivalent to that of (Ai , Bi ). Since i
is full-rank it follows that (Ai+1 , Bi+1 ) is controllable. In a similar way one
can show that (Ai+1 , Ci+1 ) is observable. (b): since Mi has full column rank
and Pi = PiT > 0, it follows that Pi+1 = Pi+1 T
> 0. Pre and postmultiplying
the KYP Lemma set of equations for i by diag(TiT SiT ) and diag(Ti1 Si )
respectively, we obtain the KYP Lemma set of equations for i+1 . (c): pre
and postmultiplying the matrix in (A.25) by diag(Ti Im ) and diag(Ti1 Si )
respectively, and since i is full-rank, it follows that rank(i+1 (s)) = m.
One sees that the algorithm preserves the PRness, in the sense that if i
is PR then i+1 is PR. However there is no guarantee yet that the algorithm
terminates.
Lemma A.59. [213] Assume that the property i holds for the transfer func-
tion Hi (s). Consider the transfer functions Gi (s) = Wi + LTi (sIni Ai )1 Bi
and Gi+1 (s) as dened in the algorithm. Let ki and ki+1 be the inherent inte-
grations of Gi (s) and Gi+1 (s) respectively. Then ki+1 = ki 1.
 A.4 Algebraic Riccati Equations 529

Proof: the equations

rank(Ql [Ai , Bi , Li , Wi ]) = ri + rank(Ql1 [Ai+1 , Bi+1 , Li+1 , Wi+1 ]) (A.38)

for l = 0, 1, 2, ... secure that

rank(Ql1 [Ai+1 , Bi+1 , Li+1 , Wi+1 ]) rank(Ql2 [Ai+1 , Bi+1 , Li+1 , Wi+1 ])

= rank(Ql [Ai , Bi , Li , Wi ]) rank(Ql1 [Ai , Bi , Li , Wi ])

for l = 0, 1, 2, .... Then it follows from Lemma A.56 that ki+1 = ki 1.

Therefore it is sucient to prove (A.38). From the denitions one has

Wi+1 = [Li Bi2 1

i Wi1 ]
and
LTi+1 Ali+1 Bi+1 = LTi Mi (Ni Ai Mi )l Ni [Ai Bi2 1
i Bi1 ]
for l = 0, 1, 2, .... By noting these identities and Mi Ni = Ini Bi2 Ei1 Ci2 , we
can verify the equation

rank(Ql1 [Ai+1 , Bi+1 , Li+1 , Wi+1 ]) =

Li Bi2 Wi1 0 0 ... 0 0


Li Ai Bi2 Li Bi1 Li Bi2 Wi1 ... 0 0

= rank



... ...


l1 l2 l2 l3
Li Ai Bi2 Li Ai Bi1 Li Ai Bi2 Li Ai Bi2 ... Li Bi2 Wi1

We may also obtain

rank(Ql [Ai , Bi , Li , Wi ]) =

Wi1 0 0 0 ... 0 0


Li Bi1 Li Bi2 Wi1 0 ... 0 0

= rank



... ...


Li Al1
i Bi1 Li Al1
i Bi2 Li Al2
i Bi1 Li Al2
i Bi2 ... Wi1 0

Therefore we obtain (A.38).

The main result is coming now.
530 A Background Material

Theorem A.60. [213] Suppose that the realization (A, B, C, D) is mini-

mal and that H(s) = C(sIn A)1 B + D Cmm . If H(s) is PR and
rank(H(s) + H T (s)) = m, then there exists an integer k 0 such that the
inherent integration of (s) = H(s) + H T (s) is 2k and the transfer function
Hk (s), which is constructed from the algorithm applied to H0 (s) = H(s) and
(A0 , B0 , C0 , D0 ) = (A, B, C, D), satises Rk = Hk () + HkT () > 0, and
any solution P of the KYP Lemma set of equations for H(s) is given by

P = NkT Pk Nk IRn0 n0 (A.39)

(notice that n0 = n), where
1
Pk = diag(Pk , Ek1 , ...., E11 , E01 )
k1 k1
with Pk a (nk + km + i=0 ri ) (nk + km + i=0 ri ) matrix, and

Nk1 Nk2 ...N1 N0
Ck1,2 Nk2 ...N1 N0

Nk =
....

C12 N0
C02

is a (nk +km+ k1 i=0 ri )n0 matrix. The matrix Pk is any symmetric positive
denite solution of the Riccati equation:

Pk Ak + ATk + (CkT Pk Ak )Rk1 (Ck BkT Pk ) = 0 (A.40)

If H(s) = C(sIn A)1 B + D is WSPR then (A.40) has a unique positive
denite symmetric stabilizing solution Pk .

For the ease of reading let us recall that Nk IRnk+1 nk , Ck,2

(mrk )nk
IR , Ek IR(mrk )(mrk ) . Proof: From the KYP Lemma it follows
that (s) = G(s)GT (s) holds, where G(s) = W +LT (sIn A)1 B. Also since
G(s) is invertible there exists a nonnegative integer k such that the inherent
integration of G(s) is k. Then it is easy to see that the inherent integration
of (s) is 2k. Now since (A0 , B0 , C0 , D0 ) has the property 0 and since the
inherent integration of G(s) is k, Lemmas A.58 and A.59 show that we can
apply the algorithm to the system 0 . Then the system k = (Ak , Bk , Ck , Dk )
has the property k and the inherent integration of Gk (s) is zero. The last
condition is equivalent to det(Wk ) = 0, or Rk = Gk () + GTk () > 0. From
T
the property k (b) we have LTk = Wk (Ck BkT Pk )and (A.40) in the case
Pi+1 0
when nk > 0. By examination of Pi = and Pi = TiT Pi Ti1 ,
0 Ini ni+1
we get
 A.5 Some Useful Matrix Algebra 531

Pi+1 0
Pi = TiT Ti
0 Ini ni+1
  (A.41)
Pi+1 0 Ni
= [NiT T
Ci2 ]
0 Ei1 Ci2
Repeating the equation above, one can obtain (A.39). When nk = 0 it
follows from nk = nk1 m + rk1 that nk1 = m rk1 . This means that
Bk1,2 is square and from Lemma A.57 it is nonsingular. This yields
T 1 1
Pk1 = Ck1,2 Bk1,2 = Ck1,2 Ek1 Ck1,2
Combining this and (A.41) gives (A.39)
with nk = 0. Furthermore if H(s)
Ak jInk Bk
is WSPR then det = 0, for all IR, and (Ak , LTk ) is
T
Lk Wk
observable. This means that Hk (s) is SSPR and that the Riccati equation
(A.40) has a unique stabilizing solution (see for instance Theorem 3.44).

A.5 Some Useful Matrix Algebra

In this section some matrix algebra results are provided, some of which are
instrumental in the PR and dissipative systems characterization.

A.5.1 Results Useful for the KYP Lemma LMI

Theorem A.61. Let G IRnn , g IRnm , IRmm be arbitrary

matrices and vector, respectively. Then

G g
> 0 G > 0 and g T G1 g > 0
gT
(A.42)
> 0 and G g 1 g T > 0

(g T G1 g 1 ) < 1

 
G g gT
Since proving that > 0 is equivalent to proving that > 0,
gT g G
the equivalence between (3.3) and (3.17) follows from Theorem A.61, identi-
T 1
fying with P A AT P and G with D  + D . The matrix g G g is
T

G g
the so-called Schur complement of G in . Another useful result is the
gT
following:
532 A Background Material

Lemma A.62.
 [272]
Let G IRmm be an invertible matrix and be square.
G g
Then rank T = m if and only if = g T G1 g.
g
Still, another result related to the above is the following:

M11 M12
Proposition A.63. [157, 260] Let M = be a real symmetric
T
M12 M22
matrix. Then M 0 if and only if there exists real matrices L and W such
that M11 = LLT , M12 = LW , M22 W T W . Moreover M > 0 if and only if
L is full rank and M22 > W T W .
Proof: Let us prove the rst part with 0. The if sense is easy to prove.
The only if is as follows: Assume M 0. Let S be any real square matrix
such that M = S T S, i.e. S is a square root of M . Let S = QR be the QR
factorization of S with an orthonormal matrix Q and an upper triangular
T
matrix R. Then M = R R is a Cholesky
Tfactorization
of M . Let us partition
R11 R12 L W
the matrix R as R = = . From M = RT R it follows
0 R22 0 R22
that M11 = LLT , M12 = LW , M22 = W T W + R22 T
R22 W T W . Therefore L
and W satisfy the conditions of the proposition.
This proposition allows us to rewrite (3.2) as an inequality. Another result
that may be useful for the degenerate case of systems where D 0 is the
following one.

R1 0
Lemma A.64. Let Q, S, R be real matrices with R = , R1 > 0.
0 0
Then
Q S1 S 2
Q S

= S1T R1 0 0 (A.43)

ST R
S2T 0 0
if and only if
Q S1
0 (A.44)
S1T R1
and

Q 0
, (A.45)

S2 = 0
where S1 and S2 are of appropriate dimensions.
 A.5 Some Useful Matrix Algebra 533

One sees that applying Theorem A.61 the reduced order LMI can be rewrit-
ten as the Riccati inequality QS1T R1 S1 0. This is the reduced order Riccati
inequality satised by a PR system with a feedthrough term D 0.
The following is taken from [61] and also concerns the degenerate case
when D 0, where A is the Moore-Penrose pseudo inverse of the matrix A.

Q S
Lemma A.65. Suppose that Q and R are symmetric. Then 0 if
ST R
and only if R 0, Q SR S T 0, S(I RR ) = 0.

A.5.2 Inverse of Matrices

The following can be found in classical books on matrix algebra or linear sys-
tems [214, 246, 272]. Let A IRmm and C IRnn be nonsingular matrices.
Then

(A + BCD)1 = A1 A1 B(DA1 B + C 1 )1 DA1 (A.46)

so that

(I + C(sI A)1 B)1 = I C(sI A + BC)1 B, (A.47)

where I has the appropriate dimension. Let now A and B be square nonsin-
gular matrices. Then
1 
A 0 A1 0
= (A.48)
CB B CA1
1
B 1
and
1 
A D A1 A1 DB 1
= (A.49)
0 B 0 B 1
Let A be square nonsingular. Then
1 
A D A1 + EG1 F EG1
= (A.50)
C B G1 F G1
where G = B CA1 D, E = A1 D, F = CA1 . The matrix G is the Schur
complement of A.
534 A Background Material

A.5.3 Jordan Chain

Let T denote a linear transformation acting on an ndimensional linear space

S. A sequence {v0 , v1 , ..., vr1 } is called a Jordan chain of length r associated
with the eigenvalue , if


T (v0 ) = v0







T (v1 ) = v1 + v0

(A.51)

.



.



.

T (vr1 ) = vr1 + vr2
The vector vr1 is a generalized eigenvector of T of order r. Equiva-
lently, the vector (T I)r1 (vr1 ) is an eigenvector of T . Equivalently,
(T I)k (vr1 ) = 0 for k r. The length of any Jordan chain of T is nite,
and the members of a Jordan chain are linearly independent [272, 6.3].

A.5.4 Auxiliary Lemmas for the KYP Lemma Proof

The following results are used in Andersons proof of the KYP Lemma 3.1.

Lemma A.66. [11] Let (A, B, C) be a minimal realization for H(s). Suppose
that all the poles of H(s) lie in Re[s] < 0. With H(s) and W0 (s) related as
in (3.12). Suppose that W0 (s) has a minimal realization (F, G, L). Then the
matrices A and F are similar.

Proof: Since (A, B, C) is a realization for H(s), a direct calculation shows

that
T
A 0 B C
(A1 , B1 , C1 ) = , ,

0 AT CT B
is a realization of H(s) + H T (s). Since H(s) and H T (s) cannot have a pole
in common (the poles of H(s) are in Re[s] < 0 and those of H T (s) are in
Re[s] > 0) then the degree of H(s) + H T (s) is equal to twice the degree of
H(s). Thus the triple (A1 , B1 , C1 ) is minimal. By direct calculation one nds
that

W0T (s)W0 (s) = GT (sIn AT )1 LLT (sIn A)1 G

(A.52)
= C2 (sIn A2 )1 B2

with
 A.5 Some Useful Matrix Algebra 535

F 0 G 0
(A2 , B2 , C2 ) = , ,

LL F
T T
0 G
Using items (i) and (ii) below (3.12), it can then be shown that the de-
gree of W0T (s)W0 (s) is twice the degree of W0 (s) and therefore the triple
(A2 , B2 , C2 ) is minimal. Let P = P T > 0 be the unique positive denite so-
lution of F T P + P F = LLT . The existence of such a P follows from item
(i) below (3.12) and the  minimality
of (F, G, L). Then one may apply Lemma
In 0
A.69 below, choosing to obtain the following alternative realization
P In
of W0T (s)W0 (s)

F 0 G PG
(A3 , B3 , C3 ) = , ,

0 F T
PG G
Since (A1 , B1 , C1 ) and (A3 , B3 , C3 ) are minimal realizations of the same
transfer matrix, and since A has eigenvalues with strictly negative real part,
so has F from item (i) below (3.12). The result follows from Lemma A.69.

Corollary A.67. Let H(s) have a minimal realization (A, B, C) and let H(s)
and W0 (s) be related as in (3.12). Then there exists matrices K, L such
that W0 (s) has a minimal realization (A, K, L). Furthermore, two minimal
realizations of H(s) + H T (s) = W0T (s)W0 (s) are given by
T
A 0 B C
(A1 , B1 , C1 ) = , ,

0 AT CT B
and

A 0 K PK
(A3 , B3 , C3 ) = , ,

0 AT
PK K
where P is uniquely dened by P A + AT P = LLT .

Lemma A.68. [11] Let H(s) have a minimal realization (A, B, C) and let
H(s) and W0 (s) be related as in (3.12). Then there exists a matrix L such
that (A, B, L) is a minimal realization for W0 (s).

Lemma A.69. Let (A1 , B1 , C1 ) and (A2 , B2 , C2 ) be two minimal realizations

of the rational matrix H(s). Then there exists a nonsingular matrix T such
that A2 = T A1 T 1 , B2 = T B1 , C2 = (T T )1 C1 . Conversely if (A1 , B1 , C1 )
is minimal and T is nonsingular, then this triple (A2 , B2 , C2 ) is minimal.
536 A Background Material

A 0
Corollary A.70. The only matrices which commute with are of
 0 AT
T1 0
the form , where T1 and T2T commute with A.
0 T1
The next lemma is a specic version of the KYP Lemma that is needed in
its proof.

Lemma A.71. Let H(s) be PR and have only purely imaginary poles, with
H() = 0. Let (A, B, C) be a minimal realization of H(s). Then there exists
P = P T > 0 such that

P A + AT P = 0
(A.53)

P B = CT

Proof: The procedure consists in nding a minimal realization (A, B, C) for

which the matrix P has an obvious form. Then use the fact that if P satises
the set of equations (A.53) then P
= (T T )1 P T 1 satises the correspond-
ing set of equations for the minimal realization (T AT 1 , T B, (T 1)T C). Thus
if one exhibits a symmetric positive denite P for a particular minimal real-
ization, a symmetric positive denite P will exist
 Afor all minimal realizations.
i s+Bi
It is possible to write H(s) as H(s) = i s2 +i2 where the i are all
dierent and the matrices Ai and Bi satisfy certain requirements [376]. Let
us realize each term (Ai s + Bi )(s2 + i2 )1 separately with a minimal real-
ization (Fi , Gi , Hi ). Select a Pi such that (A.53) is satised, so as to obtain
a minimal reaization (F, G, H) and a P satisfying (A.53) with F = +i Fi ,
GT = [GT1 GT2 ....], H T = [H1T H2T ....] and P = +i Pi , where + is the direct
sum of the matrices [272, p.145-146]. As a consequence we can consider the
realization of the simpler transfer function
As + B
H(s) = (A.54)
s2 + 02
If the degree of H(s) in (A.54) is equal to 2k, then there exists k com-
plex vectors vi such that viT vi = 1, viT vi =: mui , 0 < i 1, i IR,
k  vi viT vi viT
and H(s) = i=1 sj0 + s+j 0
[376]. Direct sum techniques allow one to
restrict considerations to obtaining a minimal realization for the degree 2, i.e.

vv T vv T
H(s) = + (A.55)
s j0 s + j0
v+v
vv
,
Now dene y1 = 2
and y2 = 2
and check that
 
1 s 0 y1T
H(s) = [y1 y2 ]
s2 + 02 0 s y2T
 A.6 Well-posedness Results for State Delay Systems 537

and then

  T  T 8
0 0 y y
(F, G, H, P ) = , 1T , 1T , In
0 0 y2 y2
denes a minimal realization of (A.55) with the set of equations (A.53) satis-
ed.

A.6 Well-posedness Results for State Delay Systems

In this appendix we provide an existence and uniqueness of solutions for sys-
tems as in (5.82) or (5.84). Let us consider the state delay control equation


x(t) = Ax(t) + Lxt + Bu(t), t 0






x(0) = x0
(A.56)



x 0 () = ()





y(t) = Cx(t)
where A IRnn , L : C([, 0], IRn ) IRn and B : IRm IRn are bounded
linear operators. Here for a function z : [, ) IRn , the history of the
function z() is the function zt : [, 0] IRn dened by zt () = z(t + ) for
t 0 and [, 0]. It is assumed further that u Lp,e .

Denition A.72 (Mild solution). For the initial condition x0 IRn and
C([, 0], IRn ), a mild solution of the system (A.56) is the function dened
by
t
x(t) = etA x0 + 0 e(ts)A [Lxs + Bu(s)] ds, t 0
(A.57)

x(t + ) = (), 0

By using a straightforward argument from xed point theory, one can see
that the system (A.56) possesses a unique mild solution given as in Denition
A.72. An example of delay operator is given by

0
Lf = A1 f (1 ) + A2 ()f ()d (A.58)

where A1 IRnn , A2 () C([, 0], IRnn ), 1 0. More generally, let

: [, 0] L(IRnn ) be a function of bounded variation. We dene the
delay operator by
538 A Background Material

0
Lf = d()f () (A.59)

Now if we set = A1 1[,0]() + A2 () then we obtain the delay operator

dened by (A.58). Here 1[,0] () is the indicator function of the interval [, 0]
(not the same indicator as the one of convex analysis used elsewhere in this
book), i.e. the function that takes values 1 in [, 0] and 0 outside.
References

1. R. Abraham, J.E. Marsden 1978 Foundations of Mechanics, Second edition,

Benjamin Cummings, Reading, MA USA.
2. V. Acary, B. Brogliato, D. Goeleven, 2006 Higher order Moreaus sweeping
process: Mathematical formulation and numerical simulation, Mathematical
Programming, in press.
3. S. Adly, D. Goeleven, 2004 A stability theory for second-order nons-
mooth dynamical systems with application to friction problems, Journal de
Mathematiques Pures et Appliquees, vol.83, pp.17-51.
4. S. Adly, 2006 Attractivity theory for second order nonsmooth dynamical sys-
tems with application to dry friction, Journal of Mathematical Analysis and
Applications, in press.
5. S. Adly, H. Attouch, A. Cabot, 2003 Finite time stabilization of nonlinear
oscillators subject to dry friction, in Nonsmooth Mechanics and Analysis. The-
oretical and Numerical Advances, P. Alart, O. Maisonneuve, R.T. Rockafellar
(Eds.), Springer Advances in Mechanics and Mathematics, vol.12, pp.289-304.
6. A. Albu-Schafer, C. Ott, G. Hirzinger, 2004 A passivity based cartesian
impedance controller for exible joint robots Part II: Full state feedback,
impedance design and experiments, Proc. of the IEEE Int. Conference on
Robotics and Automation, New Orleans, LA, April, pp.2666-2672.
7. M. Amestegui, R. Ortega, J.M. Ibarra, 1987 Adaptive linearizing decoupling
robot control: a comparative study of dierent parametrizations, Proceedings
of the 5th Yale workshop on Applications of Adaptive Systems, New Haven,
CT.
8. B.D.O. Anderson and S. Vongpanitlerd, 1973 Network Analysis and Synthesis:
A Modern Systems Theory Approach, Englewood Clis, New Jersey; Prentice
Hall.
9. B.D.O. Anderson and J.B. Moore, 1971 Linear Optimal Control. Prentice-Hall,
Englewood Cli, N.Y..
10. B.D.O. Anderson, M. Mansour, F.J. Kraus, 1995 A new test for strict positive
realness, IEEE transactions on Circuits and Systems I- Fundamental Theory
and Applications, vol.42, no 4, pp.226-229, April.
11. B.D.O. Anderson, 1967 A system theory criterion for positive real matrices,
SIAM J. Control, vol.5, No 2, pp 171-182.
540 References

12. B.D.O. Anderson, I.D. Landau, 1994 Least squares identication and the ro-
bust strict positive real property, IEEE transaction on Cicuits and Systems I,
vol.41, no 9, pp.601-607.
13. B.D.O. Anderson, 1968 A simplied viewpoint of hyperstability, IEEE Trans-
actions on Automatic Control, vol.13, no 3, pp.292-294.
14. B.D.O. Anderson and J.B. Moore, 1967 Dual form of a positive real Lemma,
Proceedings of the IEEE, Proceedings Letters, October, pp.1749-1750.
15. M.D.S. Aliyu, J. Luttemaguzi, 2006 On the bounded real and positive real con-
ditions for ane nonlinear state delayed systems and applications to stability,
Automatica, vol.42, no 2, pp.357-362, February.
16. B.R. Andrievsky, A.N. Churilov, A.L. Fradkov, 1996 Feedback Kalman-
Yakubovich Lemma and its applications to adaptive control, Proceedings of
the 35th IEEE Conference on Decision and Control, Kobe, Japan, December,
pp.4537-4542.
17. B. dAndrea Novel, G. Bastin, B. Brogliato, G. Campion, C. Canudas, H. Khalil,
A. de Luca, R. Lozano, R. Ortega, P. Tomei, B. Siciliano, 1996 Theory of Robot
Control, C. Canudas de Wit, G. Bastin, B. Siciliano (Eds.), Springer Verlag,
CCES, London.
18. M.I. Angulo Nunez, H. Sira-Ramirez, 1998 Flatness in the passivity based
control of DC to DC power converters, Proc. of the 37th IEEE Conference on
Decision and Control, Tampa, FL, USA, pp.4115-4120.
19. A.C. Antoulas, 2005 A new result on passivity preserving model reduction,
Systems and Control Letters, vol.54, no 4, pp.361-374.
20. P. Apkarian, D. Noll, 2006 Nonsmooth H synthesis, IEEE Transactions on
Automatic Control, vol.51, no 1, pp.71-86, January.
21. P. Apkarian, D. Noll, 2006 Erratum to Nonsmooth H synthesis , IEEE
Transactions on Automatic Control, vol.51, no 2, p.382, February.
22. M. Arcak, P.V. Kokotovic, 2001 Observer-based control of systems with slope-
restricted nonlinearities, IEEE transactions on Automatic Control, vol.46, no
7, pp.1146-1150, July.
23. M. Arcak, P.V. Kokotovic, 2001 Feasibility conditions for circle criterion de-
signs, Systems and Control Letters, vol.42, no 5, pp.405-412.
24. S. Arimoto, 1996, Control Theory of Nonlinear Mechanical Systems: A Passivity-
Based and Circuit-Theoretic Approach, Oxford University Press, Oxford, UK.
25. S. Arimoto, 1999 Robotics research toward explication of everyday physics,
Int. J. of Robotics Research, vol.18, no 11, pp.1056-1063, November.
26. S. Arimoto, 1990 Learning control theory for robotic motion, International J.
of Adaptive Control and Signal Processing, vol.4, pp.543-564.
27. S. Arimoto, S.Kawamura, F. Miyazaki, 1984 Bettering operation of robots by
learning, Journal of Robotic Systems, vol.2, pp.123-140.
28. V.I. Arnold, 1973 Ordinary Dierential Equations, MIT Press, Cambridge, USA.
29. D.Z. Arov, M.A. Kaashoek, D.R. Pik, 2002/2003 the Kalman-Yakubovich-
Popov inequality and innite dimensional discrete time dissipative systems,
Reprot no 26, 2002/2203, spring, ISSN 1103-467X, ISRN IML-R-26-02/03-
SE+spring, Institut Mittag-Leer, The Royal Swedish Academy of Sciences.
30. D.Z. Arov, O.J. Staans, 2005 The innite-dimensional continuous time
Kalman-Yakubovich-Popov inequality, Operator Theory: Advances and Ap-
plications, vol.1, pp.1-37.
 References 541

31. Z. Bai, R.W. Freund, 2000 Eigenvalue based characterization and test for pos-
itive realness of scalar transfer functions, IEEE transactions on Automatic
Control, vol.45,no 12, pp.2396-2402, December.
32. A.V. Balakrishnan, 1995 On a generalization of the Kalman-Yakubovich
Lemma, Appl. Math. Optim., vol.31, pp.177-187.
33. J.A. Ball, J.W. Helton, 1996 Viscosity solutions of Hamilton-Jacobi equations
arising in nonlinear H -control, Journal of Mathematical Systems, Estimation
and Control, vol.6,no 1, pp.1-22.
34. P. Ballard, 2001 Formulation and well-posedness of the dynamics of rigid-body
systems with perfect unilateral constraints, Phil. Trans. Royal Soc., Mathemat-
ical, Physical and Engineering Siences, special issue on Nonsmooth Mechanics,
series A, vol.359, no 1789, pp.2327-2346.
35. N.E. Barabanov, 1988 On the Kalman problem, Sibirskii Matematischeskii
Zhurnal, vol.29, pp.3-11, May-June. Translated in Siberian Mathematical Jour-
nal, pp.333-341.
36. N.E. Barabanov, A.K. Gelig, G.A. Leonov, A.L. Likhtarnikov, A.S. Matveev,
V.B. Smirnova, A.L. Fradkov, 1996 The frequency theorem (Kalman-
Yakubovich lemma) in control theory, Automation and Remote Control, vol.57,
no 10, pp.1377-1407.
37. N.E. Barabanov, R. Ortega, 2004 On the solvability of extended Riccati equa-
tions, IEEE transactions on Automatic Control, vol.49, no 4, pp.598-602, April.
38. N.E. Barabanov, R. Ortega, 2002 Matrix pencils and extended algebraic Riccati
equations, European Journal of Control, vol.8, no 3, pp.251-264.
39. F.D. Barb, V. Ionescu, W. de Koning, 1994 A Popov theory based approach to
digital H control with measurement feedback for Pritchard-Salamon systems,
IMA Journal of Mathematical Control and Information, vol.11, pp.277-309.
40. F.D. Barb, W. de Koning, 1995 A Popov theory based survey in digital control
of innite dimensional systems with unboundedness, IMA Journal of Mathe-
matical Control and Information, vol.12, pp.253-298.
41. I. Barkana, 2004 Comments on Design of strictly positive real systems using
constant output feedback, IEEE transactions on Automatic Control, vol.49, no
11, pp.2091-2093, November.
42. I. Barkana, M.C.M. Teixeira, L. Hsu, 2006 Mitigation of symmetry condition
in positive realness for adaptive control, Automatica, in press.
43. S. Battilotti, L. Lanari, 1995 Global set point control via link position measure-
ment for exible joint robots, Systems and Control Letters, vol.25, pp.21-29.
44. S. Battilotti, L. Lanari, R. Ortega, 1997 On the role of passivity and out-
put injection in the ouput feedback stabilisation problem: Application to robot
control, European Journal of Control, vol.3, pp.92-103.
45. H. Berghuis, H. Nijmeijer, 1993 A passivity approach to controller-observer
design for robots, IEEE Transactions on Robotics and Automation, vol.9, no
6, pp.741-754.
46. J. Bernat, J. Llibre, 1996 Counterexample to Kalman and Markus-Yamabe
conjectures in dimension larger than 3, Dynamics of Continuous, Discrete and
Impulsive Systems, vol.2, pp.337-379.
47. G.M. Bernstein and M.A. Lieberman, 1989 A method for obtaining a canonical
Hamiltonian for nonlinear LC circuits, IEEE Trans. on Circuits and Systems,
vol.35, no 3, pp.411-420.
48. D.S. Bernstein, W.M. Haddad, A.G. Sparks, 1995 A Popov criterion for un-
certain linear multivariable systems, Automatica, vol.31, no 7, pp.1061-1064.
542 References

49. A. Betser, E. Zeheb, 1993 Design of robust strictly positive real transfer func-
tions, IEEE transactions on Circuits and Systems, I- Fundamental Theory and
Applications, vol.40, no 9, pp.573-580, September.
50. G. Bianchini, A. Tesi, A. Vicino, 2001 Synthesis of robust strictly positive real
systems with l2 parametric uncertainty, IEEE transactions on Circuits and
Systems, I- Fundamental Theory and Applications, vol.48, no 4, pp.438-450,
April.
51. G. Bianchini, 2002 Synthesis of robust strictly positive real discrete-time sys-
tems with l2 parametric perturbations, IEEE transactions on Circuits and Sys-
tems, I- Fundamental Theory and Applications, vol.49, no 8, pp.1221-1225, Au-
gust.
52. G. Blankenstein, R. Ortega, A.J. van der Schaft, 2002 The matching conditions
of controlled Lagrangians and interconnections and damping assignment passiv-
ity based control, International Journal of Control, vol.75, no 9, pp.645-665.
53. P.A. Bliman, 2001 Lyapunov-Krasovskii functionals and frequency domain:
delay-independent absolute stability criteria for delay systems, International
Journal of Robust and Nonlinear Control, vol.11, pp.771-788.
54. A.N. Bloch, N. Leonard, J.E. Mardsen, 1997 Stabilization of mechanical sys-
tems using controlled Lagrangians, Procedings of IEEE Conference on Decision
and Control, pp.2356-2361.
55. A.N. Bloch, N. Leonard, J.E. Mardsen,2000 Controlled Lagrangians and the
stabilization of mechanical systems I: The rst matching theorem, IEEE trans-
actions on Automatic Control, vol.45, no 12, pp.2253-2269.
56. A.A. Bobstov, N.A. Nikolaev, 2005 Fradkov theorem-based design of the con-
trol of nonlinear systems with functional and parametric uncertainties, Au-
tomation and Remote Control, vol.66, no 1, pp.108-118.
57. V.A. Bondarko, A.L. Fradkov, 2003 Necessary and sucient conditions for the
passivicability of linear distributed systems, Automation and Remote Control,
vol.64, no 4, pp.517-530.
58. H. Bounit, H. Hammouri, 1998 Stabilization of innite-dimensional semilinear
systems with dissipative drift, Applied Mathematics and Optimization, vol.37,
pp.225-242.
59. H. Bounit, H. Hammouri, 2003 A separation principle for distributed dissipa-
tive bilinear systems, IEEE Transactions on Automatic Control, vol.48, no 3,
pp.479-483, March.
60. J.M. Bourgeot, B. Brogliato, 2005 Tracking control of complementarity La-
grangian systems, International Journal of Bifurcation and Chaos, vol.15, no
6, pp.1839-1866, June.
61. S.P. Boyd, L.E. Ghaoui, E. Feron, V. Balakrishnan, 1994 Linear Matrix Inequal-
ities in System and Control Theory, SIAM Studies in Applied Mathematics,
vol.15.
62. F. H. Branin, 1977 The network concept as a unifying principle in engineering
and the physical sciences, in Problem Analysis in Science and Engineering,
edited by F. H. Branin and K. Huseyin, 41-111, Academic Press, New York.
63. P.C. Breedveld, 1984 Physical Systems Theory in terms of Bond Graphs, Ph.D.
thesis, University of Twente , The Netherlands.
64. A. Bressan, Viscosity solutions of Hamilton-Jacobi equations and optimal con-
trol problems, SISSA, Trieste, Italy (http://www.sissa.it/publications/hj.ps )
 References 543

65. A. Bressan, F. Rampazzo, 1993 On dierential systems with quadratic im-

pulses and their applications to Lagrangian systems, SIAM J. on Control and
Optimization, vol.31, no 5, pp.1205-1220.
66. H. Brezis, 1973 Operateurs Maximaux Monotones, North Holland Mathematics
Studies, Amsterdam, 1973.
67. M. Bridges, D.M. Dawson, 1995 Backstepping control of exible joint manip-
ulators: a survey, Journal of Robotic Systems, vol.12, no 3, pp.199-216.
68. R.W. Brockett, 1977 Control theory and analytical mechanics, in Geomet-
ric Control Theory, C. Martin and R. Herman (Eds.), pp.1-46, Math.Sci.Press,
Brookline, 1977.
69. B. Brogliato, 1999 Nonsmooth Mechanics, Springer Verlag, London, Communi-
cations and Control Engineering Series, 2nd edition. Erratum and addendum at
http://bipop.inrialpes.fr/people/brogliato/brogli.html .
70. B. Brogliato, 1991 Systemes Passifs et Commande Adaptative des Manipula-
teurs, PhD Thesis, Institut National Polytechnique de Grenoble, France, 11
January.
71. B. Brogliato, I.D. Landau, R. Lozano, 1991 Adaptive motion control of robot
manipulators: a unied approach based on passivity, Int. J. of Robust and
Nonlinear Control, vol.1, no 3, July-September, pp.187-202.
72. B. Brogliato, R. Lozano, 1996 Correction to Adaptive control of robot manip-
ulators with exible joints, IEEE Transactions on Automatic Control, vol.41,
no 6, pp.920-922.
73. B. Brogliato, R. Lozano, 1992 Passive least-squares-type estimation algorithm
for direct adaptive control, Int. J. of Adaptive Control and Signal Processing,
January, vol.6, no 1, pp.35-44.
74. B. Brogliato, R. Lozano, I.D. Landau, 1993 New relationships between Lya-
punov functions and the passivity theorem, Int. J. Adaptive Control and Signal
Processing, vol.7, pp.353-365.
75. B. Brogliato, S.I. Niculescu, M.D.P. Monteiro Marques, 2000 On tracking con-
trol of a class of complementary-slackness hybrid mechanical systems, Systems
and Control Letters, vol.39, pp.255-266.
76. B. Brogliato, S.I. Niculescu, P. Orhant, 1997 On the control of nite dimen-
sional mechanical systems with unilateral constraints, IEEE Transactions on
Automatic Control, vol.42, no 2, pp.200-215.
77. B. Brogliato, 2001 On the control of nonsmooth complementarity dynami-
cal systems, Phil. Trans. Royal Soc., Mathematical, Physical and Engineering
Sciences, special issue on Nonsmooth Mechanics, series A, vol.359, no 1789,
pp.2369-2383.
78. B. Brogliato, R. Ortega, R. Lozano, 1995 Global tracking controllers for
exible-joint manipulators: a comparative study, Automatica, vol.31, no 7,
pp.941-956.
79. B. Brogliato, D. Rey, 1998 Further experimental results on nonlinear control of
exible joint manipulators, Proceedings of the American Control Conference,
vol.4, pp.2209-2211, Philadelphia, PA, USA, June 24-26.
80. B. Brogliato, D. Rey, A. Pastore, J. Barnier, 1998 Experimental comparison
of nonlinear controllers for exible joint manipulators, Int. J. of Robotics Re-
search, vol.17, no 3, March, pp.260-281.
81. B. Brogliato, 2004 Absolute stability and the Lagrange-Dirichlet theorem with
monotone multivalued mappings, Systems and Control Letters, vol.51, pp.343-
544 References

353 (preliminary version Proceedings of the 40th IEEE Conference on Decision

and Control, 4-7 December 2001, pp.27-32, vol.1).
82. B. Brogliato, D. Goeleven, 2005 The Krakovskii-LaSalle invariance principle
for a class of unilateral dynamical systems, Mathematics of Control, Signals
and Systems, vol.17, pp.57-76.
83. B. Brogliato, 2003 Some perspectives on the analysis and control of comple-
mentarity systems, IEEE Transactions on Automatic Control, vol.48, no 6,
pp.918-935, June.
84. B. Brogliato, A. Daniilidis, C. Lemarechal, V. Acary, 2006 On the equiva-
lence between complementarity systems, projected systems and dierential in-
clusions, Systems and Control Letters, vol.55, no 1, pp.45-51, January.
85. B. Brogliato, L. Thibault, 2006 Well-posedness results for non-autonomous dis-
sipative complementarity systems, INRIA Research Report 5931, ISSN 0249-
6399, June, http://hal.inria.fr
86. O. Brune, 1931 The synthesis of a nite two-terminal network whose driving-
point impedance is a prescribed function frequency, J. Math. Phys., vol.10,
pp.191-236.
87. V.A. Brusin, 1976 The Lurie equation in the Hilbert space and its solvability
(in Russian), Prikl. Mat. Mekh., vol.40, no 5, pp.947-955.
88. A.E. Bryson, Y.C. Ho, 1975 Applied Optimal Control. Optimizaton, Estimation
and Control, Taylor and Francis.
89. C.I. Byrnes, A. Isidori, J.C. Willems, 1991 Passivity, feedback equivalence, and
the global stabilization of minimum phase nonlinear systems, IEEE Transac-
tions on Automatic Control, vol.36, no 11, pp. 1228-1240.
90. C.I. Byrnes, W. Lin, 1994 Losslessness, feedback equivalence, and the global
stabilization of discrete-time nonlinear systems, IEEE transactions on Auto-
matic Control, vol.39, no 1, pp.83-98, January.
91. A. Cabot, 2006 Stabilization of oscillators subject to dry friction: nite time
convergence versus exponential decay results, Transactions of the American
Mathematical Society, in press.
92. Z. Cai, M.S. de Queiroz, D.M. Dawson, 2006 A suciently smooth projection
operator, IEEE Transactions on Automatic Control, vol.51, no 1, pp.135-139,
January.
93. P.E. Caines, 1987 Linear Stochastic Systems, Wiley series in Probability and
Mathematical Statistics, NY.
94. K. Camlibel, W.P.M.H. Heemels, H. Schumacher, 2002 On linear passive com-
plementarity systems, European Journal of Control, vol.8, no 3, pp.220-237.
95. G. Campion, B. dAndrea-Novel and G. Bastin, 1990 Controllability and state-
feedback stabilizability of non-holonomic mechanical systems, in Advanced
Robot Control, C. de Witt (Ed.), Springer LNCIS 162, pp.106-124, 1990
96. H. Cartan, 1967 Cours de Calcul Dierentiel, Hermann, Paris (fourth edition
1990).
97. K.M. Chang, 2001 Hyperstability appproach to the synthesis of adaptive con-
troller for uncertain time-varying delay systems with sector bounded nonlinear
inputs, Proceedings of the I Mech. E Part 1, Journal of Systems and Control
Engineering, vol.215, no 5, pp.505-510.
98. D.N. Cheban, 1999 Relationship between dierent types of stability
for linear almost periodic systems in Banach spaces, Electronic Jour-
nal of Dierential Equations, no 46, pp.1-9 (http://ejde.math.unt.edu or
http://ejde.math.swt.edu).
 References 545

99. V. Chellaboina, W.M. Haddad, 2003 Exponentially dissipative dynamical sys-

tems: a nonlinear extension of strict positive realness, Mathematical Problems
in Engineering, vol.1, pp.25-45.
100. V. Chellaboina, W.M. Haddad, A. Kamath, 2005 A dissipative dynamical
systems approach to stability analysis of time delay systems, International
Journal of Robust and Nonlinear Control, vol.15, pp.25-33.
101. Z. Chen, J. Huang, 2004 Dissipativity, stabilization, and regulation of cascade-
connected systems, IEEE transactions on Automatic Control, vol.49, no 5,
pp.635-650, May.
102. Y. Cheng, L. Wang, 1993 On the absolute stability of multi nonlinear con-
trol systems in the critical cases, IMA Journal of Mathematical Control and
Information, vol.10, pp.1-10.
103. L.O. Chua, J.D. McPherson, 1974 Explicit topological formulation of la-
grangian and Hamiltonian equations for nonlinear networks, IEEE trans. on
Circuits and Systems, vol.21, no 2, pp.277-285.
104. A. Cima, A. Gasull, E. Hubbers, F. Manosas, 1997 A polynomial counterex-
ample to the Markus-Yamabe conjecture, Advances in Mathematics, vol.131,
pp.453-457, article no AI971673.
105. F.H. Clarke, 1983 Optimization and Nonsmooth Analysis, Canadian Math. Soc.
Series of Monographs and Advanced Texts, Wiley Interscience Publications.
106. D. Cobb, 1982 On the solution of linear dierential equations with singular
coecients, Journal of Dierential Equations, vol.46, pp.310-323.
107. E.A. Coddington, N. Levinson, 1955 Theory of Ordinary Dierential Equa-
tions, Tata McGraw Hill Publishing company LTD, New Delhi (sixth reprint,
1982).
108. P. Coey Duncan, C.A. Farschman, B.E. Ydstie, 2000 Distillation stability us-
ing passivity and thermodynamics, Computers Chem. Engng., vol.24, pp.317-
322.
109. J.E. Colgate, G. Schenkel, 1997 Passivity of a class of sampled-data systems:
application to haptic interface, J. Robot. Systems, vol.14, no 1, pp.37-47, Jan-
uary.
110. J. Collado, R. Lozano, R. Johansson, 2001 On Kalman-Yakubovich-Popov
Lemma for stabilizable systems, IEEE Transactions on Automatic Control,
vol.46, no 7, pp.1089-1093, July.
111. J. Collado, R. Lozano, R. Johansson, 2005 Observer-based solution to strictly
positive real problem In Nonlinear and Adaptive Control: Theory and Algo-
rithms for the User, Imperial College Press, London ( A. Astol, Editor).
112. R. Costa Castello, R. Grino, 2006 A repetitive controller for discrete-time
passive systems, Automatica, in press.
113. R.W. Cottle, J.S. Pang, R.E. Stone, 1992 The Linear Complementarity Prob-
lem, Academic Press.
114. J.J. Craig, Pl. Hsu, S. Sastry, 1986 Adaptive control of mechanical manipula-
tors, IEEE International Conference on Robotics and Automation, San Fran-
cisco.
115. J.J. Craig, 1988 Adaptive Control of Mechanical Manipulators, Addison Wesley,
reading, MA.
116. M. Cromme, 1998 On dissipative systems and set stability, MAT-Report no
1998-07, April, Technical university of Denmark, Dept. of Mathematics.
117. R.F. Curtain, 1996 The Kalman-Yakubovich-Popov Lemma for Pritchard-
Salamon systems, Systems and Control Letters, vol.27, pp.67-72.
546 References

118. R.F. Curtain, 1996 Corrections to The Kalman-Yakubovich-Popov Lemma

for Pritchard-Salamon systems , Systems and Control Letters, vol.28, pp.237-
238.
119. R.F. Curtain, J.C. Oostveen, 2001 The Popov criterion for strongly stable dis-
tributed parameter systems, International Journal of Control, vol.74, pp.265-
280.
120. R.F. Curtain, M. Demetriou, K. Ito, 2003 Adaptive compensators for per-
turbed positive real innite dimensional systems, Int. J. Applied Math. and
Computer Science,vol.13, No.4. pp.441-452.
121. M. Dalsmo, A.J. van der Schaft, 1999 On representations and integrability
of mathematical structures in energy-conserving physical systems, SIAM J.
Control and Optimization, vol.37, no 1, pp.54-91.
122. D. Danciu, V. Rasvan, 2000 On Popov-type stability criteria for neural net-
works, Electronic Journal of Qualitative Theory of Dierential Equations, Proc.
of the 6th Coll. QTDE, no 23, August 10-14 1999, Szeged, Hungary.
123. R. Datko, 1970 Extending a theorem of A.M. Lyapunov to Hilbert space, J.
Math. Anal. Appl., vol.32, pp.610-616.
124. K. Deimling, 1992 Multivalued Dierential Equations, De Gruyter Series in
Nonlinear Analysis and Applications, Berlin-New York.
125. C.A. Desoer, M. Vidyasagar, 1975 Feedback Systems: Input-Output properties,
Academic Press, New-York.
126. C.A. Desoer, E.S. Kuh, 1969 Basic Circuit Theory McGraw Hill Int.
127. J. Dieudonne, 1969 Elements dAnalyse, tome 2, Gauthier-Villars, Paris.
128. N. Diolaiti, G. Niemeyer, F. Barbagli, J.K. Salisbury, 2006 Stability of haptic
rendering: discretization, quantization, time delay, and Coulomb eects, IEEE
transactions on Robotics, vol.22, no 2, pp.256-268, April.
129. P.M. Dower, M.R. James, 1998 Dissipativity and nonlinear systems with -
nite power gain, International Journal of Robust and Nonlinear Control, vol.8,
pp.699-724.
130. M.A. Duarte-Mermoud, R. Castro-Linares, A. Castillo-Facuse, 2001 Adaptive
passivity of nonlinear systems using time-varying gains, Dynamics and Control,
vol.11, pp.333-351.
131. M. Duenas Diez, B. Lie, B.E. Ydstie, 2002 Passivity-based control of partic-
ulate processes modeled by population balance equations, 4th World Congress
on Particle Technology, Sydney, Australia, 21-25 July.
132. B. Dumitrescu, 2002 Parametrization of positive-real transfer functions with
xed poles, IEEE transactions on Circuits and Systems, I- Fundamental Theory
and Applications, voil.49, no 4, pp.523-526, April.
133. O. Egeland, J.-M. Godhavn, 1994 Passivity-based adaptive attitude control
of a rigid spacecraft, IEEE Transactions on Automatic Control, vol.39, pp.842-
846, April.
134. European Journal of Control, special issue Dissipativity of Dynamical Systems:
Application in Control dedicated to Vasile Mihai Popov, vol.8, no 3, 2002.
135. G. Escobar, H. Sira-Ramirez, 1998 A passivity based sliding mode control
approach for the regulation of power factor precompensators, Proc. of the 37th
IEEE Conference on Decision and Control, Tampa, FL, pp.2423-2424.
136. R. Fabbri, S.T. Impram, 2003 On a criterion of Yakubovich type for the
absolute stability of non-autonomous control processes, International Journal
of Mathematics and Mathematical Sciences, vol.16, pp.1027-1041.
 References 547

137. F. Facchinei, J.S. Pang, 2003 Finite-dimensional Variational Inequalities and

Complementarity Problems, vol.I and II, Springer Series in Operations Research,
New-York.
138. I. Fantoni, R. Lozano, M. Spong, 1998 Energy based control of the pendubot,
IEEE Transactions on Automatic Control, vol.45, no 4, pp.725-729.
139. I. Fantoni, R. Lozano, F. Mazenc, K. Y. Pettersen, 2000, Stabilization of
an underactuated hovercraft, International Journal of Robust and Nonlinear
Control, vol.10, pp.645-654.
140. M. Farhood, G.E. Dullerud, 2005 Duality and eventually periodic systems,
International Journal of Robust and Nonlinear Control, vol.15, pp.575-599.
141. M. Farhood, G.E. Dullerud, 2002 LMI tools for eventually periodic systems,
Systems and Control Letters, vol.47, pp.417-432.
142. N.H. El-Farra, P.D.Christodes, 2003 Robust inverse optimal control laws
for nonlinear systems, International Journal of Robust and Nonlinear Control,
vol.13, pp.1371-1388.
143. E.D. Fasse, P.C. Breedveld, 1998 Modelling of Elastically Coupled Bodies:
Part I: General Theory and Geometric Potential Function Method, ASME J.
of Dynamic Systems, Measurement and Control, vol.120, pp.496-500, December.
144. E.D. Fasse, P.C. Breedveld, 1998 Modelling of Elastically Coupled Bodies:
Part II: Exponential- and Generalized-Coordinate Methods, ASME J. of Dy-
namic Systems, Measurement and Control, Vol.120, pp.501-506, December.
145. P. Faurre, M. Clerget, F. Germain, 1979 Operateurs Rationnels Positifs. Appli-
cation a lHyperstabilite et aux Processus Aleatoires, Methodes Mathematiques
de lInformatique, Dunod, Paris.
146. G. Fernandez-Anaya, J.C. Martinez-Garcia, V. Kucera, 2006 Characteriz-
ing families of positive real matrices by matrix substitutions on scalar ra-
tional functions, Systems and Control Letters, in press (also in proceedings
of the 5th Asian Control Conference, 20-23 July 2004, Melbourne, Australia,
http://www.ee.unimelb.edu.au/conferences/ascc2004/).
147. G. Fernandez-Anaya, 1999 Preservation of SPR functions and stabilization by
substitutions in SISO plants, IEEE transactions on Automatic Control, vol.44,
no 11, pp.2171-2174, November.
148. G. Fernandez, R.Ortega, 1987 On positive-real discrete-time transfer func-
tions, proceedings of the IEEE, vol.75, no 3, pp.428-430, March.
149. G. Fernandez-Anaya, J.C. Martinez-Garcia, V. Kucera, D. Aguilar-George,
2004 MIMO systems properties preservation under SPR substitutions, IEEE
transactions on Circuits and SystemsII Express Briefs, vol.51, no 5, pp.222-227,
May.
150. A. Ferrante, 2005 Positive real Lemma: necessary and sucient conditions for
the existence of solutions under virtually no assumptions, IEEE transactions
on Automatic Control, vol.50, no 5, pp.720-724, May.
151. A. Ferrante, L. Pandol, 2002 On the solvability of the positive real lemma
equations, Systems and Control Letters, vol.47, pp.211-219.
152. E. Fossas, R.M. Ros, H. Sira-Ramirez, 2004 Passivity-based control of a biore-
actor system, Journal of Mathematical Chemistry, vol.36, no 4, pp.347-360.
153. A.L. Fradkov, 2003 Passication of non-square linear systems and feed-
back Yakubovich-Kalman-Popov Lemma, European Journal of Control, vol.6,
pp.573-582.
154. A.L. Fradkov, 1974 Synthesis of an adaptive system of linear plant stabiliza-
tion, Automat. Telemekh., no 12, pp.1960-1966.
548 References

155. A.L. Fradkov, 1976 Quadratic Lyapunov functions in a problem of adaptive

stabilization of a linear dynamical plant, Sibirskiy Math. J., no 2, pp.341-348.
156. A.L. Fradkov, D.J. Hill, 1998 Exponential feedback passivity and stabilizabil-
ity of nonlinear systems, Automatica, vol.34, no 6, pp.697-703.
157. R.W. Freund, F. Jarre, 2004 An extension of the Positive Real Lemma to
descriptor systems, Optimization Methods and Software, no 1, vol.19, pp.69-
87.
158. E. Fridman, U. Shaked, 2002 On delay-dependent passivity, IEEE transac-
tions on Automatic Control, vol.47, no 4, April.
159. Z.X. Gan, W.G. Ge, 2001 Lyapunov functional for multiple delay general
Lure control systems with multiple nonlinearities, Journal of Mathematical
Analysis and Applications, vol.259, pp.596-608.
160. R. Garrido-Moctezuma, D. Suarez, R. Lozano, 1998, Adaptive LQG control
of PR plants, Int. J. of Adaptive Control, Vol 12, pp 437-449.
161. J.P. Gauthier, G. Bornard, 1981 Stabilisation des systemes non lineaires,
Outils et modeles mathematiques pour lautomatique, lanalyse de systemes et
le traitement du signal, vol.1, Trav. Rech. coop. Programme 567, pp.307-324.
162. R. van der Geest, H. Trentelman, 1997 The Kalman-Yakubovich-Popov
Lemma in a behavioural framework, Systems and Control Letters, vol.32,
pp.283-290.
163. A. Kh. Gelig, G.A. Leonov, V.A. Yakubovich, 1978 The Stability of Nonlinear
Systems with a Nonunique Equilibrium State, (in Russian), Nauka, Moscow.
164. F. Genot, B. Brogliato, 1999 New results on Painleve paradoxes, European
Journal of Mechanics A/Solids, vol.18, no 4, pp.653-677.
165. J.C. Geromel, P.B. Gapski, 1997 Synthesis of positive real H2 controllers,
IEEE transactions on Automatic Control, vol.42, no 7, pp.988-992, July.
166. M.I. Gil, R. Medina, 2005 Explicit stability conditions for time-discrete vector
Lure type systems, IMA Journal of Mathematical Control and Information,
vol.22, no 4, pp.415-421.
167. D. Goeleven, B. Brogliato, 2004 Stability and instability matrices for linear
evolution variational inequalities, IEEE transactions on Automatic Control,
vol.49, no 4, pp.521-534, April.(rst version in Proc. of IFAC conference ADHS,
16-18 June 2003, Saint-Malo, France).
168. D. Goeleven, D. Motreanu, Y. Dumont, M. Rochdi, 2003 Variational and
Hemivariational Inequalities: Theory, Methods and Applications. Volume 1: Uni-
lateral Analysis and Unilateral Mechanics, Kluwer Academic Publishers, Dor-
drecht, Nonconvex Optimization and its Applications series.
169. D. Goeleven, G.E. Stavroulakis, G. Salmon, P.D. Panagiotopoulos, 1997 Solv-
ability theory and projection method for a class of singular variational inequal-
ities: Elastostatic unilateral contact applications, J. of Optimization Theory
and Applications, vol.95, no 2, pp.263-294.
170. D. Goeleven, M. Motreanu, V. Motreanu, 2003 On the stability of stationary
solutions of evolution variational inequalities, Advances in Nonlinear Varia-
tional Inequalities, vol.6, pp.1-30.
171. R.B. Gorbet, K.A. Morris, D.W.L. Wang, 2001 Passivity-based stability and
control hysteresis in smart actuators, IEEE transactions on Control Systems
Technlology, vol.9, no 1, pp.5-16, January.
172. P. Grabowski, F.M. Callier, 2006 On the circle criterion for boundary con-
trol systems in factor form: Lyapunov stability and Lure equations, ESAIM
Control, Optimization and Calculus of Variations, vol.12, pp.169-197, January.
 References 549

173. L. Gui-Fang, L. Hui-Ying, Y. Chen-Wu, 2005 Observer-based passive control

for uncertain linear systems with delay in state and control input, Chinese
Physics, vol.14, no 12, pp.2379-2386.
174. S.V. Gusev, 2006 The Fenchel duality, Sprocedure and the Yakubovich-
Kalman Lemma, Automation and Remorte Control, vol.67, no 2, pp.293-310.
175. C. Gutierrez, 1995 A solution to the bidimensional global asymptotic stability
conjecture, Ann. Inst. Henri Poincare, vol.12, no 6, pp.627-671.
176. J.T. Gravdahl and O. Egeland, 1999 Compressor Surge and Rotating Stall:
Modeling and Control, Advances in Industrial Control, Springer-Verlag Lon-
don.
177. J. Gregor, 1996 On the design of positive real functions, IEEE transactions
on Circuits and Systems, I- Fundamental Theory and Applications, vol.43, no
11, pp.945-947, November.
178. S. Hadd, A. Idrissi, 2005 Regular linear systems governed by systems with
state, input and output delays, IMA Journal of Mathematical Control and
Information, vol.22, pp.423-439.
179. W.M. Haddad, D.S. Bernstein, Y.W. Wang, 1994 Dissipative H2 /H con-
troller synthesis, IEEE Transactions on Automatic Control, vol.39, pp.827-831.
180. W. Haddad, D. Bernstein, 1991 Robust stabilization with positive real uncer-
tainty: Beyond the small gain theorem, Systems and Control Letters, vol.17,
pp.191-208.
181. W. M. Haddad, D. S. Bernstein, 1993 Explicit construction of quadratic Lya-
punov functions for small gain, positivity, circle, and Popov theorems and their
application to robust stability. Part I: continuous-time theory, Internat. J. Ro-
bust Nonlinear Control,vol. 3, no.4, pp.313-339.
182. W.M. Haddad, D.S. Bernstein, 1994 Explicit construction of quadratic Lya-
punov functions for the small gain, positive, circle, and Popov Theorems and
their application to robust stabilityPart II: discrete-time theory, International
Journal of Robust and Nonlinear Control, vol.4, no 2, pp.229-265.
183. W.M. Haddad, V. Chellaboina, 2001 Dissipativity theory and stability of feed-
back interconnections for hybrid dynamical systems, Mathematical Problems
in Engineering, vol.7, pp.299-335.
184. W.M. Haddad, V. Chellaboina, Q. Hui, S. Nersersov, 2004 Vector dissipativity
theory for large-scale impulsive dynamical systems, Mathematical Problems in
Engineering, vol.3, pp.225-262.
185. W.M. Haddad, Q. Hui, V. Chellaboina, S. Nersesov, 2004 Vector dissipativity
theory for discrete-time large-scale nonlinear dynamical systems, Advances in
Dierence Equations, vol.1, pp.37-66.
186. W.M. Haddad, V. Chellaboina, N.A. Kablar, 2001 Nonlinear impulsive dy-
namical systems. Part I: stability and dissipativity, International Journal of
Control, vol.74, no 17, pp.1631-1658.
187. W.M. Haddad, V. Chellaboina, 1998 Nonlinear xed-order dynamic compen-
sation for passive systems, International Journal of Robust and Nonlinear Con-
trol, vol.8, no 4-5, pp.349-365.
188. W.M. Haddad, S. Nersersov, V. Chellaboina, 2003 Energy-based control for
hybrid port-controlled Hamiltonian systems, Automatica, vol.39, pp.1425-1435.
189. W.M. Haddad, V. Chellaboina, S. Nersersov, 2001 On the equivalence between
dissipativity and optimality of nonlinear hybrid controllers, International Jour-
nal of Hybrid Systems, vol.1, no 1, pp.51-66.
550 References

190. W.M. Haddad, D.S. Bernstein, 1995 Parameter dependent Lyapunov func-
tions and the Popov criterion in robust analysis and synthesis, IEEE Transac-
tions on Automatic Control, vol.40, no 3, pp.536-543.
191. W.M. Haddad, V. Chellaboina, 2005 Stability and dissipativity theory for
nonnegative dynamical systems: a unied analysis framework for biological
and physiological systems, Nonlinear Analysis, Real World Applications, vol.6,
pp.35-65.
192. W.M. Haddad, V. Chellaboina, T. Rajpurobit, 2004 Dissipativity theory for
nonnegative and compartmental dynamical systems with time delay, IEEE
Transactions on Automatic Control, vol.49, no 5, pp.747-751, May.
193. W.M. Haddad, E.G. Collins, D.S. Bernstein, 1993 Robust stability analysis
using the small gain, circle, positivity, and Popov Theorems: a comparative
study, IEEE transactions on Control, Systems and Technology, vol.1, no 4,
pp.290-293, December.
194. W.M. Haddad, V. Chellaboina, Q. Hui, S.G. Nerserov, 2005 Thermodynamic
stabilization via energy dissipating hybrid controllers, Proc. IEEE Conference
on Decision and ControlEuropean Control Conference, pp.4879-4884, Seville,
Spain, December.
195. W.M. Haddad, V. Chellaboina, Q. Hui, S.G. Nerserov, 2006 Energy and en-
tropy based stabilization for lossless dynamical systems via hybrid controllers,
submitted.
196. G. Hagen, 2006 Absolute stability via boundary control of a semilinear
parabolic PDE, IEEE Transactions on Automatic Control, vol.51, no 3, pp.489-
493, March.
197. T. Hagiwara, G. Kuroda, M. Araki, 1998 Popov-type criterion for stability of
nonlinear sampled-data systems, Automatica, vol.34, no 6, pp.671-682.
198. T. Hagiwara, M. Araki, 1996 Absolute stability of sampled-data systems with
a sector nonlinearity, Systems and Control Letters, vol.27, pp.293-304.
199. W. Hahn, 1967 Stability of Motion, Springer-Verlag, New York, NY.
200. A. Halanay, V. Rasvan, 1991 Absolute stability of feedback systems with
several dierentiable nonlinearities, Int. J. Systs. Sci., vol.23, no 10, pp.1911-
1927.
201. J.K. Hale, S.M. Verduyn Lunel, 1991 Introduction to Functional Dierential
Equations, Applied Math. Sciences, vol.99, Springer Verlag, NY.
202. Q.L. Han, 2005 Absolute stability of time-delay systems with sector-bounded
nonlinearity, Automatica, vol.41, pp.2171-2176, December.
203. T. Hayakawa, W.M. Haddad, J.M. Bailey, N. Hovakimyan, 2005 Passivity-
based neural network adaptive output feedback control for nonlinear nonnega-
tive dynamical systems, IEEE transactions on Neural Networks, vol.16, no 2,
pp.387-398, March.
204. Y. He, M. Wu, 2003 Absolute stability for multiple delay general Lure control
systems with multiple nonlinearities, Journal of Computational and Applied
Mathematics, vol.159, pp.241-248.
205. D. Henrion, 2002 Linear matrix inequalities for robust strictly positive real
design, IEEE transactions on Circuits and Systems, I- Fundamental Theory
and Applications, vol.49, no 7, pp.1017-1020, July.
206. D.J. Hill, P.J. Moylan, 1980 Connections between nite-gain and asymptotic
stability, IEEE Transactions on Automatic Control, vol.25, no 5, pp.931-936,
October.
 References 551

207. D.J. Hill, P.J. Moylan, 1976 The stability of nonlinear dissipative systems,
IEEE Transactions on Automatic Control,vol.21, no 5, pp.708-711, October.
208. D.J. Hill, P.J. Moylan, 1975 Cyclo-dissipativeness, dissipativeness and loss-
lessness for nonlinear dynamical systems, Technical Report EE7526, Novem-
ber, The university of Newcastle, Dept. of Electrical Engng., New South Wales,
Australia.
209. D.J. Hill, P.J. Moylan, 1980 Dissipative dynamical systems: Basic input-
output ands state properties, Journal of the Franklin Institute, vol.30,9, no
5,pp.327-357, May.
210. J.B. Hiriart Urruty, C. Lemarechal, Fundamentals of Convex Analysis,
Springer, Grundlehren Text Editions, Berlin, 2001.
211. L. Hitz, B.D.O. Anderson, 1969 Discrete positive-real functions and their
application to system stability, Proceedings of the IEE, vol.116, pp.153-155.
212. M.T. Ho, J.M. Lu, 2005 H PID controller design for Lure systems and its
application to a ball and wheel apparatus, International Journal of Control,
vol.78, no 1, pp.53-64.
213. I. Hodaka, N. Sakamoto, M. Suzuki, 2000 New results for strict positive real-
ness and feedback stability, IEEE transactions on Automatic Control, vol.45,no
4, pp.813-819, April.
214. R.A. Horn, C.R. Johnson, 1985 Matrix Analysis, Cambridge University Press,
Cambridge, UK.
215. R. Horowitz, W.W. Kao, M. Boals, N. Sadegh, 1989 Digital implementation of
repetitive controllers for robotic manipulators, IEEE International Conference
on Robotics and Automation, Phoenix, AZ.
216. T.C. Hsia, 1986 Adaptive control of robot manipulators: a review, IEEE
International Conference on Robotics and Automation, San Francisco.
217. T. Hu, Z. Lin, 2005 Absolute stability analysis of discrete-time systems with
composite quadratic Lyapunov functions, IEEE transactions on Automatic
Control, vol.50, no 6, pp.781-797, June.
218. T. Hu, B. Huang, Z. lin, 2004 Absolute stability with a generalized sector
condition, IEEE transactions on Automatic Control, vol.59, no 4, pp.535-548,
April.
219. C.H. Huang, P.A. Ioannou, J. Maroulas, and M.G. Safonov, 1999 Design of
strictly positive real systems using constant output feedback, IEEE Transac-
tions on Automatic Control, vol.44, no 3, pp.569-573, March.
220. S.T. Impram, N. Munro, 2004 Absolute stability of nonlinear systems with
disc and norm-bounded perturbations, International Journal of Robust and
Nonlinear Control, vol.14, pp.61-78.
221. S.T. Impram, N. Munro, 2001 A note on absolute stability of uncertain sys-
tems, Automatica, vol.37, pp.605-610.
222. The International Journal of Adaptive Control and Signal Processing, special
issue: Yakov Z. Tsypkin, Memorial issue (S. Bittanti, Ed.), vol.15, no 2, 2001.
223. T. Ionescu, 1970 Hyperstability of linear time varying discrete systems, IEEE
Transactions on Automatic Control, vol.15, pp.645-647.
224. V. Ionescu, C. Oara, 1996 The four block Nehari problem: a generalized
Popov-Yakubovich type approach, IMA Journal of Mathematical Control and
Information, vol.13, pp.173-194.
225. V. Ionescu, M. Weiss, 1993 Continuous and discrete time Riccati theory: a
Popov function approach, Linear Algebra and its Applications, vol.193, pp.173-
209.
552 References

226. P. Ioannou, G. Tao, 1987 Frequency domain conditions for SPR functions,
IEEE Transactions on Automatic Control, vol.32, pp.53-54, January.
227. A. Isidori, 1995 Nonlinear Control Systems, 3rd edition, Springer London (4th
printing, 2002).
228. A. Isidori, 1999 Nonlinear Control Systems II, Springer, London.
229. T. Iwasaki, S. Hara, 2005 Generalized KYP Lemma: unied frequency domain
inequalities with design applications, IEEE transactions on Automatic Control,
vol.50, no 1, pp.41-59, January.
230. T. Iwasaki, G. Meinsma, M. Fu, 2000 Generalized S-procedure and nite
frequency KYP Lemma, Mathematical Problems in Engineering, vol.6, pp.305-
320.
231. B. Jakubczyk, W. Respondek, 1980 On the linearization of control systems,
Bull. Acad. Polon. Sci. Math., vol.28, pp.517-522.
232. M.R. James, 1993 A partial dierential inequality for dissipative nonlinear
systems, Systems and Control Letters, vol.21, pp.315-320.
233. M.R. James, I.R. Petersen, 2005 A nonsmooth strict bounded real lemma,
Systems and Control Letters, vol.54, pp.83-94.
234. D.J. Jeltsema, J.M.A. Sherpen, 2004 Tuning of passivity-preserving con-
trollers for switched-mode power converters, IEEE transactions on Automatic
Control, vol.49, no 8, pp.1333-1344.
235. D.J. Jeltsema, R. Ortega, J.M.A. Sherpen, 2003 On passivity and power bal-
ance inequalities of nonlinear RLC circuits, IEEE transactions on Circuits and
Systems I- Fundamental Theory and Applications, vol.50, no 9, pp.1174-1179,
September.
236. Z.P. Jiang, D.J. Hill, A.L. Fradkov, 1996 A passication approach to adaptive
nonlinear stabilization, Systems and Control Letters, vol.28, pp.73-84.
237. E.A. Johannessen, 1997 Synthesis of Dissipative Output Feedback Controllers,
Ph.D. Dissertation, NTNU Trondheim.
238. E. Johannessen, O. Egeland, 1995 Synthesis of positive real H controller,
Proceedings of the American Control Conference, Seattle, Washington, June,
pp.2437-2438.
239. R. Johansson, 1990 Adaptive control of robot manipulator motion, IEEE
Transactions on Robotics and Automation, vol.6, pp.483-490.
240. R. Johansson, A. Robertsson, R. Lozano, 1999 Stability analysis of adaptive
output feedback control, Proc. of the 38th IEEE Conference on Decision and
Control, Phoenix, Arizona, vol.4, pp.3796-3801, 7-10 December.
241. R. Johansson, A. Robertsson, 2006 The Yakubovich-Kalman-Popov lemma
and stability analysis of dynamic output feedback systems, Int. Journal of
Robust and Nonlinear Control, vol.16, no 2, pp.45-69.
242. R. Johansson, A. Robertsson, 2002 Observer-based strict positive real (SPR)
feedback control system design, Automatica, vol.38, no 9, pp.1557-1564,
September.
243. D.L. Jones and F.J. Evans, 1973 A classication of physical variables and its
application in variational methods, J. of the Franklin Institute, vol.295, no 6,
pp.449-467.
244. U. Jonsson, 1997 Stability analysis with Popov multipliers and integral
quadratic constraints, Systems and Control Letters, vol.31, pp.85-92.
245. S.M. Joshi, S. Gupta, 1996 On a class of marginally stable positive-real sys-
tems, IEEE Transactions on Automatic Control, vol.41, no 1, pp.152-155, Jan-
uary.
 References 553

246. T. Kailath, 1980 Linear Systems, Prentice-Hall.

247. R.E. Kalman, 1963 Lyapunov Functions for the Problem of Lurie in Auto-
matic Control, Proc. Nat. Acad. Sci. U.S.A., vol.49, no 2, pp.201-205.
248. R.E. Kalman, 1964 When is a linear control system optimal?, Trans. ASME
(J. of Basic Engng.), vol.86, series D, pp.51-60, March.
249. O. Kaneko, P. Rapisarda, K. Takada, 2005 Totally dissipative systems, Sys-
tems and Control Letters, vol.54, pp.705-711.
250. V. Kapila, W.M. Haddad, 1996 A multivariable extension of the Tsypkin
criterion using a Lyapunov function approach, IEEE transactions on Automatic
Control, vol.41, no 1, pp.149-152, January.
251. T. Kato, 1970 Accretive operators and nonlinear evolution equations in Ba-
nach spaces, Nonlin. Functional Analysis, Proc. Sympos. Pure Math. 18, Part
1, Chicago 1968, 138-161.
252. H. Kawai, T. Murao, M. Fujita, 2005 Passivity based dynamic visual feed-
back control with uncertainty of camera coordinate frame, Proc. of American
Control Conference, pp.3701-3706, June 8-10, Portland, OR, USA.
253. A. Kelkar, S. Joshi, 1996 Control of Nonlinear Multibody Flexible Space Struc-
tures, LNCIS 221, Springer Verlag, London.
254. R. Kelly, and V. Santibanez, 1998 Global regulation of elastic joints robots
based on energy shaping, IEEE Transactions on Automatic Control, vol.43,
pp.1451-1456.
255. R. Kelly, R. Carelli, 1988 Unied approach to adaptive control of robotic
manipulators, Proceedings of 24th IEEE Conference on Decision and Control,
Austin, TX.
256. H.K. Khalil, 1992 Nonlinear Systems, MacMillan, NY.
257. H. Kimura, 1997 Chain Scattering Approach to H Control, Boston, MA,
Birkhauser.
258. T. Kiyama, S. Hara, T. Iwasaki, 2005 Eectiveness and limitation of circle
criterion for LTI robust control systems with control input nonlinearities of
sector type, International Journal of Robust and Nonlinear Control, vol.15,
pp.873-901.
259. H.W. Knobloch, A. Isidori, D. Flockerzi, 1993 Topics in Control Theory,
Birkhauser, Basel.
260. L. Knockaert, 2005 A note on strict passivity, Systems and Control Letters,
vol.54, no 9, pp.865-869.
261. D.E. Koditschek, 1988 Application of a new Lyapunov function to global
adaptive attitude tracking, Proc. of the 27th IEEE Conference on Decision
and Control, Austin, vol.1, pp.63-68, 7-9 December.
262. E. Kohlberg, J. Mertens, 1986 On the strategic stability of equilibria, Econo-
metrica, vol.54, pp.1003-1039.
263. N.N. Krasovskii, 1959 Stability of Motion, Stanford 1963 (translated from
Nekotorye zadachi ustoichivosti dvzhenia, Moskva, 1959).
264. W.S. Koon, J.E. Marsden, 1997 Poisson reduction for nonholonomic systems
with symmetry, Proc. of the Workshop on Nonholonomic Constraints in Dy-
namics, Calgary, August 26-29.
265. A.M. Krasnoselskii, D.I. Rachinskii, 2000 The Hamiltonian nature of Lure
systems, Automation and Remote Control, vol.61, no 8, pp.1259-1262.
266. A.M. Krasnoselskii, A.V. Pokkrovskii, 2006 Dissipativity of a nonreso-
nant pendulum with ferromagnetic friction, Automation and Remote Control,
vol.67, no 2, pp.221-232.
554 References

267. M. Krstic, I. Kanellakopoulos, P. Kokotovic, 1994 Nonlinear design of adaptive

controllers for linear systems, IEEE Transactions on Automatic Control, vol.39,
pp.752-783.
268. M. Krstic, P. Kokotovic, I. Kanellakopoulos, 1993 Transient performance im-
provement with a new class of adaptive controllers, Systems and Control Let-
ters, vol.21, pp.451-461.
269. A. Kugi, K. Schaler, 2002 Passivitatsbasierte regelung piezoelektrischer struk-
turen, Automatisierungtechnik, vol.50, no 9, pp.422-431.
270. M. Kunze, M.D.P. Monteiro Marques, 2000 An introduction to Moreaus
sweeping process, in Impacts in Mechanical Systems. Analysis and Modelling,
B. Brogliato (Ed.), Springer, Lecture Notes in Physics LNP 551, pp.1-60.
271. C. Lanczos, 1970 The Variational Principles of Mechanics, Dover, NY, 4th
Edition.
272. P. Lancaster and M. Tismenetsky, 1985 The Theory of Matrices, New York,
Academic Press.
273. P. Lancaster, L. Rodman, 1995 Algebraic Riccati Equations, Oxford university
Press.
274. I.D. Landau, 1979 Adaptive Control. The Model Reference Approach, Marcel
Dekker, New York.
275. I.D.Landau, R. Horowitz, 1989 Synthesis of adaptive controllers for robot
manipulators using a passive feedback systems approach, International Journal
of Adaptive Control and Signal Processing, vol.3, pp.23-38.
276. I.D. Landau, 1972 A generalization of the hyperstability conditions for model
reference adaptive systems, IEEE Transactiosn on Automatic Control, vol.17,
pp.246-247.
277. I.D. Landau, 1974 An asymptotic unbiased recursive identier for linear sys-
tems, IEEE Conf. on Decision and Control, Phoenix, Arizona, pp.288-294.
278. I.D. Landau, 1976 Unbiaised recursive identication using model reference
adaptive techniques, IEEE transactions on Automatic Control, vol.21, pp.194-
202, April.
279. P. de Larminat, 1993 Automatique. Commande des systemes lineaires, Hermes,
Paris.
280. M. Larsen, P.V. Kokotovic, 2001 On passivation with dynamic output feed-
back, IEEE transactions on Automatic Control, vol.46, no 6, pp.962-967, June.
281. M. Larsen, P.V. Kokotovic, 2001 A brief look at the Tsypkin criterion: from
analysis to design, International Journal of Adaptive Control and Signal Pro-
cessing, vol.15, no 2, pp.121-128.
282. J. LaSalle, S. Lefschetz, 1961 Stability by Liapunovs Direct Method, Academic
Press, New York, NY.
283. N. Lechevin, C.A. Rabbath, P. Sicard, 2005 A passivity perspective for the
synthesis of robust terminal guidance, IEEE Transactions on Control Systems
Technology, vol.13, no 5, pp.760-765, September.
284. J.H. Lee, C.H. Cho, M. Kim, J.B. Song, 2006 Haptic interface through wave
transformation using delayed reection: application to a passive haptic device,
Adnaved Robotics, vol.20, no 3, pp.305-322.
285. D. Lee, P.Y. Li, 2005 Passive bilateral control and tool dynamics rendering
for nonlinear mechanical teleoperators, IEEE transactions on Robotics, vol.21,
no 5, pp.936-950, October.
 References 555

286. T.C. Lee, Z.P. Jiang, 2005 A generalization of Krasovskii-LaSalle theorem

for nonlinear time-varying systems: converse results and applications, IEEE
transactions on Automatic Control, vol.50, no 8, pp.1147-1163, August.
287. L. Lee, J.L. Chen, 2003 Strictly positive real Lemma and absolute stability for
discrete time descriptor systems, IEEE transactions on Circuits and Systems
I- Fundamental Theory and Applications, vol.50, no 6, pp.788-794, June.
288. P.H. Lee, H. Kimura, Y.C. Soh, 1996 On the lossless and Jlossless embed-
ding theorems in H , Systems and Control Letters, vol.29, pp.1-7.
289. D. Lee, M.W. Spong, 2006 Passive bilateral teleoperation with constant time
delay, IEEE Transactions on Robotics, vol.22, no 2, pp.269-281, April.
290. C.H. Lee, 2006 New upper solution bounds of the continuous algebraic Ric-
cati matrix equation, IEEE Transactions on Automatic Control, vol.51, no 2,
pp.330-169, February.
291. L. Lefevre, 1998 De lIntroduction d Elements Fonctionnels au sein de la
Theorie des Bond Graphs, Ph.D. Ecole Centrale de Lille, France.
292. S. Lefschetz, 1962 Stability of Nonlinear Control Systems, Academic Press, New
York, NY.
293. G.A. Leonov, 2005 Necessary and sucient conditions for the absolute stabil-
ity of two-dimensional time-varying systems, Automation and Remote Control,
vol.66, no 7, pp.1059-1068.
294. P. Libermann, C.M. Marle, 1987 Symplectic Geometry and Analytical Mechan-
ics, Reidel, Dordrecht.
295. X.J. Li, 1963 On the absolute stability of systems with time lags, Chinese
Mathematics, vol.4, pp.609-626.
296. P.Y. Li, R. Horowitz, 2001 Passive velocity eld control (PVFC): Part I
Geometry and robustness, IEEE transactions on Automatic Control, vol.46,
no 9, pp.1346-1359, September.
297. P.Y. Li, R. Horowitz, 2001 Passive velocity eld control (PVFC): Part II
Application to contour following, IEEE transactions on Automatic Control,
vol.46, no 9, pp.1360-1371, September.
298. X.X. Liao, 1993 Absolute Stability of Nonlinear Control Systems, Beijing: Sci-
ence Press.
299. X.X. Liao, P. Yu, 2006 Sucient and necessary conditions for absolute sta-
bility of time-delayed Lurie control systems, Journal of Mathematical Analysis
and Applications, in press.
300. A. L. Likhtarnikov, V.A.Yakubovich, 1977 The frequency theorem for one-
parameter semigroups, (in Russian), Izv. Akad. Nauk SSSR, Ser. Mat., no 5,
pp.1064-1083.
301. A. L. Likhtarnikov, V.A.Yakubovich, 1976 Frequency theorem for evolution
type equations, (Russian) Sib. Mat. Zh., vol.17, pp.1069-1085.
302. S. Lim, J.P. How, 2002 Analysis of linear parameter-varying systems using
a non-smooth dissipative systems framework, International Journal of Robust
and Nonlinear Control, vol.12, pp.1067-1092.
303. W. Lin, 1995 Feedback stabilization of general nonlinear control systems: a
passive system approach, Systems and Control Letters, vol.25, pp.41-52.
304. P.L. Lions, P.E. Souganidis, 1985 Dierential games, optimal control and di-
rectional derivatives of viscosity solutions of Bellmans and Isaacs equations,
SIAM Journal of Control and Optimization, vol.23, pp.566-583.
556 References

305. K.Z. Liu, R. He, 2006 A simple derivation of ARE solutions to the standard
H control problem based on LMI solution, Systems and Control Letters,
vol.55, pp.487-493.
306. L. Ljung, 1977 On positive real transfer functions and the convergence of
some recursive schemes, IEEE Transactions on Automatic Control, vol.22, no
4, pp.539-551, August.
307. H.L. Logemann, R.F. Curtain, 2000 Absolute stability results for well-posed
innite-dimensional systems with applications to low-gain integral control, ES-
IAM: Control, Optimization and Calculus of variations, vol. 5, pp.395-424.
308. J. Loncaric, 1987 Normal form of stiness and compliant matrices, IEEE J.
of Robotics and Automation, vol.3, no 6, pp.567-572.
309. R. Lozano, N. Chopra, M.W. Spong, 2002 Passivation of force reecting bi-
lateral teleoperators with time varying delay, Mechatronics02, Enschede, NL,
June 24-26.
310. R. Lozano, S.M. Joshi, 1990 Strictly positive real functions revisited, IEEE
Transactions on Automatic Control, vol.35, pp.1243-1245, November.
311. R. Lozano, B. Brogliato, 1992 Adaptive control of rst order nonlinear sys-
tem without a priori information on the parameters, IEEE Transactions on
Automatic Control, vol.37, no 1, January.
312. R. Lozano, S. Joshi, 1988 On the design of dissipative LQG type controllers
Proceedings of the 27th IEEE Conference on Decision and Control, Austin,
Texas, pp.1645-1646,7-9 December.
313. R. Lozano, I. Fantoni, 1998 Passivity based control of the inverted pendulum,
IFAC NOLCOS, The Netherlands, July.
314. R. Lozano, I. Fantoni, D.J. Block, 2000 Stabilization of the inverted pendulum
around its homoclinic orbit, Systems and Control Letters, vol.40., no 3, pp197-
204
315. R. Lozano, A. Valera, P. Albertos, S. Arimoto, T. Nakayama, 1999 PD Control
of robot manipulators with joint exibility, actuators dynamics and friction,
Automatica, vol.35, pp.1697-1700.
316. R. Lozano, B. Brogliato, 1991 Adaptive motion control of exible joint ma-
nipulators, American Control Conference, Boston, USA, June.
317. R. Lozano, B. Brogliato, I.D. Landau, 1992 Passivity and global stabiliza-
tion of cascaded nonlinear systems, IEEE Transactions on Automatic Control,
vol.37, no 9, pp.1386-1388.
318. R. Lozano, B. Brogliato, 1992 Adaptive control of robot manipulators with
exible joints, IEEE Transactions on Automatic Control, vol.37, no 2, pp.174-
181.
319. R. Lozano, C. Canudas de Wit, 1990 Passivity-based adaptive control for
mechanical manipulators using LS type estimation, IEEE Transactions on Au-
tomatic Control, vol.35, pp.1363-1365.
320. R. Lozano, B. Brogliato, 1992 Adaptive hybrid force-position control for re-
dundant manipulators, IEEE Transactions on Automatic Control, vol.37, no
10, pp.1501-1505, October.
321. A. Lure, V.N. Postnikov, 1945 On the theory of stability of control systems,
Applied Mathematics and Mechanics, vol.8, no 3, 1944; Prikl. Matem. i, Mekh.,
vol.IX, 5.
322. A.M. Lyapunov, 1907 The General Problem of Motion Stability, in Russian,
1892; translated in French, Ann. Faculte des Sciences de Toulouse, pp.203-474.
 References 557

323. N.H. McClamroch, 1989 A singular perturbation approach to modeling and

control of manipulators constrained by a sti environment, Proc. of the 28th
IEEE Conference on Decision and Control, vol.3, pp.2407-2411, 13-15 December.
324. N.H. McClamroch, D. Wang, 1988 Feedback stabilization and tracking of
constrained robots, IEEE Trans. on Automatic Control, vol.33, no 5, pp.419-
426, May.
325. R. Marino, P. Tomei, 1995 Nonlinear Control Design. Geometric, Adaptive,
Robust, Prentice Hall.
326. C.C.H. Ma, M. Vidyasagar, 1986 Nonpassivity of linear discrete-time sys-
tems, Systems and Control Letters, vol.7, pp.51-53.
327. M. Mabrouk, 1998 A unied variational model for the dynamics of perfect
unilateral constraints, European Journal of Mechanics A/Solids, vol.17, no 5,
pp.819-842.
328. M.S. Mahmoud, L. Xie, 2001 Passivity analysis and synthesis for uncertain
time-delay systems, Mathematical Problems in Engineering, vol.7, pp.455-484.
329. M.S. Mahmoud, L. Xie, 2000 Positive real analysis and synthesis of uncertain
discrete time systems, IEEE transactions on Circuits and Systems, I- Funda-
mental Theory and Applications, vol.47, no 3, pp.403-406, March.
330. M.S. Mahmoud, Y.C. Soh, L. Xie, 1999 Observer-based positive real control
of uncertain linear systems, Automatica, vol.35,pp.749-754.
331. M.S. Mahmoud, 2006 Passivity and passication of jump time-delay systems,
IMA Journal of Mathematical Control and Information, vol.26, pp.193-209.
332. R. Mahony, R. Lozano, 1999 An energy based approach to the regulation
of a model helicopter near to hover, Proceedings of the European Control
Conference, ECC99, Karlsruhe, Germany, September.
333. M. Mahvash, V. Hayward, 2005 High-delity passive force-reecting virtual
environments, IEEE transactions on Robotics, vol.21, no 1, pp.38-46, February.
334. F. Manosas, D. Peralta-Salas, 2006 Note on the Markus-Yamabe conjecture
for gradient dynamical systems, Journal of Mathematical Analysis and Appli-
cations, in press.
335. M. Margaliot, R.Gitizadeh, 2004 The problem of absolute stability: a dynamic
programming appraoch, Automatica, vol.40, pp.1247-1252.
336. M. Margaliot, G. Langholz, 2003 Necessary and sucient conditions for abso-
lute stability: the case of second order systems, IEEE transactions on Circuits
and Systems I, vol.50, no 2, pp.227-234.
337. C.M. Marle, 1997 Various approaches to conservative and nonconsevative non-
holonomic systems, Proc. of the Workshop on Nonholonomic Constraints in
Dynamics, Calgary, August 26-29.
338. M.D.P. Monteiro Marques, 1993 Dierential Inclusions in Nonsmooth Mechan-
ical Problems. Shocks and Dry Friction, Birkhauser, Progress in Nonlinear Dif-
ferential Equations and Their Applications, Basel-Boston-Berlin.
339. H.J. Marquez, C.J. Damaren, 1995 On the design of strictly positive real
transfer functions, IEEE transactions on Circuits and Systems I- Fundamental
Theory and Applications, vol.42, no 4, pp.214-218, April.
340. H.J. Marquez, P. Agathoklis, 2001 Comments on Hurwitz polynomials and
strictly positive real transfer functions , IEEE transactions on Circuits and
Systems I- Fundamental Theory and Applications, vol.48, no 1, p.129, January.
341. H.J. Marquez, P. Agathokis, 1998 On the existence of robust strictly positive
real rational functions, IEEE Transactions on Circuits and Systems I, vol.45,
pp.962-967, September.
558 References

342. B.M. Maschke, A.J. van der Schaft, 1992 Port controlled Hamiltonian systems:
modeling origins and system theoretic properties, Proc. 2nd IFAC Symp. on
Nonlinear Control Systems design, NOLCOS92, pp.282-288, Bordeaux, June.
343. B.M. Maschke, A. van der Schaft, P.C. Breedveld, 1992 An intrinsic Hamil-
tonian formulation of network dynamics: nonstandard Poisson structures and
gyrators, Journal of the Franklin Institute, vol.329, no 5, pp.923-966.
344. B.M. Maschke, A.J. van der Schaft, P.C. Breedveld, 1995 An intrinsic Hamil-
tonian formulation of the dynamics of LC-circuits, Trans. IEEE on Circuits
and Systems, I: Fundamental Theory and Applications, vol.42, no 2, pp.73-82,
February.
345. B.M. Maschke, 1996 Elements on the modelling of multibody systems,
Modelling and Control of Mechanisms and Robots, pp.1-38, C.Melchiorri and
A.Tornambe (Eds.), World Scientic Publishing Ltd.
346. B.M. Maschke, A.J. van der Schaft, 1997 Interconnected Mechanical Sys-
tems Part I: Geometry of interconnection and implicit Hamiltonian systems,
in Modelling and Control of Mechanical Systems, A.Astol, C.Melchiorri and
A.Tornambe (Eds.), pp.1-16, Imperial College Press.
347. B.M. Maschke, A.J. van der Schaft, 1997 Interconnected Mechanical Systems.
Part II: The dynamics of spatial mechanical networks, Modelling and Control of
Mechanical Systems, A.Astol, C.Melchiorri and A.Tornambe (Eds.), pp.17-30,
Imperial College Press.
348. I. Masubuchi, 2006 Dissipativity inequalities for continuous-time descriptor
systems with applications to synthesis of control gains, Systems and Control
Letters, vol.55, pp.158-164. .
349. A.V. Megretskii, V.A. Yakubovich, 1990 A singular linear-quadratic optimiza-
tion problem, Proc. Leningrad Math. Society, vol.1, pp.134-174.
350. G. Meisters, 1996 A biography of the Markus-Yamabe conjecture, available
at http://www.math.unl.edu/ gmeister/Welcome.html , expanded form of a talk
given at the conference Aspects of Mathematics Algebra, Geometry and Sev-
eral Complex Variables, June 10-13, the Universiy of Hong-Kong.
351. Y. Merkin, 1997 Introduction to the Theory of Stability, Springer Verlag, TAM
24.
352. W. Messner, R. Horowitz, W.W. Kao, M. Boals, 1989 A new adaptive learning
rule, SIAM conference in the Nineties and IEEE International Conference on
Robotics and Automation, Cincinnati, Ohio, 1990.
353. K.R. Meyer, 1965 On the existence of Lyapunov functions for the problem of
Lure, SIAM Journal of Control, vol.3, pp.373-383, August.
354. A. Meyer Base, 1999 Asymptotic hyperstability of a class of neural networks,
Int. J. Neural Syst., vol.9, no 2, pp.95-98, April.
355. J. De Miras, A. Charara, 1998 A vector oriented control for a magnetically
levitated shaft, IEEE Trans. on Magnetics, vol.34, no 4, pp.2039-2041.
356. J. De Miras, A. Charara, 1999 Vector desired trajectories for high rotor speed
magnetic bearing stabilization, IFAC99, 14th World Congress, July, China.
357. M.D.P. Monteiro-Marques, 1993 Dierential Inclusions in Nonsmooth Mechan-
ical Problems: Shocks and Dry Friction, Birkhauser, Boston, PNLDE 9.
358. J.J. Moreau, 1988 Unilateral contact and dry friction in nite freedom dy-
namic, CISM Courses and Lectures no 302, International Centre for Mechan-
ical Sciences, J.J. Moreau and P.D. Panagiotopoulos (Eds.), Springer-Verlag,
pp.1-82.
 References 559

359. J.J. Moreau, 2003 Fonctionnelles Convexes Istituto Poligraco e Zecca dello
Stato S.p.A., Roma, Italy (preprint Seminaire sur les Equations aux Derivees
Partielles, Paris, France, College de France, 1966-1967).
360. J.J. Moreau, M. Valadier, 1986 A chain rule involving vector functions of
bounded variation, J. Funct. Analysis, vol.74, pp.333-345.
361. A.S. Morse, 1992 High-order parameter tuners for the adaptive control of
linear and nonlinear systems, Proc. of the US-Italy joint seminar Systems,
models and feedback: theory and application, Capri, Italy.
362. C. Mosquera, F. Perez, 2001 On the strengthened robust SPR problem for
discrete time systems, Automatica, vol.37, no 4, pp.625-628, April.
363. P.J. Moylan, B.D.O. Anderson, 1973 Nonlinear regulator theory and an in-
verse optimal control problem, IEEE Transactions on Automatic Control,
vol.18, pp.460-465.
364. P.J. Moylan, D.J. Hill, 1978 Stability criteria for large-scale systems, IEEE
Transactions on Automatic Control, vol.23, no 2, pp.143-149.
365. P.J. Moylan, 1974 Implications of passivity in a class of nonlinear systems,
IEEE Transactions on Automatic Control, vol.19, non 4, pp.373-381, August.
366. R.M. Murray, Z. Li, S.S. Sastry, 1994 A Mathematical Introduction to Robotic
Manipulation, CRC Press, Boca Raton, Florida.
367. K.G. Murty, 1997 Linear Complementarity, Linear and Nonlinear
Programming, available at http://www-personal.engin.umich.edu/
murty/book/LCPbook/
368. M. Namvar, F. Aghili, 2005 Adaptive force-motion control of coordinated
robots interacting with geometrically unknown environments, IEEE Transac-
tions on Robotics, vol.21, no 4, pp.678-694, August.
369. K.S. Narendra and J.H. Taylor, 1973 Frequency Domain Criteria for Absolute
Stability, Academic Press.
370. K.S. Narendra, A. Annaswamy, 1989 Stable Adaptive Systems, Prentice Hall.
371. E.M. Navarro Lopez, 2002 Dissipativity and Passivity-related Properties in
Nonlinear Discrete-time Systems, PhD Thesis, Universidad Politecnica de
Cataluna, Instituto de Organizacion y Control de Sistemas Industriales, Spain,
May 2002.
372. E.M. Navarro Lopez, H. Sira-Ramirez, E. Fossas-Colet, 2002 Dissipativity and
feedback dissipativity properties of general nonlinear discrete-time systems,
European Journal of Control, vol.8, no 3, pp.265-274.
373. E.M. Navarro Lopez, 2005 Several dissipativity and passivity implications in
the linear discrete-time setting, Mathematical Problems in Engineering, vol.6,
pp.599-616.
374. E.M. Navarro Lopez, E. Fossas-Colet, 2004 Feedback passivity of nonlinear
discrete-time systems with direct input-output link, Automatica, vol.40, no 8,
pp.1423-1428.
375. A.W. Naylor, G.R. Sell, 1983 Linear Operator Theory in Engineering and Sci-
ence, New York, Springer Verlag.
376. R.W. Newcomb, 1966 Linear Multiport Synthesis, McGraw-Hill, New York.
377. S.I. Niculescu, R. Lozano, 2001 On the passivity of linear delay systems,
IEEE Transactions on Automatic Control, vol.46, no 3, pp.460-464, March.
378. S.I. Niculescu, 1997 Systemes a Retard: Aspects Qualitatifs sur la Stabilite et
la Stabilisation, Diderot Editeur, Arts et Sciences, Paris.
379. S.I. Niculescu, 2001 Delay Eects on Stability: A Robust Control Approach,
Springer Lecture Notes in Control and Information Sciences, vol.269, London.
560 References

380. S.I. Niculescu, E. I. Verriest, L. Dugard, J. M. Dion, 1997 Stability and robust
stability of time-delay systems: A guided tour, in Stability and Control of Time-
Delay Systems (L. Dugard and E. I. Verriest, Eds.), LNCIS 228, Springer-Verlag,
London, pp.1-71.
381. H. Nijmeier, A.J. van der Schaft, 1990 Nonlinear Dynamical Control Systems,
Springer Verlag, New-York.
382. M.C. de Oliveira, J.C. Geromel, L. Hsu, 2002 A new absolute stability test for
systems with state dependent perturbations, International Journal of Robust
and Nonlinear Control, vol.12, pp.1209-1226.
383. R. Ortega, 1993 On Morses new adaptive controller: parameter convergence
and transient performance, IEEE Transactions on Automatic Control, vol.38,
pp.1191-1202.
384. R.Ortega, M. Spong, 1989 Adaptive motion control of rigid robots: a tutorial,
Automatica, vol.25, pp.877-888.
385. R. Ortega, G. Espinosa, 1993 Torque regulation of induction motors, Auto-
matica, vol.29, pp.621-633.
386. R. Ortega, G. Espinosa, 1991 A controller design methodology for systems
with physical structures: application to induction motors, Proceedings of the
30th IEEE Conference on Decision and Control, Brighton, December 1991,
pp.23454-2349.
387. R. Ortega, A.J. van der Schaft, B. Maschke, G. Escobar, 2002 Interconnection
and damping assignment passivity-based control of port-controlled Hamiltonian
systems, Automatica, vol.38, no 4, pp.585-596.
388. D.H. Owens, D. Pratzel-Wolters, A. Ilchman, 1987 Positive-real structure and
high-gain adaptive stabilization, IMA Journal of Mathematical Control and
Information, vol.4, pp.167-181.
389. B. Paden, R. Panja, 1988 Globally asymptotically stable PD+ controller for
robot manipulators, Int. J. of Control, vol.47, pp.1697-1712.
390. L. Pandol, 2001 An observation on the positive real lemma, Journal of
Mathematical Analysis and Applications, vol.255, pp.480-490.
391. L. Pandol, 2001 Factorization of the Popov function of a multivariable linear
distributed parameter system in the non-coercive case: a penalization approach,
Int. J. Appl. Math. Comput. Sci., vol.11, no 6, pp.1249-1260.
392. L. Paoli, M. Schatzman, 1993 Mouvement a un nombre ni de degres de liberte
avec contraintes unilaterales: cas avec perte denergie, Mathematical Modelling
and Numerical Analysis (Modelisation Mathematique et Analyse Numerique),
vol.27, no 6, pp.673-717.
393. T. Pare, A. Hassibi, J. How, 2001 A KYP lemma and invariance principle
for systems with multiple hysteresis non-linearities, International Journal of
Control, vol.74, no 11, pp.1140-1157.
394. P.C. Parks, 1966 Lyapunov redesigns of model reference adaptive control sys-
tems, IEEE Transactions on Automatic Control, vol.11, pp.362-367.
395. S. Partovi, N.E. Nahi, 1969 Absolute stability of dynamic system containing
non-linear functions of several state variables, Automatica, vol.5, pp.465-473.
396. V.V. Patel, K.B. Datta, 2001 Comments on Hurwitz stable polynomials and
strictly positive real transfer functions , IEEE transactions on Circuits and
Systems I- Fundamental Theory and Applications, vol.48, no 1, pp.128-129,,
January.
 References 561

397. L. Pavel, W. Fairman, 1997 Nonlinear H control: a Jdissipative approach,

IEEE transactions on Automatic Control, vol.42, no 12, pp.1636-1653, Decem-
ber.
398. H.M. Paynter, 1961 Analysis and Design of Engineering Systems, M.I.T. Press,
Cambridge, MA, 1961.
399. A. Pazy, 1972 On the applicability of Lyapunovs theorem in Hilbert space,
SIAM J. Math. Anal., vol.3, pp.291-294
400. I.G. Polushin, A.L. Fradkov, D.J. Hill, 2000 Passivity and passication of
nonlinear systems, Automation and Remote Control, vol.61, no 3, pp.355-388.
401. I.G. Polushin, H.J. Marquez, 2002 On the existence of a continuous storage
function for dissipative systems, Systems and Control Letters, vol.46, pp.85-90.
402. I.G. Polushin, H.J. Marquez, 2004 Conditions for the existence of continu-
ous storage functions for nonlinear dissipative systems, Systems and Control
Letters, vol.54, pp.73-81.
403. I.G. Polushin, H.J. Marquez, 2004 Boundedness properties of nonlinear quasi-
dissipative systems, IEEE transactions on Automatic Control, vol.49, no 12,
pp.2257-2261, December.
404. H.R. Pota, P.J. Moylan, 1993 Stability of locally dissipative interconnected
systems, IEEE Transactions on Automatic Control, vol.38, no 2, pp.308-312.
405. V.M. Popov, 1964 Hyperstability of automatic systems with several nonlinear
elements, Revue Roumaine des Sciences et Techniques, serie electrotech. et
energ., vol.9, no 1, pp.35-45.
406. V.M. Popov, 1973 Hyperstability of Control Systems, Berlin, Springer-Verlag.
407. V.M. Popov, 1959 Criteres de stabilite pour les systemes non lineaires de
reglage automatique, bases sur lutilisation de la transformee de Laplace, (in
Romanian), St. Cerc. Energ., IX, no 1, pp.119-136.
408. V.M. Popov, 1959 Criteres susants de stabilite asymptotique globale pour
les systemes automatiques non lineaires a plusieurs organes dexecution, (in
Romanian), St. Cerc. Energ., IX, no 4, pp.647-680.
409. V.M. Popov, 1964 Hyperstability and optimality of automatic systems with
several control functions, Rev. Roum. Sci. Techn., Ser. Electrotechn. et Energ.,
vol.9, no 4, pp.629-690.
410. V.M. Popov, 1961 Absolute stability of nonlinear systems of automatic con-
trol, Avt. i Telemekh., vol.22, pp.961-979 (in Russian).
411. V.M. Popov, A. Halanay, 1962 About stability of non-linear controlled systems
with delay, Automation and Remote Control, vol.23, pp.849-851.
412. A. Rantzer, 1996, On the Kalman-Yakubovich-Popov Lemma, Systems and
Control Letters, vol 28, pp7-10.
413. V. Rasvan, S. Niculescu, R. Lozano, 2000,Delay systems: passivity, dissipa-
tivity and hyperstability, Tech. Report Heudiasyc-UTC.
414. J. Reyes-Reyes, A.S. Poznyak, 2000 Passivation and control of partially known
SISO nonlinear systems via dynamical neural networks, Mathematical Prob-
lems in Engineering, vol.6, pp.61-83.
415. R.T. Rockafellar, R.J.B. Wets, 1998 Variational Analysis, Springer,
Grundlehren der Mathematischen Wissenschaften, vol.317.
416. L. Rodman, 1997 Non-Hermitian solutions of algebraic Riccati equations,
Canadian Journal of Mathematics, vol.49, no 4, pp.840-854.
417. H.H. Rosenbrock, 1973 Multivariable circle theorems, Recent Math. Devel-
opm.Control, Proc. Univ. Bath, Somerset 1972, pp.345-365.
562 References

418. L. Rosier, E.D. Sontag, 2000 Remarks regarding the gap between continu-
ous, Lipschitz, and dierentiable storage functions for dissipation inequalities
appearing in H control, Systems and Control Letters, vol.41, pp.237-249.
419. W. Rudin, 1998 Analyse Reelle et Complexe, Dunod, Paris.
420. W. Rudin, 1976, Principles of Mathematical Analysis, McGraw Hill, 3rd Edi-
tion.
421. W. Rudin, 1987 Real and Complex Analysis, McGraw Hill series in Higher
Maths, 3rd edition.
422. J.H. Ryu, C. Preusche, B. Hannaford, G. Hirzinger, 2005 Time domain pas-
sivity control with reference energy following, IEEE transactions on Control
Systems Technology, vol.13, no 5, pp.737-742, September.
423. J.H. Ryu, B. Hannaford, D.S. Kwon, J.H. Kim, 2005 A simula-
tion/experimental study of the noisy behaviour of the time domain passivity
controller, IEEE Transactions on Automatic Control, vol.21, no 4, pp.733-741,
August.
424. A. Saberi, P. Sannuti, 1987 Cheap and singular controls for linear quadratic
regulators, IEEE Transactions on Automatic Control, vol.32, no 3, pp.208-219,
March.
425. N. Sadegh, R. Horowitz, 1987 Stability analysis of adaptive controller for
robotic manipulators, IEEE Int. Conference on Robotics and Automation,
Raleigh, USA, 1987.
426. N. Sadegh, R. Horowitz, 1990 Stability and robustness analysis of a class
of adaptive controllers for robotic manipulators, Int. J. of Robotics Research,
vol.9, no 3, pp.74-92.
427. N. Sadegh, R. Horowitz, W.W. Kao, M. Tomizuka, 1988 A unied approach
to design of adaptive and repetitive controllers fr robotic manipulators, USA-
Japan symposium on Flexible Automation, Minneapolis, MN.
428. M.G. Safonov, E.A. Jonckeere, M. Verma, and D.J.N. Limebeer, 1987 Syn-
thesis od positive real multivariable feedback systems, Int. J. Control, vol.45,
pp.817-842.
429. M.K. Sain, J.L. Massey, 1969 Invertibility of linear time-invariant dynamical
systems, IEEE Transactions on Automatic Control, vol.14, no 2, pp.141-149,
April.
430. N. Sakamoto, M. Suzuki, 1996 -passive system and its phase property and
synthesis, IEEE Transactions on Automatic Control, vol.41, no 6, pp.859-865,
June.
431. I.W. Sandberg, 1964, A frequency domain criterion for the stability of feed-
back systems containing a single time varying non linear element, Bell Syst.
tech. J., vol.43, pp.1901-1908.
432. P. Sannuti, 1983 Direct singular perturbation analysis of high-gain and cheap
control problems, Automatica, vol.19, no 1, pp.41-51.
433. P. Sannuti, A. Saberi, 1987 A special coordinate basis of multivariable lin-
ear systems, nite and innite zero structure, squaring down and decoupling,
International Journal of Control, vol.45, no 5, pp.1655-1704, May.
434. P. Sannuti, H.S. Wason, 1985 Multiple time-scale decomposition in cheap
control problems Singular control, IEEE Transactions on Automatic Control,
vol.30, no 7, pp.633-644, July,
435. G.L. Santosuosso, 1997 Passivity of nonlinear systems with input-output
feedthrough, Automatica, vol.33, no 4, pp.693-697.
 References 563

436. S.S. Sastry, 1984 Model reference adaptive control- stability, parameter con-
vergence and robustness, IMA J. Math. Control Info., vol.1, pp.27-66.
437. A.J. van der Schaft, 1984 System Theoretical description of Physical Systems,
CWI Tracts 3, CWI Amsterdam, The Netherlands.
438. A.J. van der Schaft, 1987 Equations of motion for Hamiltonian systems with
constraints, J. Phys. A: Math. Gen., vol.20, pp.3271-3277.
439. A.J. van der Schaft, 1989 System theory and mechanics, in Three Decades of
Mathematical System Theory, H. Nijmeier and J.M. Schumacher (Eds.), LNCIS
135, Springer, London.
440. A.J. van der Schaft, B.M. Maschke, 1994 On the Hamiltonian formulation of
non-holonomic mechanical systems, Reports on Mathematical Physics, vol.34,
no 2, pp.225-233.
441. A.J. van der Schaft, B.M. Maschke, 1995 The Hamiltonian formulation of
energy conserving physical systems with ports, Archiv fur Elektronik und
Ubertragungstechnik, Vol.49, 5/6, pp.362-371.
442. A.J. van der Schaft, 2000 L2-gain and Passivity Techniques in Nonlinear Con-
trol, 2nd edition, Springer, London, CCES.
443. C. Scherer, 1992 H control by state feedback for plants with zeros on the
imaginary axis, SIAM Journal on Control and Optimization, vol.30, pp.123-
142.
444. R. Scherer, W. Wendler, 1994 A generalization of the positive real Lemma,
IEEE transactions on Automatic Control, vol.39, no 4, pp.882-886, April.
445. R. Scherer, H. Turke, 1989 Algebraic characterization of Astable Runge-
Kutta methods, Appl. Numer. Math., vol.5, pp.133-144.
446. G. Schmitt, 1999 Frequency domain evaluation of circle criterion, Popov cri-
terion and o-axis circle criterion in the MIMO case, International Journal of
Control, vol.72, no 14, pp.1299-1309.
447. M. de la Sen, 2002 Preserving positive realness through discretization, Pos-
itivity, vol.6, pp.31-45.
448. M. de la Sen, 1998 A method for general design of positive real functions,
IEEE transactions on Circuits and Systems I- Fundamental Theory and Appli-
cations, vol.45, no 7, pp.764-769, July.
449. M. de la Sen, 1997 A result on the hyperstability of a class of hybrid dynamic
systems, IEEE transactions on Automatic Control, vol.42, no 9, pp.1335-1339,
September.
450. M. de la Sen, J. Jugo, 1998 Absolute stability and hyperstability of a class of
hereditary systems, Informatica, vol.9, no 2, pp.195-208.
451. M. de la Sen, 2005 Some conceptual links between dynamic physical systems
and operator theory issues concerning energy balances and stability, Informat-
ica, vol.16, no 3, pp.395-406.
452. M. de la Sen, 2006 On positivity and stability of a class of time-delay systems,
Nonlinear Analysis, Real World Applications, in press.
453. M.M. Seron, D.J. Hill, A.L. Fradkov, 1995 Nonlinear adaptive control of feed-
back passive systems, Automatica, vol.31, no 7, pp.1053-1060.
454. X. Shen, M. Goldfarb, 2006 On the enhanced passivity of pneumatically actu-
ated impedance-type haptic interfaces, IEEE Transactions on Robotics, vol.22,
no 3, pp.470-480, June.
455. R. Shorten, C. King, 2004 Spectral conditions for positive realness of single-
input single-output systems, IEEE transactions on Automatic Control, vol.49,
no 10, pp.1875-1879, October.
564 References

456. D. Siljak, 1969 Parameter analysis of absolute stability, Automatica, vol.5,

pp.385-387.
457. H. Sira-Ramirez, 2000 Passivity versus atness in the regulation of an exother-
mic chemical reactor, European Journal of Control, vol.6, no 3, pp.1-17.
458. H. Sira-Ramirez, R. Ortega, M. Garcia-Esteban, 1997 Adaptive passivity-
based control of average DC to DC power converters models, International
Journal of Adaptive Control and Signal Processing, vol.11, pp.489-499.
459. H. Sira-Ramirez, M.I. Angulo-Nunez, 1997 Passivity based control of nonlin-
ear chemical processes, International Journal of Control, vol.68, no 5, pp.971-
996.
460. H. Sira-Ramirez, R.A. Perez Moreno, R. Ortega, M. Garcia Esteban, 1997
Passivity-based controllers for the stabilization of DC to DC power converters,
Automatica, vol.33, no 4, pp.499-513.
461. J.J. Slotine, W. Li, 1988 Adaptive manipulator control: A case study, IEEE
Transactions on Automatic Control, vol.33, pp.995-1003.
462. J.J.E. Slotine, W. Li, 1989 Composite adaptive control of robot manipula-
tors, Automatica, vol.25, pp.509-520.
463. J.J.E. Slotine, W. Li, 1987 On the adaptive control of robot manipulators,
International Journal of Robotics Research, vol.6, pp.49-59.
464. A. Somolines, 1977 Stability of Lurie type functional equations, Journal of
Dierential Equations, vol.26, pp.191-199.
465. Y.I. Son, H. Shim, N.H. Jo, J.H. Seo, 2003 Further results on passication
of non-square linear systems using an input-dimensional compensator, IEICE
Trans. Fundamentals, vol.E86-A, no 8, pp.2139-2143, August.
466. E.D. Sontag, 2006 Passivity gains and the secant condition for stability,
Systems and Control Letters, vol.55, no 3, pp.177-183.
467. E. Sontag, 1998 Mathematical Control Theory: Deterministic Finite Dimen-
sional Systems. Springer-Verlag, New York 1990. Second Edition 1998.
468. E.D. Sontag, 1995 On the input-to-state stability property, European Jour-
nal of Control, vol.1, pp.24-36.
469. E.D. Sontag, 2005 Input to state stability: Basic concepts and results,
Springer Lecture notes in Mathematics (CIME Course, Cetraro, June 2004).
470. C.E. de Souza, L. Xie, 192 On the discrete-time bounded real Lemma with
application in the characterization of static state feedback H controllers,
Systems and Control Letters, vol.18, no 1, pp.61-71,January.
471. M.W. Spong, 1987 Modeling and control of elastic joint robots, ASME J. of
Dyn. Syst. Meas. and Control, vol.109, pp.310-319.
472. M.W. Spong, R. Ortega, R. Kelly, 1990 Comments on Adaptive manipula-
tor control: A case study, IEEE Transactions on Automatic Control, vol.35,
pp.761-762.
473. M.W. Spong, M. Vidyasagar, 1989 Robot Dynamics and Control, Wiley, New-
York.
474. M.W. Spong, 1994 The swing up control of the Acrobot, Proc. IEEE Int.
Conf. on Robotics and Automation, pp.616-621, San Diego, CA, 8-13 May.
475. M.W. Spong, 1989 Adaptive control of exible joint manipulators, Systems
and Control Letters, vol.13, pp.15-21.
476. M. W. Spong, 1995 Adaptive control of exible joint manipulators: comments
on two papers, Automatica, vol. 31, no 4, pp. 585-590.
477. M.W. Spong, R. Ortega, 1990 On adaptive inverse dynamics control of rigid
robots, IEEE transactions on Automatic Control, vol.35, ppp.92-95.
 References 565

478. O.J. Staans, 2001 Jpreserving well-posed linear systems, Int. J. Appl.
Math. Comput. Sci., vol.11, no 6, pp.1361-1378.
479. D.M. Stipanovic, D.D. Siljak, 2001 SPR criteria for uncertain rational matri-
ces via polynomial positivity and Bernsteins expansions, IEEE transactions
on Circuits and Systems I- Fundamental Theory and Applications, vol.48, no
11, pp.1366-1369, November.
480. W. Sun, P.P. Khargonekar, D. Shim, 1994 Solution to the positive real control
problem for linear time-invariant systems, IEEE Transactions on Automatic
Control, vol.39, pp.2034-2046, October.
481. X.M. Sun, J. Zhao, D.J. Hill, 2006 Stability and L2 -gain analysis for switched
delay systems: A delay-dependent method, Automatica, in press.
482. A. Szatkowski, 1979 Remark on Explicit topological formulation of La-
grangian and Hamiltonian equations for nonlinear networks, IEEE Trans. on
Circuits and Systems, vol.26, no 5, pp.358-360.
483. G. Szego, R.E.Kalman, 1963 Sur la stabilite absolue dun systeme dequations
aux dierences nies, C. R. Acad. Sci. Paris, vol.257, no 2, pp.388-390.
484. M. Takegaki, S. Arimoto, 1981 A new feedback method for dynamic control
of manipulators, ASME J. Dyn. Syst. Meas. Control, vol.102, pp.119-125.
485. G. Tao, P. Ioannou, 1988 Strictly positive real matrices and the Lefschetz-
Kalman-Yakubovich Lemma, IEEE Transactions on Automatic Control, vol.33,
pp.1183-1185, December.
486. J.H. Taylor, 1974 Strictly positive real functions and Lefschetz-Kalman-
Yakubovich (LKY) lemma, IEEE Transactions on Circuits Systems, pp.310-
311, March.
487. M.A.L. Thathachar, M.D. Srinath, 1967 Some aspects of the Lure problem,
IEEE transactions on Automatic Control, vol.12, no 4, pp.451-453.
488. P. Tomei, 1991 A simple PD controller for robots with elastic joints, IEEE
Transactions on Automatic Control, vol.36, pp.1208-1213.
489. H.L. Trentelman, J.C. Willems, 1997 Every storage function is a state func-
tion, Systems and Control Letters, vol.32, pp.249-259.
490. H.L. Trentelman, 1998 When does the algebraic Riccati equation have a neg-
ative semi-denite solution?, in V.D. Blondel, E.D. Sontag, M. Vidyasagar,
J.C. Willems (Editors), Open Problems in Mathematical Systems and Control
Theory, pp.229-237, Springer.
491. H.L. Trentelman, J.C. Willems, 2000 Dissipative dierential systems and the
state space H control problem, International Journal of Robust and Nonlinear
Control, vol.10, pp.1039-1057.
492. Y.Z. Tsypkin, 1964 A criterion for absolute stability of automatic pulse sys-
tems with monotonic characteristics of the nonlinear element, Sov. Phys. Dok-
lady, vol.9, pp.263-366.
493. Y.Z. Tsypkin, 1962 The absolute stability of large scale, nonlinear sampled
data systems, Doklady Akademii Nauk. SSSR, vol.145, pp.52-55.
494. Y.Z. Tsypkin, 1963 Fundamentals of the theory of nonlinear pulse control
systems, Proceedings of the second IFAC congress, Balse, CH, pp.172-180.
495. Y.Z. Tsypkin, 1964 Absolute stability of equilibrium positions and of re-
sponses in nonlinear, sampled data, automatic systems, Automation and Re-
mote Control, vol.24, no 12, pp.1457-1471.
496. Y.Z. Tsypkin, 1964 Frequency criteria for the absolute stability of nonlinear
sampled data systems, Automation and Remote Control, vol.25, no 3, pp.261-
267.
566 References

497. L. Turan, M.G. Safonov, C.H. Huang, 1997 Synthesis of positive real feed-
back systems: a simple derivation via Parotts Theorem, IEEE Transactions
on Automatic Control, vol.42, no 8, pp.1154-1157, August.
498. L. Vandenberghe, V.R. Balakrishnan, R. Wallin, A. Hansson, T. Roh, 2005 In-
terior point algorithms for semidenite programming problems derived from the
KYP Lemma, in Positive Polynomials in Control, A. Garulli and D. Henrion
(Eds.), Springer LNCIS 312, pp.195-238.
499. M. Vidyasagar, 1981 Input-Output Analysis of Large-Scale Interconnected Sys-
tems, LNCIS, Springer-Verlag, London.
500. M. Vidyasagar, 1993 Nonlinear Systems Analysis, 2nd Edition, Prentice Hall.
501. H.S. Wang, C.F. Yung, F.R. Chang, 2006 A generalized algebraic Riccati
equation, Lecture Notes in Control and Information Sciences 326, Control for
Nonlinear Descriptor Systems, pp.141-148.
502. C.J. Wan, D.S. Bernstein, 1995 Nonlinear feedback control with global stabi-
lization, Dynamics and Control, vol.5, no 4, pp.321-346.
503. R. Wang, 2002 Algebraic criteria for absolute stability, Systems and Control
Letters, vol.47, pp.401-416.
504. L. Wang, W. Yu, 2001 On Hurwitz stable polynomials and strictly positive real
transfer functions, IEEE transactions on Circuits and Systems I- Fundamental
Theory and Applications, vol.48, no 1, pp.127-128, January.
505. Q. Wang, H. Weiss, J.L. Speyer, 1994 System characterization of positive
real conditions, IEEE Transactions on Automatic Control, vol.39, pp.540-544,
March.
506. H. Weiss, Q. Wang, J.L. Speyer, 1994 System characterization of positive real
conditions, IEEE transactions on Automatic Control, vol.39, no 3, pp.540-544,
March.
507. M. Weiss, 1997 Riccati equation theory for Pritchard-Slamon systems: a
Popov function approach, IMA Journal of Mathematical Control and Infor-
mation, vol.14, pp.45-83.
508. J.T. Wen, 1988 Time domain and frequency domain conditions for strict pos-
itive realness, IEEE Transactions on Automatic Control, vol.33, pp.988-992 ,
November.
509. J.T. Wen, 1989 Finite dimensional controller design for innite dimensional
systems: the circle criterion approach, Systems and Control Letters, vol.13,
pp.445-454.
510. J.C. Willems, 1972 Dissipative dynamical systems, Part I: General Theory,
Arch. Rat. Mech. An., vol.45, pp.321-351.
511. J.C. Willems, 1972 Dissipative dynamical systems, Part II: Linear Systems
with quadratic supply rates, Arch. Rat. Mech. An., vol.45, pp.352-393.
512. J.C. Willems, 1971 The generation of Lyapunov functions for input-output
stable systems, SIAM J. Control, vol.9, pp.105-133, February.
513. J.C. Willems, 1971 Least squares stationary optimal control and the alge-
braic Riccati Equation, IEEE transactions on Automatic Control, vol.16, no 6,
pp.621-634, December.
514. J.C. Willems, 1974 On the existence of a nonpositive solution to the Ric-
cati equation, IEEE Transactions on Automatic Control, vol.19, pp.592-593,
October.
515. L. Xie, Y.C. Soh, 1995 Positive real control problem for uncertain linear time-
invariant systems, Systems and Control Letters, vol.24, pp.265-271.
 References 567

516. S. Xu, J. Lam, Z. Lin, K. Galkowski, 2002 Positive real control for uncer-
tain two-dimensional systems, IEEE transactions on Circuits and Systems, I-
Fundamental Theory and Applications, vol.49, no 11, pp.1659-1666, November.
517. S. Xu, J. Lam, 2004 New positive realness conditions for uncertain discrete
descriptor systems: analysis and synthesis, IEEE transactions on Circuits and
Systems, I- Fundamental Theory and Applications, vol.51, no 9, pp.1897-1905,
September.
518. V.A. Yakubovich, 1962 La solution de quelques inegalites matricielles ren-
contrees dans la theorie du reglage automatique, (in Russian), Doklady A.N.
SSSR, t.143, no 6, pp.1304-1307.
519. V.A. Yakubovich, 1962 The solution of certain matrix inequalities, Automat.
Control Theory Sov. Math. AMS, vol.3, pp.620-623.
520. V.A. Yakubovich, 1975 The frequency theorem for the case in which the state
space and the control space are Hilbert spaces, and its application incertain
problems in the synthesis of optimal control, II,Sib. Math. J. 16(1975), pp.828-
845 (1976).
521. V.A. Yakubovich, 1974 The frequency theorem for the case in which the state
space and the control space are Hilbert spaces, and its application incertain
problems in the synthesis of optimal control, I, Sib. Math. J. 15(1974), 457-476
(1975).
522. V.A. Yakubovich, 1962 Frequency conditions for the absolute stability of non-
linear automatic control systems, in Proceedings of Intercollegiate Conference
on the Applications of Stability Theory and Analytic Mechanics (Kazan, 1962),
Kazan Aviats. Inst., pp.123-134.
523. V.A. Yakubovich, 1966 Periodic and almost periodic limit modes of controlled
systems with several, in general discontinuous, nonlinearities, Soviet. Math.
Dokl., vol.7, no 6, pp.1517-1521.
524. V.A. Yakubovich, G.A. Leonov, A.K. Gelig, 2004 Stability of Stationary Sets in
Control Systems with Discontinuous Nonlinearities, World Scientic, NJ, series
on Stability, Vibration and Control of Systems, series A, vol.14.
525. B.E. Ydstie, A.A. Alonso, 1997 Process systems and passivity via the
Clausius-Planck inequality, Systems and Control Letters, vol.30, pp.253-264.
526. B.E. Ydstie, Y. Jiao, 2005 Passivity based inventory and ow control in at
glass manufacture, Proc. of the 43rd IEEE Conference on Decision and Control,
pp.4702-4707, December 14-17, Atlantis Paradise island, Bahamas.
527. D.C. Youla, 1961 On the factorization of rational matrices, IEEE Transac-
tions on Inform. Theory, vol.IT-7, pp.172-189.
528. W. Yu, X. Li, 2001 Some stability properties of dynamic neural networks,
IEEE transactions on Circuits and SystemsI: Fundamental Theory and Appli-
cations, vol.48, no 2, pp.256-259, February.
529. W. Yu, 2003 Passivity analysis for dynamic multilayer neuro identier, IEEE
transactions on Circuits and Systems, I- Fundamental Theory and Applications,
vol.50, no 1, pp.173-178, January.
530. W. Yu, L. Wang, 2001 Andersons claim on fourth-order SPR synthesis is
true, IEEE transactions on Circuits and Systems I- Fundamental Theory and
Applications, vol.48, no 4, pp.506-509, April.
531. S. Yuliar, M.R. James, J.W. Helton, 1998 Dissipative control systems synthe-
sis with full state feedback, Math. Control Signals Systems, vol.11, pp.335-356.
568 References

532. G. Zames, 1966 On the inputoutput stability of nonlinear time-varying feed-

back systems, part I, IEEE Transactions on Automatic Control, vol.11, pp.228-
238.
533. G. Zames, 1966 On the inputoutput stability of nonlinear time-varying
feedback systems, part II, IEEE Transactions on Automatic Control, vol.11,
pp.465-477.
534. M. Zefran, F. Bullo, M. Stein, 2001 A notion of passivity for hybrid systems,
Proc. of IEEE Conference on Decision and Control, Orlando, FL, December,
pp.768-773.
535. M. Zefran, 2001 Passivity of hybrid systems based in multiple storage func-
tions,39th Annual Allerton Conference on Communication, Control and Com-
puting, October 3-5, university of Illinois.
536. E. Zeheb, R. Shorten, 2006 A note on spectral conditions for positive realness
of single-input-single-output systems with strictly proper transfer functions,
IEEE Transactions on Automatic control, vol.51, no 5, pp.897-900, May.
537. A.A. Zevin, M.A. Pinsky, 2005 Absolute stability criteria for a generalized
Lure problem with delay in the feedback, SIAM Journal on Control and Op-
timization, vol.43, no 6, pp.2000-2008.
538. A.A. Zevin, M.A. Pinsky, 2003 A new approach to the Lure problem in
the theory of absolute stability, SIAM Journal on Control and Optimization,
vol.42, no 5, pp.1895-1904.
539. G.S. Zhai, B. Hu, K. Yasuda, A. Michel, 2001 Disturbance attenuation prop-
erties of time-controlled switched systems, Journal of the Franklin Institute,
vol.338, pp.765-779.
540. J. Zhao, D.J. Hill, 2005 Dissipativity theory for switched systems, Proc.
IEEE Conference on Decision and Control and European Control Conference,
Seville, Spain, December 12-15, pp.7003-7008.
541. L. Zhang, J. Lam, S. Xu, 2002 On positive realness of descriptor systems,
IEEE transactions on Circuits and Systems, I- Fundamental Theory and Appli-
cations, vol.49, no 3, pp.401-407, March.
542. S. Zhou, J. Lam, G. Feng, 2005 New characterization of positive realness and
control of a class of uncertin polytopic discrete-time systems, Systems and
Control Letters, vol.54, pp.417-427.
Index

(R, P, J)dissipative, 300 backstepping design method, 417

C r -dissipativity, 194 backstepping scheme, 418, 452, 461,
H (nonlinear), 239 468, 478
H supply rate, 210 Barbalats lemma, 514
Jdissipativity, 302 BIBO stable, 25
positive real, 61, 80 bilinear matrix inequality, 115
K-exponential stability, 513 biological systems, 177
Lp norms, 178 BMI, 115
Lp signals properties, 180 bounded real, 37, 40, 297
Mmatrix, 266 bounded real (strict), 297
L2 gain, 215 bounded real (strong), 297
Lp,e spaces, 190 bounded real and positive real, 36
bounded real Lemma, 295
absolute continuity, 138, 146, 149
bounded real Lemma (strict), 295
absolute stability, 140, 294
bounded variation, 146, 366
absolute stability (denition), 140
bounded variation (local), 367
absolute stability (discrete time), 173
boundedness (local), 195
AC function, 146
BV (bounded variation), 366
actuator dynamics, 349, 483
adaptive control, 62, 187, 436, 456, 457,
460, 461 Capri robot, 470
Aizermans conjecture, 140 Caratheodory conditions, 137
algebraic Riccati equation, 120 cascaded systems, 281, 427
Anderson B.D.O., 74 Casimir function, 339
applications of passivity, 177 causal operator, 197
ARE, 120 certainty equivalence principle, 456
ARI, 120 circle criterion, 160, 166, 173
asymptotic stability, 508, 512 circuit
attractivity, 384 RLC, 3
autonomous system, 507 closed-loop dynamics, 401, 415, 439, 451
available storage, 201, 240, 341, 346, collocated sensors and actuators, 131,
348, 353, 356, 378, 387, 402, 407, 424
423 combustion engine, 177
available storage (virtual), 201 compensation
570 Index

of exibilities, 480 detectable (strongly nite time), 286

of nonlinearities, 478 dieomorphism, 516
complementarity conditions, 149, 365, dierentiable storage, 219
370, 432 dierentiable storage functions, 219
compliant environment, 361, 428 dierential geometry theorem, 515
cone complementarity problem, 370 dierential inclusion, 148
conjugate function, 233 dierential inclusion (unbounded), 152
connected state space, 201 diodes (circuit with ideal), 158
connected system, 269 discrete-time (KYP Lemma), 171
constrained mechanical systems, 358, discrete-time (PR), 171
428 discrete-time systems (nonlinear), 247,
constraints 281
bilateral, 358, 430 discretization of PR functions, 175
holonomic, 430 dissipation delay, 213
kinematic, 358 dissipation function, 212, 228
non-holonomic , 361 dissipation inequality, 119, 194, 203,
unilateral, 432 212, 382
continuous (absolutely), 149 dissipation inequality (innitesimal),
continuous dependence, 138 219, 223, 239
continuous storage functions, 217 dissipative ((R, P, J)), 300
controllability, 204 dissipative (exponential), 298
controllable (sign), 104, 524 dissipative (quasi), 197
controllable (strongly), 218 dissipative (weakly), 195, 197
controlled Hamiltonian, 374 dissipative LQG controllers, 130
controlled Lagrangian, 374 dissipative systems
convex analysis, 143, 365 denition, 194
convex function, 143 stability, 257
convexity, 375, 377 dissipativity (local), 197, 269
copositive matrices, 155 domain
Coulomb friction, 151 function, 144, 153
cross terms cancellation, 281, 396, 404, multivalued mapping, 144, 153
409, 415, 419, 425, 439, 460, 463 DRI (degenerate Riccati inequality),
cycle energy, 202 233
cyclo-dissipative, 204 duality, 78
cyclo-dissipativity, 196, 228
elastic transmission, 47, 51
Datko-Pazy Theorem, 186 electrical circuit (nonsmooth), 158
DC motors equilibrium, 508, 511
armature-controlled, 349, 426 error dynamics, 401, 415, 439, 451
eld-controlled, 354, 428 evolution variational inequalities, 152,
passivity, 356 274
position feedback, 424 existence of solutions (ODE), 137
de Larminats method, 475 experimental results
decrescent function, 513 Capri robot, 467
degenerate Riccati inequality, 233 inverted pendulum, 503
delay systems, 288, 291, 305 pulleys system, 467
density, 381 exponential dissipativity, 251
descriptor system, 91 exponential passivity, 298
descriptor variable system, 213 exponential stability, 508, 512
 Index 571

extended spaces, 190 function, 231, 235, 246, 309

extended SPR, 62, 134 matrix, 78
systems, 351, 355
FDI, 120 haptic interface, 177
feedback equivalence, 276 heat equation, 255
feedback gain tuning, 482 hermitian form, 243
feedback interconnection, 14, 24, 263 Hill D.J., 211
feedback interconnection of PR systems, Hill D.J. and Moylan P., 195, 196
116 HJB equation, 243
feedback linearization, 518 holonomic constraints, 358
Filippovs inclusions, 148 homoclinic orbit, 500
nite power gain, 262, 307 hyperstability, 310
nite-gain (exponential), 298 hysteresis (Preisach), 152
nite-gain stable, 190, 212, 216, 258,
259 iISS system, 284
nite-time convergence, 385 iISS-Lyapunov function, 285
rst integral, 241 impedance matching, 31
at glass manufacture, 177 impulse-free system, 92
foliated state-space, 389 inclusion (dierential), 148
force-position control, 440 indicator function, 142, 156
four-block Nehari problem, 299 induced norms, 188
Fradkovs Theorem, 114 induction motors, 357
frequency-domain inequality, 120 inequality (frequency domain), 120
friction forces, 363 innite-dimensional systems, 252
Frobenius theorem, 516 inherent integration, 525
function input-to-state ultimately bounded
absolutely continuous, 146 (ISUB), 286
lower semi-continuous, 146 integrability (complete), 516
function (proper), 270 interconnection
function of class K, 513 feedback, 388, 402, 406, 421, 424
function of class KL, 513 of positive real systems, 57
terms, 430
gain of an operator, 190 invariant set, 510
gains tuning, 474 inverse optimal control, 243
general supply rate, 208 inverse transfer function, 58
generalized equation, 139, 148 inverted pendulum, 496
global asymptotic stability, 512 invertible system, 280, 525
global stability, 510 involutivity, 516
gradient systems, 142 irreversible system, 202
gradient update law, 436438, 457, 461 ISP (Input Strictly Passive), 12, 13, 21,
graph (of a mapping), 144 28, 30, 59, 209, 213, 229, 258, 259,
Gronwalls Lemma, 145 290
gyroscopic forces, 325 ISS (input-to-state stability), 282
ISS-Lyapunov function, 283
Hamilton-Jacobi equality, 306 ISUB (input-to-state ultimately
Hamilton-Jacobi inequality, 235 bounded), 286
Hamilton-Jacobi-Bellman equation,
243, 246 Johanssons scheme, 405
Hamiltonian Jordan chain, 103, 534
572 Index

Kalmans conjecture, 141, 142 Lipschitz continuity (local), 182

Kalman-Bucy lter, 131 Lipschitz continuous, 146
Kalman-Yakubovich-Popov lemma, 69, LMI, 92, 120
346, 347, 393, 394 local w- reachability, 204
nonlinear systems, 222, 223 local boundedness, 195
PR systems, 70 local dissipativity, 197
SPR systems, 81 local stability, 510
Kalman-Yakubovich-Popov lemma locally reachable, 269
positive systems, 256 loop transformations, 162
non-minimal systems, 100 lossless system, 208, 263, 376
time-varying systems, 116 lossless system (discrete-time), 247
unobservable systems, 104 lower semi-continuity, 146
Katos Theorem, 153 lower semi-continuous, 235
kinetic energy loss, 366 Lozano-Brogliato scheme, 414
kinetic metric, 365 Lozano-Brogliato scheme (adaptive),
Krasovskii-LaSalles invariant set 442
theorem, 160, 510 LQ control, 76, 244, 475, 503
KYP Lemma (discrete-time), 171 LQG controllers (passive), 131
KYP Lemma (nite frequency range), Lure equations, 81
134 Lure problem, 135
KYP Lemma (innite dimension), 252 Lyapunov
KYP Lemma (nonnegative systems), direct method, 509, 513
256 equation, 509
KYP Lemma (proof), 74 function, 391, 393, 403, 510
KYP Lemma (stabilizable systems), 107 linearization method, 512
KYP Lemma (time-varying systems), stability, 375, 507, 514
116 linearization method, 508

Lagrange-Dirichlet theorem, 375, 378 magnetic energy, 354

Lagrangian (hyperregular), 322 marginal stability, 508
Lagrangian control systems, 318 Markov parameter, 71
Lagrangian systems, 315, 316 Markus-Yamabes conjecture, 141
Landaus scheme, 172 mass-spring system, 2
large-scale systems, 264 Matrosovs theorem, 399, 514
LaSalles invariance principle, 384 maximal monotone mappings, 144, 254
LaSalles theorem, 160, 510 McClamroch-Wangs coordinates, 359
LBV (local bounded variation), 367 Measure Dierential Equations (MDE),
LCS (linear complementarity system), 383
149, 159 measure dierential inclusion, 367
least-squares update law, 440 mechanical analogs for PD controllers,
Lefschetz-Kalman-Yakubovich lemma, 25
79 mechanical resonances, 47
Legendre transformation, 317, 328 Meyer-Kalman-Yakubovich Lemma,
Lie bracket, 516 102, 460
Lie derivative, 515 mild solution, 537
linear complementarity systems (LCS), MIMO systems, 27, 37, 40, 43, 47
149 minimal stability, 311
linear matrix inequality (LMI), 120 minimum phase systems, 277
Lipschitz continuity, 183 missile guidance, 177
 Index 573

mixed+observable subsystem, 105 output feedback to obtain SPR systems,

monotone (strongly), 139 65
monotone mapping, 139, 144, 254, 370,
385 P + observer control, 408
Moore-Penrose pseudo-inverse, 233 Paden-Panja scheme, 398, 410
Moreaus inclusion, 367 parallel interconnection, 14
Moreaus Lemma, 368 parameter adaptation law, 443
Moreaus sweeping process, 149, 367 parameter projection (dierentiable),
Morses high order tuners, 463 443
MSPR functions, 62 Parsevals theorem, 15, 42
multivalued functions, 142 partial dierential inequalities, 219, 231
multivalued mapping, 144, 220, 370 partial dierential inequality, 206, 235,
multivariable systems, 27, 37, 40, 43, 47 309, 402
particulate process, 177
natural outputs, 328, 337 passication, 65, 276
negative cone, 144 passication by pole shifting, 276
negative denite, 509 passive (exponential), 298
negative semi-denite, 509 passive output, 345
neural networks, 177 passive PID-regulator, 24
neutral subspace, 523 passive systems, 207
non-ane in the input systems, 230 passive systems
non-autonomous system, 511 feedback equivalence to , 276
non-minimal systems (KYP Lemma), passive systems (discrete-time), 247
100 passivity, 207
nonlinear discrete-time systems, 247 passivity
nonlinear-in-the-input systems, 230 denition, 11
nonnegative matrix, 255 input strict, 12, 389, 390
nonnegative systems, 255 linear multivariable systems, 27
nonsmooth bouncing system, 347, 378 linear systems, 16, 37
nonsmooth Lagrangian system, 365, 432 linear systems with delay, 288
nonsmooth mechanical systems, 432 linear systems with distributed delay,
nonsmooth systems, 142, 152, 274, 309, 291
363 of PID controllers, 24
normal cone, 139, 144, 368 of the inverted pendulum, 499
normal form, 276, 517 output strict, 12, 391, 410
normal rank (polynomial matrix), 102 very strict, 12, 402
Normed spaces, 178 passivity theorem, 393, 421
Nyquist criterion, 166 passivity theorem
one channel, 257
observer design, 113 two channels, 259
one-sided dierentials, 521 passivity-based control, 386, 398, 404,
operator (input-output), 116 414, 442, 468, 501
optimal control, 76, 119, 244 PD + gravity compensation, 436
optimal control (inverse), 243 PD control, 25, 386, 398, 422, 428, 476
optimal control (singular), 234 pedantic language, 146
OSP (Output Strictly Passive), 12, 13, persistent excitation, 274
19, 209, 213, 216, 258, 259, 356, phase lag, 212
460 physiological systems, 177
OSP and SPR, 213 PID control, 24, 391
574 Index

piecewise continuous functions, 366 reachability (local wuniform), 219

piecewise continuous systems, 349 reachability (uniform), 198
polar cone, 144 reachable (locally), 269
Popov function, 41, 100, 121, 126, 134 reduced order dynamics, 359
Popov triple, 121 regular point, 277
Popov V.M., 121, 167, 178, 310 regular subgradient, 220
Popovs criterion, 166, 173 regularity of storage functions, 217, 382
Popovs hyperstability, 310 regularization (storage functions), 221
Popovs line, 170 relative degree, 276, 460, 517
Port controlled Hamiltonian systems, relative degree (uniform), 517
359 relay characteristic, 143
port controlled Hamiltonian systems, required supply, 201, 240, 342, 346, 353,
332 402
position feedback, 408 resonances, 47
positive denite, 509 reversible system, 202
positive denite function, 513 Riccati equality, 77, 207
positive matrix, 255 Riccati equation, 120, 125, 131, 203,
positive real (discrete-time), 171 240, 243, 295, 523
positive real and bounded real, 36 Riccati inequality, 76, 120, 215, 232,
positive real lemma, 70 266, 289
positive real transfer function, 40, 44, Riccati inequality (degenerate), 233
45 Riccati inequality (reduced order), 533
positive semi-denite, 509 robot manipulators
positive systems, 255 exible joint, 343, 414, 452, 467
potential energy, 375 rigid, 340, 386, 408, 436
potential energy shaping, 388 RORE (Reduced Order Riccati
power converters, 177 Equation), 234, 525
power gain, 262
PR functions (discretization of), 175 Sadegh and Horowitz controller, 399
PR tangent linearization, 249 sampled systems, 175
PR transfer function, 40, 45, 389, 392 Sannutis canonical form, 518
principle of optimality, 119 satellites, 177
prismatic joints, 345 scattering formulation, 28, 34, 37
process systems, 177 Schur complement, 297, 531
proper convex function, 521 SDP and KYP Lemma, 134
proper function, 270 secant condition, 263
proper system, 185 second order sweeping process, 367
pseudo inverse (Moore-Penrose), 233 sector condition, 294
pulley system, 470 semi-continuous, 146
semi-global stability, 410
quadratic matrix inequality(QMI), 120 semidenite programming (SDP), 134
quadratic storage functions, 71, 194 Sen (De La) M., 175
quasi-dissipative, 197, 262, 286 separation principle, 410
quasi-dissipativity, 197 sign controllable, 104, 524
singular optimal control, 234
rational transfer function, 19 singular system, 91, 213
Rayleigh dissipation, 341 Slotine and Li (switching), 432
reachability, 201 Slotine-Li scheme, 399, 412, 438
reachability (local w), 204 small gain theorem, 29, 191, 310
 Index 575

smart actuators, 177 transfer function (rational), 19

spectral factorization, 109 Tsypkin criterion, 173
spectral function, 41, 100, 311 tuning of gains, 474
SPR (extended), 62, 134
SPR (test for), 57 ultimate dissipativity, 196
SPR and OSP, 213 uncontrollable systems (KYP lemma),
SPR dynamic feedback, 113 100
SPR system, 79 underactuated mechanical system, 497
SPR transfer function, 54, 298, 457, 458 uniform asymptotic stability, 512
stability, 508, 511 uniform continuity, 183
stability (minimal), 311 uniform reachability, 204
stability on K, 154 uniform relative degree, 517
stabilization by output feedback, 272 uniform stability, 512
stabilization of feedback linearizable uniformly observable, 269
systems, 519 unilateral constraint, 364, 365
stabilizing solution (ARE), 124, 125, uniqueness of solutions (ODE), 137
211, 530 unmixed matrix, 104, 524
stabilizing solutions (ARE), 243 unobservable systems (KYP lemma),
storage function, 119, 152, 382, 389, 438 100
storage function
denition, 194 value function, 119
storage function (virtual), 217 variational inequalities, 152, 274, 364,
strict passivity (weak), 208 365
strictly passive (discrete-time), 247 variational inequalities (PR), 156
strictly passive systems, 208 variational test, 204
strictly positive real systems, 53 velocity potential, 369
strong controllability, 218 virtual available storage, 201
strong SPR transfer function, 59, 78, virtual storage function, 217
213 virtual work principle, 362
strongly nite time detectable, 286 viscosity solution, 235
subdierential, 143 viscosity solutions, 309, 520
subgradient, 143, 220 viscosity subgradient, 220
supply rate, 13, 353 visual feedback control, 177
supply rate (H ), 210 VSP (pseudo), 261
supply rate (general), 209 VSP (Very Strictly Passive), 12, 13,
sweeping process (second order), 367 209, 217, 259, 402
switching control, 503
switching Slotine and Li scheme, 432 wave equation, 253
switching systems, 293, 349 weak solution, 235
systems (large-scale), 264 weak SPR transfer function, 59, 95, 263,
270
tangent cone, 144 weak strict passivity, 208
tangent linearization, 249 weakly dissipative, 195, 197, 198
Thomson-Taits formula, 382 Weierstrass form, 91
time-invariant system, 507 well-posedness (ODE), 137
time-varying system, 507 WFGS (nite-gain stable), 191, 212,
time-varying systems, 115, 229, 273 216
total variation, 366 Willems J.C., 119, 194
tracking control, 398 WSPR, 95, 270, 530
576 Index

Yakubovich-Kalman-Popov Lemma, 69 zero-state observable (ZSO), 266

ZSD, 266
zero dynamics, 172, 277, 520
zero-state detectable (ZSD), 266 ZSD (locally uniformly), 269
zero-state observable, 391 ZSO, 266