Nonlinear Dynamics Between Linear and Impact Limits PDF

Lecture Notes in Applied
and Computational Mechanics

Volume 52
Series Editors
Prof. Dr.-Ing. Friedrich Pfeiffer

Prof. Dr.-Ing. Peter Wriggers
Lecture Notes in Applied and Computational Mechanics
Edited by F. Pfeiffer and P. Wriggers
Further volumes of this series found on our homepage: springer.com
Vol. 52: Pilipchuk, V.N. Vol. 40: Pfeiffer, F.

Nonlinear Dynamics: Between Linear and Impact Limits Mechanical System Dynamics
364 p. 2010 [978-3-642-12798-4] 578 p. 2008 [978-3-540-79435-6]
Vol. 51: Besdo, D., Heimann, B., Klüppel, M., Vol. 39: Lucchesi, M., Padovani, C., Pasquinelli, G., Zani, N.
Kröger, M., Wriggers, P., Nackenhorst, U. Masonry Constructions: Mechanical
Elastomere Friction Models and Numerical Applications
249 p. 2010 [978-3-642-10656-9] 176 p. 2008 [978-3-540-79110-2]
Vol. 50: Ganghoffer, J.-F., Pastrone, F. (Eds.) Vol. 38: Marynowski, K.

Mechanics of Microstructured Solids 2 Dynamics of the Axially Moving Orthotropic Web
102 p. 2010 [978-3-642-05170-8] 140 p. 2008 [978-3-540-78988-8]
Vol. 37: Chaudhary, H., Saha, S.K.

Vol. 49: Hazra, S.B.
Dynamics and Balancing of Multibody Systems
Large-Scale PDE-Constrained Optimization
200 p. 2008 [978-3-540-78178-3]
in Applications
224 p. 2010 [978-3-642-01501-4]
Vol. 36: Leine, R.I.; van de Wouw, N.
Stability and Convergence of Mechanical Systems
Vol. 48: Su, Z.; Ye, L.
with Unilateral Constraints
Identification of Damage Using Lamb Waves
250 p. 2008 [978-3-540-76974-3]
346 p. 2009 [978-1-84882-783-7]
Vol. 35: Acary, V.; Brogliato, B.
Vol. 47: Studer, C.
Numerical Methods for Nonsmooth Dynamical Systems:
Numerics of Unilateral Contacts and Friction
Applications in Mechanics and Electronics
191 p. 2009 [978-3-642-01099-6]
545 p. 2008 [978-3-540-75391-9]
Vol. 46: Ganghoffer, J.-F., Pastrone, F. (Eds.)
Vol. 34: Flores, P.; Ambrósio, J.; Pimenta Claro, J.C.;
Mechanics of Microstructured Solids
Lankarani Hamid M.
136 p. 2009 [978-3-642-00910-5]
Kinematics and Dynamics of Multibody Systems
with Imperfect Joints: Models and Case Studies
Vol. 45: Shevchuk, I.V. 186 p. 2008 [978-3-540-74359-0
Convective Heat and Mass Transfer in Rotating Disk
Systems
V ol. 33: Nies ony, A.; Macha, E.
300 p. 2009 [978-3-642-00717-0]
Spectral Method in Multiaxial Random Fatigue
146 p. 2007 [978-3-540-73822-0]
Vol. 44: Ibrahim R.A., Babitsky, V.I., Okuma, M. (Eds.)
Vibro-Impact Dynamics of Ocean Systems and Related
Vol. 32: Bardzokas, D.I.; Filshtinsky, M.L.;
Problems
Filshtinsky, L.A. (Eds.)
280 p. 2009 [978-3-642-00628-9]
Mathematical Methods in Electro-Magneto-Elasticity
530 p. 2007 [978-3-540-71030-1]
Vol.43: Ibrahim, R.A.
Vibro-Impact Dynamics Vol. 31: Lehmann, L. (Ed.)
312 p. 2009 [978-3-642-00274-8] Wave Propagation in Infinite Domains
186 p. 2007 [978-3-540-71108-7]
Vol. 42: Hashiguchi, K.
Elastoplasticity Theory Vol. 30: Stupkiewicz, S. (Ed.)
432 p. 2009 [978-3-642-00272-4] Micromechanics of Contact and Interphase Layers
206 p. 2006 [978-3-540-49716-5]
Vol. 41: Browand, F., Ross, J., McCallen, R. (Eds.)
Aerodynamics of Heavy Vehicles II: Trucks, Buses, Vol. 29: Schanz, M.; Steinbach, O. (Eds.)
and Trains Boundary Element Analysis
486 p. 2009 [978-3-540-85069-4] 571 p. 2006 [978-3-540-47465-4]
Nonlinear Dynamics
Between Linear and Impact Limits
Valery N. Pilipchuk
123
Valery N. Pilipchuk
Professor (Research)
Mechanical Engineering
Wayne State University
5050 Anthony Wayne Dr., 2140
Detroit, Michigan 48202
USA
E-mail: pilipchuk@wayne.edu
ISBN: 978-3-642-12798-4 e-ISBN: 978-3-642-12799-1
DOI 10.1007/ 978-3-642-12799-1
Lecture Notes in Applied and Computational Mechanics ISSN 1613-7736
e-ISSN 1860-0816
Library of Congress Control Number: 2010925555
© Springer-Verlag Berlin Heidelberg 2010
This work is subject to copyright. All rights are reserved, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilm or in any other ways, and storage in data banks.
Duplication of this publication or parts thereof is permitted only under the provisions of the German
Copyright Law of September 9, 1965, in its current version, and permission for use must always be
obtained from Springer. Violations are liable for prosecution under the German Copyright Law.
The use of general descriptive names, registered names, trademarks, etc. in this publication does not
imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
Typeset & Cover Design: Scientific Publishing Services Pvt. Ltd., Chennai, India.
Printed on acid-free paper
9876543210
springer.com
Preface
The main objective of this book is to introduce a unified physical basis

for analyses of vibrations with essentially unharmonic, non-smooth or may
be discontinuous time shapes. It is known that possible transitions to non-
smooth limits can make investigations especially difficult. This is due to the
fact that the dynamic methods were originally developed within the paradigm
of smooth motions based on the classical theory of differential equations.
From the physical standpoint, these represent low-energy approaches to mod-
eling dynamical systems. Although the impact dynamics has also quite a long
pre-history, any kind of non-smooth behavior is often viewed as an exemption
rather than a rule. Similarly, the classical theory of differential equations usu-
ally avoids non-differentiable and discontinuous functions. To-date, however,
many theoretical and applied areas cover high-energy phenomena accom-
panied by strongly non-linear spatio-temporal behaviors making the classi-
cal smooth methods inefficient in many cases. For instance, such phenom-
ena occur when dealing with dynamical systems under constraint conditions,
friction-induced vibrations, structural damages due to cracks, liquid sloshing
impacts, and numerous problems of nonlinear physics. Similarly to the well-
known analogy between mechanical and electrical harmonic oscillators, some
electronic instruments include so-called Schmitt trigger circuits generating
nonsmooth signals whose temporal shapes resemble mechanical vibro-impact
processes. In many such cases, it is still possible to adapt different smooth
methods of the dynamic analyses through strongly non-linear algebraic ma-
nipulations with state vectors or by splitting the phase space into multiple
domains based on the system specifics. As a result, the related formulations
are often reduced to discrete mappings in a wide range of the dynamics from
periodic to stochastic. Possible alternatives to such approaches can be built
on generating models developing essentially nonlinear/unharmonic behaviors
as their inherent properties. Such models must be general and simple enough
in order to play the role of physical basis. As shown in this book, new generat-
ing systems can be found by intentionally imposing the ‘worst case scenario’
on conventional methods in anticipation that failure of one asymptotic may
VI Preface
point to its complementary counterpart. However, the related mathematical

formalizations are seldom straightforward and require new principles. For in-
stance, the tool developed here employs nonsmooth (impact) systems as a
basis to describing not only impact but also smooth or even linear dynamics.
This is built on the idea of non-smooth time substitutions/transformations
(NSTT) proposed originally for strongly nonlinear but still smooth models.
On the author’s view, the methodological role of NSTT is to reveal explicit
links between impact dynamics and hyperbolic algebras analogously to the
link between harmonic vibrations and conventional complex analyses. In par-
ticular, this book gives the first systematic description for NSTT and related
analytical and numerical algorithms. The text focuses on methodologies and
discussions of their physical and mathematical basics. Detailed applications
are mostly excluded from this book, however, necessary references on journal
publications are provided.
The material of this book was prepared during several years of work at
Dnipropetrovsk National University and Technological University (Ukraine),
Wayne State University, and Research and Development of General Motors
Corporation (Michigan, USA).
The author greatly appreciates discussions on different subjects related to
this book during different periods of time with Professors I.V. Andrianov,
R.A. Ibrahim (Wayne State University), L.I. Manevich (Russian Academy of
Science), A.A.Martynyuk (National Academy of Science of Ukraine), Yu.V.
Mikhlin (Kharkov Polytechnic Institute), A.F. Vakakis (University of Illi-
nois), A.A. Zevin (National Academy of Science of Ukraine), and V.F. Zhu-
ravlev (Russian Academy of Science). The author also would like to thank Dr.
Kelly Cooper (ONR) for her support and attitude to fundamental research
that was very helpful at the final stage of work on this book.
Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Brief Literature Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Asymptotic Meaning of the Approach . . . . . . . . . . . . . . . . . . . . 4
1.2.1 Two Simple Limits of Lyapunov Oscillator . . . . . . . . . . 4
1.2.2 Oscillating Time and Hyperbolic Numbers,
Standard and Idempotent Basis . . . . . . . . . . . . . . . . . . . 6
1.3 Quick ‘Tutorial’ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3.1 Remarks on the Basic Functions . . . . . . . . . . . . . . . . . . . 9
1.3.2 Viscous Dynamics under the Sawtooth Forcing . . . . . . 9
1.3.3 The Rectangular Cosine Input . . . . . . . . . . . . . . . . . . . . 11
1.3.4 Oscillatory Pipe Flow Model . . . . . . . . . . . . . . . . . . . . . . 12
1.3.5 Periodic Impulsive Loading . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.6 Strongly Nonlinear Oscillator . . . . . . . . . . . . . . . . . . . . . 15
1.4 Geometrical Views on Nonlinearity . . . . . . . . . . . . . . . . . . . . . . 17
1.4.1 Geometrical Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4.2 Nonlinear Equations and Nonlinear Phenomena . . . . . 19
1.4.3 Rigid-Body Motions and Linear Systems . . . . . . . . . . . 21
1.4.4 Remarks on the Multi-dimensional Case . . . . . . . . . . . . 23
1.4.5 Elementary Nonlinearities . . . . . . . . . . . . . . . . . . . . . . . . 24
1.4.6 Example of Simplification in Nonsmooth Limit . . . . . . 25
1.4.7 Non-smooth Time Arguments . . . . . . . . . . . . . . . . . . . . . 26
1.4.8 Further Examples and Discussion . . . . . . . . . . . . . . . . . . 28
1.4.9 Differential Equations of Motion and
Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.5 Non-smooth Coordinate Transformations . . . . . . . . . . . . . . . . . 33
1.5.1 Caratheodory Substitution . . . . . . . . . . . . . . . . . . . . . . . 33
1.5.2 Transformation of Positional Variables . . . . . . . . . . . . . 33
1.5.3 Transformation of State Variables . . . . . . . . . . . . . . . . . 35
VIII Contents
2 Smooth Oscillating Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

2.1 Linear and Weakly Non-linear Approaches . . . . . . . . . . . . . . . . 37
2.2 A Brief Overview of Smooth Methods . . . . . . . . . . . . . . . . . . . . 38
2.2.1 Periodic Motions of Quasi Linear Systems . . . . . . . . . . 38
2.2.2 The Idea of Averaging . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.2.3 Averaging Algorithm for Essentially Nonlinear
Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.2.4 Averaging in Complex Variables . . . . . . . . . . . . . . . . . . . 43
2.2.5 Lie Group Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3 Nonsmooth Processes as Asymptotic Limits . . . . . . . . . . . . . 51

3.1 Lyapunov’ Oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.2 Nonlinear Oscillators Solvable in Elementary Functions . . . . . 54
3.2.1 Hardening Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.2.2 Localized Damping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.2.3 Softening Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.3 Nonsmoothness Hiden in Smooth Processes . . . . . . . . . . . . . . . 61
3.3.1 Nonlinear Beats Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.4 Nonlinear Beat Dynamics: The Standard Averaging
Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.4.1 Asymptotic of Equipartition . . . . . . . . . . . . . . . . . . . . . . 69
3.4.2 Asymptotic of Dominants . . . . . . . . . . . . . . . . . . . . . . . . 71
3.4.3 Necessary Condition of Energy Trapping . . . . . . . . . . . 73
3.4.4 Sufficient Condition of Energy Trapping . . . . . . . . . . . . 74
3.5 Transition from Normal to Local Modes . . . . . . . . . . . . . . . . . . 74
3.6 System Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.7 Normal and Local Mode Coordinates . . . . . . . . . . . . . . . . . . . . 76
3.8 Local Mode Interaction Dynamics . . . . . . . . . . . . . . . . . . . . . . . 81
3.9 Auto-localized Modes in Nonlinear Coupled Oscillators . . . . . 85
4 Nonsmooth Temporal Transformations (NSTT) . . . . . . . . . . 93

4.1 Non-smooth Time Transformations . . . . . . . . . . . . . . . . . . . . . . 93
4.1.1 Positive Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.1.2 ‘Single-Tooth’ Substitution . . . . . . . . . . . . . . . . . . . . . . . 96
4.1.3 ‘Broken Time’ Substitution . . . . . . . . . . . . . . . . . . . . . . . 96
4.1.4 Sawtooth Sine Transformation . . . . . . . . . . . . . . . . . . . . 97
4.1.5 Links between NSTT and Matrix Algebras . . . . . . . . . 101
4.1.6 Differentiation and Integration Rules . . . . . . . . . . . . . . . 102
4.1.7 NSTT Averaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
4.1.8 Generalizations on Asymmetrical Sawtooth Wave . . . . 105
4.1.9 Multiple Frequency Case . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.2 Idempotent Basis Generated by the Triangular
Sine-Wave . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.2.1 Definitions and Algebraic Rules . . . . . . . . . . . . . . . . . . . 109
4.2.2 Time Derivatives in the Idempotent Basis . . . . . . . . . . 111
Contents IX
4.3 Idempotent Basis Generated by Asymmetric Triangular

Wave . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
4.3.1 Definition and Algebraic Properties . . . . . . . . . . . . . . . . 112
4.3.2 Differentiation Rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
4.3.3 Oscillators in the Idempotent Basis . . . . . . . . . . . . . . . . 115
4.3.4 Integration in the Idempotent Basis . . . . . . . . . . . . . . . 116
4.4 Discussions, Remarks and Justifications . . . . . . . . . . . . . . . . . . 117
4.4.1 Remarks on Nonsmooth Solutions in the Classical
Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
4.4.2 Caratheodory Equation . . . . . . . . . . . . . . . . . . . . . . . . . . 119
4.4.3 Other Versions of Periodic Time Substitutions . . . . . . 122
4.4.4 General Case of Non-invertible Time and Its
Physical Meaning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
4.4.5 NSTT and Cnoidal Waves . . . . . . . . . . . . . . . . . . . . . . . . 125
5 Sawtooth Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

5.1 Manipulations with the Series . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
5.1.1 Smoothing Procedures . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
5.2 Sawtooth Series for Normal Modes . . . . . . . . . . . . . . . . . . . . . . 135
5.2.1 Periodic Version of Lie Series . . . . . . . . . . . . . . . . . . . . . 135
5.3 Lie Series of Transformed Systems . . . . . . . . . . . . . . . . . . . . . . . 138
5.3.1 Second-Order Non-autonomous Systems . . . . . . . . . . . . 138
5.3.2 NSTT of Lagrangian and Hamiltonian Equations . . . . 141
5.3.3 Remark on Multiple Argument Cases . . . . . . . . . . . . . . 144
6 NSTT for Linear and Piecewise-Linear Systems . . . . . . . . . 145

6.1 Free Harmonic Oscillator: Temporal Quantization of
Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
6.2 Non-autonomous Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
6.2.1 Standard Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
6.2.2 Idempotent Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
6.3 Systems under Periodic Pulsed Excitation . . . . . . . . . . . . . . . . 149
6.3.1 Regular Periodic Impulses . . . . . . . . . . . . . . . . . . . . . . . . 149
6.3.2 Harmonic Oscillator under the Periodic Impulsive
Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
6.3.3 Periodic Impulses with a Temporal ‘Dipole’ Shift . . . . 155
6.4 Parametric Excitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
6.4.1 Piecewise-Constant Excitation . . . . . . . . . . . . . . . . . . . . 157
6.4.2 Parametric Impulsive Excitation . . . . . . . . . . . . . . . . . . 159
6.4.3 General Case of Periodic Parametric Excitation . . . . . 161
6.5 Input-Output Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
6.6 Piecewise-Linear Oscillators with Asymmetric
Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
6.6.1 Amplitude-Phase Equations . . . . . . . . . . . . . . . . . . . . . . 166
6.6.2 Amplitude Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
X Contents
6.6.3 Phase Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168

6.6.4 Remarks on Generalized Taylor Expansions . . . . . . . . . 172
6.7 Multiple Degrees-of-Freedom Case . . . . . . . . . . . . . . . . . . . . . . . 173
6.8 The Amplitude-Phase Problem in the Idempotent Basis . . . . 177
7 Periodic and Transient Nonlinear Dynamics under

Discontinuous Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
7.1 Nonsmooth Two Variables Method . . . . . . . . . . . . . . . . . . . . . . 179
7.2 Resonances in the Duffing’s Oscillator under Impulsive
Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
7.3 Strongly Nonlinear Oscillator under Periodic Pulses . . . . . . . . 185
7.4 Impact Oscillators under Impulsive Loading . . . . . . . . . . . . . . 189
8 Strongly Nonlinear Vibrations . . . . . . . . . . . . . . . . . . . . . . . . . . . 195

8.1 Periodic Solutions for First Order Dynamical Systems . . . . . . 195
8.2 Second Order Dynamical Systems . . . . . . . . . . . . . . . . . . . . . . . 196
8.3 Periodic Solutions of Conservative Systems . . . . . . . . . . . . . . . 198
8.3.1 The Vibroimpact Approximation . . . . . . . . . . . . . . . . . . 198
8.3.2 One Degree-of-Freedom General Conservative
Oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
8.3.3 A Nonlinear Mass-Spring Model That Becomes
Linear at High Amplitudes . . . . . . . . . . . . . . . . . . . . . . . 205
8.3.4 Strongly Non-linear Characteristic with a
Step-Wise Discontinuity at Zero . . . . . . . . . . . . . . . . . . . 207
8.3.5 A Generalized Case of Odd Characteristics . . . . . . . . . 209
8.4 Periodic Motions Close to Separatrix Loop . . . . . . . . . . . . . . . 211
8.5 Self-excited Oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
8.6 Strongly Nonlinear Oscillator with Viscous Damping . . . . . . . 218
8.6.1 Remark on NSTT Combined with Two Variables
Expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
8.6.2 Oscillator with Two Nonsmooth Limits . . . . . . . . . . . . 225
8.7 Bouncing Ball . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
8.8 The Kicked Rotor Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
8.9 Oscillators with Piece-Wise Nonlinear Restoring Force
Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
9 Strongly Nonlinear Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241

9.1 Wave Processes in One-Dimensional Systems . . . . . . . . . . . . . . 241
9.2 Klein-Gordon Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
10 Impact Modes and Parameter Variations . . . . . . . . . . . . . . . . 245

10.1 An Introductory Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245
10.2 Parameter Variation and Averaging . . . . . . . . . . . . . . . . . . . . . . 249
10.3 A Two-Degrees-of-Freedom Model . . . . . . . . . . . . . . . . . . . . . . . 252
10.4 Averaging in the 2DOF System . . . . . . . . . . . . . . . . . . . . . . . . . 253
10.5 Impact Modes in Multiple Degrees of Freedom Systems . . . . 256
Contents XI
10.5.1 A Double-Pendulum with Amplitude Limiters . . . . . . . 258

10.5.2 A Mass-Spring Chain under Constraint
Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
10.6 Systems with Multiple Impacting Particles . . . . . . . . . . . . . . . . 262
11 Principal Trajectories of Forced Vibrations . . . . . . . . . . . . . . 265

11.1 Introductory Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
11.2 Principal Directions of Linear Forced Systems . . . . . . . . . . . . . 267
11.3 Definition for Principal Trajectories of Nonlinear Discrete
Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
11.4 Asymptotic Expansions for Principal Trajectories . . . . . . . . . . 269
11.5 Definition for Principal Modes of Continuous Systems . . . . . . 271
12 NSTT and Shooting Method for Periodic Motions . . . . . . . 275

12.1 Introductory Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
12.2 Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
12.3 Sample Problems and Discussion . . . . . . . . . . . . . . . . . . . . . . . . 279
12.3.1 Smooth Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
12.3.2 Step-Wise Discontinuous Input . . . . . . . . . . . . . . . . . . . . 286
12.3.3 Impulsive Loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
12.4 Other Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
12.4.1 Periodic Solutions of the Period - n . . . . . . . . . . . . . . . . 290
12.4.2 Two-Degrees-of-Freedom Systems . . . . . . . . . . . . . . . . . 293
12.4.3 The Autonomous Case . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
13 Essentially Non-periodic Processes . . . . . . . . . . . . . . . . . . . . . . 295

13.1 Nonsmooth Time Decomposition and Pulse Propagation
in a Chain of Particles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
13.2 Impulsively Loaded Dynamical Systems . . . . . . . . . . . . . . . . . . 298
13.2.1 Harmonic Oscillator under Sequential Impulses . . . . . . 301
13.2.2 Random Suppression of Chaos . . . . . . . . . . . . . . . . . . . . 303
14 Spatially-Oscillating Structures . . . . . . . . . . . . . . . . . . . . . . . . . . 305

14.1 Periodic Nonsmooth Structures . . . . . . . . . . . . . . . . . . . . . . . . . 305
14.2 Averaging for One-Dimensional Periodic Structures . . . . . . . . 312
14.3 Two Variable Expansions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
14.4 Second Order Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
14.5 Acoustic Waves from Non-smooth Periodic Boundary
Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
14.6 Spatio-temporal Periodicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
14.7 Membrane on a Two-Dimensional Periodic Foundation . . . . . 326
14.8 The Idempotent Basis for Two-Dimensional Structures . . . . . 332
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
Appendices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
Chapter 1
Introduction
Abstract. This chapter contains physical and mathematical preliminaries

with different introductory remarks. Although some of the statements are
informal and rather intuitive, they nevertheless provide hints on selecting
the generating models and corresponding analytical techniques. The idea is
that simplicity of a mathematical formalism is caused by hidden links be-
tween the corresponding generating models and subgroups of rigid-body mo-
tions. Such motions may be qualified indeed as elementary macro-dynamic
phenomena developed in the physical space. For instance, since rigid-body
rotations are associated with sine waves and therefore (smooth) harmonic
analyses then translations and mirror-wise reflections must reveal adequate
tools for strongly unharmonic and nonsmooth approaches. This viewpoint is
illustrated by physical examples, problem formulations and solutions.
1.1 Brief Literature Overview
Analytical methods of conventional nonlinear dynamics are based on the clas-

sical theory of differential equations dealing with smooth coordinate trans-
formations, asymptotic integrations and averaging [96], [27], [87], [124], [204],
[160], [79], [50], [114], [86], [8], [50], [161], [77], [89], [115], [120], [117], [116],
[204], [97], [62]. The corresponding solutions often include quasi harmonic
expansions as a generic feature that explicitly points to the physical basis of
these methods namely - the harmonic oscillator. Generally speaking, some of
the techniques are also applicable to dynamical systems close to integrable
but not necessarily linear. However, nonlinear generating solutions are seldom
available in closed form [9]. As a result, strongly nonlinear methods usually
target specific situations and are rather difficult to use in other cases.
Generating models for strongly nonlinear analytical tools with a wide
range of applicability must obviously 1) capture the most common features
of oscillating processes regardless their nonlinear specifics, 2) possess simple
enough solutions in order to provide efficiency of perturbation schemes, and 3)
V.N. Pilipchuk: Nonlinear Dynamics, LNACM 52, pp. 1–36, 2010.

springerlink.com
c Springer-Verlag Berlin Heidelberg 2010
2 1 Introduction
describe essentially nonlinear phenomena out of the scope of the weakly non-
linear methods.
So the key notion of the present work suggests possible recipes for selecting
such models among so-called non-smooth systems while keeping the class of
smooth motions still within the range of applicability.
Note that different non-smooth cases have been also considered for several
decades by practical and theoretical reasons [81], [39], [17], [42], [127], [72],
[196], [82], [204], [172], [45], [65], [204], [169], [29], [14], [93], [30], [180], [51],
[206], [94], [44], [173], [25], [162], [55], [185], [85], [130], [129], [128], [1], [68],
[69], [37]. On the physical point of view, this kind of modeling essentially em-
ploys the idea of perfect spatio-temporal localization of strong nonlinearities
or impulsive loadings. For instance, sudden jumps of restoring force charac-
teristics are represented by absolutely stiff constraints under the assumption
that the dynamics in between the constraints is smooth and simple enough to
describe. As a result, the system dynamics is discretized in terms of mappings
and matchings different pieces of solutions.
The present work however is rather close to another group of methods deal-
ing with the differential equations of motion on entire time intervals despite
of discontinuity and/or nonsmoothness points. Such methods are developed
to satisfy the matching conditions automatically by means of specific coordi-
nate transformations on preliminary stages of study. To some extent, these
can be qualified as adaptations of the differential equations of motion for fur-
ther studies by another methods. Among such kind of transformations, the
Caratheodory substitution [45] can be mentioned first. This linear substitu-
tion, which includes the unit-step Heaviside function, eliminates Dirac delta-
functions participating as summands in differential equations. Much later,
non-smooth nonlinear coordinate transformations were suggested in [199],
[204] for the class of impact systems. This strongly nonlinear transforma-
tion effectively eliminates stiff barriers by unfolding the system configuration
space in a mirror-wise manner with respect to the barrier planes. A similar
idea was implemented later regarding the system phase space [72] in order
to resolve certain problems related to non-elastic impacts. Some technical
details and discussions on these methods are included below in Section 5 for
comparison reason.
In contrast, the present approach employs time histories of impact systems
as new time arguments. Originally such a nonsmooth temporal transforma-
tion (NSTT), or a sawtooth oscillating time1 , was introduced for strongly
nonlinear but smooth periodic motions with certain temporal symmetries
[132]. Then, it was shown that such an approach still works for general
cases by generating specific algebraic structures in terms of the coordinates
[134], [144]. The occurrence of such algebraic structures seems to be essential
1
Here and further, the term ‘sawtooth’ is used for periodic piecewise-linear func-
tions regardless symmetries of their teeth; in most cases, it is the triangular sine
wave.
1.1 Brief Literature Overview 3
feature of the approach since it justifies and simplifies analytical manipula-

tions with non-invertible temporal substitutions such as NSTT.
Further, the technique was applied to different problems of theoretical and
applied mechanics [15], [101], [104], [103], [47], [189], [92], [91], [175], [170],
[177], [171], [16], [179], [109], [181], [67]. Basic ideas and some of the tools
were adapted earlier for the nonlinear normal mode analyses and included in
the monograph [190]. However, much of the material presented in this book
is either new or significantly updated. The entire text is organized as follows.
As mentioned in the Preamble, this Chapter 1 contains physical and math-
ematical preliminaries and different introductory remarks.
Chapter 2 gives a brief overview of selected analytical methods for smooth
oscillating processes. The description focuses on the ideas and technical de-
tails that are used further in combinations with non-smooth approaches. In
particular, the method of asymptotic integration of the differential equations
of motion based on the Hausdorf equation for Lie operators is reproduced.
Chapter 3 includes different examples of smooth vibrating systems that,
under some conditions, show close to non-smooth or even non-smooth time
histories. Such cases are usually most difficult for conventional analyses. Much
of the content focuses on nonlinear beat and localization phenomena. Recent
interest to this area is driven by the idea of nonlinear energy absorption.
Chapter 4 provides a new description of NSTT with proofs of the basic
identities and rules for algebraic, differential and integral manipulations. In
particular, final subsections show how to introduce nonsmooth arguments
into the differential equations. Such manipulation imposes two main features
on the dynamical systems namely generates specific algebraic structures for
unknown functions and switches formulations to boundary-value problems.
Notice that the transformation itself imposes no constraints on dynamical
systems and easily applies to both smooth and nonsmooth systems. Any
further steps, however, should account for physical properties of the related
systems.
Chapter 5 preliminary illustrates the NSTT’ advantages by introducing
power-series expansions for general periodic processes. This becomes possible
because the new temporal argument - the triangular sine wave - is itself
periodic in the original time. Therefore, such expansions can be viewed as
some alternative to Fourier series for processing periodic signals especially
with step-wise discontinuities. Then a periodic version of the Lie series is
introduced. As a result, formal analytical solutions for normal mode motions
of dynamical systems are obtained.
Chapters 6, 7 and 8 describe NSTT based analytical tools for linear, weakly
nonlinear and strongly nonlinear vibrating systems, respectively. In partic-
ular, applying NSTT to linear and weakly nonlinear systems may be very
effective in those cases when nonsmooth loadings are present and thus har-
monic or quasi harmonic approaches require multiple term expansions for
capturing essential features of the dynamics.
4 1 Introduction
Chapters 9 through 11 deal with the concept of nonlinear normal modes.

In particular, it is shown that NSTT leads to adequate formulations of the
normal mode problem for impact systems. Also the idea of nonlinear normal
modes for the case of forced vibration is formulated in terms of NSTT.
Chapter 12 presents a semi-analytical approach combining NSTT with the
shooting method that essentially extends the area of applications.
Chapter 13 describes a possible physical basis for NSTT in case of essen-
tially non-periodic processes.
Finally, Chapter 14 illustrates different applications to spatially oscillating
structures such as one-dimensional elastic rods with periodic discrete inclu-
sions and two-dimensional media with a periodic nonsmooth boundary source
of waves.
1.2 Asymptotic Meaning of the Approach
1.2.1 Two Simple Limits of Lyapunov Oscillator

Let us introduce some preliminaries and remarks based on a one-degree-of-
freedom model as shown at the top of Fig. 1.1.2 This relatively simple model
however depicts the gradual evolution from linear to extremely nonlinear
dynamics as the exponent n runs from unity to infinity. Notably, all the
temporal mode shapes of the oscillator are described by special functions
except two boundaries of the interval 1 ≤ n < ∞. Both boundaries represent
simple asymptotic limits within the class of elementary functions.
Consider first the limit of harmonic oscillator, n = 1, that generates the
sine and cosine waves as illustrated on the left column of the diagram in
Fig. 1.1. From the very general standpoint, a widely known convenience of
using this couple of functions can be explained by their certain link to the
group of rigid-body motions namely the subgroup of rotations. The standard
complex plane representation and the Euler formula can be mentioned here as
related tools. Further, taking the linear combination of harmonic waves with
different frequencies and keeping in mind the idea of parameter variations
invokes the area of harmonic and quasi harmonic analyses for both signal
processing and dynamical systems. Such tools therefore represent complicated
dynamic processes as a combination of the very simple rigid-body rotations
with certain angular speeds.
Let us consider now another limit n = ∞, when the restoring force van-
ishes inside the interval −1 < x < 1 but becomes infinitely growing as the
system reaches the potential barriers at x = ±1. The physical meaning of
this limit is introduced at the top of the right column of the diagram. De-
spite of the strong nonlinearity caused by impacts, the limit oscillator is also
2
Note that oscillators with power-form characteristics were considered for quite a
long time since possibly Lyapunov who obtained such oscillators while investigat-
ing degenerated cases in dynamic stability problems; see Chapter 3 for details.
1.2 Asymptotic Meaning of the Approach 5
.. ..
x x2 n1 0 x x2 n1 0
n1 n
Harmonic oscillator Impact oscillator
Sine and cosine waves Triangular and rectangular waves

Rotation group Translation and reflection groups
1 1 Τt
sin t
t t
Π 2Π 2 4
cos t et
1 1
Conventional complex numbers Hyperbolic complex numbers

z x yi x X Τ Y Τ e

i2 1 e2 1; e Τ t
Fourier series Power series
k0 Ak cos kt Bk sin kt k0 X k 0 Τ k Y k 0 Τ k e

1 1
k k
Quasi harmonic methods Strongly nonlinear methods
Fig. 1.1 Two asymptotic limits described by elementary functions associated with
subgroups of rigid-body motions.
described by quite simple elementary functions such as triangular sine and

rectangular cosine, say τ (t) and τ̇ (t). These two are associated with another
subgroup of the rigid-body motions namely translation and reflection. There-
fore, analogously to the above case n = 1, the upper limit n = ∞ can play
the same fundamental role by generating an hierarchy of alternative to quasi
harmonic tools. On first look, such alternative tools can still be developed
within the paradigm of Fourier expansions by taking appropriate multiple
frequency combinations of τ (t) and τ̇ (t). Although such approaches could
work for signal processing, it is unclear how to deal with a large number of
singularities if substituting the related expansions into differential equations
of motion. Indeed nonsmoothness of such Fourier basis seems to contradict
6 1 Introduction
Fig. 1.2 Complex and hyperbolic planes are shown on the left and right, respec-
tively; in contrast to the circle, each of the hyperbola branches is covered exactly
once as the hyperbolic angle is varying in the infinite interval.
the very language of dynamical systems involving time derivatives.3 In ad-

dition, the Fourier expansions are closely linked to the linear superposition
principle which is inadequate to nonlinear cases.
Finally, the major role of Fig. 1.1 is to convince the reader that any pres-
ence of functions τ (t) and τ̇ (t) in further developed analytical algorithms is
not a simple match of different pieces of solutions4 . On the contrary, it has
its real physical basis and invokes specific mathematical tools.
1.2.2 Oscillating Time and Hyperbolic Numbers,

Standard and Idempotent Basis
So instead let us introduce another formalism dealing with same basic fre-
quency but adjusting the triangular sine shape to the desired one through
3
Generally speaking, it is possible to start with an expansion for high-order deriva-
tives and then come backward to coordinates by integrations. However, algebraic
complexity of such approaches may overshadow any advantages as compared to
the regular Fourier expansions.
4
It is quite clear that harmonic waves can also be interpreted as those combined
of the same pieces of curves, but such a viewpoint would eliminate much of the
vibration theory and many physical effects.
1.2 Asymptotic Meaning of the Approach 7
polynomials and other simple functions of the argument τ . Such formalism

is based on the following statement [133]:
Any periodic process x (t) of the period normalized to T = 4 can be ex-
pressed through the dynamic state of the impact oscillator, {τ (t), τ̇ (t)}, in the
form of ‘hyperbolic complex number’
x = X (τ ) + Y (τ ) τ̇ (1.1)
On the right-hand side, the functions X and Y are easily expressed through
the original function x(t), if this function is known; see Chapter 4 for further
details.
In case x(t) is an unknown periodic motion of some dynamical system,
equations for X and Y components are obtained by substituting (1.1) into
the corresponding differential equation of motion. Then either analytical or
numerical procedures can be applied. For instance, one may seek solutions
in the form of power series with respect to the ‘oscillating time’ τ . Therefore
expression (1.1) can be qualified as non-smooth time transformation, t → τ ,
on the manifold of periodic motions.
Note that the structure of hyperbolic numbers has been known for quite
a long time mostly as a formal extension of the regular complex numbers
with no relation to either vibrating systems or nonsmooth functions; see the
references below. In the mathematical literature, such an extension (which
is often regarded to as a simple particular case of Clifford’s algebras5) is
introduced as follows. Whereas the algebraic equation p2 = 1 has the real
number solutions p = ±1, the existence of a unipotent u is assumed such
that u = +1 and u = −1 but u2 = 1. Then by considering the elements
{1, u} as a standard basis, any hyperbolic number w ∈ H can be written in
the form w = x + uy, where x and y are real numbers [176]. √ The hyperbolic
conjugate of w is defined by w− = x−uy so that |w|H = ww− is the norm of
w. Fig. 1.2 illustrates the difference between complex and hyperbolic planes.
More details and some, rather abstract, applications of this algebraic theory
can be found in references [90], [176], [7], [187]. As to applied areas, the same
kind of algebraic structures occurred in hydrodynamics in connection with
characteristics of partial differential equations [90].
In our case, the unipotent is not a number but the discontinuous function
of certain physical nature i.e. the rectangular cosine wave e(t). Indeed since
t is running then there is no unique choice for the magnitude of e, whereas
always6 e2 = 1. Therefore, identity (1.1) generates the hyperbolic structure
from the very general properties of periodic processes.
Note that both terms on the right-hand side of (1.1) are essential as those
responsible for components with different temporal symmetries. For instance,
5
William Kingdon Clifford (1845-1879), English mathematician who, in particular,
developed the idea that space may not be independent of time.
6
More precisely, for almost any t.
8 1 Introduction
Fig. 1.3 Geometrical interpretation of the particular case with a sine-wave tem-
poral symmetry: observing the coordinate x does not allow to conclude which of
the two temporal variables, τ or t, is actually ‘running.’
Fig. 1.3 illustrates geometrical meaning of the particular case of temporal

symmetry with no Y -component, where x(t) ≡ x[aτ (t/a)] = X(τ ).
It is important to note that, under some conditions on X and Y , combi-
nation (1.1) can be of any class of smoothness even though the couple {τ, τ̇ }
has singularities at such time instances t where τ = ±1.
Finally, let us mention that the hyperbolic plane has another natural basis
associated with the two isotropic lines separating the hyperbolic quadrants
as shown in Fig. 1.2. The transition from one basis to another is given by
e± = (1 ± e)/2 or, inversely, 1 = e+ + e− and e = e+ − e− . Therefore,
x = X + Y e = X(e+ + e− ) + Y (e+ − e− )
= (X + Y )e+ + (X − Y )e−
≡ X+ (τ )e+ + X− (τ )e−
where x = x(t) any periodic function as defined in (1.1). whose period is

normalized to T = 4.
On one hand, the advantage is that the elements e+ and e− are mutually
annihilating (idempotents) so that e+ e− = 0, e2− = e− and e2+ = e+ . It is
clear also that ee+ = e+ and ee− = −e− . Due to the annihilation property,
the idempotent basis significantly eases different algebraic manipulations, for
instance,
(X+ e+ + X− e− )2 = X+ 2
e+ + X−2
e−
On the other hand, this basis usually couples the corresponding smoothness
(boundary) conditions; see Chapter 4 for further details and examples.
1.3 Quick ‘Tutorial’ 9
1.3 Quick ‘Tutorial’

1.3.1 Remarks on the Basic Functions
First note that the basic functions, τ (t) and τ̇ (t) are expressed through the
standard elementary functions in the closed form as
2 πt
τ (t) = arcsin sin (1.2)
π 2
and −1
πt πt
τ̇ (t) = cos cos = e(t) (1.3)
2 2
Obviously τ (t) is a triangular sine wave whose amplitude is unity and the
period is T = 4, whereas e(t) is a rectangular cosine wave with step-wise
discontinuities as shown in Fig. 1.1. Therefore, expression (1.3) holds ‘al-
most everywhere.’ For calculating purposes, it can be represented in the form
e(t) = sgn [cos(πt/2)]. Practically, for given t, one can calculate the functions
τ (t) and e(t) based on their piecewise-linear graphs even with no tables nor
calculators involved.
Necessary mathematical details regarding the discontinuities of e(t) are
discussed later; see also [144]. At this stage however, solutions can be com-
pared with exact solutions obtained in a different way or numerical ones in
order to validate the technique.
Moreover, during the derivations, there is no need in keeping in mind ex-
pressions (1.2) and (1.3). It is sufficient to take into account linear indepen-
dence of the elements 1, e and ė, as those from different classes of smoothness,
and the following properties
τ̇ = e
∞

ė = 2 [δ (t + 1 − 4k) − δ (t − 1 − 4k)] (1.4)
k=−∞
e2 = 1
where, strictly speaking, the equalities should be interpreted in terms of the

distribution theory due to the presence of singularities that occur whenever
τ = ±1.
Let us consider few examples illustrating the technique and giving a hint
regarding those cases when the NSTT is reasonable to apply.
1.3.2 Viscous Dynamics under the Sawtooth Forcing

Consider first the very simple one-dimensional case. A light particle in a vis-
cous fluid is subjected to the sawtooth force so that the differential equation
of motion is reduced to
10 1 Introduction
ẋ(t) = qτ (t) (1.5)

where q represents the force amplitude per unit coefficient of viscosity.
The general solution of equation (1.5) is given by
t
x(t) = x(0) + q τ (ϕ)dϕ (1.6)
0
where x(0) is an arbitrary initial position, and a part of the problem, which
is calculating the integral, still persists.
Now let us represent the particular (periodic) solution in the form
x = X(τ ) + Y (τ )e (1.7)
where τ = τ (t) and e = e(t).

Substituting (1.7) in (1.5) and taking into account properties (1.4), gives
Y (τ ) − qτ + X (τ )e + Y (τ )ė = 0 (1.8)
Based on the linear independence of 1, e and ė, equation (1.8) is equivalent

to the boundary value problem
Y (τ ) = qτ , X (τ ) = 0, Y (±1) = 0 (1.9)
where the boundary condition provides zero factor for all δ−functions of the
derivative ė.
In contrast to (1.5), the integration of equation (1.9) is strightforward since
the variable of integration is τ . So substituting obvious solution of problem
(1.9) into (1.7), gives general solution of equation (1.5)
q
x = C + (τ 2 − 1)e (1.10)
2
where C is an arbitrary constant.
Then taking into account that τ (0) = 0 and e(0) = 1, gives x(0) = C − q/2
and therefore solution (1.10) takes the form
q q
x = x(0) + + (τ 2 − 1)e (1.11)
2 2
As compared to the direct approach (1.6), the NSTT allowed for conducting
the integration of the differential equation of motion with no dealing with the
piece-wise structure of the integrand. Moreover, comparing solutions (1.6)
and (1.11), gives the result of direct integration (1.6) in the form
t
1
τ (ϕ)dϕ = {1 + [τ 2 (t) − 1]e(t)} (1.12)
0 2
As a simple test, expression (1.12) can be verified by taking first derivative

of the both sides.
Note that there are two boundary conditions for Y in (1.9). Since the
Y − component of the solution appeared to be even with respect to τ , then
both of the conditions are satisfied even though one arbitrary constant only
is available for Y . If the forcing function in (1.5) were of even degree with
respect to τ , for instance, qτ 2 , then the corresponding boundary-value prob-
lem would have no solution. This fact is explained by the absence of periodic
solutions under the loading qτ 2 , which has a non-zero mean value, whereas
representation (1.7) imposes periodicity on the function x(t).
However, the NSTT is applicable in case of any odd degree polynomial of
τ on the right-hand side of equation (1.5),

n
ẋ(t) = qk τ 2k−1 (t) (1.13)
k=1
where qk are constant coefficients.

Analogously to the above particular case, taking into account the modified
equation,

n
Y (τ ) = qk τ 2k−1
k=1
gives periodic solution

n

n
qk qk
x = x(0) + + (τ − 1) e
2k
(1.14)
2k 2k
k=1 k=1
This solution is verified by the direct substitution of (1.14) in (1.13).
1.3.3 The Rectangular Cosine Input

Consider the case of step-wise discontinuous periodic loading
ẋ(t) = pe(t) (1.15)
Substituting (1.7) in (1.15), gives
Y + (X − p)e + Y ė = 0
The boundary value problem therefore takes the form
Y (τ ) = 0, X (τ ) = p, Y (±1) = 0 (1.16)
In this case, Y ≡ 0 and the solution of equation (1.15) is
x = x(0) + pτ (t) (1.17)

12 1 Introduction
In this simple case, the above solution could be written though based on the
definition of e(t) (1.3). As a generalization of the right-hand side of equation
(1.15), let us consider equation
m

ẋ(t) = pi τ i (t) e(t) (1.18)
i=0
where pi are constant coefficients.

In this case, the periodic solution
m
pi i+1
x = x(0) + τ (1.19)
i=0
i + 1
does exist for both odd and even exponents of the polynomial in (1.18).
Finally, combining the right-hand sides of equations (1.13) and (1.18) and
considering the corresponding infinite series, gives
∞
∞

f (t) = qk τ 2k−1
(t) + pi τ (t) e(t) ≡ Q(τ ) + P (τ )e
i
(1.20)
k=1 i=0
It will be shown later expansion (1.20) represents a very general class of

zero mean periodic functions with the period normalized to T = 4. The
corresponding periodic solution of the equation ẋ(t) = f (t) is obtained by
combining solutions (1.14) and (1.19).
1.3.4 Oscillatory Pipe Flow Model

As a possible application, consider a simplified model of pipe flow driven by a
regularly repeating rectangular pressure wave [183]. The pipe flow is assumed
to have lumped inertance L and quadratic-law resistance with the coefficient
K. The flow Q(t) is described by first-order non-linear differential equation
LQ̇ = P0 + P1 e(t/a) − KQ2 (1.21)
where P0 and P1 are constants characterizing the pressure drop, and 4a is

the period of the rectangular pressure wave.
Introducing parameters k = K/L, p0 = P0 /L and p1 = P1 /L, brings
equation (1.21) to the form
Q̇ + kQ2 = p0 + p1 e(t/a) (1.22)
Temporal behavior of the flow essentially depends on the model parameters

and initial conditions. Let us assume however that the flow becomes even-
tually periodic with the period of rectangular pressure wave. The objective
is to find the average steady state flow. The corresponding periodic solution
can be represented in the form
Q(t) = X(τ ) + Y (τ )e (1.23)
where τ = τ (t/a) and e = e(t/a) are triangular and rectangular waves,

respectively, with the period T = 4a.
a−1 (Y + X e + Y e ) + k(X 2 + Y 2 + 2XY e) = p0 + p1 e (1.24)
where e = de(t/a)/d(t/a), and therefore
Y + ak(X 2 + Y 2 ) = ap0
X + 2akXY = ap1 (1.25)
Y (±1) = 0
Introducing the new unknowns U = X + Y and V = X − Y , brings the

boundary value problem (1.25) to the form
U + akU 2 = aF
V − akV 2 = −aG (1.26)
U (±1) − V (±1) = 0
where F = p0 + p1 and G = p0 − p1 are constant.

Both equations in (1.26) are separable and thus admit general solutions of
the form
√
F 2 exp(−2a kF τ )
U (τ, C1 ) = 1− √
k C1 + exp(−2a kF τ )
√
G 2 exp(−2a kGτ )
V (τ, C2 ) = − 1− √ (1.27)
k C2 + exp(−2a kGτ )
where C1 and C2 are arbitrary constants of integration to be determined from

the boundary conditions
U (1, C1 ) − V (1, C2 ) = 0
U (−1, C1 ) − V (−1, C2 ) = 0 (1.28)
Equations (1.28) appear to be nonlinear algebraic set of equations that, gen-

erally speaking, admits multiple solutions of which some solutions may be
real, whereas others are complex. Each real solution for the constants C1
and C2 gives a periodic solution of differential equation (1.22). However, sim-
ple numerical tests show that some of the periodic solutions may appear to
be unstable. (Detailed parametric study and stability analysis go beyond of
this introductory illustration.) After a set of the arbitrary constants has been
14 1 Introduction
determined, the pipe flow function can be represented in either standard or

idempotent basis as follows
1 1
Q(t) = X + Y e = [U + V ] + [U − V ]e
2 2
= U (τ, C1 )e+ + V (τ, C2 )e− (1.29)
where τ = τ (t/a), and e± = [1 ± e(t/a)]/2 are elements of the idempotent

basis; algebraic properties and further details of transition from one basis to
another are discussed in Chapter 4.
Also, it will be shown in Chapter 4 that, temporal mean value of a periodic
function can be found by averaging its X(τ )-component with respect to τ .
Applying this statement to solution (1.29), gives the average steady state
flow
T 1
1 1 G 1F
< Q >≡ Q(t)dt = −
X(τ )dτ =
T 2 k 2 k
0 −1
√ √
1 [1 + C1 exp(2a kF )][1 + C2 exp(−2a kG)]
+ ln √ √ (1.30)
4ak [1 + C1 exp(−2a kF ][1 + C2 exp(2a kG)]
10 T8
9 T4
Q 8
0.0 0.5 1.0 1.5 2.0

tT
Fig. 1.4 Profiles of steady state pipe flows obtained for two different periods of
the rectangular pressure wave.
Fig. 1.4 shows what happens to the steady state flow profile as the period
of pressure wave becomes twice longer. The model parameters are k = 0.03,
p0 = 2.0, and p1 = 1.5. In cases T = 4 and T = 8, the arbitrary constants are
C1 = 8.2248, C2 = −0.3125 and C1 = 11.7200, C2 = −0.2651, respectively.
Note that the average flow for the period T = 4 is < Q >= 8.12292, whereas
the longer period T = 8 gives somewhat smaller average, < Q >= 8.02996.
1.3.5 Periodic Impulsive Loading

Let us consider the one-dimensional motion of a material point under the
periodic impulsive loading inside the linearly viscous fluid. The corresponding
differential equation of motion can be represented in the form
∞

v̇ + λv = 2p [δ (t + 1 − 4k) − δ (t − 1 − 4k)] = pė (1.31)
k=−∞
where λ and p are a coefficient of viscosity and a half of the pulse amplitude
per unit mass, respectively.
Equation (1.31) has a simple form involving the distribution. Such kind
of problem is usually solved by applying either the generalized Fourier series
or Laplace transform, or by considering the equation between the pulses by
matching different pieces of the solution at the pulse times. Alternatively, the
solution can be obtained in few quick steps by using identity (1.1).
Indeed, substituting (1.1) in (1.31) and taking into account (1.4), gives
(Y + λX) + (X + λY ) e + (Y − p)ė = 0
or
Y + λX = 0, X + λY = 0, Y |τ =±1 = p (1.32)
Solving boundary value problem (1.32), gives the closed form particular
solution
p
v=− (sinh λτ − e cosh λτ ) (1.33)
cosh λ
The entire general solution is obtained by adding the term with an arbitrary
constant, C exp(−λt), due to linearity of equation (1.31).
1.3.6 Strongly Nonlinear Oscillator

Now let us consider the nonlinear oscillator shown in Fig. 1.1,
ẍ + x2n−1 = 0 (1.34)
where n is an arbitrary positive integer.

Note that this example gives the asymptotic basis for introducing the saw-
tooth temporal argument τ justified by the limit n → ∞, in which x(t) takes
the triangular sine shape [132]. The root of the mathematical problem here
is that the limit is nonsmooth, whereas large but finite numbers n generate
close to the sawtooth but smooth oscillations. Let us show that changing the
temporal variable t → τ facilitates a natural transition to the limit n → ∞.
Taking into account the temporal symmetry of the oscillation, reduces the
substitution to the particular case X(−τ ) ≡ −X(τ ) and Y ≡ 0 so that
16 1 Introduction
x = X(τ ), τ = τ (t/a) (1.35)

where a = T /4 is an unknown quarter of the period.
Substituting (1.35) in (1.34) and taking into account (1.4), gives
1
(X + X e ) + X 2n−1 = 0 (1.36)
a2
Due to the oddness of X(τ ), equation (1.36) is equivalent to a one-point
boundary value problem
X = −a2 X 2n−1 , X |τ =1 = 0 (1.37)
Now the idea is to take advantage of the fact that the new temporal argument
is periodic and bounded, −1 ≤ τ ≤ 1. So, in order to design an analogous
of the quasi harmonic approaches, successive iterations can be applied. This
tool requires no small parameter to be present explicitly. As a result, one gen-
erally must rely on convergence of the procedure. Although the convergence
is usually slow7 , the physical basis of the convergence can always be deter-
mined whenever the problem has indeed some physical content. Note that the
harmonic balance method also uses no explicit small parameter by assuming
however that temporal mode shapes are close to harmonic. In other words,
high frequency terms just correct but not exceed the principal harmonic. In
the present case, solution is approximated by the triangular sine, which is
corrected by higher powers of the same triangular sine. On the physical point
of view, the model under consideration has to be close to the impact oscillator
rather then the harmonic one. In terms of the new time variable τ , such an
assumption simply means that the right-hand side of the differential equation
of motion (1.37) is small enough to justify the following generating system
X0 = 0 (1.38)
Indeed, this equation describes a family of impact oscillators with the trian-
gular sine wave time histories
X0 = Aτ (t/a) (1.39)
where A is an arbitrary constant and another constant is zero due to the

symmetry X(−τ ) ≡ −X(τ ).
There are different ways to formalizing the entire procedure. For instance,
next term of the series can be obtained by substituting the generating solution
(1.39) into the right-hand side of equation (1.37) and then integrating twice
so that
τ 2n+1
X1 = Aτ − a2 A2n−1 (1.40)
2n(2n + 1)
7
This is rather a side effect of the generality of successive approximation
techniques.
1.4 Geometrical Views on Nonlinearity 17
Note that the linear term Aτ occurred again as a result of integration in

(1.37). Keeping the same arbitrary constant A automatically includes gener-
ating solution (1.39) into first-order approximation. This represents a certain
mathematical constraint, which actually enables one of determining the pa-
rameter a by satisfying the boundary condition in (1.37). This gives
2n
a2 = (1.41)
A2n−2
and therefore

τ 2n+1
X1 = A τ − (1.42)
2n + 1
High-order algorithms and the corresponding error estimates are described in
Chapter 8. In particular, expression (1.41) is sequentially improved as follows

2 2n 2n − 1
a = 2n−2 1 +
A 4n + 2
Note that analogous successive approximation algorithms can be applied to

equation (1.37) with a quite general right-hand side. In the above case, how-
ever, the physical meaning of the parameter, n, shown in Fig. 1.1, directly
relates to the idea of algorithm. As a result, solution (1.42) performs better
as the exponent n increases.
Note that, a physically meaningful transition to the asymptotic limit n →
∞ can be implemented by replacing the parameter A in (1.42) with the initial
velocity v0 = A/a which in contrast to A is independent of n.
Finally, despite of the manipulations with nonsmooth and discontinuous
functions, solution (1.42) is twice continuously differentiable with respect to
the argument t; this can be verified by taking first two formal derivatives of
(1.42).
Generally speaking, the boundary value problem, such as (1.37), may
appear to be complicated for any analytical method. In such cases, a com-
bination of NSTT with the shooting method can be effectively used as a
semi-analytical approach [153], [144].
1.4 Geometrical Views on Nonlinearity
1.4.1 Geometrical Example

As shown in the previous sections, the asymptotic of linearity has a strongly
nonlinear but simple enough counterpart so that both may complement each
other as possible approaches to vibration problems. In other words, two cou-
ples of basic functions, {sin t, cos t} and {τ (t), e(t)}, naturally associate with
opposite boundaries of parameter intervals of certain physical systems. Note
that the oscillator (1.34) is not a unique example showing such a property.
18 1 Introduction
Fig. 1.5 Rigid-body rotation that generates both sine and triangular waves within
the same class of elementary functions.
Another example is given by a strongly nonlinear oscillator which is exactly

solvable in terms of elementary functions [78], [122]. Below, a geometrical
interpretation for this case is introduced.
Let us consider a geometrical model generating the sine and triangular
sawtooth waves as two asymptotic limits of the same family of periodic func-
tions. With reference to Fig. 1.5, the distance between two fixed points C
and O is equal to unity. A disc of the radius CP = α is rotating around its
geometrical center C with some angular speed ω so that the angle between
CO and CP is ϕ = ωt. The edge point P of the rotating disc has its image
Q on the co-centered circle of the unit radius. The image is obtained under
the condition that, during the rotation, P Q remains parallel to CO. Then
position of the image Q is given by either its Cartesian coordinate Y or by
the arc length y as follows
Y = α sin ϕ (1.43)
y = arcsin(α sin ϕ) (1.44)
Expression (1.44) is obtained by equating projections of CP and CQ on OY
and taking into account that the angle QCO is equal to the corresponding
arc length y.
Obviously, (1.44) becomes equivalent to (1.43) and thus describes the har-
monic sine wave as α → 0.
Suppose now that α → 1 − 0. In this case, (1.44) takes the triangular sine
wave shape

π 2ϕ
y→ τ (1.45)
2 π
Interestingly enough, both types of observation of the above rigid-body model
are associated with different elastic oscillators. Namely, on one hand, as a
function of time, expression (1.43) satisfies equation
Ÿ + ω 2 Y = 0 (1.46)
On the other hand, function y(t) (1.44) satisfies the differential equation of
motion of conservative oscillator
tan y
ÿ + =0 (1.47)
cos2 y
under condition,
α2 + ω −2 = 1 (1.48)
In particular, condition (1.48) implies that the frequency ω depends upon
the disc radius monotonically in such a way that 1 ≤ ω < ∞ as 0 ≤ α <
1. Therefore, nonsmooth limit (1.45) is reached under the infinitely large
frequency ω, when the total energy of oscillator (1.47) also becomes infinitely
large,
ẏ 2 tan2 y
H= + →∞ (1.49)
2 2
The reasons for considering oscillator (1.47) as a generating model in non-
linear dynamics will be described later. At this stage, let us bring attention
to the fact that both equations (1.46) and (1.47) actually describe the same
rigid-body rotation but in different coordinate systems.
1.4.2 Nonlinear Equations and Nonlinear Phenomena

The purpose of next few subsections (1.4.2 through 1.4.6) is to find some basis
for the idea of using elementary nonsmooth systems as generating systems
in nonlinear dynamics. The discussion below may provide a hint for selecting
nonlinear generating systems within the class of nonsmooth systems even
though some statements of the discussion may have no rigorous proof.
As a starting point, it could be useful to clarify why the harmonic oscillator
represents a good generating model. Indeed, defining the nonlinearity regard-
less mathematical formalizations appears to be a challenging task. Namely
it is difficult to find appropriate physical principals for such a definition that
would qualify nonlinearity as a natural phenomenon rather than the specific
form of mathematical expressions. Briefly speaking, nonlinear phenomena
may occur at high energy levels. The following scenario is described in the
book [126]: “...once the power or violence of a system is increased, it leaves
20 1 Introduction
the familiar linear region and enters the more complex world of nonlinear
effects: rivers become turbulent, amplifiers overload and distort, chemicals
explode, machines go into uncontrollable oscillations, plates buckle, metals
fracture, and structures collapse.” The author goes even further by saying
that “within such an approach, mind may no longer appear as an alien stuff
in a mechanical universe; rather the operation of mind will have resonances
to the transformations of matter, and indeed, the two will be found to emerge
from a deeper ground.”
However, mathematical approaches can still be helpful based on the fol-
lowing logical identity
Nonlinear systems = All systems - Linear systems (1.50)
Although this relationship seems to bring no physical contents, nevertheless

it shows how physical understanding of nonlinearity can possibly be achieved.
In other words, by clarifying the physical basis that provides the well known
simplicity of the linear systems and harmonic analyses, one could find what
actually makes systems nonlinear.
The standard mathematical definition for linear systems based on the su-
perposition principle. Generally, a linear system can be defined in terms of
operators. Let L be an operator acting on state or may be coordinate vectors
{q}. Then the operator L is linear, if for any two states, say q1 and q2 , and
any two constants, C1 and C2 , the following relationship holds
L(C1 q1 + C2 q2 ) = C1 Lq1 + C2 Lq2 (1.51)
Now, according to (1.50), a system is nonlinear if it is not linear. This math-

ematical definition, however, may appear to be confusing from physical view-
points. For instance, based on definition (1.51), system
2
ρ̈(t) − ρ(t) ϕ̇(t) = ρ(t) [sin 2 ϕ(t) − 2]
ρ(t)ϕ̈(t) + 2ρ̇(t) ϕ̇(t) = ρ(t) cos 2 ϕ(t) (1.52)
does not satisfy its conditions and therefore is nonlinear, whereas
ẍ1 (t) + 2x1 (t) − x2 (t) = 0

ẍ2 (t) + 2x2 (t) − x1 (t) = 0 (1.53)
is a linear system whose operator reads

2 2
d /dt + 2 −1
L= (1.54)
−1 d2 /dt2 + 2
so that Lq = 0, where q = [x1 (t) , x2 (t)]T .

The linear superposition principle is therefore applicable to equations
(1.53) whereas it does not work for equations in the nonlinear form (1.52).
However, both sets of equations (1.52) and (1.53) describe the same mechan-
ical system in polar and Cartesian coordinates, respectively. On the plane of
configurations, the corresponding coordinate transformation is given by
x1 = ρ cos ϕ, x2 = ρ sin ϕ
In this example, it is clear that the system is linear from the physical point
of view; indeed nonlinearity of equations (1.52) is due to specific choice for
coordinates. Therefore a ‘physical definition’ for linear systems must specify
the type of reference coordinate system, for instance, as follows:
A mechanical system is linear if its differential equations of motions in
Cartesian coordinates are linear 8 .
This definition involves already some physical principles since the Cartesian
coordinates are uniquely associated with general properties of the physical
space. Note that the above definition still involves the mathematical notion
of coordinate systems. As known from observations however, some dynamic
phenomena are perceived as nonlinear with no explicit system coordinates.
So next two subsections make an attempt to clarify why such a perception
of nonlinearity is actually possible.
1.4.3 Rigid-Body Motions and Linear Systems

As mentioned at the beginning of this introduction, sine and cosine waves
are associated with the subgroup of rigid-body rotations. On the other hand,
same temporal shapes describe vibrations of deformable linearly elastic bodies
such as harmonic oscillators. Therefore, one-dimensional dynamics generated
by linearly elastic restoring forces can be represented as free rigid-body rota-
tions however in the two-dimensional space. In other words, linearly elastic
forces are effectively eliminated by expanding the dimension of space. There-
fore, linear dynamics can be viewed as kinematics of freely rotating discs,
where the number of discs is the number of modes. Such a viewpoint can be
formalized as follows.
Let z be a complex vector frozen into a rigid body (disc), whose position is
observed in the empty space. Note that the notion of ‘position’ becomes quite
vague if there is only one body. Therefore it is assumed that the observer
represents another physical body, which is a single point. Then a straight
line between this point and the center of the disc gives a reference line for
determining the vector z direction.
Since the disc is assumed to be rigid then
z z̄ = z z̄ (1.55)
8
This definition was suggested by V.Ph.Zhuravlev during a private discusion at
International Conference “Nonlinear Phenomena,” Moscow, 1989.
22 1 Introduction
where the over bar means complex conjugate, and the prime denotes any new
position of the body.
Expression (1.55) recalls mathematical objects generated by the Galilean
group of rotations. The corresponding operator G is introduced through re-
lationship
z = Gz (1.56)
where G must depend on some parameter, say ϕ, characterizing the transition
z → z.
2
GḠ = |G| = 1 =⇒ G = exp(iϕ), i2 = −1 (1.57)
Since ϕ is the only parameter of the only one ‘process’ then ϕ must be
qualified as time with possibly some scaling factor. Introducing an arbitrary
scaling factor, say ω, gives then
G = exp (iωt) = cos ωt + i sin ωt (1.58)
Now a simple calculation shows that

d d
− iω G = 0 and + iω Ḡ = 0
dt dt
Therefore both operators, G and Ḡ, satisfy the differential equation of har-
monic oscillator associated with the operator

d d d2
− iω + iω = 2 + ω 2 (1.59)
dt dt dt

By representing the scaling factor in the form ω = k/m, one can interpret
then m and k as, respectively, mass and stiffness of a mass-spring model
associated with the ‘product’ of clockwise and counter-clockwise rotations of
the disc.
Let us consider now n rotating discs with the angular frequencies
{ω1 , ..., ωn } by generalizing product (1.59) as follows

n
2
d d d d d
− iω1 + iω1 ... − iωn + iωn = + ω 2
j
dt dt dt dt j=1
dt2
(1.60)
Operator (1.60) represents an arbitrary n-degrees-of-freedom linear elastic
system. Replacing d/dt → λ, gives the corresponding characteristic equation
in the form
n
2
λ + ωj2 = 0 (1.61)
j=1
Finally, let us consider the asymptotic limit n → ∞. In order to calculate

the limit, the frequency dependence on its index j must be specified. If, for
instance, ωj = ja with some constant a then equation (1.61) gives
πλ
sinh =0 (1.62)
a
as n → ∞.
Equation (1.62) associates with a linearly elastic string of the length l = π
with two fixed ends,
∂ 2 u (t, x) ∂ 2 u (t, x)
2
− a2 = 0; u (t, 0) = u (t, π) (1.63)
∂t ∂x2
Therefore, it is shown that basic linearly elastic dynamic models can be logi-
cally obtained from the kinematics of freely rotating discs or even from a more
fundamental concept given by expression (1.55). To some extent, this may
explain why nonlinearity is perceptible as a physical phenomenon. Indeed the
linear dynamics is simply matching with the usual perception (experience) of
the space. All other motions give therefore an impression of unusual, in other
words, nonlinear phenomena.
1.4.4 Remarks on the Multi-dimensional Case

The multi-dimensional example below gives another (purely geometrical)
viewpoint on the link between rigid-body motions and the notion of linearity.
So, by definition, the mapping f : Rn −→ Rn is an isometry if relation

f (v) − f (w)
=
v − w
(1.64)
holds for all v, w ∈ Rn .

Definition (1.64) requires distances between any two images and their pre-
images be same; this is a generalization of relationship (1.55). Let us consider
such isometries with a fixed point O in Rn , that is symbolically f (O) = O.
Let us show that an isometry that fixes the origin is a linear mapping
f (av + bw) = af (v) + bf (w) (1.65)
where a and b are arbitrary scalars.

First, note that the relations

f (v)
=
v
=
−v
=
f (−v)
and

f (v) − f (−v)
=
v− (−v)
= 2
v
hold due to (1.64), and thus f (v) and f (−v) are antipodal: f (−v) = −f (v).
24 1 Introduction
Second, taking into account the latter result and definition (1.64), trans-
forms the polarization identity as follows
1 2 2 2

v·w =
v + w
−
v
−
w
2
1
=
f (v) − f (−w)
2 −
f (v)
2 −
f (−w)
2
2
1 2 2 2

=
f (v) + f (w)
−
f (v)
−
f (w)
2
= f (v) · f (w)
Therefore, the inner product is preserved. Now, if {ui }is orthogonal basis in
Rn , then {f (ui )} is another orthogonal basis. Taking the dot product of the
both sides of identity (1.65) with the arbitrary basis vector f (ui ), gives
f (av + bw) · f (ui ) = af (v) · f (ui ) + bf (w) · f (ui )
or
(av + bw) · ui ≡ av · ui + bw · ui
The above relation proves identity (1.65) in terms of the coordinates associ-
ated with the basis {f (ui )}.
All linear isometries of Rn are denoted by the symbol O (n). By choosing
a basis for Rn , one can represent every element of O (n) as a matrix. It can
be shown that O (n) consists of all n × n matrixes such that A−1 = AT :

1 = det I = det AA−1 = det AAT
= det A det AT = (det A)2 (1.66)
Relationship (1.66) is therefore a multidimensional matrix analogy of the

relationship (1.57).
1.4.5 Elementary Nonlinearities

Let us consider now the temporal coordinate of Galilean spatio-temporal
continuum. In particular, the idea of ‘perfectly rigid time’ is expressed by
relationship

2
dt
dt2 = dt2 ⇐⇒ =1 (1.67)
dt
By considering (1.67) as a differential equation with respect to t = s (t), one
obtains two possible solutions
t = ± (t − a) (1.68)
where a is an arbitrary temporal shift.

Now, combining different branches of (1.68) for t < a and t > a, gives
t = s (t) = |t − a| (1.69)
Function (1.69) also satisfies equation (1.67), for all t except may be single
point t = a, where the classic derivative of s (t) has no certain value. Namely,
equality holds for almost all t except may be t = a.
Nevertheless, solution (1.69) admits an obvious physical interpretation
since it describes a free material point moving in the space splitted into two
subspaces by a perfectly stiff plane, were the normal to the plane velocity is
scaled by
ṡ2 = 1 (1.70)
Since the perception of empty space rejects any built in stiff planes then a
sudden V -turn of the particle will represent unusual, in other words, strongly
non-linear event. This is obviously the most elementary nonlinear event
whose simplicity nevertheless will be employed further in less trivial cases.
1.4.6 Example of Simplification in Nonsmooth Limit

The idea that transition to most severe nonlinearity may actually simplify
a system response finds its support in very different examples. Consider, for
instance, the differential equation of motion

2x
ẍ = exp − 2 (1.71)

The nonlinear force on the right-hand side gives the ‘rigid body’ limit as the
parameter approaches zero. Let us assume that ẋ = −1 and x → ∞ as
t → −∞. In this case, equation (1.71) has exact solution

t−a
x = 2 ln cosh 2 (1.72)

where a is an arbitrary parameter such that ẋ(a) = 0; see Fig. 1.6, where
a = 1.
Now let us consider the asymptotic limit → 0. Taking into account
evenness of cosh-function, brings solution (1.72) to the form

2 |t − a| 2 s
x = ln cosh = ln cosh (1.73)
2 2
This simple manipulation represents a useful preliminary step for asymptotic

estimates because the new temporal argument, s, remains always positive as
the original time runs in the interval −∞ < t < ∞.
26 1 Introduction
x
2.0
Ε0.1
1.5 Ε0.5
Ε1.0
1.0
0.5
t
1 1 2 3
Fig. 1.6 Time histories of the particle when the barrier approaches rigid body
limit, → 0.
Further algebraic manipulations bring solution (1.73) to the form

2s
2
x = s + ln 1 + exp − 2 (1.74)
2

Taking into account that 0 < exp −2s/2 < 1, gives the ‘rigid-body’ limit
x → s = |t − a| (1.75)
as → 0.
The above example shows that (1.69) can be viewed as a natural time
associated with the temporal symmetry of the V-turn. Note however that
the new temporal argument is nonsmooth and non-invertible with respect to
the original time. Therefore, using (1.69) for temporal substitutions in the
differential equations of motion has certain specifics as discussed below.
1.4.7 Non-smooth Time Arguments

As mentioned above, piece-wise linear function (1.69), s(t), can play the
role of natural temporal argument associated with the elementary nonlin-
ear events such as dynamic V-turns caused by potential barriers. Namely the
temporal symmetry of V-turns is captured by the function s(t) regardless
possible shapes of the potential barriers. Therefore, using the function s(t)
as a temporal argument should simplify analyses since the global informa-
tion about V-turns is a priory built into the new argument. The correspond-
ing time substitution however is not straightforward since its inverse version
does not exist in the form t = t(s) but requires the following generalization
t = t(s, ṡ). Below, this case is discussed first. Then the periodic version is
introduced that reveals structural similarity of expressions as far as periodic

motions are in fact regular sequences of v-turns.
Non-periodic case. Let us start with the inverse version of (1.69), which
includes both the image s and its time derivative ṡ as follows
t = t(s, ṡ) = a + sṡ (1.76)
Now relation (1.70) reveals that combination (1.76) represents specific al-
gebraic structure with the basis {1, ṡ}. In other words, time belongs to the
algebra of hyperbolic ‘complex numbers’ with the table of products gener-
ated by (1.70). For instance, taking (1.76) squared, gives another hyperbolic
‘number’
t2 = a2 + s2 + 2asṡ (1.77)
Moreover, it is easy to prove for any function x(t) that
x (t) = X (s) + Y (s) ṡ (1.78)
where
1
X (s) = [x (a + s) + x (a − s)]
2
1
Y (s) = [x (a + s) − x (a − s)]
2
For instance, setting a = 0 in the case x(t) = exp t, gives
exp (ṡs) = cosh s + ṡ sinh s (1.79)
In contrast to the conventional complex algebra, division is not always pos-

sible, so the following relationship is meaningless for |a| = s or t = 0:
1 1 (a − sṡ) a s
= = = 2 − 2 ṡ
t a + sṡ (a + sṡ) (a − sṡ) a − s2 a − s2
In particular case of even function x (t), with respect to t = a, one has

Y (s) ≡ 0; see (1.73) for example.
Periodic case. The periodic version of nonsmooth time transformations as
that preliminary introduced in Section 1 involves triangular and rectangular
waves described by the periodic piecewise-linear function

2 πt t for − 1 ≤ t ≤ 1 ∀t
τ (t) = arcsin sin = , τ (t) = τ (4 + t)
π 2 −t + 2 for 1 ≤ t ≤ 3
and its Schwartz derivative e(t) = τ̇ (t); see Fig. 1.1.

Remind that from the physical point of view, the functions τ (t) and e (t)
describe dynamic states of a particle oscillating between two perfectly rigid
28 1 Introduction
Fig. 1.7 Mechanical model generating the saw-tooth sine and rectangular cosine,
τ (t) and τ̇ (t)
planes with no energy loss as shown in Fig. 1.7, where spatio-temporal coor-
dinates are normalized so that
dt2 = dτ 2 ⇐⇒ e2 = τ̇ 2 = 1 (1.80)
Note that relationship (1.80) is a periodic version of (1.70) whereas, for any
periodic function of the period T = 4, representation (1.1) is a periodic
version of (1.78). In other words, any periodic motion uniquely associates
with the standard impact vibration, for instance as follows
πt πt πτ πτ
x (t) = A sin + B cos = A sin + B cos e (1.81)
2 2 2 2
where τ = τ (t), e = τ̇ (t), and A and B are arbitrary constants.
Exercise 1. By taking sequentially derivatives, show that differentiation
keeps the result within the set of hyperbolic numbers provided that the cor-
responding process is smooth enough. In particular,
πτ πτ π 2n−1 πτ πτ
d2n−1 A sin + B cos e /dt2n−1 = −B sin + A cos e
2 2 2 2 2
πτ πτ
π 2n πτ πτ
d2n A sin + B cos e /dt2n = A sin + B cos e
2 2 2 2 2
1.4.8 Further Examples and Discussion

The Fourier analysis is based on the standard trigonometric pair of functions
{sin ωt, cos ωt} or {exp (iωt) , exp (−iωt)} so that periodic processes are de-
scribed by linear combinations of these functions with appropriate frequencies
{ωk }. For instance, the dynamic states of impact oscillators are represented
as follows

8 πt 1 3πt 1 5πt
τ (t) = 2 sin − 2 sin + 2 sin − ...
π 2 3 2 5 2

4 πt 1 3πt 1 5πt
e (t) = cos − cos + cos − ... (1.82)
π 2 3 2 5 2
The well known convenience of Fourier series for handling partial and or-
dinary differential equations is due to the fact that exp(iωt) is an eigen
function of the time derivative, in other words, d exp(iωt)/dt = iω exp(iωt).
This important advantage is unfortunately missing when using the set
{τ (ωk t), e(ωk t)} as a nonsmooth basis for Fourier expansions.
The alternative way is based on power series expansions. Indeed, in terms
of the oscillating time τ , polynomial approximations can be applied with no
periodicity loss. For example,
πt πτ πτ 1 πτ 3 1 πτ 5
sin = sin = − + − ...
2 2 2 3! 2 5! 2
πt πτ
1 πτ 2
1 πτ 4
cos = cos e= 1− + − ... e (1.83)
2 2 2! 2 4! 2
While being polynomials, truncated series of (1.83) preserve periodicity at

cost of smoothness loss though; see Fig. 1.8 for explanation. Fortunately,
nonsmoothness times Λ = {t : τ (t) = ±1} are same for every term of the
series and this enables one of smoothing the series by re-ordering their terms
as follows

3
πt πτ π τ3 π π3 τ τ5
sin = sin = τ− + + −
2 2 2 3 2 16 3 5

3 5

5 7

π π π τ τ
+ + + − + ... (1.84)
2 16 768 5 7

πt πτ π2 2
cos = cos e = [1 − τ + 1 −
2
τ − τ4
2 2 8

π2 π4 4
+ 1− + τ − τ 6 + ...]e
8 384
Fig. 1.9 illustrates both convergence and smoothness of the transformed series
as compared with the diagrams in Fig.1.8. Moreover, it will be shown that
the power series of τ can be re-ordered in such a manner that their particular
sums become as smooth as necessary. Still the convenience of such kind of
series is that they can accurately approximate non-smooth or close to them
processes just by first few terms.
Finally, there are many physical processes easily described by very
simple combinations of the triangular sine and rectangular cosine waves,
which otherwise would require quite long Fourier sums; see Fig. 1.10, for
examples.
30 1 Introduction
Fig. 1.8 Truncated Maclaurin NSTT expansions with respect to τ for sin(πt/2)
and cos(πt/2) including one, two, and three first terms of the expansions as indi-
cated on the curves.
1.4.9 Differential Equations of Motion and

Distributions
Any nonsmooth substitution in the differential equations of motion must be
carefully examined since differentiation is involved. In many physically mean-
ingful cases though sufficient justifications can be achieved by understanding
equalities as integral identities. For illustration purposes, let us consider con-
servative oscillator
ẍ + Px (x) = 0 (1.85)
where P (x) is the potential energy.
Fig. 1.9 First term of the modified NSTT series with respect to τ for sin(πt/2)
and cos(πt/2); 0 - original functions, 1 - first term of the series.
Classical pointwise interpretations of differential equations require no dis-

continuities to occur on the left-hand side of (1.85). Therefore, ‘real motions’
must be described by at least twice continuously differentiable functions of
time, x (t) ∈ C 2 (R). Practically, however, physical systems cannot be ob-
served at every time instance. In other words, point-wise definition of equal-
ity (1.85) appears to be somewhat restrictive from the physical standpoint.
As an extension of classical approaches, the left-hand side of (1.85) can be
considered as some force producing zero work on arbitrary path variations
δx during the observation interval, t1 < t < t2 , so that
32 1 Introduction
t2
(ẍ + Px (x)) δxdt = 0 (1.86)
t1
or
t2

1 2
−δ ẋ − P (x) dt = −δS = 0 (1.87)
2
t1
where S is the action by Hamilton.
Fig. 1.10 Sample temporal shapes of periodic processes easily described by com-
binations of the triangular sine and rectangular cosine
1.5 Non-smooth Coordinate Transformations 33
Expression (1.87) represents the Hamiltonian principle, where the inte-

gral can be calculated for less smooth functions x (t) ∈ C (R). Therefore, it
is possible to essentially extend the class of solutions by ‘keeping in mind’
interpretation (1.86) when considering equation (1.85).
Note that equation (1.85) may be strongly nonlinear, however the highest
derivative must participate in a linear way as a summand, otherwise transi-
tion from (1.86) to (1.87) may become impossible.
1.5 Non-smooth Coordinate Transformations
1.5.1 Caratheodory Substitution

In general, including Dirac’s δ-functions in nonlinear differential equations
complicates mathematical justifications of the modeling. It is important how-
ever how δ-functions participate in equations [46]. For instance, let the dif-
ferential equation include δ-impulse as a summand
ẋ = kx3 + qδ (t − t1 ) (1.88)
where k, q and t1 are constant parameters.

In this case, the δ-input generates step-wise discontinuity of the response
x(t) at t = t1 , however the nonlinear operation in (1.88) still can be con-
ducted. Moreover, the δ-pulse is easily excluded from equation (1.88) by sub-
stitution
x (t) = y (t) + qH (t − t1 ) (1.89)
where y (t) is a new unknown function and H (t − t1 ) is the unit-step Heavi-
side function.
Indeed, substituting (1.89) in (1.88) and taking into account that Ḣ (t − t1 )
= δ (t − t1 ), H 2 (t − t1 ) = H (t − t1 ), and H 3 (t − t1 ) = H (t − t1 ), gives
ẏ = k[y 3 + (q 3 + 3yq 2 + 3y 2 q)H (t − t1 )] (1.90)
where y (t) is now continuous on the interval of consideration including the

point t1 .
1.5.2 Transformation of Positional Variables

Now let us discuss the method of nonsmooth coordinate transformation
(NSCT) for mechanical systems with perfectly stiff constraints [200]. Ac-
cording to this approach, the new coordinates are introduced in order to
automatically satisfy the constraint conditions. Let us reproduce the idea
based on one-degree-of-freedom Langrangian system
1 2
L= ẋ − P (x) (1.91)
2
34 1 Introduction
whose motion is limited by interval
−1≤x≤1 (1.92)
It is assumed that the particle collides with the obstacles x = ±1 with no

energy loss.
Note that relationships (1.91) and (1.92) give no unique differential equa-
tion of motion on the entire time domain since every collision indicates transi-
tion from one system to another. Indeed, even though the form of Lagrangian
(1.91) remains the same, one must deal with different solutions before and
after the collision, say x− (t) and x+ (t). Matching such solutions is usually
not straightforward since the collision time is a priory unknown.
The major reason for applying NSCT is that it gives a single differential
equation of motion for the entire time interval with no constraint conditions.
As a result, the mentioned above matching procedure simply becomes un-
necessary. In addition, after the transformation, a new system appears to be
well suited for averaging since no impact forces are involved any more.
In the above case (1.91) and (1.92), the NSCT is introduced as follows9
x = τ (s) (1.93)
where s(t) is a new positional coordinate.

Substituting (1.93) into (1.91), gives
1 1
L= [τ (s)ṡ]2 − P (τ (s)) = ṡ2 − P (τ (s)) (1.94)
2 2
where the prime indicates differentiation with respect to s, and the relation-
ship [τ (s)]2 = e2 (s) = 1 has been taken into account.
Now, condition (1.92) is satisfied automatically since −1 ≤ τ (s) ≤ 1,
whereas the Hamiltonian principle gives the differential equation of motion
with no constrains
dP
s̈ + e(s) = 0 (1.95)
dτ
On geometrical point of view, transformation (1.93) unfolds the configura-
tion space by switching the coordinate direction on opposite whenever the
particle collides with an obstacle. As a result, the new configuration space
acquires a cell-wise non-local structure as illustrated in [191] and [148]. This
usually makes the differential equation essentially nonlinear even when the
system in between the constraints is linear. Let, for instance, the potential
energy function represents the harmonic oscillator, P (x) = ω 2 x2 /2. Then the
coordinate transformation (1.93) brings system
ẍ + ω 2 x = 0, −1≤x≤1 (1.96)
9
Note that both notations and normalization for the period for the triangular
wave function differ from the original work [200].
1.5 Non-smooth Coordinate Transformations 35
to the form
s̈ + ω 2 τ (s)e(s) = 0, −∞<s<∞ (1.97)
Note that, as a side effect of the elimination of constraints, the differential
equation lost its linearity. In contrast to (1.97) however, the linear differential
equation in (1.96) does not describe the entire system, which is strongly non
linear from the very beginning due to impact interactions with constraints.
Finally, results of comparison between NSCT and NSTT can be summa-
rized as follows [143]:
Coordinate transformation Time transformation
Unfolds the space with no effect on Folds the time, and generates the
the time variable hyperbolic algebraic structures in
space
Targets rigid barriers (constraints) Barriers do no matter
Dynamic regime independent Assumes certain temporal symme-
tries of motion
Applies to a function (image) Transforms an argument (pre-
image)
Essentially changes the ODE struc- Preserves most structural properties
ture, for instance, linear on strongly of ODEs
nonlinear
As follows from the table, NSCT is quite opposite to NSTT from the very
different viewpoints.
1.5.3 Transformation of State Variables

Under some conditions, the NSCT can still be adapted to the case of non-
elastic interactions with constraints in a purely geometrical way [204]. How-
ever, generalized approaches to such cases should involve both coordinates
and velocities [73], [72]. For illustrating purposes, let us consider the case of
harmonic oscillator under the constraint condition
ẋ = Ax, x1 > 0 (1.98)
where x = [x1 (t), x2 (t)]T is the state vector such that x2 = ẋ1 , and

0 1
A= (1.99)
−ω 2 0
It is assumed that every collision with the constraint x1 = 0 results in a

momentary energy loss characterized by the restitution coefficient κ.
The idea is to unfold the phase space in such way that the energy loss
would automatically occur whenever the system crosses a preimage of the
line x1 = 0. The corresponding transformation is
36 1 Introduction
x = Sy (1.100)
where y = [s(t), v(t)]T is a new state vector, and

10
S= sgn(s) (1.101)
0 1 − ksgn(sv)
where k = (1 − κ)/(1 + κ).

Transformation (1.100) is of-course strongly nonlinear due to non-smooth
dependence (1.101).
Substituting (1.100) in (1.98), gives equation
−1
ẏ = (S AS)y (1.102)
In the component-wise form, expressions (1.100) and (1.102) are written as,
respectively,
x1 = x1 (s, v) ≡ ssgn(s)
x2 = x2 (s, v) ≡ sgn(s)[1 − ksgn(sv)]v (1.103)
and
ṡ = [1 − ksgn(sv)]v
v̇ = −ω 2 s[1 + ksgn(sv)]/(1 − k 2 ) (1.104)
Now both unknown components of the state vector are continuous, whereas
specifics of non-elastic collisions are captured by transformation (1.103).
Finally, consider the general case of one-degree-of-freedom nonlinear
oscillator
ẋ1 = x2
ẋ2 = −f (x1 , x2 , t) (1.105)
with non-elastic constraint x1 = 0 of the restitution coefficient κ.

Applying transformation (1.103) to system (1.105), gives
ṡ = [1 − ksgn(sv)]v
v̇ = −f (x1 (s, v), x2 (s, v), t)sgn(s)[1 + ksgn(sv)]/(1 − k 2 ) (1.106)
Although the above illustrations is one-dimensional, similar coordinate trans-

formations can be introduced also for multiple degree-of-freedom systems
in which one of the coordinates is normal to the constraint. The corre-
sponding analytical manipulations can be conducted in terms of Routh
descriptive functions such that the normal to the constraint coordinate is La-
grangian whereas other generalized coordinates and associated momenta are
Hamiltonian.
Chapter 2
Smooth Oscillating Processes
Abstract. This chapter gives a brief overview of selected analytical methods

for smooth oscillating processes. Most of such methods are indeed quasi-
linear. In other words, the corresponding technical implementations employ
harmonic oscillators as generating models. The description focuses only on
the ideas and technical details that are further combined with non-smooth
methods. As most effective way, procedures of asymptotic integration of the
differential equations of motion bring original systems to such simple form
that further solution becomes straightforward. In particular, the method of
asymptotic integration of the differential equations of motion based on the
Hausdorff equation for operators Lie is reproduced.
2.1 Linear and Weakly Non-linear Approaches
By both practical and theoretical reasons, the quantitative methods of dy-

namics were developed first for smooth processes. As a rule, smooth oscil-
lations can be directly observed under no special conditions. For instance,
projection of any fixed point of a body rotating with constant angular speed,
makes a perfect impression about harmonic oscillations. Interestingly, in 1693,
Leibniz derived the differential equation for sine geometrically by considering
a circle. Much later, original analytical ideas of nonlinear vibrations emerged
from the celestial mechanics considering perturbations of circular orbits of
rigid-body motions rather than any mass-spring oscillators. Robert Hooke
(1635-1703) was probably first who suggested the basic elastic mass-spring
model, whereas Galileo and Huygens were investigating the pendulum. Later,
d’Alambert, Daniel Bernoulli and Euler considered a one-dimensional contin-
ual model of a string. It was found that the vibrating string represents the
infinity of harmonic oscillators corresponding to different mode shapes of the
string. It is well known that a serious discussion arised about whether or not

springerlink.com
c Springer-Verlag Berlin Heidelberg 2010
38 2 Smooth Oscillating Processes
the sum of smooth functions, such as sines, can represent a non-smooth shape
of the string. These discussions were finalized by the Fourier theorem.
Let us reproduce the result for a periodic function of time f (t) of the
period T in the complex form
∞

f (t) = ck exp(iωk t) (2.1)
k=−∞
T /2
1 2π
ck = f (t) exp(−iωk t)dt, ωk = k
T −T /2 T
This relation generates a one-to-one mapping between function and its

Fourier coefficients
f (t) ←→ {...c−2 , c−1 , c1 , c2 , ...} (2.2)
Note that mathematical expressions (2.1) do not necessarily imply that

the periodic process f (t) must be produced by linear systems even though
the right-hand side of (2.1) combines free vibrations of linear oscillators,
in other words - ‘rigid-body rotations’ as discussed in Chapter 1. There-
fore, the Fourier analysis and associated analytical tools provides a ‘linear
language’ for nonlinear systems regardless specifics of algorithm implemen-
tations. Indeed, most quantitative methods for weakly nonlinear periodic mo-
tions, actually estimate Fourier coefficients of the corresponding solutions. As
a result, on one hand, such tools possess a high level of generality. On the
other hand, even ‘elementary’ strongly nonlinear phenomena (as qualified in
Chapter 1) may become quite difficult to describe in terms of the ‘linear lan-
guage.’ Nevertheless, the quantitative theory of nonlinear vibration has been
advanced by new asymptotic techniques developed originally for solving non-
linear differential equations. Most traditional methods are essentially based
on perturbation or averaging methods [50]. Similar results can be obtained
within the theory of Poincare’ normal forms [118], which retains resonance
terms, whereas all non-resonance terms are eliminated by means of a coordi-
nate transformation. Such a normal form is qualified as the simplest possible
form of the equations of motion.
2.2 A Brief Overview of Smooth Methods
2.2.1 Periodic Motions of Quasi Linear Systems

Consider a weakly unharmonic oscillator of the form
ẍ + ω02 x = εf (x, ẋ) (2.3)

2.2 A Brief Overview of Smooth Methods 39
where ε is a small parameter, 0 < ε 1, and f (x, ẋ) is smooth enough

function.
Periodic solutions of equation (2.3) can be found by splitting the nonlinear
system into the sequence of linear oscillators by means of the power series
x = x0 + εx1 + ε2 x2 + ... (2.4)
The perturbation on the right-hand side of equation (2.3) changes the prin-
cipal frequency of the oscillator so that
ω 2 = ω02 (1 + εγ1 + ε2 γ2 + ...) (2.5)
The new frequency is introduced explicitly into the differential equation of

motion by re-scaling the temporal argument
ϕ = ωt (2.6)
As a result, series (2.4) appears to be composed of trigonometric functions

of multiple phases ϕ, 2ϕ, 3ϕ,... .
A similar idea was implemented by Lyapunov for systems of first-order
equations, for instance
ẋ1 = a11 x1 + a12 x2 + f1 (x1 , x2 ) (2.7)

ẋ2 = a21 x1 + a22 x2 + f2 (x1 , x2 )
where f1 and f2 are nonlinear functions.

It is assumed that system (2.7) admits first analytical integral and the
corresponding linearized system has only periodic solutions. Then periodic
solutions of (2.7) admit power series expansions with respect to the amplitude
parameter.
There exist at least two extensions of Lyapunov’s theory, such as local and
global approaches to nonlinear normal modes, see for instance [100], [190],
[119].
2.2.2 The Idea of Averaging

Let us illustrate different implementations of the idea of averaging by repro-
ducing some technical details. The following description focuses on such tools
that remains applicable to non-hamiltonian systems.
Following Van-der-Pol’s approach, let us transform system (2.3) by chang-
ing the variables {x, ẋ} → {a, ϕ}:
x = a cos ϕ, ẋ = −aω0 sin ϕ (2.8)

As a result, one obtains

ε
ȧ = − f (a cos ϕ, −aω0 sin ϕ) sin ϕ
ω0
ε
ϕ̇ = ω0 − f (a cos ϕ, −aω0 sin ϕ) cos ϕ (2.9)
ω0 a
Despite of a formal complexity, system (2.9) has essential advantage due to
different time scales of the new variables. This enables one of eliminating the
fast phase ϕ on the right-hand side of the system by applying the averaging
2π
1
< · · · >ϕ ≡ · · · dϕ
2π 0
as follows
ε
ȧ = − < f (a cos ϕ, −aω0 sin ϕ) sin ϕ >ϕ
ω0
ε
ϕ̇ = ω0 − < f (a cos ϕ, −aω0 sin ϕ) cos ϕ >ϕ (2.10)
ω0 a
Solutions of system (2.10) are considered then as approximate solutions of
the original system (2.9).
This method was essentially generalized in thirties [28] by incorporating
the Lindstedt-Poincare and Van-der-Pol’s ideas as follows.
Let us consider the general system with one fast phase
ẋ = εX(x, y)
ẏ = ω(x) + εY (x, y) (2.11)
where y and x are scalar and vector variables respectively.

In contrast to (2.9), the frequency in (2.11) depends on the slow vector-
function x. Sometimes, such kind of systems is called essentially nonlinear
since the condition ε = 0 does not make the frequency state independent.
However, if ε = 0 then system (2.11) has no fast phase on the right-hand
side. The problem is to find close to identical transformation
x = q + εu(q, ψ) + O(ε2 )
y = ψ + εv(q, ψ) + O(ε2 ) (2.12)
which eliminates the fast phase entirely from the system by bringing equation
(2.11) to the form
q̇ = εA(q) + O(ε2 )
ψ̇ = ω0 (q) + εω1 (q) + O(ε2 ) (2.13)
This problem is solved by substituting expansions (2.12) into equations (2.11)

and enforcing then equations (2.13).
2.2.3 Averaging Algorithm for Essentially Nonlinear

Systems
In order to illustrate the corresponding procedure, let us specify system (2.11)
as follows
l˙ = μR(μl, s, ϕ, θ)
ṡ = μ2 S(μl, s, ϕ, θ)
θ̇ = μQ(l, s) + μ2 Θ(μl, s, ϕ, θ) (2.14)
ϕ̇ = Ω(μl, s) + μ2 G(μl, s, ϕ, θ)
where all the coordinates and functions are scalars, and μ 1.

Such kind of equations may occur when considering ‘essentially nonlinear’
systems under different resonance conditions. This is the reason for using
√ In resonance cases, original small pa-
another notation for small parameter.
rameters are often modified as μ = ε to capture specifics of the dynamics
near resonance surfaces [9], [205].
The basic approximation is obtained from system (2.14) by applying the
averaging procedure directly to the terms of order μ on the right hand side.
This gives,
l˙ = μ < R(0, s, ϕ, θ) >ϕ

θ̇ = μQ(l, s) (2.15)
where s should be considered as a constant.

System (2.15) is easily integrated and the result is known to give an er-
ror of order μ on time intervals of order 1/μ. In many cases however, first
approximation gives incomplete characterizations of systems.
In order to illustrate the basic stages of second approximation, consider
the first equation only. It is sufficient for illustration of the procedure, which
is sequentially applied in the same way to other equations.
Let us represent the first equation of system (2.14) in the form
l˙ = μ < R(0, s, ϕ, θ) >ϕ +μ[R(0, s, ϕ, θ)− < R(0, s, ϕ, θ) >ϕ ] (2.16)

+μ2 lRμl (0, s, ϕ, θ) + O(μ3 )
Following the idea of averaging, one eliminates the second term on the right-
hand side by means of the coordinate transformation
l = q + μf (q, s, ϕ, θ) (2.17)
Then, substituting (2.17) into (2.16), gives

∂f ∂f ∂f ∂f
q̇ + μ q̇ + ṡ + ϕ̇ + θ̇
∂q ∂s ∂ϕ ∂θ
= μ < R(0, s, ϕ, θ) > ϕ + μ[R(0, s, ϕ, θ)− < R(0, s, ϕ, θ) > ϕ] (2.18)

+μ2 qRμl (0, s, ϕ, θ) + O(μ3 )
Now the fast phase ϕ is eliminated from the equation in first order of μ by
taking into account (2.14) and imposing condition
∂f
Ω(0, s) = R(0, s, ϕ, θ)− < R(0, s, ϕ, θ) >ϕ (2.19)
∂ϕ
Further, f (q, s, ϕ, θ) is independent of q because the terms of order μ on the

right-hand side of equation (2.18) are independent of q. As a result, equation
(2.18) takes the form

q̇ = μ < R(0, s, ϕ, θ) > ϕ + μ2 qRμl (0, s, ϕ, θ) (2.20)

∂f ∂f
−μ 2
qΩ (0, s) + Q(q, s) + O(μ3 )
∂ϕ μl ∂θ
Since the fast phase ϕ is eliminated from the terms of order μ then the
averaging procedure is applied to the terms of order μ2 analogously to the first
stage of the method. As follows from (2.19), < ∂f /∂ϕ >ϕ =< ∂f /∂θ >ϕ = 0,
therefore after the averaging, equation (2.20) takes the form

q̇ = μ < R(0, s, ϕ, θ) >ϕ +μ2 < qRμl (0, s, ϕ, θ) >ϕ +O(μ3 )
or
q̇ = μ < R(μq, s, ϕ, θ) >ϕ +O(μ3 ) (2.21)
The second approximation therefore is obtained by applying the operator of
averaging to original equation (2.14). Note however that the meaning of the
coordinate q is now different. Namely, the original coordinate l is expressed
through the new coordinate q by relationship (2.17), which due to (2.19)
takes the form
ϕ
μ
l=q+ (R(μq, s, ϕ, θ)− < R(μq, s, ϕ, θ) >ϕ )dϕ + O(μ2 ) (2.22)
Ω(μq, s)
0
In this expression, the variable μq was put back into the expression R in-
stead of zero. Although such a manipulation has no effect on the order of
approximation, the new form is in a better match with the form of equation
(2.21).
Expressions (2.21) and (2.22) summarize the averaging procedure in second
order of μ.
Note that the case of multiple fast phases turns out to be more complicated
in many respects due to the well known problem of small denominators.
2.2.4 Averaging in Complex Variables

In the physical literature, vibration problems are usually considered in terms
of complex variables [89]. The idea of using the complex variables may be
suggested by the standard manipulations of the variation of constants for
oscillator (2.3) as follows.
If ε = 0 then general solution of equation (2.3) is represented in the com-
plex form
1
x = [A exp(iω0 t) + Ā exp(−iω0 t)] (2.23)
2
where A and Ā are arbitrary complex conjugate constants.
The velocity is
iω0
ẋ = [A exp(iω0 t) − Ā exp(−iω0 t)] (2.24)
2
If ε = 0 then the constants are assumed to be time dependent whereas ex-
pressions (2.23) and (2.24) are considered as a change of the state variables
{x, ẋ} → {A, Ā} (2.25)
under the compatibility condition
dA dĀ
exp(iω0 t) + exp(−iω0 t) = 0 (2.26)
dt dt
By solving equations (2.23) and (2.24) with respect to A one obtains
1
A= exp(−iω0 t)(ẋ + iω0 x) (2.27)
iω0
Similar kind of complex amplitudes is used in both physics [89] and nonlinear
mechanics [102].
Now equation (2.3) gives
dA ε
= exp(−iω0 t)f (2.28)
dt iω0
where f = f (x, ẋ) is expressed trough (2.23) and (2.24).
Equation (2.28) is still exactly equivalent to (2.3). If the parameter ε is
small then the amplitude A is slow, and one can apply the averaging
0
2π/ω
dA ε
= exp(−iω0 t)f dt (2.29)
dt 2πi
0
On theoretical point of view, complex amplitudes may bring some conve-

nience compared to the traditional Van-der-Pol variables. Firstly, until the
certain stage of manipulations, it is usually possible to keep only one equation
since another one is its complex conjugate. Secondly, such a symmetry of the
equations helps sometimes to reveal interesting features of the dynamics.
Note that the above manipulations remain valid in degenerated cases of
multiple degrees of freedom systems. For instance, equation (2.3) can be
interpreted as a vector equation with the scalar factor ω02 .
2.2.5 Lie Group Approaches

The one-parameter Lie1 group approaches are motivated by the idea of
matching the tool and the object of study as explained in works [201]
and [204]. Briefly, it is suggested to seek transformation (2.12) among so-
lutions of dynamical systems rather than the class of the arbitrary nonlinear
transformations.
Original materials and overviews of the mathematical structure of Lie
groups, Lie algebras and Lie transforms with applications to nonlinear dif-
ferential equations can be found in [38], [61], [22], [95], [35].
An essential ingredient of this version is the Hausdorff formula, which
relates the Lie group operators of the original and new systems, and the
operator of coordinate transformation. According to [202] and [204], most of
the averaging techniques just reproduce this formula, each time implicitly,
during the transformation process. But, there is no need of doing this, since
it is reasonable to start the transformation using Hausdorff’s relationship.
The corresponding algorithms therefore enable one of optimizing the number
of manipulations for high-order approximations of asymptotic integration.
The theory of Lie groups deals with a set of transformations. In other
words, some dynamical system ẏ = f (y, ε) is transformed into its simplest
form ż = g (z, ε) by means of a coordinate transformation y → z produced
by solution z = z(y, ε) of the third dynamical system
dz
= T (z, ε) , z |ε=0 = y
dε
where the choice for vector-function T (z, ε) depends upon desired properties
of the transformed system.
1
Marius Sophus Lie ( 1842-1899 ), Norwegian mathematician; different mathe-
matical objects are named after him, for instance, groups, operators, algebras,
and series.
As mentioned, one of the advantages of the group formulation is that it

specifies a general class of near identical transformations. Specifically, one
should select the expression z = z(y, ε) among solutions of a dynamical sys-
tem, but not among all classes of the near identical transformations. Another
basic advantage is that all manipulations of the scheme can be done in linear
terms of the monomial Lie group operators. Moreover, the result of trans-
formation in general terms of operators is well-known and is given by the
Hausdorff formula.
The description below presents all the stages starting with the traditional
Newtonian form of the differential equations of motion as implemented in
[147]. The original system will be reduced to its normal form by Poincare.
In terms of the principal coordinates qk , a nonlinear dynamical system
of n-degrees of freedom may be described by a set of n + 1 autonomous
differential equations written in the standard form
q̈k + ωk2 qk = εFk (q1 , ..., qn+1 , q̇1 , ..., q̇n+1 ); k = 1, ..., n + 1 (2.30)
where an overdot denotes differentiation with respect to time t, ε is a small

parameter, and an external excitation has been replaced by the coordinate
qn+1 . The functions Fk include all nonlinear terms and possibly parametric
excitation terms, and ωk are the principal mode frequencies. It is assumed
that the functions Fk admit Taylor expansions near zero.
The Poincare normal form theory deals with sets of first-order differential
equations written in terms of normal form coordinates. In this case it is
convenient to transform the n + 1 second-order differential equations (2.30)
into n + 1 first-order differential equations plus their conjugate set. This can
be done by introducing the complex coordinates
yk = q̇k + iωk qk (2.31)
1 1
qk = (yk − ȳk ), q̇k = (yk + ȳk ) (2.32)
2iωk 2
Introducing the transformation (2.31) into the equations of motion (2.30),
gives
d
q̈k + ωk2 qk = (q̇k + iωk qk ) − iωk (q̇k + iωk qk ) =
dt
dyk
= − iωk yk = εFk (y1 , ..., yn+1 ; ȳ1 , ..., ȳn+1 )
dt
or
ẏk = iωk yk + εFk (y1 , ..., yn+1 ; ȳ1 , ..., ȳn+1 ) (2.33)
and the corresponding complex conjugate (cc) set of equations, where the
functions Fk (y1 , ..., yn+1 ; ȳ1 , ..., ȳn+1 ) are obtained by substituting (2.32) in
the right hand side of equation (2.30). These terms can be represented in the
polynomial form
m l
Fk = Fkσ y1m1 · · · yn+1 ȳ1 · · · ȳn+1
n+1 l1 n+1
(2.34)
|σ|=2,3,...
where the Taylor coefficients are
1 ∂ |σ| Fk
Fkσ = m
|
ln+1 y=0
σ! ∂y1m1 · · · ∂yn+1
n+1
∂ ȳ1l1 · · · ∂ ȳn+1
and multiple-index notations have been introduced as follows
σ = {m1 , ..., mn+1 , l1 , ..., ln+1 }

|σ| = m1 + · · · + mn+1 + l1 + · · · + ln+1
σ! = m1 ! · · · mn+1 !l1 · · · ln+1 !
Equations (2.33) correspond to the standard form, which is ready for analysis
in terms of Lie group operators.
To apply the theory of the Lie groups we rewrite equations (2.33) in the
form
ẏ = Ay, A = A0 + εA1 (2.35)
where y = (y1 , ..., yn+1 ; ȳ1 , ..., ȳn+1 )T , and

n+1
∂
n+1
∂
A0 = iωk yk + cc and A1 = Fk + cc (2.36)
∂yk ∂yk
k=1 k=1
are operators of linear and nonlinear components of the system, respectively.

In order to bring the equations of motion to their simplest (Poincare) form,
we introduce the coordinate transformation y → z in the Lie series form
ε2 2
y = e−εU z = z−εU z+ U z − ... (2.37)
2!
T
where z = (z1 , ..., zn+1 ; z̄1 , ..., z̄n+1 ) , and the operator of transformation U
is represented in the power series form with respect to the small parameter ε
U = U0 + εU1 + · · · (2.38)
The coefficients of this series are

n+1
∂
Uj = Tj,k + cc (2.39)
∂zk
k=1
where
Tj,k = Tj,k (z1 , ..., zn+1 ; z̄1 , ..., z̄n+1 ) (2.40)
are unknown functions to be determined.
One of the advantages of this process is that the inverse coordinate trans-
formation to the form (2.37) can be easily written as
eεU y = z (2.41)
where one should simply replaces z with y in the operator of transformation,

U.
If ε = 0, transformation (2.37) becomes identical, y = exp (0) z = z. In
this case, equation (2.35) has already the simplest linear form and there is
no need to transform the system. For ε = 0 transformation (2.37) converts
the system (2.35) into the following one:
ż = Bz (2.42)
where the new operator B is given by the Hausdorff formula [22]:
ε2
B = A + ε [A, U ] + [[A, U ] , U ] + ... (2.43)
2!
where [A, U ] = AU − U A is the commutator of operators A and U .
An optimized iterative algorithm for high-order solutions of equation (2.43)
was suggested in [202] and [204]. In order to illustrate just the leading order
terms of asymptotic expansions, let us follow the direct procedure though.
Substituting the power series expansions for A and U given by relations (2.35)
and (2.38) into (2.43) gives
B = A0 + ε (A1 + [A0 , U0 ]) (2.44)

2 1
+ε [A0 , U1 ] + [A1 , U0 ] + [[A0 , U0 ] , U0 ] + ...
2!
A simple calculation gives

n+1
∂
B= {iωk zk + ε [Fk + (A0 − iωk ) T0,k ]} + O ε2 + cc (2.45)
∂zk
k=1
where the terms of order ε2 have been ignored, Fk = Fk |y→z and A0 =

A0 |y→z .
The above relationships show that a transformation of the system
ẏ = Ay → ż = Bz
can be considered in terms of operators A → B. In order to bring the system

into its normal form, one must eliminate as many nonlinear terms as possible
from the transformed system such that the system dynamic characteristics
are preserved. It follows from (2.45) that all nonlinear terms of order ε could
be eliminated under the condition
Fk + (A0 − iωk ) T0,k = 0
Representing the unknown functions in the polynomial form

mn+1 l1 ln+1
T0,k = σ m1
T0,k z1 · · · zn+1 z̄1 · · · z̄n+1 (2.46)
|σ|=2,3,...
and taking into account (2.34), gives

m1 mn+1 l1 ln+1
Fk + (A0 − iωk ) T0,k = Fkσ + iΔσk T0,k
σ
z1 · · · zn+1 z̄1 · · · z̄n+1
|σ|=2,3,...
where
Δσk = (m1 − l1 − δ1k ) ω1 + ... + (mn+1 − ln+1 − δn+1,k ) ωn+1 (2.47)

m l
To reach zero-th coefficient of the monomial z1m1 · · · zn+1 z̄1 · · · z̄n+1
n+1 l1 n+1
, one
must put
σ Fσ
T0,k = i kσ
Δk
under the condition that Δσk = 0.
If Δσk = 0 for some k and σ then the corresponding nonlinear term can-
not be eliminated from the transformed equation since it is qualified as a
resonance term.
Finally, the result of transformation is summarized as follows.
The original set:
mn+1 l1 ln+1
ẏk = iωk yk + ε Fkσ y1m1 · · · yn+1 ȳ1 · · · ȳn+1 (2.48)
|σ|=2,3,...
The transformation of coordinates:

Fkσ m1 mn+1 l1 ln+1
yk = zk − ε i σ z1 · · · zn+1 z̄1 · · · z̄n+1 + O ε2 (2.49)
Δk
|σ|=2,3,...
Δσk =0
The transformed set:

mn+1 l1 ln+1
żk = iωk zk + ε Fkσ z1m1 · · · zn+1 z̄1 · · · z̄n+1 + O ε2 (2.50)
|σ|=2,3,...
Δσk =0
Equations (2.50) represent the normal form of the system, where the sum-
mation is much simpler than that in the original set (2.48). Namely, the
summation in (2.50) contains only those terms that give rise to resonance
while the first term on the right hand side stands for the fast component of the
motion. The fast component of the motion can be extracted by introducing

the complex amplitudes ak (t) as follows
zk = ak (t) exp(iωk t) (2.51)
Substituting (2.51) into (2.50) and taking into account the resonance condi-
tion, Δσk = 0, gives
mn+1 l1 ln+1 2
ȧk = ε Fkσ am 1
1 ...an+1 ā1 ...ān+1 + O ε (2.52)
|σ|=2,3,...
Δσk =0
System (2.52) describes the dynamics in terms of slowly varying amplitudes

and, as a result, reveals global properties of the dynamics in a much easier
way then the original system. After solution of system (2.52) is obtained, the
coordinate transformations (2.51) and (2.49) can interpret the result in terms
of the original coordinates.
Chapter 3
Nonsmooth Processes as Asymptotic
Limits
Abstract. In this chapter, we consider different physical models generating

both smooth and nonsmooth temporal mode shapes as appropriate conditions
occur. The objective is to bring attention to the fact that nonsmooth pro-
cesses may naturally occur as high-energy asymptotics in different oscillatory
models with no intentionally introduced stiff constraints or external impacts.
In other words, nonsmooth temporal mode shapes may be as natural as sine
waves generated by oscillators under low-energy conditions. Essentially non-
linear phenomena, such as nonlinear beats and energy localization are also
considered. In particular, it is shown that energy exchange between two os-
cillators may possess hidden nonsmooth behaviors.
3.1 Lyapunov’ Oscillator

Let us consider a family of oscillators described by the differential equation
ẍ + x2n−1 = 0 (3.1)
where n is a positive integer.

In the particular case n = 1 one has the simplest harmonic oscillator.
However, when n > 1 the system becomes essentially nonlinear and cannot be
linearized within the class of vibrating systems. Moreover, as the parameter
n increases, the temporal mode shape of oscillator (3.1) while remaining
smooth is gradually approaching the triangular wave non-smooth limit. In
general, such transition represents a challenging problem from both physical
and mathematical viewpoints. Therefore, it is important to understand some
basic cases, such as oscillator (3.1) and those considered in the next section.
These special cases admit exact solutions, so that it is possible to actually see
how smooth motions are approaching the non-smooth limit. There are also
methodological reasons for considering equation (3.1) as a simple example
of oscillators including both linear and strongly nonlinear cases. It is known
that, for an arbitrary positive integer n, general solution of equation (3.1)

springerlink.com c Springer-Verlag Berlin Heidelberg 2010
52 3 Nonsmooth Processes as Asymptotic Limits
can be expressed in terms of special Lyapunov’s function [96], [77], [56] such
as snθ and csθ as defined by expressions1
snθ
1−2n
θ= 1 − nz 2 2n
dz, cs2n θ + n sn2 θ = 1
0
These functions possess the properties

dsnθ dcsθ
cs0 = 1, sn0 = 0, = cs2n−1 θ, = −snθ
dθ dθ
The period is given by
1 √ 1
√ dx π Γ 2n
T =4 n √ =2
1 − x2n n Γ n+1
2n
0
Now, the general solution of equation (3.1) can be written as

x = Acs An−1 t + α (3.2)
where A and α are arbitrary constants.

Note that the scaling factors A and An−1 are easily predictable based
on the form of equation (3.1). Indeed, equation (3.1) admits the group of
transformations x = Ax̄(t̄), where t̄ = An−1 t.
For n = 1 the functions snθ and csθ give the standard pair of trigonomet-
ric functions sin θ and cos θ, respectively. Interestingly enough, the strongly
nonlinear limit n → ∞ also gives a quite simple pair of periodic functions.
Despite some mathematical challenges, this case admits interpretation by
means of the total energy
ẋ2 x2n 1
+ = (3.3)
2 2n 2
where the number 1/2 on the right-hand side corresponds to the initial con-
ditions x (0) = 0 and ẋ (0) = 1.
Taking into account that the coordinate of the oscillator reaches its am-
plitude value at zero kinetic energy, gives the estimate −n1/(2n) ≤ x (t) ≤
n1/(2n) for any time t. Since n1/(2n) −→ 1 as n −→ ∞ then the limit motion
is restricted by the interval −1 ≤ x (t) ≤ 1. Inside of this interval, the second
term on the left-hand side of expression (3.3) vanishes and hence, ẋ = ±1 or
x = ±t + α± , where α± are constants. By manipulating with the signs and
constants one can construct the sawtooth sine τ (t) - triangular wave - since
there is no other way to providing the periodicity condition.
1
Another version of special functions for equation (3.1) was considered in [167].
3.1 Lyapunov’ Oscillator 53
So the family of oscillators (3.1) includes the two quite simple cases asso-
ciated with the boundaries of the interval 1 ≤ n < ∞. Respectively, one has
the two couples of periodic functions
{x, ẋ} = {sint, cost}, if n = 1 (3.4)
and
{x, ẋ} → {τ (t), τ̇ (t)}, if n → ∞ (3.5)
where τ̇ (t) is a generalized derivative of the sawtooth sine and will be named
as a rectangular cosine.
Earlier, the power-form characteristics with integer exponents were em-
ployed for phenomenological modeling amplitude limiters of vibrating elastic
structures [191] and illustrations of impact asymptotics [132], [137]. It should
be noted that such phenomenological approaches to impact modeling are de-
signed to capture the integral effect of interaction with physical constraints
bypassing local details of the dynamics near constraints. Such details first
of all depend upon both the vibrating body and amplitude limiter physical
properties. In many cases, Hertz model of interaction may be adequate to
describe the local dynamics near constraint surfaces [64]. Note that direct
replacement of the characteristic x2n−1 by the Hertzian restoring force kx3/2
in (3.1) gives no oscillator. The equation,
ẍ + kx3/2 = 0 (3.6)
which is a particular case of that used in [64], must be obviously accompanied

by the condition 0 ≤ x, where x = 0 corresponds to the state at which the
moving body and constraint barely touch each other with still zero interaction
force.
The following modification brings system (3.6) into the class of oscillators
with odd characteristics
ẍ + ksgn(x)|x|3/2 = 0 (3.7)
However, oscillator (3.7) essentially differs from oscillator (3.1) since equation
(3.7) describes no gap (clearance) between the left and right constraint sur-
faces. In other words models (3.1) and (3.7) still represent physically different
situations. The gap 2Δ with its center at the origin x = 0 can be introduced
in equation (3.7) as follows
ẍ + k[H(x − Δ)|x − Δ|3/2 − H(−x − Δ)|x + Δ|3/2 ] = 0 (3.8)
where H means Heaviside unit-step function.

This is a generalization of model (3.7), which is now obtained from (3.8)
by setting Δ = 0. Equation (3.8) can be viewed as a physical impact oscil-
lator that accounts for elastic properties of its components. As compared to
phenomenological model (3.1), equation (3.8) was obtained on certain phys-

ical basis given by the Hertz contact theory.
Finally, oscillators with power-form characteristics, including their gener-
alizations, can be found in physical literature[26], [121], [58], [60], [98] and
different areas of applied mathematics and mechanics [112], [11], [3], [4], [43],
[107], [5], [34], [6], [63], [193], [48], [108], [164]. In [146], the power-form restor-
ing forces were introduced to simulate the effect of liquid container’ walls on
liquid sloshing impacts; see also review article [66].
3.2 Nonlinear Oscillators Solvable in Elementary

Functions
A class of strongly nonlinear oscillators admitting surprisingly simple exact
general solutions at any level of the total energy is described below. Although
the fact of exact solvability of these oscillators has been known for quite a long
time [78], it did not attract much attention possibly due to the specific form
of the oscillator characteristics with uncertain physical interpretations. It is
clear however that, in a phenomenological way, such characteristics capture
sufficiently general physical situations with hardening and softening behavior
of the elastic forces. For instance, these oscillators were recently used as a
phenomenological basis for describing different practically important physi-
cal and mechanical systems [122], [40], [41]. The hardening characteristic is
close to linear for relatively small amplitudes but becomes infinity growing
as the amplitude approaches certain limits. As a result, the corresponding
temporal mode of vibration changes its shape from smooth quasi harmonic
to nonsmooth triangular wave. In contrast, the softening characteristic be-
haves in a non-monotonic way such that the vibration shape is approaching
the rectangular wave as larger amplitudes are considered.
Earlier, amplitude-phase equations were obtained for a coupled array of
the hardening oscillators [157]. It will be shown below that such oscillators
admit explicit introduction of the action-angle variables within the class of
elementary functions. As a result, conventional averaging procedures become
applicable to a wide range of nonlinear motions including transitions from
high- to low-energy dynamics. In particular, analytical solutions are obtained
under small damping conditions. These solutions show a good match with the
corresponding numerical solutions at any energy level even within the first-
order asymptotic approximation.
Hardening and softening cases of these oscillators are, respectively,
p2 tan2 x tan x
H= + ⇒ ẍ + =0 (3.9)
2 2 cos2 x
and
p2 tanh2 x tanh x
H= + ⇒ ẍ + =0 (3.10)
2 2 cosh2 x
where p = ẋ is the linear momentum of the Hamiltonian H.
3.2 Nonlinear Oscillators Solvable in Elementary Functions 55
Further objectives are to investigate the high-energy asymptotics with

transitions to nonsmooth temporal mode shapes and to show that both of
the above oscillators can play the role of generating systems for regular per-
turbation procedures within the class of elementary functions.
tan x cos2 x
100
50
Π Π Π Π
x
2 4 4 2
50
100
Fig. 3.1 Hardening characteristic.
tanh x cosh2 x
0.5
Π Π Π Π
x
2 4 4 2
0.5
Fig. 3.2 Softening characteristic.
Notice that oscillators (3.9) and (3.10) complement each other as those
with stiff and soft characteristics represented in Figs. 3.1 and 3.2, respectively.
These oscillators can be represented also in the form
ẍ + tan x + tan3 x = 0 (3.11)

ẍ + tanh x − tanh3 x = 0 (3.12)

Further analyses of equations (3.11) and (3.12) can be quite easily conducted
by means of substitutions q = tan x and q = tanh x, respectively. Interestingly
enough, oscillators (3.11) and (3.12) without the qubic terms, namely ẍ +
tan x = 0 and ẍ + tanh x = 0, were considered by Timoshenko and Yang
[182]. But, despite of the simplified form, the corresponding solutions were
found to be special functions.
3.2.1 Hardening Case

Consider first stiff oscillator (3.9), whose solution is

t
x = arcsin sin A sin (3.13)
cos A
where A is an arbitrary constant, and another constant is introduced as an

arbitrary time shift t− > t + const., since the equations admits the group of
temporal shifts.
Therefore, (3.13) represents a general periodic solution of the period T =
2π cos A, and the total energy is expressed through the amplitude A as
1
E= tan2 A (3.14)
2
In zero energy limit, when the amplitude A is close to zero, the oscillator lin-
earizes whereas solution (3.13) gives the corresponding sine-wave temporal
shape. On the other hand, the energy becomes infinitely large as the pa-
rameter A approaches the upper limit π/2. In this case, the period vanishes
while the oscillation takes the triangular wave shape, as it is seen from ex-
pression (3.13). Fig. 3.3 illustrates the evolution of the vibration shape in the
normalized coordinates.
Below, the action-angle variables are introduced in terms of elementary
functions. This enables one of considering non-periodic motions by using
exact solution (3.13) as a starting point of the averaging procedure. For a
single degree-of-freedom conservative oscillator, the action coordinate I is
known to be the area bounded by the system’ trajectory on the phase plane
divided by 2π whereas the angle ϕ coordinate is simply phase angle [8], [124].
In the case of stiff oscillator (3.9), one obtains

1 1
I= pdx = −1 (3.15)
2π cos A
and,
t
ϕ= (3.16)
cos A
respectively.
1.0
Α0.001
0.5
Α0.999
Ns 0.0
0.5
1.0
0 1 2 3 4 5 6

Fig. 3.3 Normalized temporal mode shapes of the stiff oscillator, N s(ϕ, α) =
arcsin(α sin ϕ)/ arcsin α.
The original coordinate and the velocity are expressed by the action-angle
variables as follows [158]
√ √
2I + I 2 (1 + I) 2I + I 2 cos ϕ
x = arcsin sin ϕ , p = (3.17)
1+I 1 + (2I + I 2 ) cos2 ϕ
In order to observe the convenience of action-angle coordinates, let us chose

the Hamiltonian description of the oscillator. Taking into account expressions
(3.14) and (3.15), and eliminating the amplitude A, gives the total energy
and thus the Hamiltonian in the form
1
H = I + I2 (3.18)
2
The corresponding differential equations of motion are derived as follows
∂H
ϕ̇ = =1+I (3.19)
∂I
∂H
I˙ = − =0
∂ϕ
As it is seen, the differential equation of the oscillator (3.9) takes the linear
form with respect to the action-angle coordinates, and thus possess the exact
general solution
I = I0 , ϕ = (1 + I0 ) t + ϕ0 (3.20)
where I0 > 0 and ϕ0 are arbitrary constants. By substituting (3.20) in (3.17),
one can express the solution via the original coordinates. The meaning of the
initial action is clear from the energy relationship
1 1
E = I0 + I02 = tan2 A (3.21)
2 2
Note that the linearity of the Hamiltonian equations is due to the specific
strongly non-linear form of the coordinate transformation (3.17). In other
words, the system nonlinearity has been ‘absorbed’ in a purely geometric way
by the nonlinear coordinate transformation.
As mentioned at the beginning, simplicity of the transformed system and
that of the corresponding solution can be essentially employed for the purpose
of perturbation analysis. Let us consider, however, the differential equation
of motion in the Newtonian form
tan x
ẍ + = εf (x, ẋ) (3.22)
cos2 x
were ε is a small parameter.
This system is weakly non-hamiltonian. However, it is still possible to
consider expressions (3.17) as a change of the coordinates {x, p} −→ {I, ϕ}
by imposing the compatibility condition dx/dt = p, where x and p must be
taken from (3.17). This gives
εf (x, p) sin ϕ
ϕ̇ = 1 + I −
(1 + I) (2I + I 2 ) [1 + (2I + I 2 ) cos2 ϕ]
√
εf (x, p) 2I + I 2 cos ϕ
I˙ = (3.23)
1 + (2I + I 2 ) cos2 ϕ
where the function f (x, p) must be expressed through the action-angle coor-
dinates by means of (3.17).
For instance, in the case of linear damping, f (x, p) ≡ −p, one obtains
ε cos ϕ sin ϕ
ϕ̇ = 1 + I + (3.24)
1 + (2I + I 2 ) cos2 ϕ

ε (1 + I) 2I + I 2 cos2 ϕ
I=−
˙
1 + (2I + I 2 ) cos2 ϕ
At this stage, let us implement just one step of the procedure and evaluate its
effectiveness. Applying the operator of averaging with respect to the phase
variable gives the corresponding first-order averaged system in the linear form
ϕ̇ = 1 + I, I˙ = −εI (3.25)
Substituting the general solution of system (3.25) in (3.17), finally gives

2I0 exp (−εt) + I02 exp (−2εt) 1 − exp (−εt)
x = arcsin sin t + I0 + ϕ0
1 + I0 exp (−εt) ε
(3.26)
where I0 and ϕ0 are arbitrary constants. The corresponding time history

records and phase plane diagrams for different damping coefficients are shown
in Fig. 3.4. Even the first order approximation appears to be perfectly match-
ing with numerical solution for all range of amplitudes. The analytical and
numerical curves can be distinguished only at relatively large magnitudes of
the damping parameter ε. Also, the diagrams show that the temporal mode
shape is gradually changing from the triangular to harmonic as time increases
and thus the amplitude decays.
1.5 1.5
1.0 1.0
0.5 0.5
x 0.0 x 0.0
0.5 0.5
1.0 1.0
1.5 1.5
0 10 20 30 40 50 0 10 20 30 40 50
t t
10 10
5 5
v 0 v 0
5 5
10
1.5 1.0 0.5 0.0 0.5 1.0 1.5 1.5 1.0 0.5 0.0 0.5 1.0 1.5
x x
Fig. 3.4 The dynamics of the linearly damped stiff oscillator under the initial
conditions I0 = 10 and ϕ0 = 0, and two different damping parameters: ε = 0.2 (on
the left,) and ε = 0.8 (on the right.) Analytical and numerical solutions show a
slight mismatch only on the top right diagram.
3.2.2 Localized Damping

Let us consider the case of nonlinear damping
tan x
ẍ + + 2εẋ tan2 x = 0 (3.27)
cos2 x
In this case, the perturbation is given by f (x, p) ≡ −2p tan2 x. Such a damp-
ing is rapidly growing near the boundaries of the interval −π/2 ≤ x ≤ π/2,
but it becomes negligible when the amplitude is small, |x| << 1.
In the action-angle coordinates, first order averaging gives
ϕ̇ = 1 + I, I˙ = −εI 2
and thus
1 I0
ϕ=t+ ln (1 + εI0 t) + ϕ0 , I=
ε 1 + εI0 t
Using the coordinate transformation (3.17), gives solution

I0 (2 + I0 + 2εI0 t) ln(1 + εI0 t)
x = arcsin sin t + + ϕ0 (3.28)
1 + I0 + εI0 t ε
where I0 and ϕ0 are arbitrary constants.

Note that the amplitude decay of solutions (3.26) and (3.28) is qualitatively
different. For instance, the amplitude of vibration (3.28) originally decays in a
fast rate and then becomes very slow. In contrast, the amplitude of vibration
(3.26) first decays slowly then the decay rate abruptly increases and then
slows down again.
3.2.3 Softening Case

Let us consider now softening oscillator (3.10), whose exact solution is

t
x = arc sinh sinh A sin (3.29)
cosh A
As Fig. 3.5 shows, the high-energy vibration shape approaches the rectangular
wave and thus essentially differs of that observed in the stiff case.
Based on solution (3.29), the action-angle coordinates are introduced by
means of expressions
√ √
2I − I 2 (1 − I) 2I − I 2 cos ϕ
x = arc sinh sin ϕ , p = (3.30)
1−I 1 − (2I − I 2 ) cos2 ϕ
where
1
I =1− (3.31)
cosh A
Nevertheless, all the analytical manipulations are analogous to those in the
stiff case. For instance, taking into account (3.31), gives the total energy as
a function of the action coordinate
1 1
E= tanh2 A = I − I 2 (3.32)
2 2
3.3 Nonsmoothness Hiden in Smooth Processes 61
1.0
Α106
0.5
Α0.001
Nh 0.0
0.5
1.0
0 1 2 3 4 5 6

Fig. 3.5 Normalized temporal mode shapes of the soft oscillator,

N h(ϕ, α) =arcsinh(α sin ϕ)/arcsinh α.
In the presence of viscous damping,

tanh x
ẍ + = −εẋ (3.33)
cosh2 x
one obtains, compare to (3.25),
ϕ̇ = 1 − I, I˙ = −εI (3.34)
and general solution of the original equation takes the form

2I0 exp (−εt) − I02 exp (−2εt) 1 − exp (−εt)
x = arc sinh sin t−I0 +ϕ0
1 − I0 exp (−εt) ε
(3.35)
The corresponding time history records and phase plane diagrams are shown
in Fig. 3.6 for different damping coefficients. The first order approximation
appears to perfectly match the corresponding numerical solution for all range
of amplitudes, unless the initial action I0 approaches the magnitude 1. As
follows from expressions (3.32), this magnitude corresponds to the maximum
value of the total energy of the oscillator. Note that the energy of the hard-
ening oscillator has no maximum.
3.3 Nonsmoothness Hiden in Smooth Processes

In this section, we consider nonlinear beats phenomena as another source of
nonsmooth behavior that brings certain physical meaning to oscillator (3.22).
Note that nonlinear beats became of growing interest just few decades ago
0.8 0.6
0.6
0.4 0.4
0.2
x 0.0 x 0.2
0.2 0.0
0.4
0.6 0.2
0 10 20 30 40 50 0 10 20 30 40 50
t t
1.0 1.0
0.5
0.5
v v
0.0
0.0
0.5
0.5
0.60.40.2 0.0 0.2 0.4 0.6 0.8 0.2 0.0 0.2 0.4 0.6
x x
Fig. 3.6 The dynamics of the linearly damped softening oscillator under the initial
conditions I0 = 0.5, ϕ0 = 0, and two different damping parameters: ε = 0.2 (on
the left,) and ε = 0.8 (on the right.) Analytical and numerical curves practically
coincide.
from different viewpoints of physics and nonlinear dynamics [84], [59], [99],
[188], [88]. Interestingly enough, phase variables of interacting oscillators with
close natural frequencies may show non-smoothness of temporal behavior
during the beating [59], for instance similar to that of a vibro-impact process
[101], [104].
3.3.1 Nonlinear Beats Model

Consider two linearly coupled Duffing oscillators
ẍ1 + Ω 2 x1 = −β(x1 − x2 ) − αx31 ≡ f1

ẍ2 + Ω 2 x2 = −β(x2 − x1 ) − αx32 ≡ f2 (3.36)
where α and β are nonlinearity and linear coupling parameters, respectively.

Let us introduce complex coordinates Aj (t) and Āj (t) as follows
1
xj = [Aj exp(iΩt) + Āj exp(−iΩt)]
2
iΩ
ẋj = [Aj exp(iΩt) − Āj exp(−iΩt)] (3.37)
2
where j = 1, 2.
3.3 Nonsmoothness Hiden in Smooth Processes 63
Convenience of using the complex amplitudes for linear and nonlinear

mode analyses have been known for a long time [131], [174]. In physical
literature though, complex amplitudes are introduced more often as vectors
rotating on complex phase planes of oscillators, but the resultant equations
are usually similar to those obtained below. Expressions (3.37) implement
indeed a complex version of the parameter variation method based on the so-
lution of the corresponding linear system. The related compatibility condition
is imposed in the form
dAj dĀj
exp(iΩt) + exp(−iΩt) = 0 (3.38)
dt dt
By substituting (3.37) in (3.36) and taking into account (3.38) gives
dAj 1
= exp(−iΩt)fj (3.39)
dt iΩ
Assuming that the coupling and nonlinearity parameters are sufficiently small
and averaging the right-hand side with respect to Ωt, gives

βi 3α 2
Ȧ1 = A1 − A2 + A1 Ā1 (3.40)
2Ω 4β

βi 3α 2
Ȧ2 = −A1 + A2 + A Ā2
2Ω 4β 2
Further, following work [104], a slowly rotating subcomponent on the complex

plane is eliminated by means of substitution

iβ
Aj = ψj (t) exp t (3.41)
2Ω

βi 3α 2
ψ̇ 1 = − ψ2 − ψ ψ̄ (3.42)
2Ω 4β 1 1

βi 3α 2
ψ̇ 2 = − ψ1 − ψ ψ̄
2Ω 4β 2 2
This system has two integrals as follows
K = ψ 1 ψ̄ 1 + ψ 2 ψ̄ 2 (3.43)
3α 4 4

G = ψ 1 ψ̄ 2 + ψ 2 ψ̄ 1 − |ψ 1 | + |ψ 2 | (3.44)
8β
Integral (3.43) admits substitution

√ 1 π
ψ 1 = K cos θ(t) + exp[i δ1 (t)] (3.45)
2 4

√ 1 π
ψ 2 = K sin θ(t) + exp[i δ 2 (t)]
2 4
Substituting (3.45) in (3.42) and (3.44), gives
3 K2 α
G = −K cos Δ cos θ − (3 − cos 2 θ) = const. (3.46)
32 β
β 3K α
Δ̇ = − cos Δ tan θ + sin θ (3.47)
Ω 8Ω
β
θ̇ = sin Δ (3.48)
Ω
where Δ = δ 2 − δ 1 + π.
It will be shown below that the phase variable θ, which determines the
process of energy flow between the oscillators, is described by the oscillator
(3.22).
Equations (3.47)-(3.48) are similar to those obtained in [101], [104], where
it was noticed that temporal shapes of the phase variables θ and Δ may resem-
ble the behavior of the state variables of impact oscillator. This observation
seems to be important since it is hard to expect any “impact oscillators” in
weakly nonlinear systems of type (3.36).
Below, the corresponding ‘conservative oscillator’ admitting the impact
limit will be explicitly obtained and analyzed by using the methodology de-
scribed in the previous section. However, the approach below deals with a
general class of nonlinear restoring force characteristics admitting power-
series expansions. It will be shown also that equations (3.47)-(3.48) can be
derived by introducing the standard set of amplitude-phase variables and
applying then the traditional one fast phase averaging technique.
3.4 Nonlinear Beat Dynamics: The Standard

Averaging Approach
Let us consider two identical linearly coupled oscillators
ü1 + b(u1 − u2 ) + p(u1 ) = 0

ü2 + b(u2 − u1 ) + p(u2 ) = 0 (3.49)
3.4 Nonlinear Beat Dynamics: The Standard Averaging Approach 65
where b is the coupling stiffness per unit mass, and p(u) is the restoring force
characteristic, which is assumed to be an analytic function that admits a
power series expansion.
Assuming that the system has equilibrium at zero and introducing the
notation
Ω2 = b + k
ε = b/Ω 2 = b/(b + k) (3.50)
f (u) = [(b + k)/b][p(u) − ku]
where f (u) is a nonlinear component of the characteristic, and k=p’(0).

Note that the power series expansion for f (u) starts from at least second
degree of u. Therefore, the order of magnitude of the function f (u) can be
manipulated by making appropriate assumptions as to the magnitude of the
total energy of the system.
Taking into account the above notations and introducing the velocities
v1 (t) and v2 (t) brings the original system to the form
u̇1 = v1
u̇2 = v2
v̇1 = −Ω 2 u1 + ε[Ω 2 u2 − f (u1 )] (3.51)
v̇2 = −Ω u2 + ε[Ω u1 − f (u2 )]
2 2
As ε → 0, the system degenerates into two identical harmonic oscillators

whose total energies are conserved of-course since neither damping nor exter-
nal loading are present. At non-zero ε, the oscillators become non-linear and
interact with each other in such a way that one of the oscillators is loaded by
the force proportional to the displacement of another oscillator. Since system
(3.51) is still perfectly symmetric and conservative, it is reasonable to assume
a relatively slow periodic energy exchange between the oscillators. In order
to describe this process in physically meaningful terms, let us introduce new
set of variables as follows {u1 , v1 , u2 , v2 }− > {K, ϕ, δ1 , δ2 }:
√
u1 = K cos ϕ cos(Ωt + δ1 )
√
v1 = − KΩ cos ϕ sin(Ωt + δ1 )
√
u2 = K sin ϕ cos(Ωt + δ2 ) (3.52)
√
v2 = − KΩ sin ϕ sin(Ωt + δ2 )
In case ε = 0 and constant {K, ϕ, δ1 , δ2 }, expressions (3.52) gives an ex-

act general solution of system (3.51). Therefore, relationships (3.52) simply
implement the idea of parameter variations; the corresponding differential
equations will be given below.
Now, in order to track the oscillator energies during the vibration process,
let us use quantities
1 2 1 2
E1 = (v + Ω 2 u21 ) = Ω K cos2 ϕ
2 1 2
1 1 2
E2 = (v22 + Ω 2 u22 ) = Ω K sin2 ϕ (3.53)
2 2
and
1 2
E0 = E1 + E2 = Ω K (3.54)
2
Besides, expressions (3.53) and (3.54) clarify the physical meaning of the
variables K and ϕ participating in transformation (3.52), where other two
variables, δ1 and δ2 , are phases of the vibrating oscillators. In particular, K
is proportional to the total energy of the decoupled and linearized oscillators,
whereas the phase ϕ characterizes the energy split between the oscillators. In
case ε = 0, the energy parameter K will have small temporal fluctuations due
to coupling and nonlinear terms in (3.51). Nevertheless expressions (3.53) and
(3.54) still will be used as the energy related quantities for characterization
of the energy exchange process between the oscillators.
In order to conduct the transition to the new variables, let us substitute
(3.52) in (3.51) and then solve the set of equations with respect to the deriva-
tives as follows
√
2ε K
K̇ = −εKΩ sin 2ϕ sin(2Ωt + δ1 + δ2 ) +
√ Ω
×{f [ K cos ϕ cos(Ωt + δ1 )] cos ϕ sin(Ωt + δ1 )
√
+f [ K sin ϕ cos(Ωt + δ2 )] sin ϕ sin(Ωt + δ2 )}
ε ε
ϕ̇ = Ω[sin(δ1 − δ2 ) − cos 2ϕ sin(2Ωt + δ1 + δ2 )] − √
2 KΩ
√
×{f [ K cos ϕ cos(Ωt + δ1 )] sin ϕ sin(Ωt + δ1 )
√
−f [ K sin ϕ cos(Ωt + δ2 )] cos ϕ sin(Ωt + δ2 )} (3.55)
δ̇ 1 = −εΩ cos(Ωt + δ1 ) cos(Ωt + δ2 ) tan ϕ
ε √
+√ cos(Ωt + δ1 ) sec ϕf [ K cos ϕ cos(Ωt + δ1 )]
KΩ
δ̇ 2 = −εΩ cos(Ωt + δ2 ) cos(Ωt + δ1 ) cot ϕ
ε √
+√ cos(Ωt + δ2 ) csc ϕf [ K sin ϕ cos(Ωt + δ2 )]
KΩ
System (3.55) is still an exact equivalent to system (3.49) and represents a
standard dynamic system with a single fast phase, ψ = Ωt. As a next natural
stage, let us apply the direct averaging to the right-hand side of (3.55) with
respect to the fast phase ψ:
K̇ = 0
ε
ϕ̇ = Ω sin(δ1 − δ2 )
2
ε ε
δ̇ 1 = − Ω cos(δ1 − δ2 ) tan ϕ + F1 (K, ϕ) (3.56)
2 Ω
ε ε
δ̇ 2 = − Ω cos(δ1 − δ2 ) cot ϕ + F2 (K, ϕ)
2 Ω
where the residue theorem has been applied so that
2π √
1 1
F1 (K, ϕ) = √ f ( K cos ϕ cos ψ) cos ψdψ
K cos ϕ 2π 0
1 √ 1 1 1 1
= √ Res{f [ K cos ϕ (z + )] (z + 2 )}
K cos ϕ 2 z 2 z
2π √
1 1
F2 (K, ϕ) = √ f ( K sin ϕ cos ψ) cos ψdψ
K sin ϕ 2π 0
1 √ 1 1 1 1
= √ Res{f [ K sin ϕ (z + )] (z + 2 )}
K sin ϕ 2 z 2 z
First equation in (3.56) shows that the energy parameter K introduced in

(3.54) remains averagely constant regardless the magnitude of coupling and
nonlinearity parameter ε. This gives a justification for using quantities (3.53)
and (3.54) for describing the energy exchange between the oscillators: in-
deed, neither the coupling nor nonlinear stiffness in (3.51) can accumulate
the energy during one vibration cycle.
Further complete description of the dynamics can be conducted in terms
of the two phase shift parameters, Δ(t) and θ(t), introduced as follows
δ2 = δ1 + Δ − π
1 1
ϕ= θ+ π (3.57)
2 4
Substituting (3.57) in (3.56) and introducing the slow time parameter t1 =
εΩt, gives
dθ
= sin Δ
dt1
dΔ
= − cos Δ tan θ + F (θ) (3.58)
dt1
where
1 1 1 1 1
F (θ) = [F2 (K, θ + π) − F1 (K, θ + π)]
Ω2 2 4 2 4
It can be shown that system (3.58) has the integral
G = − cos Δ cos θ + h(θ) = const. (3.59)

where
h(θ) = − F (θ) cos θdθ (3.60)
Now let us show that system (3.58) can be reduced to a single strongly
nonlinear oscillator with respect to the coordinate θ. Taking time derivative
of both sides of the first equation in (3.58) and eliminating from the result
dΔ/dt1 and cos Δ by means of the second equation in (3.58) and the integral
of motion, (3.59), respectively, gives
d2 θ tan θ
+ = R(θ) (3.61)
dp2 cos2 θ
where p = |G|t1 = ε|G|Ωt is a new slow temporal argument, θ = θ(p) and

tan θ F (θ)
R(θ) = G−2 h(θ)[2G − h(θ)] 2 − [G − h(θ)] (3.62)
cos θ cos θ
Equation (3.61) represents a principal equation describing the energy ex-

change in the coupled set of oscillators (3.49) through the phase shift (3.57).
Note that the right-hand side in (3.61) is due to only nonlinearity associ-
ated with the nonlinear stiffness f (u); see relationships (3.50) and (3.51).
In case R(θ) = 0, equation (3.61) has exact analytical solution

p
θ(p) = arcsin sin θ0 sin (3.63)
cos θ0
where θ0 is the amplitude of θ, whereas another constant can be introduced

as an arbitrary temporal shift, which is admitted by equation (3.61).
Generally speaking, it is still possible to find implicit solutions of equation
(3.61) in terms of quadratures for nonzero R(θ). In most cases, however, the
corresponding expressions appear to be technically complicated for analy-
ses. Therefore secondary asymptotic approaches to oscillator (3.61) may be
reasonable for understanding its behaviors.
For illustrating purposes, let us consider the example when the nonlinear
stiffness in (3.50)-(3.51) consists of cubic and fifth-degree terms
f (x) = α3 x3 + α5 x5 (3.64)
In this case, integration in (3.60) gives
K(6α3 + 5Kα5 )
h= cos2 θ (3.65)
32Ω 2
whereas equation (3.61) takes the form
d2 θ tan θ
2
+ = μ sin 2θ (3.66)
dp cos2 θ
where
K 2 (6α3 + 5Kα5 )2
μ= (3.67)
2048G2 Ω 4
Substituting (3.57) in (3.53), gives the corresponding energy values versus
phase θ:
1
E1 = KΩ 2 (1 − sin θ)
4
1
E2 = KΩ 2 (1 + sin θ) (3.68)
4
3.4.1 Asymptotic of Equipartition

For sufficiently small amplitudes of θ, equation (3.66) can be reduced to the
following Duffing equation
d2 θ 4
+ (1 − 2μ)θ + (1 + μ)θ3 = 0 (3.69)
dp2 3
As follows from (3.68) on physical point of view, the equilibrium point θ = 0

corresponds to equal energy distribution between the oscillators (3.49). So
when the linear stiffness is positive 1 − 2μ > 0, equation (3.69) describes
periodic energy exchange between the oscillators (3.49) provided that the
initial energy distribution is close to equal and the cubic approximation for
the characteristic is justified. The period of the energy exchange process can
be easily estimated based on the corresponding solution of equation (3.69).
However, the linear stiffness becomes negative when
1
μ> (3.70)
2
Condition (3.70) says that the equal energy distribution may become unstable
if the parameter μ is sufficiently large. As a result, system (3.69) can stay in
one of the two new stable equilibrium positions so that a larger portion of the
total energy is localized on one of the two identical oscillators (3.49). Note
that the equilibrium points of system (3.69) correspond to nonlinear normal
mode regimes. The phase-plane diagrams of oscillator (3.66) are shown in
Figs. 3.7, 3.8, and 3.9 for different magnitudes of μ. It is seen how two more
equilibria occur when μ exceeds the critical value (3.70).
Note that condition (3.70) is obviously necessary but not sufficient to guar-
antee that the localization will actually occur. Necessary and sufficient con-
ditions will be discussed below.
In order to determine the initial state of oscillator (3.66) and its parame-
ter μ, let us consider transformation (3.52) at t = 0. With no loss of generality,
1
dΘdp
1
2
3
1.5 1.0 0.5 0.0 0.5 1.0 1.5
Θ
Fig. 3.7 Periodic energy exchange case, μ = 0.2.
1
dΘdp
1
2
3
1.5 1.0 0.5 0.0 0.5 1.0 1.5
Θ
Fig. 3.8 Energy trapping bifurcation, μ = 0.5
one can select δ1 (0) = 0. Then, taking into account expressions (3.57), (3.53)
and (3.54), gives

√ π θ(0)
u1 (0) = K cos +
4 2
v1 (0) = 0

√ π θ(0)
u2 (0) = − K cos Δ(0) sin + (3.71)
4 2

√ π θ(0)
v2 (0) = KΩ sin Δ(0) sin +
4 2
dΘdp 1
1
2
3
1.5 1.0 0.5 0.0 0.5 1.0 1.5
Θ
Fig. 3.9 Super-critical energy trapping diagrams, μ = 0.9
and

E1 (0) − E2 (0)
θ(0) = − arcsin
E1 (0) + E2 (0)

v2 (0)
Δ(0) = − arctan
Ωu2 (0)
2
K= [E1 (0) + E2 (0)] (3.72)
Ω2
K(6α3 + 5Kα5 )
G= − cos Δ(0) cos θ(0) + cos2 θ(0)
32Ω 2
where
1 2
E1 (0) = v1 (0) + Ω 2 u21 (0)
2
1 2
E2 (0) = v2 (0) + Ω 2 u22 (0)
2
Since the initial velocity of the first oscillator is fixed v1 (0) = 0 then the
system initial state is determined by the three quantities K, θ(0), and Δ(0);
see (3.71). Alternatively, one can specify u1 (0), u2 (0) and v2 (0) and then find
K, θ(0), and Δ(0) from (3.72).
3.4.2 Asymptotic of Dominants

As the amplitude of θ is getting closer to π/2 then the phase θ oscillations
acquire nonsmooth temporal shapes. Expression (3.63), for instance, shows
that, at amplitudes near π/2, the energy exchange phase will be close to
the triangular wave shape with a relatively small wave-length. In this case,
Duffing equation (3.69) seems to be not an adequate model. So assuming that

μ is sufficiently small, let us introduce action-angle variables as described
earlier in reference [158]
√
2I + I 2
θ = arcsin sin φ
1+I
√
(1 + I) 2I + I 2 cos φ
θ = (3.73)
1 + (2I + I 2 ) cos2 φ
Substituting (3.73) in (3.66) under the compatibility condition θ = dθ/dp,

gives still exact equivalent of oscillator (3.66)
dI I(2 + I)
=μ sin 2φ
dp (1 + I)2
dφ 2μ
= 1+I − sin2 φ (3.74)
dp (1 + I)3
Note that the coordinate transformation (3.73) still would be valid for general
case (3.61) although with different to (3.74) result. In contrast to (3.66),
system (3.74) is weakly nonlinear with a very simple solution at μ = 0.
In [158], the direct averaging was applied to the right-hand side of (3.74)
in order to obtain the first-order solution. The idea of averaging can be also
implemented as asymptotic integration of system (3.74) by means of the
coordinate transformation
J(2 + J)
I = J −μ cos 2ψ + O(μ2 )
2(1 + J)3
(J 2 + 2J − 2)
φ = ψ−μ sin 2ψ + O(μ2 ) (3.75)
4(1 + J)4
Transformation (3.75) is obtained from the condition eliminating the fast

phase ϕ from the terms of order μ on the right-hand side in such a way that
the new system takes the form
dJ
= O(μ2 )
dp
dψ μ
= 1+J − + O(μ2 ) (3.76)
dp (1 + J)3
System (3.76) is easily integrated as follows

μ
ψ = 1+J − p (3.77)
(1 + J)3
where J = const.
Then the reversed chain of transformations back to (3.73) is applied. So-

lution (3.73) through (3.77) appears to have a good match with the corre-
sponding numerical solution when μ << 1/2; see (3.70) for interpretation.
The solution can work well even under condition (3.70), but the correspond-
ing initial conditions must keep the oscillator out of the triple equilibria
region. The oscillators’s motion within such region is better approximated by
Duffing’s equation (3.69). On physical point of view, the applicability loss for
solution (3.73) through (3.77) is due to the energy localization phenomenon,
which is not captured by the above solution despite of its strong nonlinearity.
Indeed, the term, which is responsible for occurring the triple equilibria is
ignored in the leading-order approximation.
3.4.3 Necessary Condition of Energy Trapping

On physical point of view, necessary condition of localization is the presence
of triple equilibrium positions of oscillator (3.66) within the basic interval
−π/2 < θ < π/2, which is provided by condition (3.70). However, to guar-
antee the energy localization, the initial conditions must keep the oscillator
within one of the two branches of the separatrix loop. Let us bring both of
the above conditions to the explicit form.
Necessary condition. For simplicity reason, let us consider the case of cubic
nonlinearity and introduce two dimensionless parameters
3E0 α3
ν= (3.78)
8Ω 4
and
2
ΔE0
ζ= (3.79)
E0
where E0 = E1 (0) + E2 (0) is the total initial energy defined by (3.54), and
ΔE0 = E1 (0) − E2 (0), so that
√ ν estimates the weight of nonlinearity in the
system dynamics, whereas ζ is the initial energy disbalance per total initial
energy E0 .
Calculating the constants K and G from (3.72) and making algebraic ma-
nipulations, eventually brings the necessary condition of localization (3.70)
to the form
ζν (2 − ζ)ν
−√ < cos Δ(0) < √ (3.80)
1−ζ 1−ζ
Let us recall that condition (3.80) provides the presence of two stable equi-
librium position of oscillator (3.66) near the origin (θ, dθ/dp) = (0, 0) which
itself becomes unstable. However (3.80) does no guarantee that oscillator
(3.66) is in the neighborhood of a stable equilibrium condition.
3.4.4 Sufficient Condition of Energy Trapping

Now let us define that the energy localization takes place whenever the initial
conditions keep oscillator (3.66) within one of the two separatrix loops sur-
rounding the stable equilibrium points. A manifold of such initial conditions
is obtained from first integral of oscillator (3.66) as follows

2
dθ
+ tan2 θ + μ cos 2θ < μ (3.81)
dp
Taking into account equation (3.58), dθ/dp = |G|−1 sin Δ, and calculating
the left - hand side of (3.81) at p = 0, gives

ζ[cos Δ(0) − ν 1 − ζ]2 + sin2 Δ(0) < ζν 2 (3.82)
Note that both conditions (3.80) and (3.82) include the same set of param-
eters, such as the initial phase shift from the out-of-phase mode, Δ(0), the
parameter characterizing the initial energy distinction between the oscilla-
tors, ζ, and the parameter characterizing the total energy and thus strength
of nonlinearity, ν.
3.5 Transition from Normal to Local Modes

The transient mode localization phenomenon is considered in a mechanical
model combined of a simply supported beam and transverse nonlinear springs
with hardening characteristics. Two different approaches to the model reduc-
tion, such as normal and local mode representations for the beam’s center
line, are discussed. It is concluded that the local mode discretization brings
advantages for the transient localization analysis. Based on the specific co-
ordinate transformations and the idea of averaging, explicit equations de-
scribing the energy exchange between the local modes and the corresponding
localization conditions are obtained. It was shown that when the energy is
slowly pumped into the system then, at some point, the energy equipartition
around the system suddenly breaks and one of the local modes becomes the
dominant energy receiver. The phenomenon is interpreted in terms of the
related phase-plane diagram which shows qualitative changes near the image
of out-phase mode as the total energy of the system has reached its critical
level. A simple closed form expression is obtained for the corresponding crit-
ical time estimate. The text below is an update of recent publication by the
author [159].
3.6 System Description

The model under investigation represents a simply supported elastic beam
of length l with two masses attached to the beam and connected to the base
3.6 System Description 75
Fig. 3.10 The mechanical model admitting both normal and local mode motions;
all the springs have hardening restoring force characteristics.
by nonlinear springs; see Fig. 3.10. The corresponding differential equation

of motion and boundary conditions are, respectively,
∂2w ∂4w
ρA 2
+ EI 4 = f1 (t)δ(y − y1 ) + f2 (t)δ(y − y2 ) (3.83)
∂t ∂y
and
∂ 2 w(t, y)
w(t, y)|y=0,l = 0, |y=0,l = 0 (3.84)
∂y 2
where
∂w(t, yi ) ∂ 2 w(t, yi )
fi (t) = −f [w(t, yi )] − c −m ; i = 1, 2 (3.85)
∂t ∂t2
are transverse forces applied to the beam from masses attached at the two
points y = y1,2 .
It will be assumed that the model is perfectly symmetric with respect to
y = l/2 so that the springs are attached at points
y1 = l/3 and y2 = 2l/3 (3.86)
Below we consider the case of the hardening restoring force characteristics

of the springs and show that, under appropriate conditions, a slow energy
in-flow leads to the localization of vibration modes. As a result, the system
energy is spontaneously shifted to either the left or right side of the beam
- the symmetry break. The adiabatically ‘slow’ energy increase means that
the energy source has a minor or no direct effect on the mode shapes. For
simulation purposes, such an energy in-flow is provided by the assumption
that the viscous damping coefficient c is sufficiently small and negative; the
physical basis for such an assumption was discussed in [158], [157]. This
remark, which is substantiated below by the corresponding numerical values
of the parameters, is important to follow, otherwise the phenomenon, which
is the focus of this paper, may not be developed. In contrast, the dissipation
(c > 0) can lead to a spontaneous dynamic transition from local to normal
modes, when the total energy reaches its sub-critical level.
Note that the presence of Dirac δ-functions in equation (3.83) requires
a generalized interpretation of the differential equation of motion in terms
of distributions [166]. The corresponding compliance is provided by further
model reduction based on the Bubnov-Galerkin approach, which actually
switches from the point-wise to integral interpretation of equations.
3.7 Normal and Local Mode Coordinates

Normal mode coordinates. Let us evaluate two possible ways to discretizing
the model (3.83). In this paper, the reduced-order case of two degrees-of-
freedom is considered, when the conventional normal mode representation
for the boundary value problem (3.83) - (3.84) is
πy 2πy
w(t, y) = W1 (t) sin + W2 (t) sin (3.87)
l l
Substituting (3.87) in (3.83) and applying the standard Bubnov-Galerkin
procedure, gives, after dropping the time arguments,
Ẅ1 + ζ̄ Ẇ1 + λ2 W1 + F (W1 + W2 ) + F (W1 − W2 ) = 0 (3.88)

Ẅ2 + ζ̄ Ẇ2 + 16λ2 W2 + F (W1 + W2 ) − F (W1 − W2 ) = 0
where
3c π 4 EI
ζ̄ = , λ2 = 3 (3.89)
3m + Alρ l (3m + Alρ)
and √ √
3 3
F (z) = f z (3.90)
3m + Alρ 2
are constant parameters and a re-scaled restoring force function, respectively.
3.7 Normal and Local Mode Coordinates 77
Equations (3.88) are decoupled in the linear terms related to the elastic
beam centre line, whereas the modal coupling is due to the spring nonlinear-
ities included in F (z).
Local mode coordinates. Alternatively, the model can be discretized by
introducing the ‘local mode’ coordinates determined by the spring locations
ui (t) = w(t, yi ); i = 1, 2 (3.91)
Taking into account (3.86) and (3.87), and substituting in (3.91) reveals sim-
ple links between the normal and local coordinates as
√ √
3 3
u1 = (W1 + W2 ), u2 = (W1 − W2 ) (3.92)
2 2
or, inversely, √ √
3 3
W1 = (u1 + u2 ), W2 = (u1 − u2 ) (3.93)
3 3
Substituting (3.93) in (3.87), gives the ‘local mode expansion’ for the beam’s
centre line πy πy
w(t, y) = u1 (t)ψ1 + u2 (t)ψ2 (3.94)
l l
where the local mode shape functions are
√
ψ1 (x) 3 11 sin x
= (3.95)
ψ2 (x) 3 1 −1 sin 2x
Both normal and local mode shape functions are shown in Figs. 3.11 and 3.12,
respectively. Transformation (3.95) can be generalized for a greater number of
modes. Note that functions (3.95) satisfy the following ortogonality condition
π
π
ψi (x) ψj (x) dx = δij (3.96)
0 3
where δij is the Kronecker symbol.

However, the differential equations of motion for u1 (t) and u2 (t) are ob-
tained directly by substituting (3.93) in (3.88) and making obvious algebraic
manipulations that gives
√ √
ü1 + ζ̄ u̇1 + (λ2 /2)(17u1 − 15u2 ) + 3F (2u1 / 3) = 0 (3.97)
√ √
ü2 + ζ̄ u̇2 − (λ /2)(15u1 − 17u2 ) + 3F (2u2 / 3) = 0
2
This kind of discretization seems to be similar to that given by the finite

element approaches.
1.0
sin x
0.5
sin 2
x
0.0
0.5
1.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0

x
Fig. 3.11 Normal mode shape functions; here and below dashed lines correspond
to second mode.
1.0
1 x 2 x
0.5
0.0
0.5
1.0
0.0 0.5 1.0 1.5 2.0 2.5 3.0

x
Fig. 3.12 Local mode shape functions.
Further, equations (3.97) are represented as a set of first order equations
u̇1 = v1
u̇2 = v2
v̇1 = −ω 2 u1 + ε[ω 2 u2 − ζv1 − p(u1 )] (3.98)
v̇2 = −ω 2 u2 + ε[ω 2 u1 − ζv2 − p(u2 )]
where

2
17 15 λ ζ̄
ω= 2k + λ2 , k = F (0), ε = , ζ=
2 2 ω ε
3.7 Normal and Local Mode Coordinates 79
and √ √ √
3 2 3 2 3
p(ui ) = F ui − kui = βu3i
ε 3 3
are new constant parameters and the nonlinear component of the spring
characteristic; it is assumed that the damping coefficient and the nonlinear
component are small enough to provide the order of magnitude ζ = O(1) and
p(ui ) = O(1).
For calculation purposes, the spring characteristic is taken in the form
F (u) = u + (4/3)u3 which brings the nonlinearity parameter β to the form
64ω 2
β= (3.99)
135λ2
Equations (3.97), and analogously (3.98), possess advantages for transient
analysis because the corresponding linearized system has the same natu-
ral frequencies and the nonlinear components are decoupled. As a result,
the one-frequency perturbation tool becomes applicable. The corresponding
amplitude-phase variables are introduced as follows
ui = αi (t) cos[ωt + δi (t)] (3.100)

vi = −ωαi (t) sin[ωt + δi (t)]
(i = 1, 2)
Substituting (3.100) in (3.98) and applying the averaging procedure with

respect to the fast phase z = ωt, gives
ε
α̇1 = − [ζα1 + ωα2 sin(δ1 − δ2 )]
2
ε
α̇2 = − [ζα2 − ωα1 sin(δ1 − δ2 )]
2
εω α2 3εβ 2
δ̇ 1 = − cos(δ1 − δ2 ) + α (3.101)
2 α1 8ω 1
εω α1 3εβ 2
δ̇ 2 = − cos(δ1 − δ2 ) + α
2 α2 8ω 2
The result of the work of [157] as well as further analysis show that the
localization may occur as the system vibrates in the out-of-phase mode,
u1 (t) ≡ −u2 (t). In order to investigate the dynamics near the out-of-phase
vibration mode, let us introduce three new variables s, ρ and θ,
α1 = −s(t) + ρ(t)
α2 = s(t) + ρ(t) (3.102)
δ2 = δ1 + Δ(t)
where s and ρ characterize the amplitudes of the out-of-phase and in-phase

modes, respectively, and Δ is a phase shift between the local modes so that
the variables ρ and Δ describe small deviations from the out-of-phase mode.
Substituting (3.102) in (3.101), linearizing the result with respect to ρ and
Δ and then eliminating the phase variable Δ, gives

1 2 3 2
ρ̈ + εζ ρ̇ + ε ω + ζ − βs ρ = 0
2 2
(3.103)
4 4
whereas the equation obtained for s gives the solution

1
s = s0 exp − εζt (3.104)
2
In the case |ζ| 1, equation (3.103) describes an oscillator with a slow vary-
ing frequency. Making the frequency ‘frozen’ enables one of the determining
roots of the corresponding ‘characteristic equation’

1 3 2
k1,2 = ε − ζ ± i ω − βs
2 (3.105)
2 4
If the viscosity is negative, ζ < 0, then equations (3.103) through (3.105)

qualitatively describe the transition to the local mode as the system energy
increases. In particular, expression (3.105) shows that when the amplitude of
the out-phase mode, which is associated with s, becomes large enough then
the amplitude of the in-phase mode, ρ, looses its oscillatory character and
grows monotonically.
As a result, one of the local mode increases its amplitude, whereas another
one decays; see expressions (3.102). This is an onset of the dynamic transition
to a localized mode. The corresponding critical time follows from explicit
solution of (3.104) and (3.105)
1 4ω 2
t∗ = ln (3.106)
ε|ζ| 3βs20
In order to provide numerical evidence for the dynamic transition from nor-
mal to local mode vibrations, let us introduce an indicator of the energy
partition calculated as
⎧
E1 − E2 ⎨ −1 if E1 = 0 and E2 = 0
P = = 0 if E1 = E2 (3.107)
E1 + E2 ⎩
1 if E1 = 0 and E2 = 0
where Ei = (vi2 + ω 2 u2i )/2 is the total energy of i-th oscillator under the
condition ε = 0.
3.8 Local Mode Interaction Dynamics 81
Quantity (3.107) is varying within the interval −1 ≤ P ≤ 1. The ends

of the interval obviously correspond to the local modes, whereas its center
P = 0 corresponds to the normal modes.
The time history of the energy partition (3.107) is illustrated by Fig. 3.13.
The following parameters were taken for numerical simulations: λ = 0.05,
ζ̄ = −0.002, k = 1.0, ω = 2k + (17/2)λ2 = 1.4217, β = 32/(9ε) = 383.29,
ε = 15(λ/ω)2 /2, and therefore ζ = ζ̄/ε = −0.2156. The initial normal mode
amplitudes at zero velocities are W1 (0) = 0.0001 and W2 (0) = −0.003. The
critical time estimate based on expression (3.106) t∗ = 3474.29 is in quite a
good match with Fig. 3.13.
0.0
0.2
0.4
P
0.6
0.8
t
1.0
0 1000 2000 3000 4000 5000
t
Fig. 3.13 ‘Sudden’ transition from normal to local mode vibration as the system
energy has reached its critical value.
3.8 Local Mode Interaction Dynamics

Let us introduce new variables, K, θ and Δ, as follows

1 π
α1 = K(t) cos θ(t) +
2 4

1 π
α2 = K(t) sin θ(t) + (3.108)
2 4
Δ = δ2 − δ1 + π (3.109)
Further, considering the local mode total energies Ei under no interaction
condition, and taking into account (3.100), (3.108) and (3.110), gives
1 2
Ei = (v + ω 2 u2i ); i = 1, 2 (3.110)
2 i
1 2 2 1
E = E1 + E2 = ω (α1 + α22 ) = ω 2 K (3.111)
2 2
1 1
ΔE = E1 − E2 = ω 2 (α21 − α22 ) = − ω 2 K sin θ (3.112)
2 2
The variable K therefore is proportional to the total energy of the degen-
erated system, whereas the phase angle θ characterizes the energy partition
between the local modes (3.107) as follows
ΔE
P = = − sin θ (3.113)
E
The third variable (3.109) describes the phase shift in the high-frequency
vibrations between the local modes so that Δ = 0 corresponds to the out-
phase motions of the masses attached to the beam; note the difference with
(3.102).
Differentiating (3.109), (3.111) and (3.112), and enforcing equations (3.101),
gives
dκ ζ
=− κ
dt1 ω
dθ
= sin Δ (3.114)
dt1
dΔ
= − cos Δ tan θ + κ sin θ
dt1
where t1 = εωt is a new temporal argument, and
3β
κ= K (3.115)
8ω 2
In the conservative case, ζ = 0, the first equation in (3.114) gives the energy
integral κ = const, whereas another two equations admit the integral
1
G = − cos Δ cos θ + κ cos 2θ = const (3.116)
4
This particular case matches the results obtained for a linearly coupled set of
Duffing’s oscillators in [101], and later reproduced in [158] however by means
of different complex variable approaches. In particular, it was shown in [158]
that the last two equations in (3.114) are equivalent to a strongly nonlinear
conservative oscillator
d2 θ tan θ
+ G2 =0 (3.117)
dt21 cos2 θ
as κ → 0.
3.8 Local Mode Interaction Dynamics 83
Oscillator (3.117) appears to be exactly solvable with general solution
θ = arcsin[sin θ0 sin(|G|t1 / cos θ0 )] (3.118)
where θ0 is the amplitude and another arbitrary constant can be introduced

as a temporal shift.
As already mentioned in this chapter, oscillator (3.117) was considered
in [78] and [122] as a phenomenological model for quite different kinds of
problems. Since no direct physical meaning of such a unique ‘restoring force
characteristic’ was found, the fact of exact solvability not attracted much
attention for quite a long time.
In the case κ = 0, but still ζ = 0, some perturbation occurs on the right-
hand side of (3.117); the corresponding perturbation tool based on the action-
angle variables was introduced in [158].
Let us show now that equations (3.114) can describe the transition to local
modes under the assumption of small negative viscosity
|ζ/ω| 1 and ζ < 0 (3.119)
Under condition (3.119), the factor κ in the third equation of (3.114) can
be viewed as a slowly growing quasi constant. In this case, making κ ‘frozen’
and linearizing the last two equations in (3.114) near the equilibrium (θ, Δ) =
(0, 0), gives
d2 θ
+ (1 − κ)θ = 0 (3.120)
dt21
Note that small θ and Δ bring the original system close to the out-of-phase
vibration mode as follows from (3.113) and (3.109). When the growing en-
ergy parameter κ passes the critical point κ = 1, the type of equilibrium
is changing from a focus to a saddle point and thus the variable θ becomes
exponentially growing. Practically, however, the exponential growth will be
suppressed by the nonlinearity. As a result two limit phase trajectories (sep-
aratrix loops) occur around two new stable equilibrium points. These two
points represent two new stable modes of the original system - local modes.
The phase plane diagrams for sub- and super-critical energy levels are shown
in Figs. 3.14 and 3.15, respectively. Note that the equilibrium subjected to
such qualitative change corresponds to the out-of-phase vibration mode of
the model, whereas another two equilibrium points, (θ, Δ) = (0, ±π), corre-
spond to the same in-phase mode and remain stable. Therefore, out-of-phase
vibrations appear to be less favorable to energy equipartition as the energy
reaches its critical level. Note that links between localization and specifics
of phase trajectories was discussed also in [101] based on the system of two
coupled Duffing oscillators. In particular, the limit phase trajectories were
interpreted as nonlinear beats of infinitely long period, keeping the energy
near one of the two oscillators.
4
I
0 O
2
4
1.5 1.0 0.5 0.0 0.5 1.0 1.5

Θ
Fig. 3.14 Phase plane structure at undercritical system energy, κ = 0.5.
0 L L
2
4
1.5 1.0 0.5 0.0 0.5 1.0 1.5

Θ
Fig. 3.15 Phase plane structure at postcritical system energy, κ = 1.5.
3.9 Auto-localized Modes in Nonlinear Coupled Oscillators 85
As follows from expression (3.113), the growth of θ increases the energy

unbalance between the local modes, and that is onset of the mode localization.
The corresponding critical time, at which the localization begins, is obtained
from (3.115) as follows
3β ∗ ∗ 1 8ω 2
K 0 exp (ε |ζ| t ) = 1 =⇒ t = ln (3.121)
8ω 2 ε |ζ| 3βK0
As follows from (3.102) and (3.111), K0 = 2(s20 +ρ20 ) therefore, under the
condition ρ20 1, expressions (3.106) and (3.121) give the same result.
Note that the developed analytical approach, describing the local mode
interaction in terms of the energy and phase variables, appears to be inde-
pendent of the individual features of the illustrating model and this can be
used in other similar cases.
Compared to publications [101] and [158] introducing the same set of de-
scriptive variables, K, θ and Δ, current approach has distinctive features as
follows:
1) Instead of general mass-spring models, the elastic beam supported by
nonlinear springs is considered in this work. This provides clear geometrical
interpretations for both the normal and local modes through the correspond-
ing shape functions of the beam centre line.
2) Instead of using a quite complicated system reduction in terms of com-
plex coordinates, it is shown that the same result can be achieved by means of
the traditional set of amplitude-phase variables and one-frequency averaging
procedure.
3) The non-conservative case is considered in order to describe qualitative
changes in the dynamics as the total energy of the system adiabatically in-
creases or decreases. Based on such a generalization, new quantitative and
qualitative results are obtained.
In particular, explicit expressions have been obtained for the critical time
at which onset of the localization occurs. The phenomenon is explained in
terms of the related phase-plane diagram subjected to a qualitative change
(center-saddle transition) as the total energy of the system reaches its critical
level.
3.9 Auto-localized Modes in Nonlinear Coupled

Oscillators
Below, the term ‘auto-localized’ means that the system itself may come into
the nonlinear local mode regime and stay there regardless initial energy dis-
tribution among its particles. As follows from the Poincare’s recurrence the-
orem, such phenomena are rather impossible within the class of conservative
systems [9]. However, interactions between the system particles can be de-
signed in specific ways in order to achieve desired phenomena. It is assumed
that such a design can be implemented practically by using specific elec-

tric circuits and possibly mechanical actuators. On macro-levels, the auto-
localization may help to optimize vibration suppression.
Some results from the previous publication [155] are reproduced below
after some notation modifications in order to make the description coherent
with the current text. Let us consider an array of N harmonic oscillators
such that each of the oscillators interacts with only the nearest neighbors.
The corresponding differential equations of motion are of the form
ẍj + Ω 2 xj = β(xj−1 − 2xj + xj+1 )+
+ α[(Ej − Ej−1 )Ej−1 − (Ej+1 − Ej )Ej+1 ]ẋj (3.122)

1 2
Ej = (ẋ + Ω 2 x2j ); j = 1, ..., N (3.123)
2 j
where Ej = Ej (t) is the total energy of the j-th oscillator under the boundary
conditions of fixed ends E0 (t) ≡ EN +1 (t) ≡ 0, and Ω, β, and α are constant
parameters of the model.
On the right-hand side of equation (3.122), two groups of terms describe
coupling between the oscillators. If α = 0 then the only linear coupling re-
mains. In this case, under special initial conditions, N different coherent pe-
riodic motions i.e. linear normal modes, can take place. It is well known that
any other motion is combined of the linear normal mode motions, whereas the
energy is conserved on each of the modes the way it was initially distributed
between the modes. In other words, no energy localization is possible on
individual particles if α = 0.
Another group of terms, including the common factor α, has the opposite
to linear elastic interaction effect. These nonlinear terms are to simulate pos-
sible ‘competition’ between the oscillators, in other words, one-way energy
flow to the neighbor whose energy is lager. Such kind of interaction domi-
nates when the total system energy is large enough to essentially involve high
degrees of the coordinates and velocities.
For future analysis let us introduce the complex conjugate variables
{Aj (t) , Āj (t)} into equations (3.122) according to relationships (3.37) and
(3.38). In term of the complex amplitudes, the total energy of individual
oscillator (3.123), excluding the energy of coupling, takes the form
1 2 1 2
Ej = Ω Aj Āj = Ω 2 |Aj | (3.124)
2 2
When β = α = 0 the system is decomposed into the N uncoupled oscillators,
and one has a constant solution in the new variables. In general case, substi-
tuting (3.37) in (3.122), taking into account (3.38), and applying averaging
with respect to the phase z = Ωt, gives the following set of equations (the
complex conjugate set is omitted below)
iβ
Ȧj = − (Aj−1 − 2Aj + Aj+1 )+
2Ω
αΩ 4 2 2

2

2 2

2

+ |Aj | − |Aj−1 | |Aj−1 | − |Aj+1 | − |Aj | |Aj+1 | Aj (3.125)
8
Let us consider first, the simplest model of two coupled oscillators (N = 2).
In this case, system (3.125) is reduced to
iβ αΩ 4
Ȧ1 = − (A2 − 2A1 ) + |A1 |2 − |A2 |2 |A2 |2 A1 (3.126)
2Ω 8
iβ αΩ 4
Ȧ2 = − (A1 − 2A2 ) + |A2 |2 − |A1 |2 |A1 |2 A2
2Ω 8
Despite of the presence velocities ẋj in the original equations (3.122), sys-
tem (3.126) still has the integral
2 2
K = |A1 | + |A2 | = 2(E1 + E2 )/Ω 2 = const.
As a result, the dimension of system’ phase space is reduced by introducing

the angular variables ϕ1 (t), ϕ2 (t) and ψ(t),
√ √
A1 = K cos ψ exp(iϕ1 ), A2 = K sin ψ exp(iϕ2 ) (3.127)
where the angle ψ determines the energy distribution between the oscillators
as follows

|A2 | E2
tan ψ = = (3.128)
|A1 | E1
0 ≤ ψ < π/2
Substituting (3.127) into (3.126) and considering separately real and imagi-
nary parts, gives
β β
ϕ̇1 = − tan ψ cos (ϕ2 − ϕ1 )
Ω 2Ω
β β
ϕ̇2 = − cot ψ cos (ϕ2 − ϕ1 ) (3.129)
Ω 2Ω
β 1
ψ̇ = − sin (ϕ2 − ϕ1 ) − αK 2 Ω 4 sin 4ψ
2Ω 32
Introducing the phase shift Δ = ϕ2 −ϕ1 and new temporal variable p = Ωt/β,
gives
dΔ
= − cot 2ψ cos Δ (3.130)
dp
dψ 1
= − (sin Δ + λ sin 4ψ)
dp 2
Π
2
Π
4
0
Π Π
2 0 2

Fig. 3.16 Low energy transition to the nonsmooth limit cycle; numerical solution
obtained for the following system parameter and initial conditions: λ = 0.2; Δ(0) =
0.0, Ψ (0) = π/4 + 0.1.
where λ is a dimensionless parameter linked with the total energy of both

oscillators as follows
αK 2 Ω 5 Ωα
λ= = (E1 + E2 )2 (3.131)
16β 4β
System (3.130) is periodic with respect to both phase coordinates Δ and ψ,

as a result, its phase plane has periodic cell-wise structure. Let us consider
just one cell,
π π π!
R0 = − < Δ < , 0 < Ψ < (3.132)
2 2 2
including the equilibrium (critical) point
(Δ, ψ) = (0, π/4) (3.133)
As follows from (3.127) and (3.128), at point (3.133), both oscillators vibrate
in-phase with the same energy, E1 = E2 . Linearized (near (3.133)) system
(3.130) has the following couple of roots of characteristic equation

r1,2 = λ ± i 1 − λ2 (3.134)
Π
2
Π
4
0
0 25 50
p
Π
2
Π
2
0 25 50
p
Fig. 3.17 Low energy transition to the “impact” limit cycle of phase variables at
λ = 0.2.
Expression (3.134) determines the ‘low energy’ interval 0 < λ < 1 with a
qualitatively similar system behavior. Equilibrium point (3.133) is unstable
for positive λ while no other equilibrium points exist within the rectangular
(3.132). As a result, the system trajectory is eventually attracted to the
boundary of rectangular R0 (3.132) as shown in Figs. 3.16 and 3.18. This is
a periodic limit cycle whose period is found in a closed form,
π/2 0
dψ dψ 2π
P =2 −2 = √ (3.135)
1 − λ sin 4ψ 1 + λ sin 4ψ 1 − λ2
0 π/2
where the horizontal pieces of the boundary ∂R0 have zero contribution as
those passed momentarily by the system (3.130). This is confirmed also by
the diagrams in Figs. 3.17 and 3.19 showing step-wise jumps of the variable
Δ(p) in steady state limits.
Π
2
Π
4
0
Π Π
2 0 2

Fig. 3.18 Transition to the nonsmooth limit cycle under the energy level approach-
ing its critical value; the numerical solution obtained for the following system pa-
rameter and initial conditions: λ = 0.8; Δ(0) = 0.0, Ψ (0) = π/4 + 0.1
Expression (3.135) shows that P → ∞ as λ → 1. The infinity long pe-

riod means that there is only one-way energy flow in the system, in other
words, the energy is eventually be localized on one of the oscillators. The
corresponding total critical energy value is determined by substituting λ = 1
in (3.131). This gives
β
E1 + E2 = 2 = E∗ (3.136)
Ωα
If E1 + E2 < E ∗ then periodic energy exchange with the period T = βP/Ω
takes place, but no localization is possible. Therefore, in order to be localized
on one of the oscillators, the total system energy must be large enough.
Interestingly enough, the transition to localized mode of this model hap-
pens through non-smooth limit cycle along which the dynamics of phase vari-
ables, Ψ and Δ, resembles the behavior of coordinate and velocity of impact
oscillator2; see Figs. 3.17 and 3.19.
2
As already mentioned, the possibility of ‘vibro-impact dynamics’ of phase vari-
ables was noticed later in [101] when considering another model of nonlinear
beats.
Π
2
Π
4
0
0 15 30
p
Π
2
Π
2
0 15 30
p
Fig. 3.19 Transition to the “impact” limit cycle of phase variables at λ = 0.8.
Chapter 4
Nonsmooth Temporal Transformations
(NSTT)
Abstract. In this chapter, different versions of nonsmooth argument sub-

stitutions, specifically - nonsmooth time, are introduced with proofs of the
related identities. Basic rules for algebraic, differential and integral manipu-
lations are described. In particular, final subsections show how to implement
nonsmooth argument substitutions in the differential equations. These im-
pose two principal features on the dynamical systems by generating specific
algebraic structures and switching formulations to boundary-value problems.
Notice that the transformation itself imply no constraints on dynamical sys-
tems and easily applies to both smooth and nonsmooth systems. Any further
steps, however, should account for physical properties of the related systems.
Indeed, linear coordinate transformations can significantly simplify the lin-
ear dynamic problems. Weakly nonlinear coordinate transformations often
play the major role in the quasi-linear theory. Further, dynamical systems
with discontinuities can be simplified by means of appropriate non-smooth
transformations of variables.
4.1 Non-smooth Time Transformations

Major features induced by nonsmooth temporal substitutions can be briefly
listed as follows:
• Introducing non-smooth temporal variables, in particular triangular sine
wave, brings the coordinates into the algebra of hyperbolic numbers1 ;
• Under appropriate conditions, differentiation or integration of the coor-
dinates keeps the result within the same algebra and therefore eases the
corresponding manipulations with the dynamic systems;
• Explicit time argument can be used together with the nonsmooth time in
order to describe amplitude and/or frequency modulated processes.
1
As mentioned in Introduction, - complex numbers X + Y e, where e2 = 1; see
details in the next subsections.

94 4 Nonsmooth Temporal Transformations (NSTT)
Notice that the transformation itself is a preliminary stage of analysis final-

ized by specific boundary value problems on standard intervals. Then, appro-
priate methods must be applied to the boundary value problems according
to the related physical content.
4.1.1 Positive Time

To begin with a simple illustration of the non-smooth positive time, consider
the series

t t 1 1
ln 2 cosh = + exp (−t) − exp (−2t) + exp (−3t) − ... (4.1)
2 2 2 3
which is convergent for t ≥ 0 but obviously becomes divergent if t < 0; the

convergence is conditional at t = 0.
Nevertheless, replacing t −→ |t|, makes this series convergent for any t as
follows

t |t|
ln 2 cosh ≡ ln 2 cosh
2 2
|t| 1 1
= + exp (− |t|) − exp (−2 |t|) + exp (−3 |t|) − ... (4.2)
2 2 3
An ‘side effect’ of such manipulation is that finite sums of the series are losing
differentiability at t = 0, whereas the original function is smooth everywhere.
However, it will be shown later that the series can be rearranged in such a
manner that any truncated series becomes differentiable at t = 0 as many
times as needed.
Note that the transformation t −→ |t| is non-invertible. As a result, the
manipulation illustrated by example (4.2) may not work for other cases. In
the above example though, the substitution t −→ |t| holds for both positive
and negative time t due to evenness of the original function.
In order to extend the above idea on the general case, let us represent the
time argument in the form
.
t = |t| |t| (4.3)
. 2
Now, taking into account the relationship (|t| ) = 1, gives sequentially
t2n = |t|2n and t2n+1 = (|t|)2n+1 |t|. , n = 1, 2, ... (4.4)
Example 1. Combining identities (4.4) and the power series expansion of the
exponential function, gives

. |t|2 |t|3
exp (t) ≡ exp (|t| |t| ) = 1 + + ... + |t| + + ... |t|.
2! 3!
4.1 Non-smooth Time Transformations 95
or
exp (t) = cosh (|t|) + sinh (|t|) |t|. (4.5)
Expression (4.5) represents the exponential function as a sum of the even and
odd components.
Let us take a general function, x (t), and assume that
. .
x (t) = x (|t| |t| ) = X (|t|) + Y (|t|) |t| (4.6)
This gives the following set of equations for the components, X (|t|) and
Y (|t|),
.
x (|t|) = X (|t|) + Y (|t|) for |t| = 1
.
x (− |t|) = X (|t|) − Y (|t|) for |t| = −1
and, finally,
1
X (|t|) = [x (|t|) + x (− |t|)] (4.7)
2
1
Y (|t|) = [x (|t|) − x (− |t|)]
2
Therefore, identity (4.6) holds under condition (4.7). It is seen that the right-
hand side of expression (4.6) represents an element of the algebra with the
.
basis {1, |t| }. The corresponding ‘table of products’ is generated by the re-
. 2
lationship (|t| ) = 1. This leads to other useful algebraic relationships, for
instance,
.
X + Y |t| = 0 ⇐⇒ {X = 0, Y = 0}
and
. .
f (X + Y |t| ) = Rf (X, Y ) + If (X, Y ) |t| (4.8)
where
1
Rf (X, Y ) = [f (X + Y ) + f (X − Y )]
2
1
If (X, Y ) = [f (X + Y ) − f (X − Y )] (4.9)
2
Therefore, introducing the nonsmooth positive time |t|, imposes specific com-
plexification on the system coordinate
x (t) −→ {X (|t|) , Y (|t|)} (4.10)
As shown in different sections through this chapter, such complexifications

always accompany non-smooth argument substitutions.
4.1.2 ‘Single-Tooth’ Substitution

A a simple generalization of the positive time, |t|, let us introduce the tem-
poral shift, say a, as follows
s = s (t) = |t − a| (4.11)
Obviously, ṡ = −1 for t − a < 0 and ṡ = 1 for t − a > 0 , and therefore ṡ2 = 1

for, at least, almost all t.
Proposition 1. A general function x (t) can be represented in the form
x (t) = X (s) + Y (s) ṡ (4.12)
where s = s (t) is given by (4.11).
Proof. By analogy to (4.3),

t = a + sṡ (4.13)
x (a − s) = X (s) − Y (s) for ṡ = −1

x (a + s) = X (s) + Y (s) for ṡ = 1
and thus,
1
X (s) = [x (a + s) + x (a − s)] (4.14)
2
1
Y (s) = [x (a + s) − x (a − s)]
2
Therefore, identity (4.12) holds for any x (t) under condition (4.14).
4.1.3 ‘Broken Time’ Substitution

Another generalization of the positive time |t|, is given by the piece-wise
linear function
v1 t for t ≤ 0
τ= (4.15)
v2 t for t ≥ 0
where v1 = v2 .
The inverse relationship can be represented in the form
t = A (τ ) + B (τ ) τ̇ (4.16)
where

1 1 1
A= + τ and B = − τ
v1 v2 v1 v2
and
τ̇ 2 = −v1 v2 + (v1 + v2 ) τ̇ (4.17)
Further, applying some function x to both sides of equality (4.16), gives
x (t) = X (τ ) + Y (τ ) τ̇ (4.18)
where both components X and Y are determined analogously to (4.6) through

(4.7) in the form

1 τ τ
X (τ ) = v2 x − v1 x
v2 − v1 v1 v2

1 τ τ
Y (τ ) = − x −x (4.19)
v2 − v1 v1 v2
Now, assuming that another function f is applied to both sides of (4.18),

gives
f (x) = Rf (X, Y ) + If (X, Y ) τ̇ (4.20)
where
1
Rf = [v2 f (X + Y v1 ) − v1 f (X + Y v2 )]
v2 − v1
1
If = − [f (X + Y v1 ) − f (X + Y v2 )] (4.21)
v2 − v1
Comparing the right-hand sides of expressions (4.16), (4.18) and (4.20), shows
that the algebraic structure generated by the nonsmooth time substitutions
is preserved after different functional manipulations with the corresponding
elements of the algebra.
4.1.4 Sawtooth Sine Transformation

Let us consider the periodic version of nonsmooth time (4.11) which is based
on the sawtooth sine - triangular wave function

t for − 1 ≤ t ≤ 1 ∀t
τ (t) = , τ (t) = τ (4 + t) (4.22)
−t + 2 for 1 ≤ t ≤ 3
Recall that function (4.22) describes position of a bead oscillating between

two perfectly stiff parallel barriers with no energy loss. The distance between
the barriers is two spatial units, and the velocity vector has the unit length
and is normal to the barrier planes. In other words, function (4.22) describes
the motion of the standard impact oscillator; see Chapter 1 for illustrations.
This function can be also expressed in the closed form by means of the
trigonometric functions as
τ (t) = (2/π)arcsin[sin(πt/2)] (4.23)
The period is normalized to four in order to provide the unit slope,
τ̇ 2 = 1 (4.24)
Proposition 2. [133] Any periodic process, whose period is normalized to

T = 4, is expressed through the sawtooth sine τ (t) and the rectangular cosine
τ̇ (t) in the form
x (t) = X (τ ) + Y (τ ) τ̇ (4.25)
where the X- and Y -components are given by
1
X (τ ) = [x (τ ) + x (2 − τ )]
2
1
Y (τ ) = [x (τ ) − x (2 − τ )] (4.26)
2
In other words, any periodic process is uniquely expressed through the state
variables, such as the coordinate τ and velocity τ̇ , of the standard impact
oscillator.
Proof [144]. It can be verified by inspection that
t = 1 + (τ − 1) τ̇ if − 1 < t < 3 (4.27)
Note that the function τ̇ (t) has a step-wise discontinuity at t = 1, which is

suppressed however by the continuous factor, τ (t) − 1, of zero value at t = 1.
Based on this remark, property (4.24) will be considered as true everywhere
on the interval −1 < t < 3, since τ̇ is either explicitly or implicitly accompa-
nied by the factor τ (t) − 1 whenever it appears in algebraic manipulations.
For instance, identity
2
t2 = 1 + (τ − 1) + 2 (τ − 1) τ̇
holds everywhere on the interval −1 < t < 3.

Further, applying the method of mathematical induction, gives
tn = An (τ ) + Bn (τ )τ̇ (4.28)
where n is any positive integer, An and Bn are polynomials of the degree n

or n − 1.
Expression (4.28) shows that any analytic function x(t), that admits the
power series expansion on the interval −1 < t < 3, can be represented in the
form
x(t) = x [1 + (τ − 1) τ̇ ] = X (τ ) + Y (τ ) τ̇ (4.29)
where X (τ ) and Y (τ ) are power series of τ .
Now, let us assume that expression (4.29) holds even though the func-
tion x(t) is not analytic. In this case, one must show that the X- and Y -
components can be determined with no power series expansions. Indeed, since
either τ̇ = 1 or τ̇ = −1 on the entire interval −1 < t < 3, except may be the
point t = 1, then expression (4.29) gives two equations, x (τ ) = X (τ ) + Y (τ )
and x (2 − τ ) = X (τ ) − Y (τ ). Solving these equations for X and Y and
substituting the solution in (4.29), gives the identity,
1 1
x(t) = [x (τ ) + x (2 − τ )] + [x (τ ) − x (2 − τ )] τ̇ (4.30)
2 2
which obviously holds on the interval −1 < t < 3, except may be t = 1.
If the function x(t) is continuous at t = 1 then identity (4.30) is also true
at t = 1 because the step-wise discontinuity of the rectangular cosine τ̇ (t) at
t = 1 is suppressed by the factor
2Y (τ ) = x (τ ) − x (2 − τ ) → x (1 − 0) − x (1 + 0) → 0 (4.31)
as t → 1 ± 0; see definition (4.22).

Now if the function x(t) is periodic of the period T = 4 then identity
(4.30) holds almost everywhere on the interval −∞ < t < ∞, because the
right-hand side of (4.30) depends on the time argument t through the pair of
periodic functions τ (t) and τ̇ (t) of the same period T = 4.
Finally, if the function x(t) is continuous also at t = −1, and thus t = 3 due
to the periodicity, then identity (4.30) is true for every t. So let us consider
the point t = −1 at which the function τ̇ (t) has the step-wise discontinuity.
Taking into account that τ (−1 ± 0) = −1 + 0 and the periodicity condition
x(t) = x(t − 4), gives
2Y (τ ) = x[τ (−1 ± 0)] − x[2 − τ (−1 ± 0)] (4.32)

= x (−1 + 0) − x (3 − 0) = x (−1 + 0) − x (−1 − 0) → 0
as t → −1 ± 0 due to continuity of the function x(t) at t = −1. Therefore, the

step-wise discontinuity of the function τ̇ (t) at the point t = −1 is suppressed
as well due to (4.32).
It is proved that if x(t) is continuous then identity (4.30) or (4.25) holds
everywhere on the interval −∞ < t < ∞. The periodic version of conditions
(4.31) and (4.32) is
Y |τ =±1 = 0 (4.33)
Remark 1. If the function x(t) is step-wise discontinuous at the instances

Λ={t : τ (t) = ±1} then limits (4.31) and (4.32) are non-zero. In this case,
the discontinuities of the function τ̇ (t) are not suppressed but describe the
real behavior of the original function x(t).
Proposition 3. Elements (4.25) belong to the algebra of hyperbolic numbers

due to (4.24). As a result,
f (X + Y τ̇ ) = Rf + If τ̇ (4.34)
1
Rf = [f (X + Y ) + f (X − Y )]
2
1
If = [f (X + Y ) − f (X − Y )]
2
provided that f (X ± Y ) are defined.
Relationship (4.34) can be easily verified by setting sequentially τ̇ = 1 and

τ̇ = −1. The right-hand side of expression (4.25) therefore admits interpreta-
tion as a specific (hyperbolic) ‘complex number’ with ‘real’ X and ‘imaginary’
Y components. The corresponding ‘imaginary unity’ τ̇ creates the cycling
group so that τ̇ 2 = 1, τ̇ 3 = τ̇ , τ̇ 4 = 1,... . This remark significantly simplifies
all mathematical manipulations with representation (4.25). For example,
2
(X + Y τ̇ ) = X 2 + Y 2 + 2XY τ̇
3
(X + Y τ̇ ) = X 3 + 3XY 2 + Y 3 + 3Y X 2 τ̇
The right-hand sides of the above expressions appear to have the same hy-
perbolic structure of the element X + Y τ̇ .
Another example resembles Euler formula in the complex analysis, however
hyperbolic functions are involved as follows
exp (X + Y τ̇ ) = exp (X) [cosh (Y ) + sinh (Y ) τ̇ ] (4.35)
Further, by taking into account that τ̇ 2 = 1, one obtains

1 X Y
= 2 − 2 τ̇ if X = ±Y
X + Y τ̇ X −Y 2 X −Y2
This example explains why division is not always possible, and this is essential
difference with the conventional complex analysis.
Very often, the triangular wave function is supposed to depend on some
phase variable, say ϕ(t). In all such cases, the ‘imaginary element’ is given
by the derivative with respect to the phase variable τ (ϕ). In order to avoid
confusion, let us introduce a new notation for the rectangular cosine of the
normalized period and amplitude as follows e(ϕ) = τ (ϕ). Now the basic
relation
e2 = 1 (4.36)
holds regardless argument of the sawtooth sine.
Example 2. Introduce the sawtooth time into the functions sin t and cos t.
Solution. In order to deal with the period normalized to four, let us introduce
new argument ϕ = 2t/π as

π
π 2t
sin t = sin = sin ϕ
2 π 2
Now, applying representation (4.25) to the new argument ϕ , gives

1 π π ! π
X (τ ) = sin τ + sin (2 − τ ) = sin τ
2 2 2 2
1 π π !
Y (τ ) = sin τ − sin (2 − τ ) = 0
2 2 2
and therefore

π 2t
sin t ≡ sin τ
2 π
Analogously,

π 2t 2t
cos t = cos τ e
2 π π
Example 3. Combining the results from previous example with the Euler for-
mula for complex exponential functions, gives

kπ kπ
exp (ikt) = sin τ i + cos τ e for k = 1, 3, 5, ...
2 2

kπ kπ
exp (ikt) = cos τ + sin τ ie for k = 0, 2, 4, ...
2 2
where τ and e still depend on the same argument 2t/π. The right-hand sides
of these equalities can be viewed as elements of a more complicated algebra,
z = α + βe + γi + δei, with the basis elements {1, e, i, ei}. The corresponding
table of products is given by
× 1 e i ei
1 1 e i ei
e e 1 ei i
i i ei −1 −e
ei ei i −e −1
4.1.5 Links between NSTT and Matrix Algebras

Hyperbolic numbers represent a very simple example of so-called Clifford ge-
ometric algebras which is isomorphic to a matrix algebra. For instance, it
is known [105] that complex numbers associate with skew-symmetric 2 × 2
matrixes with equal diagonal entries, whereas the hyperbolic numbers corre-
spond to the symmetric matrixes as follows

a b
a + ib ←→
−b a

XY
X + eY ←→
Y X
The above correspondences are isomorphisms because both summations and

multiplications with the numbers associate with the corresponding matrix
operations, for instance
(X1 + eY1 )(X2 + eY2 ) = (X1 X2 + Y1 Y2 ) + (X1 Y2 + Y1 X2 )e (4.37)

X1 Y1 X2 Y2 X1 X2 + Y1 Y2 X1 Y2 + Y1 X2
= (4.38)
Y1 X1 Y2 X2 X1 Y2 + Y1 X2 X1 X2 + Y1 Y2
As was already mentioned, the algebra of hyperbolic numbers and Clifford

algebras were developed in an abstract way as manipulations with specific
numbers with no relation to any nonsmooth transformations. However, the
uncovered links may appear to be useful for conducting automatic symbolic
manipulations with computers. Indeed, most of the corresponding packages
have built-in tools to handling the matrix operations whereas defining the
operations with hyperbolic numbers may require some programming work.
4.1.6 Differentiation and Integration Rules

From the point of view of applications to the differential equations, it is
essential that, under some conditions, differential operations also preserve
the hyperbolic structure of elements [133]. For example,
ẋ = Y + X τ̇ + Y τ̈ (4.39)
where primes mean derivatives with respect to τ .

The term underlined in (4.39) should be ignored whenever the function x(t)
is continuous and therefore condition (4.33) is satisfied. Indeed, the derivative
τ̈ represents the periodic system of pulses, acting at those points Λ where the
factor Y is equal to zero due to (4.33),
∞

τ̈ = 2 [δ(t + 1 − 4k) − δ(t − 1 − 4k)]
k=−∞
In a similar manner, one can sequentially consider high-order derivatives. For

example, the second derivative is given by
ẍ = X + Y τ̇ (4.40)
under the boundary condition
X |τ =±1 = 0 (4.41)
Further, an arbitrary odd derivative is given by
x(2k−1) (t) = Y (2k−1) (τ ) + X (2k−1) (τ ) e (4.42)
provided that
Y (τ ) |τ =±1 = 0
X (τ ) |τ =±1 = 0
... (4.43)
Y (2k−2)
(τ ) |τ =±1 =0
Also an arbitrary even derivative is
x(2k) (t) = X (2k) (τ ) + Y (2k) (τ ) e (4.44)
if
Y (τ ) |τ =±1 = 0
X (τ ) |τ =±1 = 0
... (4.45)
X (2k−1)
(τ ) |τ =±1 = 0
Finally, integration also gives the hyperbolic ‘number’

(X + Y τ̇ )dt = Q + P τ̇ (4.46)
whose components are

τ τ
Q(τ ) = Y dτ + C and P (τ ) = Xdτ
0 −1
where C is an arbitrary constant, and the following condition must be satisfied
1
X(τ )dτ = 0 (4.47)
−1
Relationship (4.46) can be easily verified by differentiation with respect to t.

The role of condition (4.47) is to provide zero mean value of the integrand in
(4.46).
4.1.7 NSTT Averaging

Lemma 1. Let x(t) be a general periodic function of the period T = 4a so
that presentation (4.25) holds
x(t) = X(τ (φ)) + Y (τ (φ))e(φ) (4.48)
where φ = t/a, e(φ) = τ (φ) and

1
X (τ ) = [x (aτ ) + x (2a − aτ )]
2
1
Y (τ ) = [x (aτ ) − x (2a − aτ )] (4.49)
2
Then the mean value of x(t) over its period is
T 1
1 1
x(t)dt = X(τ )dτ (4.50)
T 0 2 −1
In other words, the ‘imaginary’ component Y e of the ‘hyperbolic number’

(4.48) gives zero contribution into the mean value.
Proof. With no loss of generality, let us assume that a = 1 and thus T = 4.

Then, during one period −1 < t < 3,
τ (t) = e(t) = 1 and dt = dτ for − 1 < t < 1

τ (t) = e(t) = −1 and dt = −dτ for 1 < t < 3
Taking into account (4.48), gives

T 1 3
1 1
x(t)dt = x(t)dt + x(t)dt
T 0 4 −1 1
1 1 1
1 1 1
= [X(τ ) + Y (τ )]dτ − [X(τ ) − Y (τ )]dτ = X(τ )dτ
4 −1 4 −1 2 −1
Example 4. Let x(t) be a periodic function of the general period T . Taking

into account (4.48) and the above remark, gives
x2 = (X + Y e)2 = X 2 + Y 2 + 2XY e
Then, based on the above Lemma,

T 1
1 1
< x2 >≡ x2 dt = (X 2 + Y 2 )dτ (4.51)
T 0 2 −1
Example 5. Consider the case x(t) = A sin t + B cos t, where T = 2π and

a = π/2. In this case, relationships (4.56) give X(τ ) = A sin(πτ /2) and
Y (τ ) = A cos(πτ /2). Therefore, (4.51) gives < x2 >= (A2 + B 2 )/2.
4.1.8 Generalizations on Asymmetrical Sawtooth

Wave
It has been shown in [138] and [149] that an arbitrary periodic function x (t)
whose period is normalized to four can still be represented in the form (4.25),
even though the triangular wave τ - saw-tooth sine - has asymmetrical slopes
such that

t/ (1 − γ) for − 1 + γ ≤ t ≤ 1 − γ
τ (t, γ) = (4.52)
(−t + 2) / (1 + γ) for 1 − γ ≤ t ≤ 3 + γ
∂τ (t, γ)
τ̇ (t, γ) = = e (t, γ) (4.53)
∂t
where γ ∈ (−1, 1) is a parameter characterizing the inclination of the saw as
shown in Fig. 4.1.
Τ
1
t
1 2 3 4 5 6 7 8
1
e
1 Γ1
t
1 2 3 4 5 6 7 8
1
1 Γ
Fig. 4.1 Asymmetric saw-tooth sine and rectangular cosine (γ = 0.5).
In this case, the inverse transformation of time on the period has the form

1 − γ2 1 1
t= [(1−γ) τ ] +e +[2−(1+γ) τ ] −e (4.54)
2 1+γ 1−γ
(−1 + γ ≤ t ≤ 3 + γ)
By using this expression, one can show that
x (t) = X [τ (t, γ)] + Y [τ (t, γ)] e (t, γ) (4.55)


1 1 1
X= x [(1 − γ) τ ] + x [2 − (1 + γ) τ ]
2α 1 + γ 1−γ
1
Y = {x [(1 − γ) τ ] − x [2 − (1 + γ) τ ]} (4.56)
2α
e2 = α + βe (4.57)

where α = 1/ 1 − γ 2 and β = 2γα.
As compared to (4.36), relationship (4.57) somewhat complicates algebraic
and differential operations so that, respectively,
f (X + Y e) = Rf (X, Y ) + If (X, Y ) e (4.58)

1 1 1
Rf (X, Y ) = f (Z+ ) + f (Z− )
2α 1 + γ 1−γ
1
If (X, Y ) = [f (Z+ ) − f (Z− )]
2α
Y
Z± = X ±
1∓γ
and
∂e
ẋ = αY + (X + βY ) e + Y (4.59)
∂t
where f (Z+ ) and f (Z− ) are defined, primes indicate differentiation with
respect to τ , and
∞

∂e
= 2α [δ (t + 1 − γ − 4k) − δ (t − 1 + γ − 4k)] (4.60)
∂t
k=−∞
If the function x (t) is continuous at time instances Λ = {t : τ (t, γ) = ±1}

then the singular term ∂e/∂t is suppressed by the condition Y |t∈Λ =
Y |τ =±1 = 0 analogously to the symmetric case. Otherwise, the periodic sin-
gular term remains in expression (4.59).
It is convenient to calculate high-order derivatives by means of the matrix-
operator D acting on the vector-column (X, Y )T sequentially as follows

0α d
D=
1 β dτ

X αY
D = (4.61)
Y X + βY

X αX + αβY
D2 =
Y βX + (α + β 2 )Y
Therefore, applying D and D2 , gives
ẋ = αY + (X + βY )e
ẍ = αX + αβY + [βX + (α + β 2 )Y ]e
under the conditions
Y |τ =±1 = 0
(X + βY )|τ =±1 = 0

Note that rules (4.61) are still valid in the symmetric case γ = 0, when the
differentiation matrix operator takes the form

01 d
D=
1 0 dτ
Finally, the result of integration (4.46) has the components

τ
β 1
Q= Y (τ ) − X (τ ) dτ and P = X (ξ) dξ
α α
−1
under the zero-mean value condition (4.47).
4.1.9 Multiple Frequency Case

Practical applications of NSTT to the class of multiple frequency motions face
problems similar by its nature to those caused small denominators in quasi-
linear approaches. However, formal generalizations of the basic identities are
quite simple as illustrated below.
Let Pϕ and Nϕ be operators acting on some periodic function x(ϕ) of the
period T = 4 as follows
1
Pϕx (ϕ) = {x[τ (ϕ)] + x[2 − τ (ϕ)]} ≡ X[τ (ϕ)] (4.62)
2
1
Nϕx (ϕ) = {x[τ (ϕ)] − x[2 − τ (ϕ)]} ≡ Y [τ (ϕ)]
2
where functions X (τ ) and Y (τ ) are defined according to (4.26).
In such notations, representation (4.25) takes the form
x(ϕ) = (Pϕ + eNϕ)x(ϕ) (4.63)
where e (ϕ) = τ (ϕ).

Let us consider now a function of multiple arguments x = x (ϕ1 , ..., ϕn ) of
the period T = 4 with respect to each of its n arguments. This function de-
scribes a multiple frequency quasi-periodic process if ϕ1 = ω1 t,..., ϕn = ωn t
with a set of positive incommensurable numbers ω1 ,..., ωn . In this case, trans-

formation (4.63) is independently applicable to each of the n arguments as
follows

n
x (ϕ1 , ..., ϕn ) = (Pϕj + ej Nϕj )x (ϕ1 , ..., ϕn ) (4.64)
j=1
where new notations τj = τ (ϕj ) and ej = τ (ϕj ) are introduced.

Note that all the operators included in the product are commutating as
those applied to different arguments of the function. Moreover, the number
of co-factors of the product can differ from one to n.
Let us consider the case of two arguments in details. Introducing the no-
tations e0 ≡ 1 and e3 = e1 e2 , and taking the product as shown in (4.64),
gives
x (ϕ1 , ϕ2 ) = X (τ1 , τ2 ) e0 + Y (τ1 , τ2 ) e1 + Z (τ1 , τ2 ) e2 + W (τ1 , τ2 ) e3 (4.65)
where
X = Pϕ1 Pϕ2 x (ϕ1 , ϕ2 )

Y = Nϕ1 Pϕ2 x (ϕ1 , ϕ2 )
Z = Pϕ1 Nϕ2 x (ϕ1 , ϕ2 )
W = Nϕ1 Nϕ2 x (ϕ1 , ϕ2 )
The basis of elements (4.65) obeys the table of products
× e0 e1 e2 e3
e0 e0 e1 e2 e3
e1 e1 e0 e3 e2 (4.66)
e2 e2 e3 e0 e1
e3 e3 e2 e1 e0
Further, applying some function f to the element (4.65), gives
f (Xe0 + Y e1 + Ze2 + W e3 ) = Rf e0 + If1 e1 + If2 e2 + If3 e3 (4.67)
where the coefficients at the right-hand side are defined by the system of
linear equations
Rf + If1 + If2 + If3 = f (X + Y + Z + W )

Rf + If1 − If2 − If3 = f (X + Y − Z − W )
Rf − If1 + If2 − If3 = f (X − Y + Z − W ) (4.68)
Rf − If1 − If2 + If3 = f (X − Y − Z + W )
4.2 Idempotent Basis Generated by the Triangular Sine-Wave 109
Equations (4.68) are obtained by substituting different combinations of e1 =

±1 and e3 = ±1 in (4.67). Some application of the multiple frequency case
will be illustrated at the end of Chapter 5 and also in Chapter 14.
4.2 Idempotent Basis Generated by the Triangular

Sine-Wave
4.2.1 Definitions and Algebraic Rules
In addition to the standard basis {1, e}, the hyperbolic plane has another
natural basis {e+ , e− }associated with the two isotropic lines separating the
hyperbolic quadrants as described in Chapter 1. The transition from one basis
to another is given by (see Fig. 4.2)
1
e+ = (1 + e)
2
1
e− = (1 − e) (4.69)
2
or
1 = e+ + e− (4.70)
e = e+ − e−
Therefore,
x = X + Y e = X(e+ + e− ) + Y (e+ − e− )
= (X + Y )e+ + (X − Y )e−
where x = x(t) is any periodic function whose period is normalized to T = 4,

and therefore e = e(t).
Taking into account (4.26), gives
x = X+ (τ )e+ + X− (τ )e− (4.71)
where τ = τ (t), e+ = e+ (t), e− = e− (t), and
X+ = X + Y = x(τ )
X− = X − Y = x(2 − τ ) (4.72)
The elements e+ and e− are mutually annihilating, and they are called idem-
potents because
e+ e− = 0
e2− = e− (4.73)
e2+ = e+
Fig. 4.2 The standard basis and idempotent basis (on the left and right, respec-
tively.)
As follows from (4.70), ee+ = e+ and ee− = −e− . Obviously, the matrix
representation of (4.71) is

X+ 0
X+ e+ + X− e− ←→
0 X−
Properties (4.73) make the idempotent basis very convenient to use in differ-
ent calculations. For instance, for any real number α,
(X+ e+ + X− e− )α = X+
α α
e+ + X− e− (4.74)
One can extend relationship (4.74) on a general function f as follows
f (X+ e+ + X− e− ) = f (X+ )e+ + f (X− )e− (4.75)
provided that f (X+ ) and f (X− ) are defined.

As seen from (4.75), any function is acting as linear if applied to a
hyperbolic element represented in the idempotent basis. As a result, the
major advantage of using the idempotent basis is that the differential equa-
tions of motion in terms of the components X+ (τ ) and X− (τ ) appear to
be decoupled regardless nonlinearity. However, the corresponding boundary
value problems still remain coupled because the idempotent basis makes
4.2 Idempotent Basis Generated by the Triangular Sine-Wave 111
unfortunately boundary conditions coupled. The nature of this fact is ex-

plained in the next subsection.
The time variable t can be represented in the idempotent basis during one
period of a periodic process, say T = 4. This can be shown by substituting
(4.70) in (4.27), where τ̇ (t) = e(t), and then conducting some manipulations
as follows
t = 1 + (τ − 1) e = (e+ + e− ) + (τ − 1) (e+ − e− )
= τ e+ + (2 − τ )e− (−1 < t < 3) (4.76)
Further, using the basis’ properties (4.73), gives, for instance,
t2 = τ 2 e+ + (2 − τ )2 e−
tα = τ α e+ + (2 − τ )α e− (4.77)
(α > 0)
Note that expressions (4.77) provide the direct proof of representation (4.71),
at least, in the class of analytical periodic functions x(t).
4.2.2 Time Derivatives in the Idempotent Basis

Taking into account definitions (4.69) and (4.70), gives first time derivatives
of the basis elements
1
ė+ = −ė− = ė (4.78)
2
where
∞

ė+ = [δ(t + 1 − 4k) − δ(t − 1 − 4k)]
k=−∞
Therefore, differentiating (4.71) gives

ẋ = X+ ee+ + X− ee− + X+ ė+ + X− ė−
1
= X+ e+ − X− e− + (X+ − X− )ė
2
Assuming the continuity condition
(X+ − X− )|τ =±1 = 0 (4.79)
gives

ẋ = X+ e+ − X− e− (4.80)
Analogously, assuming the continuity condition for ẋ(t),

(X+ + X− )|τ =±1 = 0 (4.81)
gives then

ẍ = X+ e+ + X− e−
Example 6. Suppose the equation ẋ = f (x), where x(t) ∈ Rn , has a family

of periodic solutions of the period T = 4a. Introducing the sawtooth time
τ = τ (t/a) and using the idempotent basis, gives
1
(X+ e+ − X− e− ) = f (X+ e+ + X− e− ) = f (X+ )e+ + f (X− )e−
a
or

X+ = af (X+ )

X− = −af (X− ) (4.82)
under condition (4.79). As mentioned above, the sawtooth time substitution

implemented in the idempotent basis leads to the decoupled set of equations
such as (4.82), whereas boundary condition (4.79) is coupled. Note that it
is practically sufficient to solve only first equation of system (4.82). Then
solution of the second equation is obtained by making replacement τ → −τ
in the first solution.
4.3 Idempotent Basis Generated by Asymmetric

Triangular Wave
4.3.1 Definition and Algebraic Properties

Let us introduce basic algebraic manipulations with the idempotent basis
generated by the asymmetric triangular (sawtooth) wave (4.52). The standard
basis is given by the set {1, e} and shown on the left in Fig. 4.3, where
e = e(t, γ) is defined by (4.53) periodic function with step-wise discontinuities
at the points Λγ = {t : τ (t, γ) = ±1}.
In this case, we introduce the idempotent basis as follows
1
e+ = [1 − γ + (1 − γ 2 )e]
2
1
e− = [1 + γ − (1 − γ 2 )e] (4.83)
2
or, inversely,
1 = e+ + e− (4.84)
1 1
e= e+ − e−
1−γ 1+γ
where e+ = e+ (t, γ) and e− = e− (t, γ), and the parameter of asymmetry γ

is included in order to normalize the range of change for the basis elements
as 0 ≤ e+ ≤ 1 and 0 ≤ e− ≤ 1; see the diagrams on the left in Fig. 4.3.
4.3 Idempotent Basis Generated by Asymmetric Triangular Wave 113
Fig. 4.3 The standard basis and idempotent basis (on the left and right, respec-
tively) generated by the asymmetric triangular wave with γ = 0.3.
Definition (4.83) brings (4.54) to the form
t = (1 − γ) τ e+ + [2 − (1 + γ) τ ] e− (4.85)
If γ = 0 then definition (4.83) becomes equivalent to (4.69). However, the

elements e+ and e− are mutually annihilating, and create idempotents re-
gardless the magnitude of γ so that table of products (4.73) still holds for
any γ ∈ (−1, 1). Also, as follows from (4.73) and (4.84),
1
ee+ = e+
1−γ
1
ee− = − e− (4.86)
1+γ
Now, substituting (4.84) into identity (4.55), gives

1 1
x = X + Y e = X(e+ + e− ) + Y e+ − e−
1−γ 1+γ

1 1
= X+ Y e+ + X − Y e− (4.87)
1−γ 1+γ
where x = x(t) is any periodic function whose period is normalized to T = 4.

Substituting (4.56) in (4.87) and conducting algebraic manipulations, gives
x = X+ (τ, γ)e+ + X− (τ, γ)e− (4.88)
where τ = τ (t, γ) is defined above by (4.52), and
X+ (τ, γ) = x((1 − γ)τ ) and X− (τ, γ) = x(2 − (1 + γ)τ ) (4.89)
Obviously, basic algebraic properties (4.74) and (4.75) remain valid.
4.3.2 Differentiation Rules

Taking into account definitions (4.83) and (4.84), gives first time derivatives
of the basis elements
∂e+ ∂e− 1 ∂e
=− = (1 − γ 2 ) (4.90)
∂t ∂t 2 ∂t
where ∂e/∂t is a sequence of Dirac δ-functions taking effect whenever t ∈ Λγ.
Taking into account that ∂τ /∂t = e and differentiating (4.88), gives
∂X+ ∂X− ∂e+ ∂e−

ẋ = ee+ + ee− + X+ + X− (4.91)
∂τ ∂τ ∂t ∂t
Substituting (4.86) and (4.90) in (4.91), brings (4.91) to the form
1 ∂X+ 1 ∂X− 1 ∂e
ẋ = e+ − e− + (1 − γ 2 )(X+ − X− ) (4.92)
1 − γ ∂τ 1 + γ ∂τ 2 ∂t
Assuming the continuity of x(t), leads to boundary conditions
(X+ − X− )|τ =±1 = 0 (4.93)
These conditions eliminate the singular term ∂e/∂t from expression (4.92),
which, as a result, takes the form
1 ∂X+ 1 ∂X−
ẋ = e+ − e− (4.94)
1 − γ ∂τ 1 + γ ∂τ
As seen from (4.94), the algebraic structure of representation (4.88) is pre-

served after the differentiation. In other words, under continuity condition
(4.93), the result of differentiation just leads to the following replacements in
(4.88)
1 ∂X+ 1 ∂X−
X+ → and X− → − (4.95)
1 − γ ∂τ 1 + γ ∂τ
Property (4.95) enables one of writing down high derivatives iteratively by
making substitutions (4.95) in (4.92). For instance, second derivative is given
by
4.3 Idempotent Basis Generated by Asymmetric Triangular Wave 115
1 ∂ 2 X+ 1 ∂ 2 X−
ẍ = e+ + e−
(1 − γ) ∂τ
2 2 (1 + γ) ∂τ 2
2

1 1 ∂X+ 1 ∂X− ∂e
+ (1 − γ 2 ) + (4.96)
2 1 − γ ∂τ 1 + γ ∂τ ∂t
The singular term ∂e/∂t in (4.96) is eliminated by condition

1 ∂X+ 1 ∂X−
+ |τ =±1 = 0 (4.97)
1 − γ ∂τ 1 + γ ∂τ
As a result, the second derivative ẍ takes the form of expansion with respect
to the basis {e+ , e− }. Such a process can be continued as soon as derivatives
remain continuous in Λγ . Otherwise, time derivatives of the function e(t, γ)
acquire non-formal meaning and must be present in the entire expression.
4.3.3 Oscillators in the Idempotent Basis

In order to illustrate the use of idempotent basis for vibration problems,
consider oscillator
ẍ + f (x) = pe+ (t/a, γ) + qe− (t/a, γ) (4.98)
where p and q are constant amplitudes of the external loading represented in

the idempotent basis with the period T = 4a.
No specific conditions are imposed on the characteristic f (x), however,
it is assumed that oscillator (4.98) possesses periodic solution with the pe-
riod of external loading, T . Such a solution admits the form (4.88), where
τ = τ (t/a, γ). Due to the presence of time scaling factor, expressions for
time derivatives derived in the previous subsection must be modified by the
replacement ∂τ → a∂τ . Now, substituting (4.88) in (4.98), taking into ac-
count (4.93) through (4.97) and the basic algebraic property of idempotents,
f (X+ e+ + X− e− ) = f (X+ )e+ + f (X− )e− , gives equations
1 ∂ 2 X+
+ f (X+ ) = p (4.99)
(1 − γ)2 a2 ∂τ 2
1 ∂ 2 X−
+ f (X− ) = q (4.100)
(1 + γ)2 a2 ∂τ 2
under the boundary conditions (4.93) and (4.97).

The obvious advantage of using the idempotent basis is that equations
(4.99) and (4.100) are in a better match with the form of original equa-
tion (4.98), appear to be decoupled and therefore can be solved in a similar
way. Although the entire boundary value problem is still coupled through
the boundary conditions, the problem caused by coupling is eased, however,
since algorithms for solving equations are often more complicated than those
applied to the boundary conditions.
4.3.4 Integration in the Idempotent Basis

Let x(t) be a periodic function of the period normalized to T = 4 with zero
mean value on the period T . Then, integrating the right-hand side of (4.88),
gives

[X+ (τ, γ)e+ + X− (τ, γ)e− ]dt = P+ (τ, γ)e+ + P− (τ, γ)e− (4.101)
where
τ
P+ (τ, γ) = (1 − γ) X+ (z, γ)dz + C
−1
τ
P− (τ, γ) = −(1 + γ) X− (z, γ)dz + C (4.102)
−1
Note that both expressions in (4.102) have the same arbitrary constant of
integration C. As a result, substituting (4.102) in (4.101) and taking into
account (4.84), gives a single constant of integration as follows: Ce+ + Ce−
= C(e+ + e− ) = C.
Proof of (4.102) is obtained by taking time derivative of both sides of
equality (4.101). Under the condition of continuity of the right-hand side of
(4.101), the differentiation leads to the boundary value problem
1 ∂P+ (τ, γ)
= X+ (τ, γ)
1−γ ∂τ
1 ∂P− (τ, γ)
− = X− (τ, γ) (4.103)
1+γ ∂τ
(P+ − P− )|τ =±1 = 0
This boundary value problem has solution (4.102) under the following con-
dition though
1 1
(1 − γ) X+ (τ, γ)dτ + (1 + γ) X− (τ, γ)dτ = 0 (4.104)
−1 −1
The meaning of this condition is clarified by substituting (4.89) in (4.104) and

taking into account the standard properties of definite integrals as follows
4.4 Discussions, Remarks and Justifications 117
1 1
x[(1 − γ)τ ]d[(1 − γ)τ ] + x[2 − (1 + γ)τ ]d[(1 + γ)τ ]
−1 −1

(1−γ)
(1−γ)
3+γ T
= x(z)dz − x(z)dz = x(z)dz = x(t)dt = 0
−(1−γ) 2+(1+γ) −1+γ 0
In other words, condition (4.104) requires zero mean value for x(t) on the
period. Otherwise, the result of integration would appear to be out of the
class of periodic functions by making the algebraic structure of right-hand
side of (4.101) invalid.
4.4 Discussions, Remarks and Justifications

Let us show that the saw-tooth temporal argument τ (t) associates with the
very general group properties of the conservative oscillator
d2 x
+ f (x) = 0 (4.105)
dt2
Suppose that another oscillating time parameter, say g, is introduced as
x(t) = X[g (t)], where g (t) is a periodic function, which is not necessarily
sawtooth. This kind of substitution brings the differential equation of motion
to the form
2 2
dg d X d2 g dX
+ + f (X) = 0 (4.106)
dt dg 2 dt2 dg
A side effect of such substitution is that, generally speaking, the system has
lost its original Newtonian form. In order to keep the differential equation
of motion in its original Newtonian form, the following conditions must be
imposed on equation (4.106)

2
dg d2 g dX
=1 and =0 (4.107)
dt dt2 dg
Conditions (4.107) are satisfied by
g =α+t or g = β − t (4.108)
where α and β are arbitrary constants.

Relationships (4.108) simply represent the group of time transformations
which is admitted by the original system (4.105). Let us recall however that
the function g (t) was assumed to be periodic. Therefore, according to (4.108),
g (t) must be a piece-wise linear periodic function. The simplest one-frequency
case is given by the triangular wave, g (t) ≡ τ (t). In this case the first equality
of (4.107) is equivalent to the basic algebraic relationship (4.24) whereas the
second equality holds for all t under the condition (4.41). As a result, the
new equation (4.106) takes the same form as (4.105).
Therefore, the triangular wave time substitution τ (t) possesses the unique
property among all periodic time substitutions, namely, it preserves the form
of differential equations of conservative oscillators.
4.4.1 Remarks on Nonsmooth Solutions in the

Classical Dynamics
Problems of classical dynamics are usually formulated in terms of second
order differential equations of motion with respect to the system coordi-
nates. Therefore, the corresponding solutions must be at least twice continu-
ously differentiable functions of time. The existence and uniqueness theorem
imposes also special conditions on the system characteristics and external
forcing functions. As a result, using discontinuities or distributions for mod-
eling dynamical systems takes the corresponding differential equations out of
frames of the classic theory of differential equations. This requires additional
examination of such formulations in order to insure that solutions actually
exist. One of the NSTT roles is to bring models back into the area of classic
theory. However, some preliminary analyses of correctness of original models
is still required especially in non-linear cases. In this and next subsections,
some details and examples are introduced on nonlinear formulations with
discontinuities and δ-functions.
For illustration purposes, let us consider a one-dimensional forced oscillator
ẍ + f (x, ẋ) = q (t) (4.109)
where the functions q (t), f (x, ẋ) and ∂f (x, ẋ) /∂ ẋ must be continuous ac-
cording to the Cauchy theorem.
Let the system be subjected to an impulsive external loading applied at
t = t0 . In this case, the external forcing function can be expressed by the
Dirac δ-function2
ẍ + f (x, ẋ) = Iδ (t − t0 ) (4.110)
where I is a constant linear momentum per unit mass.
The differential equation of motion (4.110) cannot be treated within the
classic theory of differential equations. It is also impossible to find a solution
in the class of twice differentiable functions. However, a physically meaningful
interpretation of equation (4.110) is suggested by the theory of distributions.
For example, in terms of function-theoretic approaches, distributions are not
classical functions but linear functionals acting on manifolds of appropriate
differentiable functions {ϕ (t)}. Further, equalities are thought to be integral
2
The Dirac δ-function belongs to the class of so-called generalized functions intro-
duced by Sergei Sobolev in 1935 and re-introduced by Laurent Schwartz in the
late 1940s, who developed a theory of distributions.
identities rather then point-wise relationships. Therefore, a ‘real meaning’ of

equation (4.110) is given by the identity
∞ ∞
[ẍ + f (x, ẋ)] ϕ (t) dt = I δ (t − t0 ) ϕ (t) dt (4.111)
−∞ −∞
which is supposed to be true for any ‘testing function’ ϕ (t).

Taking into account the definition of δ-function and integrating by parts,
gives
∞
[−ẋϕ̇ (t) + f (x, ẋ) ϕ (t)] dt = Iϕ (t0 ) (4.112)
−∞
On one hand, this expression allows to weaken the smoothness condition on

the class of solutions x (t). On the other hand, the integral identity (4.111)
appears to have quite clear physical meaning since its both sides represent
works done on arbitrary virtual displacements ϕ (t).
Note, that the solution x (t) itself is not considered as a functional unless
the function f is linear. For example, let us consider the case f (x, ẋ) = 2λẋ
+ω 2 x, where λ and ω are constant parameters. In this case, by proceeding
with integration by parts in equation (4.112), one obtains
∞

ϕ̈ (t) − 2λϕ̇ (t) + ω 2 ϕ (t) x (t) dt = Iϕ (t0 )
−∞
"∞
Therefore, x (t) generates a linear functional of the form (...) x (t) dt and
−∞
thus can be qualified as a generalized solution3 .
In nonlinear cases, the concept of generalized solution does not work, but
equation (4.112) still holds and enables one to use the concept of ‘weak so-
lution’ [46]. This concept is involved to justify manipulations with the nons-
mooth functions at intermediate steps of the approach.
4.4.2 Caratheodory Equation

Since δ-function is defined as a linear functional then nonlinear operations
with δ-functions are generally ‘illegal.’ However, δ-functions still can par-
ticipate in nonlinear differential equations in a linear way - as summands.
3
Sometimes the very presence of δ-functions in any nonlinear equation is rejected
by the reason that nonlinearity is incompatible with the notion of linear function-
als. In differential equations of motion, however, singular forces and the corre-
sponding accelerations usually participate as summands, whereas velocities and
coordinates include no δ-type singularities and thus can be subjected to nonlinear
operations.
Such cases were examined in details by Caratheodory [45]. Let us consider

the differential equation ẋ = f (t, x), where the right-hand side satisfies the
Caratheodory conditions. Namely, in the domain D of the (t, x) space: the
function f (t, x) is defined and continuous with respect x for almost all t;
the function f (t, x) is measurable in t for each x; and |f (t, x)| ≤ m (t),
where function m (t) is summable.
The above conditions are less restrictive than those required by the classi-
cal existence theorem namely the function f (t, x) is allowed to be step-wise
discontinuous in t. Such an extension becomes possible if the right-hand side
of the equivalent integral equation is calculated by Lebesgue,
t
x (t) − x (t0 ) = f (ξ, x (ξ)) dξ
t0
Now, let us the right-hand side of the equation includes the δ-impulse as a
summand
ẋ = f (t, x) + νδ (t − a) (4.113)
where a and v are constant parameters.
In this case, the right-hand side does not satisfy any more the Caratheodory
conditions. However, changing the variable
x (t) −→ y (t) : x (t) = y (t) + vH (t − a) (4.114)
where H is the Heaviside’s unit-step function, brings the differential equation

back into the form
ẏ = f [t, y + vH (t − a)] ≡ F (t, y) (4.115)
where the Dirac’s function has been eliminated due to the equality Ḣ (t − a) =
δ (t − a); as a result the right-hand side, F (t, y), satisfies the Caratheodory
conditions.
Note that, due to the presence of discontinuous function H (t − a), the
right hand side of the transformed equation, F (t, y), is generally piecewise
continuous in t, even though the function f (t, x) may be continuous.
Relationships (4.113) through (4.115) admit direct extensions on the vector
space so that many mechanical and physical models can be represented in
the form (4.113).
Example 7. Parametrically excited Duffing’s oscillator subjected to external

impact at t = a is described by equation
ü + ω02 (1 + ε cos 2t) u − βu3 = 2pδ (t − a) (4.116)
Equation (4.116) is transformed to (4.113) by introducing the vector-functions

as follows

u x1 0
x= = , ν=
u̇ x2 2p

x2
f (t, x) =
βx31 − ω02 (1 + ε cos 2t) x1
Example 8. Center lines of linearly elastic beams resting on elastic founda-

tions can be described by the dimensionless differential equation
d4 W
+ γ (ξ) W = qδ (ξ − a) (4.117)
dξ 4
where W = W (ξ) is the center line coordinate, γ (ξ) is a variable stiffness of

the foundation, q is a transverse force localized at ξ = a. Equation (4.117)
is brought to the form (4.113) by considering t as a spatial coordinate ξ and
introducing the matrixes
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
W 0 100 0
⎢ dW/dξ ⎥ ⎢0 0 1 0 ⎥ ⎢0⎥
x=⎢ ⎥
⎣ d2 W/dξ 2 ⎦ , f (ξ, x) = ⎢
⎣0
⎥ x, ν=⎢ ⎥
0 0 1⎦ ⎣0⎦
d3 W/dξ 3 −γ (ξ) 0 0 0 q
In order to completely formulate the problem, the corresponding boundary

conditions must be added.
Let us show now that introducing the nonsmooth argument (4.11) eliminates
the delta-pulse in equation (4.113) in such a way that no step-wise disconti-
nuity occurs in the new equations. Indeed, substituting (4.12) and (4.13) into
equation (4.113) and taking into account expression s̈ = 2δ (t − a), gives
Y + Rf (s, X, Y ) + [X + If (s, X, Y )] ṡ + (2Y − v) δ (t − a) = 0 (4.118)
where expressions
1
Rf = [f (a + s, X + Y ) + f (a − s, X − Y )]
2
1
If = [f (a + s, X + Y ) − f (a − s, X − Y )]
2
are obtained analogously to (4.34).
Eliminating the delta-pulse in (4.118) and equating separately both com-
ponents of the remaining hyperbolic element to zero, gives the boundary-value
problem
X + If (s, X, Y ) = 0 (4.119)
Y + Rf (s, X, Y ) = 0
v
Y |s=0 =
2
Although equations (4.119) have a more complicated form as compared to

(4.113), the step-wise discontinuous function H (t − a) is not involved any
more, also it may become important that the new argument s is half-limited:
0 ≤ s < ∞.
4.4.3 Other Versions of Periodic Time Substitutions

In this subsection, it will be shown that the symmetric sawtooth sine gener-
ates the most simple algebraic structure among other possible versions of
non-invertible time substitutions. Note that periodic temporal arguments
have been considered in the literature for a long time. The main idea of
these approaches is to make a new temporal argument limited and therefore
expand class of usable algorithms. For example, the harmonic transformation
of time in combination with the power series method was used by Ince [70]
for investigation of periodic motions. In particular, by introducing the new
variables
τ = sin t and x (t) = X[τ (t)] (4.120)
in the Mathieu equation,
ẍ + (a + b cos 2t) x = 0
one obtains the equation

1 − τ 2 Xτ2 − τ Xτ + a + b − 2bτ 2 X = 0
which admits periodic solutions in terms of the power series with respect to
the new temporal argument |τ | ≤ 1 [195].
Transformation (4.120) with the power series methods were employed for
non-linear vibrating systems as well [198], [110], [163]. Non-harmonic time
transformations dealing with Jacobian functions can be also found in the
literature [21]. As it is known, however, such transformations of time are
restricted by special cases and cannot be applied to any periodic motion.
From the mathematical point of view, it is caused by the fact that an inverse
transformation does not exist on the whole period.
In this section, different versions of periodic time are introduced in such a
way that the corresponding transformations are valid for any periodic motion.
This is reached by a special complexification of the coordinates.
Let us start with a generalization of substitution (4.120). For the sake of
compactness, notations
τ = τ (ϕ) ≡ sin ϕ and e = e (ϕ) ≡ cos ϕ (4.121)
will be used below, where e (ϕ) = τ (ϕ).

Proposition 4. Any sufficiently smooth periodic function x (ϕ) of the period
T = 2π, can be represented in the form
x (ϕ) = X[τ (ϕ)] + Y [τ (ϕ)]e (ϕ) (4.122)
where X and Y are of the power series form with respect to τ .
Proof. By collecting separately terms with odd and even wave numbers in
the corresponding Fourier series, one obtains
∞
A0
x (ϕ) = + [A2n cos 2nϕ + A2n−1 cos (2n − 1) ϕ]
2 n=1
∞

+ [B2n sin 2nϕ + B2n−1 sin (2n − 1) ϕ] (4.123)
n=1
Then, introducing notations (4.121) into the tabulated expressions [52] (For-
mulas No. 1.332), gives
n
n
cos 2nϕ = a2i τ 2i , cos[(2n − 1) ϕ] = a2i−1 τ 2i−2 e
i=0 i=1
n

n
sin 2nϕ = b2i−2 τ 2i−1 e, sin[(2n − 1) ϕ] = b2i−1 τ 2i−1(4.124)
i=1 i=1
where the coefficients are listed in [52].

Substituting (4.124) in (4.123) and reordering the terms, completes the
proof.
As it is seen from identities (4.124), the second component in representa-
tion (4.122) is due to the odd cosine-terms and even sine-terms of the Fourier
expansion.
Note that combination (4.122) possesses algebraic properties similar to
those generated by the sawtooth time substitution. Namely, differentiation,
integration or any sufficiently smooth function of representation (4.122) gives
an element of the same two-component structure as (4.122). This is due to
the fact that none of the listed above operations destroys periodicity of the
function, and hence, identity (4.122) can be applied to the result of the op-
erations as well. Practically, results of the operations are obtained by taking
into account the trigonometric identities
τ (ϕ) = e (ϕ) , e (ϕ) = −τ (ϕ) (4.125)
and
e2 = 1 − τ 2 (4.126)
Note also that the components of representation (4.122) apparently are lin-
early independent and thus the whole combination is zero if and only if its
both components are zero.
In order to illustrate manipulations with (4.122), let us introduce the tem-

poral argument τ = sin ωt into the Duffing oscillator with no linear stiffness
term [186]
ẍ + ζ ẋ + x3 = F sin ωt (4.127)
Considering periodic solutions and taking into account (4.125) and (4.126)
on every step of the transformation, gives

1 − τ 2 X − τ X ω 2 + ζ 1 − τ 2 Y − τ Y ω (4.128)

+3 1 − τ 2 XY 2 + X 3 = F τ

1 − τ 2 Y − 3τ Y − Y ω 2 + ζX ω (4.129)

+3X 2 Y + 1 − τ 2 Y 3 = 0
The unknown functions X and Y must satisfy conditions of analytical con-
tinuation on the boundaries of the interval −1 ≤ τ ≤ 1. These conditions are
obtained by substituting τ = ±1 in equations (4.128) and (4.129) as follows

−τ X ω 2 − τ ζY ω + X 3 − F τ |τ =±1 = 0 (4.130)

− (3τ Y + Y ) ω 2 + ζX ω + 3X 2 Y |τ =±1 = 0 (4.131)
The above system does not admit a family of solutions on which Y (τ ) ≡ 0
due to the damping, therefore transformation (4.120) is not valid in this case.
Interestingly enough, the form of representation (4.122) remains the same
in the case of triangular wave, although the basic algebraic operation (4.126)
is different.
For the comparison reason, let us consider equation (4.127) with the forcing
function F τ (t/a), where τ is assumed to be the triangular wave, i.e. sawtooth
sine of the period 4a. Let us represent periodic solutions in the form (4.122),
where the new temporal argument is the sawtooth sine (4.22) of the phase
variable ϕ = t/a. Then, substituting (4.122) into equation (4.127) and con-
sidering the result as a two-component element of the algebra, one obtains
the boundary value problem
X a−2 + ζY a−1 + XY 2 + X 3 = F τ (4.132)
Y a−2 + ζX a−1 + 3Y X 2 + Y 3 = 0 (4.133)

Y |τ =±1 = 0, X |τ =±1 = 0 (4.134)
where the boundary conditions (4.134) stay for elimination of the peri-
odic series of Dirac functions from the first and second derivatives of the
coordinate.
4.4.4 General Case of Non-invertible Time and Its

Physical Meaning
Let us consider now a general class of functions {τ (ϕ) , e (ϕ)} produced by
the conservative oscillator ẍ + Π (x) = 0, where Π (x) is the potential energy
of the oscillator. In order to make the amplitude normalized to unity, let us
normalize the
potential energy and the phase variable as P (x) = Π (x) /Π (1)
and ϕ = 2Π (1)t, respectively. As a result, the differential equation of
motion and the energy integral take the form, respectively,

2
d2 x 1 dx
+ P (x) = 0 and = 1 − P (x) (4.135)
dϕ2 2 dϕ
Let x = τ (ϕ) be the system coordinate determined implicitly from the energy
integral
τ(ϕ)
ds
=ϕ (4.136)
1 − P (s)
0
Then second expression from (4.135) gives
e2 = 1 − P (τ ) (4.137)
where
1
e (ϕ) = τ (ϕ) and e (ϕ) = − P (τ ) (4.138)
2
Now let us formulate (without proof)
Proposition 5. Any periodic function x (ϕ) whose period is normalized to
1
ds
T =4
1 − P (s)
0
can be represented in the form (4.122), where the functions τ (ϕ) and e (ϕ)
are given by (4.136) and (4.138).
For example, one can transform equation (4.127) based on the potential
energy function P (x) = x2n . Then the boundary value problems (4.128)
through (4.131) and (4.132) through (4.134) can be derived as particular
cases n = 1 and n −→ ∞, respectively.
4.4.5 NSTT and Cnoidal Waves

There are other methodological sources of approximate solutions in terms of
power series with respect to different type of periodic functions. In case of
oscillators with even nonlinearities, it was suggested to approximate solutions
by power series of the elliptic Jacobi cn-function [36]

N
x(t) = A + Bn cn2n (ωt + αn , k 2 ) (4.139)
n=1
where A and Bn are constants, ω is the frequency parameter, and k is the

modulus.
The reason for using series (4.139) is that the Jacobi function effectively
captures strongly unharmonic temporal shapes of periodic motions near sep-
aratrix loops. However, mathematical properties of Jacobi functions may re-
quire certain efforts while manipulating with solutions especially when the
solutions are involved into further perturbation procedures.
Note the particular case of expression (4.139) N = 1 resembles the well
known case of cnoidal wave [184].
E(m)
x(t) = − + dn2 [2K(m)t] (4.140)
K(m)
where K(m) and E(m) are compete elliptic integrals of first kind and sec-
ond kind, respectively, and m = k 2 is the parameter (Jacobi), the period is
normalized to unity.
If the parameter m is small enough, function (4.140) takes almost harmonic
shapes, however, when m is getting larger then x(t) describes so-called cnoidal
waves as shown in Fig. 4.4.
From the physical point of view, function (4.140) with specific scaling
factors exactly describes temporal behavior of the interaction force between
particles of the Toda lattice
f = a[exp(−br) − 1], ab > 0 (4.141)
where r is the distance between adjacent particles, a and b arbitrary

parameters.
Periodic function (4.140) admits exact Fourier series expansions
∞

π2 l cos(2πlt)
x(t) = (4.142)
[K(m)]2 sinh[πlK(1 − m)/K(m)]
l=1
Interestingly enough, function (4.140) can be also represented as a sum of

localized waves (solitons), however, shifted one with respect other on the
same interval
∞

π π2
x(t) =− + cosh−2 [π(t−l)K(m)/K(1−m)]
2K(m)K(1 − m) [2K(1 − m)]2
l=−∞
(4.143)
Practical reasons for using either of series (4.142) or (4.143) is that terms
of both series consist of elementary functions4 of time, while executing
4
Actually, exponential functions with real and imaginary exponents.
Fig. 4.4 Transition from quasi-harmonic to cnoidal wave.
different principles of approximation. Namely, each term of the Fourier series

is carrying global information about the periodic process, whereas each term
of series (4.143) provides just local description in some interval near the time
point determined by the number of term, l. Although both series describe
the process exactly from the theoretical standpoint, the only limited number
of terms is possible to keep in calculations. As a result, the above mentioned
difference in principles of approximation may become essential as discussed
further in this subsection.
Now, adapting transformation (4.71) to the case T = 1 and then applying
the result to periodic function (4.143), gives
π(e+ + e− ) π2
xn,p (t) = − + (4.144)
2K(m)K(1 − m) [2K(1 − m)]2

n
p+1
×{ cosh−2 [λ(m)(τ − 4l)]e+ + cosh−2 [λ(m)(τ − 2 + 4l)]e− }
l=−n l=−p
πK(m)
λ(m) =
4K(1 − m)
Fig. 4.5 NSTT of the periodic cnoidal wave at different Jacobi parameters m:
3-three terms truncation (n = 0, p = 0), 5-five terms truncation (n = 1, p = 0),
and ∞ - exact expression.
where the infinite limits of summation have been replaced by the integers
n > 0 and p > 0; the oscillating triangular wave time, τ = τ (4t), and the
idempotent basis e+ = [1 + e(4t)]/2 and e− = [1 − e(4t)]/2 are introduced;
substitution 1 = e+ + e− has been made in order to emphasize that the
entire expression (4.144) has the form of hyperbolic number represented in
the idempotent basis.
Obviously, x∞,∞ (t) = x(t), so that, in this case, (4.144) becomes exact
equivalent to (4.143). However, convergence properties of series (4.144) are
significantly improved as compared to (4.143). Indeed, as time t is growing
in (4.143), one must switch from one term to another to keep sufficient preci-
sion of approximation so that on the infinite time interval, −∞ < t < ∞, the
entire series (4.143) is needed. In contrast, the temporal argument of series
(4.144) is always bounded by the standard interval −1 ≤ τ ≤ 1, whereas the
periodicity of wave is captured by the basis functions rather than multiple
terms of the series. As a result, terms of series (4.144) are exponentially de-
caying as the summation index increases. As follows from the diagrams in
Fig. 4.5, for the range of Jacobi parameter m > 0.2 just five terms in (4.144)
are enough to capture both quasi-harmonic and cnoidal wave shapes with a
very good precision. For greater parameters m, when the wave becomes essen-
tially cnoidal, only three terms provide quite a perfect match with the exact
shape. As seen from the upper fragment in Fig. 4.5, the three term approx-
imation gives essential discontinuous error for very small Jacobi parameter
m = 0.08. However, in this almost harmonic range, it is more effective to
use few or even one term Fourier series (4.142) rather than series (4.143).
Another essential advantage of series (4.144) is that, in the cnoidal range,
different algebraic manipulations with truncated series are essentially eased
due to the idempotent properties.
Chapter 5
Sawtooth Power Series
Abstract. In this chapter, we introduce polynomials and power series ex-

pansions with respect to the triangular sine-wave. These can be used for
approximations of periodic signals and unknown periodic solutions of dy-
namical systems. Such approximations may appear to be effective in those
cases when trigonometric series converge slowly due to step-wise discontinu-
ities or spikes. Another reason for using polynomial expansions is that they
are usually more convenient for algebraic manipulations. If the process under
consideration is smooth then sufficient class of smoothness of approxima-
tions is achieved by imposing specific constraints on the coefficients. Other
equations for the coefficients may appear either as a result of optimization
procedures, that minimize the error of approximation, or as an outcome of it-
erative procedures dictated by the differential equations of motion. It is also
shown in this chapter that using operators Lie associated with dynamical
systems essentially facilitates construction of the periodic power series.
5.1 Manipulations with the Series

5.1.1 Smoothing Procedures
Consider a smooth periodic function x (t) of the period T = 4 represented in
the form x (t) = X (τ )+Y (τ ) τ̇ , where τ = τ (t). Obviously, both components,
X and Y , admit power series expansions with respect to the argument τ
with no loss of periodicity in the original time t. However, keeping the finite
number of terms in such series may violate the smoothness conditions at τ =
±1. Indeed, functions and their truncated series may behave differently near
the boundaries τ = ±1. Therefore, let us introduce formal algorithms that
can be applied to the truncated series in order to improve their smoothness
properties. First, consider the expansion

2N
X (i) (0)
X (τ ) = τ i + O τ 2N +1 (5.1)
i=0
i!

132 5 Sawtooth Power Series
Obviously, the power terms τ i and, generally speaking, finite sums in (5.1)
are non-differentiable at the set of of points Λ = {t : τ (t) = ±1}. In contrast,
the binomials
τi τ i+2
φi (τ ) = − ; i = 1, 2, ... (5.2)
i i+2
are twice differentiable with respect to t on Λ that can be verified by taking
derivatives with respect to t. For instance, first two generalized derivatives
are
dφi (τ )
= τ i−1 (1 − τ 2 )τ̇ (t)
dt
d2 φi (τ )
= (i − 1) τ i−2 − (i + 1) τ i
dt2
where the term τ i−1 (1 − τ 2 )τ̈ has been removed from the second derivative
due to the presence factor 1 − τ 2 that takes zero value whenever τ̈ = 0.
It is seen therefore that functions (5.2) are twice continuously differentiable
with respect to t even though each of the functions is combined of two non-
differentiable terms. This is achieved by the specific choice for the coefficients
and signs in (5.2) such that the power terms have same jumps of slopes but
with opposite signs.
Now, considering (5.2) as equations with respect to different powers of τ ,
gives, respectively odd and even powers, as
τ 2N +3
τ = φ1 (τ ) + ... + φ2N +1 (τ ) +
2N + 3

τ 2N +3
τ 3 = 3 φ3 (τ ) + ... + φ2N +1 (τ ) + (5.3)
2N + 3
2N +3

τ
τ 2N +1 = (2N + 1) φ2N +1 (τ ) +
2N + 3
and

2 τ 2N +2
τ = 2 φ2 (τ ) + ... + φ2N +2 (τ ) +
2N + 2

4 τ 2N +2
τ = 4 φ4 (τ ) + ... + φ2N +2 (τ ) + (5.4)
2N + 2

2N τ 2N +3
τ = 2N φ2N (τ ) +
2N + 2
where N is an arbitrary positive integer.

Substituting (5.3) and (5.4) in (5.1) and setting N → ∞, gives
i (2k−1)
∞

X (0) X (2k) (0)
X (τ ) = X (0) + φ2i−1 (τ ) + φ2i (τ ) (5.5)
i=1 k=1
(2k − 2)! (2k − 1)!
5.1 Manipulations with the Series 133
In contrast to (5.1), particular sums of series (5.5) are twice continuously

differentiable functions of t.
Since the Y -component usually appears with the step-wise discontinuous
factor τ̇ (t), the related series must be re-organized in a somewhat different
way by taking into account the corresponding necessary condition of conti-
nuity: Y (±1) = 0. In this case, appropriate polynomials are designed as
ψi (τ ) = τ i − τ i+2 ; i = 0, 1, 2, ... (5.6)
These polynomials provide continuity for the term ψi [τ (t)]τ̇ (t) as well as its
first derivative
d[ψi (τ )τ̇ ]
= ψi (τ ) τ̇ 2 = iτ i−1 − (i + 2) τ i+1
dt
where the term ψi (τ ) τ̈ has been eliminated due to the boundary conditions
ψi (±1) = 0.
However, second derivative appears to be a step-wise discontinuous func-
tion at every time t ∈ Λ,
d2 [ψi (τ )τ̇ ]
2
= i (i − 1) τ i−2 − (i + 2) (i + 1) τ i τ̇
dt
Therefore, particular sums of the re-built series
i (2k−1)
∞

Y (0) Y (2k) (0)
Y (τ ) τ̇ = ψ2i−1 (τ ) + ψ2i (τ ) τ̇ (5.7)
i=0 k=0
(2k − 1)! (2k)!
are less smooth than those obtained for the X-component (5.5).
Combining (5.5) and (5.7) and considering an arbitrary period, T = 4a,
gives
∞

∞

τi τ i+2
x (t) = X (0) + KX (i) − +e KY (i) τ i − τ i+2 (5.8)
i=1
i i+2 i=0
where τ =τ (t/a) and e =τ (t/a), and the coefficients are calculated as follows

i
X (2k−s) (0)
KX (2i − s) =
(2k − 1 − s)!
k=1

i
Y (2k−s) (0)
KY (2i − s) = (5.9)
(2k − s)!
k=0
(s = 0, 1)
Example 9

ln 3 π 2 τ2 τ4 π2 π4 τ4 τ6
ln (2 + cos t) = + − + − − + ...
2 12 2 4 12 48 4 6

ln 3 ln 3 π 2 2
+ 1 − τ2 + − τ − τ4
2 2 12

ln 3 π 2 π4 4
+ − + τ − τ 6 + ... e
2 12 192
where τ = τ (2t/π) and e = τ (2t/π).
Series (5.8) may represent periodic solutions of dynamical systems. Formal

solutions are obtained by taking into account the differential equations of
motion when calculating the derivatives in coefficients (5.9). It can be done
by means of the operator Lie as discussed in the next section.
Finally, the above smoothness procedures can be repeated as many times
as needed until necessary smoothness of the particular sums is achieved. For
instance, the expressions (5.2) and (5.6) reveal next steps of the smoothing
as
τi τ i+2
φi (τ ) = − ∈ C2
(τ i ) |τ =1 (τ i+2 ) |τ =1
φi (τ ) φi+2 (τ )
ϕi (τ ) = (3)
− (3)
∈ C4 (5.10)
φi (τ ) |τ =1 φi+2 (τ ) |τ =1
and,

ψi (τ ) τ̇ = τ i − τ i+2 τ̇ ∈ C 1

ψi (τ ) ψi+2 (τ )
χi (τ ) τ̇ = (2)
− (2)
τ̇ ∈ C 3 (5.11)
ψi (τ ) |τ =1 ψi+2 (τ ) |τ =1
respectively, where the symbol C indicates the class of smoothness in the

interval −∞ < t < ∞.
Second sets of equations in (5.10) and (5.11) have to be inverted for φi (τ )
and ψi (τ ) in a similar to (5.3) and (5.4) way. Then, by using the correspond-
ing expressions, one can introduce functions ϕi (τ ) and χi (τ ) into series (5.5)
and (5.7), respectively.
5.2 Sawtooth Series for Normal Modes 135
5.2 Sawtooth Series for Normal Modes
5.2.1 Periodic Version of Lie Series

Lie series with respect to the physical time parameter are considered below.
Note that the corresponding procedure essentially differs of those used for
asymptotic integration of the differential equations of motion, where Lie series
associate with a small parameter of perturbation [62] , [204].
Let us consider the differential equation of motion with respect to the
position vector-function x (t) ∈ Rn
ẍ + f (x, ẋ, t) = 0 (5.12)
under the initial conditions
x0 = x|t=0 and ẋ0 = ẋ|t=0 = v0
where vector-function f is assumed to be differentiable with respect to every

argument as many times as needed.
The standard Cauchy form of dynamical system (5.12) with respect to the
coordinate and velocity vectors is
ẋ = v
v̇ = −f (x, v, t) (5.13)
ṫ = 1
Then, the dynamics of system (5.13) can be locally described by the Lie series
[70]

1
x = exp[(t − t0 )G]x0 ≡ 1 + (t − t0 )G + (t − t0 )2 G2 + · · · x0 (5.14)
2!
∂ ∂ ∂
G = v0 · − f (x0 , v0 , t0 ) · + (5.15)
∂x0 ∂v0 ∂t0
where G is Lie operator associated with system (5.13), and {x0 , v0 , t0 } is
some initial point in the system’ phase space.
The dot between two quantities indicates dot products. For example,
∂ ∂ ∂
v0 · ≡ v01 + · · · + v0n
∂x0 ∂x01 ∂x0n
Series (5.14) is simply Taylor series whose coefficients are calculated by en-
forcing equations (5.13). Unfortunately, this general idea is still of little use for
oscillatory processes probably due to locality of expansion (5.14). In other
words, even entire expansion (5.14) does not explicitly reveal such global
characteristics of oscillations as their amplitude and period. Moreover, the
corresponding truncated series produce increasingly growing errors as the

time t runs away from the selected initial point t0 . In order to overcome
these disadvantages, it is suggested to adapt the Lie series solution for the
class of periodic motions as follows.
Theorem 1. [145] Assume that system (5.13) admits a periodic solution x(t)
of the period T = 4a so that x(t + 4a) = x(t) for any t, and some point
{x0 , v0 , t0 } belongs to this solution. Then such a solution can be expressed in
the form
x = exp(aG){cosh[a(τ − 1)G] + e sinh[a(τ − 1)G]}x0 (5.16)
where τ and e are saw-tooth sine and rectangular cosine, whose periods are
normalized to four and amplitudes are normalized to unity as
τ (ϕ) = (2/π) arcsin sin(πϕ/2) (5.17)
and,
e(ϕ) = sgn[ cos(πϕ/2)] (5.18)
respectively, and ϕ = (t−t0 )/a is a re-scaled time. If, in addition, the solution
is odd with respect to one half of the period, x(t+2a) = −x(t), then expression
(5.16) simplifies to
x = [sinh(aτ G) + e cosh(aτ G)]x0

1 1
≡ aτ G + (aτ G)3 + · · · x0 + e 1 + (aτ G)2 + · · · x0 (5.19)
3! 2!
Proof of expression (5.16) is obtained by substituting the identity [144]
ϕ = 1 + [τ (ϕ) − 1]e(ϕ), (−1 < ϕ < 3) (5.20)
in (5.14) and taking into account that
e2 = 1 (5.21)
at almost every time instance1 . In order to prove the particular case (5.19),
one should keep in mind that exp(2aG)x0 = x(t0 + 2a) = −x0 , as it follows
from (5.14), and the oddness condition assumed.
Note that τ and e are indeed quite simple piece-wise linear functions; the
above analytical expressions (5.17) and (5.18) just define them in the unit-
form which enables one to avoid conditioning of computation in the original
temporal scale, t0 ≤ t < ∞. This possibility becomes essential when the
dynamics includes some evolutionary component.
1
The set of isolated points {ϕ : τ (ϕ) = ±1}appears to have no effect on the results
[144].
5.2 Sawtooth Series for Normal Modes 137
Physical meaning of relationship (5.20) is that, during the whole period,

the time variable ϕ is expressed through the coordinate τ and velocity e of
a classic particle freely oscillating between the two absolutely stiff barriers
with no energy loss. Due to (5.21), this relationship possesses the algebraic
structure of ‘hyperbolic complex numbers’ as revealed by (5.16).
Let us outline possible applications of expressions (5.16) and (5.19). For
the sake of simplicity, consider the particular case (5.19). Of course, formal
expression (5.19) does not guarantee the existence of periodic solutions. In
case some periodic solution does exist, one should be able to find the corre-
sponding vectors x0 and v0 from appropriate conditions. In autonomous case,
the scalar parameter, a, is also unknown and must be determined.
The related conditions are formulated as a requirement of smoothness of
expression (5.19), which is generally non-smooth or even discontinuous due
to the presence of non-smooth and discontinuous functions τ and e, respec-
tively. The ‘smoothing’ relations are obtained by eliminating the step-wise
discontinuities of the coordinate and velocity vectors imposing the constraints
cosh(aG)x0 = 0
cosh(aG)v0 = 0 (5.22)
In autonomous case, algebraic equations (5.22) represent a nonlinear eigen-

value problem, where a is an eigen-value, and {x0 , v0 } is a combined (state)
eigen-vector.
By narrowing the class of periodic motions to those on which the system
passes its trajectory twice in the configurations space during the same pe-
riod, one obtains a subclass of normal mode motions. For more physically
meaningful definitions and discussions, see reference [190]. Let us formulate
the corresponding problem based on the periodic Lie series solutions.
Consider the vibrating system
ẍ + f (x) = 0, x ∈ Rn (5.23)
where f (−x) = −f (x), and the initial conditions are x|t=0 = x0 = 0 and
ẋ|t=0 = v0 .
The normal mode solutions of system (5.23) are obtained as a particular
case of (5.19) and (5.22)
x = sinh(aτ G)x0 |x0 =0 (5.24)
cosh(aG)v0 |x0 =0 (5.25)

where the initial vector x0 = 0 is substituted into the expressions only after
all degrees of the differential operator
∂ ∂
G = v0 · − f (x0 ) ·
∂x0 ∂v0
have been applied.
Relationship (5.24) can be interpreted as a parametric equation of normal
mode trajectories of the system with the parameter interval −1 ≤ τ ≤ 1.
Let us illustrate relationships (5.24) and (5.25) based on the linear sys-
tem so that the result could be compared with the well known conventional
solution.
Example 10. Suppose that f (x) = Kx, where K is positively defined sym-
metric n × n-matrix with eigen-system {v0 , ω 2 } so that Kv0 = ω 2 v0 . In
this case, by applying the operator G twice, one obtains that v0 is also an
eigen-vector of the operator G2 , namely, G2 v0 = −ω 2 v0 . Then, keeping in
mind the power series form of expressions (5.24) and (5.25) as those in (5.19)
and sequentially applying the operator G2 , gives x = (v0 /ω) sin(aωτ ) and
cos(aω) = 0, respectively. Notably, the last equation shows that there ex-
ist an infinite number of roots {a} related to the same eigen-frequency ω!
However, it is easily to find that all the roots produce the same solution in
terms of the original time t. The minimal quarter of the period is a = π/(2ω),
therefore x = (v0 /ω) sin(πτ /2), and τ = (2/π) arcsin sin ωt.
Nonlinear cases and the related problems dealing with truncated expansions
of (5.25) will be further discussed in a full-length paper.
5.3 Lie Series of Transformed Systems
5.3.1 Second-Order Non-autonomous Systems

In the previous section, the sawtooth temporal argument was introduced
into Lie series solutions with respect to time t. Alternatively, the sawtooth
argument can be introduced first into the differential equations of motion
before the Lie series procedure is applied.
Let the dynamical system be described by the set of second-order equations
(5.12), where the vector-function f is periodic with respect to t with the
period T = 4a. Periodic motions of the period T are considered. Note that,
in the autonomous case, the period is a priory unknown and thus must be
determined.
Following the rules introduced in Chapter 4 (see also Chapters 8 and 12
for further details) and making the substitutions t → τ (t/a) and x (t) =
X(τ ) + Y (τ )e in (5.12), gives the boundary value problem
X + a2 Rf (X, Y, X , Y , τ ) = 0 (5.26)
Y + a2 If (X, Y, X , Y , τ ) = 0
5.3 Lie Series of Transformed Systems 139
Y |τ =±1 = 0, X |τ =±1 = 0 (5.27)

where

Rf 1
= {f [X + Y, (Y + X ) /a, aτ ] ± f [X − Y, (Y − X ) /a, 2a − aτ ]}
If 2
(5.28)
Analogously to (5.15), the operator Lie of system (5.26) at some initial point
is represented as
∂ ∂ ∂
G = X0 · + Y0 · − a2 Rf (X0 , Y0 , X0 , Y0 , τ0 ) ·
∂X0 ∂Y0 ∂X0
∂ ∂
−a2 If (X0 , Y0 , X0 , Y0 , τ0 ) · +
∂Y0 ∂τ0
The idea of Lie series enables one of representing solution of the system (5.26)
in the power series form with respect to the sawtooth argument τ under the
initial conditions at P0 (X0 , Y0 , X0 , Y0 , τ0 ); it is assumed that τ0 = 0. Then,
the corresponding solution of the original equation is represented in the form
x (t) = X(X0 , X0 , Y0 , Y0 , τ ) + Y (X0 , X0 , Y0 , Y0 , τ )e (5.29)
where τ = τ (t/a) and e = τ (t/a).

Now, the arbitrary quantities X0 , X0 , Y0 , and Y0 can be determined from
the smoothness conditions (5.27)
∂X(X0 , X0 , Y0 , Y0 , τ )

Y (X0 , X0 , Y0 , Y0 , τ )|τ =±1 = 0, |τ =±1 = 0 (5.30)
∂τ
However, instead of solving equations (5.30), the smoothing procedure de-
scribed in the previous section can be applied. In this case, conditions (5.30)
are satisfied automatically, but the initial quantities X0 , X0 , Y0 , and Y0 re-
main undetermined. This enables one of improving the solution’ smoothness
by imposing stronger smoothness conditions as follows
∂ 2 Y (X0 , X0 , Y0 , Y0 , τ ) ∂ 3 X(X0 , X0 , Y0 , Y0 , τ )

|τ =±1 = 0, |τ =±1 = 0
∂τ 2 ∂τ 3
(5.31)
Equations (5.30) or (5.31) for X0 , X0 , Y0 and Y0 are generally quite compli-
cated. In different particular cases however, the problem can be simplified by
taking into account possible symmetries of the system.
So let us consider equation
ẍ + f (x, t) = 0 (5.32)
where the vector-function f (x, t) is odd with respect to the positional vector
x and even with respect to the quarter of the period t = a. In this case,
boundary-value problem (5.26) and (5.27) reduces to
X + a2 f (X, aτ ) = 0
X |τ =1 = 0 (5.33)
Y ≡0
The operator Lie and the corresponding solution are, respectively,

∂ ∂ ∂
G = X − a2 f (X, aτ ) +
∂X ∂X ∂τ
and

∞
i
G2k−1 X|τ =0 τ 2i−1 τ 2i+1
x (t) = − ≡ X(X0 , τ )
(2k − 2)! 2i − 1 2i + 1
i=1 k=1
where τ = τ (t/a), the initial position is X0 = 0, whereas the constant vector

X0 is determined from equation
∂ 3 X(X0 , τ )
|τ =1 = 0
∂τ 3
Example 11. Consider now a two degree of freedom oscillating system

d2 x1 ζ (2x1 − x2 ) + x31 − P sin Ωt 0
+ =
dt2 x2 ζ (2x2 − x1 ) + x32 0
In this case, the parameter a = π/(2Ω) is the quarter of the period of the
external forcing function. The operator Lie of the system is
∂ ∂ πτ ∂
G = X1 + X2 − a2 ζ (2X1 − X2 ) + X13 − P sin
∂X1 ∂X2 2 ∂X1
∂ ∂
−a2 [ζ (2X2 − X1 ) + X23 ] +
∂X2 ∂τ
Let the initial position be X0 = 0. Then, introducing the notation X0 =

T
[A1 , A2 ] and calculating the coefficients of the first three terms of series
(5.8), gives

3
τ3 τ τ5 1
X1 = τ − A1 + − A1 + (a2 P π − 4 a2 ζ A1 + 2a2 ζ A2 )
3 3 5 4

5 7

τ τ 1
+ − A1 + (a2 P π − 4 a2 ζ A1 + 2a2 ζ A2 )
5 7 4

1 2
− [a P π 3 + 8a4 P π ζ − 8A1 5 a4 ζ 2 − 6 a2 A1 2 + 32 a4 ζ 2 A2 ]
192
and

3
τ3 τ τ5 1
X2 = τ − A2 + − A2 + (a2 ζ A1 − 2 a2 ζ A2 )
3 3 5 2

τ5 τ7 1
+ − A2 + (a2 ζ A1 − 2 a2 ζ A2 )
5 7 2

1 4
+ [a P π ζ − 8 a4 ζ 2 A1 + 2A2 5 a4 ζ 2 − 6 a2 A2 2 ]
48
In this case, equations (5.31) give two scalar equations for A1 and A2 ,
∂ 3 X1 (A1 , A2 , τ ) ∂ 3 X2 (A1 , A2 , τ )
|τ =1 = 0, |τ =1 = 0 (5.34)
∂τ 3 ∂τ 3
where the first and the second equations guarantee smoothness of the ac-
celeration components ẍ1 (t) and ẍ2 (t), respectively. Intersections of curves
(5.34) on the Cartesian plane A1 A2 allow to locate the roots, as it is
shown√in Figs. 5.1 and 5.2. For instance, setting ζ = 1.0, P = 0.2 and
Ω = 1.01, gives (A1 , A2 ) = (0.7711, 0.6677). This solution corresponds to
a weakly nonlinear perturbation of the in-phase √ linear mode. Similarly, in
the neighborhood of the second frequency Ω = 3.01, one finds the solution
(A1 , A2 ) = (0.5971, −0.8181), which is close to the out-of-phase linear mode;
see Fig. 5.1. Interestingly enough, Fig. 5.2 shows that there are also two
roots, (2.3315, 2.2864) and (−2.2065, −2.2548), corresponding to in-phase vi-
brations even though the frequency of the external forcing function is close
to that of the out-phase linear mode. This effect may take place due to large
amplitudes such that the in-phase nonlinear frequency becomes close to the
out-of-phase linear frequency.
Since the initial conditions, corresponding to the periodic regimes are known,
one can integrate the differential equations of motion numerically in order to
check the analytical solutions. In the latter example, one would obtain that
both results are perfectly matching. It must be noted, however, that some of
the periodic solutions may appear to be unstable. As a result, even a very
small imperfection in the initial conditions will lead to a significant divergence
of the results. Moreover, the ‘tails’ of polynomial expansions may give roots
which do not correspond to any solution. Nevertheless, the above approaches
seems to make sense as compared to those based on the direct Fourier expan-
sions. Indeed, the number of algebraic equations in (5.22), (5.30) or (5.31) is
independent on the number of terms in the series.
5.3.2 NSTT of Lagrangian and Hamiltonian

Equations
In different theoretical areas, describing systems in terms of the analytical
dynamics brings some advantages because, until certain stage, it is sufficient
to deal with a single scalar function such as Lagrangian or Hamiltonian rather
then manipulate with a set of differential equations.
Suppose the sawtooth temporal argument is introduced into Lagrangian
or Hamiltonian. Let us show that the corresponding differential equations of
Fig. 5.1 The curves of smoothness for√ the accelerations ẍ1 (t) and ẍ2 (t), respec-
tively thin and solid lines, when Ω = 1.01.
Fig. 5.2 The curves of smoothness for√ the accelerations ẍ1 (t) and ẍ2 (t), respec-
tively thin and solid lines, when Ω = 3.01.
motion are still derivable from the transformed functions in the standard way
of the analytical dynamics.
Consider first Lagrangian that depends on time t periodically with the
period T = 4a,
L = L(x, ẋ, t) (5.35)
On the manifold of smooth periodic motions of the same period T , the vector-
function x(t) is represented as x (t) = X (τ ) + Y (τ ) e, where τ = τ (t/a) and
e = e(t/a). As a result, Lagrangian (5.35) takes the form
L(x, ẋ, t) = RL (X, Y, X , Y , τ ) + IL (X, Y, X , Y , τ ) e (5.36)
where both components on the right-hand side are determined analogously

to expressions (5.28).
It can be verified by inspection that the differential equations of periodic
motions can be represented now in any of the two equivalent forms
d ∂RL ∂RL

− =0
dτ ∂X ∂X
d ∂RL ∂RL

− =0
dτ ∂Y ∂Y
or
d ∂IL ∂IL
− =0
dτ ∂Y ∂Y
d ∂IL ∂IL

− =0
dτ ∂X ∂X
Now let us consider the Hamiltonian
H = H(p, q, t) (5.37)
which may be periodic with respect to time with the period T = 4a.
On the manifold of periodic motions of the period T , the coordinates and
the linear momenta are represented as
q = X (τ ) + Y (τ ) e and p = U (τ ) + V (τ ) e
respectively, where τ = τ (t/a) and e = e (t/a).

Let us transform the Hamiltonian (5.37) as
H −→ aH(p, q, t) = RaH (U, V, X, Y, τ ) + IaH (U, V, X, Y, τ ) e
where

RaH a
= [H (U + V, X + Y, aτ ) ± H (U − V, X − Y, 2a − aτ )]
IaH 2
The corresponding differential equations of periodic motions are

∂RaH ∂RaH
X = , U = −
∂U ∂X
∂R ∂R
Y = , V =−
aH aH
∂V ∂Y
or
∂IaH ∂IaH
X = , V =−
∂V ∂X
∂IaH ∂IaH
Y = , U =−
∂U ∂Y
Besides, in autonomous cases, the operator Lie associates with the Poisson
bracket, for instance,
GX = {X, RaH }
Therefore, introducing the sawtooth temporal argument for periodic motions
preserves both Lagrangian and Hamiltonian structures of the differential
equations of motion.
5.3.3 Remark on Multiple Argument Cases

In multiple frequency cases, the smoothing procedures can be applied sequen-
tially to each of the arguments. For instance, in the case of two arguments,
τ1 = τ (t/a1 ) and τ2 = τ (t/a2 ) , the X-component would take the form

N1 N2
i j
τ1 τ1i+2 τ2 τ2j+2
X (τ1 , τ2 ) = X (0, 0) + KX (i, j) − − (5.38)
i=1 i=1
i i+2 j j+2
where KX (i, j) are constant coefficients.

Analytical manipulations with such 2D polynomials as (5.38) are quite
complicated because algebraic and differential operations affect the structure
of binomials. However, let us bring attention to a possibility of using polyno-
mial expansions (5.38), and the related complete versions, for different kinds
of approximations by determining coefficients of expansions through appro-
priate numerical optimization procedures. Besides of multiple phase cases
in dynamics, similar expansions may be useful for multidimensional spatial
problems dealing, for instance, with elastic cell-wise periodic structures; see
also Chapter 14 for possible formulations. In such cases, the triangular sine
waves depend on the spatial coordinates, say x and y as τ1 = τ (x/a1 ) and
τ2 = τ (y/a2 ). In static problems, the parameters a1 and a2 are usually given
by design, whereas the coefficients can be obtained through variational prin-
ciples by minimizing the corresponding functionals.
Chapter 6
NSTT for Linear and Piecewise-Linear
Systems
Abstract. Remind that the tool of nonsmooth argument substitutions was

introduced first to describe strongly nonlinear vibrations whose temporal
mode shapes are asymptotically close to non-smooth ones. Such cases are
known to be most difficult for analyses because different quasi-harmonic
methods are already ineffective whereas nonsmooth mapping tools are still
inapplicable. It is quite clear however that the non-smooth arguments can
be introduced regardless the strength of nonlinearity or the form of dynam-
ical systems in general. For instance, it is shown in this chapter that the
non-smooth substitutions can essentially simplify analyses of different linear
models with non-smooth or discontinuous inputs. It is also shown that, in
piecewise-linear cases, the nonsmooth temporal transformation provides an
automatic matching the motions from different subspaces of constant stiff-
ness and justifies quasi-linear asymptotic solutions for the specific nonsmooth
case of piece-wise linear characteristics.
6.1 Free Harmonic Oscillator: Temporal Quantization

of Solutions
Introducing the sawtooth temporal argument into the differential equations
of motion may bring some specific features into the corresponding solutions.
For illustrating purposes, let us consider the harmonic oscillator
ẍ + ω02 x = 0 (6.1)
First, let us obtain exact general solution of the oscillator (6.1) in terms of
the sawtooth temporal argument by using the substitution
x = X (τ ) + Y (τ ) e (6.2)

146 6 NSTT for Linear and Piecewise-Linear Systems
where τ = τ (t/a) and e = e (t/a) are the standard triangular and rectangular
wave functions, respectively.
Substituting (6.2) in (6.1), gives the boundary value problem
a−2 X (τ ) + ω02 X (τ ) = 0 (6.3)

a−2 Y (τ ) + ω02 Y (τ ) = 0 (6.4)
X (±1) = 0, Y (±1) = 0 (6.5)

By considering the parameter a as an eigen value of the problem, one obtains
the set of eigen values and the corresponding solutions as, respectively,
jπ
aj = (6.6)
2ω0
and

jπτ jπτ
Xj = sin + ϕj , Yj = cos − ϕj (6.7)
2 2
where ϕj = (π/4) [1 + (−1)j ] , τ = τ (t/aj ), and j is any positive real integer.
Therefore, introducing the sawtooth oscillating time produced the discrete
family of solutions for harmonic oscillator (6.1).
The nature of such kind of quantization is due to the specific temporal
symmetry of periodic motions. In other words, the quantization is associated
with a multiple choice for the period
Tj = 4aj = jT (6.8)
where T = 2π/ω0 is the natural period of oscillator (6.1).

In terms of the original temporal variable t, the number j plays no role for
the temporal mode shape, given by

jπ 2ω0 t
x (t) = A sin τ + ϕj (6.9)
2 jπ

jπ 2ω0 t 2ω0 t
+B cos τ − ϕj e
2 jπ jπ
where A and B are arbitrary constants, and x(t) is the same harmonic wave
regardless the number j.
In this section, the free linear oscillator was considered for illustrating
purposes. Of course, there is no other pragmatic reason for introducing the
sawtooth time into equation (6.1). The situation drastically changes however
in non-autonomous cases of non-smooth or discontinuous inputs. It is shown
below that, in such cases, the sawtooth time variable can help to facilitate
determining particular solutions. The effect of ‘temporal quantization’ repre-
sented by expression (6.9), which seems to be just identical transformation
in the autonomous case, acquires helpful meaning at the presence of external
6.2 Non-autonomous Case 147
excitations. For instance, according to (6.9) the so-called combination reso-

nances will appear to be an inherent property of oscillators.
6.2 Non-autonomous Case
6.2.1 Standard Basis

Consider the linear harmonic oscillator under the external forcing described
by the linear combination of triangular and rectangular wave functions

2 t t
ẍ + ω0 x = F τ + Ge (6.10)
a a
where F and G are constant amplitudes, and a is a quarter of the period.

Substituting (6.2) in (6.10), leads to the boundary value problem
a−2 X (τ ) + ω02 X (τ ) = F τ (6.11)

a−2 Y (τ ) + ω02 Y (τ ) = G (6.12)
under the boundary conditions (6.5).

In contrast to autonomous case (6.1), the parameter a is known. However,
the equations (6.11) and (6.12) are non-homogeneous, and thus a non-zero
solution exists for any a and can be found in few elementary steps. As a
result, the particular periodic solution of the original equation (6.10) takes
the form

F t sin [aω0 τ (t/a)]
xp (t) = X (τ ) + Y (τ ) e = 2 τ −
ω0 a aω0 cos aω0

G cos [aω0 τ (t/a)] t
+ 2 1− e (6.13)
ω0 cos aω0 a
The corresponding general solution is x (t) = A cos (ω0 t − ϕ) + xp (t), where

A and ϕ are arbitrary amplitude and phase parameters. Note that solu-
tion (6.13) immediately shows all possible resonance combinations aω0 =
(2k + 1) π/2 or
ω0
= 2k + 1 (6.14)
Ω
where k = 1, 2, 3... , and Ω = 2π/T = π/(2a) is the principal circular fre-
quency of the external forcing.
It is interesting to compare the solution (6.13) with those obtained by
the conventional methods such as Fourier series. So, taking into account
expansion,

∞
t 8 (−1)k (2k + 1) πt
τ = 2 2 sin (6.15)
a π (2k + 1) 2a
k=0
gives the particular solution of equation (6.10) in the form

∞
1
xp (t) = 2 × (6.16)
(2k+1)π
k=0 ω02 − 2a

k k
8F (−1) (2k + 1) πt 4G (−1) (2k + 1) πt
× 2 sin + cos
2
π (2k + 1) 2a π (2k + 1) 2a
Solution (6.16) indicates the same resonance conditions, (6.14). However,

infinite trigonometric series are less convenient for calculations, especially
when dealing with derivatives of the solutions; indeed, differentiation slows
down convergence of series (6.16).
6.2.2 Idempotent Basis

Consider the linear oscillator including viscous damping under the rectangu-
lar wave external loading

t
ẍ + 2ζω0 ẋ + ω02 x = pe (6.17)
a
The purpose is to obtain periodic (particular) solution with the period of

external loading T = 4a. Recall that the idempotent basis is introduced by
means of the linear transformation (see Chapters 1 and 4)
1
{1, e} −→ {e+ , e− } : e± = (1 ± e) (6.18)
2
or, inversely, 1 = e+ + e− and e = e+ − e− , where e2± = e± and e+ e− = 0.
Now, the periodic solution and external loading are represented in the form
x(t) = U (τ )e+ + V (τ )e− (6.19)

pe = p(e+ − e− )
where e± = e± (t/a), and U (τ ) and V (τ ) are unknown functions of the tri-

angular wave τ = τ (t/a).
Substituting (6.19) in (6.17), and sequentially eliminating derivatives of
the rectangular wave e(t/a) as described in Chapter 4, gives equations
U + 2ζωaU + (ωa)2 U = pa2

V − 2ζωaV + (ωa)2 V = −pa2 (6.20)
and boundary conditions
(U − V )|τ =±1 = 0
(U + V )|τ =±1 = 0 (6.21)
6.3 Systems under Periodic Pulsed Excitation 149
All the coefficients and right-hand sides of both equations in (6.20) are con-
stant, and the equations are decoupled. As a result, solution of boundary
value problem (6.20) and (6.21) is easily obtained in the form
p 2p exp(−ατ )
U (τ ) = 2
− 2
(6.22)
ω βω (cos 2β + cosh 2α)
×[cos β cosh α(β cos βτ + α sin βτ ) + sin β sinh α(α cos βτ − β sin βτ )]
p 2p exp(ατ )
V (τ ) = − + (6.23)
ω2 βω 2 (cos 2β + cosh 2α)
×[cos β cosh α(β cos βτ − α sin βτ ) + sin β sinh α(α cos βτ + β sin βτ )]

where α = ωaζ and β = ωa 1 − ζ 2 .
Substituting (6.22) and (6.23) in (6.19), gives closed form particular so-
lution of original equation (6.17). Transition to the original temporal vari-
able is given by the functions τ (ϕ) = (2/π) arcsin[sin(πt/2)] and e(ϕ) =
sgn[cos(πt/2)]. Since the system under consideration is linear, the general
solution of equation (6.17) can be obtained by adding general equation of the
corresponding equation with zero right-hand side. Finally, note that neither
trigonometric expansions nor any integral transforms were involved into the
solution procedure.
6.3 Systems under Periodic Pulsed Excitation

Instantaneous impulses acting on a mechanical system can be modeled ei-
ther by imposing specific matching conditions on the system state vector at
pulse times or by introducing Dirac’s functions into the differential equa-
tions of motion. The first approach deals with the differential equations of a
free system separately between the impulses, therefore a sequence of systems
under the matching conditions are considered. The second method gives a
single set of equations over the whole time interval without any conditions
of matching. In this case however the analysis can be carried out correctly in
terms of distributions, which unfortunately requires additional mathematical
justifications in non-linear cases. Both of the above approaches are actually
employed for different quantitative and qualitative analyses. The analytical
tool, which is described below, on the one hand, eliminates the singular terms
from the equations and, on the other hand, brings solutions to the unit-form
of a single analytic expression for the whole time interval.
6.3.1 Regular Periodic Impulses

Introducing the sawtooth temporal argument may significantly simplify so-
lutions whenever loading functions are combined of the triangular wave and
its derivatives. For instance, let us seek a particular solution of the first order
differential equation
∞

v̇ + λv = μ [δ (t + 1 − 4k) − δ (t − 1 − 4k)] (6.24)
k=−∞
where λ and μ are constant parameters.

For positive λ, equation (6.24) describes the velocity of a particle moving
in a viscous media under the periodic impulsive force. The corresponding
physical model is shown in Fig. 6.1, where the freely moving massive tank
experiences perfectly elastic reflections from the stiff obstacles. By scaling the
variables, one can bring the differential equation of motion of the particle to
the form (6.24), where v (t) = ẋ (t).
Fig. 6.1 If the particle’ mass is very small compared to the total mass of the
tank then the inertia force applied to the particle inside the tank has the periodic
pulse-wise character.
First, note that the right-hand side of equation (6.24) can be expressed
through the generalized derivative of the rectangular wave function as follows
μ
v̇ + λv = ė (t) (6.25)
2
Now let us represent the particular solution in the form
v (t) = X (τ (t)) + Y (τ (t)) e (t) (6.26)

μ
Y + λX + (X + λY ) e (t) + Y − ė (t) = 0 (6.27)
2
Apparently, the elements {1, e} and ė in combination (6.27) are linearly

independent as functions of different classes of smoothness. Therefore,
μ
Y + λX = 0, X + λY = 0, Y |τ =±1 = (6.28)
2
In contrast to equation (6.24) or (6.25), boundary value problem (6.28) in-
cludes no discontinuities whereas the new independent variable belongs to
the standard interval, −1 ≤ τ ≤ 1.
Solving the boundary value problem (6.28) and taking into account sub-
stitution (6.26), gives periodic solution of equation (6.24) in the form
μ
v = X +Ye = (− sinh λτ + e cosh λτ )
2 cosh λ
or
μ
v= exp [−λτ (t) e (t)] e (t) (6.29)
2 cosh λ
Fig. 6.2 illustrates solution (6.29) for μ = 0.2 and different magnitudes of λ.
0.2
Λ 0.4
0.1
v 0.0
Λ 1.9
0.1
0.2
0 2 4 6 8
t
Fig. 6.2 The family of discontinuos periodic solutions.
Note that the discontinuous solution v (t) is described by the unit-form

expression (6.29) through the two elementary functions τ (t) and e (t).
6.3.2 Harmonic Oscillator under the Periodic

Impulsive Loading
Let us consider the harmonic oscillator subjected to periodic pulses
∞

ẍ + ω02 x = 2p [δ (ωt + 1 − 4k) − δ (ωt − 1 − 4k)] (6.30)
k=−∞
where p, ω0 and ω are constant parameters.

The right-hand side of equation (6.30) can be expressed through first

derivative of the rectangular wave as follows
de (ωt)
ẍ + ω02 x = p (6.31)
d (ωt)
Let us seek a periodic solution of the period T = 4/ω in the form
x (t) = X (τ (ωt)) + Y (τ (ωt)) e (ωt) (6.32)
Substituting (6.32) in (6.31) under the necessary condition of continuity for

x (t), gives
de (ωt)
ω 2 X + ω02 X + ω 2 Y + ω02 Y e + ω 2 X − p =0 (6.33)
d (ωt)
Analogously to the previous subsection, equation (6.33) gives the boundary

value problem
ω 2 ω 2
X + Y +
0 0
X = 0, Y =0 (6.34)
ω ω
p
X |τ =±1 = 2, Y |τ =±1 = 0
ω
Solving boundary value problem (6.34) and taking into account (6.32), gives
the periodic solution of the original equation (6.30) in the form
p sin [(ω0 /ω) τ (ωt)]

x = X (τ (ωt)) = (6.35)
ωω0 cos (ω0 /ω)
where Y ≡ 0.
Solution (6.35) is continuous, but nonsmooth at those times t where
τ (ωt) = ±1. All possible resonances are given by
2 ω0
ω= ; k = 1, 3, 5, ... (6.36)
π k
where the factor 2/π is due to different normalization of the periods for
trigonometric and sawtooth sines.
Now let us consider the case of viscous damping described by the differen-
tial equation of motion
de (ωt)
ẍ + 2ζ ẋ + ω02 x = p (6.37)
d (ωt)
where ζ is the damping factor.

In this case, the boundary value problem becomes coupled

ζ ω 2
X + 2 Y +
0
X=0 (6.38)
ω ω
ζ ω0 2
Y + 2 X + Y =0
ω ω
p
X |τ =±1 =
, Y |τ =±1 = 0
ω2
As a result, the periodic solution has both X and Y components
p
x = X +Ye=
cos β cosh α + sin2 β sinh2 α
βω 22 2
×[cosh α cos β cosh ατ sin βτ − sinh α sin β sinh ατ cos βτ (6.39)

+ (sinh α cos βτ cosh ατ sin β − sinh ατ sin βτ cosh α cos β) e]

where τ = τ (ωt), e = e (ωt); α = ζ/ω and β = ω02 − ζ 2 ω.
0.2
0.1
x 0.0
0.1
0.2
0 2 4 6 8
t
Fig. 6.3 Response of the damped harmonic oscillator under the periodic impulsive
excitation for p = 0.1, ζ = 0.5, ω0 = 4 and ω = 0.2 (low-frequency pulses.)
Figs. 6.3 through 6.5 illustrate qualitatively different responses of the sys-
tem when varying the input frequency. In different proportions, the responses
combine properties of the harmonic damped motion and the non-smooth mo-
tion due to the impulsive loading. For instance, when ω >> ω0 and ω >> ζ,
the system is near the limit of a free particle under the periodic impulsive
force. In this case, the boundary value problem is reduced to
p
X = 0, Y = 0; X |τ =±1 = , Y |τ =±1 = 0 (6.40)
ω2
0.06
0.04
0.02
x 0.00
0.02
0.04
0.06
0 2 4 6 8
t
Fig. 6.4 System response on ‘resonance’ pulses ω = (2/π)ω0 = 2.5465.
0.003
0.002
0.001
x 0.000
0.001
0.002
0.003
0 2 4 6 8
t
Fig. 6.5 Response on high-frequency pulses; ω = 6.
This gives the triangular temporal shape of the motion, x = pτ (ωt) /ω 2 ,

which is approached by the time history record on Fig. 6.5.
Finally, let us consider N -degrees-of-freedom system
de (ωt)
M ÿ + Ky = p (6.41)
d (ωt)
where y (t) is N -dimensional vector-function, p is a constant vector, M and

K are constant N × N mass and stiffness matrixes respectively.
Let {e1 , ..., eN } and ω1 ,...,ωN be the normal mode basis vectors and the
corresponding natural frequencies, respectively, such that
Kej = ωj2 M ej , eTk M ej = δkj
for any k = 1, ..., N and j = 1, ..., N .

Introducing the principal coordinates xj (t),

N
y= xj (t) ej (6.42)
j=1
gives a decoupled set of impulsively forced harmonic oscillators of the form

(6.31),
de (ωt)
ẍj + ωj2 xj = pj (6.43)
d (ωt)
where pj = eTj p.
Therefore, making use of solution (6.35) for each of the oscillators (6.43)
and taking into account (6.42), gives
N
(eTj p)ej sin [(ωj /ω) τ (ωt)]
y= (6.44)
j=1
ωωj cos (ωj /ω)
The corresponding resonances are determined by the condition

2 ωj
ω=
π k
where k = 1, 3, 5, ... and j = 1, ..., N .
6.3.3 Periodic Impulses with a Temporal ‘Dipole’

Shift
Let us consider the impulsive excitation with a dipole shift of pulse times. In
this case, the right-hand side of equation (6.25) can be expressed by second
derivative of the saw-tooth function with some incline described the param-
eter γ as shown in Fig. 6.6
∂ 2 τ (ωt, γ) ∂e (ωt, γ)
v̇ + λv = p 2 =p (6.45)
∂ (ωt) ∂ (ωt)
∞
2p
= [δ (ωt + 1 − γ − 4k) − δ (ωt − 1 + γ − 4k)]
1 − γ2
k=−∞
Fig. 6.6 Basic NSTT asymmetric wave functions.
Based on the NSTT identities introduced in Chapter 4, periodic solutions of

equation (6.45) still can be represented in the form
v = X (τ ) + Y (τ ) e (6.46)
where τ = τ (ωt, γ) and e = e (ωt, γ); see Fig. 6.6 for graphic illustrations.
Substituting (6.46) in equation (6.45), gives
∂e(ωt, γ)
ωαY + λX + [ω (X + βY ) + λY ] e + (ωY − p) =0 (6.47)
∂(ωt)

where α = 1/ 1 − γ 2 , β = 2γα, and the identity e2 = α + βe has been taken
into account.
Equation (6.47) is equivalent to the boundary-value problem
ω (X + βY ) = −λY
ωαY = −λX (6.48)
ωY |τ =±1 = p
The corresponding solution is

p λ cosh ωλ τ λ sinh ωλ τ λ
Y = cosh γ + sinh γ exp γ τ
ω ω cosh ωλ ω sinh ωλ ω
ωα
X=− Y (6.49)
λ
where the X-component is defined by differentiation due to the second equa-
tion in (6.48).
6.4 Parametric Excitation 157
6.4 Parametric Excitation

In this section, two different cases of parametric excitation are consid-
ered based on relatively simple linear models. Piecewise-constant and im-
pulsive excitations are described by means of the functions e(ωt, γ) and
∂e(ωt, γ)/∂ (ωt), respectively. There are at least two reasons for using NSTT
as a preliminary analytical step. First, NSTT automatically gives conditions
for matching solutions at discontinuity points. Second, due to the automatic
matching through the NSTT functions, the corresponding solutions appear
to be in the closed form that is important feature when further manipulations
with the solutions are required by problem formulations.
6.4.1 Piecewise-Constant Excitation

Let us consider the linear oscillator under periodic piecewise-constant
excitation
ẍ + ω02 [1 + εe(ωt, γ)]x = 0 (6.50)
where ω0 , ω, γ and ε are constant parameters.
We will seek periodic solutions with the period of excitation T = 4/ω in
the form
x = X (τ ) + Y (τ ) e (6.51)
where τ = τ (ωt, γ) and e = e(ωt, γ).
As follows from the form of equation (6.50), the acceleration ẍ may have
step-wise discontinuities due to the presence of the function e(ωt, γ), whereas
the coordinate x (t) and the velocity ẋ (t) must be continuous. So neither
velocity ẋ (t) nor acceleration ẍ (t) can include Dirac δ-functions.
Taking first derivative of (6.51), gives

∂e(ωt, γ)
ẋ (t) = αY + (X + βY )e + Y ω (6.52)
∂ (ωt)
where the last term, that consists of the periodic sequence of δ-functions,
must be excluded by imposing the boundary condition for Y -component
Y |τ =±1 = 0 (6.53)
Under condition (6.53), the second derivative takes the form

ẍ (t) = ω 2 [α(X + βY )] + ω 2 [βX + (α + β 2 )Y ]e
∂e (ωt, γ)
+ω 2 (X + βY ) (6.54)
∂ (ωt)
In this case, the singular term, which is underlined in (6.54), is eliminated by

condition
(X + βY ) |τ =±1 = 0 (6.55)
Substituting (6.51) and (6.54) in the differential equation of motion (6.50)
and taking into account the algebraic properties, brings the left-hand side of
the equation to the algebraic form {· · ·} + {· · ·}e. Then, setting separately
each of the two algebraic components to zero, gives the set of differential
equations for X (τ ) and Y (τ ) in the following matrix form

α αβ X 1 αε X
+ r2 =0 (6.56)
β α + β2 Y ε 1 + βε Y
where r = ω0 /ω.
Further, any particular solution of linear differential equations with con-
stant coefficients (6.56) can be represented in the exponential form

X 1
=B exp (λτ ) (6.57)
Y μ
where B, μ and λ are constant parameters.

Substituting (6.57) in (6.56), leads to a characteristic equation which two
pairs of roots determined by the relationships

2
λ2 = − (1 − γ) ε − (1 − γ) r2 ≡ ±k 2 (6.58)

2
λ2 = (1 + γ) ε − (1 + γ) r2 ≡ ±l2
where signs of the notations ±k 2 and ±l2 depend on the parameters ε

and γ.
Let us consider the case of negative signs, when the following condition
holds
− (1 − γ) < ε < (1 + γ) (6.59)
Due to condition (6.59), the stiffness coefficient in equation (6.50) is always
positive, whereas (6.58) gives λ = ±ki and λ = ±li. As a result, the general
solution of equations (6.56) takes the form
X = B1 sin kτ + B2 cos kτ + B3 sin lτ + B4 cos lτ

Y = μ1 (B1 sin kτ + B2 cos kτ ) + μ2 (B3 sin lτ + B4 cos lτ ) (6.60)
where B1 ,...,B4 are arbitrary constants, and
1 αk 2 − r2 1 αl2 − r2
μ1 = − and μ2 = −
α βk 2 − εr2 α βl2 − εr2
Substituting (6.60) in boundary conditions (6.53) and (6.55), gives the ho-
mogeneous set of four linear algebraic equations with respect to the arbitrary
constants. Setting the corresponding determinant to zero, gives condition for

non-zero solutions in the form
[μ1 (1 + βμ2 ) l cos k sin l − μ2 (1 + βμ1 ) k cos l sin k]
× [μ1 (1 + βμ2 ) l cos l sin k − μ2 (1 + βμ1 ) k cos k sin l] = 0 (6.61)
10
6
10Ε
0
0 2 4 6 8 10
r
Fig. 6.7 Instability zones for piecewise constant parametric excitation when
γ = 0.7
One the parameter plane, ε − r, equation (6.61) describes the family of curves
separating stability and instability zones as shown in Fig. 6.7, where the
instability zones are shadowed.
6.4.2 Parametric Impulsive Excitation

Let us consider the case of parametric impulsive excitation whose temporal
shape is given by first derivative of the basic function, e (ωt, γ),

2 ∂e (ωt, γ)
ẍ + ω0 1 + ε x=0 (6.62)
∂ (ωt)
This case was considered in [138] based on the saw-tooth transformation of

time. In particular, it was shown that the periodic solutions of the period
T = 4/ω exists under the condition
2
2 2r2 1 − γ 2 sin2 2r
p = (6.63)
cos 4r − cos 4γr
where r = ω0 /ω and p = εr2 .
1.0
0.8
0.6
p
0.4
0.2
0.0
0 2 4 6 8 10
r
Fig. 6.8 ‘Collapse’ of the instability zones at γ = 1/5: each fifth zone is missing;
here and below, only the upper half-plane is shown due to the symmetry.
1.0
0.8
0.6
p
0.4
0.2
0.0
0 2 4 6 8 10
r
Fig. 6.9 γ = 1/2: each second zone is missing.
The dependence of p on r for fixed γ has the branched zone-like structure

which is typical for different cases of parametrically excited oscillators.
Interestingly enough, different subsequences of zones may disappear as the
parameter γ varies. For instance, if γ = 1/5 then each fifth zone is missing
and, if γ = 1/2 then each second zone is missing; see Figs. 6.8 and 6.9,
respectively. Such an effect was discussed in [138].
6.4.3 General Case of Periodic Parametric Excitation

Below, the problem formulation only is discussed for the case of periodic
parametric loading with both regular and singular components. It is assumed
that there are two discontinuities and singularities on each period located at
the same points. The differential equation of motion is represented in the
vector form
∂e
ẍ + Q (τ ) + P (τ ) e + p x=0 (6.64)
∂ϕ
where x(t) ∈ Rn is the coordinates vector-column,τ = τ (ϕ, γ), e = e (ϕ, γ),
ϕ = ωt is the phase variable, p is a constant n × n matrix, and Q (τ (ϕ, γ))
and P (τ (ϕ, γ)) are periodic matrixes of the period T = 4 with respect to
the phase ϕ.
In equation (6.64), the first two terms of the coefficient can repre-
sent any periodic function q (ϕ) with step-wise discontinuities on Λ =
{t : τ (ϕ, γ) = ±1}. In case the original function q (ϕ) is continuous, one has
P = 0 on Λ.
Let us represent periodic solutions of the period T = 4 in the form (6.51).
Substituting (6.51) in equations (6.64), taking into account the equality
e2 = α + βe, the necessary condition of continuity of the vector function x (t),
(6.53), and using (6.52) and (6.54) gives equations
ω 2 (αX + αβY ) + QX + αP Y = 0

ω 2 α + β 2 Y + βX + P X + QY + βP Y = 0 (6.65)
and the boundary condition

2
ω (X + βY ) + pX |τ =±1 = 0 (6.66)
where, in the case of fixed sign of impulses, the matrix p should be provided
with the factor sgn(τ ).
Together with (6.53), relations (6.65) and (6.66) represent a boundary-
value problem for determining the vector functions X and Y and the corre-
sponding conditions for existence of periodic solutions.
Note that substitution (6.51) in equation (6.64) generates the specific term
e∂e/∂ϕ. Let us show that, within the theory of distributions, this terms can
be interpreted as follows
∂e 1 ∂e
e = β (6.67)
∂ϕ 2 ∂ϕ
First, note that, at this point, the relationship (6.67) is a result of formal
differentiation of both sides of the relation e2 = α + βe with respect to the
phase ϕ. To justify (6.67), let us assume that ω = 1 so that ϕ ≡ t and consider
expression (6.53) locally, near the point t = 1 − γ, which is a typical point of
the entire set of discontinuity points Λ = {t : τ (t) = ±1}.
Generally speaking, the ‘product’ f (t)δ(t) requires the function f (t) to be

at least continuous at t = 0. However, it is possible to provide the left-hand
side of (6.67) with a certain meaning due to the fact that both terms of the
product are generated by the same sequence of smooth functions.
In order to illustrate the above remark and prove equality (6.67), let us
consider a family of smooth functions {δε (t)} such that
ε
δε (t) dt = 1 (6.68)
−ε
for all positive ε, and δε (t) = 0 outside the interval −ε < t < ε.
Therefore, in terms of weak limits, δε (t) → δ (t) as ε → 0.
Now, sequences of smooth functions approximating e and ∂e/∂t in the
neighborhood of point t = 1 − γ can be chosen as, respectively,
1 β ∂eε β
eε = − θε (t − 1 + γ) and = − δε (t − 1 + γ) (6.69)
1−γ γ ∂t γ
"t
where θε (t) = δε (ξ) dξ and −1 + γ < t < 3 + γ.
−∞
Based on the above definitions for eε and ∂eε /∂t, one has eε → e and
∂eε /∂t → ∂e/∂t as ε → 0 in the interval −1 + γ < t < 3 + γ.
Substituting (6.69) in equality (6.67) instead of e and ∂e/∂ϕ, reduces the
problem to the proof of identity
1
θε δ ε = δε (6.70)
2
as ε → 0.
For simplicity reason, let us move the origin to the point t = 1 − γ and
show that the left-hand side of (6.70) gives δ (t) /2 as ε → 0 in the sense of
weak limit.
First, the area bounded by θε δε is
ε ε
dθε 1 1
θε δε dt = θε dt = θε2 |ε−ε =
−ε −ε dt 2 2
Then, let φ (t) belongs to the class of continuous testing functions, which
is usually considered in the theory of distributions. By definition, in some
ε-neighborhood of the point t = 0, one has | φ (t) − φ (0) |< 2η, where η is as
small as needed whenever ε is sufficiently small. Therefore,
ε ε
1
| θε (t) δε (t) φ (t) dt − φ (0) |≤ θε (t) δε (t) | φ (t) − φ (0) | dt ≤ η
−ε 2 −ε
6.5 Input-Output Systems 163
In other words, ε
1
θε (t) δε (t) φ (t) dt → φ (0)
−ε 2
as ε → 0.
This completes the proof.
6.5 Input-Output Systems

The input-output form of dynamical systems may be convenient for different
reasons, for instance, when dealing with control problems. In many linear
cases, input-output systems are represented in the form of a single high order
equation
dn y dy dm u du
an n
+ ... + a1 + a0 y = bm m + ... + b1 + b0 u (6.71)
dt dt dt dt
where u = u(t) and y = y(t) are input and output, respectively, and an , ... ,
a1 , a0 , bm , ... ,b1 , b0 are constant coefficients.
For illustration purposes, a two-degrees-of-freedom model as shown in
Fig. 6.10 is considered, although the general case (6.71) can be handled in
the same way.
Fig. 6.10 Two mass-spring model.
Eliminating x2 (t) from the system, gives a single higher-order equation

with respect to the another coordinate, x1 (t), in the form
d4 x1 d3 x1 m1 d2 x1 c1 dx1 k1 k2
m1 4
+ c1 3 + (k1 + k2 + k2 ) 2 + k2 + x1
dt dt m2 dt m2 dt m2
d2 F1 k2
= 2
+ F1 (6.72)
dt m2
System (6.72) is a particular case of (6.71), where n = 4 and m = 2.
Let us consider the step-wise discontinuous periodic function F1 (t) =
u(t) = e(ωt) and represent equation (6.72) in the form
d4 y dy
a4 4
+ ... + a1 + a0 y = b2 ω 2 e + b1 ωe + b0 e (6.73)
dt dt
where ≡ d/d(ωt), and all the coefficients and variables are identified by
comparing (6.72) to (6.73).
The right-hand side of equation (6.73) contains discontinuous and singular
functions, therefore equation (6.73) must be treated in terms of distributions.
Nevertheless, let us show that, on the manifold of periodic solutions, equation
(6.73) is equivalent to some classic boundary-value problem.
Let us represent the output in the form
y(t) = X(τ ) + Y (τ )e (6.74)
where τ = τ (ωt) and e = e(ωt).

When differentiating expression (6.74) step-by-step one should eliminate
the singular term e in the first two derivatives by sequentially setting bound-
ary conditions as follows
dy
= (Y + X e)ω, Y |τ =±1 = 0 (6.75)
dt
d2 y
= (X + Y e)ω 2 , X |τ =±1 = 0
dt2
However, it is dictated by the form of the input in (6.73), that the singular
terms e and e must be preserved on the next two steps given by
d3 y
= (Y + X e + Y e )ω 3 (6.76)
dt3
d4 y
= (X (4) + Y (4) e + X e + Y e )ω 4
dt4
The fourth-order derivative in (6.76) takes into account the equality ee = 0,
which easily follows from (6.53) in the symmetric case β = 0.
Substituting (6.75) and (6.76) in (6.73), and considering {1, e, e, e } as a
linearly independent basis, gives equations
a4 ω 4 X IV + a3 ω 3 Y + a2 ω 2 X + a1 ωY + a0 X = 0 (6.77)
a4 ω 4 Y IV + a3 ω 3 X + a2 ω 2 Y + a1 ωX + a0 Y = b0
under the boundary conditions at τ = ±1:
Y = 0, X = 0 (6.78)

b2 1 a3
ω 2 Y = , ω 3 X = b1 − b2
a4 a4 a4
In contrast to equation (6.73), the boundary value problem (6.77) and (6.78)
does not include discontinuous terms any more.
6.6 Piecewise-Linear Oscillators with Asymmetric Characteristics 165
Although the number of equations in (6.77) is doubled as compared to

(6.73), such a complication is rather formal due to the symmetry of the
equations. Indeed, introducing the new variables, U = X +Y and V = X −Y ,
decouples system (6.77) in such a way that the corresponding roots of the
characteristic equations differ just by signs. (Besides, this fact reveals the
possibility of using the idempotent basis for decoupling the resultant set
of equations as discussed in Chapter 4 and will be discussed later in this
chapter.) In addition, the type of the symmetry suggests that X(τ ) and Y (τ )
are odd and even functions, respectively. This enables one of reducing the
general form of solution to a family of solutions with four arbitrary constants
2 α
α

j βj j βj
X= Aj cosh τ sin τ + Bj sinh τ cos τ (6.79)
j=1
ω ω ω ω
2 α
α

j βj j βj b0
Y = Aj sinh τ sin τ + Bj cosh τ cos τ +
j=1
ω ω ω ω a 0
where αj ± βj i are complex conjugate roots of the characteristic equation
a4 p4 + ... + a1 p + a0 = 0 (6.80)
The assumption that both of the roots are complex reflects the physical mean-
ing of the example, however other cases would lead to even less complicated
expressions.
Finally, substituting (6.79) in (6.78) gives a linear algebraic set of four
independent equations with respect to four constants: A1 , A2 , B1 and B2 .
Although the corresponding analytical solution is easy to obtain by using the
R
standard Mathematica commands, the result is somewhat complicated for
reproduction. Practically, it may be reasonable to determine the constants by
setting the system parameters to their numerical values moreover that only
numerical solution are often possible for characteristic equations.
6.6 Piecewise-Linear Oscillators with Asymmetric

Characteristics
Piecewise-linear oscillators are often considered as finite degrees-of-freedom
models of cracked elastic structures [32],[2],[192], but may occur also due to
specific design solutions. In many cases, the corresponded periodic solutions
can be combined of different pieces of linear solutions valid for two different
subspaces of the configuration space [33], [75], [192]. In this section, it will
be shown that the nonsmooth transformation of time results in a closed form
analytical solution matching both pieces of the solution automatically by
means of elementary functions.
6.6.1 Amplitude-Phase Equations

Let us consider a piece-wise linear oscillator of the form
mq̈ + k[1 − εH(q)]q = 0 (6.81)
where H(q) is Heaviside unit-step function, m and k are mass and stiffness
parameters, respectively, and |ε| 1.
Therefore, k− = k and k+ = k(1 − ε) are elastic stiffness of the oscillator
for q < 0 and q > 0, respectively.
The exact general solution of oscillator (6.81) can be obtained by satisfying
the continuity conditions for q and q̇ at the matching point q = 0, where the
characteristic has a break. Such approaches are often facing quite challenging
algebraic problems, however, as the number of degrees of freedom increases
or external forces are involved. This is mainly due to the fact that times of
crossing the point q = 0 are a priory unknown.
In this section, it will be shown that, applying a combination of asymp-
totic expansions with respect to ε and nonsmooth temporal transformations,
gives a unit-form solution for oscillator (6.81) with a possibility of general-
ization on the normal mode motions of multiple degrees-of-freedom systems.
In particular, the nonsmooth temporal transformation:
1) provides an automatic matching the motions from different subspaces
of constant stiffness, and
2) justifies quasi-linear asymptotic solutions for the specific nonsmooth
case of piece-wise linear characteristics.
Let us clarify the above two remarks. Introducing the notation ω 2 = k/m,
brings equation (6.81) to the standard form of a weakly non-linear oscillator
q̈ + ω 2 q = εω 2 H(q)q (6.82)
The non-linear perturbation on the right-hand side of oscillator (6.82) is a

continuous but non-smooth function of the coordinate q. Since the major
algorithms of quasi-linear theory assume smoothness of non-linear pertur-
bations, then such algorithms are not applicable in this case unless appro-
priate modifications and extensions have been made. Even though deriving
first-order asymptotic solutions usually require no differentiation of charac-
teristics, dealing with two pieces of the solution may complicate any further
stages.
Let us show that combining quasi-linear methods of asymptotic integra-
tion, such as Krylov-Bogolyubov averaging, with nonsmooth temporal trans-
formations results in a closed form analytical solution for piece-wise linear
oscillator (6.81). Note that oscillator (6.81) plays an illustrative role for the
approach developed below. Then a more complicated case will be considered.
At this stage, let us introduce the amplitude-phase coordinates {A(t), ϕ(t)}
on the phase plane of oscillator (6.81) through relationships
q = A cos ϕ
q̇ = −ωA sin ϕ (6.83)
The following compatibility condition is imposed on transformation (6.83)
Ȧ cos ϕ − A sin ϕϕ̇ = −ωA sin ϕ (6.84)

1
Ȧ = − εωAH(A cos ϕ) sin 2ϕ
2
ϕ̇ = ω − εωH(A cos ϕ) cos2 ϕ (6.85)
The right-hand sides of equations (6.85) are 2π-periodic with respect to the
phase variable, ϕ. Therefore, nonsmooth transformation of the phase variable
applies through the couple of functions
τ = τ (2ϕ/π) and e = e(2ϕ/π) (6.86)
Assuming that A ≥ 0 and taking into account the obvious identities,
sin ϕ = sin(πτ /2)

cos ϕ = cos(πτ /2)e
H(A cos ϕ) = (1 + e)/2 (6.87)
e2 = 1
brings (6.85) to the form

1
Ȧ = − εω(1 + e)A sin πτ (6.88)
4
1 πτ
ϕ̇ = ω − εω(1 + e) cos2 (6.89)
2 2
Note that the right-hand sides of (6.88) and (6.89) are nonsmooth but contin-
uous with respect to the phase ϕ since the step-wise discontinuities of the rect-
angular cosine e(2ϕ/π) are suppressed by the factors sin πτ and cos2 (πτ /2),
respectively.
6.6.2 Amplitude Solution

Let us show that equation (6.88) has an exact 2π-periodic solution with
respect to the phase variable, ϕ.
According to the idea of NSTT, any periodic solution can be represented
in the form
A = X(τ ) + Y (τ )e (6.90)
where τ and e are defined by (6.86).
Substituting (6.90) in (6.88) and taking into account (6.89), gives boundary-
value problem
(X − Y ) = 0
(X + Y ) επ sin πτ
=− (6.91)
X +Y 4 1 − ε cos2 πτ
2
Y |τ =±1 = 0 (6.92)

where ≡ d/dτ .
Solution of the boundary value problem, (6.91) and (6.92), is obtained by
elementary integration. Then representation (6.90) gives
A(ϕ) = α[1 + ζ(τ )] − α[1 − ζ(τ )]e (6.93)

πτ −1/2
ζ(τ ) = (1 − ε cos2 )
2
where τ = τ (2ϕ/π), e = e(2ϕ/π), and α is an arbitrary positive constant.
Note that solution (6.93) exactly captures the amplitude in both subspaces
q < 0 and q > 0. However, the temporal mode shape and the period essen-
tially depend on the phase variable ϕ described by equation (6.89).
Generally speaking, the phase equation (6.89) admits exact integration,
but the result would appear to have implicit form. Alternatively, it is shown
below that solution for the phase variable can be approximated by asymptotic
series in the explicit form
1
ϕ = φ − ε[πτ + (1 + e) sin πτ ]
8
1 2
− ε {4(2 − cos πτ )(πτ + sin πτ ) (6.94)
128
−[4πτ (1 + cos πτ ) − 8 sin πτ + sin 2πτ ]e} + O(ε3 )
where τ = τ (2φ/π), e = e(2φ/π), and

1 3
φ = ω[1 − ε − ε2 + O(ε3 )]t (6.95)
4 32
Note that the functions τ and e in (6.93) and (6.94) depend on the different
arguments.
6.6.3 Phase Solution

In this subsection, a second-order asymptotic procedure for phase equations
with non-smooth periodic perturbations is introduced. If applied to equation
(6.89), the developed algorithm gives solution (6.94).
Let us consider some phase equation of the general form
ϕ̇ = ω[1 + εf (ϕ)] (6.96)
where f (ϕ) is a 2π-periodic, nonsmooth or even step-wise discontinuous func-

tion, and ε is a small parameter, |ε| 1.
Using the basic NSTT identity for f (ϕ), brings equation (6.96) to the form
ϕ̇ = ω + εω{G[τ (2ϕ/π)] + M [τ (2ϕ/π)]e} (6.97)
where the functions G(τ ) and M (τ ) are expressed through f (ϕ).

Note that the class of smoothness of the periodic perturbation in equa-
tion (6.97) depends on the behavior of functions G(τ ) and M (τ ) and their
derivatives at the boundaries τ = ±1. If, for instance, M (±1) = 0 then the
perturbation is step-wise discontinuous whenever τ (2ϕ/π) = ±1.
Let us introduce the asymptotic procedure for equation (6.97). Note that,
in case ε = 0, the right-hand side of equation (6.97) is constant. So, following
the idea of asymptotic integration, let us find phase transformation
ϕ = φ + εF1 (φ) + ε2 F2 (φ) + ... (6.98)
where functions Fi (φ) are such that the new phase variable also has a constant
temporal rate even though ε = 0.
In other words, transformation (6.98) should bring equation (6.97) to the
form
φ̇ = ω(1 + εγ1 + ε2 γ2 + ...) (6.99)
where γi are constant coefficients to be determined together with Fi (φ) during
the asymptotic procedure.
Note that the procedure, which is described below, has several specific
features due to the presence of nonsmooth periodic functions. In particu-
lar, high-order approximations require a non-conventional interpretation for
power series expansions; see the next subsection for the related remarks.
Other modifications occur already in the leading order approximation.
Substituting (6.98) into equation (6.97), then enforcing equation (6.99)
and collecting the terms of order ε, gives
F1 (φ) = G(τ ) + eM (τ ) − γ1 (6.100)
where the triangular and rectangular waves depend now on the new phase
variable φ as τ = τ (2φ/π) and e = e(2φ/π), respectively.
According to the conventional averaging procedure, the constant γ1 is se-
lected to achieve zero mean on the right-hand side of equation (6.100) and
thus provide periodicity of solution, F1 (φ). In the algorithm below, the peri-
odicity is due to the form of representation for periodic solutions, whereas the
operator of averaging occurs automatically from the corresponding conditions
of smoothness that is boundary conditions for the solution components.
So we seek solution of equation (6.100) in the form
F1 (φ) = U1 (τ ) + eV1 (τ ) (6.101)
Substituting (6.101) in (6.100) and following the NSTT procedure, gives the
boundary-value problem
π
U1 (τ ) = M (τ )
2
π
V1 (τ ) = [G(τ ) − γ1 ] (6.102)
2
V1 (±1) = 0
Note that there are two conditions on the function V1 (τ ) described by the
first-order differential equation in (6.102). However, there is a choice for γ1 ,
which is to satisfy one of the two conditions. As a result, solution of boundary-
value problem (6.102) is obtained by integration in the form

π τ
U1 (τ ) = M (z)dz
2 0

π τ
V1 (τ ) = [G(z) − γ1 ]dz (6.103)
2 −1

1 1
γ1 = G(τ )dτ
2 −1
Further, collecting the terms of order ε2 , gives
F2 (φ) = G2 (τ ) + eM2 (τ ) + P2 (τ )e − γ2 (6.104)
where
2
M2 (τ ) = [U1 (τ )G (τ ) + V1 (τ )M (τ )] − M (τ )γ1
π
2
G2 (τ ) = U1 (τ )M (τ ) − G(τ )γ1 + γ12 (6.105)
π
2
P2 (τ ) = U1 (τ )M (τ )
π
e ≡ de(2φ/π)/d(2φ/π)
In contrast to first-order equation (6.100), equation (6.104) includes the sin-

gular term P2 (τ )e produced by the power series expansion of the perturba-
tion in equation (6.97). If the perturbation is smooth then P2 (±1) = 0 and
such singular term disappear; see the example below for illustration. Never-
theless, the second-order approximation remains valid even in discontinuous
case, when P2 (±1) = 0.
So let us represent solution of equation (6.104) in the form
F2 (φ) = U2 (τ ) + eV2 (τ ) (6.106)
Then, substituting (6.106) in (6.104), gives boundary-value problem

π
U2 (τ ) = M2 (τ )
2
π
V2 (τ ) = [G2 (τ ) − γ2 ] (6.107)
2
π
V2 (±1) = P2 (±1)
2
In contrast to (6.102), boundary-value problem (6.107) has, generally speak-
ing, non-homogeneous boundary conditions for V2 . These conditions com-
pensate the singular term e from differential equation (6.104). As a result
equations (6.107) are free of any singularities and admit solution analogously
to first-order equations (6.102),

π τ
U2 (τ ) = M2 (z)dz
2 0

π τ π
V2 (τ ) = [G2 (z) − γ2 ]dz + P2 (−1) (6.108)
2 −1 2

1 1 1
γ2 = G2 (τ )dτ + [P2 (−1) − P2 (1)]
2 −1 2
Now, we return to the illustrating model. In particular case (6.89), one has
1 πτ
G(τ ) ≡ M (τ ) ≡ − cos2 (6.109)
2 2
and
G(±1) = M (±1) = 0
G (±1) = M (±1) = 0 (6.110)
G (±1) = M (±1) = −π 2 /4
where ≡ d/dτ .
First two lines of conditions (6.110) provide continuity for the right hand
side of (6.97) and its first derivative at those ϕ where τ (2ϕ/π) = ±1. As
follows from (6.105), for this class of smoothness one has P2 (±1) = 0 and
thus no singular terms occur in the first two steps of asymptotic procedure.
Finally, taking into account (6.109) and (6.110) and conducting integration
in (6.103) and (6.108), brings solution (6.98) to the form (6.94) and (6.95).
6.6.4 Remarks on Generalized Taylor Expansions

Nonsmoothness of the triangular sine is similar to that function |t| has at
zero. So let us consider its formal power series
1 2
|t + ε| = |t| + |t| ε + |t| ε + ... (6.111)
2!
where ε > 0 and −∞ < t < ∞, and prime indicates Schwartz derivative.
It is clear that equality (6.111) has no regular point-wise meaning. For
instance, equality (6.111) is obviously not true on the interval −ε < t < 0.
In addition, the right-hand side of (6.111) is uncertain at t = 0, whereas the
left-hand side gives ε. Nevertheless, let us show that equality (6.111) admits
a generalized interpretation and holds in terms of distributions. Let ψ(t) be
a test function in terms of the distribution theory, more precisely, ψ(t) is
infinitely differentiable with compact support that is identically zero outside
of some bounded interval. Integrating by parts and then shifting the variable
of integration, gives
∞

1
|t| + |t| ε + |t| ε2 + ... ψ(t)dt
−∞ 2!
∞
1
= |t| ψ(t) − ψ (t)ε + ψ (t)ε2 − ... dt (6.112)
−∞ 2!
∞ ∞
= |t|ψ(t − ε)dt = |t + ε|ψ(t)dt
−∞ −∞
Therefore, equality (6.111) holds in the integral sense of distributions.
q
0
-1
-2
0 10 20 30 40
t
Fig. 6.11 Second-order asymptotic and numerical solutions shown by solid and
dashed lines, respectively
6.7 Multiple Degrees-of-Freedom Case 173
Fig. 6.11 compares analytical solution (6.83), (6.93) and (6.94) shown by
the solid line and numerical solution shown by the dashed line. As expected,
the amplitude show the perfect match, whereas some phase shift develops
after several cycles.
6.7 Multiple Degrees-of-Freedom Case

Let us consider a multiple degrees-of-freedom piecewise-linear system of the
form
M ẍ + Kx = εH(Sx)Bx (6.113)
where x(t) ∈ Rn is a vector-function of the system coordinates, M is a mass
matrix, H denotes the Heaviside unit-step function, S is a normal vector to
the plane splitting the configuration space into two parts with different elastic
properties, so that the stiffness matrix is K when Sx < 0 and K − εB when
Sx > 0. It is assumed that the stiffness jump is small, |ε| 1.
The number of possible iterations of the classic perturbation tools usually
depends on a class of smoothness of the perturbation. The perturbation term
on the right-hand side of (6.113) is continuous but nonsmooth. Therefore, only
first-order asymptotic solution can be obtained within the classic theory of
differential equations. Moreover, the piecewise character of the perturbation
complicates the form of the solution due to the necessity of matching the
different pieces of the solution.
Fig. 6.12 Two degrees-of-freedom piecewise-linear system as a model of a rod with

a small crack.
However, we show that the idea of nonsmooth time transformation gives

a unit-form solution by automatically matching the pieces of solution in two
different configuration subspaces with different stiffness properties.
Let seek a 2π-periodic normal mode solution of (6.113) with respect to the
phase ϕ in the form
x(ϕ) = Aj cos ϕ + εx(1) (ϕ) + O(ε2 )

)
ϕ = ωj 1 + εγ (1) + O(ε2 )t (6.114)
where ωj and Aj are arbitrary eigen frequency and eigen vector (normal
mode) of the linearized system:
(−ωj2 M + K)Aj = 0; j = 1,...,n (6.115)
Substituting (6.114) in (6.113), taking into account (6.87) and (6.115) in the
first order of ε, gives

d2 x(1) 1 1 πτ
ωj2 M 2
+ Kx (1)
= BAj + ( BAj + γ (1)
KAj )e cos (6.116)
dϕ 2 2 2
where τ = τ (2ϕ/π), e = e(2ϕ/π), and the relationship (1 + εγ (1) )−1 =

1 − εγ (1) + O(ε2 ) has been used.
Since the function x(1) (ϕ) is sought to be 2π-periodic with respect to ϕ,
we represent it in the form
x(1) = X(τ ) + Y (τ )e (6.117)
This gives the boundary-value problem

2
2ωj 1 πτ
M X + KX = BAj cos , X |τ =±1 = 0 (6.118)
π 2 2

2

2ωj 1 πτ
M Y + KY = BAj + γ (1) KAj cos (6.119)
π 2 2
Y |τ =±1 = 0
Representing the corresponding solution in terms of the normal mode coor-

dinates
n n
X= Ai Xi (τ ), Y = Ai Yi (τ ) (6.120)
i=1 i=1
and taking into account M -orthogonality of the set of eigen vectors, gives

2
2ωj πτ
Xi + ωi2 Xi = βij cos , Xi |τ =±1 = 0 (6.121)
π 2

2
2ωj πτ
Yi + ωi2 Yi = (βij + γ (1) κij ) cos , Yi |τ =±1 = 0 (6.122)
π 2
where
1 Ai BAj Ai KAj
βij = , κij = (6.123)
2 Ai M Ai Ai M Ai
are dimensionless coefficients.
Note that despite of the similar representation for solution (6.114),
there is a noticeable difference between the classic Poincare-Lindshtedt
method and current procedure due to (6.117). Namely, according to the
6.7 Multiple Degrees-of-Freedom Case 175
Poincare-Lindshtedt method, the frequency correction term γ (1) is to

kill the so called secular terms in the asymptotic expansions. In our case,
the secular terms appear to be periodic due to the ‘built in’ periodicity of the
new temporal argument. However, periodicity of solutions is provided by the
existence of solutions of the boundary-value problems, such as (6.121) and
(6.122). Due to the linearity, the existence of solutions allows the direct veri-
fication. So if i = j then both problems (6.121) and (6.122) are solved in the
standard way with no presence of γ (1) because κij = 0. The corresponding
solution is given by

βij πτ ωj πωi τ πωi
Xi = 2 cos − cos csc (6.124)
ωi − ωj2 2 ωi 2ωj 2ωj
βij πτ
Yi = cos (6.125)
ωi2 − ωj2 2
In the particular case i = j, problem (6.121) still has a solution, but problem
(6.122) generally speaking does not. Fortunately, in this case, we have κjj = 0
and thus the problem is set to have the trivial solution by condition
βjj
γ (1) = − (6.126)
κjj
Therefore,

πβjj πτ 2 πτ
Xj = τ sin + cos (6.127)
4ωj2 2 π 2
Yj = 0 (6.128)
So expressions (6.117), (6.120), and (6.124) through (6.128) completely de-

termine the first order approximation x(1) (ϕ).
Let us consider the example of mass-spring model
m1 ẍ1 + (k1 + k2 )x1 − k2 x2 = εk1 H(x1 )x1 (6.129)

m2 ẍ2 − k2 x1 + (k2 + k3 )x2 = 0
Equations (6.111) can be represented in the form (6.113), where

m1 0 k + k2 −k2 k 0
M = , K= 1 , B= 1
0 m2 −k2 k2 + k3 0 0

x1
x= , S= 10
x2
In this case, the first-order asymptotic solution for the in-phase (j = 1) and
out-of-phase (j = 2) takes the form, respectively,

πτ επ 2 πτ πτ
x1 = e cos + cos + τ sin + O(ε2 )
2 16 π 2 2
πτ εk1 πτ πτ
x2 = e cos − e cos + cos (6.130)
2 8k2 2 2

−1/2
k2 k2 πτ k2 π
− 1+2 cos 1+2 / sin 1+2 + O(ε2 )
k1 k1 2 k1 2

k1 ε
ϕ= 1 − + O(ε2 )t
m 4
and

πτ εk1 πτ πτ k2
x1 = −e cos + e cos + cos − 1+2
2 8k2 2 2 k1

πτ k2 π k2
× cos / 1+2 / sin / 1+2 + O(ε2 ) (6.131)
2 k1 2 k1

πτ εk1 π 2 πτ πτ
x2 = e cos + cos + τ sin + O(ε2 )
2 16(k1 + 2k2 ) π 2 2
*
k1 + 2k2 εk1
ϕ= 1− + O(ε2 )t
m 4(k1 + 2k2 )
where it is assumed that m1 = m2 = m. Solutions (6.130) and (6.100) show
that a bi-linearity may have quite different effect on different modes. In par-
ticular, solution (6.130) reveals the possibility of internal resonances, when

πω2 ω2 k2
sin = 0, = 1+2 (6.132)
2ω1 ω1 k1
If, for instance, the system is close to the frequency ratio ω2 /ω1 = 2 then the
in phase mode may be affected significantly by a crack even under very small
magnitudes of the parameter ε. In contrary, solution (6.100) has the denom-
inator sin[(π/2)ω1 /ω2 ], which is never close to zero because 0 < ω1 /ω2 < 1.
Therefore, in current asymptotic approximation, the influence of crack on the
out-of phase mode is always of order ε provided that k2 /k1 = O(1).
The influence of the bilinear stiffness on inphase mode trajectories in the
closed to internal resonance case is seen from Fig. 6.13, where both analytical
and numerical solutions are shown for comparison reasons. The frequency ra-
tio ω2 /ω1 = 2.0025 is achieved by conditioning the spring stiffness parameters
as follows k2 = (3/2)k1 + 0.005.
6.8 The Amplitude-Phase Problem in the Idempotent Basis 177
1.0
0.5
Ε 0.01 Ε0
x2 0.0
0.5
1.0
1.0 0.5 0.0 0.5 1.0
x1
Fig. 6.13 The influence of a small “crack” ε = 0.01 on the in-phase mode tra-
jectory near the frequency ratio ω2 /ω1 = 2; the dashed line shows the numerical
solution, and the thin solid line corresponds to the perfect (linear) case ε = 0.
6.8 The Amplitude-Phase Problem in the Idempotent

Basis
Recall that the idempotent basis is given by e+ = (1+e)/2 and e− = (1−e)/2
so that e2+ = e+ , e2− = e− and e+ e− = 0. Equations (6.88) and (6.89)
therefore take the form
1
Ȧ = − εωe+ A sin πτ (6.133)
2
πτ
ϕ̇ = ω − εωe+ cos2 (6.134)
2
Let us represent the amplitude as a function of ϕ in the form
A(ϕ) = X+ (τ )e+ + X− (τ )e− (6.135)
where e+ = e+ (2ϕ/π), e− = e− (2ϕ/π) and τ = τ (2ϕ/π).

2 πτ 1
(X+ e+ − X− e− )(ω − εωe+ cos2 ) = − εωe+ (X+ e+ + X− e− ) sin πτ
π 2 2
or
πτ π
(1 − ε cos2 )X+ = − εX+ sin πτ
2 4

X− =0 (6.136)
(X+ − X− )|τ =±1 = 0 (6.137)
The boundary-value problem (6.136) and (6.137) admits exact solution so

that (6.135) gives finally
πτ −1/2
A(ϕ) = α[(1 − ε cos2 ) e+ + e− ] (6.138)
2
where α is an arbitrary positive constant.
Chapter 7
Periodic and Transient Nonlinear
Dynamics under Discontinuous
Loading
Abstract. In this chapter, two variable expansions are introduced, where

the fast temporal scale is represented by the triangular wave. In contrast to
the conventional two variable procedure, the differential equations for slow
dynamics emerge from the boundary conditions eliminating discontinuities
rather than resonance terms. For illustrating purposes, an impulsively loaded
single degree-of-freedom model with qubic damping and no elastic force is
considered. Further, the method is applied to the Duffing’s oscillator un-
der the periodic impulsive excitation whose principal frequency is close to
the linear resonance frequency. Note that the illustrating models are weakly
nonlinear, nevertheless the triangular wave temporal argument adequately
captures specifics of the impulsive loading and, as a result, provides closed
form asymptotic solutions. Following the conventional quasi-harmonic ap-
proaches, in such cases, would face generalized Fourier expansions for the
external forcing function with no certain leading term.
7.1 Nonsmooth Two Variables Method

Solutions of Cauchy problems under arbitrary initial conditions are, gen-
erally speaking, non-periodic. However, in case of amplitude or frequency
modulated oscillating motions, the idea of averaging can be involved in order
to obtain the corresponding solutions. A proper formalization can be devel-
oped by introducing slow temporal variables in addition to the fast oscillating
time τ .
Consider, for instance, first order impulsively loaded system
de (t)
v̇ + εv 3 = p (7.1)
dt
under the initial condition
v |t=0 = v 0 (7.2)

180 7 Periodic and Transient Nonlinear Dynamics under Discontinuous Loading
where ε is a small enough positive parameter, and p is a constant parameter

characterizing the strength of pulses.
The Cauchy problem (7.1) and (7.2) describes one-dimensional motions of
a material point under the periodic series of impulses and non-linear damping
forces. Note that any standard formulation of the idea of averaging is difficult
to apply directly to equation (7.1) since its right hand side is neither smooth
nor even continues function of time.
Analyzing the both sides of equation (7.1), shows that solutions should
have step-wise discontinuities at the points Λ = {t : τ (t) = ±1}. Let us rep-
resent such solutions in the form

v = X τ, t0 + Y τ, t0 e (7.3)
where τ = τ (t), e = e (t) and t0 = εt.

Therefore, the solution is assumed to depend upon the two time scales,
such as the fast oscillating time τ and the slow time t0 .
Substituting now (7.3) into (7.1), gives

∂Y ∂X 3 2
+ε + X + 3XY
∂τ ∂t0

∂X ∂Y
+ +ε 2
+ 3Y X + Y 3
e + (Y − p)ė = 0
∂τ ∂t0
This finally leads to the set of partial differential equations

∂X ∂Y
= −ε 2
+ 3Y X + Y 3
(7.4)
∂τ ∂t0

∂Y ∂X
= −ε 3
+ X + 3XY 2
∂τ ∂t0
under boundary conditions

Y |τ =±1 = p (7.5)
Following the standard procedure of the two-variable expansions, we repre-
sent solution of the boundary value problem (7.4) and (7.5) is in the form of
asymptotic series
∞
∞

X= εi Xi τ, t0 , Y = εi Yi τ, t0 (7.6)
i=0 i=0
Substituting (7.6) into (7.4) and (7.5) and matching coefficients of the same
powers of ε, gives a sequence of simplified boundary value problems. In par-
ticular, zero order problem is
∂X0 ∂Y0
= 0, =0
∂τ ∂τ
7.1 Nonsmooth Two Variables Method 181
Y0 |τ =±1 = p
Both equations and the boundary condition are satisfied by solution

X0 = A0 t0 , Y0 = p (7.7)
where A0 is an arbitrary function of the slow time.

Taking into account (7.7), gives the following first order problem
∂X1
= − 3pA20 + p3
∂τ

∂Y1 dA0
=− 3
+ A0 + 3A0 p 2
(7.8)
∂τ dt0
Y1 | τ =±1 = 0
Integrating equations (7.8), gives general solution

X1 = − 3pA20 + p3 τ + A1 t0 (7.9)

dA0
Y1 = − + A0 + 3A0 p τ + C1 t0
3 2
dt0

The boundary condition for Y1 in (7.8) dictates C1 t0 ≡ 0 and
dA0
+ A30 + 3A0 p2 = 0 (7.10)
dt0

Therefore Y1 ≡ 0, whereas function A1 t0 in the expression for X1 remains
unknown.
Equation (7.10) admits exact solution in the following explicit form

1 2 0 −1/2
A0 = − 2 + const · exp 6p t
3p
Finally, the leading order asymptotic solution of the Caushy problem is writ-
ten as

1 1 0 −2 2 −1/2
v(t) = ± − 2 + + v − p exp 6p εt + pe (t) + O (ε)
3p 3p2
(7.11)
where plus or minus sign are taken if v 0 − p > 0 or v 0 − p < 0, respectively.
Substituting asymptotic solution (7.11) in equation (7.1), leads to the er-
ror εR(εt)e (t), where R(εt) is a function of slow time scale. Although the
error is of the same order of magnitude as the perturbation, it relates to the
singular component of the equation with a scaling factor of order one. In
other words, for such kind of systems, regular and singular error components
0.8
0.6
v
0.4
0.2
0.0
0 10 20 30 40
t
Fig. 7.1 Velocity of the nonlinearly damped material point under the pulsed pe-
riodic excitation.
should be considered separately. Note that the first term in expression (7.11)
is a transient response of the system that vanishes as time increases; see
Fig. 7.1 for illustration.
In conclusion, it is seen from solution (7.11) that the amplitude of impulses
may have a strong effect on the time scale of transient response of the system.
7.2 Resonances in the Duffing’s Oscillator under

Impulsive Loading
Let us consider the Duffing’s oscillator under the periodic impulsive excitation

de (ωt)
ẍ + ω0 x = −ε x − p
2 3
(7.12)
d (ωt)
Introducing the detuning parameter σ into the differential equation (7.12) as

nπ 2
ω02 = ω + εσ (7.13)
2
gives
nπ 2 de (ωt)
ẍ + ω x = −ε σx + x − p
3
(7.14)
2 d (ωt)
where n is an odd positive integer.
Recall that, in the near resonance condition (7.13), the factor π/2 occurs
due to the specific normalization of the triangular sine wave period, T = 4.
7.2 Resonances in the Duffing’s Oscillator under Impulsive Loading 183
Let us represent solutions of equation (7.14) in the form

x(t) = X τ (ωt) , t0 + Y τ (ωt) , t0 e (ωt) (7.15)
where t0 = εt is the slow time used in the previous section.

Based on the continuity condition for x (t),
Y |τ =±1 = 0 (7.16)
one obtains

∂Y ∂X ∂X ∂Y
ẋ = ω +ε 0 + ω +ε 0 e (7.17)
∂τ ∂t ∂τ ∂t
Then, substituting (7.15) and (7.17) in (7.14), eventually gives the boundary
value problem (7.16),
∂X
ω2 |τ =±1 = εp (7.18)
∂τ
and
∂2X ∂2Y 2 nπ 2
2∂ X
ω 2 2 + 2εω + ε + ω X
∂τ
∂τ ∂t0 ∂t02 2
= −ε σX + X + 3Y X
3 2
∂2Y ∂2X 2 nπ 2
2∂ Y
ω 2 2 + 2εω + ε + ω Y (7.19)
∂τ
∂τ ∂t0 ∂t02 2
= −ε σY + Y 3 + 3X 2 Y
Note that, due to substitution (7.15), the boundary value problem, (7.16),
(7.18) and (7.19), includes no singular functions. Therefore, further averaging
procedures can be correctly applied.
Following the formalism of two variable expansions, we seek asymptotic
solutions in the form of power series

X = X0 τ, t0 + εX1 τ, t0 + ε2 X2 τ, t0 + ...

Y = Y0 τ, t0 + εY1 τ, t0 + ε2 Y2 τ, t0 + ... (7.20)
Substituting (7.20) in (7.16), (7.18) and (7.19), and matching coefficients of

the same degrees of ε, gives a series of boundary value problems, where the
leading order problem is
∂ 2 X0 nπ 2 ∂X0
+ X0 = 0, |τ =±1 = 0
∂τ 2 2 ∂τ
∂ 2 Y0 nπ 2
+ Y0 = 0, Y0 |τ =±1 = 0 (7.21)
∂τ 2 2
These equations and the boundary conditions are satisfies at any t0 by
solution
nπ nπ
X0 = A0 t0 sin τ, Y0 = D0 t0 cos τ (7.22)
2 2

where A0 t0 and D0 t0 are arbitrary functions of the slow time scale to
be determined on the next step of iteration.
So, collecting all terms of order ε, gives the boundary value problem of the
next asymptotic order as
2 nπ 2

2 ∂ X1 ∂ 2 Y0
ω + X1 = − 2ω 3 2
+ σX0 + X0 + 3Y0 X0
∂τ 2 2 ∂τ ∂t0
∂X1
ω2 |τ =±1 = p (7.23)
∂τ

∂ 2 Y1 nπ 2 ∂ 2 X0
ω2 + Y1 = − 2ω + σY0 + Y0
3
+ 3X 2
0 Y0
∂τ 2 2 ∂τ ∂t0
Y1 |τ =±1 = 0
The right-hand sides of equations (7.23)

implicitly depend
of the slow time
t0 through still unknown functions A0 t0 and D0 t0 , and their derivatives

A0 t0 and D0 t0 .
Solution of the differential equations (7.23) is found to be
nπτ nπτ
X1 = A1 t0 sin + B1 t0 cos

2 2
1 A0 3 2 3 2 nπτ
+ σ + A0 + D0 − D0 τ cos
ω nπω 4 4 2

1 A0 7 3 nπτ
− σ + A20 + D02 − D0 sin (7.24)
nπω nπω 8 8 2
A0 3nπτ
− 2 2 2 A20 − 3D02 sin
8n π ω 2
and
nπτ nπτ
Y1 = C1 t0 sin + D1 t0 cos

2 2
1 D0 3 2 3 2 nπτ
− σ + A0 + D0 + A0 τ sin
ω nπω 4 4 2

1 D0 3 7 nπτ
− σ + A20 + D02 + A0 cos (7.25)
nπω nπω 8 8 2
D0 2 3nπτ
− 2 2 2 3A0 − D02 cos
8n π ω 2
Now the boundary condition for X1 gives

B1 t0 ≡ 0
7.3 Strongly Nonlinear Oscillator under Periodic Pulses 185
and
A0 3 2 2
2p nπ
D0 = σ+ A0 + D0 + sin (7.26)
nπω 4 nπω 2
Then, the boundary condition for Y1 gives

C1 t0 ≡ 0
and
D0 3 2
A0 = − σ+ A0 + D02 (7.27)
nπω 4
In current approximation, another
two arbitrary
functions of the slow time
scale can be chosen as A1 t0 ≡ 0 and D1 t0 ≡ 0, at least, because the
same type of terms are included already into the generating solution (7.22).
Substituting derivatives from (7.26) and (7.27) in (8.24) and (7.25) and taking
into account (7.22), gives
nπτ ε nπτ
X = A0 sin 1 − 2 2 2 A20 − 3D02 cos2
2 2n π ω 2
2εp nπ nπτ nπτ
− 2 2 2 sin nπτ cos − sin + O(ε2 ) (7.28)
n π ω 2 2 2
nπτ ε nπτ
Y = D0 cos 1 + 2 2 2 3A20 − D02 sin2 + O(ε2 ) (7.29)
2 2n π ω 2
Substituting (7.28) and (8.87) in (7.15), finally gives the closed form approx-
imate solution x(t), despite of the discontinuous impulsive loading.
Figs. 7.2, 7.3, and 7.4 illustrate the results of calculations under the fol-
lowing parameters and initial conditions: n = 1, ω = 1.0, σ = 0.1, p = 1.0,
ε = 0.1, A0 (0) = 0.1, D0 (0) = 0.0. The coordinate-velocity diagram shows
two velocity jumps per one cycle due to the periodic impulsive excitation.
7.3 Strongly Nonlinear Oscillator under Periodic

Pulses
Let us consider the case of strongly nonlinear exactly solvable oscillator de-
scribed in Chapter 3 by adding periodic impulsive loading on the right-hand
side as follows
F de (t/a)
ẍ + tan x + tan3 x =
a d (t/a)
∞
= 2F (−1)k δ[t − (2k − 1)a] (7.30)
k=−∞
1.5
1.0
B0
0.5
0.0
0.5
A0
1.0
1.5
2.0
0 5 10 15 20 25
0
t
Fig. 7.2 The amplitudes A0 (t0 ) and D0 (t0 ) in the slow time scale, t0 = εt, shown
by the gray and black lines, respectively.
x 0
1
2
0 50 100 150 200 250
t
Fig. 7.3 The time history of the oscillator near the principal resonance.
Here 2F > 0 and 4a = T > 0 characterize the amplitude and period of the
impulsive loading.
Making substitutions x = X(τ ) and τ = τ (t/a) in equation (7.30), gives

d2 X dX de (t/a)
a−2 2 + tan X + tan3 X = a−2 aF −
dτ dτ d (t/a)
7.3 Strongly Nonlinear Oscillator under Periodic Pulses 187
v 0
1
2
3
2 1 0 1 2
x
Fig. 7.4 The coordinate-velocity plane trajectory on the interval 0 < t < 40.
or
d2 X
a−2 + tan X + tan3 X = 0 (7.31)
dτ 2
under the following continuity condition for x(t),
dX
|τ =±1 = aF (7.32)
dτ
Boundary value problem (7.31) and (7.32) describe the class of steady state
periodic motions of the period T .
With reference to Chapter 3, equation (7.31) has exact solution of the form
X(τ ) = arcsin[sin A sin(aτ / cos A)] (7.33)

where A is an arbitrary constant, which is sufficient to satisfy both conditions

in (7.32) due to the symmetry of solution (7.33).
Substituting (7.33) in (7.32) and conducting analytical manipulations with
elementary functions, gives

α2 F
T = 4α kπ ± arccos , (k = 1, 2, 3, ...) (7.34)
(1 − α2 )(1 − α2 F 2 )
α2 F
T = 4α arccos , (k = 0) (7.35)
(1 − α2 )(1 − α2 F 2 )
where T = 4a is the period of forced vibration, and α = cos A.
35
30
25
20
T
15
k3
10
k2
5 k1
0 k0
1.1 1.2 1.3 1.4 1.5
A
Fig. 7.5 First few branches of the period versus amplitude diagram under the fixed
pulse amplitudes, F = 1.8.
It can be shown that solutions (7.34) and (7.35) exist in the interval
1 π
arccos √ = Amin < A < Amax = (7.36)
1+F 2 2
Fig. 7.5 illustrates the sequence of branches of solutions (7.34) and (7.35) at
different numbers k, and the parameter F = 1.8. The diagram gives such
combinations of the period and amplitude at which oscillator (7.30) can have
periodic motions with the period of external impulses, T = 4a. The upper and
7.4 Impact Oscillators under Impulsive Loading 189
lower branches of each loop correspond to plus and minus signs in expression
(7.34), respectively. Solution (7.35) corresponds to the number k = 0 and
has the only upper branch. For the selected magnitude of F , the minimal
amplitude is found to be Amin = 1.0637, which corresponds to the left edges
of the amplitude-period loops in Fig. 7.5. As a result, further slight increase
of the amplitude is accompanied by bifurcation of the solutions as shown
in Fig.7.6. The diagram includes only first three couples of new solutions
(k = 1, 2, 3) from the infinite set of solutions. Further evolution of temporal
shapes of solutions, related to the amplitude increase, is illustrated by Figs.
7.7 and 7.8. As follows from these diagrams, the influence of external pulses
on the temporal shapes is decreasing as the amplitude grows. This is the
result of dominating the restoring force over the external pulses. When the
amplitude becomes close to its maximum Amax = π/2, the oscillator itself
generates high-frequency impacts; see Chapter 3 for details.
7.4 Impact Oscillators under Impulsive Loading

Consider nonlinear oscillator under periodic pulse excitation of the following
general form
de (ωt)
ẍ + 2ζ ẋ + f (x, ωt) = p (7.37)
d (ωt)
where ζ, ω and p are constant parameters, and the function f (x, ωt) is peri-
odic with respect to the argument ωt with the period equal to four.
It is assumed that the coordinate x is subjected to either the constraint
condition
0 ≤ x (t) (7.38)
or
− 1 ≤ x (t) ≤ 1 (7.39)
According to the idea of non-smooth transformation of positional coordinates
[199], [204] (see also the end Chapter 1 of this book), the constraints (7.38)
or (7.39) are eliminated by unfolding the coordinate
x (t) −→ l (t) : x = S (l) (7.40)
where S (l) ≡ |l| in case (7.38) or S (l) ≡ τ (l) in case (7.39), so that the
new coordinate l belongs to the entire infinite interval, l (t) ∈ (−∞, ∞), in
contrast to (7.38) or (7.39).
Taking into account the identity [S (l)]2 = 1, gives the final result of
transformation (7.40) in the form
d2 l dl de (ωt)
2
+ 2ζ + S (l) f [S (l) , ωt] = pS (l) (7.41)
dt dt d (ωt)
where, as mentioned above, the new coordinate l (t) is free of any constraints.
x x
1.0 k1 1.0
0.5 0.5
t t
1 2 3 4 5 1 2 3 4 5 6
0.5 0.5
1.0 1.0
x k2 x
1.0 1.0
0.5 0.5
t t
2 4 6 8 10 12 2 4 6 8 10 12
0.5 0.5
1.0 1.0
x k3
x
1.0 1.0
0.5 0.5
t t
5 10 15 5 10 15
0.5 0.5
1.0 1.0
Fig. 7.6 Periodic solutions corresponding to the first three closed loops of the
diagram shown in Fig. 7.5 at A = 1.0647; the left and right columns relate to lower
and upper branches of the loops, respectively.
Note that the right-hand side of equation (7.41) makes sense within the
distribution theory under the condition that times of interaction with con-
straints (7.38) or (7.39) never coincide with the external pulse times.
Consider now the class of periodic motions of the period of external load-
ing, T = 4/ω. As follows from Chapter 6 and the previous sections of this
chapter, within such class of motions, the external pulses are eliminated by
introducing the new time argument as follows
t −→ τ : τ = τ (ωt) , l = X (τ ) + Y (τ ) e (7.42)
x x
1.0 1.0
k1
0.5 0.5
t t
0.5 1.0 1.5 2.0 2.5 3.0 1 2 3 4 5 6 7
0.5 0.5
1.0 1.0
x x
k2
1.0 1.0
0.5 0.5
t t
2 4 6 8 2 4 6 8 10 12
0.5 0.5
1.0 1.0
x x
k3
1.0 1.0
0.5 0.5
t t
2 4 6 8 10 12 5 10 15
0.5 0.5
1.0 1.0
Fig. 7.7 Same as Fig. 7.6 for A = 1.15.
So, substituting (7.42) into equation (7.41), gives
ω 2 X (τ ) + 2ωζY (τ ) + R (X, Y, τ )

+ ω 2 Y (τ ) + 2ωζX (τ ) + I (X, Y, τ ) e (7.43)
de (ωt)
= pRS (X, Y ) + pIS (X, Y ) e − ω 2 X (τ )
d (ωt)
where

R 1
= {S (X + Y ) f [S (X + Y ) , τ ] ± S (X − Y ) f [S (X − Y ) , 2 − τ ]}
I 2
x x
k1
1.0 1.0
0.5 0.5
t t
0.5 1.0 1.5 1 2 3 4
0.5 0.5
1.0 1.0
x x
k2
1.0 1.0
0.5 0.5
t t
1 2 3 4 5 2 4 6 8
0.5 0.5
1.0 1.0
x x
k3
1.0 1.0
0.5 0.5
t t
2 4 6 8 2 4 6 8 10
0.5 0.5
1.0 1.0
Fig. 7.8 Same as Fig. 7.6 for A = 1.3.
and
RS 1
= {S (X + Y ) ± S (X − Y )}
IS 2
under the necessary condition of continuity for l (t),
Y (±1) = 0 (7.44)
Equation (7.43) is equivalent to the following boundary value problem
ω 2 X (τ ) + 2ωζY (τ ) + R (X, Y, τ ) = 0
ω 2 Y (τ ) + 2ωζX (τ ) + I (X, Y, τ ) = 0 (7.45)
[X (τ ) − ω −2 pS (X)]|τ =±1 = 0 (7.46)

where boundary condition (7.46) has been simplified by enforcing condition

(7.44).
After solution of the boundary value problem (7.44) through (7.46) has
been obtained, the original coordinate is given by composition of transfor-
mations (7.40) and (7.42) as
x (t) = S (X (τ ) + Y (τ ) e) (7.47)
where τ = τ (ωt) and e = e (ωt).

The advantage of such a boundary value problem formulation is that the
effect of both internal and external pulses are captured by the composition of
two transformations (7.47). As a result, the final system is free of any singular
terms. Some particular cases and examples were considered earlier in [140].
Chapter 8
Strongly Nonlinear Vibrations
Abstract. This chapter presents analytical successive approximations al-

gorithms for different oscillators with strongly nonlinear characteristics. In
general terms, such algorithms approximate temporal mode shapes of vibra-
tions by polynomials and other simple functions of the triangular sine wave.
In order to develope the algorithms, the triangular wave is introduced into
dynamical systems as a new temporal argument. The corresponding manip-
ulations with dynamical systems are described in the first three sections.
Then the description focuses on the algorithm implementations for different
essentially unharmonic cases including oscillators whose characteristics may
approach nonsmooth or even discontinuous limits.
8.1 Periodic Solutions for First Order Dynamical

Systems
Let us consider a dynamical system described by first-order differential equa-

tion with respect to the vector-function x(t) ∈ Rn ,
ẋ = f (x) (8.1)
where f (x) is a continuous vector-function, and the over dot indicates time
derivative.
We consider the class of periodic motions of the period T = 4a. Note that,
in the autonomous case, the period is a priori unknown. Periodic solutions
usually require specific, a priory unknown, initial conditions. Practically, how-
ever, such kind of the specific initial conditions are determined in a backward
way, after some periodic family of solutions is obtained under the assumption
of periodicity. In our case, the assumption of periodicity is imposed automat-
ically by the form of representation for periodic solutions
x = X(τ ) + Y (τ )e (8.2)

196 8 Strongly Nonlinear Vibrations
where τ = τ (t/a) and e = e(t/a) are the standard triangular sine and rect-
angular cosine, respectively, and X(τ ) and Y (τ ) are unknown components of
the solution.
(Y − aRf ) + (X − aIf )e + Y e = 0
where
1
Rf = Rf (X, Y ) = [f (X + Y ) + f (X − Y )]
2
1
If = If (X, Y ) = [f (X + Y ) − f (X − Y )]
2
Eliminating the periodic singular term e = de(t/a)/d(t/a) by means of the
boundary condition for Y (τ ), gives the non-linear boundary value problem
on the standard interval, −1 ≤ τ ≤ 1,
Y = aRf (X, Y )
X = aIf (X, Y ) (8.3)
Y |τ =±1 = 0
Note that, the entire interval −1 ≤ τ ≤ 1 is completely covered by a half of the

period, −a ≤ t ≤ a, however, representation (8.2) unfolds the corresponding
fragment on the entire time interval −∞ < t < ∞.
8.2 Second Order Dynamical Systems

Consider now the differential equation of motion in the standard Newtonian
form
ẍ + f (x, ẋ, t) = 0 (8.4)
where x(t) ∈ Rn is a positional vector-function, and the vector-function f is
assumed to be sufficiently smooth and periodic with respect to the explicit
time t with the period T = 4a; the autonomous case, when the period is a
priory unknown, is under consideration as well.
Substituting (8.2) into (8.4), using the differential and algebraic properties
of substitution (8.2), and imposing the boundary (smoothness) conditions,
gives
(X + a2 Rf ) + (Y + a2 If )e = 0
where

1 X + Y X − Y
Rf = f X + Y, , aτ + f X − Y, − , 2a − aτ
2 a a
(8.5)
8.2 Second Order Dynamical Systems 197

1 X + Y X − Y
If = f X + Y, , aτ − f X − Y, − , 2a − aτ
2 a a
(8.6)
This leads to the boundary value problem
X + a2 Rf (X, Y, X , Y , τ ) = 0, X |τ =±1 = 0 (8.7)
Y + a2 If (X, Y, X , Y , τ ) = 0, Y |τ =±1 = 0 (8.8)

Let us discuss the form of equations (8.7) and (8.8).
Firstly, despite of the obvious formal complication, equations (8.7) and
(8.8) possess certain symmetries dictated by substitution (8.2). For instance,
introducing the new unknown variables,
U (τ ) = X(τ ) + Y (τ )
V (τ ) = X(τ ) − Y (τ )
brings the boundary value problem, (8.7) and (8.8), to the form, in which
the differential equations are decoupled at cost of coupling the boundary
conditions though,
U + a2 f (U, U /a, 2a − aτ ) = 0
V + a2 f (V, − V /a, 2a − aτ ) = 0 (8.9)

U + V |τ =±1 = 0
U − V |τ =±1 = 0
In case when analytical methods are applied, the differential equations (8.9)
for U (τ ) and V (τ ) can be usually solved in a similar way. Nevertheless, the
previous boundary value problem, (8.7) and (8.8), may appear to have some
advantages for analyses. For instance, the problem may admit families of
solutions with either Y (τ ) ≡ 0 or X(τ ) ≡ 0.
Secondly, due to substitution (8.2), the major qualitative property of solu-
tions, such as periodicity, is a priory captured by the new argument, τ . As a
result, the following simplified system can be employed as a generating model
for analytical algorithms of successive approximations
X = 0 (8.10)

Y =0
Indeed, substituting obvious solutions of equations (8.10) in (8.2), gives a

family of nonsmooth periodic motions with respect to the original time pa-
rameter, t,
x(t) = X(0) + X (0)τ (t/a) + (Y (0) + Y (0)τ (t/a))e(t/a) (8.11)

Thirdly, from the physical standpoint, linear equations (8.10) describe a

strongly nonlinear (nonsmooth) generating model. In particular, if Y (0) = 0
and Y (0) = 0, then vector-function (8.11) describes vibrations of basic vibro-
impact models.
The analytical algorithms developed in few sections below are based on the
idea of approximation of smooth vibrating systems by the basic vibroimpact
models. In other words, the triangular sine wave is assumed to be a domi-
nant component of temporal mode shapes of vibrations. Such an idea indeed
follows the analogy with the quasi harmonic approaches. In particular, the
harmonic balance method approximates vibrating systems by effective har-
monic oscillators regardless types or magnitudes of the system nonlinearities.
This is justified by the fact that Fourier coefficients usually decay in a fast
enough rate so that, for instance, second term can be considered as a small
correction to the first term. The corresponding “small parameter” is therefore
hidden in the iterative procedure itself rather than explicitly present in the
differential equations of motion.
Finally, equations (8.10) make sense due to the temporal substitution t −→
τ (t/a). In terms of the original variables, the corresponding equation, ẍ = 0,
contains too little information about the original system (8.4) and captures
no global properties of the dynamics.
8.3 Periodic Solutions of Conservative Systems
8.3.1 The Vibroimpact Approximation

Let us consider the case of n-degrees-of-freedom conservative system
ẍ + f (x) = 0 (8.12)
where f (x) is an odd analytical vector-function of the positional vector-

column x(t) ∈ Rn .
A one-parameter family of periodic solutions will be built such that
X(−τ ) ≡ −X(τ ). Since equation (8.12) admits the group of time transla-
tions then another arbitrary parameter can be always added to the time
variable. Taking into account the symmetry of system (8.12), enables one of
considering the particular case of substitution (8.2)
x(t) = X(τ (t/a)), Y ≡ 0 (8.13)
Based on the conditions assumed, the boundary value problem (8.7) and (8.8)
is reduced to the following one
X + a2 f (X) = 0 (8.14)
X |τ =1 = 0
8.3 Periodic Solutions of Conservative Systems 199
We seek solutions of the boundary value problem (8.14) in the form of series
of successive approximations
X = X 0 (τ ) + X 1 (τ ) + X 2 (τ ) + ... (8.15)
a2 = h0 (1 + γ1 + γ2 + ...) (8.16)
In order to organize the corresponding iterative procedure, it is assumed that
O(
X i
) O(
X i+1
)
O(γi+1 ) O(γi+2 ) (8.17)
(i = 0, 1, 2, ...)
where the norm of vector-functions is defined by

X
=max
X
Rn .
τ
Based on assumptions (8.17), series (8.15) and (8.16) generate the se-
quences of equations and boundary conditions as, respectively,
X 0 = 0 (8.18)
X 1 = −h0 f (X 0 ) (8.19)

X 2
= −h0 [γ1 f (X ) +0
fx (X 0 )X 1 ] (8.20)
···
and
(X 0 + X 1 )|τ =1 = 0 (8.21)
X 2 |τ =1 = 0 (8.22)
···
Note that condition (8.21) includes first two approximations as the only way
to proceed with a non-zero generating solution. In particular, the generating
solution is found from equation (8.18) in the form
X 0 = A0 τ (8.23)
where A0 ∈ Rn is an arbitrary constant vector, and the oddness condition

has been enforced in order to set to zero another constant vector.
In line with the discussion at the end of the previous section, solution
(8.23) describes a multi-dimensional vibro-impact oscillator between two ab-
solutely stiff and perfectly elastic barriers such that A0 is the normal vector
to both barriers. Direction of the vector A0 will be defined on the next step of
successive approximations, whereas its length will appear to be coupled with
the parameter h0 by some relationship due to boundary condition (8.21).
So substituting (8.23) in (8.19) and integrating, gives solution

τ
X = A τ − h0
1 1
(τ − ξ)f (A0 ξ)dξ (8.24)
0
where A1 is another arbitrary constant vector.

Note that the first term in expression (8.24) is similar to generating solu-
tion (8.23) and thus contributes nothing new into the entire solution within
the first two steps of the procedure. Therefore, we take A1 = 0 and then
substitute the combination X 0 + X 1 in the boundary condition (8.21). This
gives a nonlinear eigen value problem with respect to the vector A0 in the
form
1
1 0
f (A0 τ )dτ = A (8.25)
h0
0
Equation (8.25) represents a set of n scalar equations relating the components

of vector A0 and the parameter h0 . The combination A0 and 1/h0 will be
interpreted as an eigen vector and eigen value of the nonlinear eigen vector
problem (8.25).
Taking scalar product of both sides of equation (8.25) with A0T , gives
A0T A0
h0 = (8.26)
"1
A0T f (A0 τ )dτ
0
where the upper index T stays for transpose operation.

In order to clarify the meaning of expressions (8.25) and (8.26), let us
consider the linear case f (x) ≡ Kx, where K is an n × n stiffness matrix. The
corresponding relationships will differ from those of the exact linear theory
by specific constant factors because the temporal mode shape of vibrations
is not exact but approximated by the triangular sine wave. Nevertheless,
in nonlinear cases, expression (8.26) can provide estimates for amplitude-
frequency response characteristics.
Further, integrating equation (8.20), gives
τ
X = A τ − h0
2 2
(τ − ξ)[γ1 f (A0 ξ) + fx (A0 ξ)X 1 (ξ)]dξ (8.27)
0
where A2 is an arbitrary constant vector, and fx ( A0 ξ ) is the n × n − matrix

of first partial derivatives (Jacobian).
Then boundary condition (8.22) gives
1
2
A = h0 [γ1 f (A0 τ ) + fx (A0 τ )X 1 (τ )]dτ (8.28)
0
where the coefficient γ1 is yet unknown.

In order to determine the coefficient γ1 , an additional condition for the
vector A2 can be imposed, for instance, as follows
A0T A2 = 0 (8.29)
This condition means that the vector A2 must be orthogonal to the corre-
sponding vector of the generating solution A0 in order to keep the amplitude
fixed.
"1
A0T fx (A0 τ )X 1 dτ
0
γ1 = − (8.30)
"1
A0T f (A0 τ )dτ
0
This completes the second step of successive approximations. All the further
steps can be passed in the same way.
In general terms, convergence properties of the above procedure are due
to the following integral operator
⎧ 1 ⎫
⎨ τ ⎬
F [X] ≡ a2 τ f (X(ξ))dξ + ξf (X(ξ))dξ (8.31)
⎩ ⎭
τ 0
where
"1
A0T f (A0 τ )dτ
0
a2 = h 0 (8.32)
"1
A0T f (X)dτ
0
Based on definition (8.31), the original boundary value problem (8.14) admits
representation in the form X = F [X]. Therefore, the convergence condition
is

FX [X 0 ]δX
<1 (8.33)

δX
where δX is an arbitrary vector-function from a small enough neighborhood

of X 0 .
In the case of linearized system, condition (8.33) leads to the set of in-
equalities ωi /ωj < 1 for all i = j, where ωj is the eigen frequency of the
linear normal mode, which is chosen to be a generating solution.
8.3.2 One Degree-of-Freedom General Conservative

Oscillator
In the one-degree-of-freedom case with odd characteristic, the boundary con-
dition at τ = 1 is reduced to a single equation, which is sequentially satisfied
by the factor h0 and terms γ1 , γ2 ,... of series (8.16). As a result, the process
of successive approximations therefore eases by setting A0 = A and Ai = 0
for i = 1, 2,... .
Let us introduce notations hi = h0 γi and represent series (8.15) and (8.16)
in the form
X = X0 (τ ) + X1 (τ ) + X2 (τ ) + ...
a2 = h0 + h1 + h2 + ... (8.34)
where Xi (τ ) are scalar functions of the triangular sine wave τ = τ (t/a).

Due to the reduction of one-dimensional case, all terms of the expansions
are iteratively determined by the explicit relationships. First two steps of the
iterative procedure are coupled by the smoothing boundary condition (8.21)
that provides the leading order smooth estimate for the temporal mode shape
by coupling the parameters, h0 = h0 (A), as follows
X0 = Aτ (8.35)
τ
X1 = −h0 (τ − ξ)f (Aξ)dξ
0
1
h0 = A/ f (Aξ)dξ (8.36)
0
All the next steps of the procedure are passed then in a similar way based
on relationships

i τ
Xi = − hj−1 (τ − ξ)Ri−j dξ
j=1 0

i−1
hi−1 = − αi−j hj−1 (8.37)
j=1
(i = 2, 3, ...)
where the coefficients and integrands are generated by means of the formal
auxiliary parameter, ε, as follows
1 1
αi = Ri dξ/ R0 dξ (8.38)
0 0
1 di f (X 0 + εX 1 + ε2 X 2 + · · ·)
Ri = |ε=0
i! dεi
(i = 0, 1, 2, . . .)
The way of using the parameter ε is in compliance with assumptions (8.17).

Respectively, such parameter splits the restoring force according to (8.38) and
then disappears from expressions. The convergence of such iterative series of
successive approximations is illustrated by the following example.
Example 12. Let us consider the oscillator
ẍ + xm = 0
where m is an odd positive integer. This oscillator was already discussed in

Chapter 3 under the notation m = 2n − 1. Now, applying two iterations
according to the above scheme, (8.36) and (8.37), gives solution

2m+3
τ m+2 m τ τ m+2
X =A τ− + − − R3 − R4 − · · ·
m + 2 2(m + 2) 2m + 3 m + 2
2
(8.39)
m + 1 m (m + 1)
a2 = m−1 1 + 1+ + r3 + r4 + · · · (8.40)
A 2(m + 2) (m + 2)(2m + 3)
where expressions
m+2
m |τ |
0 < Ri (m, τ ) < (8.41)
2 (m + 2)2
i−1
m
0 < ri (m) < i
2 (m + 2)
provide estimates for high order terms of the successive approximations. In

particular, expressions (8.41) indicate that series (8.39) and (8.40) may con-
verge quite slowly. However, the asymptotic of large exponents m essentially
improves precision of the truncated series even though first few terms of
the series are included. The temporal mode shapes of different iterations are
shown in Fig. 8.1, whereas Fig. 8.2 illustrates the period as a function of
the number m under the fixed energy in the first iteration only. For com-
parison reason, the exact result and first order approximation according to
the harmonic balance method are also shown in the diagram. In particular,
Fig. 8.1 shows that high-order iterations are localized near the time points
corresponding to amplitude positions of the oscillator. For instance, the first
Fig. 8.1 The first three terms of iteration (thin lines) and their sum (solid line)
for the temporal mode shape of the oscillator ẍ + x5 = 0.
5
Period, T
harmonic balance
4
exact solution
3
sawtooth
0 10 20 30 40 50
Degree of nonlinearity, m
Fig. 8.2 The period of the oscillator with power characteristic at different expo-
nents m obtained by three different methods under the total energy E = 2.
iteration X1 just compensates the discontinuities of slope of the generating

solution X0 with a minor effect on the rest of the triangular sine wave. Fur-
ther, Fig. 8.2 confirms that expansion (8.40) gives a better estimate for the
period than the harmonic balance as the exponent m increases. Note that the
entire series (8.39) and (8.40) are not asymptotic with respect to m or 1/m
in the sense of Poincare, however the series perfectly capture the asymptotic
of impact oscillator as m → ∞.
8.3.3 A Nonlinear Mass-Spring Model That Becomes

Linear at High Amplitudes
As another example of conservative oscillator, we consider a single mass vi-
brating system illustrated by Fig. 8.3 and described by the Lagrangian
2
mẇ2 w2
L= − kl2 1+ 2 −1
2 l
Here, m is mass, k is the linear stiffness of each spring, l is the length of each
spring at the equilibrium position at which the springs are horizontal, and w
is the particle vertical coordinate.
In terms of the dimensionless coordinate x = w/l and phase ϕ =
(2k/m)1/2 t, the corresponding differential equation of motion takes the form
d2 x x
+x− √ =0 (8.42)
dϕ2 1 + x2
Then, applying substitution (8.13) as x = X(τ ) and τ = τ (ϕ/a), leads to the
boundary problem

X
X = −h X − √ ≡ −hf (X)
1 + X2
X |τ =1 = 0 (8.43)
where h = a2 .
wt
l l
Fig. 8.3 The system which becomes weakly non-linear at large amplitudes.
First two steps of the successive approximation procedure give
X0 = Aτ
√ −1
1 1 + A2 − 1
h0 = −
2 A2
and

h0 1
X1 (τ ) = − 2 (Aτ ) + 2Aτ − Aτ 1 + (Aτ ) − arcsinh(Aτ ) (8.44)
3 2
2A 3

h0 6A + A3 9A − A3 1
γ1 = 3 + √ − 1+ √ arcsinh(A)
A 12 6 1 + A2 1 + A2
Interestingly enough, this model is essentially nonlinear at small amplitudes,
but it becomes linear as the amplitudes are infinitely large. Indeed, taking
the corresponding limits, shows that
X1 1 3
→ − τ 5 , h0 A2 → 8, γ1 → as A → 0 (8.45)
A 5 10
and
X1 1 1
→ − τ 3 , h0 → 2, γ1 → as A → ∞ (8.46)
A 3 6
The asymptotic (8.45) obviously corresponds the nonlinear oscillator, whereas
the limit case (8.46) associates with the harmonic oscillator. Nevertheless,
solution (8.44) is valid for both large and small amplitudes. Both limit cases
follow from equation (8.42). In the case of small amplitudes, |x| 1, by using
the estimate (1 + x2 )−1/2 ∼ 1 − x2 /2 in equation (8.42), we obtain
d2 x/dϕ2 + x3 /2 = 0 (8.47)
In the case of large amplitudes, it follows even from Fig. 8.3 that the distance
between the spring fixed ends becomes negligible if as compared to l. As
a result, the mechanical model becomes effectively close to a mass-spring
oscillator of mass m with a single spring of stiffness 2k. In terms of the
differential equations of motion, we also obtain the corresponding limit from
exact equation (8.42) by√assuming that, during ‘most of the time of vibration
cycle,’ |x| 1 and thus 1 + x2 ∼ |x|. As a result, equation (8.42) is replaced
by
d2 x/dϕ2 + x − sgn(x) = 0 (8.48)
where the term sgn(x) has to be neglected due to the same condition |x| 1
that gives the standard linear oscillator.
Alternatively, the discontinuous term in equation (8.48) can be saved and
then considered as a perturbation of the harmonic oscillator. Note that equa-
tion (8.48) admits another form as follows
d2 x/dϕ2 + sgn(x)(|x| − 1) = 0 (8.49)
The restoring force characteristic of oscillator (8.49) represents a particular

case of the characteristic, p(x) = sgn(x)f (|x|), which is considered later in
this chapter.
8.3.4 Strongly Non-linear Characteristic with a

Step-Wise Discontinuity at Zero
Let us consider the case of symmetric exponentially growing restoring force
characteristic with a step-wise discontinuity at zero such that

exp(x) for x > 0
f (x) = (8.50)
− exp(−x) for x < 0
Although force (8.50) has no certain value at the point x = 0, this still can
play the role of equilibrium position. From the physical standpoint, this is
equilibrium of a small bead at the bottom of V -shaped potential well. The
local dynamics in a small neighborhood of such type of equilibria is considered
later in this chapter; see the text to Fig. 8.13.
It follows from (8.50) that f (−x) = −f (x). Therefore, periodic motions of
the corresponding oscillator can be described by the function x =
X(τ ), where X(−τ ) = −X(τ ) and τ = τ (t/a). In terms of these NSTT
variables, the oscillator of a unit mass is described by the boundary value
problem
X + h exp(X) = 0, X |τ =1 = 0 for τ ∈ (0, 1]

(8.51)
X − h exp(−X) = 0, X |τ =−1 = 0 for τ ∈ [−1, 0)
where h = a2 .
This problem is exactly solvable, and solution that satisfies the continuity
of state condition at τ = 0 has the form
X± (τ ) ≡ Aτ ± 2 ln (1 + exp(−A)) (8.52)

h
∓2 ln 1 + exp(±Aτ − A)
2h0
where X+ and X− are taken for positive and negative subintervals of τ ,

respectively, and
2A2
h = 2h0 = (8.53)
exp(A)[1 + exp(−A)]2
Note that both the differential equation of oscillator and its solution admit
unit form representations as, respectively,
ẍ + sgn(x) exp(|x|) = 0 (8.54)

and

1 + exp(−A)
X = sgn(τ ) A|τ | + 2 ln (8.55)
1 + (h/(2h0 )) exp(A|τ | − A)
The parameter h is obtained from equation
X |τ =1 = 0 (8.56)
or
−1
h h
1− 1+ =0 (8.57)
h0 2h0
This exactly solvable case can play the role of a majorant for evaluation of
convergence properties of successive approximations. For that reason, we in-
troduce a formal ‘small’ parameter, ε = 1, and represent solution of equation
(8.57) in the form
ε −1
h = 2h0 = εh0 1 − (8.58)
2
Taking into account (8.58), brings (8.55) to the form

1 − (ε/2) (1 − exp(−A))
X = sgn(τ ) A|τ | + 2 ln (8.59)
1 − (ε/2) (1 − exp(A|τ | − A))
It can be shown by direct calculations that the power series expansions of

(8.58) and (8.59) with respect to ε lead to the same series as those obtained
by means of the iterative procedure introduced in this section for a general
one-degree-of-freedom oscillator. Moreover, the structure of expression (8.58)
suggests that considering the modified series,
εh0
h= (8.60)
1 − ελ1 − ε2 λ2 − ...
leads to the exact value h already on the second step of the procedure. This
fact can be employed for other cases in order to improve efficiency of the
successive approximation series (8.16). For instance, according to the idea of
Padè transform [19], the following equality must hold in every order of ε
εh0
= εh0 (1 + εγ1 + ε2 γ2 − ...) (8.61)
1 − ελ1 − ε2 λ2 − ...
This is equivalent to
(1 + εγ1 + ε2 γ2 − ...)(1 − ελ1 − ε2 λ2 − ...) = 1 (8.62)
Taking the product of series on the left-hand side of (8.62) and consider-
ing different orders of ε, generates a sequence of equations for the coefficients
λ1 , λ2 , ... . Then, substituting the corresponding solutions in (8.60), gives a

particular case of Padè transform of (8.16) in the form
εh0
h= (8.63)
1 − εγ1 − ε2 (γ2 − γ12 ) − ...
In many cases, expansion (8.63) appears to be more effective than (8.16). Note
that it is also possible to organize the successive approximations procedure
by using expansion (8.60) instead of (8.16).
8.3.5 A Generalized Case of Odd Characteristics

This subsection deals with some generalization of the standard one-degree-
of-freedom conservative oscillator
ẍ + f (x) = 0 (8.64)
where f (x) is a smooth odd characteristic,
f (−x) = −f (x) (8.65)
It was shown in this chapter that periodic solutions of the oscillator (8.64)
admit the form x = X(τ ), where
X(−τ ) = −X(τ ) (8.66)
and, in addition, X(τ )τ ≥ 0 for −1 ≤ τ ≤ 1.

We consider now the following class of oscillators
ẍ + sgn(x)f (|x|) = 0 (8.67)
In the case of odd characteristic, f (x), oscillator (8.67) is equivalent to the

original one (8.64). The extension is due to the fact that the oscillator (8.67)
always has an odd characteristic regardless whether or not the function f (x)
itself is odd. In general case, however, the characteristic sgn(x)f (|x|) may
appear to be nonsmooth at the equilibrium point x = 0. As a result, direct
implementation of iterative procedures with high-order derivatives of oscilla-
tor’ characteristics becomes quite limited. Nevertheless, as illustrated below,
the group properties of equation (8.67) can help to effectively build solution
of equation (8.67) based on solution of equation ẍ + f (x) = 0 for x > 0 by
ignoring the point x = 0.
Obviously, if P (x) is the potential energy of oscillator (8.64), then P (|x|) is
the potential energy corresponding to oscillator (8.67). The following example
explains why equation (8.67) covers a broader class of oscillators than (8.64).
Example 13. ẍ+sgn(x)|x|3/2 = 0 is an oscillator, but ẍ + x3/2 = 0 is not; see
also Chapter 3 for the related discussion.
Based on the transition from equation (8.64) to equation (8.67) and the
general symmetry properties (8.65) and (8.66), we introduce the following
representation for periodic solutions of equation (8.67)
x = sgn(τ )X(|τ |) (8.68)
Such an extension enables one to obtain ‘closed form’ analytical solutions

for a large number of oscillators by a simple adaptation of already known
solutions, X(τ ), for different cases of smooth characteristics.
Example 14. Applying transformation (8.68) to solution (8.39), which was

derived for the power form characteristic xα with an odd positive exponent
α = m, gives

2α+3
|τ |α+2 α |τ | |τ |α+2
X = Asgn(τ ) |τ | − + − (8.69)
α+2 2(α + 2) 2α + 3 α+2
where τ = τ (t/a) and the expansion (8.40) for a2 requires only the replace-
ment m → α. Expansion (8.69) represents an approximate solution of the
equation
ẍ + sgn(x)|x|α = 0 (8.70)
where the notation α substitutes m in order to emphasize that the new expo-
nent can take any positive real value, such as even, odd, rational or irrational.
Figs. 8.4 and 8.5 illustrate solution (8.69) compared to numerical solution for
two different exponents α and the same parameter A = 1. As both figures
show, the analytical and numerical solutions are in a better match under the
large exponent α due to the influence of vibroimpact asymptotic, α → ∞.
0.6
3 2
0.4
0.2
x 0.0
0.2
0.4
0.6
0 1 2 3 4 5 6 7
t
Fig. 8.4 Analytical and numerical solutions of the modified oscillator shown by
continuous and dashed lines respectively.
8.4 Periodic Motions Close to Separatrix Loop 211
1.0
2010
0.5
x 0.0
0.5
1.0
0 5 10 15 20 25 30 35
t
Fig. 8.5 Analytical and numerical solutions of the generalized oscillator under the
large exponent α.
The common feature of the algorithms and examples of this section is that
generating solutions for successive approximations are represented by trian-
gular sine waves of proper amplitudes and periods. Such generating solutions
belong to the ‘real’ component of the hyperbolic ‘number,’ x = X + Y e. In
contrast, the next section introduces algorithms of successive approximations
based on the ‘imaginary’ component. It will be seen that these two approaches
have different physical contents.
8.4 Periodic Motions Close to Separatrix Loop
In this section, the classic mathematical pendulum is considered as an ex-

ample, although the developed algorithm may be applicable to other cases
of one degree-of-freedom systems with multiple equilibrium positions. So we
illustrate the algorithm based on the differential equation of motion
ẍ + sin x = 0 (8.71)
It is assumed that the pendulum oscillates inside the separatrix loop around
the stable equilibrium (x, ẋ) = (0, 0) in between two physically identical
unstable saddle-points (x, ẋ) = (±π, 0). The separatrix loop also represents
trajectory of the system with the total energy
π
Es ≡ sin xdx = 2 (8.72)
0
The pendulum remains close to the separatrix loop in the area of periodic
motions if the total energy E is within the range
E
0<1− << 1 (8.73)
Es
Let us show that, under condition (8.73), a successive approximation solution
can be derived from the particular case of boundary value problem (8.7) and
(8.8). Such particular case is given by setting X ≡ 0 so that
x(t) = Y (τ (t/a))e(t/a) (8.74)
Substituting (8.74) in equation (8.71) and following the NSTT procedure,

yields
Y = −a2 sin Y (8.75)
and
Y |τ =1 = 0, Y (−τ ) ≡ Y (τ ) (8.76)
We seek solution of the boundary value problem, (8.75) and (8.76), in the
form of series of successive approximations
Y = π + εY1 (τ ) + ε3 Y3 (τ ) + ε5 Y5 (τ ) + ... (8.77)

a2 = p2 / 1 − ε2 λ2 − ε4 λ4 − ... (8.78)
where ε = 1 is an auxiliary “parameter” that helps to organize the iterative
process.
According to the formal expansion (8.77), the generating solution is rep-
resented by the rectangular cosine of the amplitude π, x(t) = πe(t/a), which
is a step-wise discontinuous function. Such temporal mode shapes occur near
the separatrix loop in the natural time scale of the pendulum because the
system spends most of the time during one period near the unstable equi-
librium positions x = π and its physically identical, x = −π. Therefore,
expansion (8.77) is designed to be a high-energy expansion near the unstable
equilibrium, rather than around the stable equilibrium position, x = 0.
Substituting expansions (8.77) and (8.78) into the equation (8.75) and
collecting terms with the same power of ε, leads to the sequence of equations
d2 Y1
− p2 Y1 = 0 (8.79)
dτ 2

d2 Y3 1 3
− p Y3 = p λ2 Y1 − Y1
2 2
(8.80)
dτ 2 6

d2 Y5 2 λ2 3 1 5 1 2
− p 2
Y5 = p 2
λ2 + λ4 Y1 − Y + Y + λ Y
2 3 − Y Y3 (8.81)
dτ 2 6 1 120 1 2 1
...
8.4 Periodic Motions Close to Separatrix Loop 213
Further, the family even solutions of equation (8.79) can be represented in

the form
cosh pτ
Y1 = −A (8.82)
cosh p
where A is an arbitrary constant accompanied by the factor − cosh−1 p, which
is convenient for further calculations due to the relationship Y1 (1) = −A.
In particular, this provides the same order of magnitude for the arbitrary
constant A as the parameter p and period T = 4a both go to infinity.
Further procedure is formally similar to the standard Poincare-Lindstedt
algorithm for nonlinear conservative oscillators with positive linear stiffness.
For example, substituting (8.82) into the right part of the equation (8.80),
gives a ‘resonance term’ on the right-hand side of the equation, which is pro-
portional to cosh pτ . This generates ‘hyperbolic secular terms’ of the form
τ cosh pτ and τ cosh pτ in the particular solution of equation (8.80). Occur-
rence of such terms can be prevented, however, analogously to the Poincare-
Lindstedt method by setting
A2
λ2 = (8.83)
8 cosh2 p
As a result, the particular solution of equation (8.80) takes the form
A3 cosh 3pτ
Y3 = (8.84)
192 cosh3 p
At the next stage, equation (8.81) gives solution

A5 1
Y5 = cosh 3pτ − cosh 5pτ (8.85)
4096 cosh5 p 5
under the condition
3A4
λ4 = − (8.86)
512 cosh4 p
Substituting (8.82) through (8.83) in (8.77) and (8.78), and setting ε = 1,
gives approximate solution
A cosh pτ A3 cosh 3pτ

Y =π− + (8.87)
cosh p 192 cosh3 p

A5 1
+ cosh 3pτ − cosh 5pτ
4096 cosh5 p 5
and
−1
A2 3A4
h=p 1−2
+ (8.88)
8 cosh p 512 cosh4 p
2
√
where τ = τ (t/a) and a = h.
The truncated series of successive approximations (8.87) and (8.88) de-

pend upon two parameters, A and p, coupled by the boundary (continuity)
condition (8.76) as follows

A3 cosh 3p A5 1
A=π+ + cosh 3p − cosh 5p (8.89)
192 cosh3 p 4096 cosh5 p 5
Equation (8.89) should be interpreted as implicit function A = A(p) near

the point A = π. Therefore, expansions (8.87) and (8.88) represent a one-
parameter family of periodic solutions with the parameter p.
x 0
1
2
0 5 10 15
t
Fig. 8.6 Analytical and numerical solutions of the period T=8.5 (p=2) shown by
solid and thin lines, respectively.
Note that the equation (8.71) admits the group of time translations. As
a result, another arbitrary parameter, say t0 , is introduced by substitution
t → t + t0 .
Figures 8.6 and 8.7 show that the analytical and numerical solutions are
matching better for larger periods as the system trajectory becomes closer to
the separatrix loop.
8.5 Self-excited Oscillator

This section illustrates the case when both X and Y components of solutions
participate in the iterative process.
In particular, we consider periodic self-sustained vibrations described the
differential equation of motion
ẍ + g(x)ẋ + f (x) = 0 (8.90)

8.5 Self-excited Oscillator 215
x 0
1
2
3
0 10 20 30 40
t
Fig. 8.7 Analytical and numerical solutions of the period T=24.5 (p=6) shown by
the solid and thin lines, respectively.
where f (x) and g(x) are analytic functions, such that (Lienard’ conditions)
"x
a) G(x) = g(u)du is an odd function such that G(0) = G(±μ) = 0 for some
0
μ > 0,
b) G(x) → ∞ if x → ∞, and G(x) is a monotonously increasing function
for x > μ,
c) f (x) is an odd function such that f (x) > 0 for x > 0.
The above conditions guarantee that system (8.90) has a single stable limit
cycle. In this case, the boundary-value problem (8.7) and (8.8) takes the form
X = −a2 Rf − a (Rg Y + Ig X ) ≡ −εFX , X |τ =±1 = 0 (8.91)

Y = −a2 If − a (Ig Y + Rg X ) ≡ −εFY , Y |τ =±1 = 0
where the period of limit cycle T = 4a is unknown, expressions Rf and If

as well as Rg and Ig are obtained by applying (8.5) and (8.6) to each of the
functions f (x) and g(x), and notations εFX and εFY are introduced with the
formal factor ε = 1 for further convenience.
We seek solution of the boundary value problem (8.91) in the form of series
of successive approximations
X = X0 (τ ) + εX1 (τ ) + ε2 X2 (τ ) + · · · (8.92)
Y = Y0 (τ ) + εY1 (τ ) + ε Y2 (τ ) + · · ·
2
a = q0 + εq1 + ε2 q2 + · · · (8.93)
Further solution procedure can be simplified by taking into account the sym-
metry properties X(−τ ) ≡ −X(τ ) and Y (−τ ) ≡ Y (τ ) due to the above
conditions (a) through (c). Substituting (8.92) and (8.93) into (8.91) and
matching the coefficients of the respective powers of ε, gives the following
sequence of boundary value problems
X0 = 0 (8.94)
Y0 = 0, Y0 |τ =1 = 0
X1 = −FX,0 , (X0 + X1 ) |τ =1 = 0

Y1 = −FY,0 , Y1 |τ =1 = 0 (8.95)

Xi+1 = −FX,i , Xi+1 |τ =1 = 0

Yi+1 = −FY,i , Yi+1 |τ =1 = 0 (8.96)
(i = 1, 2, ...)
where
1 di FX 1 di FY
FX,i = |ε=0 , FY,i = |ε=0 (8.97)
i! dεi i! dεi
Note that zero-order and first-order approximations are coupled through the
boundary condition for X-component in (8.95), whereas no boundary condi-
tion is imposed on X0 in (8.94). This specific represents a formalization of
the physical assumption regarding the dominating component in the tempo-
ral mode shape of vibration, which is assumed to be close to the sawtooth
sine rather than rectangular cosine. As a result, the generating system (8.94)
gives solution
X0 = Aτ , Y0 ≡ 0 (8.98)
where A is an arbitrary constant.
Substituting (8.98) in the right-hand side of equations (8.95) and integrat-
ing, yields
τ τ
X1 = −q02 (τ − ξ)f (Aξ)dξ, Y1 = −Aq0 (τ − ξ)g(Aξ)dξ (8.99)
0 0
Then, substituting (8.99) in the boundary conditions in (8.95), gives the

following two equations for parameters q0 and A
1 1
q02 f (Aξ)dξ = A, (1 − ξ)g(Aξ)dξ = 0 (8.100)
0 0
8.5 Self-excited Oscillator 217
Relationships (8.98) through (8.100) complete first basic steps of the iterative
procedure. Further iterations are organized then in a similar way as follows
τ τ ζ
Xi+1 = − (τ − ξ)FX,i (ξ)dξ, Yi+1 = − dζ FY,i (ξ)dξ (8.101)
0 1 0
1
FX,i (ξ)dξ = 0; i = 1, 2, .... (8.102)
0
Note that the boundary conditions for Yi+1 are satisfied automatically due
to the lower limit of the outer integral in (8.101), whereas the boundary con-
ditions for Xi+1 generates equations (8.102) for determining the coefficients
of series (8.93). Practically, high-order approximations can be obtained by
using computer systems of symbolic manipulations.
Example 15. Consider the self-excited oscillator with the power form stiffness
of the degree m = 3,
ẍ + (bx2 − 1)ẋ + x3 = 0
In this case, g(x) = bx2 − 1 and f (x) = x3 . Conducting elementary integra-
tions in (8.100), gives the algebraic system
1 2 3 1
q A = A, q0 A 6 − bA2 = 0
4 0 12
with non-trivial solution

2b 6
q0 = , A=
3 b
As a result, integrating (8.99), yields

6 τ5
X1 = − , Y1 = τ 2 − τ 4
b 5
All further steps of the procedure are conducted according to the same scheme
(8.101) and (8.102). For instance, first two steps of the procedure give ap-
proximate solution

6 τ5 1
x= {τ − + [105τ 9 + 900τ 7 b − 21τ 5 (70b + 9) + 350τ 3 b]}
b 5 3150
1
+(1 − τ 2 ){τ 2 − [20 − 43τ 2 + 20τ 4 + 216τ 6 ]}e
420
and the period

b 3
T = 4a = 8 1+
6 20
Two more steps of the procedure correct the above expression for the period
as follows

b 3 2960b + 2121 7367360b2 + 4554992b + 8659035
T = 4a = 8 1+ + +
6 20 50400 605404800
Figs.8.8 and 8.9 show limit cycle trajectories described by the analytical so-
lutions in one and two iterations, respectively. For comparison, the numerical
solution for transition to the limit cycle is also presented. Then, Fig. 8.10
illustrates dependence of the quarter of period parameter, a = T /4, versus
the quantity b−1/2 , which can be viewed as an estimate for the amplitude of
limit cycle.
v
3 b 1.0
m3
2
0 x
1 analytical, 1 iteration
numerical
2
3
2 1 0 1 2
Fig. 8.8 Trajectories of numerical solution and the one iteration analytical limit-
cycle solution.
8.6 Strongly Nonlinear Oscillator with Viscous

Damping
This section describes the successive approximation procedure combined with
the asymptotic of small energy dissipation that leads to a slow amplitude
decay. The scheme of the algorithm is closed to that was introduced earlier
in [137].
8.6 Strongly Nonlinear Oscillator with Viscous Damping 219
v
3 b 1.0
m3
0 x
analytical, 2 iterations
1 numerical
2
3
1.5 1.0 0.5 0.0 0.5 1.0 1.5
Fig. 8.9 Trajectories of numerical solution and approximate (two iterations) ana-
lytical limit-cycle solution.
a
analytical, four iterations
2
numerical estimates
1
0
0 2 4 6 8
12
b
Fig. 8.10 Illustration of convergence of the iterative procedure on the parameter
plane.
So consider a strongly nonlinear oscillator under the viscous damping
ẍ + 2μẋ + f (x) = 0 (8.103)
where f (x) is an odd function such that xf (x) ≥ 0, and 0 < μ << 1.
The idea of two variable expansions will be employed below in combination
with the sawtooth time substitution. Let us assume that τ (ϕ) is a ‘fast’ time
scale, whose phase ϕ depends on the ‘slow’ time scale t0 = μt according to
the following differential equation
ϕ̇ = ω(t0 ) (8.104)
where the right-hand side is a priory unknown.

Let us represent unknown solution of equation (8.103) in the form
x = x(ϕ, t0 ) = X(τ (ϕ), t0 ) + Y (τ (ϕ), t0 )e(ϕ) (8.105)
Substituting (8.105) in equation (8.103), and imposing ‘smoothness condi-

tions,’
∂X
Y |τ =±1 = 0, |τ =±1 = 0 (8.106)
∂τ
gives two partial differential equations
∂2X ∂Y
ω2 2
= −Rf − μH − μ2 LX
∂τ ∂τ
∂2Y ∂X
ω 2 2 = −If − μH − μ2 LY (8.107)
∂τ ∂τ
where, as follows from (8.5) and (8.6),
1
Rf = Rf (X, Y ) = [f (X + Y ) + f (X − Y )]
2
1
If = If (X, Y ) = [f (X + Y ) − f (X − Y )]
2
and two linear differential operators are introduced

∂ dω
H ≡ 2ω 1 + 0 + 0 (8.108)
∂t dt
∂2 ∂
L ≡ 02 + 2 0
∂t ∂t
Note that the fast and slow temporal scales are associated with different
physical processes developed in the system. The slow energy dissipation pro-
cess is represented by the explicit small parameter μ, but there is no explicit
parameter associated with perturbations of the triangular sine wave, which
is supposed to be a generating solution of the iterative process. However, as
discussed earlier in this section, introducing the sawtooth temporal argument,

implies that the entire right-hand side of system (8.107) is small. Otherwise,
the triangular sine wave cannot be considered as a dominating component of
the temporal mode shape of oscillations. Recall that, in a similar way, select-
ing the harmonic wave as a dominating solution in quasi harmonic approaches
implies that nonlinearities are small regardless system’ parameters. Therefore,
iterative procedure for boundary value problem (8.106) and (8.107) should
incorporate two different procedures, as those described above in this sec-
tion, and a proper asymptotic procedure related to the dissipation process.
Once again, the quasi-harmonic methods face similar situation when dealing
with weakly nonlinear systems under small damping conditions. For instance,
if being applied to such cases, the method of multiple scales accounts for
both unharmonicity and dissipation, after appropriate assumption regarding
the relation between non-linearity and damping parameters has been made.
Very often though, such parameters are assumed to be of the same order of
magnitude. As to the boundary value problem (8.106) and (8.107), similar
assumption can be introduced by providing the terms Rf and If , and the
parameter μ with the same formal ‘small factor’ ε = 1. Then, the multiple
scales or two variables expansions can be organized by using the auxiliary
parameter ε [137]. In the case of linear oscillator, such an algorithm recovers
the exact solution of the linear differential equation of motion however in the
specific form

ω exp(−t0 ) ε(1 − εμ2 )
x = X(τ, t ) = C
0
sin τ (ωt) (8.109)
ε(1 − εμ2 ) ω
and
ε(1 − εμ2 )
ω2 =
4 arcsin2 ε/2
where C is an arbitrary constant, and another arbitrary constant can be
introduced through the arbitrary time shift.
Note that solution (8.109) includes no Y -component because, at every
stage of the iterative process, it appears to be possible to satisfy condition
H(∂X/∂τ ) = 0 (8.110)
Therefore, the second equation of system (8.107) is satisfied by setting Y ≡ 0.

Practically, condition (8.110) generates the common factor exp(−t0 ) for all
successive approximations.
In general nonlinear case, however, it is rather impossible to satisfy condi-
tion (8.110) at every stage of the process, but it still works for leading order
approximate solutions.
Example 16. Consider the weakly damped oscillator of the m degree power
form restoring force characteristic
ẍ + 2μẋ + xm = 0 (8.111)
At this stage, the exponent m is an odd positive number. (It will be shown
later that a broader class of power characteristics can also be considered.)
Note that, for this kind of oscillators, whether or not the damping is small
depends on the level of amplitude and the exponent m. This is due to the
fact that, under small enough amplitudes, the elastic force becomes negli-
gible regardless the magnitude of damping. By assuming that the influence
of damping is negligible during one cycle of vibration, one can use expres-
sions (8.40) for estimations of magnitudes of damping and elastic forces. As
a result, the condition of ‘small damping’ derives in the form
1
μ2 (m + 1)Am−1 (8.112)
4
One step of the procedure gives approximate solution [137]

−4μt τ m+2
x = Cexp τ− (8.113)
m+3 m+2
where τ = τ (ϕ) and the phase variable is approximated by

m−1
ϕ = ϕ∞ 1 − exp −2μ t (8.114)
m+3
1 m + 3 C (m−1)/2
ϕ∞ =
2μ m − 1 2(m + 1)
Interestingly enough, the above approximate solution predicts that the oscil-
lator makes only a finite number of waves as m > 1 and t → ∞.
Figs. 8.11 and 8.12 illustrate damped responses of the oscillator with two
different degrees of nonlinearity, m = 3 and m = 7, respectively. As follows
from the diagrams, the approximate analytical solution and numerical one
are matching relatively well, especially at higher exponent, m = 7. In partic-
ular, this justifies the idea of using the sawtooth wave in strongly nonlinear
cases, when the oscillator becomes close to the standard vibroimpact model.
8.6.1 Remark on NSTT Combined with Two

Variables Expansion
In general, the iterative process of sawtooth expansions and the averaging
procedure can be separated. Moreover, the stage of sawtooth time substitu-
tion does not impose any specific method of analyses. So let us apply the two
variables method directly to the nonlinear boundary value problem (8.106)
and (8.107) by means of the asymptotic series
X = X0 (τ, t0 ) + μX1 (τ, t0 ) + μ2 X2 (τ, t0 ) + · · ·

Y = Y0 (τ, t0 ) + μY1 (τ, t0 ) + μ2 Y2 (τ, t0 ) + · · · (8.115)
Fig. 8.11 Approximate analytical and numerical solutions of the damped oscillator
with cubic power form characteristic.
Fig. 8.12 Approximate analytical and numerical solutions of the damped oscillator
with the seven-th degree power form characteristic.
and
ω = ω0 (t0 ) + μω1 (t0 ) + μ2 ω2 (t0 ) + · · ·

H = H0 + μH1 + μ2 H2 + ... (8.116)

where Hi = 2ωi 1 + ∂/∂t0 + dωi /dt0 .
Substituting (8.115) and (8.116) into (8.106) and (8.107) and matching
coefficients of like powers of μ, gives, in particular,
∂ 2 X0 (0) ∂X0
ω02 = −Rf (X0 , Y0 ), |τ =±1 = 0
∂τ 2 ∂τ
2
∂ Y0 (0)
ω02 = −If (X0 , Y0 ), Y0 |τ =±1 = 0 (8.117)
∂τ 2
and
∂ 2 X1 (1) ∂ 2 X0 ∂Y0
ω02 = −Rf (X 0 , Y0 , X 1 , Y1 ) − 2ω 0 ω 1 − H0
∂τ 2 ∂τ 2 ∂τ
∂X1
|τ =±1 = 0
∂τ
∂ 2 Y1 (1) ∂ 2 Y0 ∂X0
ω02 2
= −If (X0 , Y0 , X1 , Y1 ) − 2ω0 ω1 2
− H0 (8.118)
∂τ ∂τ ∂τ
Y1 |τ =±1 = 0
(0) (1) (0) (1)
where Rf , Rf , ... and If ,If , ... are determined by the expansions
(0) (1) (2)

R = Rf + μRf + μ2 Rf + ...
(0) (1) (2)
I = If + μIf + μ2 If + ...
By taking into account the assumptions on f (x) in equation (8.103), one can
represent solution of problem (8.117) in the following general form
X0 = X0 (τ, A, ω0 ), Y0 ≡ 0 (8.119)
where A = A(t0 ) is an arbitrary function of the slow time scale, which is

coupled with the frequency ω0 through the boundary condition
∂X0 (τ, A, ω0 )
|τ =1 = 0 (8.120)
∂τ
In general, this relationship determines the implicit function ω0 = ω0 (t0 ).
Now substituting solution (8.119) in (8.118), gives equations
∂ 2 X1 ∂ 2 X0
ω02 + f (X 0 )X 1 = −2ω 0 ω 1 (8.121)
∂τ 2 ∂τ 2
and
∂ 2 Y1 ∂X0
ω022
+ f (X0 )Y1 = −H0 (8.122)
∂τ ∂τ
Let us consider equation (8.122). The best choice would be achieved by setting
the right-hand side to zero and therefore making possible the solution Y1 ≡ 0
which is consistent with zero-order solution (8.119) and provides a better
smoothness property of the corresponding solution at this stage; see condition
(8.110). Note that the right-hand side cannot be always made zero for any
τ unless the zero-order solution admits separation of the variables t0 and
τ. However, it is still possible to ‘minimize’ the right-hand side of equation
(8.122) by making it orthogonal to solution of the corresponding homogeneous
equation, ∂X0 /∂τ , in other words,
1 . /
1 ∂X0 ∂X0 ∂X0 ∂X0
H0 dτ ≡ H0 =0 (8.123)
2 ∂τ ∂τ ∂τ ∂τ
−1

Taking into account the expression H0 = 2ω0 1 + ∂/∂t0 + dω0 /dt0 and
condition (8.123), gives
0
2 1
∂X0
ω0 = C exp(−2t0 ) (8.124)
∂τ
where C is an arbitrary constant.

It can be shown that the ‘minimization condition’ (8.123) occurs also in a
rigorous mathematical way based on the boundary conditions for Y1 ; at least,
this can be easily verified in the linear case f (x) ≡ x.
8.6.2 Oscillator with Two Nonsmooth Limits

Consider the following generalization of equation (8.111)
ẍ + 2μẋ + sgn(x)|x|α = 0 (8.125)
where α is a non-negative real number; see the comments to equation (8.70).

In this case, zero-order solution (8.119) can be obtained in the form (8.69),
x(t) = A(t0 )sgn(τ (ϕ)) (8.126)

|τ (ϕ)|α+2 α |τ (ϕ)|2α+3 |τ (ϕ)|α+2
× |τ (ϕ)| − + −
α+2 2(α + 2) 2α + 3 α+2
ϕ̇(t) = ω0 (t0 ), t0 = μt
where the functions A(t0 ) and ω0 (t0 ) are coupled by relation (8.40) as follows
−1/2
1 A(α−1)/2 α (α + 1)2
ω0 = = √ 1+ 1+ (8.127)
a α+1 2(α + 2) (α + 2)(2α + 3)
Equations (8.127) and (8.124) admit exact solution

4μt α−1
A = C exp − , ϕ = ϕ∞ 1 − exp −2μ t (8.128)
3+α α+3
where C is a new arbitrary constant and

1 α+3 C (α−1)/2 (2 + α) 2(3 + 2α)
ϕ∞ = (8.129)
2μ α − 1 (α + 1) (7α3 + 31α2 + 47α + 24)
It follows from expressions (8.128) and (8.129) that the linear system α = 1
plays the role of a boundary between the two strongly nonlinear areas
N0 = {α : 0 ≤ α << 1} and N∞ = {α : 1 << α < ∞} (8.130)
In other words, we show that α = 1 separates two qualitatively different

regions of the dynamics determined by the influence of different nonsmooth
limits of the potential well; see Fig. 8.13 for illustration. In particular, if α > 1
then the phase variable ϕ has the finite limit ϕ∞ as t → ∞. In contrast, if
α < 1 then the phase with its temporal rate are exponentially growing, as
the amplitude decays and the system approaches the bottom of the potential
well. The physical meaning of this effect is most clear from the limit case
α → 0, which is discussed below.
Figs. 8.14, 8.15 and 8.16, 8.17 illustrate solution (8.126) through (8.129)
for large and small exponents α, respectively. The diagrams suggest quite a
good match with numerical solution in both branches of the exponent (8.130).
The numerical solutions shown by dashed lines were obtained by the stan-
R
dard solver NDSolve built in Mathematica . Fig. 8.14 also shows that some
divergence between the curves occurs when the amplitude is decreased to the
level about A = 0.6. Below this level, the condition of small damping (8.112)
is not guaranteed any more. In contrast, the curves are in a better match
for smaller amplitudes if α < 1, see Fig. 8.16. In this case, the amplitude
decay just strengthens condition (8.112). The phase plane diagrams shown
in Figs. 8.15 and 8.17 have qualitatively different shapes as dictated by the
influence of different nonsmooth limits of the potential well, see Fig. 8.13. Let
us show now that solution (8.126) captures both nonsmooth limits α → 0
and α → ∞.
For a physically meaningful transition to the limits, let us express the
arbitrary parameter C through the initial velocity v0 = ẋ|t=0 ,
1/(α+1)
v02 (α + 1) 7α3 + 31α2 + 47α + 24
C=
2(α + 2)2 (2α + 3)
and consider the two different cases.

1) As α → ∞, the solution (8.126) through (8.129) gives
x = τ (ϕ) (8.131)
v0
ϕ= [1 − exp (−2μt)]
2μ
Solution (8.131) exactly describes the system motion in the square potential
well.
2) When α → 0, expressions (8.126) through (8.129) are reduced to

4 |τ (ϕ)|
x = v0 exp − μt τ (ϕ) 1 −
2
(8.132)
3 2

3 2
ϕ= exp μt − 1
2μv0 3
where the identity sgn[τ (ϕ)]|τ (ϕ)| ≡ τ (ϕ) has been taken into account.
If, in addition μ = 0, then solution (8.132) also exactly describes the
system dynamics with another nonsmooth limit of the potential energy, |x|,
as shown in Fig. 8.13. However, if μ = 0 then substituting solution (8.132)
into the differential equation of motion gives an error O(μ2 ). In terms of first-
order asymptotic solutions, the error of order μ occurs on the time period of
order 1/μ. Therefore, solution (8.132) exactly captures the carrying shape of
the vibration, but gives only asymptotic estimate for the exponential decay.
Fig. 8.13 Potential energy representation for the two limit oscillators.
Fig. 8.14 Temporal mode shape of the vibration for α ∈ N∞ , C = 1.5 and μ =
0.04; here and below, the dashed line represents numerical solutions.
v 0
2
4
1.0 0.5 0.0 0.5 1.0
x
Fig. 8.15 Phase plane diagram for α ∈ N∞ .
Note that the error of solution (8.126) is due to the error of the iterative
procedure for elastic vibrations and the error of asymptotic for energy dissi-
pation. As shown above, the error of successive approximations vanishes as
either α → ∞ or α → 0, but the error of asymptotic vanishes only as α → ∞.
Finally, let us discuss the qualitative difference of the dynamics in the pa-
rameter intervals N0 and N∞ . As follows from equation (8.128), for α ∈ N0 ,
the phase of vibration and the corresponding frequency are exponentially in-
creasing in the slow time scale μt . The physical meaning of this phenomenon
1.0
Α 1 14
analytical, numerical
0.5
x 0.0
0.5
1.0
0 20 40 60 80 100 120 140
t
Fig. 8.16 Temporal mode shape of the vibration for α ∈ N0 , C = 2.5 and μ = 0.04.
Fig. 8.17 Phase plane diagram for α ∈ N0 .
becomes most clear in the limit case α = 0. In this case, according to solution
(8.132), the amplitude and frequency are, respectively

v02 4 1 2 1
A(μt) = exp − μt and ϕ̇ = exp μt = (8.133)
2 3 v0 3 2A(μt)
Expressions (8.133) describe increasingly rapid vibrations -‘dither’- near the

corner of the potential energy |x| as the amplitude approaches zero. This
result is confirmed by the much earlier analysis of the corresponding conser-
vative case [78]. In contrast, when α ∈ N∞ , the oscillator makes a limited
number of cycles such that the phase ϕ remains bounded for any time t.
Again, the most clear interpretation is obtained in the limit case α → ∞,
when, as follows from (8.131), the phase variable ϕ(t) represents the total
distance passed by the particle by time t, and ϕ̇ = v is the absolute value
of the velocity. Since the barriers are perfectly elastic the particle reflects
with no energy loss, the velocity v (t) remains continuous function of time
described by the linear differential equation v̇ + 2μv = 0 or ϕ̈ + 2μϕ̇ = 0.
Under the initial conditions ϕ(0) = 0 and ϕ̇(0) = v0 , one obtains exactly
solution (8.131).
In conclusion, the explicit analytical solution for a class of strongly nonlin-
ear oscillators with viscous damping is introduced. Two different nonsmooth
functions involved into the solution are associated with two different nons-
mooth limits of the oscillator. As a result, the solution is drastically simplified
to give the best match with numerical tests if approaching any of the two
limits.
8.7 Bouncing Ball

In this section, we consider a small ball of mass m falling under the gravity
force onto a horizontal plane. In addition to the gravity, the ball is subjected
to the linear damping with the coefficient c. Impacts with the plane are in-
elastic with the restitution coefficient κ. The vertical coordinate z(t) therefore
is described by the following equations of motion
mz̈ = −mg − cż (z = 0) (8.134)

ż+ = −κż− (z = 0)
where ż− and ż+ are velocities right before and immediately after the impact,
respectively.
Let h be a natural spatial scale of the system. This can be, for instance,
the maximal height that has been reached by the ball during the very first
cycle. Introducing
the coordinate transformation z = h|x| and re-scaling the
time as t = h/gp, brings equations (8.134) to the form [204]
d2 x dx
+ 2μ + sgnx = 0 (8.135)
dp2 dp

dx dx dx
− = (1 − κ)
dp + dp − dp −
where *
1 c h
μ=
2m g
8.7 Bouncing Ball 231
In the particular case κ = 1, solution (8.132) becomes applicable to equation

(8.135). Then, returning back to the original notations of equation (8.134),
gives

2c 1 2
z(t) = C exp − t |τ (ϕ)| − τ (ϕ) (8.136)
3m 2

g 3m c
ϕ(t) = exp t −1
C c 3m
where C is a new arbitrary constant.

If C = ż02 /g then solution (8.136) satisfies the specific initial conditions
z(0) = 0 and ż(0) = ż0 . One more arbitrary constant can be introduced by
shifting the time, t− > t + t0 , that would allow to consider non-zero initial
height of the ball.
In order to compare solution (8.136) with numerical solution, let us repre-
sent equation (8.134) in the form
ż = u
c
u̇ = −g − u (8.137)
m
where u+ = −κu− whenever z = 0.
Further, introduce new unknown state variables {s, v} according to [73]
z = ssgn(s)
u = sgn(s)[1 − ksgn(sv)]v (8.138)
1−κ
k=
1+κ
Applying (8.138) to (8.137), gives1
ṡ = [1 − ksgn(sv)]v (8.139)
c g
v̇ = − v − [sgn(s) + ksgn(v)]
m 1 − k2
As compared to (8.137), system (8.138) automatically accounts for the veloc-
ity jump condition at z = 0. In other words, transformation (8.138) makes the
strong nonlinearity (due to non-elastic impacts) explicitly present in (8.139).
From the standpoint of numerical procedures, such a transformation enables
one of using built-in solvers of different packages with no impact conditioning
at z = 0.
Note that, in the particular case κ = 1 (k = 0), system (8.139) becomes
equivalent to (8.135). In this particular case, the direct numerical integration

R
of equations (8.139), using the NDSolve procedure built in Mathematica
1
Note that differentiation of sgn-functions will produce Dirac’s delta-functions
with effectively zero factors however.
package, gives solution in a perfect match with analytical solution (8.136);

see Fig.8.18.
Let us assume now that the energy loss happens only due to impact in-
teractions of the ball with the plane z = 0. The question is whether or not
solution (8.136) can still be adapted by interpreting the parameter c as some
“effective damping coefficient,” such that the energy loss between two impacts
is equal to that happens in one impact.
0.05
0.04
0.03
analytical, numerical
z
0.02
0.01
0.00
0 1 2 3 4 5 6
t
Fig. 8.18 NSTT analytical and direct numerical solutions for the height of bounc-
ing ball under the linear dissipation condition; m = 1.0, c = 0.5, z(0) = 0 and
ż(0) = 1.0.
Assuming that the damping is small enough and using the classical
parabolic approximation for the ball height during one cycle, z(t) = −gt2 /2+
v0 t, yields
2v0 /g
2cef f v03 1
cef f ż 2 dt = = m(1 − κ2 )v02
0 3g 2
or
3mg
cef f = (1 − κ2 ) (8.140)
4v0
Expression (8.140) shows that, in this case, effective linear damping cannot
be introduced on the entire time interval since the “initial velocity” v0 de-
creases from cycle to cycle. Choosing v0 = ż(0) in (8.140), provides a good
enough match between analytical (8.136) and numerical solutions during first
few cycles of the motion, as follows from Fig. 8.19. Then, the divergence
between the curves accelerates. The result can be improved by using some
8.7 Bouncing Ball 233
0.05
0.04
0.03 analytical
z numerical
0.02
0.01
0.00
0 1 2 3 4 5 6
t
Fig. 8.19 NSTT analytical solution with effective damping corresponding to the
coefficient of restitution κ = 0.97, and direct numerical solutions for the height of
bouncing ball; the parameters are: m = 1.0, c = 0.0, z(0) = 0 and ż(0) = 1.0.
0.05
0.04
0.03 numerical
z analytical
0.02
0.01
0.00
0 1 2 3 4 5 6
t
Fig. 8.20 NSTT analytical solution with effective damping adjusted by “variable
initial velocity,” and direct numerical solutions for the height of bouncing ball; the
parameters are: m = 1.0, c = 0.0, κ = 0.97, z(0) = 0 and ż(0) = 1.0.
estimate for the “initial velocity” decay based on solution (8.136), where the
effective damping is still constant, however. This one-step iteration gives
c
ef f
c̃ef f = cef f exp t (8.141)
3m
Fig. 8.20 shows a better match between numerical and analytical solutions
achieved due to modification (8.141).
8.8 The Kicked Rotor Model

The so-called kicked rotor model is introduced in physics as a relatively sim-
ple essentially nonlinear model for chaotic behavior of systems, where one
variable may be either bounded or unbounded in phase space [95], [20], [71].
The kicked rotor is described, in some units, by the Hamiltonian
∞
1 2
H= I + K cos θ δ (t − n) (8.142)
2 n=−∞
where I is the angular momentum, θ is the conjugate angle, and K is the

stochasticity parameter that determines qualitative features of the dynamics.
The sequence of pulses in (8.142) can be expressed through the sawtooth
sine τ = τ (2t − 1) in the form,
∞
1
δ (t − n) = − τ (2t − 1) τ (2t − 1) (8.143)
n=−∞
2
where primes denotes differentiation with respect to the entire argument of

a function, 2t − 1.
Note that the only role of the first multiplier, τ (2t − 1), on the right-hand
side is to provide pulses with the same sign.
Therefore, (8.142) takes the form
1 2 1
H= I − Kτ (2t − 1) τ (2t − 1) cos θ (8.144)
2 2
The corresponding differential equation of motion is
1
θ̈ = − Kτ τ sin θ (8.145)
2
We seek a family of solutions with the period T = 2 by introducing the
sawtooth time argument τ ,
θ = θ(τ ) (8.146)

d2 θ 1 dθ
= − Kτ sin θ + τ
dτ 2 8 dτ
Eliminating the singular term τ , leads to the boundary condition

dθ 1
|τ =±1 = − Kτ sin θ|τ =±1 (8.147)
dτ 8
8.9 Oscillators with Piece-Wise Nonlinear Restoring Force Characteristics 235
and the differential equation

d2 θ
=0 (8.148)
dτ 2
The boundary-value problem (8.147) and (8.148) admits solution
θ = Aτ (2t − 1) + B (8.149)
where A and B appear to be coupled by the set of equations

1
A = − K sin(A ± B) (8.150)
8
or
1
A = − K sin A cos B (8.151)
8
and
cos A sin B = 0 (8.152)
In particular, equation (8.151) show that the number of periodic solutions of
the period T = 2 is growing as the parameter K increases.
8.9 Oscillators with Piece-Wise Nonlinear Restoring

Force Characteristics
Consider the case of asymmetric restoring force characteristics described by
two pieces of smooth monotone increasing functions f (x) and g(x) of which
one or both may be nonlinear (Fig. 8.21)

f (x), −∞ < x < x1
F (x) = (8.153)
g(x), x1 < x < ∞
It is assumed that the entire characteristic F (x) is at least continuous in the

interval −∞ < x < ∞, in other words, the following matching condition
holds
f (x1 ) = g(x1 ) (8.154)
and f (0) = 0.
Expression (8.153) admits the unit form
F (x) = f (x) + [g(x) − f (x)]H(x − x1 ) (8.155)
where H indicates the unit-step Heaviside function.

Let us outline analytical procedure for periodic solutions of equation
ẍ + F (x) = 0 (8.156)
Fig. 8.21 Asymmetric piece-wise nonlinear restoring force characteristic.
Introducing the triangular wave temporal argument τ = τ (t/a) with un-

known scaling parameter a and making the substitution x = X(τ ) in equation
(8.156), gives boundary value problem
X + hF (X) = 0 (8.157)
X |τ =±1 = 0 (8.158)
where, as usually, h = a2 = (T /4)2 .
Following the physical reasoning discussed in Introduction and the second
section of this chapter, we seek solution in the form of successive approxima-
tion series with no explicit small parameter
X = X0 (τ ) + X1 (τ ) + X2 (τ ) + ... (8.159)
h = h0 + h1 + ...
where the generating solution X0 (τ ) obeys the differential equation X0 (τ ) =

0 and therefore describes the dynamics of a two-parameter family of free
impact oscillators with arbitrary amplitudes A0 and the origin shift as follows
X0 (τ, τ1 , A0 ) = x1 + A0 (τ − τ1 ) (8.160)
The idea behind approximation (8.160) is to choose its parameters in order to

make the motion x(t) = X0 (τ (t/a)) in ‘some sense’2 close to that of oscillator
(8.156). The form of (8.160) implies that the system is at the matching point
whenever τ = τ1 .
Next term of series (8.159) is found then from the equation X1 =
−h0 F (X0 ) in the integral form
2
Note that the harmonic balance analogy of generating solution (8.160) would be
x(t) = A sin ωt + B.
τ
X1 (τ, τ1 , A0 , A1 ) = A1 (τ − τ1 ) − h0 (τ − ξ)F (X0 (ξ, τ1 , A0 ))dξ (8.161)
τ1
where the combination of arbitrary constants is similar to that in generating

solution (8.160).
Since X1 (τ1 , τ1 , A0 , A1 ) = 0 then τ = τ1 corresponds to the matching
point x = x1 in first two terms of the successive approximations, (8.160) and
(8.161),
τ
X = x1 + (A0 + A1 )(τ − τ1 ) − h0 (τ − ξ)F (X0 (ξ, τ1 , A0 ))dξ (8.162)
τ1
Therefore, after the first correction to generating solution (8.160), we have

three arbitrary constants A0 , A1 and τ1 , and one still unknown parameter h0
related to the period of free vibration. Now, substituting (8.162) in boundary
conditions (8.158), gives
1
A0 + A1 − h0 F (X0 (τ, τ1 , A0 ))dτ = 0
τ1
τ1
A0 + A1 + h0 F (X0 (τ, τ1 , A0 ))dτ = 0 (8.163)
−1
where the variable of integration ξ has been formally replaced by τ .

Equations (8.163) are equivalent to
1
F (X0 (τ, τ1 , A0 ))dτ = 0 (8.164)
−1
1
F (X0 (τ, τ1 , A0 ))dτ = h−1
0 (A0 + A1 ) (8.165)
τ1
If no more iterations are planned then we set A1 = 0, because the correspond-

ing term contributes nothing qualitatively new into the approximate solution.
In this case, equation (8.164) is used to express τ1 through another constant
A0 and the matching point coordinate x1 . Then, equation (8.165) √ provides
the link between the parameters of amplitude A0 and period T = 4 h0 .
The next step of iteration employs the parameter A1 , however. The form
of differential equation for the next step of procedure is analogous to (8.20),
where γ1 h0 = h1 . Therefore, on the next step, the general solution is given
by
τ
X2 = −h1 (τ − η)F (X0 (η, τ1 , A0 ))dη (8.166)
τ1
τ
−h0 (τ − η)F (X0 (η, τ1 , A0 ))X1 (η, τ1 , A0 , A1 )dη + A2 (τ − τ1 )
τ1
where zero lower limit of integration provides X2 (τ1 ) = 0, and the prime indi-
cates derivative with respect to the entire argument, F (X0 ) ≡ dF (X0 )/dX0 .
Note that the characteristic F (x) is, generally speaking, nonsmooth at the
matching point x1 so that the derivative F (x) may not exist at x = x1 . Al-
though integration of step-wise discontinuous functions is still possible, the
current procedure is designed to avoid calculating derivatives of the char-
acteristic at the point x1 . For that reason, the lower limit of integration in
(8.166) is associated with the non-smoothness point by expression (8.160).
In addition, as follows from (8.161), the uncertainty F (x1 ) in the integrand
is suppressed by zero factor X1 (τ1 ) = 0. Obviously, on the next step of iter-
ation, this factor will accompany the second derivative F (x1 ), whereas the
first derivative acquires the factor X2 (τ1 ).
Now, applying boundary conditions (8.158) to (8.166), yields
1
−h1 F (X0 (τ, τ1 , A0 ))dτ
τ1
1
− h0 F (X0 (τ, τ1 , A0 ))X1 (τ, τ1 , A0 , A1 )dτ + A2 = 0 (8.167)
τ1
τ1
h1 F (X0 (τ, τ1 , A0 ))dτ
−1
τ1
+ h0 F (X0 (τ, τ1 , A0 ))X1 (τ, τ1 , A0 , A1 )dτ + A2 = 0 (8.168)
−1
Substracting the both sides of equation (8.167) from (8.168) and taking into
account equation (8.164), gives
1
F (X0 (τ, τ1 , A0 ))X1 (τ, τ1 , A0 , A1 )dτ = 0 (8.169)
−1
Then, taking into account (8.165), brings equation (8.167) to the form
1
F (X0 (τ, τ1 , A0 ))X1 (τ )dτ = −h1 h−2 −1
0 (A0 + A1 ) + h0 A2 (8.170)
τ1
If no more iterations needed, then we set A2 = 0 in expression (8.166) and

equation (8.170). Further, substituting (8.161) in (8.169), gives equation for
A1 , whereas (8.170) gives equation for h1 . This completes two steps of suc-
cessive approximations, although the algebraic problem still persists, and its
complexity depends on the functions f (x) and g(x).
Chapter 9
Strongly Nonlinear Waves
Abstract. This short chapter deals with modulated waves described by

strongly nonlinear Klein-Gordon equation. Mechanical equivalent for this
model can be represented by the infinite linear string on elastic foundation.
In this case, the principal (carrying) wave can be approximated by poly-
nomials or other elementary functions of the triangular sine wave based on
the algorithms of the previous chapter. Then the differential equations for
slow varying modulating components are derived by using the Whitham as-
sumptions and the specific boundary conditions associated with nonsmooth
argument substitutions.
9.1 Wave Processes in One-Dimensional Systems

A one-dimensional wave can be described by the function
u = u(θ, x, t) (9.1)
where θ = θ(x, t) is a phase variable, x and t are the coordinate and time,
respectively.
In the stationary case, the phase variable is usually the only argument of
the wave shape function u = u(θ), where θ = ωt − kx with temporal and
spatial wave numbers, ω and k, respectively, so that
∂θ ∂θ
=ω and = −k (9.2)
∂t ∂x
Function (9.1) includes also the variables x and t explicitly in order to describe
wave modulation effects. Therefore, we assume that the dependence upon
explicit x and t is slow. Such an assumption can be formalized by means of
the formal auxiliary parameter ε [194], [116]
u = u(θ, x0 , t0 ) (9.3)
where x0 = εx, t0 = εt, and relationships (9.2) take now the form

242 9 Strongly Nonlinear Waves
∂θ ∂θ
= ω(x0 , t0 ) and = −k(x0 , t0 ) (9.4)
∂t ∂x
Let the wave shape function u be periodic with respect to θ with the period
T = 4. In this case, the sawtooth phase argument can be introduced as follows
τ = τ (θ),
u = U (τ, x0 , t0 ) + V (τ, x0 , t0 )e (9.5)
where e = e(θ) = τ (θ) is the rectangular cosine of the phase variable.
9.2 Klein-Gordon Equation
For illustration purposes, we consider the Klein-Gordon equation
∂2u ∂2u
− 2 + f (u) = 0 (9.6)
∂t2 ∂x
where f (u) is an odd function.
2 ∂2U ∂V
ω − k2 = −Rf (U, V ) − εH − ε2 LU
∂τ 2 ∂τ
2 ∂ 2V ∂U
ω − k2 2
= −If (U, V ) − εH − ε2 LV (9.7)
∂τ ∂τ
∂U
|τ =±1 = 0 (9.8)
∂τ
V |τ =±1 = 0
where

∂ ∂ ∂ω ∂k
H ≡ 2 ω 0 +k 0 + 0 + 0
∂t ∂x ∂t ∂x
∂2 ∂2
L ≡ 02 − 02
∂t ∂x
and
1
Rf = [f (U + V ) + f (U − V )]
2
1
If = [f (U + V ) − f (U − V )]
2
Analogously to the case of strongly nonlinear damped oscillator considered
in the previous chapter, we represent solution in the form of series
9.2 Klein-Gordon Equation 243
U (τ, x0 , t0 ) = U0 (τ, x0 , t0 ) + εU1 (τ, x0 , t0 ) + ...

V (τ, x0 , t0 ) = V0 (τ, x0 , t0 ) + εV1 (τ, x0 , t0 ) + ... (9.9)
Substituting (9.9) into (9.7) and (9.9) and equating coefficients of like powers
of ε, gives in first two steps
2 ∂ 2 U0
ω − k2 = −f (U0 )
∂τ 2
V0 ≡ 0 (9.10)
∂U0
|τ =1 = 0 (9.11)
∂τ
where U0 (−τ ) = −U0 (τ ) due to oddness of the function f (U0 ), and
2 ∂ 2 U1
ω − k2 = −f (U0 )U1
∂τ 2
2 ∂ 2 V1 ∂U0
ω − k2 = −f (U0 )V1 − H (9.12)
∂τ 2 ∂τ
∂U1
|τ =±1 = 0
∂τ
V1 |τ =±1 = 0 (9.13)
The generating solution is obtained from (9.10) in the following general form
U0 = U0 (τ, A, ω 2 − k 2 ) (9.14)
where A, ω, and k are functions of the slow variables x0 and t0 .

The functions A(x0 , t0 ), ω(x0 , t0 ), and k(x0 , t0 ) are coupled by the bound-
ary condition (9.11)
∂U0 (τ, A, ω 2 − k 2 )
|τ =1 = 0 (9.15)
∂τ
and by the condition of solvability of the boundary value problem (9.12) for
V1 ,
1 . /
∂U0 ∂U0 ∂U0 ∂U0
H dτ = 2 H =0
∂τ ∂τ ∂τ ∂τ
−1
or 0
2 1 0
2 1
∂ ∂U0 ∂ ∂U0
ω + 0 k =0 (9.16)
∂t0 ∂τ ∂x ∂τ
Example 17. Consider the equation
∂2u ∂2u
− 2 + sgn(u)|u|α = 0 (9.17)
∂t2 ∂x
244 9 Strongly Nonlinear Waves
where α is a real positive number. In this case, the generating solution is

obtained analogously to the case of oscillator with two non-smooth limits,
considered in Chapter 8, by replacing ω02 with ω 2 − k 2 . In particular,
−1
Aα−1 α (α + 1)2
ω −k =
2 2
1+ 1+ (9.18)
α+1 2(α + 2) (α + 2)(2α + 3)
whereas equation (9.16) gives
∂(ωA2 ) ∂(kA2 )
+ =0 (9.19)
∂t ∂x
Equations (9.18) and (9.19) have been written in terms of the original time
and coordinate by setting the auxiliary parameter to unity, ε = 1. However,
according to the basic assumption, A = A (t, x), ω = ω (t, x) and k = k (t, x)
still should be treated as slow functions as compared to τ (θ).
Chapter 10
Impact Modes and Parameter
Variations
Abstract. In this chapter, a new parameter variation and averaging tools

are introduced for impact modes. It is also shown that a specific combina-
tion of two impact modes gives another impact mode1 . The corresponding
manipulations with impact modes become possible due to the availability of
closed form exact solutions obtained by means of the triangular sine temporal
substitution for impulsively loaded and vibroimpact systems. In particular,
the idea of Van-der-Pol and averaging tool are adapted for the case of im-
pact oscillator. For illustrating purposes, a model of coupled harmonic and
impact oscillators is considered. Further, mass-spring systems with multiple
impacting particles are considered in order to illustrate impact localization
phenomena on high-energy levels.
10.1 An Introductory Example

Vibration modes with impacts have been under study for several years [191],
[197], [15], [156]. In practical terms, such studies deal with the dynamics of
elastic structures whose amplitudes are limited by stiff constraints. These
may be designed intentionally or occur due to a deterioration of joints. As
a result, such kinds of dynamics are often accompanied by a rattling noise
or dither during operating regimes of vehicles or machine tools. From the
theoretical standpoint, the interest to such problems is driven by the question
what happens to linear normal modes as the energy of elastic vibrations
becomes sufficient for reaching the constraints. Interestingly enough, some of
the analytical approaches developed in the area recently found applications in
molecular dynamics [49]. However, due to strong nonlinearities of the impact
dynamics, most of the results relates to periodic particular solutions according
to the idea of nonlinear normal modes [190]. Let us recall that the importance
of linear normal modes is emphasized by the linear superposition principle as
1
Notice that the number of impact modes depends on the number of constraints
and therefore can significantly exceed the number of degrees of freedom.

246 10 Impact Modes and Parameter Variations
well as the parameter variation and averaging methods for weakly nonlinear
cases.
Let us consider a one degree of freedom free harmonic oscillator between
two absolutely rigid barriers. A mechanical model of such an oscillator can
be represented as a mass-spring model with two-sided amplitude limiters as
shown in Fig. 10.1 (a).
Fig. 10.1 The oscillator with bilateral rigid barriers (on the left) is replaced by
the oscillator under the periodic series of external impulses (on the right).
The interaction with the barriers at x = ±Δ is assumed to be perfectly

elastic, and the system is represented in the form
ẍ + ω02 x = 0, |x| ≤ Δ (10.1)
Since the normal mode regimes are periodic by their definition then the re-
action of constraints can be treated as a periodic series of external impulses
acting on the masses of the system.
10.1 An Introductory Example 247
Applying this remark to the one degree of freedom system as it is shown

in Fig. 10.1 (b), the related differential equation of motion is written in the
linear form
∞

ẍ + ω02 x = 2p [δ (ωt + 1 − 4k + α) − δ (ωt − 1 − 4k + α)] (10.2)
k=−∞

= pτ (ωt + α)
where δ (ξ) is the Dirac function, τ (ξ) is the triangular sine wave, 2p, ω, and
α will be interpreted as arbitrary parameters.
For further convenience, the right-hand side of equation (10.2) is expressed
through second-order generalized derivative of the triangular sine wave with
respect to the entire argument, ωt + α. The parameter ω will be called a
frequency parameter, although it differs by the factor π/2 from the regular
trigonometric frequency, Ω = (π/2) ω. Both parameters, ω and Ω, may be
used below.
In contrast to system (10.1), the auxiliary system (10.2) is linear but not
completely equivalent to the original one; see the analyses below.
Representing unknown steady-state periodic solution in the form,
x = X (τ ) , τ = τ (ωt + α) (10.3)
gives the boundary value problem with no singular terms,
ω 2 X (τ ) + ω02 X (τ ) = 0 (10.4)
−2
X (τ ) |τ =±1 = pω
and the related solution is represented in the saw-tooth time form [139]
p sin [(ω0 /ω) τ (ωt + α)]

x= (10.5)
ωω0 cos (ω0 /ω)
This solution can be verified by direct substitution of expression (10.5) into

the equation of motion (10.2).
A connection between solution (10.5) and vibration of the original system
with stiff constraints is established by imposing the conditions:
• The impulses on the right-hand side of equation (10.2) act when the mass
strikes the limiters,
x = ±Δ if τ = ±1 ⇐⇒ τ = 0 (10.6)
• The system cannot penetrate through the limiters, therefore,
|x| ≤ Δ for all τ ∈ [−1, 1] (10.7)

Substituting solution (10.5) into condition (10.6), determines the parameter

p,
p = Δωω0 cot (ω0 /ω) (10.8)
Substituting now (10.8) in (10.5), gives solution in the final form
sin [(ω0 /ω) τ (ωt + α)]

x(t) = Δ (10.9)
sin (ω0 /ω)
Obviously, solution (10.9) satisfies condition (10.6) automatically. The related

parameter p (10.8) will further be treated as an “eigen value” of the nonlinear
(impact) problem.
Other parameters, ω and α, are determined by the initial conditions. Let
us assume that x (0) = 0 and thus α = 0. As a result, the total energy
of the oscillator per unit mass is expressed through the initial velocity as
2
E = [ẋ (0)] /2. Then, taking into account (10.9) and making some analytical
manipulations, gives

ω0 1 ω 2 Δ2
= ± arccos 1 − 0 + kπ (10.10)
ω 2 E
where k = 0,1, ... .

The right-hand side of expression (10.10) gives a sequence of real numbers
if the total energy is above its critical value, E ≥ E∗ = ω02 Δ2 /2, so that the
oscillator can reach the constraints. However, not all of the real numbers ω
lead to real motions of the original system. Indeed, since the auxiliary sys-
tem (10.2) has no limiters then condition (10.6) does not guarantee that the
oscillator will remain inside the region |x| ≤ Δ during the period of vibra-
tion. Therefore, condition (10.7) must be verified as well. Such a verification
implemented for solution (10.9) shows that condition (10.7) is satisfied only
for the smallest root in set (10.10). Fig. 10.2 illustrates the temporal mode
shapes corresponding to the first two roots ω0 /ω.
It is seen that the second solution (on the right), violates condition (10.7)
while satisfies condition (10.6).
Note that the above approach can be applied to the case of unilateral
limiters. Let us remove, for instance, the left limiter and consider the oscillator
(10.1) under the condition x ≤ Δ. In this case, the boundary conditions in
(10.4) should be modified as
X (τ ) |τ =±1 = ±pω −2 (10.11)
Such a periodic change of sign effectively switches the directions of positive

δ-pulses on the right-hand side of equation (10.2). As a result, the solution
takes the form,
cos [(ω0 /ω) τ (ωt + α)]
x(t) = Δ (10.12)
cos (ω0 /ω)
10.2 Parameter Variation and Averaging 249
2 a 2 b
1 1
x 0 x 0
1 1
2 2
0 5 10 15 20 0 5 10 15 20
t t
Fig. 10.2 Real (a) and “phantom” (b) solutions corresponding to the first (small-
est) and second roots respectively, (ω0 /ω)1 = 1.1502 and (ω0 /ω)2 = −1.1502 + π.
The total energy level is E = 1.2E∗ .
where the period and the related basic frequency are T = 2/ω and Ω = πω,
respectively.
10.2 Parameter Variation and Averaging

In order to illustrate the idea of parameter variations for solution (10.9), let
us include the viscous damping into the model and represent the differential
equations of motion between the constraints in the form
ẋ = v
v̇ = −2ζω0 v − ω02 x (10.13)
where ζ is the damping ratio parameter.

In this case, the triangular wave frequency ω in solution (10.9) is not
constant any more, although the amplitude of the vibration remains constant
as long as the oscillator is in the impact regime. The corresponding parameter
variation is implemented as a change of the state variables {x(t), v(t)} →
{γ(t), φ(t)}, dictated by solution (10.9)
sin(γτ (φ))
x=Δ (10.14)
sin γ
cos(γτ (φ))
v = ω0 Δ e(φ)
sin γ
where τ = τ (φ) and e = τ (φ) depend upon the fast phase φ = φ(t), and
γ = γ(t) determines a relatively slow evolution of the temporal mode shape
of the vibration.
Substituting (10.14) in (10.13), gives still exact equations
γ̇ = 2ζω0 cos2 γτ tan γ

ω0
φ̇ = [1 + eζ(sin 2γτ − 2τ cos2 γτ tan γ)] (10.15)
γ
Below, the first-order averaging procedure is applied. Notice that the right-
hand side of equations (10.15) is periodic with respect to the phase φ. How-
ever, as proved in Chapter4, the averaging can be conducted with respect to
the variable τ over its interval −1 ≤ τ ≤ 1. As a result, one obtains

1
γ̇ = ζω0 1 + sin 2γ tan γ
2γ
8
= 2ζω0 γ + ζω0 γ 5 + O(γ 6 )
45
ω0
φ̇ = (10.16)
γ
Keeping the leading-order term only on the right-hand side of the first equa-
tion in (10.16), gives solution
γ = γ0 exp(2ζω0 t)
1
φ= [1 − exp(−2ζω0 t)] (10.17)
2ζγ0
A simple asymptotic analysis of expressions (10.14) and the remark after

expression (10.10) give the parameter interval, 0 < γ < π/2, within which
the impact dynamics takes place. The vibration mode shapes close to the
triangular wave near the left edge of the interval, but, as the energy dissipates
and the parameter γ approaches π/2, vibrations become close to harmonic.
The total energy is expressed through the parameter γ in the form
1 ω02 Δ2
E(t) = (10.18)
2 sin2 γ(t)
The duration of the impact stage of the dynamics is estimated via solution
(10.17),

1 π
γ(tmax ) = π/2 =⇒ tmax = ln (10.19)
2ζω0 2γ0
where γ0 = γ(0).
As follows from Fig. 10.3, the above averaging procedure leads to practi-
cally no error of the time history record within the entire interval of validity
of the approach. However, the analytical solution based on the reduced model
10.2 Parameter Variation and Averaging 251
1.0
0.5
x 0.0
0.5
1.0
0 10 20 30 40
a t
1.0
numerical, averaging
0.5
x 0.0
0.5 analytical
1.0
80 82 84 86 88 90
b t
Fig. 10.3 The time history of the impact oscillator according to exact equations,
those after the averaging, and the analytical closed form solution; the parameters
are γ0 = 0.2, ζ = 0.01, ω0 = 1.0, and Δ = 1.0.
gives some deviation from the exact curve at the end of the impact stage of
the dynamics.
Notice that there is no impact interactions with the constraints
for t > tmax , where the model becomes harmonic oscillator whose ampli-
tude exponentially decays due to the energy dissipation. At this stage, the
transformation (10.14) is not valid any more nor there is any need in trans-
formations. However, a question occurs about such solutions that would be
capable of describing both impact and non-impact stages within the same
unit-form expressions.
10.3 A Two-Degrees-of-Freedom Model

Let us obtain first the impact mode solutions for the model shown in Fig.
10.4 under no damping condition, c = 0. For the sake of simplicity, let us also
assume that k1 = k2 = k. On the normal mode (impact) motions, the system
can be effectively replaced by
ẍ1 + ω02 (2x1 − x2 ) = pe (ωt)

ẍ2 + ω02 (2x2 − x1 ) = 0 (10.20)
where ω02 = k/m, and the parameters ω and p must provide the following
condition
|x1 | ≤ Δ (10.21)
The impact mode solution is represented in the form
xn (t) = Xn (τ ) (10.22)
where τ = τ (ωt) and n = 1, 2.

Substituting (10.22) in (10.20) and eliminating the singular term e (ωt),
gives the linear boundary value problem
ω 2 X1 + ω02 (2X1 − X2 ) = 0

ω 2 X2 + ω02 (2X2 − X1 ) = 0 (10.23)
X1 |τ =±1 = pω −2
X2 |τ =±1 = 0 (10.24)
The corresponding solution has the form

√ √
p sin(ω0 τ /ω) 3 sin( 3ω0 τ /ω)
X1 = + √
2ωω0 cos(ω0 /ω) 3 cos( 3ω0 /ω)
√ √
p sin(ω0 τ /ω) 3 sin( 3ω0 τ /ω)
X2 = − √ (10.25)
2ωω0 cos(ω0 /ω) 3 cos( 3ω0 /ω)
where the ‘eigen value’ p is determined by substituting X1 into
X1 |τ =±1 = ±Δ (10.26)
10.4 Averaging in the 2DOF System 253
This gives
√ √
p = 2ωω0 Δ[tan(ω0 /ω) + ( 3/3) tan( 3ω0 /ω)]−1 (10.27)
In order to insure that solution (10.25) and (10.27) describe real motions,
condition (10.7) must be verified. As follows from detailed considerations
below, condition (10.7) is satisfied at least in the frequency range (π/2) ω >
ω2 , where ω2 is the frequency of the out-of-phase mode of the related no
impact system.
10.4 Averaging in the 2DOF System

Let us consider now the model shown in Fig. 10.4 by assuming that a rel-
atively slow energy dissipation is possible due to the viscous damping. In
order to introduce the corresponding parameter variation technique, let us
represent the differential equations of motion in the following general form
ẋ1 = v1
v̇1 = −f1 (x1 , x2 , v1 , v2 ) + pe (φ)
ẋ2 = v2 (10.28)
v̇2 = −f2 (x1 , x2 , v1 , v2 )
where pe (φ) = pτ (φ) and
f1 (x1 , x2 , v1 , v2 ) = 2ζΩ1 v1 + Ω12 x1 + β(x1 − x2 )

f2 (x1 , x2 , v1 , v2 ) = Ω22 x2 − β(x1 − x2 ) (10.29)
are impact and linear force components per unit mass, β = k/m is
the pa-
rameter of coupling, ζ = c/(2Ω1 m) is the damping ratio, and Ωi = ki /m
(i = 1, 2).
Fig. 10.4 The two-degrees-of-freedom model with viscous damping in the impact
subcomponent.
The idea of parameter variations is implemented below as a change of the

state variables, {x1 (t), v1 (t), x2 (t), v2 (t)} → {γ(t), φ(t), A(t), B(t)}, accord-
ing to expressions
sin(γτ (φ))
x1 = Δ
sin γ
cos(γτ (φ))
v1 = Ω1 Δ e(φ)
sin γ

πτ (φ) πτ (φ)
x2 = A sin + B cos e(φ) (10.30)
2 2

πτ (φ) πτ (φ)
v2 = −Ω1 B sin + Ω1 A cos e(φ)
2 2
It is assumed in (10.30) that the principal frequency of the vibration, dφ(t)/dt,

is dictated by the impact subcomponent rather then by a natural frequency
of the corresponding linearized system. However, the scaling factor Ω1 is still
used in order to indicate the dominant oscillator of the dynamic process under
consideration. Notice that x1 and v1 are transformed analogously to (10.14),
whereas the transformation of x2 are v2 is based on the standard general
solution of the harmonic oscillator represented however in the non-smooth
temporal transformation form.
e
γ̇ = cos γτ (f1 sin γ − Ω12 Δ sin γτ ) tan γ
Ω1 Δ
1
φ̇ = [f1 sin γ(sin γτ − τ cos γτ tan γ)
γΩ1 Δ
+Ω12 Δ cos γτ (cos γτ + τ sin γτ tan γ)]
1 1
Ȧ = − Ω1 B(1 − cos πτ ) + πB φ̇ (10.31)
2 2
1 1 πτ
+( Ω1 A sin πτ − f2 cos )e
2 Ω1 2
1 1
Ḃ = Ω1 A(1 + cos πτ ) − πAφ̇
2 2
1 πτ 1
+ f2 sin − e Ω1 B sin πτ
Ω1 2 2
where the functions f1 and f2 are expressed through the new variables by sub-
stitution (10.30) in (10.29), and the impact term pe (φ) has been eliminated
by setting (compare with (10.8))
p = p(t) = Δφ̇(t)Ω1 cot γ(t) (10.32)
Further reduction of system (10.31) includes two major steps, such as averag-
ing with respect to the fast phase φ and applying the power series expansion
10.4 Averaging in the 2DOF System 255
with respect to the parameter γ. Since the periodic functions in equations

(10.31) are expressed through the triangular sine wave τ (φ) then the averag-
ing can be implemented by considering τ as an argument of the averaging as
described in Chapter 4. Then, keeping the leading-order terms of the power
series expansions with respect to γ, gives
γ̇ = 2ζΩ1 γ
Ω1
φ̇ =
γ

β + Ω12 + Ω22 π
Ȧ = −B − φ̇ (10.33)
2Ω1 2

2 2

β + Ω1 + Ω2 π 4βΔ
Ḃ = A − φ̇ − 2
2Ω1 2 π Ω1
Approximate equations (10.33) describe only one-way interaction between the

oscillators so that the first two equations can be easily solved analytically; see
the above one degree-of-freedom case. Then, substituting the result into the
next two equations, gives a linear set of equations with variable coefficients
for A(t) and B(t), which can also be considered analytically. Let us skip
such kind of analysis but illustrate the final result in Figs. 10.5 and 10.6 for
0.00020 reduced model

numerical
0.00015
Ρ averaging
0.00010
0.00005
0.00000
0 20 40 60 80
t
Fig. 10.5 The total energy of the second oscillator at relatively weak coupling,
β = 0.01
0.007
numerical
0.006
0.005 averaging
0.004
Ρ
0.003
0.002 reduced model
0.001
0.000
0 20 40 60 80
t
Fig. 10.6 The total energy of the second oscillator at stronger coupling, β = 0.05
two different strengths of the coupling β. The initial conditions and other
parameters are selected as follows: γ(0) = 0.2, φ(0) = 0.0, A(0) = −0.01,
B(0) = 0, ζ = 0.01, Ω1 = Ω2 = 1.0, Δ = 1.0. The diagrams show the energy
versus time of the second oscillator based on numerical solutions for three
different equations, such as exact equations (10.31), the equations obtained
by averaging (not described here), and the reduced set (10.33). The solutions
are in quite a good match most of the time interval, however, the solution
of truncated set (10.33) show some error near the end of the interval. This
happens because the parameter γ is slowly approaching its limit magnitude
π/2, at which the first oscillator stops interacting with the constraints and the
entire system becomes linear. Remind that equations (10.33) were obtained
by truncating the polynomial expansions in the neighborhood of γ = 0; as a
result, the accuracy of the equations is low near the point γ = π/2. However,
the precision can be significantly improved by keeping few more terms of the
power series with respect to γ.
10.5 Impact Modes in Multiple Degrees of Freedom

Systems
Let us consider the N -degrees-of-freedom conservative system described by
T
the coordinate vector x = (x1 , ..., xN ) ∈ RN . The corresponding mass-spring
model is shown in Fig. 10.7. It is assumed that displacements of the ath mass
10.5 Impact Modes in Multiple Degrees of Freedom Systems 257
Fig. 10.7 The mass-spring model of a discrete elastic system with displacement
limiters.
is limited by perfectly stiff elastic constraints, such that |xa | ≤ Δa or, in the
matrix form, T
Ia x ≤ Δa (10.34)

T
where Ia = 0, ..., 1, ..., 0 .
a
Inside the domain (10.34), the differential equations of motion are assumed
to be linear,
M ẍ + Kx = 0 (10.35)
where M and K are constant mass and stiffness N ×N -matrixes, respectively.
The form of matrix equation (10.35) however is general enough to describe
different models, not necessarily mass-spring chains, see the first example
below.
In order to obtain impact mode solutions, the system (10.34) and (10.35)
is replaced by the following impulsively forced linear system under no con-
straints condition
M ẍ + Kx = pIa τ (ωt + α) (10.36)
where p is a priory unknown ‘eigen-value,’ and ω and α are arbitrary constant
parameters.
A family of periodic solutions of the period T = 4/ω can be found as
a linear superposition of solutions (10.5) for each of the N linear modes of
system (10.35) with appropriate replacement of the parameters,
N
(eTj Ia )ej sin[(ωj /ω)τ (ωt + α)]
x(t) = p (10.37)
j=1
ωωj cos(ωj /ω)
where ej and ωj are the jth normal mode and the natural frequency of linear
system (10.35); it is assumed that ωi < ωj when i < j, and the linear normal
modes are normalized as
eTj M ei = δji (10.38)

where δji is the Kronecker symbol.
The impulses act at those time instances when the ath mass interacts with
the constraints, in other terms,
ITa x = ±Δa when τ = ±1 (10.39)
Substituting (10.37) into (10.39), one obtains the related “eigen-value,”

⎡ ⎤−1

N
tan(ωj /ω) ⎦
p = Δa ω ⎣ (eTj Ia )2 (10.40)
j=1
ωj
where eTj Ia is the ath component of the jth linear mode vector.
Substitution (10.40) into (10.37) gives a two parameter family of the peri-
odic solutions for the impact modes. The parameter α is an arbitrary phase
shift, whereas the frequency parameter ω is subjected to some restrictions due
to condition (10.34). As shown in [156], condition (10.34) is satisfied when the
principal frequency of vibration, Ω = (π/2) ω, exceeds the highest frequency
of the linear spectrum, Ω > ωN . The corresponding impact mode represents
an extension of the highest linear normal mode, in which any two neighboring
masses vibrate out-of-phase. Such an impact mode becomes spatially local-
ized as Ω → ∞. This result was obtained also by qualitative methods in [197].
Condition (10.34) may be satisfied also when the frequency Ω is located in a
small enough right neighborhood of any frequency ωj and, in addition, ωi /ωj
is not an odd number for all i = j. The idea of the proof is to find such cases
when ITa x is a monotonic function of τ on the interval −1 ≤ τ ≤ 1, and hence
condition (10.39) at the boundaries guarantees that inequality (10.34) holds
inside the entire interval.
Generally, the impact modes appear to have a quite complicated spec-
tral structure. Therefore, a detailed investigation may be required in order
to formulate necessary and sufficient conditions of impact mode existence.
However, a sufficient condition of non-existence can be formulated by using
the physical meaning of the parameter p. Namely, if an impact mode exists
then the inequality p (ω) > 0 holds. Indeed, the parameter p (10.40) cannot
be negative for any real impact vibrating regime as a reaction of constraint,
because it cannot be directed to a barrier. Thus impact modes can not exist
when p (ω) < 0.
10.5.1 A Double-Pendulum with Amplitude Limiters

The top mass m1 of a free double-pendulum, as shown in Fig. 10.8, oscillates
between the two absolutely stiff constraints providing small angular ampli-
tudes of the top pendulum,
|ϕ1 | ≤ Δ1 1
Fig. 10.8 Double pendulum with bilateral constraints.
Assuming that the angle ϕ2 is also small enough, brings the differential equa-
tions of motion between the barriers to the linear form
μ2 ϕ̈1 + ϕ̈2 + μ2 ϕ1 = 0
ϕ̈1 + ϕ̈2 + ϕ2 = 0
where μ2 = 1 + m1 /m2 , over dots

indicate differentiation with respect to the
new temporal argument, t̄ = g/lt, and l is the length of each rod.
In this case, the linear mode vectors and natural frequencies are

1 1 μ
e1 = , ω1 =
2μ (μ + 1) μ μ+1

1 1 μ
e2 = , ω2 =
2μ (μ − 1) −μ μ − 1
where the modal vectors satisfy the orthogonality condition, eTj M ei = δji ,
with respect to the mass matrix

2
μ 1
M=
1 1
As a result, taking into account that Ia = I1 , Δa = Δ1 , and N = 2, brings

the formal solution (10.37) and (10.40) to the form

ϕ1 −1
= Δ1 [tan (ω1 /ω) + (ω2 /ω1 ) tan (ω2 /ω)] ×
ϕ2

1 sin [(ω1 /ω) τ (ωt + α)] ω2 1 sin [(ω2 /ω) τ (ωt + α)]
+
μ cos (ω1 /ω) ω1 −μ cos (ω2 /ω)
where the relationships μ − 1 = μω2−2 and μ + 1 = μω1−2 have been used in
manipulations.
Note that the frequency of first impact mode must be close enough to the
first linear frequency ω1 so that it is still away from the left neighborhood of
the next frequency, ω2 . In current case, ω2 is the highest frequency, therefore
its right “neighborhood” has no upper boundary. As a result, the highest
impact mode becomes spatially localized as its frequency parameter grows.
The localization admits explicit estimation by the asymptotic expansion

−1
ϕ2 ω1 ω2 ω2 ω1 ω2 ω2
|τ =1 = μ tan − tan tan + tan
ϕ1 ω ω1 ω ω ω1 ω

ω12 − ω22 2ω12 ω22
=μ 2 1+ ω −2 + O ω −4 (10.41)
ω1 + ω22 3 (ω12 + ω22 )
1
= −1 − ω −2 − O ω −4
3
It follows from (10.41) that (ϕ2 /ϕ1 ) |τ =1 → −1 as ω → ∞, so that the
amplitude of the bottom mass becomes negligibly small whereas the upper
mass has the amplitude determined by the angular limiters.
10.5.2 A Mass-Spring Chain under Constraint

Conditions
Let us consider a mass-spring chain of N identical particles under the con-
straint condition
k
ẍn + (−xn−1 + 2xn − xn+1 ) = 0; n = 1, ..., N (10.42)
m
x0 = xN +1 = 0; |xa | ≤ Δa ; 1<a<N (10.43)
where k and m are the stiffness of each spring and the mass of each particle,
respectively.
In this case, the corresponding linear modes and their frequencies are de-
scribed exactly by expressions

T
2 πj N πj
ej = sin , ..., sin (10.44)
N +1 N +1 N +1
πj
ωj = ωN +1 sin
2 (N + 1)

where notation2 ωN +1 = 2 k/m has been introduced, and the basis vectors
are normalized so that eTj ei = δji .
The ath component of the jth normal mode vector is therefore

T 2 aπj
ej Ia = sin (10.45)
N +1 N +1
Let us show that the impact mode periodic solution (10.37) and (10.40)
becomes localized as ω → ∞.
First, replacing the trigonometric functions by their asymptotic estimates
sin(ωj τ /ω) ∼ ωj τ /ω, cos(ωj /ω) ∼ 1, and tan(ωj /ω) ∼ ωj /ω, gives
2N T T
j=1 (ej Ia )(ej Ib )
xb ∼ Δa 2N τ (ωt + α) as ω→∞ (10.46)
T 2
j=1 (ej Ia )
In this particular example, the sums can easily be evaluated. So, taking into
account expression (10.45) and the standard trigonometric sums [52], gives

N
2
N
aπj bπj
(eTj Ia )(eTj Ib ) = sin sin = δab (10.47)
j=1
N + 1 j=1 N +1 N +1

N
1 2
(eTj Ia )(eTj Ib )ωj2 = ω (−δa,b−1 + 2δa,b − δa,b+1 ) (10.48)
j=1
2 N +1
Substituting then (10.47) in (10.46), gives
xb ∼ Δa δab τ (ωt + α) as ω→∞ (10.49)
Expression (10.49) shows that the ath particle of the chain vibrates accord-
ing to the saw-tooth temporal mode shape with the infinitely large frequency,
2
Note that this is just a suitable notation since the (N + 1)th frequency does not
physically exist.
whereas all the other particles are at rest. Therefore, the impact mode be-
comes spatially localized as ω → ∞.
10.6 Systems with Multiple Impacting Particles

Let us consider the case of two particles, say the ath and bth, under the
constraint conditions. These conditions are |xa | ≤ Δa and |xb | ≤ Δb or, in
the vector notations,
T
I x ≤ Δa and IT x ≤ Δb (10.50)
a b
In this case, the impulsive excitation on the right-hand side of the auxiliary
equation must act on both ath and bth particles, so that the equation takes
the form
M ẍ + Kx = (pa Ia + pb Ib )τ (ωt + α) (10.51)
where pa and pb are parameters to be determined.
The related solution includes terms related to pa and pb , and can be rep-
resented in the form
N
pa (eTj Ia )ej + pb (eTj Ib )ej sin(ωj τ (ωt + α) /ω)
x(t) = (10.52)
j=1
ωωj cos(ωj /ω)
Following the idea of normal modes, let us assume that the impact mode
periodic motion is accompanied by synchronous impacts of both particles
with the constraints according to conditions
ITa x = ±Δa and ITb x = ±Δb when τ = ±1 (10.53)
Substituting (10.52) in (10.53), gives linear algebraic equations with respect

to pa and pb in the form
kaa pa + kab pb = Δa
kba pa + kbb pb = Δb (10.54)
where

N
(eTj Ia )(eTj Ib ) ωj
kab = tan (10.55)
j=1
ωωj ω
Expressions (10.52) and (10.54) give a formal impact mode solution, which
indeed describes an impact mode when the determinant of system (10.54)
is non-zero and condition (10.50) holds. Solution (10.52) can be viewed as a
strongly non-linear superposition of the two basic impact modes with a single
impacting mass.
10.6 Systems with Multiple Impacting Particles 263
Let us consider the higher frequency domain, (π/2) ω >> ωN , for the
above example of mass-spring system. In this case, tan(ωj /ω) > 0 for all
j = 1, ..., N , and therefore, the coefficients kab (10.55) create a Gram matrix
with non-zero determinant [23]. The asymptotic estimation below confirms
this conclusion.
Namely, assuming that (π/2) ω >> ωN , provides that all the ratios ωj /ω
are small enough to justify the following asymptotic estimate

sin[(ωj /ω)τ ] ωj τ 1 ωj 3 τ3
= + τ− + O ω −5 (10.56)
cos(ωj /ω) ω 2 ω 3
Then substituting expression (10.56) in (10.52) and (10.55) for an arbitrary

cth particle, gives

1 ωN +1 2 τ3
xc = (Pa δac + Pb δbc ) τ − τ− × (10.57)
4 ω 3
−4
× [Pa (δa,c−1 − 2δac + δa,c+1 ) + Pb (δb,c−1 − 2δbc + δb,c+1 )] + O ω
where xc = xT Ic and τ = τ (ωt + α), and the parameters {Pa , Pb } =

ω −2 {pa , pb } are determined by the linear algebraic system
Kaa Pa + Kab Pb = Δa
Kba Pa + Kbb Pb = Δb (10.58)
where
1 ωN +1 2
Kab = ω 2 kab = δab − (δa,b−1 − 2δab + δa,b+1 ) + O ω −4 (10.59)
6 ω
As follows from (10.59), equations (10.58) have the solution Pa = Δa and
Pb = Δb as ω → ∞. In this limit, the vibration energy becomes localized on
the two particles vibrating between the barriers with the saw-tooth temporal
shape,
xc ∼ (Δa δac + Δb δbc ) τ (ωt + α) as ω → ∞ (10.60)
According to (10.60), the particles with numbers c = a and c = b are at rest.
However, they oscillate with amplitudes of different orders of ω −2 when the
parameter ω takes a large but finite value.
Expansion (10.57) shows that the temporal mode shapes of particles c = a
and c = b are non-smooth and getting close to the triangular sine wave as the
frequency ω increases. The temporal mode shapes of the nearest particles,
c = a ± 1 and c = b ± 1, have amplitudes of order ω −2 and appear to be twice
continuously differentiable with respect to time, t. In particular, calculating
directly first two derivatives, gives

d τ3
τ− = ω 1 − τ 2 τ ∈ C 1 (R)
dt 3

d 2
τ3 2

τ − = ω 2
−2τ (τ ) + 1 − τ 2
τ = −ω 2 2τ ∈ C (R)
dt2 3
where prime denotes differentiation with respect to the whole argument of the
saw-tooth sine, τ ≡ dτ /d (ωt + α), and the term underlined is zero3 because
1 − τ 2 = 0 on the set of points {t : τ (ωt + α) = ±1}, where τ = 0.
3
Namely, it gives zero contribution into the related integrals of the theory of
distributions.
Chapter 11
Principal Trajectories of Forced
Vibrations
Abstract. As shown earlier by Zhuravlev (1992) that harmonically loaded

linear conservative systems possess an alternative physically reasonable ba-
sis, which is generally different from that associated with conventional princi-
pal coordinates. Briefly, such a basis determines directions of harmonic loads
along which the system response is equivalent to a single oscillator. The corre-
sponding definition (principal directions of forced vibrations) is loosing sense
in nonlinear case, when the linear tool of eigen vectors becomes inapplicable.
However, it will be shown in this chapter that nonlinear formulation is still
possible in terms of eigen vector-functions of time given by NSTT boundary
value problems. Physical meaning of the corresponding nonlinear definitions
for both discrete and continual models is discussed.
11.1 Introductory Remarks

The theory of linear normal modes defines a natural basis in the configuration
space of linear conservative systems. The corresponding directions are asso-
ciated with a set of independent harmonic oscillators. The number of such
oscillators is infinite, if the original system is continuous. In the later case,
the modal analysis provides reduction of a continuous system to the related
discrete set of harmonic oscillators. As it is known, the normal modes are
defined for a class of unforced systems, therefore only initial conditions select
those oscillators that will be excited during the dynamical process. Practi-
cally, a normal mode regime must be supported by some external loading
due to inevitable energy dissipation. However, the theory does not identify
directly such external forces. Let ψj (y) be, for instance, the jth mode shape
of a beam. Generally speaking, the external loading of the same profile, ψj (y),
will excite not only the jth mode but also some others, unless the mass per
unit length of the beam is constant. From the mathematical viewpoint, this
is due to the mass density, say ρ(y), participating as a weighting factor in
the orthogonality condition

266 11 Principal Trajectories of Forced Vibrations
< ψi (y)ρ(y)ψj (y) >= 0, i = j (11.1)
The question therefore is what kind of external force must be applied to a

mechanical system in order to generate a normal mode type of motion when
all the system particles coherently vibrate with the same frequency?
Following reference [203], let us consider first the linear case assuming that
the linear n-degree-of-freedom forced system oscillates as a single harmonic
oscillator in such a manner that the coordinates vector x (t) and the force
vector p (t) are collinear to the same constant vector q with a constant length
ratio μ as follows
x = q sin ωt, p = μq sin ωt (11.2)
In the case of forced vibration, the frequency ω is rather predetermined by the
external loading and therefore should not play the role of eigen-value. It was
shown in [203] that the coefficient of proportionality μ can play such a role
instead. In a regular case, the coefficient μ has exactly n eigen-values, whereas
the vector q determines the corresponding ‘principal directions’ according to
the definition of reference [203].
Note that the principal directions are always orthogonal regardless the
mass matrix of the system. Such an approach therefore determines a new
natural basis for external forces from the standpoint of system considered.
This, of-course, should not be viewed as a substitute for the theory of normal
modes, however, some non-autonomous problems can be naturally solved by
making use of the above complementary basis.
In nonlinear cases, definition (11.2) is unapplicable and the above notion
of principal directions loses its sense. However, it was shown in [136] that
the basic idea still can be generalized by considering trajectories instead of
directions. Also a mixed spatio-temporal consideration must be applied since
spatial and temporal coordinates are not separable in nonlinear cases and
the related vibration and forcing are generally neither harmonic in time not
similar in space.
There are some practically important formulations of the problem for the
case of nonlinear forced vibration, which could be qualified as inverse or
semi-inverse approaches. The related methods select practically reasonable
external forces that generates simple enough dynamics. For example, Har-
vey [57] considered ‘natural forcing functions’ proportional to the non-linear
restoring force of the forced Duffing oscillator.
The notion of ‘exact steady state’ was defined by Rosenberg [168] for a
strongly nonlinear single degree of freedom system as a vibration with the
cosine-wave temporal shape of the period of external force. The corresponding
forcing function is determined under some initial conditions. Kinney and
Rosenberg [80] considered systems with many degrees of freedom.
11.2 Principal Directions of Linear Forced Systems 267
11.2 Principal Directions of Linear Forced Systems

Let us illustrate first the basic idea of reference [203] by considering the linear
n-degree-of-freedom forced system
M ẍ + Kx = p (ωt) , x (t) ∈ Rn (11.3)
where M and K are constant mass and stiffness n× n -matrixes, respectively;

p (ωt) is a periodic vector-force of the period T = 2π with respect to ωt, and
the upper dot means differentiation with respect to time, t.
Substituting (11.2) in (11.3), gives the eigen-value problem with respect
to the parameter μ and vector q in the form
− ω 2 M q + Kq = μq (11.4)
Let q = vs and μ = μs be the sth eigen-vector and eigen-value respectively,

s = 1, ..., n. The eigen-vectors vs are orthogonal and can be normalized by
condition
viT vj = δij (11.5)
where δij is the Kronecker symbol.
Therefore, the set of vectors vs determine a natural basis for the case
of forced vibrations. Let, for instance, the external force be p = Q sin ωt,
where Q ∈ Rn is an arbitrary constant vector. In this case, the corresponding
steady-state (particular) solution is written as

vsT Q
x= vs sin ωt (11.6)
s
μs
Now, let es and ωs be conventional linear normal modes and natural frequen-
cies of the system. (The related eigen-value problem is obtained from (11.4)
by setting μ = 0.) As follows from the linear theory, the normal mode vectors
are orthogonal with respect to the mass matrix M so that the normalization
condition can be represented in the form
eTi M ej = δij (11.7)
Using the normal mode basis for the above steady-state, gives

eTs Q
x= es sin ωt (11.8)
s
ωs2 − ω 2
Since the uniqueness theorem holds, expansions (11.6) and (11.8) must rep-
resent the same solution, and therefore,

vsT Q eTs Q
vs = es (11.9)
s
μs s
ωs2 − ω 2
Let the external force amplitude vector Q be directed along one of the prin-
cipal directions. Then, expansion (11.6) will include only one term, whereas
expansion (11.8) still includes all n terms.
Now, let us consider the case, when the mass matrix is equal to the identity
matrix, M = E. In this particular case, expression (11.4) takes the standard
form of the eigen-value problem for normal modes with respect to the eigen-
value parameter ω 2 + μ,

− ω 2 + μ Eq + Kq = 0 (11.10)
As follows from (11.10), the eigen-values of free and forced vibration are
coupled by expression
ω 2 + μs = ωs2 , s = 1, ..., n (11.11)
It is seen that each eigen-value of forced vibration, μs = ωs2 − ω 2 , is a

monotonically decreasing functions of the external frequency ω with only one
zero at ω = ωs .
11.3 Definition for Principal Trajectories of Nonlinear

Discrete Systems
Let us consider the nonlinear case
M ẍ + Kx + εf (x) = p (ωt) , x (t) ∈ Rn (11.12)
where f (x) is an analytic nonlinear vector-function such that f (−x) = −f (x),

ε is a small positive parameter, and the forcing function and matrixes are
defined in equation (11.3).
If ε = 0, then the concept of principal directions of forced vibrations is
not applicable any more, however it is still possible to consider principal
trajectories instead based on the following
Definition 1. Trajectories of periodic motions of the period T = 2π/ω on
which mechanical system (11.12) behaves as a Newtonian particle in Rn ,
namely the external force and acceleration vectors are coupled by the Newton
second law,
mẍ (t) = p (ωt) (11.13)
will be called principal trajectories of forced vibrations.
In equation (11.13), m is a priory unknown effective mass parameter. The
effective mass m and the force p (ωt) must be chosen in order to make equa-
tions (11.12) and (11.13) compatible.
Note that, in the linear case, the above definition still gives principal di-
rections of forced vibrations (11.2) after representing the mass parameter as
follows
11.4 Asymptotic Expansions for Principal Trajectories 269
μ
m=− (11.14)
ω2
Indeed, substituting expression x (t) = q sin ωt in equation (11.13) and taking
into account expression (11.14), gives definition (11.2) in the form p =μx. In
contrast to linear case (11.2), however, definition (11.13) allows non-harmonic
temporal shapes.
Current definition itself does not imply that the system is weakly nonlinear.
However, if the parameter ε is small then explicit solutions can be obtained in
terms of conventional asymptotic expansions as described in the next section.
As mentioned, the notion of principal trajectories seems to relate to the
idea of ‘natural forcing functions’ introduced in [57] for the Duffing oscillator.
Let us consider now a multidimensional case from that point of view.
Applying definition (11.13) to the general nonlinear system
M ẍ + F (x) = p (ωt) (11.15)
and eliminating the acceleration, gives the external forcing vector-function

as a linear transformation of the restoring force in the form,

−1
1
p (ωt) = E − M F (x) (11.16)
m
where the matrix of the transformation includes the effective mass parame-
ter m.
Relationship (11.16) can be viewed as a vector version of the concept of
natural forcing functions.
On the other hand, using the definition for principal trajectories and ex-
cluding the external forcing vector p (ωt) from the equation of motion, gives
an auxiliary free system described by the differential equation of motion
(M − mE) ẍ + F (x) = 0
The idea of transforming the forced problem to a free vibration problem

by imposing the form of excitation was used also in [31] with illustrations on
two degrees of freedom systems based on an essentially different methodology
though.
11.4 Asymptotic Expansions for Principal Trajectories
In order to make equations (11.12) and (11.13) compatible, let us eliminate

the forcing vector-function p (ωt) and thus consider equation
M ẍ + Kx + εf (x) = mẍ (t) (11.17)

A family of periodic solutions, that give principal directions of linearized

system as ε → 0, will be considered. Let us represent such solutions (principal
trajectories) in the following parametric form
x = X (τ ) (11.18)
where is τ = τ ((2ω/π)t) is the triangular sine wave of the period of external

loading, T = 2π/ω.

2
2ω
LX+εf (X) = mX (11.19)
π

2
2ω d2
L≡ M +K
π dτ 2
X (τ ) |τ =±1 = 0 (11.20)
As mentioned above, the temporal and spatial variables generally are not
separable any more in nonlinear cases, therefore it is impossible to obtain an
exact nonlinear version of the eigenvector problem (11.4). As a result, both
temporal and spatial mode shapes must be corrected on each step of the
related asymptotic process as described below.
Remind that the differential operator L in equation (11.19) includes the
frequency parameter ω fixed, whereas the mass m is an eigen value to be
determined.
Let ma and ea (τ ) be the eigen value and eigen vector of the linearized
problem, ε = 0, respectively,

2
2ω
Lea = ma ea (11.21)
π
ea | τ =±1 = 0
where the index a = {s, j} consists of spatial and temporal mode shape
numbers, s = 1, ..., n and j = 1, ..., respectively.
The scalar product of any two vector-functions x = x (τ ) and y = y (τ )
will be defined as follows

1 1 T
x, y = x ydτ
2 −1
Let us represent solution of the weakly nonlinear eigen value problem (11.19)
and (11.20) in the following form of asymptotic expansions
11.5 Definition for Principal Modes of Continuous Systems 271

X(τ ) = Aea (τ ) + εX(1) (τ ) + O ε2 (11.22)

m = ma + εη1 + O ε2
Then substituting (11.22) in (11.19) and (11.20), and matching the coeffi-
cients of the first order of ε, gives equation

2
2
2ω 2ω
LX (1)
− (1)
ma X = −f (Aea ) + η1 Aea (11.23)
π π
and boundary condition

X(1) |τ =±1 = 0 (11.24)
Following the idea of perturbations for eigen-value problems [83], let us rep-
resent solution of equation (11.23)
(1)
X(1) = ab eb (τ ) (11.25)
b=a
(1)
where b = {r, i} is a double index, ab are yet unknown constant coefficients,
and boundary condition (11.24) is automatically satisfied.
Let us assume the following normalization condition for the eigen vector-
functions
0, b = a
ea (τ ), eb (τ ) = (11.26)
1, b=a
Substituting (11.25) in (11.23) and taking into account (11.26), determines
(1)
the coefficients ab and η1 . As a result, expansions (11.22) give first-order
asymptotic solution
π 2 e , f (Ae ) e
b a b
X = Aea + ε + O ε2 (11.27)
2ω mb − ma
b=a
π 2 e , f (Ae )
a a
m = ma − ε + O ε2
2ω A
As follows from the form of solution (11.27), all the coefficients are uniquely
determined under the condition that ma = mb for a = b. The possibility of
degeneration, namely ma = mb for a = b, depends on the inner properties
of the system and the frequency parameter ω. The related examples were
considered earlier [136], [141].
11.5 Definition for Principal Modes of Continuous

Systems
Let us consider a one-dimensional elastic system whose vibration is described
by some function u = u(t, y). For certainty reason, let us consider a
non-linear string of the length l under external distributed loading described

by the partial differential equation and boundary conditions
Lu+εf [u] = p(ωt, y), 0<y<l (11.28)
u(t, 0) = u(t, l) = 0 (11.29)

2 2
∂ ∂
L ≡ ρ(y) 2
−T 2 (11.30)
∂t ∂y
where L is the differential self-adjoint operator of linear string, ρ(y) is a mass
per unit length parameter, T is a constant tensile force, f [u] is a nonlinear
operator acting in the corresponding function space of configurations, ε is
a small parameter, and p(ωt, y) is the external forcing function, which is
assumed to be 2π-periodic with respect to ωt.
Now keeping in mind expressions (11.28) through (11.30), let us introduce
Definition 2. Periodic forced vibrations of a continuous system, in which
the system motion is equivalent to a particle in the function space of config-
urations described by the second Newton law,
∂ 2 u(t, y)
σ = p(ωt, y) (11.31)
∂t2
will be called a principal mode of forced vibration.
In one-dimensional cases, σ is a priory unknown effective mass per unit length.
Substituting (11.31) in (11.28), gives the following partial differential equa-
tion for principal modes of forced vibrations
∂2u
Lu+εf [u] = σ (11.32)
∂t2
Introducing the triangular wave time substitution as τ = τ ((2ω/π)t) and
u(t, y) = U (τ, y), gives

2
2ω ∂2U
LU +εf (U ) = σ
π ∂τ 2

2
2ω ∂2 ∂2
L≡ ρ(y) − T (11.33)
π ∂τ 2 ∂y 2
The boundary conditions are formulated for both temporal and spatial vari-
ables as
U (τ, 0) = U (τ, l) = 0 (11.34)
and,
∂U (τ, l)
|τ =±1 = 0 (11.35)
∂τ
respectively.
11.5 Definition for Principal Modes of Continuous Systems 273
In this case, the scalar product of two functions U = U (τ, y) and V =

V (τ, y) from the configuration space can be defined as
1 l
1
U ,V = U V dτ dy (11.36)
2l −1 0
Further, a weakly nonlinear asymptotic procedure can be developed analo-

gously to the above discrete case.
Chapter 12
NSTT and Shooting Method for
Periodic Motions
Abstract. In this chapter, two-dimensional shooting diagrams are introduced

for visualization of manifolds of periodic solutions and their bifurcations. A
general class of nonlinear oscillators under smooth, nonsmooth, and impulsive
loadings is considered. The corresponding boundary value problems are for-
mulated by introducing the triangular wave temporal argument. The Duffing
oscillator with no linear stiffness (Ueda circuit) is considered for illustration.
It is shown that the temporal mode shape of the loading is responsible for
qualitative features of the dynamics, such as transitions from regular and
random motions. The important role of unstable periodic orbits is discussed.
12.1 Introductory Remarks
Periodic solutions and their bifurcation diagrams often reveal important qual-
itative features of the dynamics even though the system motion is not ex-
pected to be periodic. In particular, the number of periodic orbits, their
distribution and properties reveal the structure of chaotic orbits; see, for in-
stance, works [12], [53], [123], and references therein. Direct numerical tools
for detection and construction of periodic orbits based on the mapping ap-
proach can be found in the book [125] and paper [54]. Different formulations
in terms of boundary value problems for ordinary differential equations are
described in [10]. Theoretical and applied results regarding periodic motions,
bifurcations and chaos are reported in [13] and [196].
In this chapter, a special two-dimensional visualization of the shooting
method is introduced in order to incorporate the general two component
NSTT as a preliminary analytical stage [144]. Note that the same approach
using the one-component NSTT was suggested earlier in [154] and imple-
R
mented in Mathematica interface [18]. In particular, subharmonic orbits of
the forced pendulum and bifurcation diagrams were obtained by examining
the shooting curves and their zeros.

276 12 NSTT and Shooting Method for Periodic Motions
Let us consider a multi-dimensional oscillator described be the differential

equation
F (ẍ, ẋ, x, t) = 0 (12.1)
where x (t) ∈ Rn , and the vector-function F ∈ Rn is periodic with respect to
time t with the period T = 4a.
In this work, different kind of temporal discontinuity in the differential
equations of motion will be considered. In order to satisfy the related math-
ematical requirements, the left-hand side of equation (12.1) must be inter-
preted in terms distributions [178]
∞
F (ẍ, ẋ, x, t)ϕ (t) dt = 0 (12.2)
−∞
where ϕ (t) is any sufficiently smooth testing function.

However, at this point, let us assume that the function F (ẍ, ẋ, x, t) is
regular with no singular terms involved.
Let us consider periodic solutions of the period T by means of the coordi-
nate complexification (NSTT)
x → {X, Y } : x = X (τ ) + Y (τ ) τ (12.3)
whereτ = τ (t/a) the triangular sine wave of the period T = 4a, and
τ = dτ (t/a) /d (t/a) is its first generalized derivative, which is a step-wise
discontinuous function at the time instances
Λ = {t : τ (t/a) = ±1} (12.4)
As discussed in this book, the above discontinuities can be suppressed by

the condition Y (±1) = 0, which is the necessary condition of continuity of
the original coordinates x(t). Then, assuming for a while no infinite discon-
tinuities in F , substituting (12.3) in (12.1), using the algebraic properties of
representation (12.3) as well as other NSTT rules, gives the boundary value
problem

X + Y X + Y X + Y
F , , , aτ = 0 (12.5)
a2 a a

X − Y X − Y X − Y
F , − , , 2a − aτ = 0 (12.6)
a2 a a
Y |τ =±1 = 0, X |τ =±1 = 0 (12.7)

where the prime used with X and Y means differentiation with respect to τ .
Both equations (12.5) and (12.6) are easily derived by the corresponding
algebraic manipulations, whereas boundary conditions (12.7) represent the
result of elimination of the singular term τ (t/a) when substituting (12.3) in
12.2 Problem Formulation 277
(12.1). In some cases, such a singular term can be employed though in order
to eliminate singularities from original equations; see below.
Despite of a relatively complicated form of equations (12.5) and (12.6),
the new formulation brings some advantages due to the fact that the new
temporal variable τ is bounded and automatically accounts for periodicity
of solutions regardless their temporal shapes. This property appears to be
important in those cases when the solutions do not represent a final stage of
investigation but must be used for further analyses. The dimension increase
is often compensated by an effective decrease of the temporal interval of
the problem, since the range −1 τ 1 is covered by the original time
domain −a t a, which is twice shorter than the whole period T =
4a. Moreover, there are many cases when the number of equations can be
reduced to that of the original system due to the symmetry of equations.
If, for instance, the vector-function F (ẍ, ẋ, x, t) is even with respect to the
velocity ẋ or includes no velocity at all, and the explicit dependence on time
t produces zero ‘imaginary component’ then boundary value problem (12.5)
through (12.7) admits a family of solutions on which

X X X
Y ≡ 0, F , , , aτ = 0 (12.8)
a2 a a
X |τ =±1 = 0 (12.9)
The particular case (12.8) and (12.9) was investigated numerically by the
shooting method in [154] and [177] based on a single- and multiple-degrees-of
freedom systems, respectively. It should be noted that no special require-
ments are imposed on numerical methods or packages for solving the above
boundary-value problems. However, the shooting algorithm in the Mathe-

R
matica interface provides a physically meaningful way of visualization of
periodic solutions due to the specific combination analytical and numerical
commands.
12.2 Problem Formulation

Let assume now that the system loading may include a periodic series of
Dirac δ-pulses acting at times Λ(12.4). As is known [45], Dirac δ-functions
can participate in nonlinear differential equations only as summands because
nonlinear manipulations with δ-functions are physically meaningless, except
special concepts [106]. Therefore, the original equation (12.1) must be con-
cretized as
ẍ + f (x, ẋ, t) = q (t) (12.10)
where
q (t) = Q (τ (t/a)) + P (τ (t/a)) τ (t/a) + p (τ (t/a)) τ (t/a) (12.11)

and
τ (t/a) = d2 τ (t/a) /d (t/a)2 (12.12)

∞

t t
=2 δ + 1 − 4k − δ − 1 − 4k
a a
k=−∞
In equation (12.10), the function f (x, ẋ, t) may still include parametric terms
of the period T = 4a with possible step-wise discontinuities on Λ. The accel-
eration ẍ also participates as a summand, since it must have the same kind
of singularities as the external forcing function, q (t). According to the distri-
bution theory [165], p (τ (t/a)) must be at least continuous on Λ, otherwise
the ‘product’ p (τ ) τ cannot be treated as a distribution. Note, that behavior
of the function p(τ (t/a)) between the times Λ is arbitrary, since only values
p (−1) and p (1) contribute into the expression
p (τ (t/a)) τ (t/a) (12.13)

∞

t t
=2 p (−1) δ + 1 − 4k − p (1) δ − 1 − 4k
a a
k=−∞
The numbers p (−1) and p (1) control the ‘amplitudes’ and directions of the
δ−pulses. For example, all the pulses can be positively co-directed by setting
p (τ ) = − sign τ .
Remark 2. Expressions (12.3) and (12.11) represent particular cases of the

truncated series

N
q (t) = Pk (τ (t/a)) dk τ (t/a) /d (t/a)k (12.14)
k=0
where Pk (τ (t/a)) must be at least k − 2 times continuously differentiable

in the neighborhood of points t = ±a. Although, physical interpretation of
the higher order terms in (12.14) is not straightforward, such terms still can
occur after reducing the number of equations from the entire system. In case
(12.10) and (12.11), one has N = 2 and therefore the velocity vector ẋ must
be step-wise discontinuous. Further, if N = 3 then the velocity ẋ includes
singular terms and the function f (x, ẋ, t) in equation (12.10) must be linear
with respect to ẋ. If N = 4 then the function f (x, ẋ, t) must be linear also
with respect to the position vector x provided that any parametric terms are
sufficiently smooth functions of time. Therefore, only linear systems can be
considered if N ≥ 4.
Since the basis elements {1, τ , τ } represent functions of different classes of

smoothness, then substituting (12.3) and (12.11) in (12.10), gives separately
a−2 X + Rf (X, Y, X , Y , τ ) = Q (τ ) (12.15)

12.3 Sample Problems and Discussion 279
a−2 Y + If (X, Y, X , Y , τ ) = P (τ ) (12.16)
and
a−2 X |τ =±1 = p (±1) (12.17)
where

Rf 1 X + Y X − Y
= f X + Y, , aτ ± f X − Y, − , 2a − aτ
If 2 a a
(12.18)
Note that the singular term a−1 Y τ is eliminated from the velocity vector
ẋ (t) by imposing another boundary condition
Y |τ =±1 = 0 (12.19)
The boundary value problem (12.15) through (12.19) includes no singular

or discontinuous functions, therefore standard numerical codes and packages
can be applied with no specific constraints on their choice.
In general, equations (12.15) and (12.16) are coupled. Although the
equations can be decoupled by introducing the new unknown functions,
X(τ ) + Y (τ ) and X(τ ) − Y (τ ), the boundary conditions will become cou-
pled. There are two special cases, however, when the entire problem can be
reduced. If, for instance, f (x, ẋ, t) = f (x, −ẋ, 2a − t), and P (τ ) ≡ 0 then
the problem admits a family of solutions such that
Y ≡ 0, a−2 X + f (X, X /a, aτ ) = Q (τ ) (12.20)
under the boundary condition (12.17).

In case f (x, ẋ, t) = −f (−x, ẋ, 2a − t), Q (τ ) ≡ 0 and p (τ ) ≡ 0, then one
can consider another family of solutions on which
X ≡ 0, a−2 Y + f (Y, Y /a, aτ ) = P (τ ) (12.21)
under the boundary condition (12.19).

This chapter nevertheless focuses on the general two-component problem
(12.15) through (12.19).
12.3 Sample Problems and Discussion

12.3.1 Smooth Loading
The Duffing-Ueda oscillator [186] under the periodic loading of different tem-
poral shapes will be considered below.
Let us start with the standard case of sine-wave voltage
ẍ + ζ ẋ + x3 = B sin ωt (12.22)
where ζ, B, and ω are constant parameters.

In this case, the differential equations (12.15) and (12.16) take the form
πτ
a−2 X + ζa−1 Y + X 3 + 3XY 2 = B sin (12.23)
2
a−2 Y + ζa−1 X + Y 3 + 3X 2 Y = 0 (12.24)
where a = π/(2ω) is a quarter of the loading period, and the boundary

conditions are
Y |τ =±1 = 0, X |τ =±1 = 0 (12.25)
The shooting method can be applied now as follows. First, the differential
equations (12.23) and (12.24) are solved under the initial conditions
X(−1) = g, X (−1) = 0 (12.26)

Y (−1) = 0, Y (−1) = h (12.27)
where g and h are numbers to be determined in order to satisfy boundary

conditions (12.25).
Let us represent solution of the initial-value problem (12.23), (12.24),
(12.26) and (12.27) in the following general form
X = X(τ ; g, h), Y = Y (τ ; g, h) (12.28)
By the idea of shooting method, the initial value problem (12.23), (12.24),
(12.26) and (12.27) must be iteratively solved multiple times at different g
and h until sufficient precision has been reached for boundary conditions
(12.25) at right end τ = 1,
∂X(τ ; g, h)
|τ =1 ≡ G(g, h) = 0 (12.29)
∂τ
Y (τ ; g, h)|τ =1 ≡ H(g, h) = 0
When dealing with the particular cases (12.20) or (12.21), such a procedure is
not difficult since one has only one equation with a singe unknown, G(g) = 0
or H(h) = 0. Multidimensional cases, such as (12.29), appear to be more
difficult and time consuming. From this point of view, the important feature
of Mathematica is that it is possible to program the functions G(g, h) and
H(g, h) ‘explicitly’ in such a way that the arguments g and h are included
into the numerical solver of differential equations. This can be done as follows.
First, the numerical solution is defined as a function of the arguments g and
h according to the command
sol[g , h ]:=NDSolve[{eqX, eqY,
X[-1]==g, X [-1] == 0, Y[-1] == 0, Y [-1] == h}, {X,Y}, {τ ,-1,1}];

where eqX and eqY are equations (12.23) and (12.24), respectively.
Then, the functions G(g, h) and H(g, h) are defined as follows
G[g , h ]:=X [1]/.sol[g, h][[1]];
H[g , h ]:=Y[1]/.sol[g, h][[1]];

As a result, the functions G(g, h) and H(g, h) can be considered as usual func-
tions of two arguments. In particular, intersections of two manifolds (12.29)
can be located and determined by using the commands ContourPlot and
FindRoot, respectively. Each of the determined roots of equations (12.29)
represents a periodic solution of the original equation. If the loading ampli-
tude B is a control parameter, then the evolution of diagrams G(g, h; B) = 0
and H(g, h; B) = 0 represents the corresponding structural changes in the
set of periodic solutions.
Fig. 12.1 gives an example of such a diagram. The parameters were chosen
in order to provide conditions for the ‘randomly transitional’ process in terms
of work [186]. The diagram clearly shows five intersections between the two
different families of curves. The corresponding solutions of the input period
T = 4a = 2π are shown in Figs. 12.2 and 12.3.
Direct numerical solutions the corresponding estimates for Floquet multi-
pliers show that first four periodic solutions, (a) through (d), are unstable,
and only one solution (e) is stable. Solution (e) was detected by direct ana-
log and digital computer simulations reported in [186], whereas solutions (a)
through (d) were unlisted. Instead, a ‘non-reproducible trajectory’ as a real-
ization of the ‘randomly transitional’ process was represented in the xv-plane.
Such a trajectory can be treated as a chaotic drift around the first three un-
stable motions (a), (b) and (c). However, high order periodic solutions may
also affect the dynamics of chaotic drift [12].
Fig. 12.4 shows what actually happens when trying to numerically repro-
duce an unstable periodic orbit, say (a). Neither the shooting algorithm nor
computer codes allow to perfectly introduce the initial conditions, therefore
it is unlikely that the oscillator will remain on the unstable orbit. After few
cycles, the system leaves the orbit (a) for the ‘randomly transitional’ drift
around the all three unstable orbits (a), (b) and (c) with ‘no certain choice’
between them. The long-term time history and the corresponding spectro-
gram of this motion, represented in Fig. 12.4, confirm its random character
during quite a long period of time.
Although preliminary qualitative information about stability or instabil-
ity of periodic solutions can be obtained by direct numerical tests, one can
quantify stability properties based on the well known Floquet theory in terms
of the characteristic multipliers [111], [56]. In order to remind the prin-
cipals, let us consider periodic solution x(t) of the equation (12.10),where
f (x, ẋ, t) = f (x, ẋ, t + T ), q (t) = q (t + T ), and the period is T = 4a.
20
10
d
c e
b
h 0
a
10
20
4 2 0 2 4
g
Fig. 12.1 The curves G(g, h) = 0 (continuous) and H(g, h) = 0 (dashed) and their
intersections for the Ueda oscillator under the sine-wave input and the following
parameters: ζ = 0.1, B = 12, and ω = 1 (0.1592 Hz).
A variation of the solution x(t), say u(t), is described by the linear differ-
ential equation with periodic coefficients
ü + q1 (t)u̇ + q2 (t)u = 0
where q1 (t) = ∂f (x, ẋ, t)/∂ ẋ is assumed to be independent on ẋ, and q2 (t) =
∂f (x, ẋ, t)/∂x.
Then substitution
⎛ ⎞
t
1
u = y(t) exp ⎝− q1 (z)dz ⎠
2
0
x x
6 6
a b
4 4
2 2
t t
1 2 3 4 5 6 1 2 3 4 5 6
2 2
4 4
6 6
x x
6 6
c d
4 4
2 2
t t
1 2 3 4 5 6 1 2 3 4 5 6
2 2
4 4
6 6
x
6
e
4
2
t
1 2 3 4 5 6
2
4
6
Fig. 12.2 The temporal mode shapes of periodic solutions.
gives
ÿ + p(t)y = 0 (12.30)
where
1 1
p(t) = q2 (t) − [q1 (t)]2 − q̇1 (t)
4 2
As known from the Floquet theory, stability of the solution x(t) is determined
by the Floquet multipliers

μ1,2 = A ± A2 − 1 (12.31)
v v
6
a
4 5 b
2
0 x 0 x
2
4 5
6
3 2 1 0 1 2 3 3 2 1 0 1 2 3
v v
5 c 5 d
0 x 0 x
5 5
3 2 1 0 1 2 3 3 2 1 0 1 2 3
v
6
e
4
2
0 x
2
4
6
3 2 1 0 1 2 3
Fig. 12.3 The projections of periodic trajectories on xv-planes.
where A = [y1 (T ) + ẏ2 (T )]/2, and y1 (t) and y2 (t) are two fundamental solu-
tions of equation (12.30) given by the initial conditions
y1 (0) = 1, ẏ1 (0) = 0

y2 (0) = 0, ẏ2 (0) = 1
Based on the number A, the solution x(t) is unstable if A2 > 1, and it is

stable if A2 < 1. In case A2 = 1, there exist a periodic solution of equation
(12.30).
0
x
−5
50 100 150 200 250 300
Time
2.5
2
Frequency
1.5
0.5
0
0 10 20 30 40 50
Time
Fig. 12.4 The time history record and its spectrogram (in Hz) for Ueda oscillator
after the direct numerical integration. The parameters are: ζ = 0.1 and B = 12.0.
Now, let x(t) be a periodic solution of equation (12.22). The corresponding

variational equation is
ü + ζ u̇ + 3x2 u = 0 (12.32)
where u = u(t) is a small variation of the solution x = x(t).
After the standard substitution u(t) = y(t) exp(−ζt/2), equation (12.32)
takes the form

ζ2
ÿ + 3x −2
y=0
4
Taking into account the form of solution (12.3), gives the variational equation
with periodic coefficient
ÿ + [U (τ (t/a)) + V (τ (t/a))τ (t/a)]y = 0 (12.33)
where U (τ ) = 3X 2 (τ ) + 3Y 2 (τ ) − ζ 2 /4 and V (τ ) = 6X(τ )Y (τ ).

Note that the periodic coefficient in equation (12.32) is continuous with
respect to time t since V (±1) = 0 due to the boundary conditions (12.25).
By using the numerical solutions of equation (12.33), one obtains the num-
ber A for every solution
Aa = 10.5155, Ab = −2.63747, Ac = −2.63749
Ad = 1.70201, and Ae = 0.143507 (12.34)

where the index correspond to the type of periodic solution of the original
equation; see Figs. 12.1, 12.2, and 12.3. These numbers confirm that only
solution (e) is stable.
12.3.2 Step-Wise Discontinuous Input

Let us consider now the case of discontinuous periodic input of the rectangular
cosine temporal shape
dτ (t/a)
ẍ + ζ ẋ + x3 = B (12.35)
d (t/a)
where a is a quarter of the input period.

In this case, the right-hand side of equations (12.23) and (12.24) are mod-
ified so that the equations take the form
a−2 X + ζa−1 Y + X 3 + 3XY 2 = 0 (12.36)

a−2 Y + ζa−1 X + Y 3 + 3X 2 Y = B (12.37)
under the homogeneous boundary conditions (12.25).

Fig. 12.5 shows the shooting diagram under the fixed parameters ω =
π/(2a) = 1 (0.1592 Hz), B = 7.4 and ζ = 0.05. In this case, there are seven
intersections between the two families of curves and therefore seven periodic
solutions of the period T = 4a as shown in Figs. 12.6 and 12.7.
Note that, under the same parameters, the system response on the rectan-
gular cosine input shows new features compared to those under the sine-wave
input [144]. For example, after few cycles along the orbit (a), the system starts
its drift around the first three solutions, (a), (b) and (c). At this stage, the
dynamics resembles that under the sine-wave input. Further, however, af-
ter several random ‘jumps’ between the orbits (a), (b) and (c), the system
becomes eventually attracted by the stable orbit (e). The direct numerical
solution, represented in Fig. 12.8, clearly shows all three stages of the time
and spectral histories of the dynamics.
In order to clarify stability, the Floquet theory can be applied analogously
to the case of the sine-wave input.
12.3.3 Impulsive Loading

Let us consider the same oscillator loaded by the periodic series of pulses
ẍ + ζ ẋ + x3 = pτ (12.38)
∞

t t
= 2p δ + 1 − 4k − δ − 1 − 4k
a a
k=−∞
where p = p (±1) = B = const.

40
20
h 0
a
c b
20
40
4 2 0 2 4
g
Fig. 12.5 The curves G(g, h) = 0 (continuous) and H(g, h) = 0 (dashed) for the
Ueda oscillator under the step-wise input and the following parameters: ζ = 0.05,
B = 7.4, and ω = 1.
In this case, both equations (12.36) and (12.37) should have zero right-
hand side, however, the non-homogeneous version of the boundary con-
dition (12.17) must be imposed in order to eliminate the pulses. The
second expression in (12.26) and the first one in (12.29) must be modified
as X (−1) =a2 B=[π/(2ω)]2 B and G (g, h)=a2 B=[π/(2ω)]2 B, respectively.
Therefore, the singular terms are eliminated from the system due to the saw-
tooth time, and the shooting procedure can be applied in the same fashion as
that under the smooth input. The shooting diagram and the corresponding pe-
riodic solutions are shown in Figures 12.9, and 12.10 and 12.11, respectively.
The projections of the phase trajectories show discontinuities of the veloc-
ity on the xv-plane caused by the external pulses. The last four projections,
x x
6 6
a b
4 4
2 2
t t
1 2 3 4 5 6 1 2 3 4 5 6
2 2
4 4
6 6
x x
6 6
c d
4 4
2 2
t t
1 2 3 4 5 6 1 2 3 4 5 6
2 2
4 4
6 6
x x
6 6
e f
4 4
2 2
t t
1 2 3 4 5 6 1 2 3 4 5 6
2 2
4 4
6 6
x
6
g
4
2
t
1 2 3 4 5 6
2
4
6
Fig. 12.6 The temporal mode shapes of periodic solutions og Ueda circuit under
the step-wise input.
(h) through (k), can be qualified as ‘quasi free’ vibrations sustained by

the pulses. In the shooting diagram represented in Fig. 12.9, the related
v v
6
4
a 4
b
2 2
0 x 0 x
2 2
4
4
6
3 2 1 0 1 2 3 3 2 1 0 1 2 3
v v
10
6
4 5
c d
2
0 x 0 x
2
4 5
6
10
3 2 1 0 1 2 3 4 2 0 2 4
v v
10
20
5 10
e
0 x 0 f x
5 10
20
10
4 2 0 2 4 6 4 2 0 2 4 6
20
10
g
0 x
10
20
6 4 2 0 2 4 6
intersections are difficult to determine due to a very small angle between the
intersecting curves.
0
x
−5
50 100 150 200 250 300
Time
2.5
2
Frequency
1.5
0.5
0
0 10 20 30 40 50
Time
Fig. 12.8 The time history record and its spectrogram (in Hz) for evolution of the
solution (a) under the step-wise voltage of the amplitude B = 7.4.
12.4 Other Applications

12.4.1 Periodic Solutions of the Period - n
The above sections deal with periodic solutions with the input period T = 4a.
In order to capture ‘subharmonic’ solutions of the period nT , the components
of representation (12.3) must be taken in the form
1
X (τ ) = [x (naτ ) + x (2na − naτ )]
2
1
Y (τ ) = [x (naτ ) − x (2na − naτ )] (12.39)
2
where τ = τ (t/(na)).
For instance, applying (12.39) to the sine wave sin ωt, gives
1 nπτ nπτ
sin ωt = sin + sin nπ −
2 2 2
1 nπτ nπτ
+ sin − sin nπ − τ (12.40)
2 2 2
1 nπτ 7 8
= sin 1 + (−1)n+1 + 1 − (−1)n+1 τ
2 2
where τ = dτ (t/(na))/d(t/(na)) and a = π/(2ω).
12.4 Other Applications 291
30
j k
20
10
g
i d h
a b
h 0 c
e
f
10
20
30
6 4 2 0 2 4 6
g
Fig. 12.9 The curves G(g, h) = 0 (continuous) and H(g, h) = 0 (dashed) for Ueda
circuit under the impulsive input and parameters: ζ = 0.05, B = 1.4, and ω = 1.
According to representation (12.40), equations (12.23) and (12.24) must

be modified as follows
(na)−2 X + ζ(na)−1 Y + X 3 + 3XY 2

B nπτ
= 1 + (−1)n+1 sin
2 2
(na)−2 Y + ζ(na)−1 X + Y 3 + 3X 2 Y (12.41)
B nπτ
= 1 − (−1)n+1 sin
2 2
x x x
a 2 b 2
1.5
1.0 c
1 1
0.5
t t t
0.5 1 2 3 4 5 6 1 2 3 4 5 6 1 2 3 4 5 6
1.0 1 1
1.5 2 2
x x x
d e
2 2 2 f
1 1 1
t t t
1 2 3 4 5 6 1 2 3 4 5 6 1 2 3 4 5 6
1 1 1
2 2 2
x x x
g 3 h 3 i
2
2 2
1 1 1
t t t
1 2 3 4 5 6 1 1 2 3 4 5 6 1 1 2 3 4 5 6
1
2 2
2 3 3
x x
6 6
4 j k
4
2 2
t t
1 2 3 4 5 6 1 2 3 4 5 6
2 2
4 4
6 6
Fig. 12.10 The temporal mode shapes of periodic solutions og Ueda circuit under
the impulsive input.
The right-hand side of these equations shows that direct replacement a → na

in (12.23) and (12.24) would not work.
If n = 1 then equations (12.41) take the form (12.23) and (12.24), but if
n > 1 equations (12.41) can give new solutions in addition to those described
by equations (12.23) and (12.24). The corresponding calculations however
become time consuming and give complicated diagrams as the number n
increases.
12.4 Other Applications 293
v v v
4 4
2
a 2 b c
1 2
0 x 0 x 0 x
1 2 2
2 4 4
1.5
1.0
0.50.00.51.01.5 2 1 0 1 2 2 1 0 1 2
v v v
4 4
e 4 f
2 d 2 2
0 x 0 x 0 x
2 2
2
4
4 4
6
2 1 0 1 2 2 1 0 1 2 2 1 0 1 2
v v v
6 10
4 g i
5 h 5
2
0 x 0 x 0 x
2 5 5
4
10
2 1 0 1 2 321 0 1 2 3 321 0 1 2 3
v v
20 20
j k
10 10
0 x 0 x
10 10
20 20
6 4 2 0 2 4 6 6 4 2 0 2 4 6
12.4.2 Two-Degrees-of-Freedom Systems

Using the above two-dimensional geometrization of shooting diagrams enables
one of considering special cases of two-degrees-of-freedom systems based on
equations (12.20) or (12.21). For example, equation (12.20) can be treated
as an equation with respect to the two-component vector-function X =
{X1 (τ ), X2 (τ )}. Such an interpretation leads to two scalar equations
a−2 X1 + f1 (X1 , X2 , X1 /a, X2 /a, aτ ) = Q1 (τ )

a−2 X2 + f2 (X1 , X2 , X1 /a, X2 /a, aτ ) = Q2 (τ ) (12.42)
In this case, the shooting procedure should be based on the initial conditions
at τ = −1,
X1 (−1) = g, X1 (−1) = 0

X2 (−1) = h, X2 (−1) = 0 (12.43)
where the numbers g and h are determined to satisfy the boundary conditions
on the right end of the interval −1 ≤ τ ≤ 1,
∂X1 (τ ; g, h)
|τ =1 ≡ G(g, h) = 0
∂τ
∂X2 (τ ; g, h)
|τ =1 ≡ H(g, h) = 0 (12.44)
∂τ
In this case, every solution g and h of system (12.44) gives the initial position
on the configuration plane X1 X2 at which the system starts with zero velocity
its periodic motion of the period T = 4a.
12.4.3 The Autonomous Case

The nonlinear normal modes represent an important class of periodic mo-
tions. The related references and description of analytical methods can be
found in [100] and [190]. Analogously to the linear theory, the basic nonlin-
ear normal mode solutions are given by the class of autonomous conservative
systems. In this case, equations (12.42) take the form
a−2 X1 + f1 (X1 , X2 ) = 0

a−2 X2 + f2 (X1 , X2 ) = 0 (12.45)
The form of equations (12.45) is easier than (12.42), but the parameter a
becomes unknown. It is possible to avoid determining the parameter a by
considering it as a control parameter for tracking the evolution of shooting
diagrams. Alternatively, the parameter a can be considered as a shooting
parameter by imposing one constraint on the parameters g or h. Let us con-
sider, for instance, the system trajectories
in the configuration plane X1 X2 .
Introducing the amplitude A = g 2 + h2 in (12.43), gives
X1 (−1) = A cos ϕ
X2 (−1) = A sin ϕ
where the angle ϕ (0 ≤ ϕ < 2π) together with the parameter a can play the
role of a new unknowns to be determined by shooting whereas the amplitude
A is considered as a control parameter.
Chapter 13
Essentially Non-periodic Processes
Abstract. This chapter describes a possible physical basis for NSTT in case
of essentially non-periodic processes. The physical time is structurised to
match the one-dimensional dynamics of rigid-body chain of identical particles.
Namely, the continuos ‘global’ time is associated with the propagation of
linear momentum, whereas a sequence of non-smooth ‘local’ times describe
behavious of individual physical particles. Such an idea helps to incorporate
temporal symmetries of the dynamics into differential equations of motion
in many other cases of regular or irregular sequences of internal impacts
or external pulses. Since the local times are bounded, a much wider set of
analytical tools becomes possible, wereas matching conditions are generated
automatically by the corresponding time substitution.
13.1 Nonsmooth Time Decomposition and Pulse

Propagation in a Chain of Particles
The periodic version of NSTT employs basis functions generated by the most
simple impact oscillator. This is based the fact that the triangular sine and
rectangular cosine waves capture general temporal symmetries of periodic
processes regardless specifics of individual vibrating systems. Below, a non-
periodic pair of nonsmooth functions is considered, such as the ramp function,
1
s (t; d) = (d + |t| − |t − d|) (13.1)
2
and its first order generalized derivative, ṡ (t; d), with respect to the temporal
argument, t; see Figs.13.1 and 13.2, respectively.
Such kind of functions play an important role in signal analyses [74].
Possible physical interpretation of these functions is represented in Fig.
13.3. Namely, the function s (t, d) can be treated as a coordinate of a particle,
say a very small perfectly stiff bead, initially located at the origin x = 0. At
the time instance t = 0, this bead is struck by the identical bead with the
velocity v = 1. After the linear momentum exchange, the reference bead
starts moving until it stopped by the third bead x = d; in our case d = 1.

296 13 Essentially Non-periodic Processes
s
1.0
0.8
0.6
0.4
0.2
t
1.0 0.5 0.5 1.0 1.5 2.0
Fig. 13.1 The unit slope ramp function at d = 1.0
s
1.0
0.8
0.6
0.4
0.2
t
1.0 0.5 0.5 1.0 1.5 2.0
Fig. 13.2 First derivative of the ramp function
Fig. 13.3 Physical meaning of the ramp function: s(t; 1) describes position of the
bead struck by another bead from the left and moving until it strikes the next bead
of the same mass.
Now, let us consider an infinite chain of the identical perfectly stiff beads
located regularly on a straight line at the points xi (i = 0, 1, ...). No energy
loss is assumed so that any currently moving bead has the same velocity. As
a result, the linear momentum is translated with the constant speed v = 1,
whereas the beads are interacting at the time instances ti = xi . Making the
temporal shift t → t − ti in function (13.1), gives
1
si (t) = s (t − ti , di ) = (di + |t − ti | − |t − ti+1 |) (13.2)
2
where di = ti+1 − ti .
13.1 Nonsmooth Time Decomposition and Pulse Propagation 297
Due to the unit velocity, function (13.2) can play the role of ‘local’ time for
the bead moving within the interval xi < x < xi+1 during the ‘global’ time
interval ti < t < ti+1 . The term “local” means that the temporal variable si
starts at zero when the “global” time, t, has reached the point t = ti .
In other words, the global temporal variable is associated with the linear
momentum, whereas all the local temporal variables are “attached” to the
physical bodies.
For any sequence of time instances, Λ={t0 , t1 , ...}, the global time, t ∈
[t0 , ∞), can be expressed through the sequence of local times, {si }, in the
form
∞
t= (ti + si ) ṡi (13.3)
i=0
where the derivatives ṡi satisfy the relationship
ṡi ṡj = ṡi δij (13.4)
Practically, (13.3) is always a finite sum because temporal intervals of physical

processes always have finite upper bounds.
Equality (13.3) can be easily verified within the arbitrary interval ti < t <
ti+1 , by using definitions (13.1) and (13.2), but the set of boundary points Λ
still require some attention. So formally differentiating both sides of equality
(13.3) with respect to t and taking into account (13.4), gives
∞
∞

1= ṡi + (ti + si ) [δ (t − ti ) − δ (t − ti+1 )] (13.5)
i=0 i=0
∞
= ṡi + t0 δ (t − t0 )
i=0
where the relationship (ti+1 + si+1 (ti+1 ) − ti − si (ti+1 )) = 0 has been used
for calculations.
Therefore, for almost all t > t0 , expression (13.5) gives
∞

1= ṡi (13.6)
i=0
Equality (13.6) holds based on the definition for ṡi as illustrated by Fig. 13.2,
therefore expansion (13.3) also holds for almost all t > t0 at least, and its
time derivative includes no singular functions.
Note that the right-hand side of expansion (13.3) can be viewed as an
element of algebra with the basis {si } and multiplication rule (13.4). This
significantly eases different manipulations with the temporal variable (13.3),
for instance
∞

n n
t = (ti + si ) ṡi , n = 1, 2, ... (13.7)
i=0
or, generally,
∞
∞

x (t) = x (ti + si ) ṡi = Xi (si ) ṡi (13.8)
i=0 i=0
Since the right-hand sides of (13.7) and (13.8) have the same structure as the
argument t itself, then considering further a general function g, gives
∞

g (x) = g (Xi ) ṡi (13.9)
i=0
Now, differentiating (13.8) with respect to time t, and taking into account
that si (ti ) = 0 and si−1 (ti ) = di−1 , gives
∞
∞

ẋ (t) = Xi (si ) ṡi + Xi (si ) [δ (t − ti ) − δ (t − ti+1 )] (13.10)
i=0 i=0
∞ ∞
= Xi (si ) ṡi + [Xi (0) − Xi−1 (di−1 )] δ (t − ti )
i=0 i=0
where X−1 (d−1 ) = 0.

Therefore, all the δ- pulses are eliminated from (13.10) under condition,
which can be qualified as a necessary condition of continuity for x (t),
Xi (0) − Xi−1 (di−1 ) = 0 (13.11)
Under condition (13.11), the derivative ẋ (t) has the same algebraic structure
as the function x (t) itself. As a result, transformation (13.3) can be applied to
a general class of dynamical systems. Moreover, in case of impulsively loaded
systems, the sequences of δ-pulses in (13.10) can be utilized for eliminating
the corresponding singularities from dynamical systems.
13.2 Impulsively Loaded Dynamical Systems

Let us consider a dynamical system subjected to an arbitrary sequence of
δ-impulses, applied to the system at time instances Λ = {t0 , t1 , ...},
∞

ẋ = f (x, t) + pi δ (t − ti ) , x (t) ∈ Rn (13.12)
i=0
x ≡ 0, t < t0 (13.13)
where f (x, t) is a sufficiently smooth vector-function, pi are vectors charac-

terizing magnitudes and directions of the impulses.
13.2 Impulsively Loaded Dynamical Systems 299
In particular case, when t0 = 0, and pi = 0 (i = 1, ... ), system (13.12)

and (13.13) becomes equivalent to the following initial value problem
ẋ = f (x, t) (13.14)
x (0) = p0 (13.15)
Below, solution of the initial value problem (13.12) and (13.13) is introduced
in the specific form based on the operator Lie associated with dynamical
system (13.15)
∂ ∂
A = f (x, t) + (13.16)
∂x ∂t
∂ ∂ ∂
= f1 (x, t) + .... + fn (x, t) +
∂x1 ∂xn ∂t
It is known, for instance, that the exponent of operator (13.16) produces
temporal shifts as follows
ezA f (x (t) , t) = f (x (t + z) , t + z) (13.17)

∂f (x, t) ∂ (x, t)
= f (x, t) + f (x, t) + z + O(z 2 )
∂x ∂t
Proposition 6. Solution of the initial value problem (13.12) and (13.13) can
be represented in the form
∞

x (t) = [ai−1 + pi + F (ai−1 + pi , ti , si (t))] ṡi (t) (13.18)
i=0
where ai = x (ti+1 ) is the sequence of constant vectors determined by the

mapping
a−1 = 0 (13.19)
ai = ai−1 + pi + F (ai−1 + pi , ti , di ) ; i = 1, 2, ...
and the function F is defined by

z
F (x, t, z) = ezA f (x, t) dz (13.20)
0
where A is the operator Lie (13.16).

Proof. Substituting vector analogs of expressions (13.3), (13.8) and (13.10)
into the differential equation of motion (13.12) and taking into account (13.9),
gives
∞

{[Xi (si ) − f (Xi (si ) , ti + si )]ṡi + (13.21)
i=0
[Xi (0) − Xi−1 (di−1 ) − pi ]δ (t − ti )} = 0 (13.22)

The left-hand side of expression (13.22) includes both regular and singular
terms. Moreover, the basis elements ṡi are linearly independent, and all the
δ-pulses are acting at different time instances.. Therefore, equation (13.22)
gives
Xi (si ) = f (Xi (si ) , ti + si ) (13.23)
Xi (0) = Xi−1 (di−1 ) + pi = ai−1 + pi (13.24)
where a−1 = 0 and ai = Xi (di ) (i = 0, 1, 2, ...).
Equation (13.23) can be represented in the integral form
si
Xi (si ) = Xi (0) + f (Xi (z) , ti + z) dz (13.25)
0
Since the variable of integration is limited by the interval 0 ≤ z ≤ si , the

integrand in (13.25) can be approximated by the easy to integrate Maclaurin’s
series with respect to z. Moreover, such a series can be represented in the
convenient form of Lie series based on the fact that Xi (z) are coordinates
of the dynamical system with the operator Lie (13.16). As a result, all the
coefficients of power series are expressed through the “initial conditions” at
z = 0 (t = ti ) by enforcing the form of the dynamical system. This eliminates
all high-order time derivatives from the coefficients of the power series. means
of the right-hand side of the dynamical system; no high order derivatives of
the coordinates are included any more into the coefficients of the series.
So, taking into account the notation Xi (si ) = x (ti + si ), and expressions
(13.17) and (13.20), brings (13.25) to the form
si
Xi (si ) = Xi (0) + ezA f (x (ti ) , ti ) dz
0
= Xi (0) + F (x (ti ) , ti , si ) (13.26)
= Xi (0) + F (Xi (0) , ti , si )
Substituting now Xi (0) from (13.24) in (13.26), gives
Xi (si ) = ai−1 + pi + F (ai−1 + pi , ti , si ) (13.27)
Finally, substituting (13.27) in expansion (13.8), gives (13.18). Then, substi-

tuting si = di in (13.27) gives (13.19) and thus completes the proof.
Solution (13.18) and (13.19) should be viewed as a semi-analytic solution,
since some numerical tool is required for calculating the discrete mapping
(13.19). The central role here belongs to the function s (t; d) (13.1), which is
automatically matching all the neighboring pieces of the solution.
Note that the distances di between times Λ are not necessary small, how-
ever, the precision of the solution can be improved by increasing the number
of terms of the Lie series ezA f (x, t) with respect to z, rather than reducing
the distances di .
13.2.1 Harmonic Oscillator under Sequential

Impulses
In order to estimate precision of the above procedure, let us consider the
particular case in which function (13.20) can be calculated exactly in the
closed form due to the presence of exact analytical solution in between the
pulses Λ.
The differential equation of motion on the entire time range is
∞

ẍ + 2ζω ẋ + ω 2 x = pi δ (t − ti ) (13.28)
i=0
In this case, the function f (x, t) in equation (13.12) becomes

x2
f (x) = (13.29)
−2ζωx2 − ω 2 x1
Using the identity ezA f (x (t) , t) = f (x (t + z) , t + z) and the exact analytical

solution of the corresponding free oscillator, gives both components of the
vector-function (13.20) in the form
) )
−z ζ ω ζe−z ζ ω
F1 (x; z) = e cos(z 1 − ζ ω) +
2
sin(z 1 − ζ ω) − 1 x1
2
1 − ζ2
)
e−z ζ ω
+ sin(z 1 − ζ 2 ω) x2
ω 1 − ζ2
)
ωe−z ζ ω
F2 (x; z) = − sin(z 1 − ζ 2 ω) x1 (13.30)
1−ζ 2
) )
−z ζ ω ζ e−z ζ ω
+ e cos(z 1 − ζ ω) −
2
sin(z 1 − ζ ω) − 1 x2
2
1 − ζ2
In this particular case, properties of mapping (13.19) depend on the following

determinant

1 + ∂F1 /∂x1 ∂F1 /∂x2
J = = e−2di ζω (13.31)
∂F2 /∂x1 1 + ∂F2 /∂x2
Let us introduce the relative error
δ = |J − Jappr | /J (13.32)
where Jappr is an approximate determinant based on the Lie series expansion

(13.17).
Figs. 13.4 and 13.5 show diagrams for the relative error δ versus the dis-
tance d between any two neighboring impulse times when the highest order
terms kept in Lie series (13.17) are O(z 2 ) and O(z 3 ), respectively.
0.030
0.025
0.020
Δ 0.015
0.010
0.005
0.000
0.0 0.2 0.4 0.6 0.8
d
Fig. 13.4 Relative error of the determinant based on the truncated Lie series
including terms of order O(z 2 ).
Fig. 13.5 Relative error of the determinant based on the truncated Lie series
including terms of order O(z 3 ).
As follows from the diagrams, precision of the discrete mapping essentially

depends on both the distance between pulse times and the number of terms
kept in the Lie series. As a result, the error due to a large distance can be
reduced by increasing the number of terms in the Lie series.
13.2.2 Random Suppression of Chaos

A specific case of the Duffing oscillator with no linear stiffness under sine
modulated random impulses was considered in [142]. The corresponding dif-
ferential equation of motion is represented in the form
∞

ẍ + ζ ẋ + x3 = B sin t δ (t − ti ) (13.33)
i=0
where ζ is a constant linear damping coefficient, and B is the amplitude of

modulation.
Distances between any two sequential impulse times are given by
π
di = ti+1 − ti = (1 + βηi )
12
where ηi are random real numbers homogeneously distributed on the interval
[−1, 1], and β is a small positive number, 0 < β 1.
Introducing the state vector x = (x, ẋ)T ≡ (x1 , x2 )T , brings system (13.33)
to the standard form (13.12), where

x2 0
f (x) = , p =
−ζx2 − x31 i
B sin ti
Note that oscillator (13.33) represents of-course a modified version of the

well known oscillator, ẍ + ζ ẋ + x3 = B sin t, considered first by Ueda [186]
as a model of nonlinear inductor in electrical circuits - the Ueda circuit. In
particular, the result of work [186], as well as many further investigations of
similar models, reveal the existence of stochastic attractors often illustrated
by the Poincare diagrams [113]. Similar diagrams obtained under non-regular
snapshots can be qualified as ‘stroboscopic’ diagrams.
The results of the computer simulations described in [142] show that some
irregularity of the pulse times can be used for the purposes of a more clear ob-
servation of the system orbits in the stroboscopic diagrams. When repeatedly
executing the numerical code, under the same input conditions, such a small
disorder in the input results some times in a less noisy and more organized
stroboscopic diagrams. However, such phenomenon itself was found to be a
random event whose ‘appearance’ depends on the level of pulse randomization
as well as the number of iterations.
Chapter 14
Spatially-Oscillating Structures
Abstract. This chapter illustrates applications of nonsmooth argument sub-

stitutions to modeling spatially oscillating structures such as one-dimensional
elastic rods with periodic discrete inclusions, and two- or three-dimensional
acoustic media with periodic nonsmooth boundary sources of waves. When-
ever the corresponding global spatial domains are infinite or cyclical, the
related analytical manipulations are similar to those conducted with dynam-
ical systems. The idea of averaging is implemented through the two-variable
expansions, where the fast scale is represented by the triangular periodic
wave. Such an approach results in closed form analytical solutions despite of
the presence discrete inclusions or external discontinuous loads.
14.1 Periodic Nonsmooth Structures

Static and dynamic problems of elasticity dealing with non-smooth periodic
structures are often considered in the literature due to their practical impor-
tance; see recent reviews [76] for introduction and references. Such problems
can also be considered by means of the non-smooth argument transformations.
In this case though, the transformation must be applied to the spatial inde-
pendent variable related to coordinate along which the structure under con-
sideration is periodic. Such an approach was introduced in [135] for a string on
discrete elastic foundation, although the complete description of the tool was
given later for the corresponding nonlinear case [151], [171], [152],
∂2u ∂2u
∞
y y
ρ 2 − T 2 + 2f (u) δ − 1 − 2k = q , y, t (14.1)
∂t ∂y ε ε
k=−∞
−∞ < y < ∞
where ρ is the mass density per unit length, T is tension, q(y/ε, y, t) is the
body force or external loading, which is assumed to be periodic in the ‘fast
scale’ y/ε (0 < ε 1) with the period normalized to four, and similar
assumption is made with respect to the transverse displacement of the string
u = u(y/ε, y, t); see Fig. 14.1 for illustration.

306 14 Spatially-Oscillating Structures
Note that equation (14.1) has no point-wise interpretation due to the pres-
ence of Dirac δ-function. Both sides of the equation therefore must be viewed
as distributions producing the same output if applied to the same testing
function. Correctness of such type of modeling was intensively discussed in
the literature; see, for instance, [46]. Omitting details, the series of δ-functions
in equation (14.1) has certain meaning if the function f (u(y/ε, y, t)) is at least
continuous at y = ε(1 + 2k) for every k. Such a continuity condition easily
follows from the very form of equation (14.1). Indeed, if the displacement u
were step-wise discontinuous at y = ε(1 + 2k) then the derivative ∂ 2 u/∂y 2
would produce uncompensated derivatives of δ-function. So the displacement
u is at least continuous function of the coordinate y that is in match with the
physical meaning of model. Moreover, it will be shown below that appropriate
nonsmooth substitution for the spatial argument y/ε eliminates the singu-
larities from equation (14.1). As a result, new equations can be considered
within the classical theory.
In equation (14.1), the co-directed set of localized forces generated by the
elastic springs can be expressed through second derivative of the triangular
wave τ = τ (y/ε) in the form
∞
y
2f (u) δ − 1 − 2k = −f (u)sgn(τ )τ (14.2)
ε
k=−∞
Both the transverse displacement and the external loading are expressed in
terms of the non-smooth variables as follows
u = U (τ, y, t) + V (τ, y, t)τ

q = Q(τ, y, t) + P (τ, y, t)τ (14.3)
Finally, substituting (14.2) and (14.3) in equation (14.1) and using the differ-
ential and algebraic rules of the non-smooth argument transformation, gives
the differential equations and boundary conditions as, respectively,

∂2U ∂2V 2
2 ρ ∂ U ∂2U Q(τ, y, t)
= −2ε +ε − − (14.4)
∂τ 2 ∂y∂τ T ∂t2 ∂y 2 T

∂2V ∂2U 2
2 ρ ∂ V ∂2V P (τ, y, t)
2
= −2ε + ε 2
− 2
− (14.5)
∂τ ∂y∂τ T ∂t ∂y T
and
V |τ =±1 = 0 (14.6)
∂U ε2
|τ =±1 = ∓ f (U )|τ =±1 (14.7)
∂τ T
where the form of boundary condition (14.7) was eased by taking into account
condition (14.6).
14.1 Periodic Nonsmooth Structures 307
Further analysis is based on representing solutions in the form of asymp-

totic series
∞

U (τ, y, t) = εk Uk (τ, y, t)
k=0
∞
V (τ, y, t) = εk Vk (τ, y, t) (14.8)
k=0
Problem formulations based on equations (14.4) through (14.7) possess cer-

tain advantage as compared to equation (14.1). For instance, the influence of
infinite discontinuities in equation (14.1) is captured by the form of substitu-
tion (14.3) so that no discontinuities are present in the new equations. Also,
the structural discreteness of the nonlinear elastic foundation (14.2) is asso-
ciated with the new spatial variable τ restricted in the range −1 ≤ τ ≤ 1.
As a result, conventional perturbation tools become applicable. Moreover,
polynomial expansions with respect to the coordinate τ will not be affecting
the regularity of asymptotic expansions (14.8) in terms of the original fast
scale y/ε.
Note that partial differential equations (14.4) and (14.5) are coupled
whereas boundary conditions (14.6) and (14.7) are decoupled with respect
to the unknowns U and V . However, introducing new unknown functions,
say X = U + V and Y = U − V , decouples equations (14.4) and (14.5) as
follows

∂2X ∂2X 2
2 ρ ∂ X ∂2X Q+P
= −2ε + ε − − =0 (14.9)
∂τ 2 ∂y∂τ T ∂t2 ∂y 2 T

∂2Y ∂2Y 2 ρ ∂ Y
2
∂2Y Q−P
= 2ε + ε − − =0 (14.10)
∂τ 2 ∂y∂τ T ∂t2 ∂y 2 T
Equations (14.9) and (14.10) are of the same structure, except opposite signs
of the partial derivatives.
The buckling of a circular ring loaded by a discrete regular set of concen-
trated compressive forces was considered in [189]; see Fig.14.1. Taking into
account identity (14.2), the differential equation of equilibrium of such ring
∞

t
ẏ(t) = f (y(t)) − λ δ − 1 − 2k (14.11)
a
k=−∞
λ
≡ f (y(t)) + sgn(τ )τ
2
0 ≤ t ≤ 2π
where t = s/R is the arc length of the undeformed ring axis per radius, y(t)
is a six-component vector-function characterizing elastic states of the ring,
Fig. 14.1 Linear string on the discrete regular set of nonlinearly-elastic springs.
τ = τ (t/a) is the triangular sine wave of the period T = 4a and prime means
its Schwartz derivative, a = ε/2 is a small parameter as compared to unity, λ
is a dimensionless parameter which is proportional to the localized forces P
applied to the ring, and conditions of periodicity are imposed on the elements
of vector-function y(t).
Let us represent the unknown vector-function in the form
y = X(t, τ ) + Y (t, τ )τ (14.12)
Substituting then (14.12) in equation (14.11), gives

1 ∂Y ∂X 1 ∂X ∂Y
+ − R(X, Y ) + + − I(X, Y ) τ
a ∂τ ∂t a ∂τ ∂t

1 λ
+ Y − sgn(τ ) τ = 0 (14.13)
a 2
or

∂Y ε ∂X
= − + R(X, Y )
∂τ 2 ∂t

∂X ε ∂Y
= − + I(X, Y ) (14.14)
∂τ 2 ∂t
1
Y |τ =±1 = ± λε (14.15)
4
where
1
R(X, Y ) = [f (X + Y ) + f (X − Y )]
2
1
I(X, Y ) = [f (X + Y ) − f (X − Y )]
2
Fig. 14.2 The circular ring under discrete regular set of compressive radial forces.
The mechanical model, which is shown in Fig. 14.3, was considered in [150]
based on the generalized (asymmetric) version of the triangular wave. In
contrast to the model shown in Fig. 14.1, the sprigs are linearly elastic and
shifted in dipole-wise manner so that the differential equation of motion with
respect to the string deflection u = u(t, y) has the form
∂ 2u ∂ 2 u ku ∂ 2 τ (y/a, γ)
ρ − T − (1 − γ 2
)sgn(τ (y/a, γ)) =0 (14.16)
∂t2 ∂y 2 a ∂(y/a)2
−∞ < y < ∞
where the triangular wave with different positive and negative slopes is given
by

z/ (1 − γ) for − 1 + γ ≤ z ≤ 1 − γ
τ = τ (z, γ) = (14.17)
(−z + 2) / (1 + γ) for 1 − γ ≤ z ≤ 3 + γ
−1 < γ < 1
Schwartz derivatives of function (14.17) satisfy the following relationships

2
∂τ (z, γ) ∂τ (z, γ)
=α+β (14.18)
∂z ∂z
∂τ (z, γ) ∂ 2 τ (z, γ) 1 ∂ 2 τ (z, γ)

= β (14.19)
∂z ∂z 2 2 ∂z 2
∞
∂ 2 τ (z, γ)
= 2α [δ (z + 1 − γ − 4k) − δ (z − 1 + γ − 4k)] (14.20)
∂z 2
k=−∞

where α = 1/ 1 − γ 2 and β = 2γα.
Let us represent the string deflection in the form
u = U (τ, y, t) + V (τ, y, t)τ (14.21)
where τ = τ (y/a, γ) and τ = ∂τ (y/a, γ)/∂(y/a).

The components of representation (14.21), U and V , depend on the coor-
dinate y both explicitly and through the triangular wave function τ in such
a way that the complete partial derivative ∂u/∂y is equivalent to applying
the differential matrix operator

∂/∂y (α/a)∂/∂τ
D= (14.22)
(1/a)∂/∂τ (β/a)∂/∂τ +∂/∂y
to the vector-column of the components U and V

∂u U
⇐⇒ D (14.23)
∂y V
under the condition

V |τ =±1 = 0 (14.24)
The regular part of the second derivative can be calculated by means of the
relationship
∂2u 2 U
⇐⇒ D (14.25)
∂y 2 V
However, second derivative of the triangular wave function must be preserved
in order to eliminate the same kind of singularity from equation (14.16).
So, substituting (14.21) in (14.16) and collecting separately terms related to
different elements of the basis {1, τ }, gives
∂2U
ρ −T
∂t2

α ∂2 ∂2 αβ ∂ 2 2α ∂ 2
× + U+ + V =0 (14.26)
a2 ∂τ 2 ∂y 2 a2 ∂τ 2 a ∂τ ∂y
∂2V
ρ 2 −T
∂t

β ∂2 2 ∂2 α + β2 ∂2 2β ∂ 2 ∂2
× + U + + + V =0
a2 ∂τ 2 a ∂τ ∂y a2 ∂τ 2 a ∂τ ∂y ∂y 2
(14.27)
under the additional to (14.24) condition

T ∂U ∂V k(1 − γ 2 )
− 2 +β |τ =±1 = ± U |τ =±1 (14.28)
a ∂τ ∂τ a
Note that some terms have been eliminated from condition (14.28) by taking
into account condition (14.24).
Further analysis of the boundary value problem (14.24) and (14.26)
through (14.28) can be implemented by using the asymptotic approach [151]
considering a as a small parameter.
Fig. 14.3 Linear string on the discrete periodic set of linearly-elastic dipole-wise
shifted springs of the stiffness k.
Similar version of the transformation was employed for statics of layered

composites in [179]. The idea is to represent the elastic constants in the form
λ∗ = λ(1 + lτ )
μ∗ = μ(1 + mτ ) (14.29)
τ = ∂τ (x/a, γ) /∂(x/a)
where x is the spatial coordinate, which is perpendicular to the layer bound-

aries, and λ, μ, l, and m are such constants that λ∗ and μ∗ remain positive
on the period of elastic structure, T = 4a.
Respectively, the vector of elastic displacements is represented in the form
{u, v, w} = {U (1) , V (1) , W (1) } + {U (2) , V (2) , W (2) }τ (14.30)
where U (i) = U (i) (τ, y, z), V (i) = V (i) (τ, y, z), and W (i) = W (i) (τ, y, z).
Then, both representations, (14.29) and (14.30), are substituted into the
static equations of three-dimensional elastic bodies with variable elastic
constants, and the NSTT algebraic and differential manipulations applied

in order to formulate boundary value problems, however with constant elas-
ticity parameters, for the components of representation (14.30).
14.2 Averaging for One-Dimensional Periodic

Structures
Let us consider a one-dimensional 4ε-periodic structure whose static elastic
states are described by the vector-function z(y) ∈ Rn that depends upon the
longitudinal coordinate y attached to the undeformed structure. The num-
ber of vector’ components n can always be increased so that the differential
equation of equilibrium takes the form of first-order differential equation, for
instance, as follows
z (y) = f (z(y), ϕ, y) + p(y)τ (ϕ) (14.31)
Here the spatial scale ϕ = y/ε associates with the structural periodicity, the
vector-function f (z, ϕ, y) ∈ Rn is continuous with respect to z and y, but
it is allowed to be step-wise discontinuous with respect to ϕ at the points
{ϕ : τ (ϕ) = ±1}, and p(y) ∈ Rn is a continuous vector-function describing
the amplitude modulation of the localized loading.
Example 18. In terms of the matrixes,

u(y) 0
z= , p(y) = (14.32)
v(y) q(y)/(2ε)

v(y)/{EF [1 + ατ (y/ε)]}
f =
0
equation (14.31) describes an elastic rod whose cross sectional area is a piece-
wise constant periodic function of the longitudinal coordinate as it is shown
in Fig. 14.4 Indeed, substituting (14.32) in (14.31) and eliminating then v(y),
gives the second-order differential equation of equilibrium for the rod
du q(y) y
d y
EF 1 + ατ = τ . (14.33)
dy ε dy 2ε ε
Fig. 14.4 An elastic rod with a periodic non-smoothly varying cross sectional area
and concentrated loading.
14.3 Two Variable Expansions 313
The averaging technique is described below based on the general form of

equation (14.31).
14.3 Two Variable Expansions

The idea of two variable expansions will be used by considering the saw-tooth
oscillating coordinate τ = τ (y/ε) and the original coordinate y 0 ≡ y as fast
and slow spatial scales, respectively, provided that the following assumption
holds
ε << 1 (14.34)
Let us represent solutions of equation (14.31) in the form
z = X(τ, y 0 ) + Y (τ, y 0 )τ (14.35)
Substituting (14.35) in equation (14.31), gives

∂Y ∂X ∂X ∂Y
+ε − R f + + ε − If τ + [Y − εp(y 0 )]τ = 0
∂τ ∂y 0 ∂τ ∂y 0
(14.36)
where

Rf X, Y, τ, y 0 1
= [f (X + Y, τ, y 0 ) ± f (X − Y, 2 − τ, y 0 )] (14.37)
If X, Y, τ, y 0 2
Expression (14.36) is equivalent to the following boundary value problem

with no discontinuities

∂X ∂Y
+ε − If =0 (14.38)
∂τ ∂y 0

∂Y ∂X
+ε − Rf =0
∂τ ∂y 0
Y |τ =±1 = εp(y 0 ) (14.39)

Let us represent solutions of the boundary value problem, (14.38) and (14.39),
in the form of asymptotic series with respect to ε,
∞

X= εi X i (τ, y 0 ) (14.40)
i=0
∞
Y = εi Y i (τ, y 0 )
i=0
where the functions X i and Y i are to be sequentially determined.

Substituting (14.40) in (14.37), generates power series expansions
Rf = Rf0 + εRf1 + ε2 Rf2 + ... (14.41)

If = If0 + εIf1 + ε2 If2 + ...
where the following notations are used
1 ∂ i Rf 1 ∂ i If
Rfi = |ε=0 , Ifi = |ε=0
i! ∂εi i! ∂εi
Substituting (14.40) in (14.38) and (14.39), and matching the coefficients
of the same powers of ε, gives the corresponding sequence of equations and
boundary conditions. In particular, zero-order problem takes the form
∂X 0 ∂Y 0
= 0, =0 (14.42)
∂τ ∂τ
and
Y 0 |τ =±1 = 0 (14.43)
As follows from (14.42), the generating solution is independent on the fast
oscillating scale τ . Therefore, taking into account (14.43), gives solution
X 0 = A0 (y 0 ), Y0 ≡0 (14.44)
where A0 is an arbitrary vector-function of the slow co-ordinate that will be

determined on the next step of the asymptotic procedure.
So, collecting the terms of order ε, gives the differential equations and
boundary conditions in the form, respectively,
∂X 1 ∂Y 0
= If0 − = If (A0 , 0, τ, y 0 ) (14.45)
∂τ ∂y 0
∂Y 1 ∂X 0 dA0
= Rf0 − = R f (A0
, 0, τ, y 0
) − (14.46)
∂τ ∂y 0 dy 0
and
Y 1 |τ =±1 = p(y 0 ) (14.47)
Integrating equations (14.45) and (14.46), gives first-order terms of the
asymptotic solution
τ
1
X = If0 dτ + A1 (y 0 ) (14.48)
0
τ
dA0
Y1 = Rf0 dτ − (τ + 1) + p(y 0 )
dy 0
−1
14.4 Second Order Equations 315
where A1 is a new arbitrary vector-function of the slow spatial scale y 0 , and

the limits of integration for Y 1 are chosen in such a manner that boundary
condition (14.47) is satisfied automatically at the point τ = −1 whereas
another point, τ = 1, gives equation
1
dA0 1
= Rf0 dτ ≡< Rf0 > (14.49)
dy 0 2
−1
Note that the ‘slow scale’ equation (14.49) was obtained by satisfying the
boundary condition in contrast to the conventional scheme of two variable
expansions in which such kind of equations are obtained by eliminating the
so-called ‘resonance terms.’
Enforcing now equation (14.49), brings the component Y 1 to the final form
τ
Y =1
(Rf0 − < Rf0 >)dτ + p(y 0 ) (14.50)
−1
At this stage, expressions (14.48) through (14.50) determine the first-order

terms of the asymptotic solution, however, the slow-scale vector-function
A1 (y 0 ) still remains unknown. The corresponding ordinary differential equa-
tion is obtained on the next stage from the boundary condition for Y 2 and
0 1
dA1 ∂Rf0
= A1 + F 1 (A0 , y 0 ) (14.51)
dy 0 ∂A0
where ∂Rf0 /∂A0 is the Jacobian matrix, and the vector-function F 1 is known.
Note, that equation (14.51) is linear. Moreover, on the next steps, equa-
tions for the vector-functions A2 , A3 ,... will be of the same linear structure,
including the same Jacobian matrix.
14.4 Second Order Equations

Let us consider now the second order differential equation with respect to
the vector-function z(y) ∈ Rn , however, in the linear form
z (y) + [q(ϕ, y) + p(y)τ (ϕ)]z = g(ϕ, y) + r(y)τ (ϕ) (14.52)
where q and p are n× n-matrixes, g and r are n-dimensional vector-functions,

and ϕ = y/ε is the fast spatial scale.
Based on the assumptions of the previous section, the functions q and g
and solutions of equation (14.52) can be represented in the form, respectively,
q(ϕ, y) = Q(τ (ϕ), y 0 ) + P (τ (ϕ), y 0 )τ (ϕ)

g(ϕ, y) = G(τ (ϕ), y 0 ) + F (τ (ϕ), y 0 )τ (ϕ) (14.53)
and
z(y) = X(τ (ϕ), y 0 ) + Y (τ (ϕ), y 0 )τ (ϕ)
where y 0 ≡ y represents the slow spatial scale.
This leads to the boundary value problem

2
∂2X ∂2Y ∂ X
= −2ε −ε 2
+ QX + P Y − G (14.54)
∂τ 2 ∂τ ∂y 0 ∂y 02

∂2Y ∂2X ∂2Y
= −2ε −ε 2
+ P X + QY − F (14.55)
∂τ 2 ∂τ ∂y 0 ∂y 02
and
∂X
|τ =±1 = ε2 [r(y) − p(y)X]|τ =±1 (14.56)
∂τ
Y |τ =±1 = 0
Further, representing the solution of the boundary value problem (14.54),

(14.55) and (14.56) in the form of asymptotic series (14.40), gives a sequence
of boundary value problems, in which first two steps appear to have quite
trivial solutions, such as
X 0 = B 0 (y 0 ), Y 0 ≡ 0
and
X 1 ≡ 0, Y 1 ≡ 0
where B 0 is an arbitrary function of the slow argument y 0 .
As a result, first two non-trivial steps of the averaging procedure give
z(y) = B 0 (y 0 ) + ε2 [X 2 (τ (ϕ), y 0 ) + Y 2 (τ (ϕ), y 0 )τ (ϕ)] + O(ε3 ) (14.57)
where, in second-order of ε, the solution components are

τ
2
X = (τ − ξ)[G(ξ, y 0 )− < G(τ, y 0 ) > (14.58)
−1
−(Q(ξ, y 0 )− < Q(τ, y 0 ) >)B 0 ]dξ + (r − pB 0 )τ + B 2 (y 0 )

τ
Y = [(τ − ξ)(F (ξ, y 0 ) − P (ξ, y 0 )B 0 )
2
(14.59)
−1
− < (1 − τ )(F (τ, y 0 ) − P (τ, y 0 )B 0 ) >]dξ

14.4 Second Order Equations 317
Here, notation < • > means averaging with respect to τ as defined in (14.49),
the vector-function B 0 = B 0 (y 0 ) satisfies equation
d2 B 0
+ < Q(τ, y 0 ) > B 0 =< G(τ, y 0 ) > (14.60)
dy 02
The new arbitrary function of the slow coordinate, B 2 (y 0 ), has to be defined

on the next step of the procedure.
Note that the δ-loads generated by the derivative τ (ϕ) are switching their
directions twice per one period of the triangular wave. In many practical cases
though the direction of impulses may remain constant. The corresponding
reformulation of the problem can be implemented by introducing the factor
-sgn(τ ) into the differential equation as follows
z (y) + [q(ϕ, y) − p(y)sgn(τ )τ (ϕ)]z = g(ϕ, y) − r(y)sgn(τ )τ (ϕ) (14.61)
Now, in equation (14.61), the term sgn(τ )τ (ϕ) generates δ-loads of the same
direction, whereas the boundary condition (14.56) for the X-component takes
the form
∂X
|τ =±1 = ∓ε2 [r(y) − p(y)X]|τ =±1 (14.62)
∂τ
The form of expressions (14.58) and (14.60) is modified as, respectively,
τ
2
X = (τ − ξ)[G(ξ, y 0 )− < G(τ, y 0 ) > (14.63)
−1
τ2
−(Q(ξ, y 0 )− < Q(τ, y 0 ) >)B 0 ]dξ − (r − pB 0 ) + B 2 (y 0 )
2
and,
d2 B 0
+ (< Q(τ, y 0 ) > +p)B 0 =< G(τ, y 0 ) > +r (14.64)
dy 02
and the component Y 2 is still described by (14.59).
Example 19. Let us consider an infinite beam resting on a discrete foundation

represented by the periodic set of linearly-elastic springs of stiffness c. The
corresponding differential equation of equilibrium is
d4 w cw x x
∞
D + δ − 1 − 2k = qload (14.65)
dx4 a a L
k=−∞
−∞ < x < ∞
Let us introduce the following dimensionless values

x y w
y= , ϕ= , W =
L ε a
cL4 L4
γ= , ψ(y) = qload (y)
aD aD
where ε = a/L << 1. As a result the above equation (14.65) for the beam’
center line takes the form
d4 W 1
4
− γsgn[τ (ϕ)]τ (ϕ) W = ψ(y) (14.66)
dy 2
This equation becomes equivalent to (14.61), after the substitutions

W (y) 0
z= , g= ≡G
W (y) ψ(y)

0 0 0 −1
p= , q= ≡Q
γ/2 0 00
P ≡ 0, F ≡ 0, r ≡ 0
After the corresponding calculations in (14.63) and (14.59), one finally

obtains

0 2 1 2 y

z = B (y) + ε pτ B (y) + B (y) + O ε3
0 2
(14.67)
2 ε
where the vector-function B 0 = [B10 , B20 ]T is determined from equation

(14.64). In terms of the vector B 0 components, equation (14.64) reads
0
d2 B10 0 −1 B1 0
+ = (14.68)
dy 2 B20 γ/2 0 B20 ψ(y)
The first equation in (14.68) gives B20 = d2 B10 /dy 2 , whereas the second equa-
tion after elimination of B20 takes the form
d4 B10 γ
4
+ B10 = ψ(y) (14.69)
dy 2
Equation (14.69) is indeed obtained by averaging equation (14.66) with re-

spect to the fast spatial scale ϕ. In other words, equation (14.69) describes
an elastic beam resting on effective continuous elastic foundation. In order
to illustrate the asymptotic solution, let us consider the beam under the
harmonic transverse loading, qload (x/L) ≡ q0 sin (πx/L), where q0 =const.
Taking into account only the leading order ‘slow’ and ‘fast’ components, gives
the bending moment in terms of the original variables
d2 w
M (x) = D (14.70)
dx2
a 2 x πx
= −M0 1−γ τ2 sin
2πL a L
14.5 Acoustic Waves from Non-smooth Periodic Boundary Sources 319
where M0 = 2q0 π 2 L2 /(2π 4 + γ). The bending moment diagram is given in

Fig. 14.5.
1.0
0.5
MxMo
0.0
0.5
1.0
0 1 2 3 4 5 6
x
Fig. 14.5 Bending moment of the beam on the discrete elastic foundation; numer-
ical values of the parameters are as follows: L = π, a = 0.2, and γ = 1948.0
Finally, note that the homogenization procedure described in [24] gives the
averaged equation in the slow spatial scale and a so-called ‘cell problem’ in
the fast scale. In the above approach, the analog of cell problem associates
with the fast oscillating spatial scale given by the triangular wave function
τ (y/ε). As a result, solution for the cell problem automatically unfolds on
the entire structure so that the fast and slow components of elastic states are
eventually expressed though the same coordinate.
14.5 Acoustic Waves from Non-smooth Periodic

Boundary Sources
This section deals with two-dimensional acoustic waves propagating from a
discontinuous periodic source located at the boundary of half-infinite space.
It is shown that introducing the triangular wave function as a specific spatial
coordinate naturally eliminates discontinuities from the boundary condition
associated with the active boundary.
For illustrating purposes, let us consider the case of two-dimensional sta-
tionary waves propagating in the half-infinite media from a piecewise-linear
periodic boundary source as shown in Fig. 14.6.
Fig. 14.6 The model
Let us describe acoustic waves by the linear wave equation in the standard
form
1 ∂2P ∂2P ∂ 2P ∂2P
= + + (14.71)
c2f ∂t2 ∂x2 ∂y 2 ∂z 2
where P is a pressure deviation from the static equilibrium pressure; x, y, z
and t are spatial coordinates and time, respectively, and cf is the speed of
sound in the media.
Further, the plane problem is considered when P = P (t, y, z), and there-
fore, ∂ 2 P/∂x2 = 0. Such an assumption can be justified by sufficiently
long piezoelectric rods whose characteristics are constant along the x-co-
ordinate. Suppose that the pressure generated by the rods near the bound-
ary is P0 = A sin ωt, where A and ω are constant amplitude and frequency,
respectively.
Let the boundary condition at z = 0 to have the form

P0 (t) for (4n − 1) a ≤ y ≤ (4n + 1) a
P (t, y, 0) = (14.72)
0 for (4n + 1) a ≤ y ≤ (4n + 3) a
n = 0, ±1, ±2, ...
Note that, based on what is actually known near the fluid-source interface,
the boundary condition can also be formulated for pressure derivatives. From
the mathematical standpoint, this do not affect much the solution procedure
though.
Let us seek the steady-state solution, which is periodic with respect to t
and y and remains bounded as z → ∞.
Since the boundary condition is periodic along y-coordinate with period
T = 4a, then, according to the idea of non-smooth argument transforma-
tion, the triangular wave periodic coordinate is introduced as y → τ (y/a).
As a result, the boundary condition (14.72) and yet unknown solution are
represented as, respectively,
1 1 y
P (t, y, z)|z=0 = P0 (t) + P0 (t) τ (14.73)
2 2 a
and
P (t, y, z) = P1 (t, τ (y/a), z) + P2 (t, τ (y/a), z)τ (y/a) (14.74)
where the components P1 and P2 are considered as new unknown functions.
14.5 Acoustic Waves from Non-smooth Periodic Boundary Sources 321
Taking into account the expression [τ (y/a)]2 = 1, gives first generalized

derivative of the original unknown function in the form
∂P 1 ∂P2 1 ∂P1 y 1 y
= + τ + P2 τ (14.75)
∂y a ∂τ a ∂τ a a a
Since the function P (t, y, z) has to be continuous with respect to y in the

unbounded open region z > 0, then the periodic singular term τ in (14.75)
must be eliminated by imposing condition
P2 |τ =±1 = 0 (14.76)
Analogously, second derivative takes the form
∂2P 1 ∂ 2 P1 1 ∂ 2 P2 y
= + τ (14.77)
∂y 2 a2 ∂τ 2 a2 ∂τ 2 a
under the condition
∂P1
|τ =±1 = 0 (14.78)
∂τ
Note that both derivatives, (14.75) and (14.77), as well as the original function
(14.74) appear to have the same algebraic structure of hyperbolic numbers.
Obviously, differentiation with respect to t and z preserve such a structure as
well. As a result, substituting the second derivatives into differential equation
(14.71) and collecting separately terms related to each of the basis elements
{1, τ }, gives two partial differential equations for the components of repre-
sentation (14.74)
1 ∂ 2 Pi 1 ∂ 2 Pi ∂ 2 Pi
= + (14.79)
c2f ∂t2 a2 ∂τ 2 ∂z 2
(i = 1, 2)
Substituting then (14.74) in (14.73), gives the corresponding set of boundary

conditions
1 1
Pi (t, τ, z) |z=0 = P0 (t) = A sin ωt (14.80)
2 2
Now equations (14.79) and boundary conditions (14.76), (14.78) and (14.80)
constitute two independent boundary value problems for the components P1
and P2 . However, the result achieved is that no discontinuous functions are
present any more in the boundary conditions.
Solving the above boundary value problems by the standard method of
separation of variables, gives finally1
1
Derived by S. Pavlyshyn.

1 z
P (t, y, z) = A sin ω t − +
2 cf

m
(−1)k−1 1 y
A sin ω (t − Kk z) cos k − πτ + (14.81)
(k − 1/2) π 2 a
k=1
∞
k−1

(−1) 1 y y
sin ωt exp (−χk z) cos k − πτ τ
(k − 1/2) π 2 a a
k=m+1
where
*

2 2
1 π 2
ω
Kk = − k− , k = 1, ..., m
cf
2 a
*
2
2
1 π 2 ω
χk = k− − , k = m + 1, ...
2 a cf
and m is the maximum number at which the expression under the first square
root is still positive.
2
6
P
0
2 4
1
z
0 2
1
y
2
0
3
Fig. 14.7 Acoustic wave surface for the set of parameters : cf = 10.0, a = 1,
ω = 172, t = 3, and A = 2.
A three-dimensional illustration of solution (14.81) is given by Figs. 14.7

and 14.8 for two different magnitudes of the frequency ω. Besides, it is seen
14.6 Spatio-temporal Periodicity 323
Fig. 14.8 Contour plot of the wave field shown in Fig. 14.7.
that shorter waves are carrying the information about the discreetness of the
wave source for a longer distance from the source.
In conclusion, solution (14.81) could be of-course obtained in terms of the
standard trigonometric expansions by applying the method of separation of
variables directly to the original problem, (14.71) and (14.72). However, the
derivation of solution (14.81) implies no integration of discontinuous func-
tions, since all the discontinuities have been captured in advance by trans-
formation (14.74).
It is also worth to note that, the terms of series (14.81) are calculated
on the standard interval, −1 ≤ τ ≤ 1, which is covered by one half of the
total period, whereas the standard Fourier expansions must be built over the
entire period. This is due to the fact that representation (14.74) automatically
unfolds the half-period domain on the infinite spatial interval.
14.6 Spatio-temporal Periodicity

As a possible generalization of the approach, let us consider, for instance, the
boundary condition in the form
P (t, y, z)|z=0 = f (t, y) (14.82)

Fig. 14.9 Acoustic wave surface for the set of parameters: cf = 10.0, a = 1,
ω = 86, t = 3, and A = 2.
Fig. 14.10 Contour plot of the wave field shown in Fig. 14.9.
14.6 Spatio-temporal Periodicity 325
where the function f is periodic with temporal period Tt = 2π/ω and spatial
period Ty = 4a.
Introducing the triangular wave spatial argument, τy = τ (y/a), gives
f (t, y) = F1 (t, τ (y/a)) + F2 (t, τ (y/a))τ (y/a) (14.83)
where
1
F1 (t, τy ) = [f (t, aτy ) + f (t, 2a − aτy )]
2
1
F2 (t, τy ) = [f (t, aτy ) − f (t, 2a − aτy )] (14.84)
2
In a similar way, introducing the triangular wave temporal argument, τt =
τ (2ωt/π), into both of the components, F1 and F2 , gives eventually expression
of the form
f (t, y) = f0 (τt , τy )e0 + f1 (τt , τy )e1 + f2 (τt , τy )e2 + f3 (τt , τy )e3 (14.85)
where components fi (τt , τy ) are uniquely determined by rule (14.84) applied

to each of the two arguments, and the following basis is introduced
e0 = 1
e1 = τ (2ωt/π)
e2 = τ (y/a) (14.86)
e3 = e1 e2
Basis (14.86) obeys the table of products
× e0 e1 e2 e3
e0 1 e1 e2 e3
e1 e1 1 e3 e2 (14.87)
e2 e2 e3 1 e1
e3 e3 e2 e1 1
Now, the acoustic pressure is represented in the similar to (14.85) form
P (t, y, z) = P0 (τt , τy , z)e0 + P1 (τt , τy , z)e1 + P2 (τt , τy , z)e2 + P3 (τt , τy , z)e3

(14.88)
Regarding the problem described in the previous section, the components
of representation (14.88) can be obtained as an exercise by introducing the
argument τt directly into solution (14.81). However, formulations based on
representation (14.88) become technically reasonable whenever the boundary
pressure is adequately described by the functions τt and τy or their different
combinations, for instance, polynomials. In such cases, polynomial approx-
imations with respect to the bounded arguments may appear to be more
effective as compared to Fourier expansions.
Let, for instance, P0 (t) describes the periodic sequence of rectangular

spikes of the amplitude A,

1 2ω 1
P0 (t) = A 1 + τ t ≡ A(1 + e1 ) (14.89)
2 π 2
Then, boundary condition (14.73) takes the form

1
P (t, y, z)|z=0 = A(1 + e1 )(1 + e2 )
4
1
≡ A(e0 + e1 + e2 + e3 ) (14.90)
4
where the basis elements {e0 , e1 , e2 , e3 } are given by (14.86), and the table
of products (14.87) is taken into account.
Now, substituting representation (14.88) in (14.90), gives the boundary
conditions for its components at z = 0 as follows
1
Pi (τt , τy , 0) = A; i = 0,...,3
4
Finally, the three-dimensional case can be considered by adding periodicity of
the source along the x-direction at the boundary z = 0 and introducing the
corresponding triangular wave argument, say τx . The corresponding rules
for algebraic manipulations would be analogous to those generated by the
arguments τt and τy . However, necessary details are illustrated below on
another model.
14.7 Membrane on a Two-Dimensional Periodic

Foundation
Consider an infinite membrane resting on a linearly elastic foundation of the
stiffness K(x, y) under the transverse load q(x, y). Assuming that both the
stiffness K and load q are measured per unit membrane tension T , the partial
differential equation of equilibrium is represented in the form the
Δu − K(x, y)u = q(x, y) (14.91)

∂2 ∂2
Δ= 2
+ 2
∂x ∂y
where u = u(x, y) is the membrane transverse deflection.

The foundation is assumed to be step-wise discontinuous and periodic
along each of the coordinates as described by the function
k x y
K(x, y) = 1 + τ 1 + τ (14.92)
4 a b
14.7 Membrane on a Two-Dimensional Periodic Foundation 327
Fig. 14.11 Fragment of the map of periodic elastic foundation; a = 1.0 and b = 2.0.
With reference to Fig. 14.11, function (14.92) is defined on the infinite plane,
such that
0 (x, y) ∈ any “dark field”
K(x, y) = (14.93)
k (x, y) ∈ any “light field”
In the same way, Fig. 14.12 provides maps for the elements of basis
e0 = 1
e1 = τ (x/a)
e2 = τ (y/b) (14.94)
e3 = e1 e2
The above table of products (14.87) still valid for basis (14.94). As a result,
function (14.92) takes eventually the form
k
K(x, y) = (e0 + e1 + e2 + e3 ) (14.95)
4
Now let us represent the membrane deflection in the form
u(x, y) = X(τx , τy , x, y)e0 + Y (τx , τy , x, y)e1 (14.96)

+Z(τx , τy , x, y)e2 + W (τx , τy , x, y)e3
Fig. 14.12 The standard basis map: each of the elements is equal to unity within
light domains and zero within dark domains.
where τx = τ (x/a) and τy = τ (y/b) are triangular waves whose lengths

are determined by the periods of foundation along x− and y− direction,
respectively; scales of the explicitly present variables, x and y, are associated
with the scales of loading q(x, y), which is assumed to be slow as compared
to the spatial rate of foundation.
Note that both linear and non-linear algebraic manipulations with com-
binations of type (14.96) are dictated by the table of products (14.87). For
example, taking into account (14.95) and (14.96), gives
k
Ku = (X + Y + Z + W )(e0 + e1 + e2 + e3 ) (14.97)
4
High-order derivatives of (14.96) are simplified by using the table of products

(14.87) and introducing specific differential operators as follows.
First, using the chain rule, gives
dτx 1 1
= τ (x/a) = e1 (14.98)
dx a a
dτy 1 1
= τ (y/a) = e2 (14.99)
dy b b
Then, taking into account (14.87), (14.94), (14.98) and (14.99), gives first
derivatives of (14.96) in the form

∂u 1 ∂Y ∂X 1 ∂X ∂Y
= + e0 + + e1
∂x a ∂τx ∂x a ∂τx ∂x

1 ∂W ∂Z 1 ∂Z ∂W
+ + e2 + + e3 (14.100)
a ∂τx ∂x a ∂τx ∂x
de1 (x/a) 1
+(Y + W e2 )
d(x/a) a

∂u 1 ∂Z ∂X 1 ∂W ∂Y
= + e0 + + e1
∂y b ∂τy ∂y b ∂τy ∂y

1 ∂X ∂Z 1 ∂Y ∂W
+ + e2 + + e3 (14.101)
b ∂τy ∂y b ∂τy ∂y
de2 (y/b) 1
+(Z + W e1 )
d(y/b) b
Last addends in (14.100) and (14.101) include derivatives of the step-wise

discontinuous functions e1 (x/a) and e2 (y/b). Such derivatives are expressed
through Dirac δ-functions and therefore must be excluded from the expres-
sions (14.100) and (14.101) due to continuity of the original function u(x, y).
The δ-functions are eliminated under the boundary conditions
Y |τx =±1 = 0
W |τx =±1 = 0 (14.102)
and
Z|τy =±1 = 0
W |τy =±1 = 0 (14.103)
The rest of terms in (14.100) and (14.101) represent linear combinations of

the basis {e0 , e1 , e2 , e3 }. In order to formalize the differentiation procedure,
let us associate expansion (14.96) with the vector-column
⎡
⎤
X
⎢Y ⎥
u =⎢
⎣Z ⎦
⎥ (14.104)
W
In a similar way, let us introduce the vector-columns ux and uy associated

with derivatives (14.100) and (14.101) under conditions (14.102) and (14.103),
respectively,2
ux = Dx u (14.105)
uy = Dy u
where
⎡ ⎤
∂/∂x a−1 ∂/∂τx 0 0
⎢ a−1 ∂/∂τx ∂/∂x 0 0 ⎥
Dx = ⎢
⎣0
⎥ (14.106)
0 ∂/∂x a−1 ∂/∂τx ⎦
0 0 a−1 ∂/∂τx ∂/∂x
and ⎡ ⎤
∂/∂y 0 b−1 ∂/∂τy 0
⎢0 ∂/∂y 0 b−1 ∂/∂τy ⎥
Dy = ⎢
⎣ b−1 ∂/∂τy
⎥
⎦ (14.107)
0 ∂/∂y 0
0 b−1 ∂/∂τy 0 ∂/∂y
These differential matrix operators automatically generate high-order deriva-
tives of combination (14.96) provided that necessary smoothness (boundary)
conditions hold. For instance, the components of expansion for Δu are given
by the elements of vector-column (D2x + D2y )u under conditions

1 ∂X ∂Y
+ |τx =±1 = 0
a ∂τx ∂x

1 ∂Z ∂W
+ |τx =±1 = 0 (14.108)
a ∂τx ∂x

1 ∂X ∂Z
+ |τy =±1 = 0
b ∂τy ∂y

1 ∂Y ∂W
+ |τy =±1 = 0 (14.109)
b ∂τy ∂y
Consider now the particular case a = b = ε 1. Following the differentia-

tion and algebraic manipulation rules as introduced above, and substituting
(14.96) in (14.91), gives
2
Note that ux is not ∂u/∂x.

∂2Y ∂ 2Z
Δτ X + 2ε + + ε2 (ΔX − F ) = ε2 q(x, y)
∂τx ∂x ∂τy ∂y

∂2X ∂ 2W
Δτ Y + 2ε + + ε2 (ΔY − F ) = 0

2
∂ W ∂2X
Δτ Z + 2ε + + ε2 (ΔZ − F ) = 0 (14.110)

2
∂ Z ∂2Y
Δτ W + 2ε + + ε2 (ΔW − F ) = 0
where Δτ = ∂ 2 /∂τx2 + ∂ 2 /∂τy2 , Δ = ∂ 2 /∂x2 + ∂ 2 /∂y 2 , and symbol F denotes

the following group of terms related to the elastic foundation
1
F ≡ k(X + Y + Z + W ) (14.111)
4
Boundary conditions (14.108) and (14.109) can be simplified due to (14.102)
and (14.103). As a result, the complete set of boundary conditions takes the
form
∂X ∂X
|τ =±1 = 0, |τ =±1 = 0
∂τx x ∂τy y
∂Y
Y |τx =±1 = 0, |τ =±1 = 0
∂τy y
∂Z
|τ =±1 = 0, Z|τy =±1 = 0 (14.112)
∂τx x
W |τx =±1 = 0, W |τy =±1 = 0
Note that equations (14.110) have constant coefficients whereas the step-wise
discontinuities of foundation have been absorbed by the triangular wave argu-
ments τx and τy . The corresponding boundary conditions (14.112) generate a
so-called “cell problem” [76] within the rectangular domain {−1 ≤ τx ≤ 1,
−1 ≤ τy ≤ 1}. Therefore, the arguments τx and τy locally describe the fast
varying component of membrane shape. In contrast, the explicitly present co-
ordinates x and y describe the slow component in the infinite plane {−∞ <
x < ∞, −∞ < y < ∞}. Finally, the increase in number of equations (14.110)
of-course complicates solution procedures from the technical standpoint. How-
ever, the obvious symmetry of equations helps to ease the corresponding deriva-
tions. For instance, equations (14.110) can be decoupled by introducing new
unknown functions Ui = Ui (τx , τy , x, y) (i = 1, .., 4) as follows
U1 = X + Y + Z + W
U2 = X + Y − Z − W
U3 = X − Y + Z − W (14.113)
U4 = X − Y − Z + W
Linear transformation (14.113) however makes boundary conditions (14.112)

coupled. New boundary conditions are given by the inverse substitution in
(14.112)
1
X= (U1 + U2 + U3 + U4 )
4
1
Y = (U1 + U2 − U3 − U4 )
4
1
Z = (U1 − U2 + U3 − U4 ) (14.114)
4
1
W = (U1 − U2 − U3 + U4 )
4
Note that transformation (14.113) can be effectively incorporated at the very
beginning of transformations by using the idempotent basis as described in
the next section.
14.8 The Idempotent Basis for Two-Dimensional

Structures
The two-dimensional idempotent basis is introduced as follows
1 1
i1 = e+ +
1 e2 = (e0 + e1 )(e0 + e2 ) = (e0 + e1 + e2 + e3 )
4 4
− 1 1
1 e2 = (e0 + e1 )(e0 − e2 ) =
i2 = e+ (e0 + e1 − e2 − e3 )
4 4
− + 1 1
i3 = e1 e2 = (e0 − e1 )(e0 + e2 ) = (e0 − e1 + e2 − e3 ) (14.115)
4 4
− − 1 1
i4 = e1 e2 = (e0 − e1 )(e0 − e2 ) = (e0 − e1 − e2 + e3 )
4 4
where the standard basis ei is defined by (14.94) and the table of products
(14.87), and the following notations for one-dimensional idempotent basis are
used
1
e±
k = (e0 ± ek ) (14.116)
2
−
e+
k ek = 0; (k = 1, 2)
The main reason for using basis (14.115) is that its table of products has the
normalized diagonal form
ik in = δkn (14.117)
where δkn is the Kronecker symbol.
The geometrical meaning of property (14.117) follows from the maps in
Fig. 14.13.
14.8 The Idempotent Basis for Two-Dimensional Structures 333
Fig. 14.13 The map of idempotent basis: each of the elements is equal to unity
within light domains and zero within dark domains.
In this basis, representation (14.96) takes the form

4
u(x, y) = Uk (τx , τy , x, y)ik (14.118)
k=1
As a result,
4
4
f( Uk ik ) = f (Uk )ik (14.119)
k=1 k=1
where f is practically any function, linear or nonlinear.

First-order partial derivatives of representation (14.119) are obtained as
follows
∂u(x, y)
(14.120)
∂x
4
1 ∂Uk (τx , τy , x, y) ∂Uk (τx , τy , x, y) ∂ik
= e1 ik + ik + Uk (τx , τy , x, y)
a ∂τx ∂x ∂x
k=1
Further, taking into account (14.115) and (14.116), gives

− + +
1 − e1 )e1 e2 = i1
e1 i1 = (e+
− + −
1 − e1 )e1 e2 = i2
e1 i2 = (e+
− − +
1 − e1 )e1 e2 = −i3
e1 i3 = (e+
− − −
1 − e1 )e1 e2 = −i4
e1 i4 = (e+
and
∂i1 ∂e+
1 + 1 ∂e1 +
= e = e
∂x ∂x 2 2 ∂x 2
∂i2 ∂e+ 1 ∂e1 −
= 1 e− = e
∂x ∂x 2 2 ∂x 2
∂i3 ∂e− 1 ∂e1 +
= 1 e+ =− e
∂x ∂x 2 2 ∂x 2
∂i4 ∂e− 1 ∂e1 −
= 1 e− =− e
∂x ∂x 2 2 ∂x 2
As a result, derivative (14.120) takes the form

∂u 1 ∂U1 ∂U1 1 ∂U2 ∂U2
= + i1 + + i2
∂x a ∂τx ∂x a ∂τx ∂x

1 ∂U3 ∂U3 1 ∂U4 ∂U4
+ − + i3 + − + i4 (14.121)
a ∂τx ∂x a ∂τx ∂x
1 ∂e1 + 1 ∂e1 −
+ (U1 − U3 ) e + (U2 − U4 ) e
2 ∂x 2 2 ∂x 2
Analogously, one obtains

∂u 1 ∂U1 ∂U1 1 ∂U2 ∂U2
= + i1 + − + i2
∂y b ∂τy ∂y b ∂τy ∂y

1 ∂U3 ∂U3 1 ∂U4 ∂U4
+ + i3 + − + i4 (14.122)
b ∂τy ∂y b ∂τy ∂y
1 ∂e2 + 1 ∂e2 −
+ (U1 − U2 ) e + (U3 − U4 ) e
2 ∂y 1 2 ∂y 1
Let us introduce vector, associated with expansion (14.118), and the corre-
sponding differential matrix operators as, respectively,
⎡ ⎤
U1
⎢ U2 ⎥
u =⎢ ⎥
⎣ U3 ⎦ (14.123)
U4
and
⎡1 ∂ ∂ ⎤
a ∂τx + ∂x 0 0 0
⎢0 1 ∂
+ ∂
0 0 ⎥
Dx = ⎢
⎣0
a ∂τx ∂x ⎥
⎦ (14.124)
0 − a1 ∂τ∂x + ∂
∂x 0
0 0 0 − a1 ∂τ∂x + ∂
∂x
⎡1 ∂ ∂ ⎤
b ∂τy + ∂y 0 0 0
⎢0 − 1b ∂τ∂y + ∂ ⎥
⎢ ∂y 0 0 ⎥
Dy = ⎢ ⎥ (14.125)
⎣0 0 1 ∂
b ∂τy + ∂
∂y 0 ⎦
0 0 0 − 1b ∂τ∂y + ∂
∂y
Substituting (14.118) into the original equation (14.91), using the differenti-
ation rules for idempotent basis, and assuming that a = b = ε, gives

2
∂ U1 ∂ 2 U1
Δτ U1 + 2ε + + ε2 ΔU1 = ε2 [q(x, y) + kU1 ]

2
∂ U2 ∂ 2 U2
Δτ U2 + 2ε − + ε2 ΔU2 = ε2 q(x, y)

2
∂ U3 ∂ 2 U3
Δτ U3 − 2ε − + ε2 ΔU3 = ε2 q(x, y) (14.126)

2
∂ U4 ∂ 2 U4
Δτ U4 − 2ε + + ε2 ΔU4 = ε2 q(x, y)
where the notations Δτ and Δ have the same meaning as those in equations
(14.110).
Equations (14.126) are decoupled, at cost of coupling the boundary con-
ditions though
∂(U1 − U3 ) ∂(U1 − U2 )
|τx =±1 = 0, |τy =±1 = 0
∂τx ∂τy
(U1 − U3 )|τx =±1 = 0, (U1 − U2 )|τy =±1 = 0
∂(U2 − U4 ) ∂(U3 − U4 )
|τx =±1 = 0, |τy =±1 = 0 (14.127)
∂τx ∂τy
(U2 − U4 )|τx =±1 = 0, (U3 − U4 )|τy =±1 = 0
Note that both boundary value problems (14.110) through (14.112) and
(14.126) through (14.127) implement the transition from two to four spa-
tial arguments: {x, y} → {τx , τy , x, y}. The arguments τx and τy naturally
relate to cell problems and incorporate the corresponding elastic components
within the class of closed form solutions.
Fig. 14.14 The map of idempotent basis generated by the asymmetric triangular
waves with parameters: a = b = 1.0, γ1 = 0.2 and γ2 = 0.6; each of the elements is
equal to unity within light domains and zero within dark domains.
Finally, let us introduce two-dimensional idempotent basis generated by

the triangular asymmetric wave; see Fig. 14.14. First, following definitions of
Chapter 4, let us introduce one-dimensional idempotents associated with x
and y coordinates
1
e+
i = [1 − γi + (1 − γi2 )ei ]
2
1
e−
i = [1 + γi − (1 − γi2 )ei ] (14.128)
2
(i = 1, 2)
where e1 = ∂τ (x/a, γ1 )/∂(x/a) and e2 = ∂τ (y/b, γ2 )/∂(y/b).

Now the two-dimensional idempotent basis is given by i1 = e+ +
1 e2 , i2 =
+ − − + − −
e1 e2 , i3 = e1 e2 and i4 = e1 e2 , although further expansions shown in
(14.115) are not valid any more.
References
1. Acary, V., Brogliato, B.: Numerical methods for nonsmooth dynamical sys-
tems: Applications in Mechanics and Electronics. Springer, Heidelberg (2008)
2. Andreaus, U., Casini, P., Vestroni, F.: Non-linear dynamics of a cracked can-
tilever beam under harmonic excitation. International Journal of Non-Linear
Mechanics 42(3), 566–575 (2007)
3. Andrianov, I.V.: Asymptotic solutions for nonlinear systems with high degrees
of nonlinearity. Prikl. Matem. Mekhanika (PMM) 57(5), 941–943 (1993)
4. Andrianov, I.V., Awrejcewicz, J.: Methods of small and large δ in the nonlinear
dynamics—a comparative analysis. Nonlinear Dynam. 23(1), 57–66 (2000)
5. Andrianov, I.V.: Asymptotics of nonlinear dynamical systems with a high
degree of nonlinearity. Doklady Mathematics 66(2), 270–273 (2002)
6. Andrianov, I.V., Awrejcewicz, J., Barantsev, R.G.: Asymptotic approaches
in mechanics: New parameters and procedures. Applied Mechanics Re-
views 56(1), 87–110 (2003)
7. Antonuccio, F.: Hyperbolic numbers and the Dirac spinor (1998),
http://arxiv.org/abs/hep-th/9812036v1
8. Arnol d, V.I.: Mathematical methods of classical mechanics. Springer, New
York (1978)
9. Arnol d, V.I.: Mathematical methods of classical mechanics. Springer, Heidel-
berg (1978)
10. Ascher, U.M., Mattheij, R.M.M., Russell, R.D.: Numerical solution of bound-
ary value problems for ordinary differential equations. Classics in Applied
Mathematics, vol. 13. Society for Industrial and Applied Mathematics (SIAM),
Philadelphia (1995); Corrected reprint of the 1988 original
11. Atkinson, C.P.: On the superposition method for determining frequencies of
nonlinear systems. In: ASME Proceedings of the 4th National Congress of
Applied Mechanics, pp. 57–62 (1962)
12. Auerbach, D., Cvitanovic, P., Eckmann, J.-P., Gunaratne, G., Procaccia, I.:
Exploring chaotic motion through periodic orbits. Phys. Rev. Lett. 58(23),
2387–2389 (1987)
13. Awrejcewicz, J., Bajaj, A.K., Lamarque, C.-H. (eds.): Nonlinearity, bifurca-
tion and chaos: the doors to the future. Part II. World Scientific Publishing
Co., Singapore (1999); Papers from the International Conference held in Do-
bieszków, September 16-18 (1996); Internat. J. Bifur. Chaos Appl. Sci. Engrg.
9(3) (1999)
340 References
14. Awrejcewicz, J., Lamarque, C.-H.: Bifurcation and Chaos in Nonsmooth Me-
chanical Systems. World Scientific, Singapore (2003)
15. Azeez, M.A.F., Vakakis, A.F., Manevitch, L.I.: Exact solutions of the problem
of vibroimpact oscillations of a discrete system with two degrees of freedom.
Prikl. Mat. Mekh. 63(4), 549–553 (1999)
16. Azeez, M.A.F., Vakakis, A.F., Manevich, L.I.: Exact solutions of the problem
of vibro-impact oscillations of a discrete system with two degrees of freedom.
Prikl. Mat. Mekh. 63(4), 549–553 (1999)
17. Babitsky, V.I.: Theory of Vibroimpact Systems and Applications. Springer,
Berlin (1998)
18. Bahler, T.B.: Mathematica for Scientists and Engineers. Addison-Wesley, New
York (1995)
19. Baker Jr., G.A., Graves-Morris, P.: Padè Approximants, 2nd edn. Encyclope-
dia of Mathematics and Its Applications, vol. 59. Cambridge University Press,
Cambridge (1987)
20. Balescu, R.: Statistical Dynamics, Matter out of Equilibrium. Imperial College
Press, Singapore (1997)
21. Bateman, H., Erdelyi, A.: Higher Transcendental Functions. McGraw-Hill,
New York (1955)
22. Belinfante, J.G.F., Kolman, B.: A survey of Lie groups and Lie algebras with
applications and computational methods. Society for Industrial and Applied
Mathematics (SIAM), Philadelphia (1989); Reprint of the 1972 original
23. Bellman, R.: Introduction to Matrix Analysis. McGraw-Hill Company, New
York (1960)
24. Bensoussan, A., Lions, J.-L., Papanicolaou, G.: Asymptotic analysis for peri-
odic structures. North-Holland Publishing Co., Amsterdam (1978)
25. Blazejczyk-Okolewska, B., Czolczynski, K., Kapitaniak, T., Wojewoda, J.:
Chaotic Mechanics in Systems with Impacts and Friction. World Scientific,
Singapore (1999)
26. Boettcher, S., Bender, C.M.: Nonperturbative square-well approximation to a
quantum theory. Journal of Mathematical Physics 31(11), 2579–2585 (1990)
27. Bogoliubov, N., Mitropollsky, Y.: Asymptotic Methods in the Theory of Non-
linear Oscillations. Gordon and Breach, New York (1961)
28. Bogoljubow, N.N., Mitropolski, J.A.: Asymptotische Methoden in der Theorie
der nichtlinearen Schwingungen. Akademie-Verlag, Berlin (1965)
29. Brogliato, B.: Nonsmooth Mechanics: Models, Dynamics and Control.
Springer, Berlin (1999)
30. Brogliato, B.: Impacts in Mechanical Systems: Analysis and Modelling.
Springer-Verlag, Berlin (2000)
31. Caughey, T.K., Vakakis, A.F.: A method for examining steady state solutions
of forced discrete systems with strong non-linearities. International Journal of
Non-Linear Mechanics 26(1), 89–103 (1966)
32. Chati, M., Rand, R., Mukherjee, S.: Modal analysis of a cracked beam. Journal
of Sound and Vibration 207, 249–270 (1997)
33. Chen, S., Shaw, S.W.: Normal modes for piecewise linear vibratory systems.
Nonlinear Dynamics 10, 135–163 (1996)
34. Cooper, K., Mickens, R.E.: Generalized harmonic balance/numerical method
for determining analytical approximations to the periodic solutions of the x4/3
potential. Journal of Sound and Vibration 250, 951–954 (2002)
References 341
35. Coppola, V.T., Rand, R.H.: Computer algebra implementation of Lie trans-
forms for hamiltonian systems: Application to the nonlinear stability of l4.
ZAMM 69(9), 275–284 (1989)
36. Cveticanin, L.: Oscillator with strong quadratic damping force. Publications
de L’institut Mathematique (Nouvelle serie) 85(99), 119–130 (2009)
37. Dankowicz, H., Paul, M.R.: Discontinuity-induced bifurcations in systems with
hysteretic force interactions. Journal of Computational and Nonlinear Dynam-
ics 4(Article 041009), 1–6 (2009)
38. Deprit, A.: Canonical transformations depending on a parameter. Celestial
mechanics (1), 1–31 (1969)
39. Dimentberg, M.F.: Statistical Dynamics of Nonlinear and Time-Varying Sys-
tems. John Wiley & Sons, New York (1988)
40. Dimentberg, M.F., Bratus, A.S.: Bounded parametric control of random vibra-
tions. R. Soc. Lond. Proc. Ser. A Math. Phys. Eng. Sci. 456(2002), 2351–2363
(2000)
41. Dimentberg, M.F., Iourtchenko, D.V., Bratus’, A.S.: Transition from planar
to whirling oscillations in a certain nonlinear system. Nonlinear Dynamics 23,
165–174 (2000)
42. Feeny, B.A., Guran, A., Hinrichs, N., Popp, K.: A historical review on dry
friction and stick-slip phenomena. ASME Applied Mechanics Reviews 51, 321–
341 (1998)
43. Ferrari, L., Boschi, C.D.E.: Nonautonomous and nonlinear effects in general-
ized classical oscillators: A boundedness theorem. Physical Review E 62(3),
R3039–R3042 (2000)
44. Fidlin, A.: Nonlinear Oscillations in Mechanical Engineering. Springer, Hei-
delberg (2005)
45. Filippov, A.F.: Differential equations with discontinuous righthand sides.
Kluwer Academic Publishers Group, Dordrecht (1988) (Translated from the
Russian)
46. Fucik, S., Kufner, A.: Nonlinear differential equations. Elsevier, Amsterdam
(1980); Studies in Applied Mechanics 2. Elsevier Scientific Publishing Com-
pany, Amsterdam
47. Gendelman, O., Manevitch, L.I., Vakakis, A.F., M’Closkey, R.: Energy pump-
ing in nonlinear mechanical oscillators. I. Dynamics of the underlying Hamil-
tonian systems. Trans. ASME J. Appl. Mech. 68(1), 34–41 (2001)
48. Gendelman, O.V.: Modeling of inelastic impacts with the help of smooth func-
tions. Chaos, Solitons and Fractals 28, 522–526 (2006)
49. Gendelman, O.V., Manevitch, L.I.: Discrete breathers in vibroimpact chains:
Analytic solutions. Physical Review E 78(026609) (2008)
50. Giacaglia, G.E.O.: Perturbation methods in non-linear systems. Springer, New
York (1972); Applied Mathematical Sciences, Vol. 8
51. Goldsmith, W.: Impact: The Theory and Physical Behaviour of Colliding.
Courier Dover Publications, North Chelmsford (2001)
52. Gradshteyn, I.S., Ryzhik, I.M.: Table of Integrals, Series, and Products, 5th
edn. Academic Press, Boston (1994)
53. Grebogi, C., Ott, E., Yorke, J.A.: Unstable periodic orbits and the dimensions
of multifractal chaotic attractorscontrolling chaos. Physical Review A 37(5),
1711–1724 (1988)
54. Guckenheimer, J., Meloon, B.: Computing periodic orbits and their bifur-
cations with automatic differentiation. SIAM J. Sci. Comput. 22(3), 951–
985(electronic) (2000)
342 References
55. Guran, A., Pfeiffer, F., Popp, K.: Dynamics with Friction: Modeling, Analysis
and Experiments. World Scientific, Singapore (2001)
56. Hahn, W.: Stability of motion. Springer Series in Nonlinear Dynamics.
Springer, New York (1967)
57. Harvey, T.J.: Natural forcing functions in nonlinear systems. ASME Journal
of Applied Mechanics 25, 352–356 (1958)
58. Hascoët, E., Herrmann, H.J., Loreto, V.: Shock propagation in a granular
chain. Phys. Rev. E 59(3), 3202–3206 (1999)
59. Holm, D.D., Lynch, P.: Stepwise precession of the resonant swinging spring.
SIAM J. Applied Dynamical Systems 1(1), 44–64 (2002)
60. Hong, J., Ji, J.-Y., Kim, H.: Power laws in nonlinear granular chain under
gravity. Phys. Rev. Lett. 82(15), 3058–3061 (1999)
61. Hori, G.: Theory of general perturbations with unspecified canonical variables.
Publ. Astron. Soc. Japan 18(4), 287–296 (1966)
62. Hori, G.: Mutual perturbations of 1: 1 commensurable small bodies with the
use of the canonical relative coordinates. I. In: Resonances in the motion of
planets, satellites and asteroids, pp. 53–66. Univ. São Paulo, São Paulo (1985)
63. Hu, H., Xiong, Z.-G.: Oscillations in an x(2m+2)/(2n+1) potential. Journal of
Sound and Vibration 259, 977–980 (2003)
64. Hunt, K.H., Crossley, F.R.E.: Coefficient of restitution interpreted as damping
in vibroimpact. ASME Journal of Applied Mechanics 97, 440–445 (1975)
65. Hutchins, C.M.: A history of violin research. J. Acoust. Soc. Am. 73(5), 1421–
1440 (1983)
66. Ibrahim, R.A., Pilipchuk, V.N., Ikeda, T.: Recent advances in liquid sloshing
dynamics. Applied Mechanics Reviews 54(2), 133–199 (2001)
67. Ibrahim, R.A.: Liquid Sloshing Dynamics. Cambridge University Press, New
York (2005)
68. Ibrahim, R.A.: Vibro-Impact Dynamics: Modeling, Mapping and Applications.
LNACM, vol. 43. Springer, Heidelberg (2009)
69. Ibrahim, R.A., Babitsky, V.I., Okuma, M. (eds.): Vibro-Impact Dynamics of
Ocean Systems and Related Problems. Springer, Heidelberg (2009)
70. Ince, E.L.: Ordinary Differential Equations. Dover, New York (1956)
71. Iomin, A., Fishman, S., Zaslavsky, G.M.: Quantum localization for a kicked
rotor with accelerator mode islands. Physical Review E 65(036215) (2002)
72. Ivanov, A.P.: Dynamics of Systems with Mechanical Collisions. International
Program of Education, Moscow (1997) (in Russian)
73. Ivanov, A.P.: Impact oscillations: linear theory of stability and bifurcations.
Journal of Sound and Vibration 178(3), 361–378 (1994)
74. Jackson, L.B.: Signals, Systems, and Transforms. Addison-Wesley Publishing
Company, New York (1991)
75. Jiang, D., Pierre, C., Shaw, S.W.: Large-amplitude non-linear normal modes of
piecewise linear systems. Journal of Sound and Vibration 272, 869–891 (2004)
76. Kalamkarov, A.L., Andrianov, I.V., Danishevskyy, V.V.: Asymptotic ho-
mogenization of composite materials and structures. Applied Mechanics Re-
views 62(030802), 1–20 (2009)
77. Kamenkov, G.V.: Izbrannye trudy v dvukh tomakh. Tom. I. Izdat, Nauka,
Moscow (1971); Ustoichivost dvizheniya. Kolebaniya. Aerodinamika. [Stability
of motion. Oscillations. Aerodynamics], With a biography of G. V. Kamenkov,
a survey article on his works by V. G. Veretennikov, A. S. Galiullin, S. A.
Gorbatenko and A. L. Kunicyn, and a bibliography, Edited by N. N. Krasovskiı̆
References 343
78. Kauderer, H.: Nichtlineare Mechanik. Springer, Berlin (1958)

79. Kevorkian, J., Cole, J.D.: Multiple scale and singular perturbation methods.
80. Kinney, W.M., Rosenberg, R.M.: On steady state harmonic vibrations of non-
linear systems with many degrees of freedom. ASME Journal of Applied Me-
chanics 33, 406–412 (1966)
81. Kobrinskii, A.E.: Dynamics of Mechanisms with Elastic Connections and Im-
pact Systems. Iliffe Books, London (1969)
82. Koch, C.: Biophysics of Computation. Oxford University Press, Oxford (1999)
Information processing in single neurons
83. Kollatz, L.: The eigen-value problems. Nauka, Moscow (1968)
84. Kosevich, A.M., Kovalev, A.S.: Introduction to Nonlinear Physical Mechanics
(in Russian). Naukova Dumka, Kiev (1989)
85. Kowalczyk, P., Di Bernardo, M., Champneys, A.R., Hogan, S.J., Homer,
M., Piiroinen, P.T., Kuznetsov, Y.A., Kuznetsov, Y.A., Nordmark, A.: Two-
parameter discontinuity-induced bifurcations of limit cycles: Classification and
open problems. International Journal of Bifurcation and Chaos 16(3), 601–629
(2006)
86. Kryloff, N., Bogoliuboff, N.: Introduction to Non-Linear Mechanics. Princeton
University Press, Princeton (1943)
87. Krylov, N.M., Bogolyubov, N.N.: Vvedeniye v nelinejnuyu mekhaniku. AN
UkrSSR, Kiev (1937)
88. Kutz, N.J.: Mode-locked soliton lasers. SIAM Review 48(4), 629–678 (2006)
89. Landau, L.D., Lifschitz, E.M.: Lehrbuch der theoretischen Physik (Landau-
Lifschitz. Band I, 12th edn. Akademie-Verlag, Berlin (1987); Mechanik. [Me-
chanics], Translated from the third Russian edition by Hardwin Jungclaussen,
Edited and with a foreword by Paul Ziesche
90. Lavrent ev, M.A., Shabat, B.V.: Problemy gidrodinamiki i ikh matematich-
eskie modeli, 2nd edn., Izdat. “Nauka”, Moscow (1977)
91. Lee, Y.S., Nucera, F., Vakakis, A.F., McFarland, D.M., Bergman, L.A.: Pe-
riodic orbits, damped transitions and targeted energy transfers in oscillators
with vibro-impact attachments. Physica D 238(18), 1868–1896 (2009)
92. Lee, Y.S., Kerschen, G., Vakakis, A.F., Panagopoulos, P., Bergman, L., McFar-
land, D.M.: Complicated dynamics of a linear oscillator with a light, essentially
nonlinear attachment. Physica D 204, 41–69 (2005)
93. Leine, R.I., Nijmeijer, H., Nijmeijer, H.: Dynamics and Bifurcations of Non-
Smooth Mechanical Systems. Springer, Heidelberg (2006)
94. Lewis, F.L., Dawson, D.M., Abdallah, C.T.: Robot Manipulator Control: The-
ory and Practice. CRC Press, Boca Raton (2004)
95. Lichtenberg, A.J., Lieberman, M.A.: Regular and Stochastic Motion. Springer,
New York (1983)
96. Lyapunov, A.M.: Investigation of a singular case of the problem of stability of
motion. Mat. Sbornik 17, 252–333 (1893)
97. Malkin, I.G.: Some problems of the theory of nonlinear oscillations. Gosu-
darstv. Izdat. Tehn.-Teor. Lit., Moscow (1956)
98. Manciu, M., Sen, S., Hurd, A.J.: Impulse propagation in dissipative and dis-
ordered chains with power-low repulsive potentials. Physica D 157, 226–240
(2001)
99. Manevich, A.I., Manevitch, L.I.: The Mechanics of Nonlinear Systems With
Internal Resonances. Imperial College Press, London (2005)
344 References
100. Manevich, L.I., Mikhlin, Y.V., Pilipchuk, V.N.: Metod normalnykh kolebanii
dlya sushchestvenno nelineinykh sistem, Nauka, Moscow (1989)
101. Manevich, L.I.: New approach to beating phenomenon in coupled nonlinear
oscillatory chains. Archive of Applied Mechanics 77, 301–312 (2007)
102. Manevitch, L.I.: The description of localized normal modes in a chain of non-
linear coupled oscillators using complex variables. Nonlinear Dynamics 25,
95–109 (2001)
103. Manevitch, L.I., Gendelman, O.V.: Oscillatory models of vibro-impact type
for essentially non-linear systems. Proceedings of the Institution of Mechanical
Engineers, Part C: Journal of Mechanical Engineering Science 222(10), 2007–
2043 (2008), doi: 10.1243/09544062JMES1057
104. Manevitch, L.I., Musienko, A.I.: Limiting phase trajectory and beating phe-
nomena in systems of coupled nonlinear oscillators. In: 2nd International Con-
ference on Nonlinear Normal Modes and Localization in Vibrating Systems,
Samos, Greece, June 19-23, pp. 25–26 (2006)
105. Marsden, J.E.: Basic complex analysis. Freeman, San Francisco (1973)
106. Maslov, V.P., Omel janov, G.A.: Asymptotic soliton-like solutions of equations
with small dispersion. Uspekhi Mat. Nauk. 36(3), 63–126 (1981)
107. Mickens, R.E.: Oscillations in an x4/3 potential. J. Sound Vibration 246, 375–
378 (2001)
108. Mikhlin, Y.V., Reshetnikova, S.N.: Dynamical interaction of an elastic system
and a vibro-impact absorber. Mathematical Problems in Engineering (Article
ID 37980), 15 (2006)
109. Mikhlin, Y.V., Volok, A.M.: Solitary transversal waves and vibro-impact mo-
tions in infinite chains and rods. International Journal of Solids and Struc-
tures 37, 3403–3420 (2000)
110. Mikhlin, Y.V., Zhupiev, A.L.: An application of the Ince algebraization to the
stability of the non-linear normal vibration modes. Internat. J. Non-Linear
Mech. 32(2), 393–409 (1997)
111. Minorsky, N.: Introduction to non-linear mechanics. J.W. Edwards, Ann Arbor
(1947)
112. Mitropl’sky, Y.A., Senik, P.M.: Construction of asymptotic solution of an au-
tonomouse system with strong nonlinearity. Doklady AN Ukr.SSR (Ukrainian
Academy of Sciences Reports) 6, 839–844 (1961)
113. Moon, F.C.: Chaotic Vibrations. John Willey & Sons, New York (1987)
114. Moser, J.: Recent developments in the theory of Hamiltonian systems. SIAM
Rev. 28(4), 459–485 (1986)
115. Moser, J.K.: Lectures on Hamiltonian systems. In: Mem. Amer. Math. Soc.
No. 81, p. 60. Amer. Math. Soc., Providence (1968)
116. Nayfeh, A.H.: Perturbation methods. John Wiley & Sons, New York (1973);
Pure and Applied Mathematics
117. Nayfeh, A.H.: Perturbation methods in nonlinear dynamics. In: Nonlinear
dynamics aspects of particle accelerators (Santa Margherita di Pula, 1985),
pp. 238–314. Springer, Berlin (1986)
118. Nayfeh, A.H.: Method of normal forms. John Wiley & Sons Inc., New York
(1993); A Wiley-Interscience Publication
119. Nayfeh, A.H.: Nonlinear interactions: analytical computational, and exper-
imental methods. John Wiley & Sons Inc., New York (2000); A Wiley-
Interscience Publication
References 345
120. Nayfeh, A.H., Balachandran, B.: Applied nonlinear dynamics. John Wiley
& Sons Inc, New York (1995) Analytical, computational, and experimental
methods. A Wiley-Interscience Publication
121. Nesterenko, V.F.: Dynamics of Heterogeneous Materials. Springer, New York
(2001)
122. Nesterov, S.V.: Examples of nonlinear Klein-Gordon equations, solvable in
terms of elementary functions. In: Proceedings of Moscow Institute of Power
Engineering, vol. 357, pp. 68–70 (1978)
123. Ott, E., Grebogi, C., Yorke, J.A.: Controlling chaos. Phys. Rev. Lett. 64(11),
1196–1199 (1990)
124. Ozorio de Almeida, A.M.: Hamiltonian systems: chaos and quantization. Cam-
bridge University Press, Cambridge (1988)
125. Parker, T.S., Chua, L.O.: Practical numerical algorithms for chaotic systems.
126. Peat, F.D.: Synchronicity: the bridge between matter and mind. Bantam
Books, New York (1988)
127. Peterka, F.: Introduction to Oscillations of Mechanical Systems with Internal
Impacts (in Czech). Academia, Prague (1981)
128. Pfeiffer, F.: Mechanical system dynamics. Springer, Heidelberg (2008)
129. Pfeiffer, F., Glocker, C.: Multibody dynamics with unilateral contacts. Wiley,
New York (1996)
130. Pfeiffer, F., Kunert, A.: Rattling models from deterministic to stochastic pro-
cesses. Nonlinear Dynamics 1(1), 63–74 (1990)
131. Pierce, J.R.: Coupling of modes of propagation. Journal of Applied
Physics 25(2), 179–183 (1954)
132. Pilipchuk, V.N.: The calculation of strongly nonlinear systems close to vi-
broimpact systems. Journal of Applied Mathematics and Mechanics 49(5),
572–578 (1985)
133. Pilipchuk, V.N.: Transformation of oscillating systems by means of a pair of
nonsmooth periodic functions. Dokl. Akad. Nauk Ukrain. SSR Ser. A (4),
37–40, 87 (1988)
134. Pilipchuk, V.N.: Transformation of the vibratory-systems by means of a pair
of nonsmooth periodic-functions. Dopovidi Akademii Nauk Ukrainskoi Rsr
Seriya A-Fiziko-Matematichni Ta Technichni Nauki (in Ukrainian) 4, 36–38
(1988)
135. Pilipchuk, V.N.: On the computation of periodic processes in mechanical sys-
tems with the impulsive excitation. In: XXXI Sympozjon “Modelowanie w
Mechanice”, Zeszyty Naukowe Politechniki Slaskiej, Z.107, Gliwice (Poland),
pp. 335–342. Politechnica Slaska (1992)
136. Pilipchuk, V.N.: On special trajectories in configuration space of non - linear
vibrating systems. Mekhanika Tverdogo Tela (Mechanics of Solids) 3, 36–47
(1995)
137. Pilipchuk, V.N.: Analytical study of vibrating systems with strong non-
linearities by employing saw-tooth time transformations. J. Sound Vibra-
tion 192(1), 43–64 (1996)
138. Pilipchuk, V.N.: On the computation of mechanical systems with impulse
excitation. Prikl. Mat. Mekh. 60(2), 223–232 (1996)
139. Pilipchuk, V.N.: Application of special nonsmooth temporal transformations
to linear and nonlinear systems under discontinuous and impulsive excitation.
Nonlinear Dynam. 18(3), 203–234 (1999)
346 References
140. Pilipchuk, V.N.: Non-smooth spatio-temporal transformation for impulsively

forced oscillators with rigid barriers. J. Sound Vibration 237(5), 915–919
(2000)
141. Pilipchuk, V.N.: Principal trajectories of the forced vibration for discrete and
continuous systems. Meccanica 35(6), 497–517 (2000)
142. Pilipchuk, V.N.: Non-smooth time decomposition for nonlinear models driven
by random pulses. Chaos Solitons Fractals 14(1), 129–143 (2002)
143. Pilipchuk, V.N.: Some remarks on nonsmooth transformations of space and
time for oscillatory systems with rigid barriers. Prikl. Mat. Mekh. 66(1), 33–40
(2002)
144. Pilipchuk, V.N.: Temporal transformations and visualization diagrams for
nonsmooth periodic motions. International Journal of Bifurcation and
Chaos 15(6), 1879–1899 (2005)
145. Pilipchuk, V.N.: A periodic version of Lie series for normal mode dynamics.
Nonlinear Dynamics and System Theory 6(2), 187–190 (2006)
146. Pilipchuk, V.N., Ibrahim, R.A.: The dynamics of a non-linear system sim-
ulating liquid sloshing impact in moving structures. Journal of Sound and
Vibration 205(5), 593–615 (1997)
147. Pilipchuk, V.N., Ibrahim, R.A.: Application of the Lie group transformations
to nonlinear dynamical systems. Trans. ASME J. Appl. Mech. 66(2), 439–447
(1999)
148. Pilipchuk, V.N., Ibrahim, R.A.: Dynamics of a two-pendulum model with
impact interaction and an elastic support. Nonlinear Dynam. 21(3), 221–247
(2000)
149. Pilipchuk, V.N., Starushenko, G.A.: On the representation of periodic solu-
tions of differential equations by means of an oblique-angled saw-tooth trans-
formation of the argument. Dopov. Nats. Akad. Nauk Ukr. Mat. Prirodozn.
Tekh. Nauki (11), 25–28 (1997)
150. Pilipchuk, V.N., Starushenko, G.A.: A version of non-smooth transformations
for one-dimensional elastic systems with a periodic structure. Journal of ap-
plied mathematics and mechanics 61(2), 265–274 (1997)
151. Pilipchuk, V.N., Vakakis, A.F.: Nonlinear normal modes and wave transmis-
sion in a class of periodic continuous systems. In: Dynamics and control of
distributed systems, pp. 95–120. Cambridge Univ. Press, Cambridge (1998)
152. Pilipchuk, V.N., Vakakis, A.F.: Study of the oscillations of a nonlinearly sup-
ported string using nonsmooth transformations. Journal of Vibration and
Acoustics 120(2), 434–440 (1998)
153. Pilipchuk, V.N., Vakakis, A.F., Azeez, M.A.F.: Study of a class of subharmonic
motions using a nonsmooth temporal transformations (NSTT). Physica D 100,
145–164 (1997)
154. Pilipchuk, V.N., Vakakis, A.F., Azeez, M.A.F.: Study of a class of subharmonic
motions using a non-smooth temporal transformation (NSTT). Phys. D 100(1-
2), 145–164 (1997)
155. Pilipchuk, V.N.: Auto-localized modes in array of nonlinear coupled oscilla-
tors. In: Manevich, A.I. (ed.) Problemy nelineinoi mekhaniki i fiziki materialov,
Dnipropetrovsk, pp. 229–235 (1999) ISBN: 966-7476-10-3
156. Pilipchuk, V.N.: Impact modes in discrete vibrating systems with bilateral
barriers. International Journal of Nonlinear Mechanics 36(6), 999–1012 (2001)
157. Pilipchuk, V.N.: Transient mode localization in coupled strongly nonlinear
exactly solvable oscillators. Nonlinear Dynamics 51(1-2), 245–258 (2008)
References 347
158. Pilipchuk, V.N.: Transitions from strongly to weakly-nonlinear dynamics in a

class of exactly solvable oscillators and nonlinear beat phenomena. Nonlinear
Dynamics 52(4), 263–276 (2008)
159. Pilipchuk, V.N.: Transition from normal to local modes in an elastic beam
supported by nonlinear springs. Journal of Sound and Vibration 322, 554–563
(2009)
160. Poincaré, H.: Les méthodes nouvelles de la mécanique céleste. Tome I. Librairie
Scientifique et Technique Albert Blanchard, Paris, Solutions périodiques. Non-
existence des intégrales uniformes. Solutions asymptotiques. [Periodic solu-
tions. Nonexistence of uniform integrals. Asymptotic solutions], Reprint of
the, original, With a foreword by J. Kovalevsky, Bibliothèque Scientifique Al-
bert Blanchard. [Albert Blanchard Scientific Library] (1987)
161. Poincaré, H.: Science and method. Thoemmes Press, Bristol (1996); Trans-
lated by Francis Maitland, With a preface by Bertrand Russell, Reprint of the
1914 edition
162. Popp, K.: Non-smooth mechanical systems. Journal of Applied Mathematics
and Mechanics (PMM) 64(5), 765–772 (2000)
163. Qaisi, M.I.: Non-linear normal modes of a lumped parameter system. Journal
of Sound and Vibration 205, 205–211 (1997)
164. Ramos, J.I.: Piecewise-linearized methods for oscillators with fractional-power
nonlinearities. Journal of Sound and Vibration 300, 502–521 (2007)
165. Richtmyer, R.D.: Principles of advanced mathematical physics, vol. I.
Springer, New York (1978); Texts and Monographs in Physics
166. Richtmyer, R.D.: Principles of Advanced Mathematical Physics. Springer,
Berlin (1985)
167. Rosenberg, R.M.: The Ateb(h)-functions and their properties. Quart. Appl.
Math. 21, 37–47 (1963)
168. Rosenberg, R.M.: Steady-state forced vibrations. Internat. J. Non-Linear
Mech. 1, 95–108 (1966)
169. Rowat, P.F., Selverston, A.I.: Oscillatory mechanisms in pairs of neurons
connected with fast inhibitory synapses. Journal of Computational Neuro-
science 4, 103–127 (1997)
170. Salenger, G., Vakakis, A.F., Gendelman, O., Manevitch, L., Andrianov, I.:
Transitions from strongly to weakly nonlinear motions of damped nonlinear
oscillators. Nonlinear Dynam. 20(2), 99–114 (1999)
171. Salenger, G.D., Vakakis, A.F.: Localized and periodic waves with discreteness
effects. Mech. Res. Comm. 25(1), 97–104 (1998)
172. Samoı̆lenko, A.M., Boı̆chuk, A.A., Zhuravlev, V.F.: Weakly nonlinear bound-
ary value problems for operator equations with impulse action. Ukraı̈n. Mat.
Zh., 49(2):272–288 (1997)
173. Scherz, P.: Practical Electronics for Inventors. McGraw-Hill, New York (2006)
174. Scott, A.C., Lomdahl, P.S., Eilbeck, J.C.: Between the local-mode and normal-
mode limits. Chemical Physics Letters 113(1), 29–36 (1985)
175. Sheng, G., Dukkipati, R., Pang, J.: Nonlinear dynamics of sub-10 nm flying
height air bearing slider in modern hard disk recording system. Mechanism
and Machine Theory 41, 1230–1242 (2006)
176. Sobczyk, G.: The hyperbolic number plane. The College Mathematics Jour-
nal 26(4), 268–280 (1995)
177. Sophianopoulos, D.S., Kounadis, A.N., Vakakis, A.F.: Complex dynamics of
perfect discrete systems under partial follower forces. Internat. J. Non-Linear
Mech. 37(6), 1121–1138 (2002)
348 References
178. Stakgold, I.: Green’s Functions and Boundary Value Problems. Wiley Inter-
science, New York (1979)
179. Starushenko, G., Krulik, N., Tokarzewski, S.: Employment of non-symmetrical
saw-tooth argument transformation method in the elasticity theory for layered
composites. International Journal of Heat and Mass Transfer 45, 3055–3060
(2002)
180. Stronge, W.J.: Impact Mechanics. Cambridge University Press, Cambridge
(2000)
181. Thomsen, J.J., Fidlin, A.: Near-elastic vibro-impact analysis by discontinuous
transformations and averaging. Journal of Sound and Vibration 311, 386–407
(2008)
182. Timoshenko, S.P., Yang, D.H., Wiver, U.: Kolebaniya v inzhenernom dele.
Mashinostroeniye, Moscow (1985)
183. Tippetts, J.R.: Analysis of idealised oscillatory pipe flow. In: 2nd Interna-
tional Symposium on Fluid - Control, Measurement, Mechanics - and Flow
Visualisation, Sheffield, England, September 5-9 (1988)
184. Toda, M.: Nonlinear lattice and soliton theory. IEEE Transactions on Circuits
and Systems 30(8), 542–554 (1983)
185. Turner, J.D.: On the simulation of discontinuous functions. Journal of Applied
Mechanics 68, 751–757 (2001)
186. Ueda, Y.: Randomly transitional phenomena in the system governed by Duff-
ing’s equation. J. Statist. Phys. 20(2), 181–196 (1979)
187. Ulrych, S.: Relativistic quantum physics with hyperbolic numbers. Physics
Letters B 625, 313 (2005)
188. Uzunov, I.M., Muschall, R., Golles, M., Kivshar, Y.S., Malomed, B.A., Led-
erer, F.: Pulse switching in nonlinear fiber directional couplers. Phys. Rev.
E 51, 2527–2537 (1995)
189. Vakakis, A.F., Atanackovic, T.M.: Buckling of an elastic ring forced by a
periodic array of compressive loads. ASME Journal of Applied Mechanics 66,
361–367 (1999)
190. Vakakis, A.F., Manevitch, L.I., Mikhlin, Y.V., Pilipchuk, V.N., Zevin, A.A.:
Normal modes and localization in nonlinear systems. John Wiley & Sons Inc.,
New York (1996); A Wiley-Interscience Publication
191. Vedenova, E.G., Manevich, L.I., Pilipchuk, V.N.: Normal oscillations of a
string with concentrated masses on nonlinearly elastic supports. Prikl. Mat.
Mekh. 49(2), 203–211 (1985)
192. Vestroni, F., Luongo, A., Paolone, A.: A perturbation method for evaluating
nonlinear normal modes of a piecewise linear two-degrees-of-freedom system.
Nonlinear Dynamics 54(4), 379–393 (2008)
193. Waluya, S.B., van Horssen, W.T.: On the periodic solutions of a generalized
non-linear Van-der-Pol oscillator. Journal of Sound and Vibration 268, 209–
215 (2003)
194. Whitham, G.B.: Linear and nonlinear waves. John Wiley & Sons Inc., New
York (1999); Reprint of the 1974 original. A Wiley-Interscience Publication
195. Whittaker, E.T., Watson, G.N.: A Course of Modern Analysis, 4th edn. Cam-
bridge University Press, Cambridge (1986)
196. Wiercigroch, M., de Kraker, B. (eds.): Applied Nonlinear Dynamics ans Chaos
of Mechanical Systems with Discontinuities, vol. 28. World Scientific, Singa-
pore (2000)
References 349
197. Zevin, A.A.: Localization of periodic oscillations in vibroimpact systems. In:

XXXV Symposium Modeling in Mechanics, Gliwice (Poland), pp. 261–266.
Politechnica Slaska (1996)
198. Zhupiev, A.L., Mikhlin, Y.V.: Stability and branching of normal oscillations
forms of nonlinear systems. Prikladnaya Matematica Mekhanika (PMM) 45,
450–455 (1981)
199. Zhuravlev, V.F.: A method for analyzing vibration-impact systems by means
of special functions. Izvestiya AN SSSR Mekhanika Tverdogo Tela (Mechanics
of Solids) 11(2), 30–34 (1976)
200. Zhuravlev, V.F.: Equations of motion of mechanical systems with ideal one-
sided links. Prikl. Mat. Mekh. 42(5), 781–788 (1978)
201. Zhuravlev, V.F.: The method of Lie series in the motion-separation problem
in nonlinear mechanics. Prikl. Mat. Mekh. 47(4), 559–565 (1983)
202. Zhuravlev, V.F.: The application of monomial Lie groups to the problem of
asymptotically integrating equations of mechanics. Prikl. Mat. Mekh. 50(3),
346–352 (1986)
203. Zhuravlev, V.F.: Particular directions in the configuration space of linear os-
cillatory systems. Prikl. Mat. Mekh. 56(1), 16–23 (1992)
204. Zhuravlev, V.F., Klimov, D.M.: Prikladnye metody v teorii kolebanii, Nauka,
Moscow (1988) Edited and with a foreword by A. Yu. Ishlinskiı̆
205. Zhuravlev, V.P., Klimov, D.M.: Applied methods in the theory of vibrations
(in Russian), Nauka, Moscow (1988)
206. Zhusubaliyev, Z.T., Mosekilde, E.: Bifurcations and Chaos in Piecewise-
smooth Dynamical Systems. World Scientific, Singapore (2003)
Appendix 351
APPENDIX 1: MathematicaR (Version 6) notebook for sawtooth power-series

solutions for the oscillator t,t x xm 0.
This module builds sawtooth power series solutions of the oscillator.

The module is easy to modify on polynomial characteristics of the oscillator.
n the number of iterations
m the exponent
STSRn, m : Modulef, h, RHS, Ε, suc, x, X, H,
fx : xm ;
n
h Εi Hi ;
i0
n
RHS Ε h f Εi Xi Ξ
;
i0
suc CancelCoefficientNormalSeriesRHS, Ε, 0, n,

TableΕi , i, 1, n;
X0 Ξ : A Ξ;
DoXk Τ Integratesuck Ξ
Τ, Ξ, 0, Τ,
Assumptions Τ 0 && Rem 1, k, 1, n;
H0 H0 . Solve Τ X0 Τ
X1 Τ 0 . Τ 1, H0 1;
DoHk Hk . Solve Τ Xk
1 Τ 0 . Τ 1, Hk 1, k, 1, n;
TableXk Τ, k, 0, n, TableHk , k, 0, n 1
Successive approximations (execution):
n 3;
AnalyticalSolution STSRn, Α;
DoXi1 AnalyticalSolution1i, i, 1, n
1;
DoHi1 AnalyticalSolution2i, i, 1, n;
DoPrint"X", i, "", Xi Simplify, i, 0, n
DoPrint"H", i, "", Hi Simplify, i, 0, n 1
X0A Τ
A Τ2Α
X1
2Α
A2 Α Α Τ A Τ1Α AΑ 3 2 Α 2 Α Τ A ΤΑ
X2
2 2 Α2 3 2 Α
A12 Α Α Τ2Α A2 Α 1 Α2 4 3 Α AΑ Α 8 10 Α 3 Α2 Τ A ΤΑ 2 Α 3 Α2 Α3 Τ2 A Τ2 Α

X3
2 2 Α3 3 2 Α 4 3 Α
352 Appendix
H0A1Α 1 Α
A1Α Α 1 Α
H1
2 2 Α
A1Α Α 1 Α3
H2
2 2 Α2 3 2 Α
Successive approximation truncated series:

n n1 1
x Xi ; h Hi ; v Τ x;
i0 i0 h
Appendix 353
APPENDIX 2: MathematicaR (Version 6) notebook for sawtooth power-series

expansion of periodic functions.
In[1]:= f ft periodic function of the period T

m "length" of the series; see the example below
smoothness True or False; use smoothness
True to smooth the series
TauSeriesf, T, m, smoothness
: Modulea,s1, s2, NLX, NLY,RE, IM, der, TauSpectrum,
a T 4;
s1 t a Τ; s2 t 2a a Τ;
1 1
RE f . s1
f . s2; IM f . s1 f . s2;
2 2
derF : ApplyD, F, Τ ;
NLX NestListder, RE, 2m
1 . Τ 0; NLY
NestListder, IM, 2m
1 . Τ 0;
i NLX2k i NLX2k
1
DoKX2i 1 , KX2i ,
k1 Factorial2k 2 k1 Factorial2k 1
i NLY2k
KY2i 1 ,
k0 Factorial2k 1
i NLY2k
1
KY2i , i, 1, m
;
k0 Factorial2k
2m Τi Τi
2
Ifsmoothness, X RE . Τ 0
KXi ,
i1 i i
2
2m 2m NLXkΤk1
Y KYiΤi Τi
2 , X ,
i0 k1 Factorialk 1
2m NLYkΤk1
Y
k1 Factorialk 1
In[2]:= Example of usage

Πt 3
ft : Sin
2
T 4;
In[4]:= Expansions with no smoothing procedure

fnosmootht, m : X
Y e . TauSeriesft, T, m, False
. Τ Τ4t T, e e4t T
fnosmootht, 1
fnosmootht, 2
354 Appendix
fnosmootht, 3
fnosmootht, 4
fnosmootht, 5
Out[5]= 0
1
Out[6]= Π3 Τt3
8
1 1
Out[7]= Π3 Τt3 Π5 Τt5
8 64
1 1 13 Π7 Τt7
8 64 15 360
1 1 13 Π7 Τt7 41 Π9 Τt9
8 64 15 360 1 548 288
In[10]:= Expansions with smoothing procedure

fsmootht, m : X
Y e . TauSeriesft, T, m, True
. Τ Τ4t T, e e4t T
fsmootht, 1
fsmootht, 2
fsmootht, 3
fsmootht, 4
Out[11]= 0
3 Τt3 Τt5
Out[12]= Π3
8 3 5
3 Τt3 Τt5 3 Π3 5 Π5 Τt5 Τt7

Out[13]= Π3
8 3 5 8 64 5 7
3 Τt3 Τt5 3 Π3 5 Π5 Τt5 Τt7

Out[14]= Π3
8 3 5 8 64 5 7
3 Π3 5 Π5 91 Π7 Τt7 Τt9

8 64 15 360 7 9
In[15]:= Defining the basis functions

2 Π
ΤΞ : ArcSin Sin Ξ
;
Π 2
Π
eΞ : Sign Cos Ξ
;
2
Appendix 355
In[17]:= Convergence with no smoothing
PlotEvaluateft, fnosmootht, 2, fnosmootht, 3,

fnosmootht, 4, fnosmootht, 5, t, 0, 2 T,
PlotStyle Thickness.005, Color Black,
Thickness.002, Dashing0.01, 0.01, Color Black,
Thickness.002,Color Black,Thickness.002,Color Black,
Thickness.003, Color Black,
AxesLabel "t", "x", PlotRange All
x
4
Out[17]=
t
2 4 6 8
2
4
In[18]:= Convergence with smoothing
PlotEvaluateft, fsmootht, 2, fsmootht, 3,

fsmootht, 4, t, 0, 2 T,
PlotStyle Thickness.005, Color Black,
Thickness.002, Dashing0.01, 0.01, Color Black,
Thickness.002, Color Black,
Thickness.003, Color Black, AxesLabel "t", "x",
PlotRange All
1.5
1.0
0.5
Out[18]=
t
2 4 6 8
0.5
1.0
1.5
Appendix 357
APPENDIX 3: MathematicaR (Version 4) notebook.

NSTT & Shooting method for periodic solutions of the oscillator
d 2 x dx t
Ζ dt
Ε x3 B e a .
dt2
px, v, t : Ζ v

Ε x3 ;
Π
a ;
2Ω
Substitution xt XΤt a

YΤt aet a
applied to the equation of motion:
1 1 1
eqX Τ,Τ XΤ
pXΤ
YΤ, Τ XΤ
Τ YΤ, a Τ
a2 2 a
1

pXΤ YΤ, Τ XΤ
Τ YΤ, 2 a a Τ
0 Simplify;
a
1 1 1
eqY Τ,Τ YΤ
pXΤ
YΤ, Τ XΤ
Τ YΤ, a Τ
a2 2 a
1
pXΤ YΤ, Τ XΤ
Τ YΤ, 2 a a Τ
B Simplify;
a
Parameters:
Ω 1.0; Ζ 0.05; Ε 1.0; B 7.4;
Clearg, h; dg 3; dh 20;

solg, h : NDSolveeqX, eqY, X1 g, X '1 0, Y1 0,
Y '1 h, X, Y, Τ, 1, 1, MaxSteps Infinity;
Xng, h : X '1 . solg, h1;
Yng, h : Y1 . solg, h1;
plx ContourPlotXng, h, g, dg, dg, h, dh, dh,

Contours 0, ContourShading False, FrameLabel "g", "h",
PlotPoints 100, RotateLabel False, DisplayFunction Identity;
ply ContourPlotYng, h,g, dg, dg,h, dh, dh, Contours 0,
ContourShading False, FrameLabel "g", "h", PlotPoints 100,
RotateLabel False,
ContourStyle Dashing0.01, 0.01, DisplayFunction Identity;
pxy Showplx, ply, DisplayFunction $DisplayFunction;
358 Appendix
15
10
0
h
-5
-10
-15
-20
-3 -2 -1 0 1 2 3
g
Magnified portion of the diagram:
Showpxy, PlotRange 0.3, 0, 5, 4;
-4
-4.2
-4.4
-4.6
-4.8
-0.25 -0.2 -0.15 -0.1 -0.05 0

g
Appendix 359
Find one of the roots:
g, h g, h . FindRootXng, h 0, Yng, h 0, g, 0.3, 0,

h, 5, 4
0.113351, 4.51957
Check precision:
sln NDSolveeqX, eqY, X1 g, X '1 0, Y1 0, Y '1 h,

X, Y, Τ, 1, 1;
Y1, X '1 . sln1
2.09795 109 , 1.47496 109
Introduce the sawtooth sine and the rectangular cosine:

2 Π
Τt : ArcSin Sin t
;
Π 2
Π
et : Sign Cos t
;
2
Graphic output compared to solution of the related Cauchy problem:
xt : XΤt a

YΤt a et a . sln1;
1
vt : Y 'Τt a
X 'Τt a et a . sln1;
a
xxtt PlotEvaluatext, t, 0, 4 a, PlotRange All,
AxesLabel "t", "x", PlotStyle Thickness.009,
TextStyle FontSize 14, DisplayFunction Identity;
xxvv ParametricPlotEvaluatext, vt, t, 0, 4a, PlotRange All,
AxesLabel "x", "v", Frame True, PlotStyle Thickness.009,
TextStyle FontSize 14, DisplayFunction Identity;
Cleary;
Tmax 4a; dirsol NDSolve t,t yt
Ζ t yt
Ε yt3 B et a,
y0 x0, y '0 v0, y, t, 0, Tmax,
MaxSteps Infinity; y y . dirsol1;
yyvv ParametricPlotEvaluateyt, y 't, t, 0, Tmax,
PlotRange All, AxesLabel "x", "v",
PlotStyle Thickness.001, TextStyle FontSize 14,
PlotPoints 500, Frame True, DisplayFunction Identity;
ShowGraphicsArrayxxtt, xxvv, yyvv;
360 Appendix
x
3
2
1
t
1 2 3 4 5 6
-1
-2
-3
4
2
0 x
-2
-4
-3 -2 -1 0 1 2 3
4
2
0 x
-2
-4
-3 -2 -1 0 1 2 3
Appendix 361
APPENDIX 4: MathematicaR notebook.

Conducts NSTT of the T-periodic differential equation of the form
t,t x f x, t x, t 0.
In[1]:= see examples below for the meaning of inputs
TauTransformequation, function, argument, period, basis :

Modulesub, eqn, RE, IM,
sub function XΤ
YΤe,
argument function Y 'Τ
X 'Τe a,
argument,argument function X ''Τ
Y ''Τe a2 ;
Q Partequation, 1 . sub;
1
eqn Q . t a Τ
Q . t 2a a Τ
2
1

Q . t a Τ Q . t 2a a Τ e . a period 4;
2
1
EQNX Simplify eqn . e 1
eqn . e 1
;
2
1
EQNY Simplify eqn . e 1 eqn . e 1
;
2
1 1
subid XΤ X
Τ
X Τ, YΤ X
Τ X Τ,
2 2
1 1
X 'Τ Τ X
Τ
Τ X Τ, Y 'Τ Τ X
Τ Τ X Τ,
2 2
1 1
X ''Τ Τ,Τ X
Τ
X Τ, Y ''Τ Τ,Τ X
Τ X Τ;
2 2
IDP SimplifyEQNX
EQNY . subid;
IDM SimplifyEQNX EQNY . subid;
Ifbasis 1, EQNX 0, EQNY 0, Y1 0, Y1 0, X '1
0, X '1 0, IDP 0, IDM 0,X
1 X 1 0,
X
1 X 1 0, X
'1
X '1 0, X
'1
X '1 0
EXAMPLES
In[2]:= If basis 1, then the output is created in the standard

basis 1,e, otherwise the output is in the idempotent basis
e
,e 1
e 2,1e 2
In[3]:= TauTransformx ''t

xt3 0, xt, t, 4a, 1
X Τ Y Τ
Out[3]= XΤ3 3 XΤ YΤ2 0, 3 XΤ2 YΤ YΤ3 0,
a2 a2
Y1 0, Y1 0, X 1 0, X 1 0

362 Appendix

xt3 0, xt, t, 4a, 0
X Τ X Τ
Out[4]= X Τ3 0, X Τ3 0, X 1 X 1 0,
a2 a2
X 1 X 1 0, X 1 X 1 0, X 1 X 1 0


xt3 P Sint 0, xt, t, 2Π, 0
ΠΤ 4 X Τ ΠΤ 4 X Τ
Out[5]= P Sin X Τ3 0,P Sin X Τ3 0,
2 Π2 2 Π2
X 1 X 1 0, X 1 X 1 0, X 1 X 1
0, X 1 X 1 0


xt3 P Sint 0, xt, t, 2Π, 1
ΠΤ 4 X Τ
Out[6]= P Sin XΤ3 3 XΤ YΤ2 0, 3 XΤ2 YΤ
2 Π2
4 Y Τ
YΤ3 0, Y1 0, Y1 0, X 1 0, X 1 0

Π2
In[7]:= If present, the function ee4t T must be shown with
no argument; derivatives de dt are not allowed by this code,
but necessary generalizations are easy to implement

1
Α e xt
Β xt3 0, xt, t, 4a, 1
X Τ
Out[8]= XΤ Β XΤ3 Α YΤ 3 Β XΤ YΤ2 0,
a2

Y Τ
Α XΤ YΤ 3 Β XΤ2 YΤ Β YΤ3 0, Y1 0,
a2
Y1 0, X 1 0, X 1 0


1
Α e xt
Β xt3 0, xt, t, 4a, 0
X Τ
Out[9]= 1 Α X Τ Β X Τ3 0, 1 Α X Τ Β X Τ3
a2
X Τ
0, X 1 X 1 0, X 1 X 1 0, X 1
a2
X 1 0, X 1 X 1 0

PROJECT: Using the idempotent basis, find T-periodic solution of the linear
oscillator under external and parametric rectangular wave periodic excitation
of the period T=4,
Appendix 363
1
t,t x 1 2 etΩ2 x pet.
In[10]:= Formulates boundary value problem in the idempotent basis:
T 4;
1
bvp TauTransformx ''t
Ω2 1
e xt p e 0, xt, t, T, 0
2
3 1
Out[11]= p Ω2 X Τ X Τ 0, p Ω2 X Τ X Τ 0,
2 2
X 1 X 1 0, X 1 X 1 0, X 1 X 1 0,
X 1 X 1 0

In[12]:= Solves the boundary value problem analytically:
sol DSolvebvp, X
Τ, X Τ, Τ Simplify
Ω 3 3 3 Ω
2p 3 Cos Sin 2 Ω Cos 2 Ω 4 Cos 2 Τ Ω Sin
2 2
Out[12]= X Τ ,
Ω 3 3 Ω
3 Ω2 3 Cos Sin 2 Ω Cos 2 Ω Sin
2 2
Ω 3 ΤΩ 3
X Τ 2 p 3 3 Cos Sin Ω 4 3 Cos Sin Ω
2 2 2 2
3 Ω Ω 3
3 Cos Ω Sin 3 Ω2 3 Cos Sin Ω
2 2 2 2
3 Ω
Cos Ω Sin
2 2
In[13]:= Extracting two components of the solution:
X
Τ X
Τ . sol1;
X Τ X Τ . sol1;
In[15]:= Back to the original xt variables:

1 1
x X
Τ 1
e
X Τ 1 e . Τ Τt, e et;
2 2
1 1 1
v Τ X
Τ 1
e Τ X Τ 1 e .Τ Τt,e et;
T 4 2 2
In[17]:= Basis functions:
364 Appendix
2 Π
Τt : ArcSin Sin t
;
Π 2
Π
et : Sign Cos t
;
2
In[19]:= Parameters and graphic output:
Ω 20.0;
p 1.0;
PlotEvaluatex, t, 0, 8, PlotRange All,
AxesLabel "t", "x ", PlotStyle Thickness.005
x
0.010
0.005
Out[21]=
2 4 6 8
t
0.005
0.010
In[22]:= Validating the code by durect numerical solution:

x0 x . t 0;
v0 v . t 0;
1
dirsol NDSolvey ''t
Ω2 1
e t ytp et 0,y0 x0,
2
y '0 v0, y, t, 0, 50, MaxSteps Infinity
;
y y . dirsol1;
In[26]:= PlotEvaluateyt, t, 0, 8, PlotRange All,

AxesLabel "x", "v", PlotStyle Thickness.003
v
0.010
0.005
Out[26]= x
2 4 6 8
0.005
0.010

Nonlinear Dynamics Between Linear and Impact Limits PDF

Încărcat de

Informații document

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Nonlinear Dynamics Between Linear and Impact Limits PDF

Încărcat de

Drepturi de autor:

Formate disponibile

Lecture Notes in Applied

and Computational Mechanics

Prof. Dr.-Ing. Friedrich Pfeiffer

Vol. 52: Pilipchuk, V.N. Vol. 40: Pfeiffer, F.

Vol. 50: Ganghoffer, J.-F., Pastrone, F. (Eds.) Vol. 38: Marynowski, K.

Vol. 37: Chaudhary, H., Saha, S.K.

Between Linear and Impact Limits

ISBN: 978-3-642-12798-4 e-ISBN: 978-3-642-12799-1

DOI 10.1007/ 978-3-642-12799-1

Lecture Notes in Applied and Computational Mechanics ISSN 1613-7736

Library of Congress Control Number: 2010925555

© Springer-Verlag Berlin Heidelberg 2010

Printed on acid-free paper

The main objective of this book is to introduce a uniﬁed physical basis

point to its complementary counterpart. However, the related mathematical

2 Smooth Oscillating Processes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

3 Nonsmooth Processes as Asymptotic Limits . . . . . . . . . . . . . 51

4 Nonsmooth Temporal Transformations (NSTT) . . . . . . . . . . 93

4.3 Idempotent Basis Generated by Asymmetric Triangular

5 Sawtooth Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

6 NSTT for Linear and Piecewise-Linear Systems . . . . . . . . . 145

6.6.3 Phase Solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168

7 Periodic and Transient Nonlinear Dynamics under

8 Strongly Nonlinear Vibrations . . . . . . . . . . . . . . . . . . . . . . . . . . . 195

9 Strongly Nonlinear Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241

10 Impact Modes and Parameter Variations . . . . . . . . . . . . . . . . 245

10.5.1 A Double-Pendulum with Amplitude Limiters . . . . . . . 258

11 Principal Trajectories of Forced Vibrations . . . . . . . . . . . . . . 265

12 NSTT and Shooting Method for Periodic Motions . . . . . . . 275

13 Essentially Non-periodic Processes . . . . . . . . . . . . . . . . . . . . . . 295

14 Spatially-Oscillating Structures . . . . . . . . . . . . . . . . . . . . . . . . . . 305

Abstract. This chapter contains physical and mathematical preliminaries

1.1 Brief Literature Overview

Analytical methods of conventional nonlinear dynamics are based on the clas-

V.N. Pilipchuk: Nonlinear Dynamics, LNACM 52, pp. 1–36, 2010.

feature of the approach since it justiﬁes and simpliﬁes analytical manipula-

Chapters 9 through 11 deal with the concept of nonlinear normal modes.

1.2 Asymptotic Meaning of the Approach

1.2.1 Two Simple Limits of Lyapunov Oscillator

Harmonic oscillator Impact oscillator

Sine and cosine waves Triangular and rectangular waves

Conventional complex numbers Hyperbolic complex numbers

Fourier series Power series

k0 Ak cos kt Bk sin kt k0 X k 0 Τ k Y k 0 Τ k e

Quasi harmonic methods Strongly nonlinear methods

described by quite simple elementary functions such as triangular sine and

the very language of dynamical systems involving time derivatives.3 In ad-

1.2.2 Oscillating Time and Hyperbolic Numbers,

polynomials and other simple functions of the argument τ . Such formalism

Fig. 1.3 illustrates geometrical meaning of the particular case of temporal

where x = x(t) any periodic function as deﬁned in (1.1). whose period is

1.3 Quick ‘Tutorial’

where, strictly speaking, the equalities should be interpreted in terms of the

1.3.2 Viscous Dynamics under the Sawtooth Forcing

ẋ(t) = qτ (t) (1.5)

where τ = τ (t) and e = e(t).

Based on the linear independence of 1, e and ė, equation (1.8) is equivalent

As a simple test, expression (1.12) can be veriﬁed by taking ﬁrst derivative

where qk are constant coeﬃcients.

gives periodic solution

This solution is veriﬁed by the direct substitution of (1.14) in (1.13).

1.3.3 The Rectangular Cosine Input

ẋ(t) = pe(t) (1.15)

k0 Ak cos kt Bk sin kt k0 X k 0 Τ k Y k 0 Τ k e