Ali Hasan Nayfeh The Method of Normal Forms
Related Titles Lalanne, C.
Mechanical Vibration and Shock Volume 1: Sinu...

Author:
Ali Hasan Nayfeh

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Ali Hasan Nayfeh The Method of Normal Forms

Related Titles Lalanne, C.

Mechanical Vibration and Shock Volume 1: Sinusoidal Vibration 2009 ISBN 978-1-84821-122-3

Gossett, E.

Discrete Mathematics with Proof 2009 ISBN 978-0-470-45793-1

Talman, R.

Geometric Mechanics Toward a Uniﬁcation of Classical Physics 2007 ISBN 978-3-527-40683-8

Zauderer, E.

Partial Differential Equations of Applied Mathematics 2006 ISBN 978-0-471-69073-3

Kahn, P. B., Zarmi, Y.

Nonlinear Dynamics Exploration Through Normal Forms 1998 ISBN 978-0-471-17682-4

Nayfeh, A. H.

Introduction to Perturbation Techniques 1993 ISBN 978-0-471-31013-6

Ali Hasan Nayfeh

The Method of Normal Forms Second, Updated and Enlarged Edition

WILEY-VCH Verlag GmbH & Co. KGaA

The Author Prof. Ali Hasan Nayfeh Virginia Polytechnic Institute and State University Department of Engineering Science and Mechanics Blacksburg, VA 24061 USA [email protected]

All books published by Wiley-VCH are carefully produced. Nevertheless, authors, editors, and publisher do not warrant the information contained in these books, including this book, to be free of errors. Readers are advised to keep in mind that statements, data, illustrations, procedural details or other items may inadvertently be inaccurate. Library of Congress Card No.: applied for British Library Cataloguing-in-Publication Data: A catalogue record for this book is available from the British Library. Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliograﬁe; detailed bibliographic data are available on the Internet at http://dnb.d-nb.de. © 2011 WILEY-VCH Verlag GmbH & Co. KGaA, Boschstr. 12, 69469 Weinheim, Germany All rights reserved (including those of translation into other languages). No part of this book may be reproduced in any form – by photoprinting, microﬁlm, or any other means – nor transmitted or translated into a machine language without written permission from the publishers. Registered names, trademarks, etc. used in this book, even when not speciﬁcally marked as such, are not to be considered unprotected by law. Typesetting le-tex publishing services GmbH, Leipzig Printing Binding Cover Design Adam-Design, Weinheim Printed in Singapore Printed on acid-free paper ISBN Print 978-3-527-41097-2 ISBN ePDF 978-3-527-63578-8 ISBN oBook 978-3-527-63580-1 ISBN ePub 978-3-527-63577-1

V

To my youngest son Nader

VII

Contents Preface XI Introduction 1 1 1.1 1.2 1.3 1.4 1.5 1.5.1 1.5.2 1.5.3 1.6 1.7 1.7.1 1.7.2 1.8

SDOF Autonomous Systems 7 Introduction 7 Dufﬁng Equation 9 Rayleigh Equation 13 Dufﬁng–Rayleigh–van der Pol Equation 15 An Oscillator with Quadratic and Cubic Nonlinearities 17 Successive Transformations 17 The Method of Multiple Scales 19 A Single Transformation 21 A General System with Quadratic and Cubic Nonlinearities 22 The van der Pol Oscillator 24 The Method of Normal Forms 25 The Method of Multiple Scales 26 Exercises 27

2 2.1 2.2 2.3 2.4 2.5 2.5.1 2.5.2 2.6 2.7

Systems of First-Order Equations 31 Introduction 31 A Two-Dimensional System with Diagonal Linear Part 34 A Two-Dimensional System with a Nonsemisimple Linear Form 39 An n-Dimensional System with Diagonal Linear Part 40 A Two-Dimensional System with Purely Imaginary Eigenvalues 42 The Method of Normal Forms 43 The Method of Multiple Scales 47 A Two-Dimensional System with Zero Eigenvalues 48 A Three-Dimensional System with Zero and Two Purely Imaginary Eigenvalues 52 The Mathieu Equation 54 Exercises 57

2.8 2.9

VIII

Contents

3 3.1 3.1.1 3.1.2 3.2 3.3 3.4 3.4.1 3.4.2 3.4.3 3.4.4 3.4.5 3.5

Maps 61 Linear Maps 61 Case of Distinct Eigenvalues 62 Case of Repeated Eigenvalues 64 Nonlinear Maps 66 Center-Manifold Reduction 72 Local Bifurcations 76 Fold or Tangent or Saddle-Node Bifurcation 76 Transcritical Bifurcation 79 Pitchfork Bifurcation 80 Flip or Period-Doubling Bifurcation 81 Hopf or Neimark–Sacker Bifurcation 85 Exercises 91

4 4.1 4.1.1 4.1.2 4.2 4.2.1 4.2.2 4.2.3 4.3 4.4 4.4.1 4.4.2 4.4.3 4.4.4 4.4.5 4.5 4.5.1 4.5.2 4.5.3 4.6 4.6.1 4.6.2 4.6.3 4.7

Bifurcations of Continuous Systems 97 Linear Systems 97 Case of Distinct Eigenvalues 98 Case of Repeated Eigenvalues 99 Fixed Points of Nonlinear Systems 100 Stability of Fixed Points 100 Classiﬁcation of Fixed Points 101 Hartman–Grobman and Shoshitaishvili Theorems Center-Manifold Reduction 103 Local Bifurcations of Fixed Points 107 Saddle-Node Bifurcation 108 Nonbifurcation Point 110 Transcritical Bifurcation 111 Pitchfork Bifurcation 113 Hopf Bifurcations 114 Normal Forms of Static Bifurcations 117 The Method of Multiple Scales 117 Center-Manifold Reduction 126 A Projection Method 132 Normal Form of Hopf Bifurcation 137 The Method of Multiple Scales 138 Center-Manifold Reduction 141 Projection Method 144 Exercises 146

5 5.1 5.2 5.3 5.4 5.4.1 5.4.2

Forced Oscillations of the Dufﬁng Oscillator 161 Primary Resonance 161 Subharmonic Resonance of Order One-Third 164 Superharmonic Resonance of Order Three 167 An Alternate Approach 169 Subharmonic Case 171 Superharmonic Case 172

102

Contents

5.5

Exercises

6 6.1 6.2 6.3 6.4 6.5

Forced Oscillations of SDOF Systems 175 Introduction 175 Primary Resonance 176 Subharmonic Resonance of Order One-Half 178 Superharmonic Resonance of Order Two 180 Subharmonic Resonance of Order One-Third 182

7 7.1 7.1.1 7.1.2 7.2 7.2.1 7.2.2 7.2.3 7.2.4 7.3 7.3.1 7.3.2 7.3.3 7.3.4 7.4 7.4.1 7.4.2 7.5 7.5.1 7.5.2 7.6

Parametrically Excited Systems 187 The Mathieu Equation 187 Fundamental Parametric Resonance 188 Principal Parametric Resonance 190 Multiple-Degree-of-Freedom Systems 191 The Case of Ω Near ω 2 C ω 1 194 The Case of Ω Near ω 2 ω 1 194 The Case of Ω Near ω 2 C ω 1 and ω 3 ω 2 194 The Case of Ω Near 2ω 3 and ω 2 C ω 1 195 Linear Systems Having Repeated Frequencies 195 The Case of Ω Near 2ω 1 198 The Case of Ω Near ω 3 C ω 1 199 The Case of Ω Near ω 3 ω 1 200 The Case of Ω Near ω 1 200 Gyroscopic Systems 205 The Case of Ω Near 2ω 1 208 The Case of Ω Near ω 2 ω 1 208 A Nonlinear Single-Degree-of-Freedom System 208 The Case of Ω Away from 2ω 209 The Case of Ω Near 2ω 211 Exercises 212

8 8.1 8.1.1 8.1.2 8.1.3 8.1.4 8.2 8.2.1 8.2.2 8.3 8.4

MDOF Systems with Quadratic Nonlinearities 217 Nongyroscopic Systems 217 Two-to-One Autoparametric Resonance 220 Combination Autoparametric Resonance 222 Simultaneous Two-to-One Autoparametric Resonances Primary Resonances 223 Gyroscopic Systems 225 Primary Resonances 226 Secondary Resonances 227 Two Linearly Coupled Oscillators 229 Exercises 232

9 9.1 9.1.1 9.1.2

TDOF Systems with Cubic Nonlinearities 235 Nongyroscopic Systems 235 The Case of No Internal Resonances 236 Three-to-One Autoparametric Resonance 238

172

223

IX

X

Contents

9.1.3 9.1.4 9.1.5 9.1.6 9.2 9.2.1 9.2.2 9.2.3

One-to-One Internal Resonance 239 Primary Resonances 239 A Nonsemisimple One-to-One Internal Resonance 240 A Parametrically Excited System with a Nonsemisimple Linear Structure 244 Gyroscopic Systems 249 Primary Resonances 250 Secondary Resonances in the Absence of Internal Resonances Three-to-One Internal Resonance 255

10 10.1 10.2 10.3 10.4 10.5 10.6 10.6.1 10.6.2 10.6.3 10.7 10.8 10.8.1 10.8.2 10.9

Systems with Quadratic and Cubic Nonlinearities 257 Introduction 257 The Case of No Internal Resonance 262 The Case of Three-to-One Internal Resonance 263 The Case of One-to-One Internal Resonance 264 The Case of Two-to-One Internal Resonance 266 Method of Multiple Scales 267 Second-Order Form 268 State-Space Form 271 Complex-Valued Form 274 Generalized Method of Averaging 276 A Nonsemisimple One-to-One Internal Resonance 279 The Method of Normal Forms 279 The Method of Multiple Scales 283 Exercises 285

11 11.1 11.1.1 11.1.2 11.2 11.2.1 11.2.2 11.3 11.3.1 11.3.2 11.4 11.5

Retarded Systems 287 A Scalar Equation 287 The Method of Multiple Scales 289 Center-Manifold Reduction 291 A Single-Degree-of-Freedom System 295 The Method of Multiple Scales 296 Center-Manifold Reduction 299 A Three-Dimensional System 304 The Method of Multiple Scales 306 Center-Manifold Reduction 308 Crane Control with Time-Delayed Feedback Exercises 313 References 315 Further Reading 319 Index 325

311

251

XI

Preface This book gives an introductory treatment of the method of normal forms. This technique has its application in many branches of engineering, physics, and applied mathematics. Approximation techniques such as these are important for people working with dynamical problems and are a valuable tool they should have in their tool box. The exposition is largely by means of examples. The readers need not understand the physical bases of the examples used to describe the techniques. However, it is assumed that they have a knowledge of basic calculus as well as the elementary properties of ordinary differential equations. For most of the examples, the results obtained with the method of normal forms are shown to be equivalent to those obtained with other perturbation methods, such as the methods of multiple scales and averaging. As such, new sections are added treating some of the examples with these methods. Moreover, exercises are added to most chapters. Because the normal forms of maps and differential equations are very useful in bifurcation analysis, I added in this edition three chapters dealing with the normal forms and bifurcations of maps, continuous systems, and retarded systems. The normal forms of continuous systems are constructed using the method of multiple scales, a combination of center-manifold reduction and the method of normal forms, and the new method of projection, which is developed ﬁrst in this edition. Also, the normal forms of retarded systems are constructed using center-manifold reduction and the method of multiple scales. In the center-manifold reduction, we represent the retarded equations as operator differential equations, decompose the solution space of their linearized form into stable and center subspaces, deﬁne an inner product, determine the adjoint of the operator equations, calculate the center manifold, carry out details of the projection using the adjoint of the center subspace, and ﬁnally calculate the normal form on the center manifold.

XII

Preface

I am very much indebted to my late parents, Hasan and Khadrah, who in spite of their lack of formal education insisted that all their sons obtain the highest degrees. If it were not for their incredible foresight on the value of an education even under the most severe conditions, I would not have ﬁnished secondary school. This book and its second edition would not have been written without the patience and continuous encouragement of my wife, Samirah. Blacksburg, VA, December 2010

Ali Hasan Nayfeh

1

Introduction The method of normal forms dates back to the days of Euler, Delaunay, Poincaré, Dulac, and Birkhoff. Moreover, the concept of using coordinate transformations to simplify mathematical problems involving algebraic, ordinary differential, partial differential, integral, and integro-differential equations has been used for a long time, as illustrated by the following examples. As a ﬁrst example, we consider Bessel’s equation of order one-half; that is, x 2 u00 C x u0 C x 2 14 u D 0 Using the transformation x 1/2 v (x), we transform this equation into the simple equation v 00 C v D 0 whose solution is v D c 1 cos x C c 2 sin x where c 1 and c 2 are arbitrary constants. Hence, Bessel’s function of order one-half J1/2 (x) is given by J1/2 (x) D x 1/2 (c 1 cos x C c 2 sin x) As a second example, we consider the vibrations of an n degree-of-freedom system governed by the following set of n coupled, linear equations of motion: xR C K x D 0 where x is a column vector of length n and K is an nn constant symmetric matrix. Using the transformation x D P v, we obtain vR C P 1 K P v D 0 Assuming the eigenvalues λ 1 , λ 2 , . . . , and λ n of K to be distinct and choosing the columns of P to be the orthonormal eigenvectors of K, we ﬁnd that P 1 K P is a The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

2

Introduction

diagonal matrix Λ with entries λ 1 , λ 2 , . . . , λ n . Hence, the system of equations can be written as vR C Λ v D 0 or in the decoupled form vR i C λ i v i D 0

and

i D 1, 2, . . . , n

which is called the normal-modal form of xR C K x D 0 As a third example, we consider the system aP D µ a 12 a sin 2β a βP D 12 σ a 12 a cos 2β where µ and σ are constants, which describes the time variation of the amplitude and phase of a parametrically excited linear oscillator in the case of a principal parametric resonance (Nayfeh and Mook, 1979). This system is nonlinear and its solution is not apparent. However, using the nonlinear transformation x D a cos β and y D a sin β, we transform the nonlinear system into the following linear system: xP D µ x C yP D µ y

1 (σ 1) y 2 1 (σ C 1) x 2

whose closed-form solution is readily obtainable. As a fourth example, we consider the nonlinear system xP D y C (ax C b y ) x 2 C y 2 yP D x C (a y b x) x 2 C y 2 where a and b are constants, which describes the motion near a Hopf bifurcation point (Marsden and McCracken, 1976; Wiggins, 1990), as described in Section 4.4.5. Again the solution of this system is not apparent. However, using the nonlinear transformation x D r cos β and y D r sin β, we transform the system into rP D ar 3 βP D 1 C b r 2 whose closed-form solution is readily obtainable. As a ﬁfth example, we consider the linear partial differential equation u t t c 2 u x x D f 0 (x c t) f 00 (x c t)

Introduction

where f is a known twice differential function, the prime denotes the derivative with respect to the argument (x c t), and subscripts denote partial derivatives. The general solution of this equation can be readily obtained if we express the independent variables x and t in terms of the characteristics ξ D x ct

and

η D x C ct

Thus, this partial differential equation is transformed into 4c 2 u ξ η D f 0 (ξ ) f 00 (ξ ) whose general solution is uD

1 02 f (ξ )η C g(ξ ) C h(η) 8c 2

where g(ξ ) and h(η) are general functions of ξ and η. As a sixth example, we consider the nonlinear partial differential equation u t C uu x D v u x x where v is a constant, which is known as Burger’s equation (Whitham, 1974). Replacing u with ψ x and integrating the result once yields ψ t C 12 ψ x2 D v ψ x x Then, using the nonlinear transformation ψ D 2v ln(φ), Hopf (1950) and Cole (1951) transformed the nonlinear equation into the linear heat transfer equation φt D v φxx which can be solved much more easily than the original nonlinear equation. As a seventh example, we consider the steady, incompressible, high-Reynolds number ﬂow over a ﬂat plate aligned with the oncoming uniform stream. The boundary layer approximation to the stream function ψ(x, y ) is governed by Van Dyke (1964) ψy y y C ψx ψy y ψy ψx y D 0 ψ(x, 0) D 0 ψ y (x, 0) D 0 and 0 < x < 1 ψ y (x, 1) D 1 This nonlinear partial differential equation can be reduced to an ordinary differential equation by using the similarity transformation p p ψ(x, y ) D 2x f (η) , η D y / 2x With this transformation, the boundary layer problem becomes f 000 C f f 00 D 0 ,

f (0) D 0 ,

which is the Blasius problem.

f 0 (0) D 0 ,

f 0 (1) D 1

3

4

Introduction

In the preceding examples, transformations were introduced to transform a difﬁcult problem into a more readily solvable problem. Next, we consider cases in which a transformation is used to transform the problem into a new “approximate” problem for which the exact solution can be readily obtained. Speciﬁcally, we consider the Liouville equation y 00 C λ 2 q(x)y D 0 when

λ1

where λ is a constant, q(x) is a known function, and the prime denotes the derivative with respect to x. To determine an approximate solution of this equation when λ 1, we transform both of the dependent and independent variables as z D φ(x) and

v (z) D ψ(x)y (x)

With this transformation, the Liouville equation becomes 2 1 2φ 0 ψ 0 d v ψ 00 2ψ 02 λ q d2 v 00 φ v D0 C C C d z2 φ 02 ψ dz φ 02 ψ φ 02 ψ 2 φ 02 We choose φ and ψ so that the dominant part of the transformed equation has the simplest possible form and, at the same time, has solutions that have qualitatively the same behavior as the solutions of the original equations. In other words, we have to insist on the transformation being regular everywhere in the interval of interest. To this end, we force the coefﬁcient of d v /d z to be zero; that is, 2φ 0 ψ 0 D0 ψ p Hence, ψ D φ 0 . In order that the transformation be regular, ψpmust be regular and have no zeros in the interval of interest. Then, because ψ D φ 0 , φ 0 must be regular and have no zeros in the interval of interest. Consequently, we set φ 00

λ 2 q D φ 02 ξ (z) so that the transformed equation becomes d2 v C ξ (z)v D δ v d z2 and choose the simplest possible function ξ (z) that yields a nonsingular transformation. In order that φ 0 be regular and have no zeros in the interval of interest, ξ (z) must have the same number, type, and order of singularities and zeros as q. For example, when q > 0 everywhere in the interval of interest, the solutions of the original equation are oscillatory, and hence φ and ψ must be chosen so that the dominant part of the transformed equation is d2 v Cv D0 d z2

Introduction

which is the simplest possible equation with oscillatory solutions. When q < 0 everywhere in the interval of interest, one of the solutions of the original equation grows exponentially with x and the other decays exponentially with x. Hence, φ and ψ must be chosen so that the dominant part of the transformed equation is d2 v v D0 d z2 which is the simplest possible equation with exponentially growing and decaying solutions. When q changes sign once in the interval of interest, the solutions of the original equation are oscillatory on one side of the sign change and exponentially growing and decaying on the other side. For example, if q D 1 x 3 , the solutions of the original equation are oscillatory for x < 1 and exponential for x > 1. Hence, φ and ψ must be chosen so that the dominant part of the transformed equation has solutions whose behavior changes from oscillatory to exponentially growing and decaying at a given point. The simplest possible equation with these properties is the Airy equation d2 v zv D 0 d z2 When z > 0 the solutions of this equation are growing and decaying with z, and when z < 0 they are oscillatory. In other words, if q(x) is regular and has only a simple zero (simple turning point) such as 1 x 3 , then ξ (z) must be chosen to be regular and have only a simple zero. The simplest possible function that satisﬁes these requirements is ξ (z) D z. If q(x) is regular and has only a double zero at a point in the interval of interest (i.e., turning point of order 2), ξ (z) must be chosen to be regular and have only a double zero. The simplest possible function satisfying these requirements is ξ (z) D z 2 . If q(x) is regular and has only a zero of order n (i.e., turning point of order n), ξ (z) must be chosen to be z n . If q(x) has two zeros at x D a and b, where b > 1, of order m and n, then one uses ξ (z) D z m (1 z) n In analyzing oscillations of a weakly nonlinear system, the method of variation of parameters is usually used to transform the equations governing these oscillations into the standard form xP D f (xI ) D where

1 X m f (x) m! m mD0

ˇ @ m f ˇˇ f m (x) D @ m ˇD0

Here x and f are vectors with N components. The vector x may represent, for example, the amplitudes and phases of the system. If we denote the components

5

6

Introduction

of the vector f m by f m n , then a component x k of the vector x is said to be a rapidly rotating phase if f 0k ¤ 0. To analyze this standard system, we introduce a near-identity transformation x D X (yI ) D y C X 1 (y) C 2 X 2 (y) C from x to y such that the system is transformed into yP D g(yI ) D

1 X n g (y) n! n nD0

where the g n contain long-period terms only. Using the generalized method of averaging (Nayfeh, 1973), one determines the X n and g n by substituting the transformation into the standard system and separating the short- and long-period terms assuming that the X n contain short-period terms only. Alternatively, we can deﬁne the transformation x D X (yI ) as the solution of the N differential equations dx D W (xI ) , d

x( D 0) D y

The vector W is called the generating vector. This equation generates the so-called Lie transforms (Kamel, 1970), which are invertible because they are close to the identity. It seems at ﬁrst that we are going in circles because we are proposing to simplify the original system of differential equations by solving a system of N differential equations. This is not the case, because we are interested in the solution of the original system for large t, whereas we need the solution of the transformation for small , which is a signiﬁcant simpliﬁcation. These examples clearly show that linear and nonlinear coordinate transformations can be used to simplify linear and nonlinear problems. A powerful method for systematically constructing these transformations is the method of normal forms. The basic idea underlying the method of normal forms is the use of “local” coordinate transformations to “simplify” the equations describing the dynamics of the system under consideration. In other words, with the method of normal forms, one seeks a near-identity coordinate transformation in which the dynamical system takes the “simplest” or so-called normal form. The transformations are generated in a neighborhood of a known solution, such as a ﬁxed point (constant, stationary, or equilibrium solution) or a periodic orbit (limit cycle) of a system. In this text, the normalization is usually carried out with respect to a perturbation parameter.

7

1 SDOF Autonomous Systems 1.1 Introduction

In this chapter, we describe the method of normal forms using single-degreeof-freedom (SDOF) autonomous systems that can be modeled by the following second-order nonlinear ordinary differential equation: P uR C ω 2 u D f (u, u)

(1.1)

where f (u, u) P can be developed in a power series in terms of u and u. P In what follows, we will refer to uP C ω 2 u D 0 as the unperturbed system and (1.1) as the perturbed system. We assume that (1.1) has an equilibrium at u D 0 and uP D 0. Equation 1.1 can be cast as a system of two ﬁrst-order equations by letting x1 D u

and

x2 D uP

(1.2)

The result is xP 1 D x2

(1.3)

xP 2 D ω 2 x1 C f (x1 , x2 )

(1.4)

It is clear that the unperturbed system xP 1 D x2

and

xP2 D ω 2 x1

has a simple pair of purely imaginary eigenvalues ˙i ω. The main idea underlying the method of normal forms is to introduce a nearidentify transformation x1 D y 1 C h 1 (y 1 , y 2 )

(1.5a)

x2 D y 2 C h 2 (y 1 , y 2 )

(1.5b)

from (x1 , x2 ) to (y 1 , y 2 ) into (1.3) and (1.4) to produce the simplest possible equations (the so-called normal form). We call the transformation (1.5) near-identity The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

8

1 SDOF Autonomous Systems

because x1 (t) y 1 (t) and x2 (t) y 2 (t) are small; that is, o(x1 (t), x2 (t)). This procedure is also called normalization. To this end, we substitute (1.5) into (1.3) and (1.4) and obtain yP1 D y 2 C h 2

@h 1 @h 1 yP1 yP 2 @y 1 @y 2

yP2 D ω 2 y 1 ω 2 h 1 C f (y 1 C h 1 , y 2 C h 2 )

(1.6a) @h 2 @h 2 yP1 yP2 @y 1 @y 2

(1.6b)

Then, we choose h 1 and h 2 such that (1.6) assume their simplest form. This task is accomplished in steps. If one decomposes f (x1 , x2 ) as f (x1 , x2 ) D

N X

f n (x1 , x2 )

(1.7)

nD1

where f n is a polynomial of degree n in x1 and x2 , then one chooses h 1 and h 2 to simplify the terms resulting from the lowest-order polynomial f m (x1 , x2 ), where m 2, in f (x1 , x2 ). In the next step, one chooses a second near-identity transformation to simplify the polynomial terms of degree m C 1, and so on. It turns out that, because the unperturbed system (1.3) and (1.4) represents an oscillator, the governing equations can conveniently be expressed as a single complexvalued equation. To this end, we follow steps similar to those used in the method of variation of parameters (Nayfeh, 1981). When f 0, the solution of (1.1) can be expressed as u D B e i ω t C BN e i ω t where B is a constant and BN is the complex conjugate of B. Hence, uP D i ω B e i ω t BN e i ω t

(1.8)

(1.9)

When f ¤ 0, we continue to represent the solution of (1.1) as in (1.8) subject to the constraint (1.9) but with time-varying rather than constant B. Next, we replace B e i ω t with ζ(t) and rewrite (1.8) and (1.9) as N N u D ζ(t) C ζ(t) and uP D i ω ζ(t) ζ(t) (1.10) N we obtain Hence, solving for ζ and ζ, i i 1 1 u uP u C uP and ζN D ζD 2 ω 2 ω Differentiating (1.11) with respect to t yields i i 1 1 uP uR D uP C i ωu ζP D f 2 ω 2 ω on account of (1.1). Hence, 1 i i ζP D i ω u uP f (u, u) P 2 ω 2ω

(1.11)

(1.12)

(1.13)

1.2 Dufﬁng Equation

which, upon using (1.10), becomes i N i ω ζ ζN ζP D i ωζ f ζ C ζ, 2ω

(1.14)

Next, we consider different polynomial forms for f.

1.2 Dufﬁng Equation

The Dufﬁng equation is uR C ω 2 u D α u3 so that, in this case, f D α u3 and (1.14) becomes 3 iα ζP D i ωζ ζ C ζN 2ω

(1.15)

We introduce a near-identity transformation from ζ to η in the form ζ D η C h (η, η) N

(1.16)

and obtain ηP D i ωη C i ωh

3 iα @h @h P ηN ηP η C h C ηN C hN @η @ ηN 2ω

(1.17)

Because the nonlinearity is cubic, we assume that h is third order in η and η; N that is, h D Λ 1 η 3 C Λ 2 η 2 ηN C Λ 3 η ηN 2 C Λ 4 ηN 3

(1.18)

and choose the Λ i so that (1.17) takes the simplest possible (normal) form. In the ﬁrst step, we eliminate ηP and ηPN from the right-hand side of (1.17). This task is accomplished by iteration. To the ﬁrst approximation, it follows from (1.17) that ηP D i ωη

ηPN D i ω ηN

and

(1.19)

Next, we replace ηP and ηNP on the right-hand side of (1.17) using (1.19), use (1.18), keep up to third-order terms, and obtain

α ηP D i ωη i ω 2Λ 1 C 2ω 2 α ηN 3 C i ω 4Λ 4 2ω 2

3i α 2 3α η η ηN 2 η ηN C i ω 2Λ 3 2ω 2ω 2 3

(1.20)

9

10

1 SDOF Autonomous Systems

Next, we choose Λ 1 , Λ 3 , and Λ 4 to eliminate the terms involving η 3 , η ηN 2 , and ηN 3 ; that is, Λ1 D

α , 4ω 2

Λ3 D

3α , 4ω 2

Λ4 D

α 8ω 2

(1.21)

However, because Λ 2 does not appear in (1.20), the term involving η 2 ηN cannot be eliminated; it is called a resonance term. Consequently, to the second approximation, the simplest possible form for ηP is ηP D i ωη

3i α 2 η ηN 2ω

(1.22)

To show that η 2 ηN is a resonance term, we ﬁnd a solution for (1.22) by iteration. To the ﬁrst approximation, η D Ae i ω t , where A is a constant. Then, (1.22) becomes ηP D i ωη

3i α 2 N i ω t A Ae 2ω

whose solution can be written as η D Ae i ω t

3α 2 N i ω t A At e 2ω

(1.23a)

It is clear that this expansion, which is also a straightforward expansion, is nonuniform for large t because of the presence of a secular term created by η 2 η. N Alternatively, we can demonstrate that the term ζ 2 ζN is a resonance term in the original equation (1.15). To the ﬁrst approximation, we neglect the nonlinear term in (1.15) and ﬁnd that ζ D Ae i ω t . Then, to the second approximation, (1.15) becomes i α 3 3i ω t N i ω t C 3A AN2 e i ω t C AN3 e 3i ω t ζP D i ωζ C 3A2 Ae A e 2ω whose solution can be written as ζ D Ae i ω t C

α 3 3i ω t 3i α 2 N i ω t 3α A e A AN2 e i ω t A At e C 2 4ω 2ω 4ω 2

α N3 3i ω t A e 8ω 2

(1.23b)

It is clear that this expansion is nonuniform because of the presence of a secular N The other three terms proportional to A3 e 3i ω t , A AN2 e i ω t , and term created by ζ 2 ζ. 3 3i ω t N A e created by ζ 3 , ζ ζN 2 , and ζN 3 do not produce secular terms and hence they are nonresonance. Consequently, one can choose a near-identity transformation to eliminate them. As a second alternative, starting with the original equation (1.15), we break the N into two parts as nonlinear part f (ζ, ζ) N D f 1 (ζ, ζ) N C f 2 (ζ, ζ) N f (ζ, ζ) where e i ω t f 1 e i ω t , e i ω t

1.2 Dufﬁng Equation

is time invariant, whereas e i ω t f 2 e i ω t , e i ω t is not time invariant. In the present case, 3 f D ζ C ζN , f 1 D 3ζ 2 ζN , f 2 D ζ 3 C 3ζ ζN 2 C ζN 3 Thus, e i ω t f 1 e i ω t , e i ω t D e i ω t 3e 2i ω t e i ω t D 3 which is time invariant, whereas e i ω t f 2 e i ω t , e i ω t D e 2i ω t C 3e 2i ω t C e 4 i ω t which does not contain any time-invariant terms. Substituting (1.16) and (1.18) into (1.10), using (1.21), and setting Λ 2 D 0 because it is arbitrary yields u D η C ηN

α 3 3α 2 η C ηN 3 C η ηN C η 2 ηN 2 2 8ω 4ω

(1.24)

where η is given by (1.22). Next, we separate the fast from the slow variations in η by introducing the transformation η D A(t)e i ω t where ω is the natural frequency of the system and A is a function of time, into (1.22) and (1.24) and obtain 3i α 2 N AP D A A 2ω

(1.25) α 3 3i ω t A e C AN3 e 3i ω t 2 8ω C AN2 Ae i ω t C

N i ω t u D Ae i ω t C Ae C

3α 2 N i ω t A Ae 4ω 2

(1.26)

Expressing A in the polar form A D 12 ae i β where a and β are functions of t, we rewrite (1.26) as 3α 3 α a3 uD aC cos(ωt C β) a cos(3ωt C 3β) C 2 16ω 32ω 2

(1.27)

(1.28)

Substituting (1.27) into (1.25) and separating real and imaginary parts, we have aP D 0 a βP D

(1.29) 3α 3 a 8ω

(1.30)

11

12

1 SDOF Autonomous Systems

In determining the normal form (1.22), we had to use an ordering scheme to indicate the relative magnitudes of the different terms in (1.15). We based the orN ζ ζN 2 , and ζN 3 dering scheme on the fact that ζ and ζN are small and hence ζ 3 , ζ 2 ζ, N In other words, we based the ordering scheme on are much smaller than ζ and ζ. the degree of the terms. This worked well in this example, but there are many physical systems where the ordering does not follow from the degree of the polynomial but from the presence of certain parameters in their models. We consider such an example in the next section. Next, we treat (1.15) by using the method of multiple scales. To this end, we introduce a small nondimensional parameter as a bookkeeping device and rewrite (1.15) as 3 iα ζP D i ωζ ζ C ζN 2ω

(1.31)

Then, we seek an approximate solution of (1.31) in the form ζ(tI ) D ζ0 (T0 , T1 ) C ζ1 (T0 , T1 ) C

(1.32)

where Tn D n t and d @ @ C C D D0 C D1 C D dt @T0 @T1

(1.33)

Substituting (1.32) and (1.33) into (1.31) and equating coefﬁcients of like powers of yields Order (0 )

D0 ζ0 i ωζ0 D 0

(1.34)

Order ()

D0 ζ1 i ωζ1 D D1 ζ0

3 iα ζ0 C ζN 0 2ω

(1.35)

The solution of (1.34) can be expressed as ζ0 D A(T1 )e i ω T0

(1.36)

Then, (1.35) becomes i α 3 3i ω T0 N i ω T0 A e C 3A2 Ae 2ω C 3A AN2 e i ω T0 C AN3 e 3i ω T0

D0 ζ1 i ωζ1 D A0 e i ω T0

(1.37)

Eliminating the terms that lead to secular terms from (1.37), we have A0 D

3i α 2 N A A 2ω

(1.38)

1.3 Rayleigh Equation

Then, a particular solution of (1.37) can be expressed as ζ1 D

α 3 3i ω T0 3α α N3 3i ω T0 A e A e C A AN2 e i ω T0 C 4ω 2 4ω 2 8ω 2

(1.39)

Substituting (1.36) and (1.39) into (1.10), we obtain N i ω t α A3 e 3i ω t C AN3 e 3i ω t u D Ae i ω t C Ae 8ω 2 3α 2 N i ω t A Ae C A AN2 e i ω t C C 2 4ω

(1.40)

Equations 1.38–1.40 are in full agreement with (1.25) and (1.26) obtained with the method of normal forms because T1 D t and can be set equal to unity.

1.3 Rayleigh Equation

The Rayleigh equation is uR C ω 2 u D uP 13 uP 3

(1.41)

where is a small, positive nondimensional parameter. Here f D uP 13 uP 3 and (1.14) becomes h 3 i ζP D i ωζ C 12 ζ ζN C 13 ω 2 ζ ζN

(1.42)

In this example, the ordering is not based on the degree of the polynomial, but on the small nondimensional parameter . Normalization is carried out in terms of the small parameter . In fact, the perturbation contains linear as well as cubic terms. Using the transformation (1.16), we rewrite (1.42) as @h @h P 1 ηP D i ωη C i ωh ηP ηN C η ηN C h hN @η @ ηN 2 3

1 2 (1.43) C ω η ηN C h hN 3 Because the perturbation in (1.43) involves linear and cubic terms, we express h in the form (1.44) h D ∆ 1 η C ∆ 2 ηN C Λ 1 η 3 C Λ 2 η 2 ηN C Λ 3 η ηN 2 C Λ 4 ηN 3 Moreover, to the ﬁrst approximation, ηP and ηPN are given by (1.19). Then, substituting (1.19) and (1.44) into the right-hand side of (1.43) and keeping terms up to O(),

13

14

1 SDOF Autonomous Systems

we obtain 1 i 1 ηN C η iω 2Λ 1 C i ω η 3 ηP D i ωη C 2iω ∆ 2 C 4ω 2 6 1 2 2 1 1 ω η ηN C iω 2Λ 3 i ω η ηN 2 C iω 4Λ 4 C i ω ηN 3 2 2 6 (1.45) We note that (1.45) is independent of ∆ 1 and Λ 2 and hence they are arbitrary. Moreover, the terms proportional to η and η 2 ηN are resonance terms and hence cannot be eliminated from (1.45). Next, we choose ∆ 2 , Λ 1 , Λ 3 , and Λ 4 to eliminate the terms involving η, N η 3 , η ηN 2 , and ηN 3 , thereby producing the simplest possible equation for η. Thus, we have ∆2 D

i , 4ω

Λ1 D

1 iω , 12

Λ3 D

1 iω , 4

Λ4 D

1 iω 24

(1.46)

With this choice, (1.45) takes the normal form ηP D i ωη C 12 η 12 ω 2 η 2 ηN

(1.47)

Again, in this case, we could have identiﬁed the resonance terms in (1.42) by one of the procedures described in Section 1.2. Because the solution of the unperturbed problem is proportional to e i ω t , the resonance terms in N 3 N D ζ ζN C 1 ω 2 (ζ ζ) f (ζ, ζ) 3 are the terms proportional to e i ω t or the time-invariant terms in e i ω t f e i ω t e i ω t , i ω e i ω t e i ω t N is the only resonance A simple calculation shows that the term 1/2(ζ ω 2 ζ 2 ζ) term. Hence, keeping only the resonance terms in (1.42), we have ζP D i ωζ C 12 ζ ω 2 ζ 2 ζN C which is formally equivalent to (1.47). Next, we treat (1.42) with the method of multiple scales. To this end, we substitute (1.32) and (1.33) into (1.42), equate coefﬁcients of equal powers of , and obtain Order (0 )

D0 ζ0 i ωζ0 D 0

(1.48)

Order ()

D0 ζ1 i ωζ1 D D1 ζ0 C

1 2

h

3 i ζ0 ζN 0 C 13 ω 2 ζ0 ζN 0

(1.49)

1.4 Dufﬁng–Rayleigh–van der Pol Equation

The solution of (1.48) can be expressed as ζ0 D A(T1 )e i ω T0

(1.50)

Then, (1.49) becomes N i ω T0 C 1 ω 2 A3 e 3i ω T0 D0 ζ1 i ωζ1 D A0 e i ω T0 C 12 Ae i ω T0 12 Ae 6 N i ω T0 C 1 ω 2 A AN2 e i ω T0 1 ω 2 AN3 e 3i ω T0 12 ω 2 A2 Ae 2 6 (1.51) Eliminating the terms that lead to secular terms from (1.51), we have A0 D 12 A 12 ω 2 A2 AN

(1.52)

Letting η D Ae i ω t in (1.47), we obtain (1.52) because T1 D t.

1.4 Dufﬁng–Rayleigh–van der Pol Equation

The Dufﬁng, Rayleigh, and van der Pol equations are special cases of uR C ω 2 u D µ uP C α 1 u3 C α 2 u2 uP C α 3 u uP 2 C α 4 uP 3

(1.53)

so that f D µ uP C α 1 u3 C α 2 u2 uP C α 3 u uP 2 C α 4 uP 3 and (1.14) becomes 3 2 i h ζP D i ωζ ζ ζN i µ ω ζ ζN C α 1 ζ C ζN C i ωα 2 ζ C ζN 2ω 2 3 i (1.54) ω 2 α 3 ζ C ζN ζ ζN i ω 3 α 4 ζ ζN Using the transformation (1.16), where h D O(), we rewrite (1.54) as ηP D i ωη C i ωh

@h @h P ηP ηN @η @ ηN

i N 2 (η η) N i µ ω (η η) N C i ωα 2 (η C η) 2ω N 3 ω 2 α 3 (η C η) N (η η) N 2 i ω 3 α 4 (η η) N 3 Cα 1 (η C η)

(1.55)

where terms of O( 2 ) and higher have been neglected. Again, because the perturbation contains linear as well as third-order terms, h has the form (1.44). Moreover, to the ﬁrst approximation, ηP and ηPN are given by

15

16

1 SDOF Autonomous Systems

(1.19). Hence, substituting (1.19) and (1.44) into (1.55) yields iµ ηN ηP D i ωη C 2iω ∆ 2 C 4ω i 3α 1 C i ωα 2 C ω 2 α 3 C 3i ω 3 α 4 η 2 ηN 2ω

3 1 1 2 3 α η C µ η C iω 2Λ 1 C i ωα ω α i ω α 1 2 3 4 2 2ω 2

1 2 3 3α η ηN 2 C iω 2Λ 3 i ωα C ω α 3i ω α 1 2 3 4 2ω 2

3 1 2 3 α ηN C iω 4Λ 4 i ωα ω α C i ω α (1.56) 1 2 3 4 2ω 2 We note that ∆ 1 and Λ 2 do not appear in (1.56) and hence they are arbitrary and the terms η and η 2 ηN are resonance terms. To produce the simplest form for (1.56), we choose ∆ 2 , Λ 1 , Λ 3 , and Λ 4 to eliminate the terms involving η, N η 3 , η ηN 2 , and ηN 3 ; that is, iµ 4ω 1 Λ 1 D 2 α 1 C i ωα 2 ω 2 α 3 i ω 3 α 4 4ω 1 3α 1 i ωα 2 C ω 2 α 3 3i ω 3 α 4 Λ3 D 2 4ω 1 α 1 i ωα 2 ω 2 α 3 C i ω 3 α 4 Λ4 D 8ω 2

∆2 D

(1.57) (1.58) (1.59) (1.60)

With these choices, (1.56) assumes the simple form 1 i 3α 1 C i ωα 2 C ω 2 α 3 C 3i ω 3 α 4 η 2 ηN ηP D i ωη C µ η 2 2ω

(1.61)

Again, we did not have to go through the lengthy algebra to arrive at the normal form (1.61). Because the solution of the unperturbed problem (1.54) is proportional to e i ω t , we could have replaced ζ with e i ω t in the perturbation and identiﬁed the terms proportional to e i ω t . In this case, they are i 1 3α 1 C i ωα 2 C ω 2 α 3 C 3i ω 3 α 4 ζ 2 ζN µ ζ 2 2ω Hence, keeping only the resonance terms in (1.54), we obtain the normal form 1 i ζP D i ωζ C µ ζ 3α 1 C i ωα 2 C ω 2 α 3 C 3i ω 3 α 4 ζ 2 ζN 2 2ω which is formally equivalent to (1.61).

1.5 An Oscillator with Quadratic and Cubic Nonlinearities

1.5 An Oscillator with Quadratic and Cubic Nonlinearities

We consider free oscillations of a single-degree-of-freedom system governed by uR C ω 2 u C δ u2 C α u3 D 0

(1.62)

To keep track of the different orders of magnitude, we use a nondimensional parameter that is the order of the amplitude of oscillations and hence rewrite (1.62) as uR C ω 2 u C δ u2 C 2 α u3 D 0

(1.63)

Thus, f D δ u2 2 α u3 and (1.14) becomes 2 3 iδ i 2 α ζP D i ωζ C ζ C ζN C ζ C ζN 2ω 2ω

(1.64)

In the next section, we use two successive transformations to produce the normal form of (1.64). In Section 1.5.3, we use a single transformation to produce the same normal form, and in Section 1.5.2, we use the method of multiple scales to determine a second-order expansion of (1.64). 1.5.1 Successive Transformations

To simplify the O() terms in (1.64), we introduce the near-identity transformation ζ D η C h 1 (η, η) N

(1.65)

and rewrite (1.64) as ηP D i ωη C iωh 1

2 @h 1 @h 1 P iδ ηP ηN C η C ηN C h 1 C hN 1 @η @ ηN 2ω

i 2 α (η C η) N 3 C 2ω The form of the O() terms suggests choosing h 1 in the form C

h 1 D Γ1 η 2 C Γ2 η ηN C Γ3 ηN 2

(1.66)

(1.67)

It follows from (1.66) that ηP D i ωη C O() and

ηPN D i ω ηN C O()

so that to O() (1.66) becomes δ δ 2 η η ηN C iω Γ C ηP D i ωη C iω Γ1 C 2 2ω 2 ω2 δ ηN 2 C O( 2 ) C iω 3Γ3 C 2ω 2

(1.68)

17

18

1 SDOF Autonomous Systems

The simplest possible form for (1.68) corresponds to the vanishing of the terms involving η 2 , η η, N and ηN 2 ; that is, choosing Γ1 , Γ2 , and Γ3 to be Γ1 D

δ , 2ω 2

Γ2 D

δ , ω2

Γ3 D

δ 6ω 2

(1.69)

Then, (1.68) reduces to ηP D i ωη C O( 2 )

(1.70)

PN we obtain Substituting (1.67) into (1.66) and using (1.70) to eliminate ηP and η, ηP D i ωη C

2δ 2 3 i 2 3 2 2 η C (1.71) C η N 5η η N 5η η N α (η C η) N 3C 2ω 3ω 2

Next, we introduce a near-identity transformation from η to ξ in the form η D ξ C 2 h 2 ξ , ξN

(1.72)

and obtain @h 2 P @h 2 NP ξ 2 ξ ξP D i ωξ C i 2 ωh 2 2 @ξ @ ξN

3 i 2 2δ 2 3 N 3 2N N2 C C ξ 5ξ ξ ξ C ξ 5ξ α ξ C ξN C 2ω 3ω 2

(1.73)

The form of the O( 2 ) terms suggests choosing h 2 in the form h 2 D Λ 1 ξ 3 C Λ 2 ξ 2 ξN C Λ 3 ξ ξN 2 C Λ 4 ξN 3

(1.74)

It follows from (1.73) that ξP D i ωξ C O( 2 )

and

ξPN D i ω ξN C O( 2 )

(1.75)

Therefore, substituting (1.74) and (1.75) into the right-hand side of (1.73) and keeping terms up to O( 2 ), we have α δ2 ξP D i ωξ C i 2 ω 2Λ 1 C ξ3 C 2ω 2 3ω 4 α δ2 i 2 10δ 2 2 3 N ξ C ξ 2 ξN C i ω 4Λ 4 C C 3α 2ω 2 3ω 4 2ω 3ω 2 3α 5δ 2 ξ ξN 2 C C i 2 ω 2Λ 3 C 2ω 2 3ω 4

(1.76)

We note that (1.76) is independent of Λ 2 and hence it is arbitrary and the term ξ 2 ξN is a resonance term. Equation 1.76 takes the simplest possible form if Λ1 D

α δ2 C , 4ω 2 6ω 4

Λ3 D

3α 5δ 2 C , 4ω 2 6ω 4

Λ4 D

α δ2 (1.77) 8ω 2 12ω 4

1.5 An Oscillator with Quadratic and Cubic Nonlinearities

Then, (1.76) becomes i 2 ξP D i ωξ C 2ω

3α

10δ 2 3ω 2

ξ 2 ξN C

(1.78)

Substituting (1.65) into (1.10), we have u D ζ C ζN D η C ηN C h 1 (η, η) N C hN 1 (η, η) N Then, substituting (1.72) into (1.79) yields u D ξ C ξN C h 1 ξ , ξN C hN 1 ξ , ξN C

(1.79)

(1.80)

Substituting for h 1 from (1.67) into (1.80) and using (1.69), we obtain δ 2 u D ξ C ξN C ξ 6ξ ξN C ξN 2 C 2 3ω

(1.81)

Substituting the polar form ξ D 12 ae i(ω tCβ) into (1.81), we ﬁnd that u D a cos (ωt C β) C

δ a 2 [cos (2ωt C 2β) 3] C 6ω 2

(1.82)

Substituting the polar form into (1.78) and separating real and imaginary parts, we have aP D 0

(1.83)

a βP D 2 a 3

3α 5δ 2 8ω 12ω 3

(1.84)

Equations 1.82–1.84 are in full agreement with those obtained by using the method of multiple scales, as shown in the next section. 1.5.2 The Method of Multiple Scales

Using the method of multiple scales, we seek a second-order uniform expansion of the solution of (1.64) in the form ζ(tI ) D

2 X

n ζ n (T0 , T1 , T2 ) C

(1.85)

nD0

where Tn D n t. In terms of these scales, the time derivative becomes d D D0 C D1 C 2 D2 C , dt

Dn D

@ @Tn

(1.86)

19

20

1 SDOF Autonomous Systems

Substituting (1.85) and (1.86) into (1.64) and equating coefﬁcients of like powers of , we obtain Order (0 )

D0 ζ0 i ωζ0 D 0

(1.87)

Order ()

D0 ζ1 i ωζ1 D D1 ζ0 C

2 iδ ζ0 C ζN 0 2ω

(1.88)

Order (2 )

D0 ζ2 i ωζ2 D D2 ζ0 D1 ζ1 C

3 iα iδ ζ0 C ζN 0 ζ1 C ζN 1 C ζ0 C ζN 0 ω 2ω (1.89)

The general solution of (1.87) can be expressed as ζ0 D A (T1 , T2 ) e i ω T0

(1.90)

where A is an undetermined function of T1 and T2 at this order; it is determined by eliminating the secular terms at the next orders of approximation. Substituting (1.90) into (1.88) yields D0 ζ1 i ωζ1 D D1 Ae i ω T0 C

i δ 2 2i ω T0 A e C 2A AN C AN2 e 2i ω T0 2ω

(1.91)

Eliminating the terms that produce secular terms in (1.91) demands that D1 A D 0 or A D A(T2 ). Then, the solution of (1.91) can be expressed as ζ1 D

δ A2 2i ω T0 δ A AN δ AN2 2i ω T0 e e 2ω 2 ω2 6ω 2

(1.92)

where the solution of the homogeneous equation has not been included so that the amplitude of the term at the frequency of oscillation is uniquely deﬁned by the zeroth-order solution (1.90). We note that the coefﬁcients in (1.92) are the same as the Γi deﬁned in (1.69). Substituting (1.90) and (1.92) into (1.89) and using the fact that D1 A D 0, we have 10δ 2 i N i ω T0 C NST (1.93) 3α D0 ζ2 i ωζ2 D D2 Ae i ω T0 C A2 Ae 2ω 3ω 2 where NST stands for the terms that do not produce secular terms. Eliminating the terms that produce secular terms from (1.93), we obtain 10δ 2 i 3α (1.94) A2 AN D2 A D 2ω 3ω 2

1.5 An Oscillator with Quadratic and Cubic Nonlinearities

Putting ξ D Ae i ω t in (1.78) and using the fact that D2 A D 2 d A/d t, we obtain exactly (1.94). We note that, for a uniform second approximation, we do not need to solve for ζ2 , but we only need to inspect (1.93) and eliminate the terms that produce secular terms. Similarly, to determine a uniform second approximation by using the method of normal forms, we do not need to determine h 2 in (1.72), but we need only keep the resonance terms in (1.73). 1.5.3 A Single Transformation

Instead of using successive transformations to produce the normal form of (1.64), one can formulate the process as a perturbation method. Thus, we expand ζ in a power series of in terms of a new variable η in the form ζ D η C h 1 (η, η) N C 2 h 2 (η, η) N C

(1.95)

ηP D i ωη C g 1 (η, η) N C 2 g 2 (η, η) N C

(1.96)

where h 1 and h 2 are smooth functions of η and ηN and g 1 and g 2 contain all of the resonance and near-resonance terms. Substituting (1.95) and (1.96) into (1.64) and equating coefﬁcients of like powers of , we obtain iδ @h 1 @h 1 (η C η) g1 C i ω η (1.97) ηN h1 D N 2 @η @ ηN 2ω @h 1 @h 1 @h 2 @h 2 iα (η C η) g2 C i ω η ηN h 2 D g 1 gN 1 C N 3 @η @ ηN @η @ ηN 2ω iδ (1.98) (η C η) C N h 1 C hN 1 ω Equations 1.97 and 1.98 are the so-called homology equations for h 1 and h 2 . Next, we need to determine g 1 and h 1 from (1.97). In order that h 1 be nonsingular (smooth), we choose g 1 to eliminate all of the resonance and near-resonance terms; otherwise, h 1 will be singular (i.e., have secular terms) if there are resonance terms and near singular (i.e., have small divisors) if there are near-resonance terms. In the present case, there are no resonance terms in (1.97). To see this, we note from (1.96) that η D B e i ω t and hence the perturbation terms on the right-hand side of (1.97) contain terms proportional to e ˙2i ω t and a constant. Because none of these terms is proportional to e i ω t , which is the solution of the ﬁrst-order problem in (1.96), there are no resonance terms. If we are in doubt, we seek a function h 1 that can be used to eliminate all of the perturbation terms. If we are successful in ﬁnding a smooth function h 1 that eliminates all of the perturbation terms, then g 1 D 0. Otherwise, we choose g 1 to eliminate all terms that produced the troublesome terms (i.e., singular and near-singular terms) in h 1 . Because the perturbation terms are of second degree, we let h 1 D Γ1 η 2 C Γ2 η ηN C Γ3 ηN 2

(1.99)

21

22

1 SDOF Autonomous Systems

choose Γ1 , Γ2 , and Γ3 to eliminate i δ(η C η) N 2 /2ω and obtain (1.69). Because the obtained Γm are regular, there are no resonance terms and g 1 D 0. Substituting (1.99) and g 1 D 0 into (1.98) and using (1.69), we obtain i δ2 @h 2 @h 2 (η C η) N η 2 C ηN 2 6η ηN g2 C i ω η ηN h2 D 3 @η @ ηN 3ω iα (η C η) N 3 C 2ω

(1.100)

As stated earlier, we do not need to determine h 2 and all that we need is to inspect (1.100) and choose g 2 to eliminate all of the resonance and near-resonance terms. Again, because η / e i ω t , only the term proportional to η 2 ηN is a resonance term. Hence, choosing g 2 to eliminate this term, we obtain i g2 D 2ω

10δ 2 3α 3ω 2

η 2 ηN

(1.101)

Substituting (1.95) and (1.99) into (1.10) and using (1.69), we obtain u D η C ηN C

δ 2 η C ηN 2 6η ηN C 2 3ω

(1.102)

Substituting for g 2 from (1.101) into (1.96) and using the fact that g 1 D 0, we have ηP D i ωη C

i 2 2ω

10δ 2 3α η 2 ηN C 3ω 2

(1.103)

in agreement through second order with the expansions obtained in Sections 1.5.1 and 1.5.2.

1.6 A General System with Quadratic and Cubic Nonlinearities

We consider free oscillations of a single-degree-of-freedom system governed by uR C ω 2 u C δ 1 u2 C δ 2 uP 2 C 2 2µ uP C α 1 u3 C α 2 u2 uP C α 3 u uP 2 C α 4 uP 3 D 0

(1.104)

so that f D δ 1 u2 C δ 2 uP 2 2 2µ uP C α 1 u3 C α 2 u2 uP C α 3 u uP 2 C α 4 uP 3 (1.105) We note that a small nondimensional parameter has been introduced as a bookkeeping device. The quadratic terms have been ordered as O(), whereas the cubic terms and the linear damping term have been ordered as O( 2 ). Then, (1.14) be-

1.6 A General System with Quadratic and Cubic Nonlinearities

comes

2 2 i i h ζP D i ωζ C 2 µ ζ ζN δ 1 ζ C ζN δ 2 ω 2 ζ ζN 2ω 3 2 i 2 h ζ ζN α 1 ζ C ζN C i ωα 2 ζ C ζN C 2ω 2 3 i ω 2 α 3 ζ C ζN ζ ζN i ω 3 α 4 ζ ζN

(1.106)

As in Section 1.5.3, we seek an expansion for (1.106) in the form (1.95) and (1.96), equate coefﬁcients of like powers of , and obtain i @h 1 @h 1 N 2 δ 2 ω 2 (η η) N 2 ηN h1 D δ 1 (η C η) g1 C i ω η @η @ ηN 2ω (1.107) @h 1 @h 1 @h 2 @h 2 g2 C i ω η ηN h 2 D g 1 gN 1 µ (η η) N @η @ ηN @η @ ηN i i i h N h 1 C hN 1 δ 2 ω 2 (η η) N h 1 hN 1 C N 3 δ 1 (η C η) α 1 (η C η) C ω 2ω i i ωα 2 (η C η) N 2 (η η) N i ω 3 α 4 (η η) N 3 ω 2 α 3 (η C η) N (η η) N 2 C 2ω (1.108) The right-hand side of (1.107) does not contain resonance or near-resonance terms, and hence we put g 1 D 0, seek h 1 in the form (1.99), and obtain δ1 1 δ1 δ1 1 δ 2 , Γ2 D 2 δ 2 , Γ3 D 2 C δ 2 (1.109) 2ω 2 2 ω 6ω 6 Because all of the Γm are regular, our conclusion that there are no resonance or near-resonance terms in (1.107) and hence g 1 D 0 is justiﬁed a posteriori. Substituting (1.99) and (1.109) into (1.108) and using the fact that g 1 D 0, we obtain 2 @h 2 δ1 @h 2 (η η) N η 2 ηN 2 g2 C i ω η δ ηN h2 D i δ2 ω 2 2 @η @ ηN 3 ω

2 δ1 i δ1 δ1 2 (η (η µ η) N C δ 2 η C ηN 6 C δ 2 η ηN C η) N 3ω ω2 ω2 i α 1 (η C η) N 3 C i ωα 2 (η C η) N 2 (η η) N ω 2 α 3 (η C η) N (η η) N 2 C 2ω (1.110) N 3 C i ω 3 α 4 (η η) Inspecting the right-hand side of (1.110), we conclude that the terms proportional to η and η 2 ηN are the only resonance terms and there are no near-resonance terms. Consequently, choosing g 2 to eliminate the resonance terms, we obtain i 2 5δ 21 2 2 3α 1 C ω 2 α 3 g 2 D µ η C C 5δ δ C 2δ ω 1 2 2 2ω 3 ω2

Ci ω α 2 C 3ω 2 α 4 η 2 ηN (1.111) Γ1 D

23

24

1 SDOF Autonomous Systems

Substituting (1.95) and (1.99) into (1.10) and using (1.109), we obtain

2 2δ 1 1 δ1 2 η η ηN C (1.112) C η N C 2δ δ u D ηC ηC N 2 2 3ω 2 3 ω2 Substituting for g 2 from (1.111) into (1.96) and using the fact that g 1 D 0, we obtain i 2 2 5δ 21 2 2 2 3α 1 C ω α 3 C 5δ 1 δ 2 C 2δ 2 ω ηP D i ωη µ η C 2ω 3 ω2

Ci ω α 2 C 3ω 2 α 4 η 2 ηN (1.113) 2

To compare the expansion (1.112) and (1.113) with that obtained by using the method of multiple scales (Nayfeh, 1984), we substitute the polar form η D 12 ae i(ω tCβ)

(1.114)

in (1.112) and (1.113) and obtain 1 u D a cos (ωt C β) C a 2 6

δ1 δ1 (2ωt cos C 2β) 3 δ C δ 2 2 ω2 ω2

C

(1.115)

where aP D 2 µ a 18 2 α 2 C 3ω 2 α 4 a 3 C a βP D

2 5δ 1 2 2 2 3α 1 C ω 2 α 3 a3 C C 5δ δ C 2δ ω 1 2 2 8ω 3 ω2

(1.116) (1.117)

which is formally equivalent to that obtained by using the method of multiple scales.

1.7 The van der Pol Oscillator

In this section, we construct a second-order approximation of the normal form of the van der Pol oscillator uR C ω 2 u D (1 u2 ) uP

(1.118)

Using the transformation (1.10), we rewrite (1.118) as N 1 (ζ C ζ) N 2 ζP D i ωζ C 12 (ζ ζ)

(1.119)

1.7 The van der Pol Oscillator

1.7.1 The Method of Normal Forms

As in the preceding two sections, we seek a second-order expansion of (1.119) in the form (1.95) and (1.96), equate coefﬁcients of like powers of , and obtain 1 @h 1 @h 1 1 3 η C η 2 ηN ηN 2 η ηN 3 ηN h 1 D (η η) N g1 C i ω η @η @ ηN 2 2 (1.120) @h 1 @h 1 @h 2 @h 2 1 ηN h 2 D g 1 gN 1 C h 1 hN 1 g2 C i ω η @η @ ηN @η @ ηN 2 3 2 1 1 3 2 2 2 η h 1 η η(h N 1 hN 1 ) η hN 1 C ηN h 1 C ηN hN 1 (1.121) 2 2 2 2 Choosing g 1 to eliminate the resonance terms in (1.120), we have g 1 D 12 η 12 η 2 ηN

(1.122)

Then, we seek h 1 in the form h 1 D ∆ 1 ηN C Λ 1 η 3 C Λ 2 η ηN 2 C Λ 3 ηN 3

(1.123)

Substituting (1.123) and (1.122) into (1.120) yields 2i ω∆ 1 12 ηN 2i ωΛ 1 C 12 η 3 C 2i ωΛ 2 C 12 η ηN 2 C 4 i ωΛ 3 C 12 ηN 3 D 0

(1.124)

Hence, ∆1 D Therefore, h1 D

i , 4ω

Λ1 D

i , 4ω

Λ2 D

i , 4ω

1 1 i ηN η 3 η ηN 2 ηN 3 4ω 2

Λ3 D

i 8ω

(1.125)

(1.126)

Substituting (1.122) and (1.126) into (1.121) yields i @h 2 @h 2 g2 C i ω η ηN h2 D 2η C 5η 3 C 5η 5 12η 2 ηN @η @ ηN 16ω (1.127) 2η 4 ηN 4η ηN 2 C 11η 3 ηN 2 C 2 ηN 3 C 5η 2 ηN 3 C η ηN 4 C 5 ηN 5 Choosing g 2 to eliminate the resonance terms (terms proportional to η, η 2 η, N and η 3 ηN 2 ) from (1.127), we have 1 (1.128) i 2η 12η 2 ηN C 11η 3 ηN 2 16ω Substituting (1.122) and (1.128) into (1.96), we obtain, to the second approximation, the normal form 1 1 ηP D i ωη C η η 2 ηN (1.129) i 2 2η 12η 2 ηN C 11η 3 ηN 2 2 16ω g2 D

25

26

1 SDOF Autonomous Systems

Substituting the polar form (1.27) into (1.129) and separating real and imaginary parts, we obtain aP D 12 a 14 a 3 1 2 3 11 4 βP D 1 a2 C a 8ω 2 32

(1.130) (1.131)

in agreement with those obtained with the generalized method of averaging (Nayfeh, 1973). 1.7.2 The Method of Multiple Scales

We seek a second-order expansion of the solution of (1.119) in the form (1.85). Substituting (1.85) and (1.86) into (1.119) and equating coefﬁcients of like powers of , we obtain Order (0 )

D0 ζ0 i ωζ0 D 0

(1.132)

Order ()

D0 ζ1 i ωζ1 D D1 ζ0 C

1 2

ζ0 ζN 0 12 ζ03 C ζ02 ζN 0 ζ0 ζN 02 ζN 03 (1.133)

Order (2 )

D0 ζ2 i ωζ2 D D2 ζ0 D1 ζ1 C 12 ζ1 ζN 1 12 3ζ02 C 2ζ0 ζN 0 ζN 02 ζ1 1 ζ 2 2ζ0 ζN 0 3 ζN 2 ζN 1 2

0

0

(1.134) The general solution of (1.132) can be expressed as in (1.90). Then, (1.133) becomes N i ω T0 1 A3 e 3i ω T0 D0 ζ1 i ωζ1 D D1 Ae i ω T0 C 12 Ae i ω T0 12 Ae 2 N i ω T0 C 1 A AN2 e i ω T0 C 1 AN3 e 3i ω T0 12 A2 Ae 2 2

(1.135)

Eliminating the terms that lead to secular terms from (1.135) yields D1 A D

1 2

A A2 AN

(1.136)

Then, the solution of (1.135) can be expressed as ζ1 D

i N i ω T0 C 2A3 e 3i ω T0 C 2A AN2 e i ω T0 C AN3 e 3i ω T0 2 Ae 8ω

(1.137)

1.8 Exercises

Substituting (1.90), (1.136), and (1.137) into (1.134), we have D0 ζ2 i ωζ2 D D2 Ae i ω T0

i 2A 12A2 AN C 11A3 AN2 e i ω T0 16ω

C NST

(1.138)

Eliminating the terms that lead to secular terms from (1.138) yields D2 A D

i 2A 12A2 AN C 11A3 AN2 16ω

(1.139)

Using the method of reconstitution, we obtain from (1.136) and (1.139) that 1 i 2 AP D A A2 AN 2A 12A2 AN C 11A3 AN2 2 16ω

(1.140)

Letting ζ D Ae i ω t in (1.129), we obtain (1.140), which means that the results obtained with the methods of normal forms and multiple scales are the same.

1.8 Exercises

1.8.1 Use the methods of normal forms and multiple scales to determine the normal forms of a) b) c) d) e)

uR C ω 2 u C α uP 3 D 0 , uR C ω 2 u C α u2 uP D 0 , uR C ω 2 u C α u5 D 0 , uR C ω 2 u C α u3 uP 2 D 0 , uR C ω 2 u C uP 5 D 0 .

1.8.2 Use the methods of multiple scales and normal forms to construct a secondorder approximation to the normal form of uR C ω 2 u C α u3 C 2µ u2 uP D 0 1.8.3 Use the methods of multiple scales and normal forms to construct a ﬁrstorder approximation to the normal form of uR C ω 2 u C α u2 uR D 0 1.8.4 Use the methods of normal forms and multiple scales to construct a secondorder approximation to the normal form of uR C ω 2 u C α uP 13 uP 3 D 0

27

28

1 SDOF Autonomous Systems

1.8.5 Consider the equation xR C x C 3x 2 C 2x 3 D 0 Determine the equilibrium points. Determine, to second order, the normal form of the system near each of these equilibrium points. 1.8.6 Consider xR C x C ax 2 C 2x 3 D 0 p Show that there is only one equilibrium p point when a < 2 2 and that there are three equilibrium points when a > 2 2. Determine, to second order, the normal forms near the equilibrium points. 1.8.7 Consider xR C x

a D0 1x

Show that there is only one equilibrium point when a 1/4. Determine, to second order, the normal form near this equilibrium point. 1.8.8 Consider xR 3x C x 3 D 2 Determine the equilibrium points and the normal forms near them. 1.8.9 Consider uR u C u4 D 0 Determine the equilibrium points and the normal forms near them. 1.8.10 Consider xR 2x x 2 C x 3 D 0 Determine the equilibrium points and the normal forms near them. 1.8.11 Consider uR C u

3 D0 16(1 u)

Determine the equilibrium points and the normal forms near them. 1.8.12 Use the methods of multiple scales and normal forms to determine a ﬁrstorder uniform expansion for the general solution of θR C ω 2 sin θ C for small but ﬁnite θ .

4 sin2 θ θP D 0 1 C 4(1 cos θ )

1.8 Exercises

1.8.13 Consider the equation uR C ω 20 u C

µ uP D0 1 u2

Use the methods of multiple scales and normal forms to determine a ﬁrst-order uniform expansion for small u. 1.8.14 Consider the equation

2 l 2 C r 2 2r l cos θ θR C r l sin θ θP C g l sin θ D 0

where g, r, and l are constants. Determine a ﬁrst-order expansion for small but ﬁnite θ by using the methods of multiple scales and normal forms. 1.8.15 Consider the equation 1 12

2 l 2 C r 2 θ 2 θR C r 2 θ θP C g r θ cos θ D 0

where r, l, and g are constants. Determine a ﬁrst-order uniform expansion for small but ﬁnite θ by using the methods of multiple scales and normal forms. 1.8.16 The motion of a simple pendulum is governed by g θR C sin θ D 0 l Use the methods of multiple scales and normal forms to determine a ﬁrst-order uniform expansion for small but ﬁnite θ . 1.8.17 Consider the equation g θR D Ω 2 sin θ cos θ sin θ R Use the methods of multiple scales and normal forms to determine a ﬁrst-order uniform expansion for small but ﬁnite θ . 1.8.18 The motion of a particle on a rotating parabola is governed by 1 C 4p 2 x 2 xR C Λ x C 4p 2 xP 2 D 0 where p and Λ are constants. Use the methods of multiple scales and normal forms to determine a ﬁrst-order expansion for small but ﬁnite x. 1.8.19 Consider the equation u uP 2 u2 g u uR C 1C C ω 20 u C p D0 2 1u (1 u2 )2 l 1 u2 Use the methods of multiple scales and normal forms to determine a ﬁrst-order expansion for small u.

29

31

2 Systems of First-Order Equations 2.1 Introduction

In this chapter, we describe in detail the problem of calculating the normal form for a system of ﬁrst-order equations having the form uP D Au C F 1 (u) C 2 F 2 (u) C

(2.1)

where u and the F m are column vectors of length n, A is an n n constant matrix, and is a small nondimensional parameter, which may represent a physical quantity, or it may be just a bookkeeping device and set equal to unity in the ﬁnal result. In this book, the normalization is carried out in terms of , whereas in the literature the normalization is usually carried out in terms of the degree of the polynomials in the nonlinear terms. In the latter case, the components of F m (u) are homogeneous polynomials of degree m C 1 in u. We assume that the F m are smooth vector ﬁelds satisfying F m (0) D 0 so that u D 0 is a ﬁxed point of (2.1). As a ﬁrst step, we introduce a linear transformation u D P x, where P is a nonsingular or invertible matrix, in (2.1) and obtain P xP D AP x C F 1 (P x) C 2 F 2 (P x) C

(2.2)

Multiplying (2.2) from the left by P 1 , the inverse of P, we obtain xP D J x C f 1 (x) C 2 f 2 (x) C

(2.3)

J D P 1 AP

(2.4)

where and

f m (x) D P 1 F m (P x)

We choose P so that J has a simple real form. The simplest possible real form for J depends on the eigenvalues λ i and eigenvectors p i of A. There are three possibilities (Coddington and Levinson, 1955; Hale, 1980; Hirsch and Smale, 1974; Walter, 1976): The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

32

2 Systems of First-Order Equations

a) All eigenvalues are real and distinct. b) Some eigenvalues occur in complex conjugate pairs. c) Some eigenvalues are repeated. In the ﬁrst case, there are n linearly independent real eigenvectors p i of A. Then, forming the matrix P by having its columns be the eigenvectors p 1 , p 2 , . . . , p n produces a diagonal matrix J with real entries λ 1 , λ 2 , . . . , λ n . For two-dimensional systems, λ1 0 JD 0 λ2 When the eigenvalues are real and some of them are repeated, there are two possibilities. First, there are n linearly independent real eigenvectors p i . Then, choosing P as in the ﬁrst case produces a diagonal matrix with real entries λ 1 , λ 2 , . . . , λ n , where some of them are repeated. For two-dimensional systems, λ 0 JD 0 λ Second, there are less than n linearly independent eigenvectors. In this case, we form the matrix P by having its columns p i be the generalized eigenvectors of A; that is, the eigenvectors of (A λ j I ) m , where m is the multiplicity of the eigenvalue λ j . With this choice for P, J takes the so-called Jordan form. For twodimensional systems, λ 1 λ 0 JD or J D 0 λ 1 λ When some of the eigenvalues occur in complex conjugate pairs, some of the eigenvectors are complex. Forming the matrix P by having its columns be the generalized real eigenvectors of A produces a matrix J that is in the so-called real Jordan form. For two-dimensional systems, α ω α ω JD or J D ω α ω α where α and ω are real constants. The ﬁxed point u D 0 is called hyperbolic if none of the eigenvalues of A has a zero real part and nonhyperbolic if at least one of the eigenvalues of A has a zero real part. The zero real parts imply the existence of constraints on the elements of A. For two-dimensional systems, there are four possible constraints: 1. A has real eigenvalues and one of them is zero; that is, jAj D 0 and Tr(A) ¤ 0 where Tr(A) is the trace of A.

2.1 Introduction

2. A has purely imaginary eigenvalues; that is, jAj ¤ 0

and Tr(A) D 0

3. Both eigenvalues of A are zero but A is not the null matrix; that is, A¤0,

jAj D 0 ,

and Tr(A) D 0

4. A D 0. The equality constraints are called degeneracy conditions. The number of degeneracy conditions indicates the level of degeneracy or codimension of the ﬁxed point (Dumortier, 1977; Guckenheimer and Holmes, 1983). Our aim is to construct a sequence of transformations that successively remove the perturbation terms f m , starting from f 1 . Ideally, we would like to be able to remove all of the f m , especially, if they are nonlinear, thereby reducing (2.3) to a linear problem. In general, this is not possible, as discussed below. Instead of using a sequence of transformations, we introduce the single nearidentity transformation x D y C h 1 (y) C 2 h 2 (y) C

(2.5)

into (2.3) and choose the h m so that it takes the simplest possible form, the so-called normal form yP D J y C g 1 (y) C 2 g 2 (y) C

(2.6)

As discussed later, the g n are referred to as resonance and near-resonance terms. Substituting (2.5) into (2.3) yields yP C D h 1 (y) yP C 2 D h 2 (y) yP C

D J y C J h 1 (y) C 2 J h 2 (y) C C f 1 y C h 1 (y) C 2 h 2 (y) C C 2 f 2 y C h 1 (y) C 2 h 2 (y) C C (2.7)

where D h is the Jacobian of h. Using (2.6) to eliminate yP from (2.7) and equating coefﬁcients of like powers of , we obtain g 1 (y) C D h 1 (y) J y J h 1 (y) D f 1 (y)

(2.8)

g 2 (y) C D h 2 (y) J y J h 2 (y) D f 2 (y) C D f 1 (y)h 1 (y) D h 1 (y)g 1 (y) (2.9) The operator L(h 1 (y)) D D h 1 (y) J y J h 1 (y) D [h 1 , J y] is called the Lie or Poisson bracket. We note that although the h n (y), which describe the transformation (2.5), are in general nonlinear functions of y, they are found by solving a sequence of linear problems, as described below. Next, we carry out the details of determining the h m and g m for speciﬁc problems.

33

34

2 Systems of First-Order Equations

2.2 A Two-Dimensional System with Diagonal Linear Part

As a ﬁrst example, we consider the two-dimensional system xP D

α x 2 C α 2 x1 x2 C α 3 x22 0 x C 1 12 λ2 α 4 x1 C α 5 x1 x2 C α 6 x22

λ1 0

(2.10)

where the λ i and α i may be real or complex, but both of λ 1 and λ 2 are not zero. Using (2.10), we rewrite (2.8) as 2

g 11 g 12

D

@h 11 6 @y 1 6 C6 4 @h 12 @y 1

α 1 y 12 α 4 y 12

3 @h 11 @y 2 7 7 λ1 7 @h 12 5 0 @y 2

C α 2 y 1 y 2 C α 3 y 22 C α 5 y 1 y 2 C α 6 y 22

0 λ2

y1 λ 1 0 y2

0 λ2

h 11 h 12

(2.11)

where (h 11 , h 12 ) and (g 11 , g 12 ) are the components of h 1 and g 1 . The right-hand side of (2.11) suggests seeking the g 1m and h 1m in the form h 11 D Γ1 y 12 C Γ2 y 1 y 2 C Γ3 y 22

(2.12)

h 12 D Γ4 y 12 C Γ5 y 1 y 2 C Γ6 y 22

(2.13)

g 11 D Λ 1 y 12 C Λ 2 y 1 y 2 C Λ 3 y 22

(2.14)

g 12 D Λ 4 y 12 C Λ 5 y 1 y 2 C Λ 6 y 22

(2.15)

Substituting (2.12)–(2.15) into (2.11) yields λ 1 y 1 (2Γ1 y 1 C Γ2 y 2 ) C λ 2 y 2 (Γ2 y 1 C 2Γ3 y 2 ) λ 1 Γ1 y 12 C Γ2 y 1 y 2 C Γ3 y 22 D (α 1 Λ 1 ) y 12 C (α 2 Λ 2 ) y 1 y 2 C (α 3 Λ 3 ) y 22

(2.16)

λ 1 y 1 (2Γ4 y 1 C Γ5 y 2 ) C λ 2 y 2 (Γ5 y 1 C 2Γ6 y 2 ) λ 2 Γ4 y 12 C Γ5 y 1 y 2 C Γ6 y 22 D (α 4 Λ 4 ) y 12 C (α 5 Λ 5 ) y 1 y 2 C (α 6 Λ 6 ) y 22

(2.17)

Equating the coefﬁcients of y 12 , y 1 y 2 , and y 22 on both sides of (2.16) and (2.17) yields BΓ D α Λ

(2.18)

2.2 A Two-Dimensional System with Diagonal Linear Part

where Γ , α, and Λ are column vectors having the components Γm , α m , and Λ m , respectively, and 2 3 0 0 0 0 λ1 0 60 λ 0 0 0 07 2 6 7 6 7 0 2λ 2 λ 1 0 0 07 60 BD6 (2.19) 7 60 0 0 2λ 1 λ 2 0 07 6 7 40 0 0 0 λ1 0 5 0 0 0 0 0 λ2 Thus, the matrix B is diagonal having the eigenvalues λ 1 , λ 2 , 2λ 2 λ 1 , and 2λ 1 λ 2 . As shown in Section 2.4, these eigenvalues are related to the eigenvalues λ 1 and λ 2 of J and hence A by λ m,i D m 1 λ 1 C m 2 λ 2 λ i

(2.20)

where m D (m 1 , m 2 ), m 1 and m 2 are positive integers such that m 1 C m 2 D 2 and m 1 , m 2 D 0, 1, and 2. Clearly, the matrix B has an inverse and hence (2.18) has a solution for any α if none of its eigenvalues vanishes. Consequently, we put Λ D 0, explicitly solve (2.18) for the Γm , and obtain α1 α2 α3 , Γ2 D , Γ3 D , λ1 λ2 2λ 2 λ 1 α4 α5 α6 Γ4 D , Γ5 D , Γ6 D 2λ 1 λ 2 λ1 λ2 Γ1 D

(2.21)

With this choice, all of the nonlinear terms are eliminated from (2.10), resulting in the linear normal form yP D J y. When any of the eigenvalues λ 1 , λ 2 , 2λ 2 λ 1 , and 2λ 1 λ 2 vanishes (i.e., m 1 λ 1 C m 2 λ 2 D λ i , for some choice of m 1 and m 2 ), the matrix B is singular and one or more of the Γm are unbounded, implying that one cannot eliminate all of the nonlinear terms in (2.10) and hence obtain a linear normal form. The terms that cannot be eliminated are called resonance terms. Moreover, the condition m 1 λ 1 Cm 2 λ 2 D λ i is called a resonance condition of order 2 because the nonlinearity is quadratic (i.e., m 1 C m 2 D 2). When any of the eigenvalues of B is small, it follows from (2.21) that at least one of the Γm has a small divisor and therefore the transformation breaks down. The terms that produce small divisors are called near-resonance terms. For example, let us consider the resonance condition 2λ 2 λ 1 0. In this case, it follows from (2.21) that all Γm except Γ3 are regular and that Γ3 is either singular or has a small divisor. Thus, we choose Λ 3 D α 3 to avoid this singularity, put the remaining Λ m equal to zero, and choose Γ1 , Γ2 , Γ4 , Γ5 , and Γ6 as in (2.21). Consequently, the normal form of (2.10) is λ α y2 0 yP D 1 y C 3 2 (2.22) 0 λ2 0 When 2λ 1 λ 2 0, we choose Λ 4 D α 4 to avoid the singularity, set the remaining Λ m equal to zero, and choose Γ1 , Γ2 , Γ3 , Γ5 , and Γ6 as in (2.21). In this case, the

35

36

2 Systems of First-Order Equations

normal form of (2.10) is yP D

λ1 0

0 0 y C α 4 y 12 λ2

(2.23)

When λ 1 0, we choose Λ 1 D α 1 and Λ 5 D α 5 to avoid the singularities, set the remaining Λ m equal to zero, and choose Γ2 , Γ3 , Γ4 , and Γ6 as in (2.21). The resulting normal form of (2.10) is yP D

λ1 0

α 1 y 12 0 y C λ2 α5 y1 y2

(2.24)

When λ 2 0, we choose Λ 2 D α 2 and Λ 6 D α 6 , set the remaining Λ m equal to zero, and choose Γ1 , Γ3 , Γ4 , and Γ5 as in (2.21). Consequently, the normal form of (2.10) becomes yP D

λ1 0

α y y 0 y C 2 1 22 λ2 α6 y2

(2.25)

Recapping, we note that the 6 6 matrix B maps six-dimensional vectors in the space 0 and for µ > 0 if f µ f x x < 0. Then, it follows from (3.80) that the upper branch is stable and the lower branch is unstable if f x x < 0 and that the upper branch is unstable and the lower branch is stable if f x x > 0. This bifurcation of the nonhyperbolic ﬁxed point at the origin as µ passes through zero is called fold or tangent or saddle-node bifurcation. Example 3.9 We consider the one-dimensional map x kC1 D x k C µ x k2

(3.81)

where µ is a scalar control parameter. When µ > 0, it is clear from Figure 3.1a that this map intersects the identity map x kC1 D x k in two points. In other words, for µ > 0, (3.81) has the nontrivial ﬁxed points x1 D

p

µ

and

p x2 D µ

The Jacobian matrix associated with the ﬁxed point x j 0 has the single eigenvalue D 1 2x j The ﬁxed point x2 is an unstable node for all µ > 0 because jj > 1. On the other hand, the ﬁxed point x1 is a stable node for 0 < µ < 1 because jj < 1. When µ is decreased to zero, the two ﬁxed points approach each other and coalesce at the single ﬁxed point x D 0, as shown in Figure 3.1b. The map (3.81) is tangent to the identity map and, hence, the associated bifurcation is called tangent bifurcation. When µ < 0, the map (3.81) does not intersect the identity map, as shown in Figure 3.1c, and hence for µ < 0, (3.81) does not have any ﬁxed points. In Figure 3.2, we show the different ﬁxed points of (3.81) and their stability in the vicinity of the origin of the x µ space. Broken and solid lines are used to represent branches

77

78

3 Maps

0 f(x) –0.5

(a)

–1 –0.5

0 x 0.5

0.5

–0.5

0 x

(b)

0.5

0 f(x) –0.5

(c)

–1 –0.5

0 x

0.5

Figure 3.1 The functions f (x) D µ C x x 2 and x for (a) µ D 0.1, (b) µ D 0.0, and (c) µ D 0.1.

x 0

0 μ Figure 3.2 Scenario in the vicinity of a saddle-node bifurcation.

of unstable and stable ﬁxed points, respectively. A saddle-node or tangent bifurcation occurs at (x, µ) D (0, 0). Diagrams such as Figure 3.2 in which the variation of solutions and their stability are displayed in the state-control space are called bifurcation diagrams. In the bifurcation diagram, a branch of stable solutions is called a stable branch and a branch of unstable solutions is called an unstable branch. In most situations, a branch of solutions either ends or begins at a bifurcation point.

3.4 Local Bifurcations

3.4.2 Transcritical Bifurcation

When f µ D 0 and f x x D 6 0, the limit of (3.77) as x ! 0 and µ ! 0 is x kC1 D x k C 12 f µ µ µ 2 C 2 f x µ x k µ C f x x x k2

(3.82)

whose ﬁxed points are q q f x µ C f x2µ f µ µ f x x f x µ f x2µ f µ µ f x x x1 D µ and x2 D µ fxx fxx (3.83) provided that f x2µ f µ µ f x x > 0. It follows from (3.83) that there are two curves of ﬁxed points in the neighborhood of (x, µ) D (0, 0), which intersect transversely at (0, 0). The multipliers associated with these ﬁxed points are q q (x1 ) D 1 C f x2µ f µ µ f x x µ and (x2 ) D 1 f x2µ f µ µ f x x µ (3.84) Therefore, x1 is stable when µ < 0 and unstable when µ > 0. On the other hand, x2 is unstable when µ < 0 and stable when µ > 0. In this case, we have two branches of ﬁxed points that interchange stability at (0, 0). The bifurcation of the nonhyperbolic ﬁxed point at the origin as µ passes through zero is called transcritical bifurcation. Example 3.10 We consider the one-dimensional map x kC1 D x k C µ x k x k2

(3.85)

where µ is again a scalar control parameter. This map has the two ﬁxed points x1 D 0W

trivial ﬁxed point

x2

nontrivial ﬁxed point .

D µW

For the ﬁxed point x j , the Jacobian matrix has the single eigenvalue D 1 C µ 2x j Hence, it follows that the trivial ﬁxed point is stable for 2 < µ < 0 and unstable for all µ > 0. On the other hand, the nontrivial ﬁxed point is unstable for all µ < 0 and stable for 0 < µ < 2. The scenario in the vicinity of (x, µ) D (0, 0) is illustrated in Figure 3.3. A transcritical bifurcation occurs at the origin.

79

80

3 Maps

x 0

0 μ Figure 3.3 Scenario in the vicinity of a transcritical bifurcation.

3.4.3 Pitchfork Bifurcation

When f µ D 0, f x x D 0, f x µ ¤ 0, and f x x x ¤ 0, the limit of (3.77) as x ! 0 and µ ! 0 is x kC1 D x k C f x µ x k µ C

1 6

f x x x x k3

(3.86)

whose ﬁxed points are s x1

D0,

x2

D

6 f x µ µ , fxxx

s and

x3

D

6 f x µ µ fxxx

(3.87)

It follows from (3.87) that there are three branches of solutions. The ﬁrst branch exists for all values of µ and the other two branches exist for µ > 0 if f x µ f x x x < 0 and for µ < 0 if f x µ f x x x > 0. The multipliers associated with these ﬁxed points are (x1 ) D 1 C f x µ µ ,

(x2 ) D 1 2 f x µ µ ,

and

(x3 ) D 1 2 f x µ µ (3.88)

When f x µ > 0, the trivial ﬁxed point is stable when µ < 0 and unstable when µ > 0. When f x x x < 0, the two nontrivial ﬁxed points exist and are stable when µ > 0. This bifurcation of the nonhyperbolic ﬁxed point at the origin as µ passes through zero is called supercritical pitchfork bifurcation. On the other hand, when f x x x > 0, the two nontrivial ﬁxed points exist and are unstable when µ < 0. This bifurcation of the nonhyperbolic ﬁxed point at the origin as µ passes through zero is called subcritical or reverse pitchfork bifurcation. Example 3.11 We consider the one-dimensional map x kC1 D x k C µ x k C α x k3

(3.89)

3.4 Local Bifurcations

x0

0 μ

(a)

(b)

0 μ

Figure 3.4 Local scenarios: (a) supercritical pitchfork bifurcation and (b) subcritical pitchfork bifurcation.

where µ is a scalar control parameter. This map has the ﬁxed points x1 D 0W

trivial ﬁxed point

x2,3

nontrivial ﬁxed points.

p D ˙ µ/αW

The Jacobian matrix associated with the ﬁxed point x j has the single eigenvalue D 1 C µ C 3α x 2 j Therefore, the trivial ﬁxed point is stable for 2 < µ < 0 and unstable for all µ > 0. For α < 0, nontrivial ﬁxed points exist only for µ > 0, and they are stable for 0 < µ < 1. For α > 0, nontrivial ﬁxed points exist only for µ < 0, and they are unstable. The scenarios for α D 1 and α D 1 near the origin (x, µ) D (0, 0) are shown in Figure 3.4a and b, respectively. There is a supercritical pitchfork bifurcation at the origin in Figure 3.4a and a subcritical pitchfork bifurcation at the origin in Figure 3.4b.

3.4.4 Flip or Period-Doubling Bifurcation

As in the cases of fold, transcritical, and pitchfork bifurcations, analysis of the dynamics of an n-dimensional map in the neighborhood of a nonhyperbolic ﬁxed point with one eigenvalue being equal to 1 and the remaining eigenvalues being inside the unit circle can be reduced, by using the center-manifold theorem, to the analysis of the dynamics of the following one-dimensional map on the center manifold: x kC1 D f (x k I µ)

(3.90)

We assume that the ﬁxed point is at x D 0 when µ D 0 and that f (0, 0) D 0 and f x (0, 0) D 1.

81

82

3 Maps

Again, we expand f (xI µ) in a Taylor series for small x and µ and obtain x kC1 D x k C f µ µ C 12 f µ µ µ 2 C 2 f x µ x k µ C f x x x k2 C 16 f µ µ µ µ 3 C 3 f x µ µ x k µ 2 C 3 f x x µ x k2 µ C f x x x x k3 C

(3.91)

Therefore, the ﬁxed points of the map (3.91) are solutions of 2x C f µ µ C 12 f µ µ µ 2 C 2 f x µ x µ C f x x x 2 C 16 f µ µ µ µ 3 C 3 f x µ µ x µ 2 C 3 f x x µ x 2 µ C f x x x x 3 C D 0

(3.92)

For (x, µ) near (0, 0), (3.92) has a single solution x given by x D

1 2

f µ µ C O(µ 2 )

(3.93)

To ascertain the stability of this ﬁxed point, we differentiate (3.91) with respect to x and obtain D x f (xI µ) D 1C f x µ µC f x x x C 12 f x µ µ µ 2 C f x x µ x µC 12 f x x x x 2 C (3.94) Substituting (3.93) into (3.94) yields the multiplier (x ) D 1 C f x µ C

1 2

f µ f x x µ C O(µ 2 )

(3.95)

Hence, if f x µ C 1/2 f µ f x x > 0, the ﬁxed point x is stable when µ > 0 and unstable when µ < 0. On the other hand, if f x µ C 1/2 f µ f x x < 0, the ﬁxed point x is stable when µ < 0 and unstable when µ > 0. For nondegenerate bifurcation, 1 d (x ) D f x µ C f µ f x x ¤ 0 dµ 2

(3.96)

In other words, the multiplier of the map at the ﬁxed point should exit the unit circle through 1 with nonzero speed. To investigate the bifurcating solutions, we ﬁrst introduce a transformation to reduce (3.91) to its normal form by shifting the ﬁxed point to x D 0 and eliminating the quadratic term. To this end, we let xD

1 2

fµ µ C y C by2

(3.97)

Substituting (3.97) into (3.91) and choosing b to eliminate the quadratic term in y, we obtain the normal form y kC1 D (1 C ν)y k C α y k3

(3.98)

where ν D fxµ C αD

1 6

fxxx C

1 µC f f 2 µ xx 3 2 f C O(µ) 8 xx

O(µ 2 ) ,

bD

1 4

f x x C O(µ) ,

3.4 Local Bifurcations

The ﬁxed point y D 0 of (3.98) ceases to be hyperbolic when ν D 0 and the map undergoes a ﬂip bifurcation there. To determine the bifurcating period-two orbit, we investigate the ﬁxed points of the second iterate of the map (3.98); that is, y kC2 D (1 C 2ν)y k 2α y k3 C

(3.99)

It follows from (3.99) that the ﬁxed points of the second iterate map are r r ν ν y D 0 , , α α The ﬁrst ﬁxed point is also a ﬁxed point of the map. The associated multipliers are D 1 C 2ν , 1 4ν , 1 4ν Therefore, the ﬁxed point y D 0 is stable when ν < 0 and unstable when ν > 0. If α > 0, the nontrivial ﬁxed points exist and are stable for ν > 0. Consequently, the period-two bifurcating orbit is stable, and the bifurcation of the nonhyperbolic ﬁxed point at the origin as µ passes through zero is called supercritical period-doubling or ﬂip bifurcation. If α < 0, the nontrivial ﬁxed points exist and are unstable for ν < 0. Consequently, the period-two bifurcating orbit is unstable, and the bifurcation of the nonhyperbolic ﬁxed point at the origin as µ passes through zero is called subcritical period-doubling or ﬂip bifurcation. Example 3.12 We consider the logistic map x kC1 D F(x k ) D 4α x k (1 x k )

(3.100)

for 0 x 1 and positive α. The ﬁxed points of this map are x1 D 0 and

x2 D 1

1 4α

The multiplier associated with the ﬁxed point x j is given by j D 4α(1 2x j ) 1

1

F(x)

F2(x)

0

(a)

0

x

1

0

(b)

0

x

Figure 3.5 Graphs to determine solutions of the logistic equation at α D 0.8: (a) ﬁxed points and (b) period-two points.

1

83

84

3 Maps

The trivial ﬁxed point x1 exists for all α. Because 1 D 4α, it is stable when α < 1/4. The nontrivial ﬁxed point x2 exists for α > 1/4. Because 2 D 2 4α, this nontrivial ﬁxed point is stable for 1/4 < α < 3/4. At α D 3/4, x2 is nonhyperbolic because 2 D 1 and it is unstable for α > 3/4. In Figure 3.5a, we plot F(x) versus x when α D 0.8. The intersections of the line y D x with the curve F(x) give the ﬁxed points of F(x). They occur at x1 D 0 and x2 D 0.6875. The ﬁxed point x2 D 0.6875 is unstable because jF 0 (x2 D 0.6875)j D 1.2 > 1. Next, we investigate the ﬁxed points of F (2) (x), which are given by F (2) (x) D F [F(x)] D F [4α x (1 x)] or F (2) (x) D 16α 2 x (1 x) 1 4α x (1 x) Hence, the ﬁxed points of F (2) (x) are given by the solutions of 16α 2 x (1 x) 1 4α x (1 x) D x which are 1 x D 0 ,1 , 4α

and

2 3 s 1 2 1 1 41 2α 15 C ˙ 2 4α 2 2

The ﬁrst two ﬁxed points are also ﬁxed points of F(x). In Figure 3.5b, we plot F (2) (x) versus x when α D 0.8. The intersections of the curve F (2) (x) with the curve y D x give the ﬁxed points of F (2) (x). We note that there are four intersections. The dot corresponds to the ﬁxed point x D 1 1/(4α) D 0.6875 of F (2) (x), which is also a ﬁxed point of F(x). The other three ﬁxed points are x D 0, x D 0.5130, and x D 0.7995. We note that F(0.5130) D 0.7995

and

F (2) (0.5130) D 0.5130

and

F (2) (0.7995) D 0.7995

and that F(0.7995) D 0.5130

Hence, x D 0.5130 and x D 0.7995 are period-two points of F(x). To determine the stability of the ﬁxed points x j of F (2) (x), we calculate its Jacobian; that is, j D

h i d h (2) i F (x j ) D 16α 2 (1 2x j ) 1 8α x j (1 x j ) dx

For the ﬁxed points x D 0 and x D 0.6875, D 10.24 and D 1.44, respectively. Hence, these ﬁxed points are unstable. This is expected because these ﬁxed points are unstable ﬁxed points of F(x); F 0 (x D 0) D 3.2 and F 0 (x D 0.6875) D 1.2. For the ﬁxed points x D 0.5130 and x D 0.7995 of F (2) (x), D 0.16, and hence both of the period-two points of F(x) are stable.

3.4 Local Bifurcations

0.8 x 0.6

0.4 0.65

0.7

0.75 α

0.8

0.85

Figure 3.6 Scenario in the vicinity of a period-doubling bifurcation of the ﬁxed point of the logistic map.

We numerically determined that the ﬁxed points x3 and x4 of F (2) (x) or equivalently the period-two points of F(x) are stable for α < 0.85. In Figure 3.6, we show the different solutions of F(x) and their stability for 0.65 < α < 0.85. The branches of stable and unstable solutions are depicted by solid and broken lines, respectively. The ﬁxed point x2 of F(x) experiences a supercritical period-doubling bifurcation at α D 0.75. As a consequence, two branches of period-two points emerge from this bifurcation point. We note that an iterate of F(x) initiated at either of the two period-two points ﬂips back and forth between them because F(x3 ) D x4

and

F(x4 ) D x3

For this reason, the period-doubling bifurcation of a ﬁxed point of a map is also called a ﬂip bifurcation. The ﬁxed point x2 and the period-two points x3 and x4 of the map F(x) are all ﬁxed points of the map F (2) (x). At α D 3/4, the ﬁxed point x2 D 2/3 of F (2) (x) is nonhyperbolic because 2 D 1. Hence, from Figure 3.6, we infer that this ﬁxed point of F (2) (x) experiences a pitchfork bifurcation at α D 3/4. In this case, the pitchfork bifurcation is supercritical. In other cases, it is possible that a perioddoubling bifurcation of F(x) can correspond to a subcritical pitchfork bifurcation of the map F (2) (x). The associated period-two points that arise due to this bifurcation will be unstable.

3.4.5 Hopf or Neimark–Sacker Bifurcation

When a ﬁxed point of (3.55) is nonhyperbolic with a pair of complex conjugate eigenvalues on the unit circle, the ﬁxed point of the map can experience what is called the Neimark–Sacker bifurcation or Hopf bifurcation (Iooss, 1979; Wiggins, 1990). This bifurcation can occur in two- and higher-dimensional maps. Again, center-manifold reduction analysis can be used to reduce the analysis of the dynamics of the n-dimensional map in the vicinity of the nonhyperbolic ﬁxed point

85

86

3 Maps

into the analysis of the dynamics of a two-dimensional map on the center manifold (Carr, 1981); that is, x kC1 D f (x k , y k I µ)

(3.101)

y kC1 D g(x k , y k I µ)

(3.102)

We assume that coordinate and parameter transformations have been used so that the ﬁxed point is at (x, y ) D (0, 0) when µ D 0. Hence, f (0, 0I 0) D 0 , and the Jacobian matrix 2 @f (0, 0I 0) 6 @x 6 AD4 @g (0, 0I 0) @x

g(0, 0I 0) D 0 3 @f (0, 0I 0) 7 @y 7 5 @g (0, 0I 0) @y

(3.103)

has a pair of complex-conjugate eigenvalues D e i β and N D e i β lying on the unit circle. In the absence of strong resonances (i.e., n ¤ 1 for n D 1, 2, 3, 4), one can follow steps similar to those in Examples 3.4 and 3.5 and the next example to obtain the following normal form of the map (3.101) and (3.102):

z kC1 D e i β C νµ e i τ z k C α z 2 zN (3.104) where ν and τ are real-valued parameters and z is a complex-valued quantity. We assume that ν cos(β τ) ¤ 0 so that the pair of complex-conjugate multipliers transversely exit the unit circle as µ varies past zero. Substituting the polar form z D r eiθ

(3.105)

into (3.104) and separating real and imaginary parts, we obtain r kC1 D r k C νµ cos(β τ)r k C Λ r r k3 C

(3.106)

θkC1 D θk C β νµ sin(β τ) C Λ i r k2 C

(3.107)

for small r and µ, where Λ r C i Λ i D α e i β . We note that, because the r component is independent of θ , the problem is reduced to studying the stability of the ﬁxed points of the one-dimensional map (3.106). We assume that Λ r ¤ 0; otherwise higher-order terms need to be included in (3.104). There are three ﬁxed points: s s νµ cos(β τ) νµ cos(β τ) r D0, r D , and (3.108) Λr Λr The origin is asymptotically stable for νµ cos(β τ) < 0, unstable for νµ cos(β τ) > 0, unstable for νµ cos(β τ) D 0 and Λ r > 0, and asymp-

3.4 Local Bifurcations

totically stable for νµ cos(β τ) D 0 and Λ r < 0. On the other hand, the nontrivial ﬁxed points exist when νµ Λ r cos(β τ) > 0 and correspond to invariant circles. They are stable for νµ cos(βτ) > 0 and Λ r < 0 and unstable for νµ cos(βτ) < 0 and Λ r > 0. Substituting for the nontrivial ﬁxed points from (3.108) into (3.107) leads to the circle map θkC1 D θk C β νµ sin(β τ)

νµ Λ i cos(β τ) Λr

(3.109)

which describes the dynamics on the invariant circle. The dynamics on the invariant circle is periodic or aperiodic (densely ﬁlls the invariant circle) depending on whether χ D β νµ sin(β τ)

νµ Λ i cos(β τ) Λr

is rational or irrational. A rational number is a number that can be written as the ratio of two integers; that is, χ D p /q where p and q are integers. An irrational number cannot be written as the ratio of two integer. p For example, 0.2 is a rational number because it can be written as 1/5, whereas 5 is an irrational number. When χ D p /q where p and q are integers, the orbit of (3.109) starting at θ0 consists of q points fθ0 , θ0 C 2π χ, θ0 C 4π χ, , θ0 C 2(q 2)π χ, θ0 C 2(q 1)π χg We note that the next point in the orbit is θ0 C 2q π χ D θ0 C 2p π D θ0 . Consequently, the orbit would repeat itself. This orbit can be represented by q points on a circle. For example, when χ D 3/5, the orbit consists of the ﬁve points ˚

θ0 , θ0 C 65 π, θ0 C

12 π, θ0 5

C

18 π, 5

θ0 C

24 π 5

which can be represented as ﬁve points on a circle, as shown in Figure 3.7a.

(a)

(b)

Figure 3.7 Dynamics of a linear p circle map: (a) periodic motion when χ D 3/5 and (b) quasiperiodic motion when χ D 1/ 7.

87

88

3 Maps

When χ is irrational, starting from any point θ0 on a circle, we ﬁnd that none of the iterates F (m) returns to the initial point θ0 for any ﬁnite value of m. Thus, the orbit wanders on the circle, ﬁlling it up without becoming periodic. In other words, the orbit for irrational χ is aperiodic and is p an example of a quasiperiodic orbit. An example is shown in Figure 3.7b for χ D 1/ 7. We note that two quasiperiodic orbits starting from two points a small distance apart remain close for all subsequent iterations; that is, there is no sensitivity to initial conditions. Example 3.13 We consider the map x kC1 D (1 C µ)y k

(3.110)

y kC1 D y k x k 2x k y k

(3.111)

which has the ﬁxed points (x, y ) D (0, 0) and (x, y ) D (1/2 1/2µ, 1/2). The Jacobian of this map is 0 1Cµ JD 1 2y 1 2x Consequently, the multipliers associated with the nontrivial ﬁxed point are D 0 and D 2 C µ and hence it is unstable for values of µ near zero. On the other hand, the multipliers associated with the trivial ﬁxed point are D

1 2

p

p 2 3 p p i tan1 3C 3 µ ˙ 12 i 3 C 4µ D 1 C µe

These complex-conjugate multipliers transversely exit the unit circle away from the real axis as µ increases past zero, and hence the trivial ﬁxed point undergoes a Neimark–Sacker bifurcation as µ increases past zero. To analyze the dynamics of the map near this bifurcation, we construct the normal form of the map (3.110) and (3.111) for small x, y, and µ. To this end, we rewrite it as x kC1 0 1 xk 0 µ xk 0 D C C (3.112) 1 1 y k 0 0 yk 2x k y k y kC1 The eigenvalues of the linear part of (3.112) when µ D 0 are p p 1 1 1 D 12 (1 C i 3) D e 3 i π and 2 D N1 D 12 (1 i 3) D e 3 i π and the associated eigenvectors are the columns of the matrix p p 1 (1 i 3) 12 (1 C i 3) 2 PD 1 1 Next, we introduce the transformation x z DP y zN

(3.113)

3.4 Local Bifurcations

Hence, p p x D 12 (1 i 3)z C 12 (1 C i 3) zN

and

y D z C zN

(3.114)

Substituting (3.113) and (3.114) into (3.112) and multiplying the outcome from the left with P 1 , we obtain p 1p 2 3 2 1 3i µ(z k C zN k ) C z kC1 D e 3 i π z k C i zk 3 3 p ! p ! 3 3 1 (3.115) i z k zN k 1 C i zN k2 3 3 with the second equation being the complex-conjugate of this equation. Next, we determine the normal form of (3.115) for small z and µ. Instead of using two successive transformations to simplify the quadratic terms and then the cubic terms, we use a single transformation to accomplish simpliﬁcation of both nonlinearities. To this end, we introduce a small nondimensional bookkeeping parameter , scale z as z and µ as 2 µ, let z D ξ C h 1 (ξ , ξN ) C 2 h 2 (ξ , ξN )

(3.116)

and choose h 1 (ξ , ξN ) and h 2 (ξ , ξN ) so that (3.115) takes on the simplest form 1 ξkC1 D e 3 i π ξk C g 1 (ξk , ξN k ) C 2 g 2 (ξk , ξN k )

(3.117)

Substituting (3.116) and (3.117) into (3.115) and equating coefﬁcients of like powers of , we obtain Order ()

1 1 1 g 1 (ξ , ξN ) C h 1 e 3 i π ξ , e 3 i π ξN e 3 i π h 1 (ξ , ξN ) p ! p ! p 3 3 2 3 2 N iξ 1 i ξξ 1C i ξN 2 D 3 3 3

(3.118)

Order (2 )

1 1 1 g 2 (ξ , ξN ) C h 2 e 3 i π ξ , e 3 i π ξN e 3 i π h 2 (ξ , ξN ) p 4 3 1p 3i µ(ξ C ξN ) C i ξ h 1 (ξ , ξN ) D 3 3 p !

1 3 @ i π e 3 g 1 (ξ , ξN ) h 1 (ξ , ξN ) 1 i ξ hN 1 (ξ , ξN ) C ξN h 1 (ξ , ξN ) @ξ 3 p ! 3 @ N 13 i π N N gN 1 (ξ , ξ ) h 1 (ξ , ξ ) 2 1 C i ξN hN 1 (ξ , ξN ) e (3.119) N 3 @ξ

89

90

3 Maps

Because e 1/3i n π ¤ 1 for n D 1 and 3, there are no strong resonances, and hence h 1 (ξ , ξN ) can be chosen to eliminate all of the quadratic terms in (3.118); that is, p ! p ! p 3 3 3 2 1 2 (3.120) h 1 (ξ , ξN ) D iξ 1 C i ξ ξN C 1 i ξN 2 3 3 2 3 Consequently, g 1 (ξ , ξN ) 0 and (3.119) becomes

1 1 1 g 2 (ξ , ξN ) C h 2 e 3 i π ξ , e 3 i π ξN e 3 i π h 2 (ξ , ξN ) p p p D 13 3i µ(ξ C ξN ) C 2ξ 3 C 2(1 i 3)ξ 2 ξN C 4ξ ξN 2 C (1 i 3) ξN 3 (3.121) One can choose h 2 (ξ , ξN ) to eliminate the nonresonance terms in (3.122), leaving g 2 (ξ , ξN ) with the resonance terms; that is, p p (3.122) g 2 (ξ , ξN ) D 13 3i µ ξ C 2(1 i 3)ξ 2 ξN Therefore, the normal form of the map is p p 1 ξkC1 D e 3 i π ξk C 13 3i µ ξk C 2(1 i 3)ξk2 ξN k

(3.123)

where the bookkeeping parameter has been set equal to unity. Equation 3.123 can be rewritten as h i p p 1 1 ξkC1 D 1 C 12 µ e 3 i πC 6 3i µ ξk (2 C 2i 3)ξk2 ξN k C

(3.124)

Letting ξ D r e i θ in (3.124), we obtain r kC1 D 1 C 12 µ r k 2r k3

(3.125)

θkC1 D θk C 13 π C

1 6

p

p 3µ 2 3r k2

(3.126)

p It follows from (3.125) that the attracting invariant circle r D 1/2 µ bifurcates from the origin as µ increases past zero. The dynamics on this p invariant circle is periodic or quasiperiodic, depending on whether 1/3(π 3µ) is rational or irrational. We note that higher-order terms in µ and r have been neglected in arriving at (3.124) and (3.125). Including these higher-order terms, one ﬁnds that the attracting smooth invariant curve is a circle only for values of µ close to zero. Moreover, for large values of µ, the attractor may be complicated. In Figure 3.8, we show phase portraits obtained numerically for the map (3.110) and (3.111) for µ D 0.01 and µ D 0.02. When µ < 0, the origin is asymptotically stable and all iterates starting within its domain of attraction spiral to it, as shown in Figure 3.8a. The six-fold rotational symmetry is the result of β being 1/3π. When µ > 0, the origin is unstable and all iterates starting close to the origin spiral out to a smooth closed invariant curve enclosing the origin, as shown in Figure 3.8b. For the chosen value of µ, the long-time iterates densely ﬁll the invariant curve. Again, the six-fold rotational symmetry is the result of β being 1/3π.

3.5 Exercises

0.2

0.2

0.1

0.1

0

0

−0.1

−0.1

−0.1

(a)

0

0.1

0.2

(b)

−0.1

0

0.1

0.2

Figure 3.8 Phase portrait of the map (3.110) and (3.111): (a) µ < 0 and the origin is stable and iterates spiral to it and (b) µ > 0 and the origin is unstable and iterates spiral away from it and onto a smooth invariant closed curve encircling it.

3.5 Exercises

3.5.1 Consider the map x kC1 D x k C µ x k2 Examine the bifurcation that occurs at (x k , µ) D (1, 1). 3.5.2 Consider the map x kC1 D x k C µ x k x k2 Study the bifurcation that takes place at (x, µ) D (0, 2). 3.5.3 Consider the following map: x kC1 D µ C x k2 Examine the bifurcation that takes place at (x, µ) D (1/2, 1/4). 3.5.4 Consider the following map: x kC1 D µ x k C x k2 Examine the bifurcation that takes place at (x, µ) D (0, 1). 3.5.5 Consider the map x kC1 D ax k2 C 1 Determine its ﬁxed points and their stability. Examine the bifurcation that takes place as a is varied. 3.5.6 Find the ﬁxed points of x nC1 D ax n x n3 Determine their stability and the bifurcations, which they undergo as a is varied.

91

92

3 Maps

3.5.7 Find the ﬁxed points of x nC1 D ax n C x n3 Determine their stability and the bifurcations, which they undergo as a is varied. 3.5.8 Consider the following map: x kC1 D µ x k C x k2 Determine whether the period-doubled orbit bifurcating from the point (x, µ) D (0, 0) is supercritical or subcritical. 3.5.9 Consider the map x nC1 D (1 C a)x n C b x n2 x n3 What is the type of bifurcation which occurs at x D 0 and a D 0? Is it supercritical or subcritical? 3.5.10 Show that a ﬁxed point of the map f (xI λ) D (1 C λ)x C x 2 undergoes a transcritical bifurcation at λ D 0. 3.5.11 Show that the map f (xI λ) D e x λ undergoes a saddle-node bifurcation at λ D 1. 3.5.12 Determine the ﬁxed points of the following cubic map and discuss their stability: x nC1 D (1 λ)x n C λx n3 For what value of λ does the ﬁrst period-doubling bifurcation occur? 3.5.13 Consider the map x kC1 D x k C ax k2 C b x k3 Compute the second iterate of this map and hence determine whether the perioddoubled orbit is stable or unstable. 3.5.14 Determine the normal form of the map x kC1 D x k C ax k2 C b x k3 3.5.15 Consider the two one-dimensional maps x nC1 D e x n λ y nC1 D 12 λ tan1 y n Find the ﬁxed points and their stability. Show that x undergoes a saddle-node bifurcation at λ D 1, whereas y undergoes a period-doubling bifurcation at λ D 3.

3.5 Exercises

3.5.16 Consider the Hénon map x kC1 D 1 C y k α x k2 y kC1 D β x k Assume that β ¤ 0 and α > 0. a) Verify that this map has a stable ﬁxed point and an unstable ﬁxed point when α < 34 (1 β)2 b) Is this map dissipative for β D 0.3? For this case, determine the period-one and period-two points of this map for α D 0.1, α D 0.5, and α D 1.3. Discuss their stability. c) For the above values of α, plot the iterates of this map. d) Examine the bifurcation that takes place at (1 β) β(1 β) 3 (x k , y k , α) D , , (1 β)2 . 2α 2α 4

3.5.17 Consider the two-dimensional map x nC1 D λ C x n C λ y n C x n2 y nC1 D 12 y n C λx n C x n2 Determine its ﬁxed points and then show that the origin undergoes a saddle-node bifurcation at λ D 0. 3.5.18 Consider the two-dimensional map: x nC1 D y n y nC1 D 12 x n C λ y n y n3 Determine its ﬁxed points. Show that the origin undergoes a pitchfork bifurcation at λ D 3/2. Analyze the bifurcation at λ D 3. 3.5.19 Consider the following map: x kC1 D 2x k y kC1 D 12 y k C 7x k2 Show that the stable manifold of its ﬁxed point is the y-axis and the unstable manifold is y D 2x 2 .

93

94

3 Maps

3.5.20 Consider the map x kC1 D y k y kC1 D α y k (1 x k ) Show that the origin is a saddle. Determine analytical expressions for the stable and unstable manifolds. 3.5.21 Consider the map x kC1 D y k y kC1 D α y k (1 x k ) Examine the bifurcation that occurs at α D 2 and determine the normal form of the map near this bifurcation. 3.5.22 Consider the map x kC1 D y k y kC1 D α y k (1 x k ) Examine the bifurcation that occurs at α D 1 and determine the normal form of the map near this bifurcation. 3.5.23 Consider the following map near the origin: x kC1 D x k C α x k y k y kC1 D 12 y k C x k2 Compute the center manifold, describe the dynamics on the center manifold, and then determine the stability of the origin. 3.5.24 Consider the map x nC1 D y n y nC1 D b x n y n C x n y n What type of bifurcation occurs at (x, y, b) D (0, 0, 0). 3.5.25 Consider the map x kC1 D 58 x k C 38 y k C x y y kC1 D 38 x k C 58 y k C x 2 y 2 Compute the center manifold near the origin, describe the dynamics on the center manifold, and then determine the stability of the origin.

3.5 Exercises

3.5.26 Consider the two-dimensional map x nC1 D (1 C µ)y n y nC1 D y n x n 2x n y n Determine the ﬁxed points of this map and their stability. Show that the origin undergoes a Hopf bifurcation at µ D 0. Calculate and plot the phase portraits for a) µ D 0.01 using the initial condition x D 0.4, y D 0.4. b) µ D 0.05 using the initial condition x D 0.01, y D 0.01. 3.5.27 Consider the map x kC1 D y k y kC1 D 12 µ C (1 C µ) (y k x k 2x k y k ) a) Determine the ﬁxed points and their stability. b) Compute the normal form of this map near the origin when µ 0. c) Use this normal form to calculate the bifurcating invariant circle from the origin. d) Calculate the phase portraits for µ D 0.01 and µ D 0.02. 3.5.28 Consider the map z kC1 D z k C b 11 z 2 C b 12 z zN C b 13 zN 2 C a 11 z 3 C a 12 z 2 zN C a 13 z zN 2 C a 14 zN 3 Show that its normal form near the origin has the form z kC1 D z k C α 1 zN 2 C α 2 z 2 zN when 3 D 1 and the form z kC1 D z k C α 1 z 2 zN C α 2 zN 3 when 4 D 1. 3.5.29 Consider the map p z nC1 D 12 ( 3 C i C 2i µ)z n C z n2 zN n What type of bifurcation occurs at (z, zN , µ) D (0, 0, 0). Determine the bifurcating solutions.

95

97

4 Bifurcations of Continuous Systems In this chapter, we construct normal forms of smooth continuous systems depending on a scalar control parameter µ near their ﬁxed or equilibrium points. Speciﬁcally, we consider the following system: xP D F(xI µ) where x 2 U R n , F 2 U R n , and µ 2 V R. The ﬁxed points or equilibrium solutions of this system are solutions of the algebraic system of equations F(xI µ) D 0 In Section 4.1, we consider linear systems; in Section 4.2, we consider the ﬁxed points of nonlinear systems and their stability; in Section 4.3, we discuss centermanifold reduction; in Section 4.4, we consider local bifurcations of ﬁxed points; and in Sections 4.5 and 4.6, we illustrate how the method of multiple scales, a combination of center-manifold reduction and the method of normal forms, and a projection method can be used to construct the normal forms of static and Hopf bifurcations in the neighborhood of a ﬁxed point.

4.1 Linear Systems

We consider the linear system xP D Ax

(4.1)

where A is an n n constant matrix. In this case, the trivial solution x D 0 is a ﬁxed point of this linear system. We denote the eigenvalues of A by λ i , i D 1, 2, . . . , n, and the corresponding eigenvectors (generalized eigenvectors) by p i , i D 1, 2, . . . , n. The eigenvalues are the roots of the characteristic equation det(A λ I ) D 0

(4.2)

The eigenvector p i corresponding to a distinct eigenvalue λ i is given by Ap i D λ i p i

(4.3)

The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

98

4 Bifurcations of Continuous Systems

and the generalized eigenvectors corresponding to an eigenvalue λ m with multiplicity n m are the nontrivial solutions of (A λ m I )p D 0 ,

(A λ m I )2 p D 0 ,

... ,

(A λ m I ) n m p D 0

(4.4)

If an eigenvalue is complex-valued, then its corresponding eigenvector and generalized eigenvectors are also complex-valued. Introducing the transformation x D Py

(4.5)

where the matrix P D [p 1 p 2 , . . . , p n ], into (4.1) yields P yP D AP y

(4.6)

Multiplying (4.6) from the left with the inverse P 1 of P, we obtain the normal of (4.1) as yP D J y

(4.7)

1

where J D P AP is called the Jordan canonical form of A. Next, we discuss two cases: systems with distinct and nondistinct eigenvalues. 4.1.1 Case of Distinct Eigenvalues

If the eigenvalues of A are distinct, then J is a diagonal matrix D with entries λ i , i D 1, 2, . . . , n; that is, 2 3 λ1 0 0 6 0 λ2 0 7 6 7 D D6 (4.8) 7 6 7 4 5 0 0 λn Then, (4.7) can be rewritten as yP (m) D λ m y (m) ,

m D 1, 2, . . . , n

(4.9)

where y (m) is the mth component of y. Therefore, y (m) D c m e λ m t ,

m D 1, 2, . . . , n

(4.10)

where c m is a constant. It follows from (4.10) that y (m) ! 0 as t ! 1 when λ m lies in the left-half of the complex plane, y (m) ! 1 as t ! 1 when λ m lies in the right-half of the complex plane, and y (m) D c m for all time when λ m lies on the imaginary axis. Therefore, the origin of (4.1) is (a) asymptotically stable if all of the eigenvalues λ i of the matrix A lie in the left-half of the complex plane, (b) unstable if one or more λ i lie in the right-half of the complex plane, and (c) neutrally or marginally stable if one or more λ i lie on the imaginary axis with the rest of the eigenvalues being in the left-half of the complex plane.

4.1 Linear Systems

4.1.2 Case of Repeated Eigenvalues

If the number of distinct eigenvalues of A is k < n, then J is diagonal if all of the p i are eigenvectors; otherwise, it has the form 3 2 J1 φ φ 6φ J φ7 2 7 6 7 6 7 6 (4.11) J D6 7 7 6 7 6 4 5 φ φ Jk where φ represents a matrix with zero entries and 3 2 1 0 0 λm 60 λm 1 07 7 6 6 0 λm 7 Jm D 6 0 7 4 5 0 0 0 λm

(4.12)

Example 4.1 We consider a system with repeated roots and a nondiagonal J; that is, 1 (a b) a 2 AD 12 (a b) b where b ¤ a. The eigenvalues of this matrix are D 1/2(a C b) with a multiplicity of two. It follows from Example 3.3. that the corresponding eigenvector and generalized eigenvector are 1 p1 D 1 and

2

3

1

p2 D 4

1C

Then,

2 5 ba 2

1 6 2 (a C b) 1 P AP D J D 4 0

3 1

7 5 1 (a C b) 2

and (4.7) yields yP (1) D λ y (1) C y (2)

(4.13)

yP (2) D λ y (2)

(4.14)

99

100

4 Bifurcations of Continuous Systems

The general solutions of (4.13) and (4.14) can be expressed as y (1) D c 1 e λ t C t c 2 e λ t

(4.15)

y (2) D c 2 e λ t

(4.16)

where c 1 and c 2 are arbitrary constants. Therefore, the origin of (4.1) is (a) asymptotically stable if Real(λ) < 0 and (b) unstable if Real(λ) 0.

4.2 Fixed Points of Nonlinear Systems

The ﬁxed points of the autonomous system xP D F(xI µ)

(4.17)

are deﬁned by the vanishing of the vector ﬁeld; that is, F(xI µ) D 0

(4.18)

A location in the state space where this condition is satisﬁed is called a singular point. At such a point, the integral curve of the vector ﬁeld F corresponds to the point itself. Also, an orbit of a ﬁxed point is the ﬁxed point itself. Fixed points are also called stationary solutions, critical points, constant solutions, and sometimes steady-state solutions. Physically, a ﬁxed point corresponds to an equilibrium position of a system. Further, ﬁxed points are examples of invariant sets of (4.17). 4.2.1 Stability of Fixed Points

To investigate the stability of a ﬁxed point x (µ ), where x 2 R n and µ 2 R, we superimpose on it a small disturbance y(t) and obtain x(t) D x C y(t)

(4.19)

Substituting (4.19) into (4.17) yields yP D F(x C yI µ )

(4.20)

We note that the ﬁxed point x D x of (4.17) has been transformed into the ﬁxed point y D 0 of (4.20). Assuming that F is at least twice continuously differentiable (i.e., C 2 ), expanding (4.20) in a Taylor series about x , and retaining only linear terms in the disturbance leads to yP D F(x I µ ) C D x F(x I µ )y C O(jjyjj2 )

4.2 Fixed Points of Nonlinear Systems

or yP D x F (x I µ )y Ay

(4.21)

where A, the matrix of ﬁrst partial derivatives, is called the Jacobian matrix. If the components of F are F1 (x1 , x2 , . . . , x n ), F2 (x1 , x2 , . . . , x n ), . . . , F n (x1 , x2 , . . . , x n ) then

2 @F

1

@x 6 @F1 6 2 6 @x1

6 6 6 AD6 6 6 6 6 4

@F n @x1

@F1 @x2 @F2 @x2

@F n @x2

3

@F1 @x n 7 @F2 7 @x n 7

7 7 7 7 7 7 7 7 5

@F n @x n

We have transformed the problem of determining the local stability of the ﬁxed point x of (4.17) into that of determining the stability of the trivial solution of the linear system (4.21). We say local stability because we have considered a small disturbance and linearized the vector ﬁeld. It follows from the preceding section that the trivial solution of (4.21) and hence the ﬁxed point x of (4.17) is (a) asymptotically stable if all of the eigenvalues λ i of the matrix A lie in the left-half of the complex plane and (b) unstable if one or more λ i lie in the right-half of the complex plane. In the case of repeated eigenvalues, the trivial solution of (4.21) and hence the ﬁxed point x of (4.17) is unstable if one or more eigenvalues lie on the imaginary axis and the Jordan form is not diagonal. 4.2.2 Classiﬁcation of Fixed Points

When all of the eigenvalues of A have nonzero real parts, the corresponding ﬁxed point is called a hyperbolic ﬁxed point, irrespective of the values of the imaginary parts; otherwise, it is called a nonhyperbolic ﬁxed point. There are three types of hyperbolic ﬁxed points: sinks, sources, and saddles. If all of the eigenvalues of A have negative real parts, then all of the components of the disturbance y decay in time, and hence x approaches the ﬁxed point x of (4.17) as t ! 1. Therefore, the ﬁxed point x of (4.17) is asymptotically stable. An asymptotically stable ﬁxed point is called a sink. If the matrix A associated with a sink has complex eigenvalues, the sink is also called a stable focus. On the other hand, if all of the eigenvalues of the matrix A associated with a sink are real, the sink is also called a stable node. A sink is stable in forward time (i.e., t ! 1) but unstable in reverse time (i.e., t ! 1). Further, all sinks qualify as attractors. If one or more of the eigenvalues of A have positive real parts, some of the components of y grow in time, and x moves away from the ﬁxed point x of (4.17) as

101

102

4 Bifurcations of Continuous Systems

t increases. In this case, the ﬁxed point x is said to be unstable. When all of the eigenvalues of A have positive real parts, x is said to be a source. If the matrix A associated with a source has complex eigenvalues, the source is also called an unstable focus. On the other hand, if all of the eigenvalues of the matrix A associated with a source are real, the source is also called an unstable node. A source is unstable in forward time but stable in reverse time. Because trajectories move away from a source in forward time, the source is an example of a repellor. When some, but not all, of the eigenvalues have positive real parts while the rest of the eigenvalues have negative real parts, the associated ﬁxed point is called a saddle point. Because a saddle point is unstable in both forward and reverse times, it is called a nonstable ﬁxed point. A nonhyperbolic ﬁxed point is unstable if one or more of the eigenvalues of A have positive real parts. If some of the eigenvalues of A have negative real parts while the rest of the eigenvalues are distinct and have zero real parts, the ﬁxed point x D x of (4.17) is said to be neutrally or marginally stable. If all of the eigenvalues of A are distinct, nonzero, and purely imaginary, the corresponding ﬁxed point is called a center. Example 4.2 We consider the system xP D x (3 x 2y )

(4.22)

yP D y (2 x y )

(4.23)

Its ﬁxed points are (0, 0), (0, 2), (3, 0), and (1, 1). The Jacobian matrix of the system (4.22) and (4.23) is AD

3 2x 2y y

2x 2 x 2y

The eigenvalues of A corresponding to the ﬁxed point (0, 0) are λ 1 D 2 and λ 2 D 3; hence, it is an unstable node. The eigenvalues of A corresponding to the ﬁxed point (0, 2) are λ 1 D 1 and λ 2 D 2; hence, it is a stable node. The eigenvalues of A corresponding to the ﬁxed point (3, 0) are λ 1 D 1 and λ 2 D 3; hence, it is a stable node. p The eigenvalues p of A corresponding to the ﬁxed point (1, 1) are λ 1 D 1 C 2 and λ 2 D 1 2; hence, it is a saddle. In this example, all of the ﬁxed points are hyperbolic.

4.2.3 Hartman–Grobman and Shoshitaishvili Theorems

Many theorems provide precise statements on what the stability of ﬁxed-point solutions of the linearized system (4.21) imply for the stability of ﬁxed-point solutions

4.3 Center-Manifold Reduction

of the full nonlinear system (4.17). The Hartman–Grobman theorem (e.g., Arnold, 1988, Chapter 3; Wiggins, 1990, Chapter 2) is applicable to hyperbolic ﬁxed points, whereas the Shoshitaishvili theorem (e.g., Arnold, 1988, Chapter 6) is applicable to nonhyperbolic ﬁxed points. From these theorems, it follows that (a) the ﬁxed point x D x of the nonlinear system (4.17) is stable when the ﬁxed point y D 0 of the linear system (4.21) is asymptotically stable; (b) the ﬁxed point x D x of the nonlinear system (4.17) is unstable when the ﬁxed point y D 0 of the linear system (4.21) is unstable; and (c) linearization cannot determine the stability of neutrally stable ﬁxed points (including centers) of (4.17). In the case of neutrally stable ﬁxed points, a nonlinear analysis is necessary to determine the stability of x . It will be necessary to retain quadratic and, sometimes, higher-order terms in the disturbance y in the Taylor-series expansion of (4.20). In a topological setting, the Hartman–Grobman theorem implies that the trajectories in the vicinity of a hyperbolic ﬁxed point x D x of (4.17) are qualitatively similar to those in the vicinity of the hyperbolic ﬁxed point y D 0 of (4.21). In other words, the local nonlinear dynamics near x D x is qualitatively similar to the linear dynamics near y D 0, and a qualitative change in the local nonlinear dynamics can be detected by examining the associated linear dynamics. According to the Hartman–Grobman theorem, there exists a continuous coordinate transformation (i.e., a homeomorphism) that transforms the nonlinear ﬂow into the linear ﬂow in the vicinity of a hyperbolic ﬁxed point. In the absence of resonances or near resonances, the method of normal forms (e.g., Sections 2.2 and 2.4) may be used to generate a coordinate transformation to transform the nonlinear ﬂow into linear ﬂow (e.g., Arnold, 1988; Guckenheimer and Holmes, 1983). Further, such a coordinate transformation would be a differentiable one because the method of normal forms yields transformations in the form of power-series expansions.

4.3 Center-Manifold Reduction

We consider the local dynamics near a nonhyperbolic ﬁxed point x of the nonlinear system (4.17), where F is an analytic vector function of x. According to the center-manifold theorem (Carr, 1981), there exists a C r local center manifold for the nonlinear system (4.17) near x . Furthermore, if none of the eigenvalues of this ﬁxed point lies in the right-half of the complex plane, the long-time dynamics of (4.17) can be reduced to determining the dynamics on the center manifold. Next, we describe how to construct the center manifold in the neighborhood of a nonhyperbolic ﬁxed point with one eigenvalue being zero and the real parts of all of the other eigenvalues being negative. We assume that the ﬁxed point has been shifted to the origin and that the linear part has been transformed into a Jordan canonical form; that is, we consider the system xP kC1 D J x k C F(x)

(4.24)

103

104

4 Bifurcations of Continuous Systems

We arrange the system (4.24) and rewrite it as xP D f (x, y)

(4.25)

yP D B y C G(x, y)

(4.26)

where B is a constant matrix with the real parts of all of its eigenvalues being negative and f and G are scalar and vector-valued nonlinear functions of x and y. According to the center-manifold theorem, there exists a center manifold y D h(x)

(4.27)

Moreover, the dynamics of the system (4.25) and (4.26) is qualitatively similar to the dynamics on this manifold; that is, xP D f x, h(x) (4.28) Substituting (4.27) into (4.26) yields P h(x) D B h(x) C G x, h(x)

(4.29)

which upon using (4.28) becomes h 0 (x) f x, h(x) D B h(x) C G x, h(x)

(4.30)

Using three examples, we describe how to construct approximate solutions of (4.30). Example 4.3 We consider the system xP D α 1 x y

(4.31)

yP D y C α 2 x 2

(4.32)

where α 1 and α 2 are constants. Clearly, the origin is a ﬁxed point of (4.31) and (4.32). Because its associated eigenvalues are λ D 0 and λ D 1, the origin is nonhyperbolic and the center and stable subspaces are one-dimensional. Consequently, the center manifold is one-dimensional. To calculate the center manifold and the dynamics on this manifold, we note that B D 1, f (x, y ) D α 1 x y , and G(x, y ) D α 2 x 2 . Hence, (4.30) becomes h 0 (x)α 1 x h(x) D h(x) C α 2 x 2

(4.33)

We seek an approximate solution of (4.33) in the form h(x) D ax 2 C

(4.34)

and obtain 2a 2 α 1 x 4 C ax 2 α 2 x 2 C D 0

(4.35)

4.3 Center-Manifold Reduction

Equating to zero the coefﬁcient of x 2 , we obtain a D α 2 . Hence, the center manifold is given by h(x) D α 2 x 2 C

(4.36)

and it follows from (4.31) that the long-time dynamics on this center manifold is given by xP D α 1 α 2 x 3 C

(4.37)

We note that linearization is not sufﬁcient for determining the stability of the origin because its associated eigenvalues are λ D 0 and λ D 1. However, including the nonlinear terms, we ﬁnd from (4.37) that the origin is unstable when α 1 α 2 > 0 and stable when α 1 α 2 < 0. Example 4.4 We consider the system xP D α 1 x y C α 2 x z

(4.38)

yP D µ y C z C α 3 x 2

(4.39)

zP D µ z y C α 4 x 2

(4.40)

where µ > 0. Clearly, the origin is a ﬁxed point of (4.38)–(4.40) and its associated eigenvalues are λ D 0 and λ D µ ˙ i. Hence, the origin is a nonhyperbolic ﬁxed point and the center and stable subspaces are one-dimensional and twodimensional, respectively. Consequently, the center manifold is two-dimensional. Thus, we seek the center manifold of (4.38)–(4.40) emanating from the origin in the form y (x) D ax 2 C

and

z(x) D b x 2 C

(4.41)

Substituting (4.41) into (4.38) leads to the following equation describing the dynamics on the manifold: xP D α 1 ax 3 C α 2 b x 3 C

(4.42)

Substituting (4.41) and (4.42) into (4.39) and (4.40) yields 2ax (α 1 ax 3 C α 2 b x 3 ) C µ ax 2 b x 2 α 3 x 2 C D 0

(4.43)

2b x (α 1 ax 3 C α 2 b x 3 ) C µ b x 2 C ax 2 α 4 x 2 C D 0

(4.44)

Equating the coefﬁcients of x 2 to zero in (4.43) and (4.44), we obtain µ a b D α3

(4.45)

a C µ b D α4

(4.46)

105

106

4 Bifurcations of Continuous Systems

Hence, µ α3 C α4 1 C µ2

aD

and

bD

µ α4 α3 1 C µ2

(4.47)

and it follows from (4.42) that the dynamics on the center manifold is given by xP D Γ x 3 C

(4.48)

where µ(α 1 α 3 C α 2 α 4 ) C α 1 α 4 α 2 α 3 1 C µ2

Γ D

We note again that linearization is not sufﬁcient for determining the stability of the origin because its associated eigenvalues are λ D 0 and λ D µ ˙ i. However, including the nonlinear terms, we ﬁnd from (4.48) that the origin is unstable when Γ > 0 and stable when Γ < 0. Example 4.5 We consider the system xP D y C a 1 x 2 C a 2 y 2 C a 3 z 2 C a 4 x y C a 5 x z C a 6 y z

(4.49)

yP D x C b 1 x 2 C b 2 y 2 C b 3 z 2 C b 4 x y C b 5 x z C b 6 y z

(4.50)

zP D z C c 1 x 2 C c 2 y 2 C c 3 z 2 C c 4 x y C c 5 x z C c 6 y z

(4.51)

Clearly, the origin x D y D z D 0 is a ﬁxed point of (4.49)–(4.51). Its associated eigenvalues are λ D i, i, 1, and hence the center subspace is two-dimensional and the stable subspace is one-dimensional. Consequently, the center manifold is one-dimensional. We seek the center manifold emanating from the origin in the form z(x, y ) D α 1 x 2 C α 2 x y C α 3 y 2

(4.52)

Substituting (4.52) into (4.49) and (4.50) yields the following two-dimensional system describing the dynamics on the center manifold: xP D y C a 1 x 2 C a 2 y 2 C a 4 x y C (a 5 x C a 6 y )(α 1 x 2 C α 2 x y C α 3 y 2 ) C

(4.53)

yP D x C b 1 x 2 C b 2 y 2 C b 4 x y C (b 5 x C b 6 y )(α 1 x 2 C α 2 x y C α 3 y 2 ) C

(4.54)

4.4 Local Bifurcations of Fixed Points

Substituting (4.52) into (4.51) and using (4.53) and (4.54), we obtain 2α 1 x y C α 2 y 2 α 2 x 2 2α 3 x y D α 1 x 2 α 2 x y α 3 y 2 C c1 x 2 C c2 y 2 C c4 x y C

(4.55)

Equating the coefﬁcients of x 2 , x y , and y 2 on both sides of (4.55) yields α1 α2 D c1

(4.56)

2α 1 2α 3 C α 2 D c 4

(4.57)

α2 C α3 D c2

(4.58)

Therefore, α 1 D 15 (3c 1 C 2c 2 C c 4 )

(4.59)

α 2 D 51 (2c 1 2c 2 c 4 )

(4.60)

α 3 D 15 (2c 1 C 3c 2 c 4 )

(4.61)

Substituting for the α i from (4.59)–(4.61) into (4.53) and (4.54), we obtain a twodimensional dynamical system describing the dynamics on the center manifold. Using a methodology similar to that in Section 2.5, we reduce this dynamical system into its normal form.

4.4 Local Bifurcations of Fixed Points

From Section 4.2, we know that the matrix A in (4.21) and the associated eigenvalues are functions of the control parameter µ. Let us suppose that, as µ is slowly varied, a ﬁxed point becomes nonhyperbolic at a certain location in the state-control space. Then, if the state-space portraits before and after this location are qualitatively different, this location is called a bifurcation point, and the accompanying qualitative change is called a bifurcation. There are two cases in which a ﬁxed point x of the continuous system (4.17) ceases to be hyperbolic at a critical value µ c of the control parameter. These cases are 1. D x F (x I µ c ) has one eigenvalue equal to zero with the remaining eigenvalues being in the left-half of the complex plane, 2. D x F (x I µ c ) has a pair of purely imaginary eigenvalues with the remaining eigenvalues being in the left-half of the complex plane.

107

108

4 Bifurcations of Continuous Systems

According to the center-manifold theorem, analysis of the the dynamics of an ndimensional continuous system near one of its ﬁxed points can be reduced to the analysis of the dynamics on its center manifold. In the ﬁrst case, the center manifold is (n 1)-dimensional and the analysis of the dynamics of the system can be reduced to that of analyzing the dynamics of a one-dimensional dynamical system. On the other hand, in the second case, the center manifold is (n 2)-dimensional and the analysis of the dynamics of the system can be reduced to that of analyzing a two-dimensional dynamical system. In Sections 4.5 and 4.6, we show how one can obtain such reduced dynamical systems. In Case 2, Hopf bifurcation can occur; and in Case 1, three types of static bifurcation can occur: saddle-node, transcritical, and pitchfork bifurcations. In Section 4.4.5, we consider Hopf bifurcation; and in Sections 4.4.1–4.4.4 we consider bifurcations of the one-dimensional dynamical system xP D f (xI µ)

(4.62)

where x and f are scalars and f (xI µ) is a smooth function of x and µ. We assume that (x D 0, µ D 0) is a nonhyperbolic ﬁxed point of (4.62); that is, f (0I 0) D 0 and f x (0, 0) D 0. Expanding f (xI µ) in a Taylor series for small x and µ, we obtain xP D f µ µ C 12 ( f µ µ µ 2 C 2 f x µ x µ C f x x x 2 ) C 16 ( f µ µ µ µ 3 C 3 f x µ µ x µ 2 C 3 f x x µ x 2 µ C f x x x x 3 ) C

(4.63)

Next, we consider the following four cases: (a) f µ ¤ 0 and f x x ¤ 0; (b) f µ ¤ 0, f x x D 0, and f x x x ¤ 0; (c) f µ D 0 and f x x ¤ 0; and (d) f µ D 0 and f x x D 0. We show below that the ﬁrst case corresponds to saddle-node bifurcation, the second case corresponds to a nonbifurcation, the third case corresponds to transcritical bifurcation, and the fourth case corresponds to pitchfork bifurcation. 4.4.1 Saddle-Node Bifurcation

We consider the case f µ ¤ 0 and f x x ¤ 0. As x ! 0 and µ ! 0, (4.63) tends to xP D f µ µ C

1 2

f xx x2 C

When f µ µ ¤ 0, the ﬁxed points of (4.64) are s 2 f µ µ x D˙ fxx

(4.64)

(4.65)

which exist only when f µ f x x µ < 0. The eigenvalues associated with these two ﬁxed points are s 2 f µ µ λ D ˙ fxx (4.66) fxx

4.4 Local Bifurcations of Fixed Points

It follows from (4.65) that there are two branches of ﬁxed points in the neighborhood of (x, µ) D (0, 0) for f µ µ < 0 if f x x > 0 and for f µ µ > 0 if f x x < 0. Then, it follows from (4.66) that the upper branch is stable and the lower branch is unstable if f x x < 0 and that the upper branch is unstable and the lower branch is stable if f x x > 0. This bifurcation of the nonhyperbolic ﬁxed point at the origin as µ passes through zero is called fold or tangent or saddle-node bifurcation. Example 4.6 We let f µ D 1 and f x x D α D ˙1 and consider the system xP D f (xI µ) D µ C α x 2

(4.67)

When α D 1, (4.67) does not have any ﬁxed points for µ < 0 but has the two nontrivial ﬁxed points p p x D µ and x D µ for µ > 0. The Jacobian matrix has a single eigenvalue given by λ D 2x p Thus, the ﬁxed point x D µ is a stable node because λ < 0, and the ﬁxed point p x D µ is an unstable node because λ > 0. In Figure 4.1a, we display the different ﬁxed-point solutions of (4.67) and their stability in the x µ space. We use broken and solid lines to depict branches of unstable and stable ﬁxed points, respectively. On the other hand, when α D 1, (4.67) does not have any ﬁxed points for µ > 0 but has the two nontrivial ﬁxed points p p x D µ and x D µ for µ < 0. The Jacobian matrix has a single eigenvalue given by λ D 2x p Thus, the ﬁxed point x D µ is an unstable node because λ > 0, and the ﬁxed p point x D µ is a stable node because λ < 0. In Figure 4.1b, we display the different ﬁxed-point solutions of (4.67) and their stability in the x µ space. We note that f (xI µ) D 0 and f x (xI µ) D 0 at (0, 0), and hence there is a nonhyperbolic ﬁxed point at µ D 0. Moreover, we note that there is a change in the number of ﬁxed points from zero to two as µ passes through zero. Therefore, the origin of the x µ space is a static bifurcation. It is clear from Figure 4.1 that the stable and unstable branches meet at the bifurcation point and have the same tangent. Therefore, this bifurcation is called a tangent bifurcation. Although branches of stable and unstable nodes meet at this bifurcation point, the tangent bifurcation is also called saddle-node bifurcation because, in higher-dimensional systems, branches of saddle points and stable nodes meet at such static bifurcation points. Equation 4.67 is the normal form for a generic saddle-node bifurcation of a ﬁxed point of a continuous system.

109

110

4 Bifurcations of Continuous Systems 0.4

0.4

0.2

0.2

x 0.0

x 0.0

–0.2

–0.2

–0.4 0.00

0.05

(a)

0.10 μ

0.15

0.20

–0.4 –0.20

(b)

–0.15

–0.10 μ

–0.05

0.00

Figure 4.1 Scenario in the vicinity of a saddle-node bifurcation: (a) xP D µ x 2 and (b) xP D µ C x 2.

4.4.2 Nonbifurcation Point

We consider the case f µ ¤ 0, f x x D 0, and f x x x ¤ 0. As µ ! 0 and x ! 0, (4.63) tends to xP D f µ µ C

1 6

f xxx x3 C

(4.68)

It has only one nontrivial ﬁxed point when f µ µ ¤ 0 given by

x D

6 f µ µ fxxx

1/3 (4.69)

The eigenvalue associated with this ﬁxed point is 1 λ D fxxx 2

6 f µ µ fxxx

2/3 (4.70)

It follows from (4.69) that this ﬁxed point exists on both sides of f µ µ and it follows from (4.70) that it is stable when f x x x < 0 and unstable when f x x x > 0. Although the ﬁxed point (4.69) is nonhyperbolic because f (xI µ) D 0 and λ D 0 at µ D 0, this ﬁxed point is not a bifurcation point because there is no qualitative change either in the number of ﬁxed-point solutions or in the stability of the ﬁxed-point solutions as µ passes through zero in the state-control space. Example 4.7 We let f µ D 1 and f x x x D 6 in (4.68) and consider xP D f (x, µ) D µ x 3 We have only one ﬁxed point, namely, x D µ 1/3

(4.71)

4.4 Local Bifurcations of Fixed Points

1.0 0.5 x 0.0 –0.5 –1.0 –1.0

–0.5

0.0 μ

0.5

1.0

Figure 4.2 Fixed-point solutions of (4.71).

This solution is depicted in Figure 4.2. At the origin of the x µ space, f (x, µ) D 0 and f x has a zero eigenvalue, implying that x D 0 is a nonhyperbolic ﬁxed point at µ D 0. However, (0, 0) is not a bifurcation point because there is no qualitative change either in the number of ﬁxed-point solutions or in the stability of the ﬁxedpoint solutions as µ passes through zero in the state-control space.

4.4.3 Transcritical Bifurcation

We consider the case f µ D 0 and f x x ¤ 0. As µ ! 0 and x ! 0, (4.63) tends to xP D 12 ( f µ µ µ 2 C 2 f x µ x µ C f x x x 2 ) C

(4.72)

whose ﬁxed points are x1 D x2

D

fxµ C fxµ

q q

f x2µ f µ µ f x x fxx f x2µ f µ µ f x x fxx

µ

and

µ

(4.73)

They exist when f x2µ f µ µ f x x > 0. Their associated eigenvalues are λ(x1 ) D

q

f x2µ f µ µ f x x µ

and

λ(x2 ) D

q

f x2µ f µ µ f x x µ

(4.74)

Therefore, x1 is stable when µ < 0 and unstable when µ > 0. On the other hand, x2 is unstable when µ < 0 and stable when µ > 0. In this case, we have two branches of ﬁxed points that interchange stability at (0, 0). Therefore, the bifurcation of the nonhyperbolic ﬁxed point at the origin as µ passes through zero is transcritical bifurcation.

111

112

4 Bifurcations of Continuous Systems

Example 4.8 We let f µ µ D 0, f µ x D 1, and f x x D 2 and consider the system xP D f (xI µ) D µ x x 2

(4.75)

There are two ﬁxed points: x D0I

trivial ﬁxed point

xDµI

nontrivial ﬁxed point

The Jacobian matrix f x D µ 2x has the single eigenvalue λDµ

at x D 0

λ D µ

at x D µ

In the corresponding bifurcation diagram shown in Figure 4.3, the ﬁxed point x D 0 is a nonhyperbolic ﬁxed point at µ D 0. At this point, a static bifurcation occurs because there is an exchange of stability between the trivial and nontrivial branches. The bifurcation point in Figure 4.3 is a transcritical bifurcation point. We point out that all of the branches that meet at this bifurcation point do not have the same tangent. Equation 4.75 is the normal form for a generic transcritical bifurcation of a ﬁxed point of a continuous system.

0.2 0.1 x 0.0 0.1 0.2

0.2

0.1

0.0 μ

0.1

0.2

Figure 4.3 Scenario in the vicinity of a transcritical bifurcation.

4.4 Local Bifurcations of Fixed Points

4.4.4 Pitchfork Bifurcation

We consider the case f µ D 0 and f x x D 0. As µ ! 0 and x ! 0, (4.63) tends to xP D f x µ x µ C

1 6

f xxx x3 C

(4.76)

whose ﬁxed points are s x1 D 0 ,

x2 D

6 f x µ µ , fxxx

s and

x3 D

6 f x µ µ fxxx

The second and third ﬁxed points exist only when f x µ f x x x µ < 0. The Jacobian matrix in this case f x D f µx µ C

1 2

f xxx x2

has the single eigenvalue λ(x1 ) D f µ x µ ,

λ(x2 ) D 2 f µ x µ ,

λ(x3 ) D 2 f µ x µ

Consequently, the trivial ﬁxed point is stable when f µ x µ < 0 and unstable when f µ x µ > 0. On the other hand, the nontrivial ﬁxed points exist and are stable when f x x x < 0 and f µ x µ > 0 and exist and are unstable when f x x x > 0 and f µ x µ < 0. Example 4.9 We let f µ µ D 0, f µ x D 1, and f x x x /6 D α D ˙1 and consider the system xP D f (xI µ) D µ x C α x 3

(4.77)

where µ is again the scalar control parameter. There are three ﬁxed points: x D 0 I trivial ﬁxed point p x D ˙ µ/α I nontrivial ﬁxed points In this case, the Jacobian matrix f x D µ C 3α x 2 has the single eigenvalue λDµ

at x D 0

λ D 2µ

p at x D ˙ µ/α

Consequently, the trivial ﬁxed point is stable when µ < 0 and unstable when µ > 0. On the other hand, when α < 0, nontrivial ﬁxed points exist only when µ > 0 and they are stable. However, when α > 0, nontrivial ﬁxed points exist only when

113

114

4 Bifurcations of Continuous Systems 0.4

0.4

0.2

0.2

x 0.0

x 0.0

–0.2

–0.2

–0.4 –0.2

–0.1

(a)

0.0 μ

0.1

0.2

–0.4 –0.2

(b)

–0.1

0.0 μ

0.1

0.2

Figure 4.4 Scenario in the vicinity of a pitchfork bifurcation: (a) supercritical pitchfork bifurcation (α D 1) and (b) subcritical pitchfork bifurcation (α D 1).

µ < 0 and they are unstable. The bifurcation diagrams of Figure 4.4a,b correspond to α D 1 and α D 1, respectively. In both cases, we note the following at (0, 0): (a) f (x, µ) D 0, (b) f x has a zero eigenvalue, (c) the number of ﬁxed-point solutions for µ < 0 is different from that for µ > 0, and (d) there is a change in the stability of the trivial ﬁxed point as we pass through µ D 0. Hence, the origin of the statecontrol space is a bifurcation point. p p When α D 1, two stable branches of ﬁxed points x D µ and x D µ bifurcate from the bifurcation point, as shown in Figure 4.4a. When α D 1, two p p unstable branches of ﬁxed points x D µ and x D µ bifurcate from the bifurcation point, as shown in Figure 4.4b. The bifurcations observed in Figure 4.4a,b are called pitchfork bifurcations because the bifurcating nontrivial branches have the geometry of a pitchfork at (0, 0). Speciﬁcally, the bifurcation in Figure 4.4a is called a supercritical pitchfork bifurcation, and the bifurcation in Figure 4.4b is called a subcritical or reverse pitchfork bifurcation. In the case of a supercritical pitchfork bifurcation, locally we have a branch of stable ﬁxed points on one side of the bifurcation point and two branches of stable ﬁxed points and a branch of unstable ﬁxed points on the other side of the bifurcation point. In the case of a subcritical pitchfork bifurcation, locally we have two branches of unstable ﬁxed points and a branch of stable ﬁxed points on one side of the bifurcation point and a branch of unstable ﬁxed points on the other side of the bifurcation point. Equation 4.77 is the normal form for a generic pitchfork bifurcation of a ﬁxed point of a continuous system.

4.4.5 Hopf Bifurcations

When a scalar control parameter µ is varied, a Hopf bifurcation of a ﬁxed point of a system such as (4.17) is said to occur at µ D µ c if the following conditions (Marsden and McCracken, 1976) are satisﬁed:

4.4 Local Bifurcations of Fixed Points

1. F(x I µ c ) D 0, 2. The matrix D x F has a pair of purely imaginary eigenvalues ˙i ω while all of its other eigenvalues have nonzero negative real parts at (x I µ c ), 3. For µ ' µ c , let the analytic continuation of the pair of purely imaginary eigenvalues be λ r ˙ i ω. Then, Real(d λ r /d µ) ¤ 0 at µ D µ c . This condition implies a transversal or nonzero speed crossing of the imaginary axis and hence is called a transversality condition. Again, the ﬁrst two conditions imply that the ﬁxed point undergoing the bifurcation is a nonhyperbolic ﬁxed point. When all of the above three conditions are satisﬁed, a periodic solution of period 2π/ω is born at (x I µ c ); bifurcating periodic solutions can also occur when the transversality condition is not satisﬁed (e.g., Marsden and McCracken, 1976). It is to be noted that bifurcating periodic solutions can also occur under certain other degenerate conditions (e.g., Golubitsky and Schaeffer, 1985). In such cases, we have degenerate Hopf bifurcations. The Hopf bifurcation is also called the Poincaré–Andronov–Hopf bifurcation (e.g., Wiggins, 1990) to give credit to the works of Poincaré and Andronov that preceded the work of Hopf. As pointed out in the literature (e.g., Abed, 1994; Arnold, 1988), Poincaré (1899) was aware of the conditions for this bifurcation to occur. (Poincaré studied such bifurcations in the context of lunar orbital dynamics.) Andronov and his coworkers studied Hopf bifurcations in planar systems before Hopf studied such bifurcations in general n-dimensional systems (Andronov et al., 1966; Arnold, 1988). In aeroelasticity, the consequence of a Hopf bifurcation is known as galloping or ﬂutter. Example 4.10 We consider the planar system xP D µ x ω y C (α x β y )(x 2 C y 2 )

(4.78)

yP D ωx C µ y C (β x C α y )(x 2 C y 2 )

(4.79)

where x and y are the states and µ is the scalar control parameter. The ﬁxed point (0, 0) is a solution of (4.78) and (4.79) for all values of µ. The eigenvalues of the corresponding Jacobian matrix there are λ1 D µ i ω

and

λ2 D µ C i ω

From these eigenvalues, we note that (0, 0) is a nonhyperbolic ﬁxed point of (4.78) and (4.79) when µ D 0. Further, at (x, y, µ) D (0, 0, 0), we note that d λ1 D 1 and dµ

d λ2 D1 dµ

115

116

4 Bifurcations of Continuous Systems

Hence, the three conditions required for a Hopf bifurcation are satisﬁed, and a Hopf bifurcation of the ﬁxed point (0, 0) of (4.78) and (4.79) occurs at µ D 0. The period of the bifurcating periodic solution at (0, 0, 0) is 2π/ω. By using the transformation x D r cos θ

and

y D r sin θ

we transform (4.78) and (4.79) into rP D µ r C α r 3

(4.80)

θP D ω C β r 2

(4.81)

The trivial ﬁxed point of (4.80) corresponds to the ﬁxed point (0, 0) of (4.78) and (4.79), and a nontrivial ﬁxed point (i.e., r ¤ 0) of (4.80) corresponds to a periodic solution of (4.78) and (4.79). In the latter case, r is the amplitude and θP is the frequency of the periodic solution that is created due to the Hopf bifurcation. A stable nontrivial ﬁxed point of (4.80) corresponds to a stable periodic solution of (4.78) and (4.79). Likewise, an unstable nontrivial ﬁxed point of (4.80) corresponds to an unstable periodic solution of (4.78) and (4.79). We note that (4.80) is identical to (4.77), so the Hopf bifurcation at (0, 0, 0) in the x y µ space is equivalent to a pitchfork bifurcation at (0, 0) in the r µ space. When α D 1, we have a supercritical pitchfork bifurcation in the r µ space and, hence, a supercritical Hopf bifurcation in the x y µ space. When α D 1, we have a subcritical pitchfork bifurcation in the r µ space and, hence, a subcritical Hopf bifurcation in the x y µ space. The bifurcation diagrams for α D 1 are shown in Figure 4.5a,c and α D 1 are shown in Figure 4.5b,d, respectively. In y

y x

x

0

μ

(a)

μ

(b)

r 0

(c)

0

μ

(d)

0

μ

Figure 4.5 Local scenarios: (a,c) supercritical Hopf bifurcation and (b,d) subcritical Hopf bifurcation.

4.5 Normal Forms of Static Bifurcations

Figure 4.5a,b, the bifurcating periodic solutions in the x y µ space are depicted as parabolic surfaces. In the case of a supercritical Hopf bifurcation, locally we have a branch of stable ﬁxed points on one side of the bifurcation point and a branch of unstable ﬁxed points and a branch of stable periodic solutions on the other side of the bifurcation point. In the case of a subcritical Hopf bifurcation, locally we have a branch of unstable periodic solutions and a branch of stable ﬁxed points on one side of the bifurcation point and a branch of unstable ﬁxed points on the other side of the bifurcation point. When α ¤ 0, (4.78) and (4.79) are the normal form for a generic Hopf bifurcation of a ﬁxed point of a continuous system. On the other hand, when α D 0 in (4.78) and (4.79), although the conditions for a Hopf bifurcation are satisﬁed, there are no periodic orbits in the vicinity of the bifurcation point to third order. This case is degenerate.

4.5 Normal Forms of Static Bifurcations

In this section, we consider reduction of the nonlinear continuous system (4.17) near a static bifurcation ﬁxed point to its normal form. We assume that the ﬁxed point has been shifted to x D 0 and its corresponding control parameter has been shifted to µ D 0. Moreover, we expand (4.17) in a Taylor series for small x and µ and obtain xP D Ax C b µ C B µ x C Q(x, x) C C (x, x, x) C

(4.82)

where A D D x F, B D D µ x F, and b D D µ F at (x, µ) D (0, µ) and Q(x, x) and C (x, x, x) are bilinear and trilinear column vectors involving quadratic and cubic terms, respectively. Because the ﬁxed point (x, µ) D (0, 0) is a static bifurcation point, one of the eigenvalues of the matrix A is zero and all of its other eigenvalues are in the left-half of the complex plane. Next, we demonstrate how one can use the method of multiple scales, a combination of center-manifold reduction and the method of normal forms, and a projection method to compute the normal forms of saddle-node, transcritical, and pitchfork bifurcations. 4.5.1 The Method of Multiple Scales

To compute the normal form of the static bifurcation of (4.82) at the origin, we introduce a small nondimensional parameter as a bookkeeping parameter and seek a third-order approximate solution of (4.82) in the form x(tI µ) D x 1 (T0 , T1 , T2 ) C 2 x 2 (T0 , T1 , T2 ) C 3 x 3 (T0 , T1 , T2 ) C

(4.83)

where the time scales Tm D m t. The time derivative can be expressed in terms of these scales as d (4.84) D D0 C D1 C 2 D2 C dt

117

118

4 Bifurcations of Continuous Systems

where D m D @/@Tm . Moreover, because µ is small, we expand it in terms of as µ D µ 1 C 2 µ 2 C 3 µ 3 C

(4.85)

Substituting (4.83)–(4.85) into (4.82) and equating coefﬁcients of like powers of , we obtain Order ()

D0 x 1 Ax 1 D bµ 1

(4.86)

Order (2 )

D0 x 2 Ax 2 D bµ 2 D1 x 1 C B x 1 µ 1 C Q(x 1 , x 1 )

(4.87)

Order (3 )

D0 x 3 Ax 3 D b µ 3 D1 x 2 D2 x 1 C B x 2 µ 1 C B x 1 µ 2 C 2Q(x 1 , x 2 ) C C (x 1 , x 1 , x 1 )

(4.88)

The general solution of (4.86) is the superposition of a homogeneous solution x 1h and a particular solution x 1p . For the homogeneous solution, we denote the eigenvalues of the matrix A by λ m , m D 1, 2, . . . , n and order them so that λ 1 D 0. We let p m be the eigenvector (generalized eigenvector) corresponding to λ m . Then, all of the terms corresponding to λ m and p m for m > 2 decay with time so that the long-time dynamics is given by x 1h D u(T1 , T2 )p

(4.89)

where the subscript in p 1 has been suppressed and the function u(T1 , T2 ) is determined by imposing solvability conditions at the higher-order approximations. We note that A is singular because one of its eigenvalues is zero. Because the term bµ 1 on the right-hand side of (4.86) is independent of T0 , the particular solution of (4.86) is given by Ax 1p D b µ 1

(4.90)

Because A is singular, (4.90) has a solution if and only if its right-hand side bµ 1 is orthogonal to every solution of the adjoint homogeneous equation; that is solutions of qT A D 0 which has a nontrivial solution because A is singular. This condition demands that q T b µ 1 D 0. There are two possibilities. First, q T b ¤ 0 and hence µ 1 D 0. Second, q T b D 0 and hence µ 1 is arbitrary. In the latter case, the solution is not unique. To make it unique, we require it to be orthogonal to the left eigenvector q and write the particular solution as x 1p D c µ 1

where

Ac D b

(4.91)

and q T c D 1. We consider these two cases separately, starting with the ﬁrst case.

4.5 Normal Forms of Static Bifurcations

The Case q T b ¤ 0 In this case, the solution of the ﬁrst-order problem, (4.86), can be expressed as

x 1 D u(T1 , T2 )p

(4.92)

Substituting (4.92) into (4.87) yields D0 x 2 Ax 2 D b µ 2 D1 u p C Q(p , p )u2

(4.93)

Again, the general solution of (4.93) is the superposition of a homogeneous term and a particular term. The homogeneous solution is the same as that given in (4.89) and hence is neglected. A particular solution of (4.93) is given by Ax 2 D bµ 2 C D1 u p Q(p , p )u2

(4.94)

Equation 4.94 has a solution if and only if its right-hand side is orthogonal to q; that is, D1 u D q T bµ 2 C q T Q(p , p )u2

(4.95)

Hence, there is a generic saddle-node bifurcation at the origin of (4.82) if q T Q(p, p ) ¤ 0. When q T Q(p , p ) D 0, it follows from (4.95) that µ 2 D 0 and D1 u D 0 or u D u(T2 ). Then, a unique particular solution of (4.94) is given by x 2 D z u2

where

Az D Q(p , p )

and

qT z D 1

(4.96)

Substituting (4.92) and (4.96) into (4.88) and using the fact that µ 1 D 0, µ 2 D 0, and D1 u D 0, we obtain D0 x 3 Ax 3 D b µ 3 D2 u p C 2Q(z, p ) C C (p , p , p ) u3

(4.97)

Equation 4.97 has a solution if and only if its right-hand side is orthogonal to q; that is, D2 u D q T bµ 3 C 2q T Q(z, p ) C q T C (p , p , p ) u3

(4.98)

Therefore, the origin of (4.82) is an inﬂection point and there is no bifurcation there if the coefﬁcient of u3 in (4.98) is different from zero. The Case q T b D 0

In this case, the general solution of (4.86) can be expressed as

x 1 D u(T1 , T2 )p C c µ 1

(4.99)

Substituting (4.99) into (4.87) yields D0 x 2 Ax 2 D b µ 2 p D1 u C B p C 2Q(c, p ) µ 1 u C B c C Q(c, c) µ 21 C Q(p, p )u2

(4.100)

119

120

4 Bifurcations of Continuous Systems

The solvability condition of (4.100) demands that its right-hand side be orthogonal to q, which yields D1 u D q T B c C q T Q(c, c) µ 21 C q T B p C 2q T Q(c, p ) µ 1 u C q T Q(p , p )u2 (4.101) Therefore, there is a generic transcritical bifurcation of (4.82) at the origin when q T Q(p , p ) ¤ 0. When q T Q(p, p ) D 0, it follows from (4.101) that µ 1 D 0 and D1 u D 0 and hence u D u(T2 ). Then, the solution of (4.100) can be expressed as x 2 D c µ 2 C z u2

(4.102)

where c and z are deﬁned in (4.91) and (4.96), respectively. Then, substituting (4.99) and (4.102) into (4.88) and using the fact that µ 1 D 0 and D1 u D 0, we obtain D0 x 3 Ax 3 D b µ 3 D2 u p C B p C 2Q(c, p ) µ 2 u (4.103) C 2Q(z, p ) C C (p, p , p ) u3 Imposing the solvability condition in (4.103) yields D2 u D q T B p C 2q T Q(c, p ) µ 2 u C 2q T Q(z, p ) C q T C (p , p , p ) u3 (4.104) Therefore, there is a generic supercritical pitchfork bifurcation at the origin of (4.82) when 2q T Q(z, p ) C q T C(p , p , p ) < 0 and a generic subcritical pitchfork bifurcation at the origin of (4.82) when 2q T Q(z, p ) C q T C (p , p , p ) > 0. Example 4.11 We construct the normal form of the system xP1 D b 1 µ C 2x1 2x2 C µ x1 C a 11 x12 C a 12 x1 x2 C a 13 x22

(4.105)

xP2 D b 2 µ C 4x1 4x2 4µ x2 C a 21 x12 C a 22 x1 x2 C a 23 x22

(4.106)

near the ﬁxed point (x1 , x2 , µ) D (0, 0, 0). We seek a third-order approximate solution of (4.105) and (4.106) in the form x n (tI µ) D x n1 (T0 , T1 , T2 ) C 2 x n2 (T0 , T1 , T2 ) C 3 x n3 (T0 , T1 , T2 ) C (4.107) Substituting (4.107), (4.84), and (4.85) into (4.105) and (4.106) and equating coefﬁcients of equal powers of , we obtain

4.5 Normal Forms of Static Bifurcations

Order ()

D0 x11 2x11 C 2x21 D µ 1 b 1

(4.108)

D0 x21 4x11 C 4x21 D µ 1 b 2

(4.109)

Order (2 ) 2 D0 x12 2x12 C 2x22 D µ 2 b 1 D1 x11 C µ 1 x11 C a 11 x11 2 C a 12 x11 x21 C a 13 x21

(4.110)

2 D0 x22 4x12 C 4x22 D µ 2 b 2 D1 x21 4µ 1 x21 C a 21 x11 2 C a 22 x11 x21 C a 23 x21

(4.111)

Order (3 )

D0 x13 2x13 C 2x23 D µ 3 b 1 D1 x12 D2 x11 C µ 1 x12 C µ 2 x11 C 2a 11 x11 x12 C a 12 x11 x22 C a 12 x12 x21 C 2a 13 x21 x22

(4.112)

D0 x23 4x13 C 4x23 D µ 3 b 2 D1 x23 D2 x21 4µ 1 x23 4µ 2 x21 C 2a 21 x11 x13 C a 23 x11 x23 C a 23 x13 x21 C 2a 23 x21 x23

(4.113)

In this case, the coefﬁcient matrix of (4.108) and (4.109) is AD

2 4

2 4

(4.114)

whose eigenvalues are λ D 0 and λ D 2. Hence, the origin of (4.105) and (4.106) may be a static bifurcation point. The right and left eigenvectors of A corresponding to λ D 0 are 1 2 pD and q D 1 1 where q T p D 1. Because A is singular, (4.108) and (4.109) have a solution if and only if their right-hand sides are orthogonal to q T ; that is, if and only if (2b 1 b 2 )µ 1 D 0. There are two possibilities: (a) (2b 1 b 2 ) ¤ 0 and hence µ 1 D 0 and (a) (2b 1 b 2 ) D 0 and hence µ 1 is arbitrary. We treat both cases separately, starting with the ﬁrst case. When (2b 1 b 2 ) ¤ 0, the solution of (4.108) and (4.109) can be expressed as x11 D u(T1 , T2 )

and

x21 D u(T1 , T2 )

(4.115)

121

122

4 Bifurcations of Continuous Systems

Substituting (4.115) into (4.110) and (4.111) yields D0 x12 2x12 C 2x22 D µ 2 b 1 D1 u C (a 11 C a 12 C a 13 )u2

(4.116)

D0 x22 4x12 C 4x22 D µ 2 b 2 D1 u C (a 21 C a 22 C a 23 )u2

(4.117)

Demanding that the right-hand side of (4.116) and (4.117) be orthogonal to q yields the normal form D1 u D (2b 1 b 2 )µ 2 C (2a 11 C 2a 12 C 2a 13 a 21 a 22 a 23 )u2

(4.118)

which indicates that the origin of (4.105) and (4.106) undergoes a generic saddlenode bifurcation as µ passes through zero if the coefﬁcient of u2 is different from zero. When this coefﬁcient is zero, one needs to carry out the expansion to higher order. When (2b 1 b 2 ) D 0, the solution of (4.108) and (4.109) that is orthogonal to q can be expressed as x11 D u(T1 , T2 ) C 12 b 1 µ 1 and x21 D u(T1 , T2 ) C b 1 µ 1

(4.119)

Substituting (4.119) into (4.110) and (4.111) yields D0 x12 2x12 C 2x22 D µ 2 b 1 D1 u C 12 C 14 a 11 b 1 C 12 a 12 b 1 C a 13 b 1 b 1 µ 21 C 1 C a 11 b 1 C 32 a 12 b 1 C 2a 13 b 1 µ 1 u C (a 11 C a 12 C a 13 ) u2

(4.120)

D0 x22 4x12 C 4x22 D µ 2 b 2 D1 u C 4 C 14 a 21 b 1 C 12 a 22 b 1 C a 23 b 1 b 1 µ 21 C 4 C a 21 b 1 C 32 a 22 b 1 C 2a 23 b 1 µ 1 u C (a 21 C a 22 C a 23 )u2 (4.121) The solvability condition of (4.120) and (4.121) yields the following normal form of (4.105) and (4.106): D1 u D Γ0 µ 21 C Γ1 µ 1 u C Γ2 u2

(4.122)

where Γ0 D 14 b 1 (20 C 2a 11 b 1 C 4a 12 b 1 C 8a 13 b 1 a 21 b 1 2a 22 b 1 4a 23 b 1 ) (4.123) Γ1 D

1 2

(12 C 4a 11 b 1 C 6a 12 b 1 C 8a 13 b 1 2a 21 b 1 3a 22 b 1 4a 23 b 1 ) (4.124)

Γ2 D 2a 11 C 2a 12 C 2a 13 a 21 a 22 a 23

(4.125)

Equation 4.122 indicates that the origin of (4.105) and (4.106) undergoes a generic transcritical bifurcation as µ passes through zero if Γ2 ¤ 0.

4.5 Normal Forms of Static Bifurcations

When Γ2 D 0, it follows from (4.122) that µ 1 D 0 and D1 u D 0 or u D u(T2 ). Then, the particular solution of (4.120) and (4.121) that is orthogonal to q T can be expressed as x12 D 12 b 1 µ 2 C 12 (a 11 C a 12 C a 13 )u2

and

x22 D b 1 µ 2 C(a 11 C a 12 C a 13 )u2 (4.126)

Substituting (4.119) and (4.126) into (4.112) and (4.113) and recalling that µ 1 D 0 and D1 u D 0, we obtain D0 x13 2x13 C 2x23 D µ 3 b 1 D2 u C C

1 4

1 2

(2 C 2a 11 b 1 C 3a 12 b 1 C 4a 13 b 1 ) µ 2 u

(2a 11 C 3a 12 C 4a 13 ) (a 21 C a 22 C a 23 ) u3

(4.127)

D0 x23 4x13 C 4x23 D µ 3 b 2 D2 u C

1 4

1 2

(8 2a 21 b 1 3a 23 b 1 4a 23 b 1 ) µ 2 u

(a 21 C a 23 C a 23 ) (2a 21 C 3a 23 C 4a 23 ) u3

(4.128)

Imposing the solvability condition for (4.127) and (4.128) yields D2 u D νµ 2 u C α u3

(4.129)

where νD αD

1 2 1 4

(12 C 4a 11 b 1 C 6a 12 b 1 C 8a 13 b 1 2a 21 b 1 3a 22 b 1 4a 23 b 1 ) (4a 11 C 6a 12 C 8a 13 2a 21 3a 22 4a 23 ) (a 21 C a 22 C a 23 )

It follows from (4.139) that, as µ passes through zero, the origin of (4.105) and (4.106) undergoes a generic supercritical pitchfork bifurcation when α < 0 and a generic subcritical pitchfork bifurcation when α > 0. Example 4.12 We consider the three-dimensional dynamic system xP 1 D b 1 µ x1 C x2 2x3 C µ x1 C α 1 (x1 C 2x2 C x3 )2

(4.130)

xP 2 D b 2 µ 2x1 x2 x3 C µ x2 C α 2 (x1 x2 C x3 )2

(4.131)

xP 3 D b 3 µ C x1 C 2x2 x3 2µ x3 C α 3 (2x1 C x2 x3 )2

(4.132)

Substituting (4.107), (4.84), and (4.85) into (4.130)–(4.132) and equating coefﬁcients of like powers of yields.

123

124

4 Bifurcations of Continuous Systems

Order ()

D0 x11 C x11 x21 C 2x31 D b 1 µ 1

(4.133)

D0 x21 C 2x11 C x21 C x31 D b 2 µ 1

(4.134)

D0 x31 x11 2x21 C x31 D b 3 µ 1

(4.135)

Order (2 )

D0 x12 C x12 x22 C 2x32 D D1 x11 C b 1 µ 2 C µ 1 x11 C α 1 (x11 C 2x21 C x31 )2

(4.136)

D0 x22 C 2x12 C x22 C x32 D D1 x21 C b 2 µ 2 C µ 1 x21 C α 2 (x11 x21 C x31 )2

(4.137)

D0 x32 x12 2x22 C x32 D D1 x31 C b 3 µ 2 2µ 1 x31 C α 3 (2x11 C x21 x31 )2

(4.138)

Order (3 )

D0 x13 C x13 x23 C 2x33 D D1 x12 D2 x11 C b 1 µ 3 C µ 2 x11 C µ 1 x12 C 2α 1 (x11 C 2x21 C x31 ) (x12 C 2x22 C x32 ) (4.139) D0 x23 C 2x13 C x23 C x33 D D1 x22 D2 x21 C b 2 µ 3 C µ 2 x21 C µ 1 x22 C 2α 2 (x11 x21 C x31 ) (x12 x22 C x32 ) (4.140) D0 x33 x13 2x23 C x33 D D1 x32 D2 x31 C b 3 µ 3 2µ 2 x31 2µ 1 x32 C 2α 3 (2x11 C x21 x31 ) (2x12 C x22 x32 ) (4.141) In this case, the coefﬁcient matrix of (4.133)–(4.135) is 1 0 1 1 2 @2 1 1A 1

2

(4.142)

1

p whose eigenvalues are λ D 0 and λ D 3/2(1˙ i 3). Hence, the origin of (4.130)– (4.133) may be a static bifurcation point. The right and left eigenvectors of A corresponding to λ D 0 are 2 3 2 3 1 1 1 p D 4 1 5 and q D 4 1 5 3 1 1

4.5 Normal Forms of Static Bifurcations

where q T p D 1. Because A is singular, (4.133)–(4.135) have a solution if and only if their right-hand sides are orthogonal to q T ; that is, if and only if (b 3 Cb 2 b 1 )µ 1 D 0. There are two possibilities: (a) (b 3 C b 2 b 1 ) ¤ 0 and hence µ 1 D 0 and (a) (b 3 C b 2 b 1) D 0 and hence µ 1 is arbitrary. We treat both cases separately, starting with the ﬁrst case. When (b 3 C b 2 b 1 ) ¤ 0, the nondecaying solution of (4.133)–(4.135) can be expressed as x11 D u(T1 , T2 ) ,

x21 D u(T1 , T2 ) ,

and

x31 D u(T1 , T2 )

(4.143)

Substituting (4.143) into (4.136)–(4.138) yields D0 x12 C x12 x22 C 2x32 D D1 u C 4u2 α 1 C b 1 µ 2

(4.144)

D0 x22 C 2x12 C x22 C x32 D D1 u C u2 α 2 C b 2 µ 2

(4.145)

D0 x32 x12 2x22 C x32 D D1 u C 4u2 α 3 C b 3 µ 2

(4.146)

Demanding that the right-hand side of (4.144)–(4.146) be orthogonal to q yields the normal form D1 u D

1 3

(b 3 C b 2 b 1 ) µ 2 C

1 3

(4α 3 C α 2 4α 1 ) u2

(4.147)

which indicates that the origin of (4.130)–(4.132) undergoes a generic saddle-node bifurcation as µ passes through zero if the coefﬁcient of u2 is different from zero. When this coefﬁcient is zero, one needs to carry out the expansion to higher order. When (b 3 C b 2 b 1 ) D 0, the solution of (4.133)–(4.135) that is orthogonal to q can be expressed as x11 D u(T1 , T2 ) C 13 b 2 µ 1 x21 D u(T1 , T2 ) 13 b 3 µ 1 x31 D u(T1 , T2 ) C 13 (b 3 C b 2 )µ 1

(4.148)

To simplify the algebra, we let b 1 D 2b, b 2 D b, and b 3 D b. Substituting (4.148) into (4.136)–(4.138) yields D0 x12 C x12 x22 C 2x32 D D1 u C 19 b (3 C b α 1 ) µ 21 C 2b µ 2 13 (3 4b α 1 ) µ 1 u C 4α 1 u2

(4.149)

D0 x22 C 2x12 C x22 C x32 D D1 u 19 b (3 16b α 2 ) µ 21 C b µ 2 C

1 3

(3 8b α 2 ) µ 1 u C α 2 u2

(4.150)

D0 x32 x12 2x22 C x32 D D1 u 19 b (12 b α 3 ) µ 21 C b µ 2 23 (3 2b α 3 ) µ 1 u C 4α 3 u2

(4.151)

The solvability condition of (4.149)–(4.151) yields the normal form of (4.130)– (4.132) as D1 u D Γ0 µ 21 C Γ1 µ 1 u C Γ2 u2

(4.152)

125

126

4 Bifurcations of Continuous Systems

where 1 Γ0 D 27 b (18 C b α 1 16b α 2 b α 3 )

Γ1 D 49 b (α 3 2α 2 α 1 ) Γ2 D

1 3

(4α 3 C α 2 4α 1 )

Equation 4.152 indicates that the origin of (4.130)–(4.132) undergoes a generic transcritical bifurcation as µ passes through zero if the coefﬁcient of u2 is different from zero; that is, (4α 3 C α 2 4α 1 ) ¤ 0. When Γ2 D 0, it follows from (4.152) that µ 1 D 0 and D1 u D 0 or u D u(T2 ). Then, the particular solution of (4.144)–(4.146) that is orthogonal to q T can be expressed as x12 D 13 b µ 2 C 13 α 2 u2 x22 D 31 b µ 2 43 α 3 u2 x32 D 23 b µ 2 C 13 α 2 C 43 α 3 u2

(4.153)

Substituting (4.148) and (4.153) into (4.139)–(4.141) and recalling that µ 1 D 0 and D1 u D 0, we obtain D0 x13 C x13 x23 C 2x33 D D2 u C 2b µ 3 C

1 3

(4b α 3 C b α 2 3) µ 2 u (4.154)

C 23 (α 2 2α 3 )(α 2 C 4α 3 )u3

D0 x23 C 2x13 C x23 C x33 D D2 u C b µ 3 13 (8b α 2 3) µ 2 u 43 α 2 (α 2 C 4α 3 ) u3 D0 x32 x12 2x22 C x32 D D2 u C b µ 3 C

2 3

(4.155)

(2b α 3 3) µ 2 u

C (8α 3 α 2 ) α 3 u3 4 3

(4.156)

Demanding that the right-hand side of (4.154)–(4.156) be orthogonal to q and using the fact that α 1 D 1/4α 2 C α 3 yields the normal form (4.157) D2 u D b α 2 µ 2 u C 13 16α 23 8α 2 α 3 2α 22 u3 It follows from (4.157) that the origin of (4.130)–(4.132) undergoes a supercritical pitchfork bifurcation as µ passes through zero when 16α 23 8α 2 α 3 2α 22 < 0 and a subcritical pitchfork bifurcation when (16α 23 8α 2 α 3 2α 22 ) > 0. 4.5.2 Center-Manifold Reduction

To compute the normal form of the static bifurcation of (4.82) at the origin by using center-manifold reduction, we ﬁrst calculate the eigenvalues λ 1 , λ 2 , . . . , λ n

4.5 Normal Forms of Static Bifurcations

and eigenvectors (generalized eigenvectors) p 1 , p 2 , . . . , p n of the matrix A. Second, we arrange the eigenvalues of A so that λ 1 D 0 and let p 1 be be its corresponding eigenvector. Third, we introduce the transformation x D P y and obtain yP D J y C P 1 b µ C P 1 B P y µ C P 1 Q(P y, P y) C P 1 C(P y, P y, P y) (4.158) where J D P 1 AP . We note that J can be rewritten as JD

0 0

0 Js

(4.159)

where J s is an (n 1) (n 1) matrix whose eigenvalues are λ 2 , λ 3 , . . . , λ n . Fourth, we let y s be the (n 1)-dimensional vector with the components y 2 , y 3 , . . . , y n and rewrite (4.158) as yP 1 D bO 1 µ C

n X

ξi y i µ C f 2 (y 1 , y s ) C f 3 (y 1 , y s )

(4.160)

iD1

and yP s D J s y s C bO s µ C ζ y 1 µ C BO y µ C F 2 (y 1 , y s ) C F 3 (y 1 , y s )

(4.161)

where bO D P 1 b, f 2 and F 2 are the ﬁrst and last (n 1) components of P 1 Q(P y, P y), and f 3 and F 3 are the ﬁrst and last (n 1) components of P 1 C (P y, P y, P y). To determine the dependence of the center manifold on µ, we augment (4.160) and (4.161) with the additional equation µP D 0

(4.162)

Because the f i and F i are polynomials and hence inﬁnitely differentiable, there exists a local center manifold of the form (Carr, 1981) y s D h(y 1 I µ)

(4.163)

where h is a polynomial function of y 1 and µ such that h(0, 0) D 0,

D y 1 h i (0I 0) D 0 ,

and

D µ h i (0I 0) D 0

(4.164)

where the h i are the scalar components of h. Substituting (4.163) into (4.161) yields " # n X 0 O b h (y 1 ) 1 µ C ξ1 y 1 µ C ξi h i (y 1 )µ C f 2 (y 1 , h(y 1 ) iD2

D J s h(y 1 ) C bO s µ C ζ y 1 µ C BO h(y 1 )µ C F 2 [y 1 , h(y 1 )] C

(4.165)

127

128

4 Bifurcations of Continuous Systems

To solve (4.165), one approximates the components of h(y 1 I µ) with polynomials. The polynomial approximations are usually taken to be quadratic to the ﬁrst approximation and do not contain constant and linear terms so that the conditions (4.164) are satisﬁed. Substituting the assumed quadratic polynomial approximations into (4.165) and equating the coefﬁcients of the different terms in the polynomials on both sides, one obtains a system of algebraic equations for the coefﬁcients of the polynomials. Solving these equations, we obtain a ﬁrst approximation to the center manifold y s D h(y 1 I µ). Finally, substituting this approximation into (4.160), we obtain the one-dimensional dynamical system yP1 D bO 1 µ C ξ1 y 1 µ C

n X

ξi y i µ C f 2 [y 1 , h(y 1 )] C f 3 [y 1 , h(y 1 )]

(4.166)

iD2

describing the dynamics of the system (4.82) on the center manifold. Next, we consider two examples. Example 4.13 To reduce the algebra, we consider a special case of Example 4.11, namely xP1 D b 1 µ C 2x1 2x2 C µ x1 C α 1 x12

(4.167)

xP2 D b 2 µ C 4x1 4x2 4µ x2 C α 2 x1 x2

(4.168)

The coefﬁcient matrix of this system at (0, 0, 0) is given by (4.114). Its eigenvalues are λ 1 D 0 and λ 2 D 2 and their corresponding eigenvectors are p1 D

1 1

and

p2 D

1 2

Hence, PD

1 1

1 2

and

P 1 D

2 1

1 1

Next, we introduce the transformation x1 1 1 y1 D 1 2 y2 x2 into (4.167) and (4.168) and obtain yP1 D (2b 1 b 2 )µ C 6y 1 µ C 10y 2 µ C (2α 1 α 2 )y 12 C (4α 1 3α 2 )y 1 y 2 C 2(α 1 α 2 )y 22

(4.169)

yP2 D 2y 2 C (b 2 b 1 )µ 5y 1 µ 9y 2 µ (α 1 α 2 )y 12 (2α 1 3α 2 )y 1 y 2 (α 1 2α 2 )y 22

(4.170)

4.5 Normal Forms of Static Bifurcations

We seek the center manifold of (4.169) and (4.170) in the form y 2 (tI µ) D h(y 1I µ) and rewrite (4.169) as yP1 D (2b 1 b 2 )µ C 6y 1 µ C 10h µ C (2α 1 α 2 )y 12 C (4α 1 3α 2 )y 1 h C 2(α 1 α 2 )h 2

(4.171)

Substituting (4.171) and the expression for the center manifold into (4.170) yields h 0 (2b 1 b 2 )µ C 6y 1 µ C 10h µ C (2α 1 α 2 )y 12 D 2h C (b 2 b 1 )µ 5y 1 µ 9h µ (α 1 α 2 )y 12 (2α 1 3α 2 )y 1 h (α 1 2α 2 )h 2 C (4.172) whose approximate solution can be expressed as h D 12 (b 2 b 1 )µ 12 (α 1 α 2 )y 12 C

(4.173)

Substituting (4.173) into (4.171) yields the following one-dimensional equation describing the dynamics on the center manifold: yP1 D (2b 1 b 2 )µ C 5(b 2 b 1 ) C 14 (α 1 α 2 )(b 2 b 1 )2 µ 2 C (2α 1 α 2 )y 12 C 6 C 12 (b 2 b 1 )(4α 1 3α 2 ) y 1 µ 12 (α 1 α 2 )(4α 1 3α 2 )y 13 (4.174) There are two possibilities: (a) b 2 ¤ 2b 1 and (b) b 2 D 2b 1 . In the ﬁrst case, as y 1 ! 0 and µ ! 0, (4.174) tends to yP1 D (2b 1 b 2 )µ C (2α 1 α 2 )y 12

(4.175)

Therefore, the origin of (4.167) and (4.168) undergoes a saddle-node bifurcation as µ passes through zero when α 2 ¤ 2α 1 . When α 2 D 2α 1 , the origin of (4.167) and (4.168) is not a bifurcation point. When b 2 D 2b 1 , as y 1 ! 0 and µ ! 0, (4.174) tends to 1 1 yP1 D 5b 1 C (α 1 α 2 )b 21 µ 2 C 6 C b 1 (4α 1 3α 2 ) y 1 µC(2α 1 α 2 )y 12 4 2 (4.176) and therefore the origin of (4.167) and (4.168) undergoes a transcritical bifurcation as µ passes through zero when α 2 ¤ 2α 1 . When α 2 D 2α 1 , as y 1 ! 0 and µ ! 0, (4.174) tends to yP1 D (6 α 1 b 1 ) y 1 µ α 21 y 13

(4.177)

and therefore the origin of (4.167) and (4.168) undergoes a supercritical pitchfork bifurcation as µ passes through zero.

129

130

4 Bifurcations of Continuous Systems

Example 4.14 We reconsider the system (4.130)–(4.132) discussed in Example 4.12. In this case, the coefﬁcient matrix of the linear at (0, 0, 0I 0) is given by (4.142). Its eigen system p

values are λ D 0 and λ D 3/2 1 ˙ i 3 and their corresponding eigenvectors can be expressed as 2 2 p 3 p 3 2 3 1 1 1Ci 3 1i 3 1 2 2 6 6 p 7 p 7 7 7 6 1 1 p1 D 4 1 5 p2 D 6 4 2 i i C 3 5 , and p 3 D 4 2 i i C 3 5 1 1 1 We note that p 3 D pN 2 . Then, 2 p

1 12 1 C i 3 6 p

1 P D6 i iC 3 41 2 1 1 and

2

P 1

1 3 p

61 1i 3 D6 46 p

1 1Ci 3 6

p 3 1i 3 p 7 7 12 i i C 3 5 1 2

1

1

3 p

16 i i C 3 p

1 i iC 3 6

3

1 3 17 37 5 1 3

Next, we introduce the transformation x D P y into (4.130)–(4.132) and after some algebraic manipulations obtain yP1 D 13 (b 1 b 2 b 3 ) µ 13 (4α 1 α 2 4α 3 ) y 12 µ(y 2 C yN2 ) h

p p 23 α 1 C 3i 3α 1 C 2α 2 α 3 C 3i 3α 3 y 1 y 2

p p 16 13α 1 3i 3α 1 C 8α 2 13α 3 3i 3α 3 y 22 C 13 (7α 1 4α 2 7α 3 ) y 2 yN2 C cc

(4.178)

p

p p yP2 D 32 1 i 3 y 2 C 16 b 1 i 3b 1 b 2 i 3b 2 C 2b 3 µ

p p (y 1 C yN2 )µ C 16 4α 1 4 i 3α 1 α 2 i 3α 2 C 8α 3 y 12

p p p C 23 5α 1 C i 3α 1 C α 2 C i 3α 2 C α 3 3i 3α 3 y 1 y 2

p p p C 23 4α 1 2i 3α 1 C α 2 C i 3α 2 C α 3 C 3i 3α 3 y 1 yN2

p p p C 16 i 2i α 1 C 8 3α 1 C 4 i α 2 4 3α 2 C 13i α 3 3 3α 3 y 22

p p C 13 7α 1 7i 3α 1 4α 2 4 i 3α 2 C 14α 3 y 2 yN 2

p p p C 16 i 11i α 1 C 5 3α 1 C 4 i α 2 4 3α 2 C 13i α 3 C 3 3α 3 yN22 (4.179)

4.5 Normal Forms of Static Bifurcations

where y 3 D yN2 . We note that the equation governing y 3 is the complex conjugate of (4.179). We seek the center manifold of (4.178) and (4.179) in the form y 2 (tI µ) D h(y 1 I µ), where h is a complex-valued function, and rewrite (4.178) as yP 1 D 31 (b 1 b 2 b 3 ) µ 13 (4α 1 α 2 4α 3 ) y 12 µ h µ hN h

p p 23 α 1 C 3i 3α 1 C 2α 2 α 3 C 3i 3α 3 y 1 h C 13 (7α 1 4α 2 7α 3 ) h hN i

p p 16 13α 1 3i 3α 1 C 8α 2 13α 3 3i 3α 3 h 2 C cc

(4.180)

Substituting y 2 (tI µ) D h(y 1I µ) into (4.179) and using (4.180), we obtain i h h 0 13 (b 1 b 2 b 3 ) µ 13 (4α 1 α 2 4α 3 ) y 12 µ h µ hN

p

p p D 32 1 i 3 h C 16 b 1 i 3b 1 b 2 i 3b 2 C 2b 3 µ

p p C 16 4α 1 4 i 3α 1 α 2 i 3α 2 C 8α 3 y 12 C

(4.181)

An approximate solution of (4.181) can be expressed as p p

1 h D 18 2b 1 C b 2 i 3b 2 C b 3 C i 3b 3 µ

p p 1 8α 1 C α 2 i 3α 2 C 4α 3 C 4 i 3α 3 y 12 C C 18

(4.182)

Substituting (4.182) into (4.180) yields the following one-dimensional equation describing the dynamics on the center manifold: yP 1 D 31 (b 1 b 2 b 3 ) µ 13 (4α 1 α 2 4α 3 ) y 12

4 27

[b 3 (4α 1 C α 2 5α 3 )

Cb 1 (α 1 C 2α 2 α 3 ) C b 2 (5α 1 C α 2 C 4α 3 )] µ y 1 C Γ0 µ 2 2 4 4α 1 C α 22 C α 1 (13α 2 20α 3 ) C 8α 2 α 3 20α 23 y 13 27

(4.183)

where Γ0 D

1 243

b 2 (27 C 8b 3 (5α 1 C α 2 5α 3 )) C b 21 (α 1 C 16α 2 C α 3 )

2b 1 (27 C b 2 (5α 1 8α 2 C 4α 3 ) b 3 (4α 1 C 8α 2 C 5α 3 )) Cb 22 (25α 1 C 4 (α 2 C 4α 3 )) Cb 3 (27 C b 3 (16α 1 C 4α 2 C 25α 3 ))

There are two possibilities: (a) b 1 ¤ b 2 C b 3 and (b) b 1 D b 2 C b 3 . In the ﬁrst case, as y 1 ! 0 and µ ! 0, (4.183) tends to yP 1 D 31 (b 1 b 2 b 3 ) µ 13 (4α 1 α 2 4α 3 ) y 12

(4.184)

131

132

4 Bifurcations of Continuous Systems

Therefore, the origin of (4.130)–(4.132) undergoes a saddle-node bifurcation as µ passes through zero when 4α 1 ¤ α 2 C 4α 3 . When 4α 1 D α 2 C 4α 3 , the origin of (4.130)–(4.132) is not a bifurcation point. Equation 4.184 is in full agreement with (4.147) obtained with the method of multiple scales. When b 1 D b 2 C b 3 , as y 1 ! 0 and µ ! 0, (4.183) tends to yP 1 D Γ0 µ 2

4 9

[b 3 (α 1 C α 2 2α 3 ) C b 2 (2α 1 C α 2 C α 3 )] µ y 1

(4α 1 α 2 4α 3 ) y 12 1 3

(4.185)

where Γ0 D

1 b [9 C 4b 3 (α 1 C 27 2 1 b 3 [9 C b 3 (α 1 C 27

2α 2 α 3 )] C

1 2 b (4α 1 27 2

C 4α 2 C α 3 )

C 4 (α 2 C α 3 ))]

Therefore, the origin of (4.130)–(4.132) undergoes a transcritical bifurcation as µ passes through zero when 4α 1 ¤ α 2 C 4α 3 . Equation 4.185 is in full agreement with (4.152) obtained with the method of multiple scales. When 4α 1 D α 2 C 4α 3 , as y 1 ! 0 and µ ! 0, (4.183) tends to yP1 D

1 3

[b 3 (4α 3 α 2 ) 2b 2 (α 2 C 2α 3 )] µ y 1

2 3

α 22 C 4α 2 α 3 8α 23 y 13 (4.186)

Therefore, the origin of (4.130)–(4.132) undergoes a supercritical or subcritical pitchfork bifurcation as µ passes through zero, depending on whether α 22 C 4α 2 α 3 8α 23 is positive or negative, respectively. Equation 4.186 is in full agreement with (4.157) obtained with the method of multiple scales. 4.5.3 A Projection Method

In this section, we propose a method of reducing (4.82) directly without calculating the center manifold. We decompose its solution as follows: x D p u(t) C y (t)

(4.187)

where p is the right eigenvector of A, q T y(t) D 0

(4.188)

and q is the left eigenvector of A normalized so that q T p D 1. Substituting (4.187) into (4.82) yields p uP C yP D Ay C bµ C B p µ u C B y µ C Q(p , p )u 2 C 2Q(p , y)u C Q(y, y) C C (p , p , p )u3 C

(4.189)

4.5 Normal Forms of Static Bifurcations

Multiplying (4.189) from the left with q and using (4.188) leads to uP D q T bµ C q T B p µ u C q T B y µ C q T Q(p, p )u2 C 2q T Q(p , y)u C q T Q(y, y) C q T C (p , p , p )u3 C Substituting for uP from (4.190) and (4.189), we obtain yP D Ay C b µ (q T b)p µ C Q(p , p )u2 q T Q(p , p ) p u2 C

(4.190)

(4.191)

There are two possibilities: q T b ¤ 0 and q T b D 0. In the ﬁrst case, as u ! 0 and µ ! 0, (4.190) tends to uP D q T bµ C q T Q(p , p )u2

(4.192)

and therefore the origin of (4.82) undergoes a generic saddle-node bifurcation as µ passes through zero when q T Q(p , p ) ¤ 0. When q T b D 0, we express the solution of (4.191) and (4.188) as y D cµ C z u2

(4.193)

where Ac D b

and

Az D Q(p , p ) C q T Q(p, p ) p

(4.194)

and q T c D 0 and

qT z D 0

(4.195)

Substituting (4.193) into (4.190), we ﬁnd that, as u ! 0 and µ ! 0, (4.190) tends to uP D q T B c C q T Q(c, c) µ 2 C q T B p µ u C q T Q(p , p )u2 (4.196) when q T Q(p , p ) ¤ 0. Therefore, the origin of (4.82) undergoes a generic transcritical bifurcation as µ passes through zero. When q T Q(p , p ) D 0, as u ! 0 and µ ! 0, (4.190) tends to uP D q T B p µ u C 2q T Q(p , z) C q T C (p , p , p ) u3 (4.197) Therefore, the origin of (4.82) undergoes a generic supercritical or subcritical pitchfork bifurcation as µ passes through zero depending on whether [2q T Q(p , z) C q T C (p , p , p )] is negative or positive. Example 4.15 We reconsider (4.167) and (4.168). Their Jacobian matrix evaluated at (0, 0, 0) is given in (4.114) and its eigenvalues are λ D 0 and λ D 2. The right and left eigenvectors of A are 1 pD 1

and

2 qD 1

133

134

4 Bifurcations of Continuous Systems

Therefore, we express the solution of (4.167) and (4.168) as x D p u(t) C y(t) or x1 D u(t) C y 1 (t) and x2 D u(t) C y 2 (t)

(4.198)

and impose the condition q T y D 0. Substituting (4.198) into (4.167) and (4.168) yields uP C yP1 D b 1 µ C 2y 1 2y 2 C µ(u C y 1 ) C α 1 (u C y 1 )2

(4.199)

uP C yP2 D b 2 µ C 4y 1 4y 2 4(u C y 2 )µ C (u C y 1 )(u C y 2 )

(4.200)

The condition q T y D 0 yields 2y 1 y 2 D 0

(4.201)

Multiplying (4.199) and (4.200) from the left with q T and using the condition q T y D 0, we obtain uP D (2b 1 b 2 )µ C (2α 1 α 2 )u2 C 6µ u C 2µ y 1 C 4µ y 2 C (4α 1 α 2 )u y 1 α 2 u y 2 C 2α 1 y 12 α 2 y 1 y 2

(4.202)

Substituting (4.202) into (4.199) and (4.200) yields yP1 D (b 2 b 1 )µ C 2y 1 2y 2 C (α 2 α 1 )u2 C

(4.203)

yP2 D 2(b 2 b 1 )µ C 4y 1 4y 2 C 2(α 2 α 1 )u2 C

(4.204)

Because u(t) varies slowly with time, the solution of (4.204) and (4.201) can be expressed as y 1 D 12 (b 2 b 1 )µ C 12 (α 2 α 1 )u2 C

and

y 2 D 2y 1

(4.205)

It follows from (4.202) that there are two possibilities: b 2 ¤ 2b 1 and b 2 D 2b 1 . In the ﬁrst case, as u ! 0 and µ ! 0, (4.202) tends to uP D (2b 1 b 2 )µ C (2α 1 α 2 )u2

(4.206)

Therefore, the origin of (4.167) and (4.168) undergoes a saddle-node bifurcation as µ passes through zero if α 2 ¤ 2α 1 . We note that (4.206) is in full agreement with (4.118) and (4.175) obtained with the methods of multiple scales and centermanifold reduction, respectively.

4.5 Normal Forms of Static Bifurcations

When b 2 D 2b 1 , substituting (4.205) into (4.202) yields uP D 12 b 1 (10 C b 1 α 1 b 1 α 2 ) µ 2 C C (2α 1 α 2 )u 2

1 (4α 1 2

1 2

(12 C 4b 1 α 1 3b 1 α 2 ) µ u

3α 2 )(α 1 α 2 )u3 C

(4.207)

As u ! 0 and µ ! 0, (4.207) tends to uP D 12 b 1 (10 C b 1 α 1 b 1 α 2 ) µ 2 C C (2α 1 α 2 )u

1 2

(12 C 4b 1 α 1 3b 1 α 2 ) µ u

2

(4.208)

Therefore, the origin of (4.167) and (4.168) undergoes a transcritical bifurcation as µ passes through zero if α 2 ¤ 2α 1 . We note that (4.208) is in full agreement with (4.122) and (4.176) obtained with the methods of multiple scales and centermanifold reduction, respectively. When α 2 D 2α 1 , as u ! 0 and µ ! 0, (4.207) tends to uP D (6 b 1 α 1 )µ u α 21 u3

(4.209)

Therefore, the origin of (4.167) and (4.168) undergoes a supercritical pitchfork bifurcation as µ passes through zero if α 2 D 2α 1 . We note that (4.209) is in full agreement with (4.129) and (4.177) obtained with the methods of multiple scales and center-manifold reduction, respectively. Example 4.16 We reconsider the system (4.130)–(4.132) treated in Examples 4.12 and 4.14. The coefﬁcient matrix of this system evaluated p at (0, 0, 0I 0) is given by (4.142). Its eigenvalues are λ D 0 and λ D 3/2(1 ˙ i 3). Hence, the origin of (4.130)–(4.133) may be a static bifurcation point. The right and left eigenvectors of A corresponding to λ D 0 are 2 3 2 3 1 1 1 p D 4 1 5 and q D 4 1 5 3 1 1 where q is normalized so that q T p D 1. We express the solution of (4.130)–(4.132) as x D p u(t) C y(t) where

qT y D 0

or x1 D u C y 1 , y1 y2 y3 D 0

x2 D u C y 2 ,

x3 D u C y 3

(4.210) (4.211)

135

136

4 Bifurcations of Continuous Systems

Substituting (4.210) into (4.130)–(4.132), we have yP1 uP D b 1 µ y 1 C y 2 2y 3 µ u C µ y 1 C α 1 (2u C y 1 C 2y 2 C y 3 )2

(4.212)

yP2 C uP D b 2 µ 2y 1 y 2 y 3 C µ u C µ y 2 C α 2 (u C y 1 y 2 C y 3 )2

(4.213)

yP3 C uP D b 3 µ C y 1 C 2y 2 y 3 2µ u 2µ y 3 C α 3 (2u C 2y 1 C y 2 y 3 )2

(4.214)

Multiplying (4.212)–(4.213) from the left with q T yields uP D 31 (b 1 b 2 b 3 )µ C 13 (4α 3 C α 2 4α 1 )u2 (y 1 y 2 C 2y 3 )µ 23 (4α 3 C α 2 C 2α 1 )u y 1 23 (2α 3 α 2 C 4α 1 )u y 2 C 23 (2α 3 α 2 2α 1 )u y 3 C 13 (4α 3 C α 2 α 1 )y 12 C 13 (α 3 C α 2 4α 1 )y 22 C 13 (α 3 C α 2 4α 1 )y 22 C 13 (α 3 C α 2 α 1 )y 32 C 23 (2α 3 α 2 2α 1 )y 1 y 2 23 (2α 3 α 2 C α 1 )y 1 y 3 23 (α 3 C α 2 C 2α 1 )y 2 y 3

(4.215)

Substituting (4.215) into (4.212)–(4.213) yields yP 1 D 13 (4b 1 b 2 b 3 )µ y 1 C y 2 2y 3 C 23 (2α 1 C α 2 2α 3 )u2 C

(4.216)

yP 2 D 13 (b 1 C 2b 2 b 3 )µ 2y 1 y 2 y 3 C 13 (4α 1 α 2 C 8α 3 )u2 C

(4.217)

yP 3 D 13 (b 1 b 2 C 2b 3 )µ C y 1 C 2y 2 y 3 C 13 (4α 1 α 2 C 8α 3 )u2 C

(4.218)

Because u(t) is a slowly varying function of time, the particular solution of (4.216)– (4.218) orthogonal to q can be expressed as y 1 D 19 (b 1 C 2b 2 b 3 )µ C 29 (2α 1 C α 2 2α 3 )u2 C

(4.219)

y 2 D 19 (b 2 b 1 2b 3 )µ 19 (4α 1 α 2 C 8α 3 )u2 C

(4.220)

y 3 D 19 (2b 1 C b 2 C b 3 )µ C 19 (8α 1 C α 2 C 4α 3 )u2 C

(4.221)

There are two possibilities: b 1 b 2 b 3 ¤ 0 and b 1 b 2 b 3 D 0. In the ﬁrst case, as u ! 0 and µ ! 0, (4.215) tends to uP D 13 (b 1 b 2 b 3 )µ C 13 (4α 3 C α 2 4α 1 )u2

(4.222)

4.6 Normal Form of Hopf Bifurcation

Therefore, the origin of (4.130)–(4.132) undergoes a saddle-node bifurcation as µ passes through zero when 4α 1 ¤ α 2 C 4α 3 . When 4α 1 D α 2 C 4α 3 , the origin of (4.130)–(4.132) is not a bifurcation point. Equation 4.222 is in full agreement with (4.147) and (4.184) obtained with the methods of multiple scales and centermanifold reduction, respectively. When b 1 D b 2 C b 3 , substituting (4.219)–(4.221) into (4.215), we have 1 9b 2 C 9b 3 C 4b 22 α 1 4b 2 b 3 α 1 C b 23 α 1 4b 22 α 2 8b 2 b 3 α 2 uP D 27 2 4b 3 α 2 b 22 α 3 C 4b 2 b 3 α 3 4b 23 α 3 µ 2

4 27

(b 1 α 1 C 5b 2 α 1 4b 3 α 1 C 2b 1 α 2 C b 2 α 2

Cb 2 α 2 C b 3 α 2 b 1 α 3 C 4b 2 α 3 5b 3 α 3 ) µ u C 13 (4α 3 C α 2 4α 1 )u2 2 4 (4.223) 4α 1 C 13α 1 α 2 C α 22 20α 1 α 3 C 8α 2 α 3 20α 23 u3 27 When 4α 1 ¤ α 2 C 4α 3 , as u ! 0 and µ ! 0, (4.223) tends to 1 uP D 27 9b 2 C 9b 3 C 4b 22 α 1 4b 2 b 3 α 1 C b 23 α 1 4b 22 α 2 8b 2 b 3 α 2 4b 23 α 2 b 22 α 3 C 4b 2 b 3 α 3 4b 23 α 3 µ 2

4 27

(b 1 α 1 C 5b 2 α 1 4b 3 α 1 C 2b 1 α 2 C b 2 α 2

Cb 2 α 2 C b 3 α 2 b 1 α 3 C 4b 2 α 3 5b 3 α 3 ) µ u C 13 (4α 3 C α 2 4α 1 )u2

(4.224)

Therefore, the origin of (4.130)–(4.132) undergoes a transcritical bifurcation as µ passes through zero when 4α 1 ¤ α 2 C 4α 3 . Equation 4.224 is in full agreement with (4.152) and (4.185) obtained with the methods of multiple scales and centermanifold reduction, respectively. When α 1 D α 3 C 1/4α 2 , as u ! 0 and µ ! 0, (4.223) tends to uP D

1 3

[b 3 (4α 3 α 2 ) 2b 2 (α 2 C 2α 3 )] µ u

2 3

α 22 C 4α 2 α 3 8α 23 u3 (4.225)

Therefore, the origin of (4.130)–(4.132) undergoes a supercritical or subcritical pitchfork bifurcation as µ passes through zero, depending on whether α 22 C 4α 2 α 3 8α 23 is positive or negative, respectively. Equation 4.225 is in full agreement with (4.157) and (4.186) obtained with the methods of multiple scales and center-manifold reduction, respectively.

4.6 Normal Form of Hopf Bifurcation

In this section, we describe methods for simplifying the dynamical system (4.17) near a Hopf bifurcation point. We assume that this point has been shifted to x D 0 and that the control parameter has been shifted to µ D 0. Moreover, we expand

137

138

4 Bifurcations of Continuous Systems

(4.17) for small x and µ and obtain (4.82). Furthermore, because A is nonsingular, we can shift the ﬁxed point by A1 bµ and hence rewrite (4.82) as xP D Ax C µ B x C Q(x, x) C C (x, x, x) C

(4.226)

Because the origin is a Hopf bifurcation, two of the eigenvalues are purely imaginary complex conjugates (i.e., ˙i ω) and the remaining eigenvalues are in the lefthalf of the complex plane. Next, we demonstrate how one can use the methods of multiple scales, projection method, and center-manifold reduction to compute the normal form of (4.226) near x D 0 and small µ. 4.6.1 The Method of Multiple Scales

To compute the normal form of the Hopf bifurcation of (4.226) at the origin, we introduce a small nondimensional parameter as a bookkeeping parameter and seek a third-order approximate solution of (4.226) in the form x(tI µ) D x 1 (T0 , T2 ) C 2 x 2 (T0 , T2 ) C 3 x 3 (T0 , T2 ) C

(4.227)

where the time scales Tm D m t. The time derivative can be expressed in terms of these scales as d D D0 C 2 D2 C dt

(4.228)

where D m D @/@Tm . As shown below, there is no dependence on T1 because the second-order problem is solvable. Moreover, because µ is small, we scale it as 2 µ. Substituting (4.227) and (4.228) into (4.226) and equating coefﬁcients of like powers of , we obtain Order ()

D0 x 1 Ax 1 D 0

(4.229)

Order (2 )

D0 x 2 Ax 2 D Q(x 1 , x 1 )

(4.230)

Order (3 )

D0 x 3 Ax 3 D D2 x 1 C µ B x 1 C 2Q(x 1 , x 2 ) C C (x 1 , x 1 , x 1 )

(4.231)

The general solution of (4.229) is the superposition of n linearly independent solutions corresponding to the n eigenvalues of A: n 2 of these eigenvalues have negative real parts and two are purely imaginary. The n 2 solutions corresponding

4.6 Normal Form of Hopf Bifurcation

to the eigenvalues with negative real parts decay with time and hence the longtime dynamics of the system depends on the center space corresponding to the two purely imaginary eigenvalues. Therefore, we express the solution of (4.229) as x 1 D z(T2 )p e i ω T0 C z(T N 2 ) pN e i ω T0

(4.232)

where the function z(T2 ) is determined by imposing the solvability condition at third order and p is the eigenvector of A corresponding to the eigenvalue i ω; that is, Ap D i ω p

(4.233)

Substituting (4.232) into (4.230) yields D0 x 2 Ax 2 D Q(p , p )z 2 e 2i ω T0 C 2Q(p , pN )z zN C Q( pN , pN ) zN 2 e 2i ω T0 (4.234) The solution of (4.234) can be expressed as x 2 D ζ 2 z 2 e 2i ω T0 C 2ζ 0 z zN C ζN 2 zN 2 e 2i ω T0

(4.235)

where [2i ω I A] ζ 2 D Q(p , p )

and

Aζ 0 D Q(p, pN )

(4.236)

Substituting (4.232) and (4.235) into (4.231), we have D0 x 3 Ax 3 D D2 z p µ B p z 4Q(p , ζ 0 )z 2 zN 2Q( pN , ζ 2 )z 2 zN 3C (p , p , pN )z 2 zN e i ω T0 C cc C NST . (4.237) Because p e i ω T0 is a solution of the homogeneous equation 4.237, the nonhomogeneous equation has a solution only if a solvability condition is satisﬁed. We let q be the left eigenvector of A corresponding to the eigenvalue i ω; that is, AT q D i ωq

(4.238)

We normalize it so that q T p D 1. Then, the solvability condition demands that the term proportional to e i ω T0 in (4.237) is orthogonal to q. Imposing this condition, we obtain the following normal form of the Hopf bifurcation: D2 z D q T B p µ z C 4q T Q(p , ζ 0 ) C 2q T Q( pN , ζ 2 ) C 3q T C(p , p , pN ) z 2 zN (4.239)

139

140

4 Bifurcations of Continuous Systems

Example 4.17 We consider the system xP1 D µ x1 x2 C x12 α 1 x1 x3

(4.240)

xP2 D µ x2 C x1 C α 2 x2 x3

(4.241)

xP3 D x3 C x12 C 2x22

(4.242)

Substituting (4.227) and (4.228) into (4.240)–(4.242), scaling µ as 2 µ, and equating coefﬁcients of like powers of on both sides, we obtain Order () D0 x11 C x21 D 0

(4.243)

D0 x21 x11 D 0

(4.244)

D0 x31 C x31 D 0

(4.245)

Order ( 2 ) 2 α 1 x11 x31 D0 x12 C x22 D x11

(4.246)

D0 x22 x12 D α 2 x21 x31

(4.247)

2 2 D0 x32 C x32 D x11 C 2x21

(4.248)

Order ( 3 ) D0 x13 C x23 D D2 x11 C µ x11 C 2x11 x12 α 1 x11 x32 α 1 x12 x31

(4.249)

D0 x23 x13 D D2 x21 C µ x21 C α 2 x21 x32 C α 2 x22 x31

(4.250)

D0 x33 C x33 D D2 x31 C 2x11 x12 C 4x21 x22

(4.251)

The nondecaying solutions of (4.243)–(4.245) can be expressed as x11 D i z(T2 )e i T0 i zN (T2 )e i T0 , N 2 )e i T0 , x21 D z(T2 )e i T0 C z(T x31 D 0

(4.252)

4.6 Normal Form of Hopf Bifurcation

Substituting (4.252) into (4.246)–(4.248) yields D0 x12 C x22 D z 2 e 2i T0 2z zN C zN 2 e 2i T0

(4.253)

D0 x22 x12 D 0

(4.254)

D0 x32 C x32 D z 2 e 2i T0 C 6z zN C zN 2 e 2i T0

(4.255)

whose solutions can be expressed as x12 D 32 i z 2 e 2i T0 C 23 i zN 2 e 2i T0

(4.256)

x22 D 31 z 2 e 2i T0 2z zN 13 zN 2 e 2i T0

(4.257)

x32 D 15 (1 2i)z 2 e 2i T0 C 6z zN C 15 (1 C 2i) zN 2 e 2i T0

(4.258)

Substituting (4.252) and (4.256)–(4.258) into (4.249) and (4.250), we have D0 x13 C x23 D i D2 z C i µ z C 43 C 25 α 1 (2 29i) z 2 zN e i T0 C cc C NST

(4.259)

D0 x23 x13 D D2 z C µ z C 15 α 2 (31 2i)z 2 zN e i T0 C cc C NST (4.260) The solvability condition of (4.259) and (4.260) demands that their right-hand sides be orthogonal to q, where q T D 1/2(i, 1). Imposing this condition, we obtain the following normal form of the system (4.240)–(4.242) near its origin: D2 z D µ z C

1 30

93α 2 87α 1 (20 C 6α 1 C 6α 2 )i z 2 zN

(4.261)

Letting z D 1/2e i θ in (4.261) and separating real and imaginary parts, we obtain the following alternate normal form of the Hopf bifurcation: D2 r D 12 µ r C

1 (31α 2 40

29α 1 )r 3

1 D2 θ D 60 (10 C 3α 1 C 3α 2 )r 2

(4.262) (4.263)

4.6.2 Center-Manifold Reduction

To determine the normal form of the Hopf bifurcation of the system (4.226) at the origin, we ﬁrst calculate the eigenvalues λ 1 , λ 2 , . . . , λ n and eigenvectors (generalized eigenvectors) p 1 , p 2 , . . . , p n of the matrix A. Second, we arrange the eigenvalues of A so that λ 1 D i ω and λ 2 D i ω and let p 1 and pN 1 be their corresponding eigenvectors. Third, we introduce the transformation x D P y and obtain yP D J y C P 1 B P y µ C P 1 Q(P y, P y) C P 1 C (P y, P y, P y)

(4.264)

141

142

4 Bifurcations of Continuous Systems

where J D P 1 AP . We note that J can be rewritten as Jc 0 iω 0 JD and J c D 0 Js 0 i ω

(4.265)

and J s is an (n 2) (n 2) matrix whose eigenvalues are λ 3 , λ 4 , . . . , λ n . Fourth, we let y c be the two-dimensional vector with the components y 1 and y 2 and let y s be the (n 2)-dimensional vector with the components y 3 , y 4 , . . . , y n and rewrite (4.264) as yP c D J c y c C µ B1c y c C µ B1s y s C G 2 (y c , y s ) C G 3 (y c , y s )

(4.266)

yP s D J s y s C µ B2c y c C µ B2s y s C F 2 (y c , y s ) C F 3 (y c , y s )

(4.267)

and

We note that y c and y s are linearly uncoupled but nonlinearly coupled. Further, F j (0, 0) D 0, G j (0, 0) D 0, and the Jacobian matrices D F j (0, 0) and D G j (0, 0) are matrices with zero entries. Because F j and G j are polynomials and hence inﬁnitely differentiable, there exists a local center manifold of the form y s D h(y c ) where h is a polynomial function of y c and h(0) D 0

and

D y c h(0) D 0

(4.268)

Fifth, we determine the (n2)-dimensional function h by constraining the center manifold to be two-dimensional in the n-dimensional space. Substituting for y s into (4.266) yields yP c D J c y c C µ B1c y c C µ B1s h(y c ) C G 2 y c , h(y c ) C G 3 y c , h(y c ) (4.269) Substituting for y s into (4.267) and using (4.268), we have ˚ D y c h(y c ) J c y c C G 2 y c , h(y c ) C G 3 y c , h(y c ) D J s h(y c ) C F 2 [y c , h(y c )] C F 3 [y c , h(y c )] C

(4.270)

To solve (4.270), one approximates the components of h(y c ) with polynomials. The polynomial approximations are usually taken to be quadratic to the ﬁrst approximation and do not contain constant and linear terms so that the conditions on h are satisﬁed. Substituting the assumed quadratic polynomial approximations into (4.270) and equating the coefﬁcients of the different powers in the polynomials on both sides, one obtains a system of algebraic equations for the coefﬁcients of the polynomials. Solving these equations, we obtain a ﬁrst approximation to the center manifold y s D h(y c ). Finally, substituting this approximation into (4.269), we obtain a two-dimensional dynamical system describing the dynamics on the center manifold.

4.6 Normal Form of Hopf Bifurcation

Example 4.18 We use a combination of center-manifold reduction and the method of normal forms to compute the normal form of the Hopf bifurcation near the origin of (4.240)–(4.242). In this case, the local manifold is one-dimensional and tangent to the x1 x2 plane at the origin. We approximate this manifold by a quadratic function in the form x3 (x1 , x2 ) D h(x1 , x2 ) D a 1 x12 C a 2 x22 C a 3 x1 x2 C

(4.271)

Substituting (4.271) into (4.242) yields 2a 1 x1 xP 1 C 2a 2 x2 xP 2 C a 3 x1 xP 2 C a 3 x2 xP 1 D a 1 x12 C a 2 x22 C a 3 x1 x2 C x12 C 2x22 C

(4.272)

Substituting (4.240) and (4.241) into (4.272) and using (4.271), we have 2(a 2 a 1 )x1 x2 C a 3 x12 a 3 x22 D a 1 x12 C a 2 x22 C a 3 x1 x2 C x12 C2x22 C (4.273) Equating the coefﬁcients of each of x12 , x22 , and x1 x2 on both sides of (4.273), we obtain a3 C a1 D 1 ,

a2 a3 D 2 ,

a 3 C 2a 2 2a 1 D 0

whose solution is a1 D

7 5

,

a2 D

8 5

,

a 3 D 25

(4.274)

Substituting (4.274) into (4.271) and then substituting the result into (4.240) and (4.241), we obtain the following two-dimensional system describing the dynamics of (4.240)–(4.242) on the center manifold: xP 1 D µ x1 x2 C x12 75 x13 α 1 C 25 x12 x2 α 1 85 x1 x22 α 1

(4.275)

xP 2 D µ x2 C x1 C 75 x12 x2 α 2 25 x1 x22 α 2 C 85 x23 α 2

(4.276)

Next, we use the method of normal forms to simplify (4.275) and (4.276). First, we calculate the Jacobian matrix of (4.275) and (4.276) evaluated at the origin. The result is 0 1 1 0

143

144

4 Bifurcations of Continuous Systems

Its eigenvalues are λ D ˙i. The right and left eigenvectors of this matrix corresponding to λ D i are pD

i 1

qD

and

1 2

i 1

Second, we introduce the transformation x1 D p u(t) C pN u(t) N x2 and obtain x1 (t) D i u(t) i u(t) N

and

x2 (t) D u(t) C u(t) N

(4.277)

Substituting (4.277) into (4.275) and (4.276) and multiplying the result from the left with q yields (31 2i)α 2 (29 C 2i)α 1 u2 uN (4.278) C nonresonance cubic and higher-order terms

uP D i u C µ u C 12 i(u2 2u uN C uN 2 ) C

1 10

Third, we introduce a near-identity transformation of the form u(t) D v (t) C b 1 v 2 (t) C b 2 v (t) vN (t) C b 3 vN 2

(4.279)

into (4.278), approximate vP and vPN with i v and i vN , respectively, and obtain vP D i v C µ v C 12 i(1 2b 1 )v 2 i(1 b 2 )v vN C 12 i(1 C 6b 3 ) vN 2 1 (31 2i)α 2 (29 C 2i)α 1 C 10i(b 3 b 1 ) v 2 vN C 10

(4.280)

Fourth, we choose the b i to eliminate the quadratic terms and obtain b1 D

1 2

,

b2 D 1 ,

b 3 D 61

(4.281)

Finally, we substitute for the b i in (4.280) and obtain the normal form vP D i v C µ v C

1 30

93α 2 87α 1 (20 C 6α 1 C 6α 2 )i v 2 vN

(4.282)

which is in full agreement with (4.261) obtained with the method of multiple scales. 4.6.3 Projection Method

To determine the normal form of the Hopf bifurcation at the origin of the system (4.226) as µ increases past zero with the projection method, we assume that x(t) D p u(t) C pN u(t) N C y (t)

(4.283)

4.6 Normal Form of Hopf Bifurcation

where p and pN are the right eigenvectors of A corresponding to the eigenvalues ˙i ω. We assume that all of the other eigenvalues of A are in the left-half of the complex plane. Moreover, we constrain y(t) to be orthogonal to the left eigenvectors q and qN corresponding to the eigenvalues ˙i ω normalized so that q T p D 1 and qN T pN D 1; that is, we require qT y D 0

and

qN T y D 0

(4.284)

Substituting (4.283) into (4.226) yields p uP C pN uPN C yP D i ω p u i ω pN uN C Ay C µ uB p C Q(p , p )u2 C 2Q(p , pN )u uN C Q( pN , pN ) uN 2 C 3C (p, p , pN )u2 uN C 2Q(p , y)u C Q( pN , y) uN C

(4.285)

where the three dots stand for nonresonance and higher-order terms. Multiplying (4.285) from the left with q T , we have uP D i ω p u C µ uB p C q T Q(p, p )u2 C 2q T Q(p , pN )u uN C q T Q( pN , pN ) uN 2 (4.286) C 3q T C (p , p , pN )u2 uN C 2q T Q(p , y)u C q T Q( pN , y) uN C Substituting (4.286) and its complex conjugate into (4.285), we obtain yP D Ay C Q(p , p ) q T Q(p , p )p qN T Q(p, p ) pN u2 C 2 Q(p , pN ) q T Q(p , pN )p qN T Q(p , pN ) pN u uN C Q( pN , pN ) q T Q( pN , pN )p qN T Q( pN , pN ) pN uN 2 C A particular solution of (4.287) can be expressed as i i T qN Q(p , p ) pN u2 y D C η 2 C q T Q(p , p )p C ω 3ω i T i T C 2 η 0 q Q(p , pN )p C qN Q(p , pN ) pN u uN ω ω i T i q Q( pN , pN )p qN T Q( pN , pN ) pN uN 2 C C ηN 2 3ω ω

(4.287)

(4.288)

Next, we substitute (4.288) into (4.284) and determine constraint conditions on η 2 and η 0 . Finally, we substitute (4.288) into (4.286) and introduce a near-identity transformation to eliminate the quadratic terms, thereby obtaining the normal form of the bifurcation. We note that, out of the three methods, the method of multiple scales seems to be the best for simplifying high-dimensional nonlinear systems near a Hopf bifurcation.

145

146

4 Bifurcations of Continuous Systems

Example 4.19 We use the projection method to compute the normal form of the Hopf bifurcation at the origin of the system (4.240)–(4.242) as µ passes through zero. In this case, 0 1 i p D @1 A 0

and

0 1 i [email protected] A qD 1 2 0

(4.289)

and (4.283) and (4.284) lead to x1 (t) D i u(t) i u(t) N C y 1 (t) , N C y 2 (t) , x2 (t) D u(t) C u(t)

and

x3 (t) D y 3 (t) y 1 (t) D 0 and

(4.290) y 2 (t) D 0

(4.291)

Substituting (4.290) and (4.291) into (4.240)–(4.242) yields i uP i uNP D u uN C i µ(u uN ) u2 C 2u uN uN 2 i α 1 u y 3 C i α 1 uN y 3 (4.292) uP C uNP D i uN C µ(u u) N C α 2 u y 3 C α 2 uy N 3

(4.293)

yP3 D y 3 C u2 C 6u uN C uN 2

(4.294)

Multiplying (4.292)–(4.294) from the left with q, we have uP D i u C µ u C 12 i u2 i u uN C 12 i u2 12 u y 3 α 1 C 12 uN y 3 α 1 C 12 u y 3 α 2 C 12 uN y 3 α 2

(4.295)

Substituting (4.295) and its complex conjugate into (4.292)–(4.294), we ﬁnd that the ﬁrst two equations reduce to zero. Then, solving (4.294) yields y 3 D 15 (1 2i)u2 C 6u uN C 15 (1 C 2i) uN 2

(4.296)

Substituting (4.296) into (4.295), we obtain (4.282) obtained with center-manifold reduction.

4.7 Exercises

4.7.1 Consider the following one-dimensional systems: a) xP D µ x C x 2 b) xP D µ C x 2

4.7 Exercises

c) xP D µ x C x 3 d) xP D µ x x 3 e) xP D µ x 3 . In each case, x is the state variable and µ is the control parameter. Construct the bifurcation diagrams for all cases and discuss them. 4.7.2 Consider the following one-dimensional systems: a) b) c) d)

xP xP xP xP

D 2µ 2 C µ x x 2 D xP D 2µ 2 C µ x x 3 D µ C µx x3 D µ C µx x2 C x3 .

What type of bifurcation does each of these systems undergo with increasing µ at µ D 0? Determine the bifurcating solutions. 4.7.3 Determine the ﬁxed points of xP D x 2 4x C 3 Determine their stability. 4.7.4 Find the equilibrium solutions of xP D x 2ax 2 C x 3 Determine their stability and the bifurcations, which they undergo as a is varied. 4.7.5 Sketch the bifurcation diagram of xP D x 3 C a 3 3ax in the (a, x) plane, indicating which equilibrium solutions are stable and identifying a turning point and a pitchfork bifurcation. 4.7.6 Sketch the bifurcation diagram of xP D 2(a 2 x 2 ) (a 2 C x 2 )2 in the (a, x) plane, indicating which equilibrium solutions are stable and identifying two turning points and a transcritical bifurcation. 4.7.7 Sketch the bifurcation diagram of xP D x (x 2 2b x a) in the (a, x) plane for a given positive b, indicating which solutions are stable. 4.7.8 Consider the one-dimensional system xP D ax C b x 3 C c x 5 Determine the ﬁxed points and their stability.

147

148

4 Bifurcations of Continuous Systems

4.7.9 Consider the system xP D x 4 C ax 2 C b x C c Sketch the intersection of the bifurcation set and three planes a D constant for a < 0, a D 0, and a > 0 (Thom, 1975). 4.7.10 Consider the one-dimensional system (Drazin, 1992) xP D x 3 2ax 2 (b 3)x C c for real a, b, and c. a) Show that, if c D 0, then there is a transcritical bifurcation, but if c ¤ 0, there are two (nonbifurcating) branches of equilibria. b) Show that the loci of the bifurcation points is given by the curve (27c 18b C 38)2 D 4(3b 5)3 which has a cusp at b D 3/5 and c D 8/27. 4.7.11 Determine the ﬁxed points and their types for the following systems, and for each case sketch the trajectories and the separatrices in the phase plane: a) b) c) d) e)

xR xR xR xR xR

C 2 xP C x C x 3 D 0 C 2 xP C x x 3 D 0 C 2 xP x C x 3 D 0 C 2 xP x x 3 D 0 a C x 2 D 0 for a > 0, a D 0, and a < 0 .

4.7.12 Consider the following system (Drazin, 1992): xP D x 3 C δ x 2 µ x Determine the ﬁxed points of this system and study the bifurcations in the (x, µ) plane for zero and nonzero values of δ. Show that the pitchfork bifurcation at (0, 0) for δ D 0 becomes a transcritical bifurcation for small δ and that there is a turning point at (1/2δ, 1/4δ 2 ). Sketch the bifurcation diagram in the (x, µ) plane for δ > 0. 4.7.13 Consider the following single-degree-of-freedom system with quadratic and cubic nonlinearities: xR C x C δ x 2 C α x 3 D 0 Sketch the potential energy V(x) for the system and the associated phase portrait for each of the following cases: (a) δ D 3 and α D 4, (b) δ D α D 4, and (c) δ D 5 and α D 4.

4.7 Exercises

It is common to refer to the ﬁrst case as a single-well potential system because there is a well in the graph of V(x) versus x. The third case is referred to as a twowell potential system. From the phase portraits, one can discern a qualitative change as one goes from the ﬁrst case to the third case. 4.7.14 Consider the system xP D x (3 x 2y ) yP D y (2 x y ) where x > 0 and y > 0. Determine all of the ﬁxed points of this system and determine their types. 4.7.15 Consider the system xP D x 2 µ 2 yP D y C x 2 µ 2 Determine the equilibrium solutions and their stability. 4.7.16 Consider the system xP 1 D x2 13 λ(x13 3x1 ) xP 2 D x1 Show that the origin is the only equilibrium point. Determine its stability as a function of λ. 4.7.17 Show that the origin is an unstable equilibrium point for each of the two systems a) xP 1 D x2 , xP 2 D x1 C 2x23 . b) xP 1 D x1 C 5x2 C x12 x2 , xP 2 D 5x1 C x2 x23 . 4.7.18 Show that the origin is a saddle point for each of the two systems a) xP 1 D x2 , xP 2 D x2 C x13 . b) xP 1 D x2 , xP 2 D x1 C x13 . Determine the stable and unstable manifolds of the origin for the linearized as well as the nonlinear systems. 4.7.19 Determine an approximation to the stable and unstable manifolds of the saddles of the system xP 1 D 1 x1 x2 xP 2 D x1 x23

149

150

4 Bifurcations of Continuous Systems

4.7.20 Consider the system xP 1 D λ C 2x1 x2 xP 2 D 1 C x12 x22 Show that there are two saddles when λ D 0 and that the x2 -axis is invariant. Hence, show that there is a heteroclinic connection. Sketch the phase plane. Show that, when jλj ¤ 0 and is small, there are still two saddle points but the saddle connection is no longer present. Sketch the phase plane for 1 λ > 0 and 1 λ < 0. 4.7.21 Show that the trivial solution is the only equilibrium solution of xP D x y 2 C x 2 y C x 3

yP D y 3 x 3

and

and that it is unstable. 4.7.22 Show that the trivial solution of the system xP D 2x y 2 x 3 and yP D

2 2 x y y3 5

is asymptotically stable. 4.7.23 Show that the origin is a stable equilibrium point of the system xP D y x 3

yP D x 2

and

4.7.24 Show that the origin is an asymptotically stable equilibrium point of the system xP D y x (x 4 C y 4 )

and

yP D x y (x 4 C y 4 )

4.7.25 The origin of each of the following systems is a degenerate saddle point: a) xP D x 2 , yP D y b) xP D x 2 y 2 , yP D 2x y . Sketch the phase portrait for each case. 4.7.26 Show that the origin is an unstable equilibrium point of the system xP D 2x 2 y

and

yP D 2x y 2

Hint: Show that x y is constant on each orbit. 4.7.27 Consider the system xP D x y

and

yP D 2 x y

Find the ﬁxed points and determine their stability.

4.7 Exercises

4.7.28 Consider the planar system xP D 12 (x C x 3 ) 2y C yP D 1 2x 2 where 1. Show that there are three saddle points. For D 0 and ¤ 0, sketch the phase portrait and indicate any heteroclinic connections. 4.7.29 Consider the system xR C ω 2 x C δ x 2 cos(Ω t) D 0 where is a small nondimensional parameter. Use the method of multiple scales to determine the modulation equations when Ω 3ω. 4.7.30 Consider the system xR C ω 2 x C δ x 2 cos(Ω t) D 0 where is a small nondimensional parameter. Use the method of averaging to determine the modulation equations when Ω 3ω. 4.7.31 Consider the following speed-control system investigated by Fallside and Patel (1965): xP1 D x2

x2 xP 2 D K d x2 x1 G x12 C x1 C 1 Kd a) For K d D 1 and G D 6, determine the ﬁxed points and their stability. b) Plot the stable manifolds of the unstable ﬁxed points and a few other trajectories in the (x1 , x2 ) space. c) Discuss the phase portrait. 4.7.32 Calculate the ﬁxed points of the system xP D α x y

and

yP D y C x 2

Determine their stability. 4.7.33 The free oscillation about the upright position of an inverted pendulum constrained to oscillate between two closely spaced rigid barriers is described by (e.g., Shaw and Rand, 1989) xR C 2µ xP x D 0 jxj < 1 xP ! r xP

jxj D 1

151

152

4 Bifurcations of Continuous Systems

where x describes the position of the pendulum; the locations of the rigid barriers are x D 1 and x D 1; µ is a measure of the friction; and r 1 is a reﬂection coefﬁcient representing energy loss during impact with either of the rigid barriers. Assuming elastic impact (i.e., r D 1), construct and discuss the phase portraits for the following two cases: (a) µ D 0 and (b) µ > 0. 4.7.34 In studying the forced response of a van der Pol oscillator with delayed amplitude limiting, Nayfeh (1968) encountered the following system of equations: xP 1 D x1 (1 x12 ) C F cos x2 xP 2 D σ C νx12

F sin x2 x1

a) Show that the ﬁxed points (x10 , x20 ) of this system satisfy (1 )2 C (σ C ν)2 D F 2 2 where D x10 . b) For ν D 0.15, plot the loci of the ﬁxed points in the σ plane for F 2 D 1, 1/3, 4/27, and 1/10. What is the signiﬁcance of the value 4/27? c) Show that the interior points of the ellipse deﬁned by

(1 )(1 3) C (σ C ν)(σ C 3ν) D 0 are saddle points and hence unstable. Also, show that the exterior points are nodes if D 0 and foci if D < 0, where D D 4 (1 3ν 2 )2 4νσ σ 2 . d) Finally, show that the exterior points are stable if > 1/2 and unstable if < 1/2. 4.7.35 A bead of mass m sliding on a rotating circular hoop of radius R is described by g θR C 2µ θP C sin θ ω 2 sin θ cos θ D 0 R Here, θ describes the angular position of the bead on the hoop, g is the acceleration due to gravity, µ is a measure of the friction experienced by the bead, and ω is the angular velocity of the hoop. a) For µ D 0, determine the ﬁxed points (equilibrium positions) of the system and sketch the phase portrait in each of the following cases: (i) ω 2 < g/R, (ii) ω 2 D g/R, and (iii) ω 2 > g/R. b) For µ > 0, choose ω as a control parameter and examine the different local bifurcations of the ﬁxed points that occur as ω is increased from zero. Construct appropriate bifurcation diagrams.

4.7 Exercises

4.7.36 Mingori and Harrison (1974) studied the following system for analyzing the motion of a particle constrained to move on a circular path that is spinning and coning: uP D v vP D µ 1 (v 1) C µ 2 µ 3 sin u C 12 µ 23 sin 2u Let µ 1 D 0.1 and µ 2 D 2.0. Then, as µ 3 is varied from zero, bifurcations take place at 0.0502, 0.3, and 2.265. Examine the qualitative changes that take place in the (u, v ) space due to these bifurcations. 4.7.37 Consider the nonlinear oscillator uR C u C 2µ 1 uP C µ 2 uj P uj P C α u3 C 2K u cos(Ω t) D 0 where is a small, positive parameter. Further, the parameters µ 1 , µ 2 , and K are all independent of while the parameter Ω is such that Ω D 2 C σ A ﬁrst approximation obtained for this system has the form u D p cos 12 Ω t C q sin 12 Ω t C O() where

q 1 4µ 2 3α p 0 D µ 1 p (σ C K )q C q(p 2 C q 2 ) p p 2 C q2 2 8 3π q 1 4µ 2 3α 0 2 2 q D µ 1 q C (σ K )p p (p C q ) q p 2 C q2 2 8 3π

In the above equations, the prime denotes the derivative with respect to the time scale τ D t. a) Simplify the dynamical system governing p and q to its normal form for a transcritical bifurcation in the vicinity of q (p, q, K c ) D 0, 0, 4µ 21 C σ 2 . b) What happens to the bifurcation at the above-mentioned bifurcation point when µ 2 D 0? c) Construct the frequency-response curves when µ 2 D 0 and discuss them. 4.7.38 Consider the system x xP D y C p (1 x 2 y 2 ) x2 C y2 y (1 x 2 y 2 ) yP D x C p 2 x C y2 Determine its limit cycle and indicate whether it is stable or unstable.

153

154

4 Bifurcations of Continuous Systems

4.7.39 Consider the system xP D µ x C y C x y yP D 2µ C x y y 2 where µ is the control parameter. Determine the normal form of this system in the vicinity of the bifurcation point (x, y, µ) D (0, 0, 0). What is the type of this bifurcation? 4.7.40 Consider the system xP 1 D µ C x1 x2 C 2µ x2 C x12 xP 2 D b µ C 2x1 2x2 C µ x1 C α x22 where µ is the control parameter. Determine the normal form of this system in the vicinity of the bifurcation point (x1 , x2 , µ) D (0, 0, 0). a) Show that this system undergoes a generic saddle-node bifurcation as µ increases past zero when b ¤ 2 and α ¤ 2. b) Is there a bifurcation at (x1 , x2 , µ) D (0, 0, 0) when b ¤ 2 and α D 2? c) Show that this system undergoes a generic transcritical bifurcation as µ increases past zero when b D 2 and α ¤ 2. d) Show that this system undergoes a generic pitchfork bifurcation as µ increases past zero when b D 2 and α D 2. 4.7.41 Consider the system xP 1 D x1 2x2 C µ b C α 1 (x1 C 2x2 )2 xP 2 D 2x1 4x2 C µ b C α 2 (x1 x2 )2 Determine the type and normal form of the bifurcation that takes place at (x1 , x2 , µ) D (0, 0, 0) 4.7.42 Consider the system xP 1 D x1 2x2 C µ b C α(x1 C 2x2 )2 xP 2 D 2x1 4x2 C 2µ b C 29α(x1 x2 )2 Determine the type and normal form of the bifurcation that takes place at (x1 , x2 , µ) D (0, 0, 0) 4.7.43 Consider the system xP 1 D x1 2x2 C µ b C µ x1 C α(x1 C 2x2 )2 xP 2 D 2x1 4x2 C 2µ b C µ x2 C 32α(x1 x2 )2

4.7 Exercises

Determine the type and normal form of the bifurcation that takes place at (x1 , x2 , µ) D (0, 0, 0) 4.7.44 Consider the system xP D y C µ x C α x (x 2 C y 2 ) yP D x C µ y C α y (x 2 C y 2 ) What type of bifurcation occurs at (x, y, µ) D (0, 0, 0). Determine the bifurcating solutions. 4.7.45 Consider the system xP1 D µ C x1 x2 C 2µ x2 C x12 xP 2 D 2µ C 2x1 2x2 C µ x1 C x22 where µ is the control parameter. Determine the normal form of this system in the vicinity of the bifurcation point (x1 , x2 , µ) D (0, 0, 0). What is the type of this bifurcation? 4.7.46 Consider the system xP 1 D µ C x1 x2 C 2µ x2 C x12 xP 2 D µ C 2x1 2x2 C µ x1 C x22 where µ is the control parameter. Determine the normal form of this system in the vicinity of the bifurcation point (x1 , x2 , µ) D (0, 0, 0). What is the type of this bifurcation? 4.7.47 Consider the system xP 1 D µ C x1 x2 C 2µ x2 C x12 xP 2 D µ C 2x1 2x2 C µ x1 C 2x22 where µ is the control parameter. Determine the normal form of this system in the vicinity of the bifurcation point (x1 , x2 , µ) D (0, 0, 0). What is the type of this bifurcation? 4.7.48 Consider the system xP 1 D µ C x1 x2 C 2µ x2 C x12 xP 2 D b µ C 2x1 2x2 C µ x1 C α x22 where µ is the control parameter. Determine the normal form of this system in the vicinity of the bifurcation point (x1 , x2 , µ) D (0, 0, 0) for each of the following cases:

155

156

4 Bifurcations of Continuous Systems

a) b) c) d)

b b b b

¤ 2 and ¤ 2 and D 2 and D 2 and

α α α α

¤2 D2 ¤2 D2.

4.7.49 Consider the system xP 1 D x1 2x2 C µ b C α(x1 C 2x2 )2 xP 2 D 2x1 4x2 C 2µ b C 29α(x1 x2 )2 Determine the type and normal form of the bifurcation that takes place at (x1 , x2 , µ) D (0, 0, 0) 4.7.50 Consider the system xP 1 D x1 2x2 C µ b C µ x1 C α(x1 C 2x2 )2 xP 2 D 2x1 4x2 C 2µ b C µ x2 C 32α(x1 x2 )2 Determine the type and normal form of the bifurcation that takes place at (x1 , x2 , µ) D (0, 0, 0) 4.7.51 Consider the system xP 1 D x1 2x2 C µ b C α 1 (x1 C 2x2 )2 xP 2 D 2x1 4x2 C µ b C α 2 (x1 x2 )2 Determine the type and normal form of the bifurcation that takes place at (x1 , x2 , µ) D (0, 0, 0) 4.7.52 Consider the dynamical system xP D 32 µ C x 2y C x 2 yP D 2x 4y C x y Examine the bifurcation that takes place when (x, y, µ) D (0, 0, 0) and determine the normal form of the system near this bifurcation. 4.7.53 Consider the system xP 1 D µ C x1 x2 C 2µ x2 C x12 xP 2 D 2µ C 2x1 2x2 C µ x1 C 2x22

4.7 Exercises

where µ is the control parameter. Determine the normal form of this system in the vicinity of the bifurcation point (x1 , x2 , µ) D (0, 0, 0). What is the type of this bifurcation? 4.7.54 Consider the system xP D µ x y yP D x C µ y C α x 3 Examine the bifurcation that takes place at (x, y, µ) D (0, 0, 0) Is the bifurcation supercritical or subcritical? 4.7.55 Consider the system xP D µ x y C x y 2 yP D µ y C x C 2x 2 y where µ is the control parameter. Determine the normal form of this system in the vicinity of the bifurcation point (x, y, µ) D (0, 0, 0). What is the type of this bifurcation? 4.7.56 Consider the system xP 1 D x2 xP 2 D x1 C λx2 x12 Discuss the bifurcations of the ﬁxed points of this system as a function of the control parameter λ. Sketch the state space for λ D 0, λ > 0, and λ < 0. 4.7.57 Consider the system xP D µ x y C x y 2 yP D µ y C x where µ is the control parameter. Determine the normal form of this system in the vicinity of the bifurcation point (x, y, µ) D (0, 0, 0). What is the type of this bifurcation? 4.7.58 Consider the system xP D 2y C µ x C α x (x 2 C y 2 ) yP D 2x C µ y C α y (x 2 C y 2 ) What type of bifurcation occurs at (x, y, µ) D (0, 0, 0). Determine the bifurcating solutions. 4.7.59 Consider the Rössler equations (Rössler, 1976a): xP D (y C z) yP D x C a y zP D b C (x c)z

157

158

4 Bifurcations of Continuous Systems

Assume that the parameters a, b, and c are positive. a) When a is used as a control parameter, verify that a ﬁxed point of this system experiences a saddle-node bifurcation at c c c c2 . (x, y, z, a) D , , , 2 2a 2a 4b b) Simplify the three-dimensional system to the normal form for a saddle-node bifurcation in the vicinity of the bifurcation point. 4.7.60 Consider the Lorenz equations (Lorenz, 1963): xP D σ(y x) yP D x y x z zP D β z C x y Assume that the parameters σ, β, and are positive. a) Choose as the control parameter and examine the different local bifurcations experienced by the different ﬁxed points. Verify that a Hopf bifurcation of a ﬁxed point occurs at c D

σ(σ C β C 3) . σβ1

b) Construct the bifurcation diagram for 0 < c . c) Simplify the three-dimensional system to its normal form for a pitchfork bifurcation in the vicinity of (x, y, z, ) D (0, 0, 0, 1) and obtain uP D

σ σ(1 ) u u3 . σC1 β(σ C 1)

4.7.61 Consider the three-dimensional dynamical system xP 1 D b 1 µ x1 C x2 2x3 C Cb 11 µ x1 C α 1 (x1 C 2x2 C x3 )2

(4.297)

xP 2 D b 2 µ 2x1 x2 x3 C b 22 µ x2 C α 2 (x1 x2 C x3 )2

(4.298)

xP 3 D b 3 µ C x1 C 2x2 x3 C b 33 µ x3 C α 3 (2x1 C x2 x3 )2

(4.299)

Show that the origin of this system undergoes a saddle-node bifurcation as µ increases past zero when (b 3 C b 2 b 1 ) ¤ 0. Show that this system can be simpliﬁed to uP D

1 3

(b 3 C b 2 b 1 ) µ 2 C

1 3

(4α 3 C α 2 4α 1 ) u2

(4.300)

4.7 Exercises

Show that the origin of this system undergoes a transcritical bifurcation as µ increases past zero when (b 3 C b 2 b 1 ) D 0. Show that this system can be simpliﬁed to uP D Γ0 µ 21 C Γ1 µ 1 u C Γ2 u2

(4.301)

where Γ0 D Γ1 D

3b 2 b 11 3b 3 b 22 C 3b 2 b 33 C 3b 3 b 33 4b 22 α 1 C 4b 2 b 3 α 1 b 23 α 1 C4b 22 α 2 C 8b 2 b 3 α 2 C 4b 23 α 2 C b 22 α 3 4b 2 b 3 α 3 C 4b 23 α 3 1 27

1 9

(3b 11 C 3b 22 C 3b 33 8b 2 α 1 C 4b 3 α 1

4b 2 α 2 4b 3 α 2 4b 2 α 3 C 8b 3 α 3 ) Γ2 D

1 3

(4α 3 C α 2 4α 1 )

4.7.62 Consider the system xP D µ x 2y yP D µ y C 2x C λ 1 x z C λ 2 y z zP D z C α x 2 where α, µ, λ 1 , and τ are constants. Use a combination of center-manifold reduction and the method of normal forms to construct periodic solutions of this system for small µ. 4.7.63 Consider the system xP 1 D x1 C h(x3 ) xP 2 D h(x3 ) xP 3 D ax1 C b x2 c h(x3 ) where a, b, and c are positive constants and h(0) D 0 and y h(y ) > 0

for 0 < jy j < k

for some

k>0

a) Show that the origin is an isolated equilibrium point. b) Is the origin an asymptotically stable equilibrium point? c) Suppose that y h(y ) > 0. Is the origin globally asymptotically stable? 4.7.64 Consider the system xP D y yP D x C µ y α y z zP D z C x 2

159

160

4 Bifurcations of Continuous Systems

Show that the center manifold of the origin (x D 0, y D 0, µ D 0) is given by zD

1 5

3x 2 2x y C 2y 2

Then, determine the equations governing the dynamics on this manifold. 4.7.65 Consider the system xP D µ x y C x 2 α 1 x z yP D µ y C x C α 2 y z zP D z C x 2 C 2y 2 a) Show that the origin of this system undergoes a Hopf bifurcation as µ increases through zero. b) Determine the normal form of this bifurcation. c) Is the bifurcation supercritical or subcritical? d) Calculate the amplitude and frequency of the bifurcating limit cycle. 4.7.66 Consider the system xP 1 D µ x1 x2 xP 2 D µ x2 C x1 C 2x1 x3 xP 3 D x3 C x12 where µ is a small parameter. Design a controller to convert the bifurcation at (x1 , x2 , x3 , µ) D (0, 0, 0, 0) from subcritical to supercritical. 4.7.67 Consider the system uR C ω 2 u D µ uP α 1 uv α 2 u vP vP C v D δ u2 Determine the normal form of this system near (u, v ) D (0, 0) as µ passes through zero.

161

5 Forced Oscillations of the Dufﬁng Oscillator In this chapter, we consider the response of a single-degree-of-freedom system to a harmonic excitation modeled by uR C ω 2 u C 2µ uP C α u3 D F cos Ω t where is a small nondimensional parameter that is used as a bookkeeping device. Carrying out a straightforward expansion, one ﬁnds that up to O(), resonances occur if Ω ω: Primary or main resonance Ω 3ω: Subharmonic resonance of order one-third Ω 13 ω: Superharmonic resonance of order three. Next, we use the method of normal forms to determine second-order uniform approximations to the solutions of this equation for these resonances.

5.1 Primary Resonance

In the case of primary resonance, the linear theory shows that a small excitation leads to a large response. Hence, we need to determine the orders of F and u that will explicitly display this observation. One way of accomplishing this is to order the excitation at O() and rewrite the governing equation as uR C ω 2 u D 2µ uP C α u3 F cos Ω t

(5.1)

To apply the method of normal forms, we let z D F eiΩ t

(5.2)

N F cos Ω t D 12 (z C z)

(5.3)

so that

The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

162

5 Forced Oscillations of the Dufﬁng Oscillator

and zP D i Ω z

(5.4)

Hence, we transform the two-dimensional nonautonomous problem into a threedimensional autonomous problem. Next, to quantitatively describe the nearness of the primary resonance, we introduce a detuning parameter σ deﬁned according to Ω 2 D ω 2 C σ

(5.5)

Using (5.5) and (5.2) to eliminate ω 2 and F cos Ω t, respectively, from (5.1), we have uR C Ω 2 u D 2µ uP σ u C α u3 12 (z C z) N

(5.6)

Equation 5.6 can conveniently be represented as a ﬁrst-order complex-valued equation. To this end, we note that when, D 0, the solution of (5.6) can be expressed as N i Ω t u D Ae i Ω t C Ae where A is a constant. Then, N i Ω t uP D i Ω Ae i Ω t Ae When ¤ 0, A will be time-varying rather than constant. To represent (5.6) as a single complex-valued equation, we identify Ae i Ω t as ζ and hence introduce the transformation u D ζ C ζN

uP D i Ω ζ ζN

and

(5.7)

so that ζD

1 2

u

i uP Ω

and

1 ζN D 2

uC

i uP Ω

(5.8)

With this transformation, (5.6) becomes 3 1 i 2i Ω µ ζ ζN σ ζ C ζN C α ζ C ζN (z C z) ζP D i Ω ζ C N 2Ω 2 (5.9) We note that the unperturbed problem is transformed into the simple equation ζP D i Ω ζ under the transformation (5.7). Other transformations may lead to an N equation involving ζ and ζ. Next, we use the near-identity transformation ζ D η C h (η, η, N z, z) N C

(5.10)

5.1 Primary Resonance

so that (5.9) takes the simple form ηP D i Ω η C g (η, η, N z, zN ) C

(5.11)

Substituting (5.10) and (5.11) into (5.9), using (5.4), and equating the coefﬁcients of on both sides, we obtain @h @h @h @h g C iΩ η ηN C z zN h @η @ ηN @z @ zN 1 i 3 (η (η (z (η σ C η) N C α C η) N C z) N (5.12) D µ η) N C 2Ω 2 Next, we choose h to eliminate all of the nonresonance terms in (5.12), leaving g with the resonance and near-resonance terms. The right-hand side of (5.12) suggests choosing h in the form h D Λ 1 η 3 C Λ 2 η 2 ηN C Λ 3 η ηN 2 C Λ 4 ηN 3 C ∆ 1 η C ∆ 2 ηN C ∆ 3 z C ∆ 4 zN (5.13) Substituting (5.13) into (5.12) and rearranging yields iσ i α 3i α 2 η3 g C µη C η η ηN C z i 2Ω Λ 1 C 2Ω 2Ω 4Ω 2Ω 3α α 2 η ηN i 4Ω Λ 4 C ηN 3 i 2Ω Λ 3 C 2Ω 2Ω iσ 1 ηN i 2Ω ∆ 4 zN D 0 2i Ω ∆ 2 C µ 2Ω 4Ω

(5.14)

We note that (5.14) does not depend on ∆ 1 , ∆ 3 , and Λ 2 , and hence they are arbitrary, and the terms proportional to η, η 2 η, N and z are resonance terms. Choosing Λ 1 , Λ 3 , Λ 4 , ∆ 2 , and ∆ 4 to eliminate the nonresonance terms, we have iµ 1 σ , ∆4 D C 2Ω 4Ω 2 8Ω 2 α 3α α , Λ3 D , Λ4 D Λ1 D 4Ω 2 4Ω 2 8Ω 2

∆2 D

(5.15) (5.16)

With the choices (5.15) and (5.16), (5.14) yields g as g D µ η

i iσ 3i α 2 ηC η ηN z 2Ω 2Ω 4Ω

(5.17)

which, upon substitution into (5.11), yields the normal form ηP D i Ω η µ η

i 3iα 2 iσ ηC η ηN z 2Ω 2Ω 4Ω

(5.18)

Substituting (5.13) into (5.10) and then substituting the result into (5.7), we obtain N 2 C Λ 3 η ηN 2 u D η C ηN C (Λ 1 C Λ 4 ) η 3 C ηN 3 C (Λ 2 C Λ 3 ) η 2 ηN C Λ N 3 C ∆ 4 zN C N 2 η C ∆2 C ∆ N 1 ηN C (∆ 3 C ∆ 4 ) z C ∆ C ∆1 C ∆ (5.19)

163

164

5 Forced Oscillations of the Dufﬁng Oscillator

If we express η in the polar form η D 12 ae i (Ω tCγ )

(5.20)

we ﬁnd that, to O(1), the amplitude of the fundamental frequency Ω of oscillations N η ηN 2 , η, η, N z, is a. However, some of the ﬁrst-order terms in (5.19), namely, η 2 η, and zN , also have the frequency Ω , and hence modify a and make it nonunique. In order to uniquely deﬁne the amplitude of the term at the fundamental frequency Ω of oscillations by a, we choose Λ 2 , ∆ 1 , and ∆ 3 to eliminate the terms involving z, zN , η, η, N η 2 η, N and η ηN 2 in (5.19). With this choice, (5.19) becomes u D η C ηN C

α 3 η C ηN 3 C 8Ω 2

(5.21)

on account of (5.16). In terms of the polar coordinates (5.20), (5.21) becomes u D a cos (Ω t C γ ) C

α a 3 cos (3Ω t C 3γ ) C 32Ω 2

(5.22)

Substituting (5.2) and (5.20) into (5.18) and separating real and imaginary parts, we obtain F sin γ 2Ω

aP D µ a a γP D

3α a 3 F σ aC cos γ 2Ω 8Ω 2Ω

(5.23) (5.24)

5.2 Subharmonic Resonance of Order One-Third

To express the nearness of Ω to 3ω, we introduce the detuning parameter σ deﬁned as 1 Ω2 9

D ω 2 C σ

(5.25)

Moreover, because this resonance is called secondary resonance, we need to order F cos Ω t at O(1) so that its inﬂuence is accounted for in the transformed equation. Consequently, using (5.3) to replace F cos Ω t with 1/2(z C zN ) and using (5.25) to replace ω 2 with 1/9Ω 2 σ, we rewrite (5.1) as uR C 19 Ω 2 u D

1 2

(z C z) N 2µ uP σ u C α u3

(5.26)

To express (5.26) in complex-valued form, we let u D ζ C ζN

and

uP D 13 i Ω ζ ζN

(5.27)

5.2 Subharmonic Resonance of Order One-Third

so that ζD

1 2

3i u uP Ω

and

ζN D

1 2

3i uC uP Ω

(5.28)

Substituting (5.27) and (5.28) into (5.26) yields 1 3i (z C z) ζP D i Ω ζ N 3 4Ω 3 3i 2 C i µ Ω ζ ζN σ ζ C ζN C a ζ C ζN 2Ω 3

(5.29)

Because the frequency of the response is 1/3Ω and because z and zN have the frequencies Ω and Ω , z C zN is not a resonance term in this case and because it appears at O(1), we ﬁrst introduce a linear transformation to remove it from (5.29) at O(1). To this end, we let ζ D η C Γ1 z C Γ2 zN

(5.30)

in (5.29) and obtain ηP D

3i 1 1 (z C z) i Ω η C i Ω (Γ1 z C Γ2 zN ) Γ1 zP Γ2 zPN N C O() (5.31) 3 3 4Ω

Using (5.4) to eliminate zP and zPN from (5.31) yields ηP D

3i 1 1 (z C zN ) C O() i Ω η C i Ω (Γ1 z C Γ2 zN ) i Ω Γ1 z C i Ω Γ2 zN 3 3 4Ω (5.32)

Next, we choose Γ1 and Γ2 to eliminate z and zN from (5.32) and obtain Γ1 D

9 8Ω 2

Γ2 D

and

9 16Ω 2

(5.33)

Substituting (5.30) into (5.29) and using (5.4) and (5.33), we obtain 1 3i ηP D i Ω η C 3 2Ω σ η C ηN

"

2 27 zN 27z C i µ Ω η ηN 3 16Ω 2 16Ω 2 3 # 9z 9z 9 zN 9 zN C α η C ηN 16Ω 2 16Ω 2 16Ω 2 16Ω 2 (5.34)

To simplify (5.34), we introduce the near-identity transformation η D ξ C h ξ , ξN , z, zN

(5.35)

165

166

5 Forced Oscillations of the Dufﬁng Oscillator

and obtain @h PN @h 1 1 @h P @h ξ ξ zP zPN ξP D i Ω ξ C iΩ h N 3 3 @ξ @z @ zN @ξ " 3 3i 9 zN 9z C α ξ C ξN 2Ω 16Ω 2 16Ω 2 9z 9 zN N σ ξ C ξ 16Ω 2 16Ω 2 # 2 27 zN 27z N C iµΩ ξ ξ C C 3 16Ω 2 16Ω 2

(5.36)

Substituting (5.35) into (5.30) and then substituting the result into (5.27), we have

9 N C (z u D ξ C ξN C z) N C h C h (5.37) 16Ω 2 on account of (5.33). To be strictly consistent, we should not include the term (h C N in (5.37) if we are not going to keep the resonance terms in (5.36) that arise h) from h. The form of the O() terms in (5.36) suggests choosing h in the form h D ∆ 1 ξ C ∆ 2 ξN C ∆ 3 z C ∆ 4 zN C Λ 1 ξ 3 C Λ 2 ξ 2 ξN C Λ 3 ξ ξN 2 C Λ 4 ξN 3 C Λ 5 z 3 C Λ 6 z 2 zN C Λ 7 z zN 2 C Λ 8 zN 3 C Λ 9 ξ 2 z C Λ 10 ξ 2 zN C Λ 11 ξN 2 z C Λ 12 ξN 2 zN C Λ 13 ξ z 2 C Λ 14 ξN z 2 C Λ 15 ξ zN 2 C Λ 16 ξN zN 2 C Λ 17 ξ z zN (5.38) C Λ 18 ξN z zN C Λ 19 ξ ξN z C Λ 20 ξ ξN zN Because zP D i Ω z and, to the ﬁrst approximation, ξP D 1/3i Ω ξ , out of all of N and ξN 2 z are resonance the terms in (5.36), only the terms involving ξ , ξ 2 ξN , ξ z z, terms and hence the coefﬁcients ∆ 1 , Λ 2 , Λ 11 , and Λ 17 , are arbitrary. Moreover, as N in (5.37) if we are going aforementioned, we are not justiﬁed in including (h C h) to eliminate only the resonance terms from (5.36). Substituting (5.38) into (5.36) and using (5.4), we can choose the Λ m to eliminate all of the terms except the resonance terms. Consequently, we obtain the simple equation 243α 1 27α N 2 3iσ 3i 3α ξ 2 ξN C C ξP D i Ω ξ µ ξ ξ z z N z ξ ξC 3 2Ω 2Ω 128Ω 4 16Ω 2 (5.39) Substituting (5.2) and the polar form ξ D 12 ae i ( 3 Ω tCγ ) 1

(5.40)

into (5.37) and (5.39) and separating real and imaginary parts, we obtain to the second approximation 1 9F cos Ω t C (5.41) u D a cos Ωt C γ 3 8Ω 2

5.3 Superharmonic Resonance of Order Three

where aP D µ a a γP D

81α F 2 a sin 3γ 64Ω 3

(5.42)

9α 3 81α F 2 3σ 729F 2 aC a cos 3γ aC a 2Ω 256Ω 5 8Ω 64Ω 3

(5.43)

5.3 Superharmonic Resonance of Order Three

To express the nearness of ω to 3Ω , we introduce the detuning parameter σ deﬁned according to 9Ω 2 D ω 2 C σ

(5.44)

As in the case of subharmonic resonance of order one-third, we need to order the excitation at O(1) so that the effect of this resonance will occur at the same order as the effects of the damping and nonlinearity. Substituting (5.44) into (5.1), using (5.3), and assuming that z D O(1), we obtain uR C 9Ω 2 u D

1 2

(z C zN ) 2µ uP σ u C α u3

(5.45)

To express (5.45) in complex-valued form, we let u D ζ C ζN

and

uP D 3i Ω ζ ζN

(5.46)

so that ζD

1 2

u

i uP 3Ω

and

1 ζN D 2

uC

i uP 3Ω

(5.47)

With this transformation, (5.45) becomes i (z C zN ) ζP D 3i Ω ζ 12Ω 3 i i h C 6i µ Ω ζ ζN σ ζ C ζN C α ζ C ζN 6Ω

(5.48)

The ﬁrst step is to introduce a linear transformation to eliminate the term proportional to z C zN at O(1). To this end, we let ζ D η C Γ1 z C Γ2 zN

(5.49)

in (5.48) and obtain N Γ1 zP Γ2 zPN ηP D 3i Ω η C 3i Ω (Γ1 z C Γ2 z)

i (z C z) N C O() (5.50) 12Ω

167

168

5 Forced Oscillations of the Dufﬁng Oscillator

Using (5.4), we express zP as i Ω z and zNP as i Ω zN in (5.50) and obtain i i z C 4 i Ω Γ2 zN C O() ηP D 3i Ω η C 2i Ω Γ1 12Ω 12Ω

(5.51)

We choose Γ1 and Γ2 to eliminate the terms involving z and zN in (5.51); that is, we put Γ1 D

1 24Ω 2

and

Γ2 D

1 48Ω 2

(5.52)

Substituting (5.49) into (5.48) and using (5.4) and (5.52), we obtain " i zN z ηP D 3i Ω η C 6i µ Ω η ηN C 6Ω 48Ω 2 48Ω 2 3 # z z zN zN C α η C ηN C σ η C ηN C C C 16Ω 2 16Ω 2 16Ω 2 16Ω 2 (5.53) To simplify (5.53), we introduce the near-identity transformation η D ξ C h ξ , ξN , z, zN

(5.54)

and obtain @h P @h PN @h @h ξP D 3i Ω ξ C 3iΩ h ξ ξ zP zPN @ξ @z @ zN @ ξN " i zN z C 6i µ Ω ξ ξN C 6Ω 48Ω 2 48Ω 2 zN z C σ ξ C ξN C 16Ω 2 16Ω 2 3 # z N z C C Cα ξ C ξN C 16Ω 2 16Ω 2

(5.55)

Next, we choose h to eliminate the nonresonance terms. As discussed in the preceding section, if we want to stop at second order, we do not need to determine h and all that we need is to keep the resonance terms in (5.55). The result is z3 Pξ D 3i Ω ξ µ ξ iσ ξ C iα 3ξ 2 ξN C 3z zN ξ C (5.56) 6Ω 6Ω 128Ω 4 4096Ω 6 Substituting (5.54) into (5.49) and then substituting the result into (5.46), we have 1 (z C zN ) C O() u D ξ C ξN C (5.57) 16Ω 2 Expressing ξ in the polar form ξ D 12 ae i (3Ω tCγ )

(5.58)

5.4 An Alternate Approach

and using (5.2) to replace z with F e i Ω t , we rewrite (5.57) as u D a cos (3Ω t C γ ) C

F cos Ω t C 8Ω 2

(5.59)

Substituting (5.58) and (5.2) into (5.56) and separating real and imaginary parts, we obtain aP D µ a C a γP D

α F 3 sin γ 12 288Ω 7

(5.60)

α 3 α F 3 σ α F 2 a C C cos γ aC a 6Ω 256Ω 5 8Ω 12 288Ω 7

(5.61)

5.4 An Alternate Approach

In the preceding sections, we used the detuning relationships (5.5), (5.25), and (5.44) to replace ω 2 in terms of Ω 2 , 1/9Ω 2 , and 9Ω 2 in (5.1). Alternatively, we can have ω as it is in the governing equation and watch for near-resonance terms. We describe this approach for the cases of subharmonic and superharmonic resonances of order one-third and three, respectively. In these cases, F is assumed to be O(1). To express (5.1) in complex-valued form we let u D ζ C ζN and uP D i ω ζ ζN (5.62) so that ζD

1 2

u

i uP ω

and

1 ζN D 2

uC

i uP ω

(5.63)

Using (5.3), (5.62), and (5.63) and the fact that F D O(1), we rewrite (5.1) as 3 i i i h (z C zN ) C ζP D i ωζ (5.64) 2i ωµ ζ ζN C α ζ C ζN 4ω 2ω Again, the ﬁrst step is to introduce a linear transformation to eliminate the term proportional to z C zN . To this end, we substitute (5.49) into (5.64) and obtain i (z C zN ) C O() ηP D i ωη C i ω (Γ1 z C Γ2 z) N Γ1 zP Γ2 zPN 4ω

(5.65)

Using (5.4) to replace zP and zPN with i Ω z and i Ω zN , we rewrite (5.65) as 1 1 z C i ωΓ2 C Ω Γ2 zN C O() ηP D i ωη C i ωΓ1 Ω Γ1 4ω 4ω (5.66) Next, we choose Γ1 and Γ2 to eliminate the terms involving z and zN in (5.66); that is, Γ1 D

1 4ω(ω Ω )

and

Γ2 D

1 4ω(ω C Ω )

(5.67)

169

170

5 Forced Oscillations of the Dufﬁng Oscillator

Substituting (5.49) into (5.64) and using (5.67), we obtain ( i Ω (z zN ) ηP D i ωη C 2i ωµ η ηN C 2ω 2ω(ω 2 Ω 2 ) ) 3 z C zN Cα η C ηN C 2(ω 2 Ω 2 )

(5.68)

To simplify (5.68), we use the near-identity transformation (5.54) and obtain @h PN @h @h P @h ξ ξ zP zPN ξP D i ωξ C iωh @ξ @z @ zN @ ξN ( i Ω (z zN ) C 2i ωµ ξ ξN C 2ω 2ω(ω 2 Ω 2 ) 3 ) z C zN N Cα ξ C ξ C C 2(ω 2 Ω 2 )

(5.69)

Then, we choose h to eliminate all of the terms in (5.69) except the resonance and near-resonance terms. The resonance terms correspond to the terms that produce secular terms, whereas the near-resonance terms correspond to terms that produce small-divisor terms in the applications of the method of multiple scales (Nayfeh, 1973, 1981) and the Krylov–Bogoliubov–Mitropolsky technique (Bogoliubov and Mitropolsky, 1961). The resonance terms in (5.69) are the terms proportional to ξ , ξ 2 ξN , and ξ z z. N The near-resonance terms depend on the resonance being considered. In the subharmonic-resonance case of order one-third, the term proportional to z ξN 2 is the near-resonance term, whereas in the superharmonic-resonance case of order three, the term proportional to z 3 is the near-resonance term. Therefore, keeping the resonance and near-resonance terms in (5.69), we obtain iα 3z zN 3z N 2 (5.70) 3ξ 2 ξN C ξ ξP D i ωξ µ ξ C ξ C 2ω 2(ω 2 Ω 2 )2 2(ω 2 Ω 2 ) when Ω 3ω (i.e., when there is a subharmonic resonance of order one-third), and 3z zN iα z3 3ξ 2 ξN C (5.71) ξP D i ωξ µ ξ C ξ C 2ω 2(ω 2 Ω 2 )2 8(ω 2 Ω 2 )3 when Ω 1/3ω (i.e., when there is a superharmonic resonance of order three). Again, to second order, we do not need to determine h. Then, substituting (5.54) and (5.49) into (5.62) and using (5.67), we obtain u D ξ C ξN C

1 (z C zN ) C 2(ω 2 Ω 2 )

(5.72)

Next, substituting the polar form ξD

1 i (ω tCβ ) ae 2

(5.73)

5.4 An Alternate Approach

into (5.72) and using the fact that z D F e i Ω t , we have u D a cos (ωt C β) C

ω2

F cos Ω t C Ω2

(5.74)

5.4.1 Subharmonic Case

For the subharmonic-resonance case, substituting (5.73) into (5.70), separating real and imaginary parts, and using the fact that z D F e i Ω t , we obtain aP D µ a C a βP D

3α Fa 2 sin 3γ 8ω(ω 2 Ω 2 )

3α F 2 a 3α Fa 2 3α 3 C a C cos 3γ 2 2 2 8ω 4ω(ω Ω ) 8ω(ω 2 Ω 2 )

(5.75) (5.76)

where 3γ D 3β (Ω 3ω) t

(5.77)

Substituting for β from (5.77) into (5.74) yields F 1 cos Ω t C Ωt C γ C 2 u D a cos 3 ω Ω2

(5.78)

while substituting for β from (5.77) into (5.76) yields 3α F 2 a 3α Fa 2 3α 3 1 C a C cos 3γ a γP D (Ω 3ω) a C 2 2 2 3 8ω 4ω(ω Ω ) 8ω(ω 2 Ω 2 ) (5.79) Therefore, to the second approximation, u is given by (5.78), where a and γ are given by the autonomous equations (5.75) and (5.79). These equations are in full agreement with those obtained by using the methods of multiple scales and averaging. To compare the expansion obtained in this section with that obtained in Section 5.2, we solve for ω in terms of Ω from (5.25) and obtain ω2 D

1 2 Ω σ 9

or

ωD

1 3σ Ω C 3 2Ω

(5.80)

Substituting for ω from (5.80) into (5.78), (5.75), and (5.79), we obtain (5.41)–(5.43) to O().

171

172

5 Forced Oscillations of the Dufﬁng Oscillator

5.4.2 Superharmonic Case

In this case, substituting (5.73) and (5.2) into (5.71) and separating real and imaginary parts, we obtain aP D µ a C a βP D

α F 3 sin γ 8ω(ω 2 Ω 2 )3

(5.81)

3α F 2 a α F 3 3α 3 C cos γ a C 2 2 2 8ω 4ω(ω Ω ) 8ω(ω 2 Ω 2 )3

(5.82)

where γ D β (3Ω ω) t

(5.83)

Using (5.83) to eliminate β from (5.74) yields u D a cos (3Ω t C γ ) C

F cos Ω t C ω2 Ω 2

(5.84)

while using (5.83) to eliminate β from (5.82) yields a γP D (3Ω ω) a C

3α F 2 a α F 3 3α 3 C cos γ a C 8ω 4ω(ω 2 Ω 2 )2 8ω(ω 2 Ω 2 )3 (5.85)

Therefore, to the second approximation, u is given by (5.84) where a and γ are given by the autonomous equations (5.81) and (5.85). This approximation is in full agreement with that obtained by using the methods of multiple scales and averaging. To compare the expansion obtained in this section with that obtained in Section 5.3, we use (5.44) to solve for ω in terms of Ω and obtain ω 2 D 9Ω 2 σ

or

ω D 3Ω

σ C 6Ω

(5.86)

Using (5.86) to eliminate ω from (5.84), (5.81), and (5.85), we obtain (5.59)–(5.61) to O().

5.5 Exercises

5.5.1 Consider the system uR C ω 2 u D uP 13 uP 3 C F cos Ω t

5.5 Exercises

Use the method of normal forms to determine an approximation to this system when Ω ω, 3ω, and 1/3ω. 5.5.2 Use the method of normal forms to determine a second-order approximation to the solution of uR C ω 2 C 2µ uP C α u2 D F cos Ω t when Ω 2ω.

173

175

6 Forced Oscillations of SDOF Systems 6.1 Introduction

In this chapter, we consider small- but ﬁnite-amplitude responses of a singledegree-of-freedom system with quadratic and cubic nonlinearities to a harmonic excitation; speciﬁcally, we consider solutions of uR C ω 2 u C 2µ uP C δ u2 C α u3 D F cos Ω t

(6.1)

The results of Section 1.5 show that, to the second approximation, the free undamped solution of (6.1) can be expressed as u D a cos ( ωt O C β) C

δ a2 [cos (2 ωt O C 2β) 3] C 6ω 2

where a and β are constants and the frequency ω O of free oscillations is given by 2 3α 5δ ω O DωC a2 C 8ω 12ω 3 Consequently, as the oscillation amplitude a increases, the undamped frequency ω O of oscillation is shifted from the undamped linear frequency ω of oscillation by 3α 5δ 2 a2 8ω 12ω 3 To keep track of the different orders of magnitude, we introduce a small nondimensional parameter as a bookkeeping device and scale the quadratic terms at O() and the cubic term at O( 2 ). The resonances produced by the excitation try to make the response very large. But as the response grows, the nonlinearity, which modiﬁes the natural frequency of the system, and the damping, which dissipates part of the input energy, are activated, thereby limiting the response at small but ﬁnite amplitude. Because the quadratic and cubic nonlinearities produce a resonance term ﬁrst at O( 2 ) according to Section 1.5, we scale the damping at O( 2 ). With these scalings, we rewrite (6.1) as uR C ω 2 u D F cos Ω t δ u2 2 2µ uP C α u3 (6.2) The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

176

6 Forced Oscillations of SDOF Systems

Because the appearance of resonance terms produced by the excitation depends on the excitation magnitude F and the type of resonance, we scale F differently, depending on the resonance being considered. Carrying out a straightforward expansion, one ﬁnds the following resonances:

Ω Ω Ω Ω Ω

ω: Primary or main resonance 2ω: Subharmonic resonance of order one-half 12 ω: Superharmonic resonance of order two 3ω: Subharmonic resonance of order one-third 13 ω: Superharmonic resonance of order three.

To treat these cases and determine the appropriate scaling of F for each resonance, we cast (6.2) in complex-valued form by letting u D ζ C ζN and uP D i ω ζ ζN (6.3) and z D F eiΩ t

(6.4)

zP D i Ω z

(6.5)

so that

Using (6.3)–(6.5), we rewrite (6.2) as 2 i 2 h 3 i iδ i (z C z)C ζ C ζN C N 2i µ ω ζ ζN Cα ζ C ζN ζP D i ωζ 4ω 2ω 2ω (6.6)

6.2 Primary Resonance

In this case Ω ω and we need to scale F at O( 2 ) so that the resonance term produced by the excitation appears at the same order as those produced by the damping and the nonlinearity. With this scaling, we rewrite (6.6) as 2 Pζ D i ωζ C iδ ζ C ζN 2 2 µ ζ ζN C i α ζ C ζN 3 1 (z C z) N 2ω 2ω 2 (6.7) To determine a second-order uniform expansion of (6.7), we let N C 2 h 2 (η, η, N z, zN ) C ζ D η C h 1 (η, η)

(6.8)

N C 2 g 2 (η, η, N z, zN ) C ηP D i ωη C g 1 (η, η)

(6.9)

where

6.2 Primary Resonance

and the g i represent resonance and near-resonance terms that cannot be eliminated by nonsingular choices of the h i . Substituting (6.8) and (6.9) into (6.7) and equating coefﬁcients of like powers of , we obtain iδ @h 1 @h 1 (η C η) (6.10) η ηN h 1 D N 2 g1 C i ω @η @ ηN 2ω @h 2 @h 2 @h 2 @h 2 η ηN h 2 C i Ω z zN g2 C i ω @η @ ηN @z @ zN @h 1 iδ @h 1 (η C η) N C g1 gN 1 µ (η η) N h 1 C hN 1 D @η @ ηN ω 1 i α (η C η) N 3 (z C zN ) (6.11) C 2ω 2 Because we will proceed to the next order, we need to explicitly determine h 1 . The right-hand side of (6.10) suggests seeking h 1 in the form h 1 D Γ1 η 2 C Γ2 η ηN C Γ3 ηN 2

(6.12)

Substituting (6.12) into (6.10) yields δ δ δ η 2 C ωΓ2 C η ηN C 3ωΓ3 C ηN 2 D 0 (6.13) g 1 ωΓ1 2ω ω 2ω N and ηN 2 can be eliminated from (6.13) if Hence, the terms proportional to η 2 , η η, Γ1 D

δ , 2ω 2

Γ2 D

δ , ω2

Γ3 D

δ 6ω 2

(6.14)

Then, (6.13) reduces to g 1 D 0. Substituting (6.12) and (6.14) into (6.11) and using the fact that g 1 D 0, we obtain @h 2 @h 2 @h 2 @h 2 η ηN h 2 C i Ω z zN D µ (η η) N @η @ ηN @z @ zN 2 i 1 i δ2 3 2 (η (η (z ) α C η) N C η) N η C η N 6η η N C C z N C 3ω 3 2ω 2 (6.15)

g2 C i ω

The usual approach for determining the normal form is to choose h 2 to eliminate as many terms as possible from (6.15), thereby leaving g 2 with the remaining terms. Alternatively, we choose g 2 to eliminate the resonance and near-resonance terms in (6.15) knowing that we can always eliminate the nonresonance terms by a proper choice of h 2 . Because we are stopping at this order (seeking a second-order approximation), we do not need to solve explicitly for h 2 and all that we need to do is choose g 2 to eliminate the resonance and near-resonance terms in (6.15); that is, we put 10δ 2 i i 3α η 2 ηN g 2 D µ η C z 2ω 3ω 2 4ω

177

178

6 Forced Oscillations of SDOF Systems

Substituting for g 1 and g 2 into (6.9), we obtain the normal form ηP D i ωη 2 µ η C

i 2 2ω

10δ 2 i 2 3α η 2 ηN z 2 3ω 4ω

(6.16)

Substituting (6.12) and (6.14) into (6.8) and then substituting the result into (6.3), we obtain to the second approximation u D η C ηN C

δ 2 η C ηN 2 6η ηN C 3ω 2

(6.17)

where η is given by (6.16). Finally, we introduce the polar representation η D 12 ae i (ω tCβ )

(6.18)

into (6.17) and obtain u D a cos(ωt C β) C

δ a 2 cos(2ωt C 2β) 3 C 6ω 2

(6.19)

Substituting (6.18) and (6.4) into (6.16) and separating real and imaginary parts, we obtain 2 F sin [(Ω ω) t β] 2ω 3α 2 F 5δ 2 3 a cos [(Ω ω) t β] a βP D 2 8ω 12ω 3 2ω aP D 2 µ a C

(6.20) (6.21)

Equations 6.19–6.21 are the same as those obtained by using the methods of multiple scales and averaging (Nayfeh and Mook, 1979).

6.3 Subharmonic Resonance of Order One-Half

Because z and zN are nonresonance terms when Ω is away from ω, we ﬁrst introduce the linear transformation ζ D η C ∆ 1 z C ∆ 2 zN

(6.22)

into (6.6) and obtain ηP D i ωη C i ω (∆ 1 z C ∆ 2 zN ) ∆ 1 zP ∆ 2 zPN

i (z C zN ) C O() 4ω

(6.23)

6.3 Subharmonic Resonance of Order One-Half

Using (6.5) to express zP and zPN as i Ω z and i Ω zN , we rewrite (6.23) as 1 1 z C i (ω C Ω ) ∆ 2 zN C O() ηP D i ωη C i (ω Ω ) ∆ 1 4ω 4ω (6.24) We choose ∆ 1 and ∆ 2 to eliminate the terms involving z and zN ; that is, ∆1 D

1 4ω(ω Ω )

and

∆2 D

1 4ω(ω C Ω )

(6.25)

It follows from (6.25) that the transformation (6.22) is singular or near singular when Ω ω; that is, z and zN would be resonance terms in this case. This is the reason we scaled F and hence z and zN at O( 2 ) in the preceding section. Substituting (6.22) and (6.25) into (6.6) yields 2 z C zN iδ η C ηN C 2 µ (η η) N ηP D i ωη C 2ω 2(ω 2 Ω 2 ) " 3 # i 2 2i µ Ω (z zN ) z C zN C (6.26) C α η C ηN C 2ω (ω 2 Ω 2 ) 2(ω 2 Ω 2 ) When Ω 2ω, the excitation produces the near-resonance term z η. N In order to have this term appear at the same order at which the nonlinear resonance term η 2 ηN appears, we rescale F and hence z as O() and rewrite (6.26) as ηP D i ωη C

iδ i 2 δ (η C η) (η C η) N N 2C N (z C z) N 2 µ (η η) 2ω 2ω(ω 2 Ω 2 )

i 2 α (6.27) (η C η) N 3 C 2ω Because there are no resonance or near-resonance terms at O() in (6.27), we can eliminate the O() terms, as in the preceding section, by using the near-identity transformation C

ηDξC

δ 2 N 2 3ξ ξ 6ξ ξN 6ω 2

(6.28)

Then, (6.27) becomes i 2 δ ξ C ξN (z C zN ) 2 µ ξ ξN 2ω(ω 2 Ω 2 ) 3 i 2 δ 2 i 2 α ξ C ξN ξ 2 C ξN 2 6ξ ξN C ξ C ξN C 3 2ω 3ω

ξP D i ωξ C

(6.29)

To simplify (6.29), we can introduce a near-identify transformation to eliminate the nonresonance terms. Because we are stopping at this order, we need to keep only the resonance terms ξ and ξ 2 ξN and near-resonance term z ξN . The result is 10δ 2 i 2 δ i 2 ξP D i ωξ 2 µ ξ C 3α ξ 2 ξN C z ξN (6.30) 2 2ω 3ω 2ω(ω 2 Ω 2 )

179

180

6 Forced Oscillations of SDOF Systems

Substituting (6.22) and (6.28) into (6.3), using (6.25), and recalling that F is scaled at O(), we obtain to the second approximation u D ξ C ξN C

(z C z) N δ 2 ξ C ξN 2 6ξ ξN C C 2 2 2 2(ω Ω ) 3ω

(6.31)

where ξ is given by (6.30). Substituting the polar forms (6.4) and (6.18) into (6.31) yields u D a cos (ωt C β) C

F δ a 2 [cos (2ωt C 2β) 3] C cos Ω t C ω2 Ω 2 6ω 2 (6.32)

while substituting (6.4) and (6.18) into (6.30) and separating real and imaginary parts yields 2 δ F sin [2β (Ω 2ω) t] 2ω(ω 2 Ω 2 ) 3α 2 δ F 5δ 2 3 a C cos [2β (Ω 2ω) t] a βP D 2 8ω 12ω 3 2ω(ω 2 Ω 2 ) aP D 2 µ a C

(6.33) (6.34)

Equations 6.32–6.34 are in full agreement with those obtained by using the method of multiple scales (Nayfeh, 1986).

6.4 Superharmonic Resonance of Order Two

In this case, Ω 1/2ω. Inspection of (6.26) reveals that the resonance term arising from the excitation is z 2 . In order that its inﬂuence balance the resonance terms arising from the damping and the nonlinearity, we scale F and hence z at O( 1/2 ) and then rewrite (6.26) as ηP D i ωη C C

iδ i 3/2 δ (η C η) (η C η) N N 2C N (z C z) N 2 µ (η η) 2ω 2ω(ω 2 Ω 2 )

i 2 α i 2 δ(z C zN )2 (η C η) C N 3 C 2 2 2 8ω(ω Ω ) 2ω

(6.35)

As in the preceding section, we introduce the near-identity transformation (6.28) to eliminate the O() terms in (6.35) and hence obtain i 3/2 δ ξ C ξN (z C zN ) 2 µ ξ ξN 2 2 2ω(ω Ω ) N 2 i 2 δ 2 i 2 δ(z C z) ξ C ξN ξ 2 C ξN 2 6ξ ξN C C 2 2 2 3 8ω(ω Ω ) 3ω 3 i 2 α ξ C ξN C C 2ω

ξP D i ωξ C

(6.36)

6.4 Superharmonic Resonance of Order Two

Because there are no resonance terms at O( 3/2 ), we can introduce a near-identity transformation ξ D χ C 3/2 h (χ, χN , z, zN )

(6.37)

to eliminate these terms and rewrite (6.36) as i 2 δ 2 (χ C χN ) χ 2 C χN 2 6χ χN 2 µ (χ χN ) 3ω 3 i 2 δ i 2 α (z C z) (χ C χN )3 C N 2 C C 2ω 8ω(ω 2 Ω 2 )2

χP D i ωχ C

(6.38)

Because our aim is to obtain an expression that is valid to O( 2 ), we do not need to determine h. Moreover, we also know that we can choose a near-identity transformation N z, z) N χ D w C 2 g (w, w,

(6.39)

to eliminate the nonresonance terms from (6.38) and obtain 10δ 2 i 2 i 2 δ 2 2 3α w w N C z 2 C (6.40) wP D i ωw µ w C 2ω 3ω 2 8ω(ω 2 Ω 2 )2 Again, for an expression that is valid to O( 2 ), we do not need to determine g. Substituting (6.22), (6.25), (6.28), (6.37), and (6.39) into (6.3), we have u D w C wN C

δ 2 1/2 (z C zN ) w C wN 2 6w wN C C 2 2 2 2(ω Ω ) 3ω

(6.41)

where use has been made of the assumption that F and hence z are O( 1/2 ). Expressing w in the polar form wD

1 i (ω tCβ ) ae 2

(6.42)

we rewrite (6.41) as u D a cos (ωt C β) C

δ a 2 1/2 F [cos (2ωt C 2β) 3] C cos Ω t C 2 2 ω Ω 6ω 2 (6.43)

Substituting (6.4) and (6.42) into (6.40) and separating real and imaginary parts, we obtain 2 δ F 2 sin [β (2Ω ω) t] 4ω(ω 2 Ω 2 )2 3α 2 δ F 2 5δ 2 3 a C cos [β (2Ω ω) t] a βP D 2 8ω 12ω 3 4ω(ω 2 Ω 2 )2

aP D 2 µ a C

(6.44) (6.45)

Equations 6.43–6.45 are in full agreement with those obtained by using the method of multiple scales (Nayfeh, 1986).

181

182

6 Forced Oscillations of SDOF Systems

6.5 Subharmonic Resonance of Order One-Third

In this case, Ω 3ω. Inspection of (6.26) shows that the resonance term produced by the excitation is z ηN 2 . Because this resonance term occurs at O( 2 ), F and hence z are O(1). Because none of the O() terms is resonance in this case, one can introduce a near-identity transformation in the form η D ξ C g ξ , ξN , z, zN

(6.46)

and eliminate them. The form of the O() terms suggests seeking g in the form g D Λ 1 ξ 2 C Λ 2 ξ ξN C Λ 3 ξN 2 C Λ 4 z 2 C Λ 5 z zN C Λ 6 zN 2 C Λ 7 ξ z C Λ 8 ξ zN (6.47) C Λ 9 ξN z C Λ 10 ξN zN Substituting (6.46) into (6.26) shows that ξP D i ωξ C O()

(6.48)

Consequently, substituting (6.46) and (6.47) into (6.26) and replacing zPN and zP with i Ω zN and i Ω z and ξPN and ξP with i ω ξN and i ωξ on the right-hand side of the resulting equation, we obtain to O() δ δ ξP D i ωξ C i ωΛ 1 C ξ 2 C i ωΛ 2 C ξ ξN 2ω ω δ N2 δ C i 3ωΛ 3 C ξ C i ωΛ 5 C z zN 2ω 4ω(ω 2 Ω 2 )2 δ C i (ω 2Ω ) Λ 4 C z2 8ω(ω 2 Ω 2 )2 δ zN 2 C i (ω C 2Ω ) Λ 6 C 8ω(ω 2 Ω 2 )2 δ ξz i Ω Λ 7 2ω(ω 2 Ω 2 ) δ C i Ω Λ 8 C ξ zN 2ω(ω 2 Ω 2 ) δ ξN z C i (2ω Ω ) Λ 9 C 2ω(ω 2 Ω 2 ) δ C i (2ω C Ω ) Λ 10 C ξN zN C O( 2 ) 2ω(ω 2 Ω 2 )

(6.49)

6.5 Subharmonic Resonance of Order One-Third

Choosing the Λ m to eliminate all of the O() terms in (6.49), we have Λ1 D

δ , 2ω 2

Λ2 D

δ , ω2

Λ3 D

δ , 6ω 2

δ , 8ω(ω 2 Ω 2 )2 (ω 2Ω ) δ δ , Λ6 D , Λ5 D 2 2 4ω (ω Ω 2 )2 8ω(ω 2 Ω 2 )2 (ω C 2Ω ) δ δ , Λ8 D , Λ7 D 2 2 2ωΩ (ω Ω ) 2ωΩ (ω 2 Ω 2 ) δ δ , Λ 10 D Λ9 D 2ω(ω 2 Ω 2 )(2ω Ω ) 2ω(ω 2 Ω 2 )(2ω C Ω ) (6.50)

Λ4 D

It is clear from (6.50) that, in addition to the small-divisor term that occurs when Ω ω (primary resonance), there are small-divisor terms when Ω 2ω (subharmonic resonance of order one-half) and Ω 1/2ω (superharmonic resonance of order two). Thus, the terms involving z ξN and z 2 are near-resonance terms. These are the cases treated in the preceding two sections. Substituting (6.46) into (6.26) and using (6.47) and (6.50) yields " 2 2 1 2 N2 z C zN Pξ D i ωξ C i δ ξ C ξN C ξ C ξ 6ξ ξN ω 2(ω 2 Ω 2 ) 3ω 2 z zN z 2 C zN 2 Ω 2 )2 (ω 2 4Ω 2 ) 2ω 2 (ω 2 Ω 2 )2 # ξ z C ξN zN ξ zN C ξN z C Ω (2ω C Ω )(ω 2 Ω 2 ) Ω (2ω Ω )(ω 2 Ω 2 ) " Ω (z z) N i 2 2i µ ω ξ ξN C C 2ω 2ω(ω 2 Ω 2 ) 3 # z C z N C α ξ C ξN C C 2(ω 2 Ω 2 )

4(ω 2

(6.51) Next, we introduce the near-identity transformation ξ D w C 2 h(w, wN , z, zN )

(6.52)

to eliminate the nonresonance terms from (6.51). The remainder depends on the type of resonance being considered. Because we are stopping at this order, we do not need to determine h. In the case of subharmonic resonance of order one-third, the excitation produces the near-resonance term z wN 2 . Keeping this term as well as

183

184

6 Forced Oscillations of SDOF Systems

the resonance terms in the transformed equation (6.51), we obtain 10δ 2 i 2 3α w 2 wN 2ω 3ω 2 3 1 1 i 2 2 z zN w C C α δ ω(ω 2 Ω 2 )2 4 2ω 2 4ω 2 Ω 2 3 1 i 2 1 2 z wN 2 C α C δ ω(ω 2 Ω 2 ) 4 6ω 2 Ω (2ω Ω )

wP D i ωw 2 µ w C

(6.53)

In the case of superharmonic resonance of order three, the excitation produces the near-resonance term z 3 . Then, keeping this term as well as the resonance terms in the transformed equation (6.51), we obtain i 2 10δ 2 w 2 wN 3α 2ω 3ω 2 3 1 1 i 2 2 z zN w C C α δ ω(ω 2 Ω 2 )2 4 2ω 2 4ω 2 Ω 2 2δ 2 i 2 α z3 C 16ω(ω 2 Ω 2 )3 ω 2 4Ω 2

wP D i ωw 2 µ w C

(6.54)

Substituting the transformations (6.52), (6.46), and (6.22) into (6.3), using the expressions (6.25), (6.47), and (6.50), and recalling that z D O(1), we obtain δ 2 z C zN w C wN 2 6w wN C u D w C wN C 2 2 2 2(ω Ω ) 3ω z 2 C zN 2 z zN 4(ω 2 Ω 2 )2 (ω 2 4Ω 2 ) 2ω 2 (ω 2 Ω 2 )2 w z C wN zN w zN C wN z C C Ω (2ω C Ω )(ω 2 Ω 2 ) Ω (2ω Ω )(ω 2 Ω 2 )

(6.55)

Substituting (6.4) and the polar form (6.42) into (6.55) yields ( F F2 u D a cos (ωt C β) C 2 cos Ω t C 2 2 ω Ω 2ω (ω 2 Ω 2 )2 F 2 cos 2Ω t δ a2 [cos (2ωt C 2β) 3] C 2(ω 2 Ω 2 )2 (ω 2 4Ω 2 ) 6ω 2

) Fa cos (Ω C ω)t C β Fa cos (Ω ω)t β C C Ω (2ω C Ω )(ω 2 Ω 2 ) Ω (2ω Ω )(ω 2 Ω 2 )

(6.56)

6.5 Subharmonic Resonance of Order One-Third

Substituting (6.4) and (6.42) into (6.53) and separating real and imaginary parts, we have 3 1 2 1 2 aP D 2 µ a C α C δ 2ω(ω 2 Ω 2 ) 4 6ω 2 Ω (2ω Ω ) Fa 2 sin [3β (Ω 3ω) t] a βP D

(6.57)

2 F 2 3 1 1 2 a C αδ ω(ω 2 Ω 2 )2 4 2ω 2 4ω 2 Ω 2 3α 2 5δ 2 a3 C C 2 3 8ω 12ω 2ω(ω 2 Ω 2 ) 3 1 1 Fa 2 cos [3β (Ω 3ω) t] α C δ2 4 6ω 2 Ω (2ω Ω ) (6.58)

for the subharmonic case. Equations 6.56–6.58 are in full agreement with those obtained by using the method of multiple scales (Nayfeh, 1986). In the case of superharmonic resonance of order three, substituting (6.4) and (6.42) into (6.54) and separating real and imaginary parts, we obtain aP D 2 µ a C

2 F 3 8ω(ω 2 Ω 2 )3

α

2δ 2 ω 2 4Ω 2

sin [β (3Ω ω) t] (6.59)

a βP D

3 1 1 2 F 2 a C α δ2 ω(ω 2 Ω 2 )2 4 2ω 2 4ω 2 Ω 2 3α 5δ 2 a3 C 2 8ω 12ω 3 2δ 2 2 F 3 α 2 cos [β (3Ω ω) t] C 8ω(ω 2 Ω 2 )3 ω 4Ω 2

(6.60)

Equations 6.56, 6.59, and 6.60 are in full agreement with those obtained by using the method of multiple scales (Nayfeh, 1986).

185

187

7 Parametrically Excited Systems In the preceding two chapters, the excitation was taken to be external or sometimes called additive. In this chapter, we treat the case in which the excitation is parametric or sometimes called multiplicative. We start with the Mathieu equation and then progress to linear multiple-degree-of-freedom systems and ﬁnally include the effect of nonlinearities.

7.1 The Mathieu Equation

In this section, we determine approximations to the solutions of the Mathieu equation uR C ω 2 u C 2µ uP C 2 u cos Ω t D 0

(7.1)

where is a small nondimensional parameter and µ, ω, and Ω are constants. As in the preceding chapters, we ﬁrst cast (7.1) into a complex-valued form by introducing the transformation u D ζ C ζN and uP D i ω ζ ζN (7.2) and 2 cos Ω t D z C zN ,

z D eiΩ t

(7.3)

so that zP D i Ω z

(7.4)

With this transformation, (7.1) becomes i ζP D i ωζ µ ζ ζN C ζ C ζN (z C z) N 2ω

(7.5)

We note that the transformation (7.2) is not valid when ω 0. This case is treated in Section 2.8. The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

188

7 Parametrically Excited Systems

To determine a second-order uniform expansion of and hence determine a normal form of (7.5), we let ζ D η C h 1 (η, η, N z, z) N C 2 h 2 (η, η, N z, zN ) C

(7.6)

N z, zN ) C 2 g 2 (η, η, N z, zN ) C ηP D i ωη C g 1 (η, η,

(7.7)

where

Substituting (7.6) and (7.7) into (7.5) and equating coefﬁcients of like powers of , we obtain i (η C η) N (z C zN ) (7.8) 2ω @h 1 @h 1 i g1 gN 1 µ h 1 hN 1 C h 1 C hN 1 (z C zN ) (7.9) g 2 C L(h 2 ) D @η @ ηN 2ω g 1 C L(h 1 ) D µ (η η) N C

where the operator L is deﬁned by @ @ @ @ L D iω η ηN 1 C iΩ z zN @η @ ηN @z @ zN

(7.10)

Next, we consider two cases: fundamental parametric resonance (i.e., Ω ω) and principal parametric resonance (i.e., Ω 2ω). 7.1.1 Fundamental Parametric Resonance

We start by choosing g 1 to eliminate the resonance and near-resonance terms in (7.8). Because Ω ω, only the term µ η is a resonance term and there are no near-resonance terms. Therefore, we choose g 1 to eliminate this resonance term; that is, g 1 D µ η

(7.11)

Then, we choose h 1 to eliminate all of the nonresonance terms in (7.8). The righthand side of (7.8) suggests seeking h 1 in the form h 1 D Γ1 ηN C Γ2 η z C Γ3 ηN zN C Γ4 η zN C Γ5 ηN z

(7.12)

Substituting (7.12) into (7.8) and using (7.11) yields 1 1 (2i ωΓ1 C µ) ηN C i Ω Γ2 C η z C i Ω Γ4 C η zN 2ω 2ω 1 1 ηN zN C i (2ω Ω ) Γ5 C ηN z D 0 C i (2ω C Ω ) Γ3 C 2ω 2ω (7.13)

7.1 The Mathieu Equation

Equating each of the coefﬁcients of η, N η z, ηN z, N η z, N and ηz N to zero, we obtain iµ 1 1 , Γ2 D , Γ3 D , 2ω 2ωΩ 2ω(Ω C 2ω) 1 1 , Γ5 D Γ4 D 2ωΩ 2ω(Ω 2ω) Γ1 D

(7.14)

Substituting (7.11) and (7.12) into (7.9) and using (7.14), we have

iµ Ω ηz (Ω 2ω) ηN zN Ω η zN η C 2 2 2 2ω 4ω (Ω C 2ω) 4ω Ω 4ω (Ω 2ω) i η z C ηN zN (Ω C 2ω) ηN z η zN C ηz N (z C z) C N C C (7.15) 4ω 2 Ω 2ω Ω (Ω C 2ω) Ω (Ω 2ω)

g 2 C L(h 2 ) D µ

We note that the terms proportional to η and η z zN are resonance terms and the term proportional to ηz N 2 is a near-resonance term because Ω ω. Hence, choosing g 2 to eliminate these resonance and near-resonance terms, we have 2η z zN i µ2 z 2 ηN i (7.16) C ηC g2 D 2ω 2ω Ω 2 4ω 2 Ω (Ω 2ω) Substituting for g 1 and g 2 from (7.11) and (7.16) into (7.7), we obtain the normal form 2η z zN i 2 µ 2 z 2 ηN i 2 (7.17) ηP D i ωη µ η C ηC 2ω 2ω Ω 2 4ω 2 Ω (Ω 2ω) Substituting (7.6) and (7.12) into (7.2) and using (7.14), we obtain η z C ηN zN η zN C ηz N iµ η ηN C C C (7.18) u D η C ηN C 2ω Ω (Ω C 2ω) Ω (Ω 2ω) Expressing η in the polar form ηD

1 i (Ω tCβ ) a 2

(7.19)

and using (7.3), we rewrite (7.18) as a cos(2Ω t C β) µ a sin(Ω t C β) C u D a cos (Ω t C β) C 2ω Ω (Ω C 2ω) a cos(β) C C Ω (Ω 2ω)

(7.20)

Substituting (7.3) and (7.19) into (7.17) and separating real and imaginary parts, we obtain aP D µ a C

2 a sin(2β) 2ωΩ (Ω 2ω)

a βP D (Ω ω) a

2 a 2 a cos(2β) 2 µ2 aC C 2ω ω(Ω 2 4ω 2 ) 2ωΩ (Ω 2ω)

(7.21) (7.22)

189

190

7 Parametrically Excited Systems

7.1.2 Principal Parametric Resonance

Because Ω 2ω in this case, z ηN is a near-resonance term. Consequently, choosing g 1 to eliminate the resonance and near-resonance terms in (7.8), we obtain g 1 D µ η C

i ηz N 2ω

(7.23)

Then, we seek the solution of (7.8) in the form h 1 D Γ1 ηN C Γ2 η z C Γ3 ηN zN C Γ4 η zN

(7.24)

Following steps similar to those used in the preceding section, we ﬁnd that the Γi for i D 1, 2, 3, 4 are given by (7.14). Substituting (7.23) and (7.24) into (7.9), using (7.14), and choosing g 2 to eliminate the resonance and near-resonance terms, we obtain g2 D

i µ2 i µ(Ω C 2ω) η η z zN z ηN 2ω 4ω 2 (Ω C 2ω) 4ω 2 Ω

(7.25)

Substituting (7.23) and (7.25) into (7.7) yields the normal form 2 iµ η i z zN η µ(Ω C2ω)z ηN i z ηN 2 C C ηP D i ωη µ η 2ω 2ω 4ω 2 (Ω C 2ω) 4ω 2 Ω (7.26) Substituting (7.6) into (7.12) and using (7.24) and (7.14), we have iµ η z C ηN zN η zN C ηz N (η η) u D η C ηN C N C C 2ω Ω (Ω C 2ω) 2Ω ω

(7.27)

Because z D e i Ω t , the form of the near-resonance term in (7.26) suggests the following polar form for η: ηD

1 i ( 1 Ω tCβ ) ae 2 2

(7.28)

We note that this choice produces a set of autonomous equations for a and β as shown below. Substituting (7.3) and (7.28) into (7.27) yields " cos 32 Ω t C β µ sin 12 Ω t C β 1 u D a cos Ω t C β C a C 2 Ω (Ω C 2ω) 2ω 1 # cos 2 Ω t β C (7.29) 2ωΩ

7.2 Multiple-Degree-of-Freedom Systems

Substituting (7.3) and (7.28) into (7.26) and separating real and imaginary parts, we obtain 2 µ(Ω C 2ω) a cos(2β) a sin 2β 2ω 4Ω ω 2 1 βP D (2ω Ω ) C cos 2β 2 2ω 1 µ(Ω C 2ω) 2 2 µ C sin(2β) 2ω 2ω(Ω C 2ω) 2ωΩ

aP D µ a C

(7.30)

(7.31)

7.2 Multiple-Degree-of-Freedom Systems

In this section, we consider parametrically excited, nongyroscopic, multiple-degreeof-freedom systems governed by xR C Ax C D xP C (2 cos Ω t) F x D 0

(7.32)

where A, D, and F are n n constant matrices and x is a column vector of length n. We assume that none of the eigenvalues of A is zero; this case is considered in Chapter 2. Then, we introduce the nonsingular linear transformation x D P u into (7.32) and obtain P uR C AP u C D P uP C (2 cos Ω t) F P u D 0

(7.33)

Multiplication of (7.33) from the left by P 1 , the inverse of P, yields uR C J u C DO uP C (2 cos Ω t) FO u D 0

(7.34)

where J D P 1 AP ,

DO D P 1 D P ,

and

FO D P 1 F P

(7.35)

One can always choose P so that J is a Jordan canonical form. In this section, we treat the case in which the eigenvalues of A are distinct and positive so that J is a diagonal matrix. We assume that DO is a diagonal matrix (the so-called modaldamping assumption). Moreover, we limit our discussion to ﬁrst-order expansions of the solutions of three-degree-of-freedom systems modeled by uR 1 C ω 21 u 1 C 2µ 1 uP 1 C (2 cos Ω t) uR 2 C ω 22 u 2 C 2µ 2 uP 2 C (2 cos Ω t) uR 3 C ω 23 u 3 C 2µ 3 uP 3 C (2 cos Ω t)

3 X nD1 3 X nD1 3 X

f 1n u n D 0

(7.36)

f 2n u n D 0

(7.37)

f 3n u n D 0

(7.38)

nD1

where the ω m have been arranged so that ω 3 > ω 2 > ω 1 .

191

192

7 Parametrically Excited Systems

As a ﬁrst step in determining uniform expansions of the solutions of (7.36)– (7.38), we recast them into a system of three ﬁrst-order complex-valued equations. To accomplish this, we note that when D 0 the solutions of (7.36)–(7.38) can be expressed as u n D A n e i ω n t C AN n e i ω n t

(7.39a)

where the A n are complex. Hence, uP n D i ω n A n e i ω n t AN n e i ω n t

(7.39b)

When ¤ 0, we continue to represent u n and uP n as in (7.39a) and (7.39b) but with time-varying rather than constant A n . Then, we identify A n e i ω n t with ζ n and rewrite (7.39a) and (7.39b) as u m D ζ m C ζN m , uP m D i ω m ζ m ζN m (7.39c) Moreover, we let z D eiΩ t

and

zP D i Ω z

(7.40)

With this transformation, (7.36)–(7.38) become 3 X i (z C z) ζP 1 D i ω 1 ζ1 µ 1 ζ1 ζN 1 C N f 1n ζ n C ζN n 2ω 1 nD1

(7.41)

3 X i (z C z) ζP 2 D i ω 2 ζ2 µ 2 ζ2 ζN 2 C N f 2n ζ n C ζN n 2ω 2 nD1

(7.42)

3 X i (z C z) ζP 3 D i ω 3 ζ3 µ 3 ζ3 ζN 3 C N f 3n ζ n C ζN n 2ω 3 nD1

(7.43)

To simplify (7.41)–(7.43), we introduce the near-identity transformation ζ m D η m C h m (η n , ηN n , z, z) N

(7.44)

and obtain ηP m D i ω m η m C iω m h m µ m (η m ηN m )

3 X @h m nD1

@η n

ηP n C

@h m P ηN n @ ηN n

@h m P i @h m (z C z) N f m1 (η 1 C ηN 1 ) C f m2 (η 2 C ηN 2 ) zP zN C @z @ zN 2ω m

(7.45) C f m3 (η 3 C ηN 3 ) C

for m D 1, 2, and 3. The form of the O() terms suggests the following form for the h m : h m D ∆ m1 η m C ∆ m2 ηN m C Γm1 z η 1 C Γm2 z ηN 1 C Γm3 z η 2 C Γm4 z ηN 2 C Γm5 z η 3 C Γm6 z ηN 3 C Γm7 zN η 1 C Γm8 zN ηN 1 C Γm9 zN η 2 C Γm10 zN ηN 2 C Γm11 zN η 3 C Γm12 zN ηN 3

(7.46)

7.2 Multiple-Degree-of-Freedom Systems

It follows from (7.45) that ηP m D i ω m η m C O()

(7.47)

Hence, substituting (7.40), (7.46), and (7.47) into the right-hand side of (7.45), we obtain ηP m D i ω m η m µ m η m C (2i ω m ∆ m2 C µ m ) ηN m f m1 z η1 i (Ω C ω 1 ω m ) Γm1 2ω m f m1 z ηN 1 i (Ω ω 1 ω m ) Γm2 2ω m f m2 z η2 i (Ω C ω 2 ω m ) Γm3 2ω m f m2 z ηN 2 i (Ω ω 2 ω m ) Γm4 2ω m f m3 z η3 i (Ω C ω 3 ω m ) Γm5 2ω m f m3 z ηN 3 i (Ω ω 3 ω m ) Γm6 2ω m f m1 zN η 1 C i (Ω ω 1 C ω m ) Γm7 2ω m f m1 zN ηN 1 C i (Ω C ω 1 C ω m ) Γm8 C 2ω m f m2 zN η 2 C i (Ω ω 2 C ω m ) Γm9 C 2ω m f m2 zN ηN 2 C i (Ω C ω 2 C ω m ) Γm10 C 2ω m f m3 zN η 2 C i (Ω ω 3 C ω m ) Γm11 C 2ω m f m3 zN ηN 3 C i (Ω C ω 3 C ω m ) Γm12 C 2ω m C

(7.48)

for m D 1, 2, and 3. Substituting (7.44) into (7.39c), we obtain to the ﬁrst approximation u m D η m C ηN m C O()

(7.49)

We note that (7.48) does not depend on the ∆ m1, and hence they are arbitrary. Choosing the Γm n to eliminate the terms involving z η m , z ηN m , zN η m , and zN ηN m , we ﬁnd that some of the Γm n have small-divisor terms when Ω ωm C ωn

for

m, n D 1, 2, and 3

Ω ωm ωn

for

m, n D 1, 2, and 3 but

m¤n

193

194

7 Parametrically Excited Systems

The case Ω 2ω m , m D 1, 2, or 3, is called principal parametric resonance of the mth mode, which is treated in the preceding section. The case Ω ω m C ω n , for m ¤ n, is called combination parametric resonance of the additive type and the case Ω ω m ω n , for m ¤ n, is called combination parametric resonance of the difference type. The cases (a) Ω 2ω s and Ω ω n ˙ ω m and (b) Ω ω s and Ω ω n ˙ ω m are called simultaneous parametric resonances. Next, we treat some of these cases. 7.2.1 The Case of Ω Near ω 2 C ω 1

When Ω ω 2 C ω 1 and there are no other resonances, the term involving z ηN 2 is near-resonance when m D 1, while the term involving z ηN 1 is near-resonance when m D 2, and there are no near-resonance terms when m D 3. Consequently, choosing the ∆ m2 and Γm n to eliminate all nonresonance terms in (7.48), we obtain i f 12 z ηN 2 C O 2 2ω 1 i f 21 ηP 2 D i ω 2 η 2 µ 2 η 2 C z ηN 1 C O 2 2ω 2 ηP 3 D i ω 3 η 3 µ 3 η 3 C O 2

ηP 1 D i ω 1 η 1 µ 1 η 1 C

(7.50) (7.51) (7.52)

We note that Γ14 and Γ22 are arbitrary. Moreover, (7.50)–(7.52) show that, although the ﬁrst and second modes are coupled, the third mode is uncoupled from the ﬁrst two modes. The polar form of (7.50)–(7.52) is in full agreement with that obtained by using the method of multiple scales (Nayfeh and Mook, 1979). 7.2.2 The Case of Ω Near ω 2 ω 1

In this case, η 2 zN and η 1 z are near-resonance terms when m D 1 and 2, respectively, and there are no resonance terms when m D 3. Consequently, to O(), the normal form of (7.48) is i f 12 zN η 2 2ω 1 i f 21 z η1 ηP 2 D i ω 2 η 2 µ 2 η 1 C 2ω 2

ηP 1 D i ω 1 η 1 µ 1 η 1 C

ηP 3 D i ω 3 η 3 µ 3 η 3

(7.53) (7.54) (7.55)

7.2.3 The Case of Ω Near ω 2 C ω 1 and ω 3 ω 2

In this case, z ηN 2 is a near-resonance term when m D 1, z ηN 1 and η 3 zN are nearresonance terms when m D 2, and z η 2 is a near-resonance term when m D 3.

7.3 Linear Systems Having Repeated Frequencies

Consequently, to O(), the normal form of (7.48) is i f 12 z ηN 2 2ω 1 i f 21 i f 23 z ηN 1 C zN η 3 ηP 2 D i ω 2 η 2 µ 2 η 2 C 2ω 2 2ω 2 i f 32 z η2 ηP 3 D i ω 3 η 3 µ 3 η 3 C 2ω 3

ηP 1 D i ω 1 η 1 µ 1 η 1 C

(7.56) (7.57) (7.58)

7.2.4 The Case of Ω Near 2ω 3 and ω 2 C ω 1

In this case, the near-resonance terms are z ηN 2 when m D 1, z ηN 1 when m D 2, and z ηN 3 when m D 3. Consequently, the normal form of (7.48) is i f 12 z ηN 2 2ω 1 i f 21 z ηN 1 ηP 2 D i ω 2 η 2 µ 2 η 2 C 2ω 2 i f 33 z ηN 3 ηP 3 D i ω 3 η 3 µ 3 η 3 C 2ω 3

ηP 1 D i ω 1 η 1 µ 1 η 1 C

(7.59) (7.60) (7.61)

We note that η 3 is uncoupled from η 1 and η 2 .

7.3 Linear Systems Having Repeated Frequencies

In contrast with the preceding section, here we consider a three-degree-of-freedom system having two repeated frequencies and the Jordan form 2

ω 21 4 JD 1 0

0 ω 21 0

3 0 05 ω 23

(7.62)

Thus, we consider uR 1 C ω 21 u 1 C 2µ 1 uP 1 C (2 cos Ω t)

3 X

f 1n u n D 0

(7.63)

nD1

uR 2 C ω 21 u 2 C u 1 C 2µ 2 uP 2 C (2 cos Ω t)

3 X

f 2n u n D 0

(7.64)

nD1

uR 3 C ω 23 u 3 C 2µ 3 uP 3 C (2 cos Ω t)

3 X nD1

We assume that ω 3 > ω 1 .

f 3n u n D 0

(7.65)

195

196

7 Parametrically Excited Systems

In the absence of damping and the parametric excitation, the system is unstable (the system is said to be in ﬂutter). To show this, we note that when µ n D 0 and f m n D 0, u 1 D a 1 cos (ω 1 t C β 1 ) u 3 D a 3 cos (ω 3 t C β 3 ) where a 1 , a 3 , β 1 and β 3 are constants. Then (7.64) becomes uR 2 C ω 21 u 2 D u 1 D a 1 cos (ω 1 t C β 1 ) Hence, u 2 D a 2 cos (ω 2 t C β 2 )

a1 t sin (ω 1 t C β 1 ) 2ω 1

which contains a secular term or resonance term. Consequently, one refers to this one-to-one resonance as a nonsemisimple one-to-one resonance and the linear operator J is said to have a generic nonsemisimple structure. We wish to determine the normal forms of (7.63)–(7.65) in the presence of damping and the parametric excitation so that one can use these forms to ascertain if the damping and parametric excitation can stabilize the system. In the stable case, one assumes that all of the three variables u 1 , u 2 , and u 3 are bounded and, if possible, determine the values of the parameters which are consistent with this assumption. Although the damping and parametric excitation might stabilize the system, we still expect the amplitude of u 2 to be much larger than those of u 1 and u 3 . We use this observation to simplify the obtained normal forms. As a ﬁrst step in the application of the method of normal forms, we recast (7.63)– (7.65) in complex-valued form by using the transformation (7.39c) and (7.40). The result is 3 X i (z C z) ζP 1 D i ω 1 ζ1 µ 1 ζ1 ζN 1 C N f 1n ζ n C ζN n 2ω 1 nD1 i ζ1 C ζN 1 µ 2 ζ2 ζN 2 ζP 2 D i ω 1 ζ2 C 2ω 1 3 X i (z C zN ) f 2n ζ n C ζN n C 2ω 1 nD1 3 X i (z C z) ζP 3 D i ω 3 ζ3 µ 3 ζ3 ζN 3 C N f 3n ζ n C ζN n 2ω 3 nD1

(7.66)

(7.67) (7.68)

First, we introduce a transformation to simplify the ﬁrst-order problem. We note that (i ζ1 )/(2ω 1 ) is a resonance term in (7.67) and hence it cannot be eliminated by any transformation. This leaves the term (i ζN 1 )/(2ω 1 ), which can be eliminated by the transformation ζ1 D η 1 ,

ζ2 D η 2 C ∆ 1 ηN 1 ,

ζ3 D η 3

(7.69)

7.3 Linear Systems Having Repeated Frequencies

Substituting (7.69) into the O(1) terms in (7.66) and (7.67) yields ηP 1 D i ω 1 η 1 C O()

(7.70)

ηP 2 D i ω 1 η 2 C i ω 1 ∆ 1 ηN 1 ∆ 1 ηPN 1 C

i (η 1 C ηN 1 ) C O() 2ω 1

(7.71)

Using (7.70) in (7.71), we have i i ηN 1 C O() η 1 C 2i ω 1 ∆ 1 C ηP 2 D i ω 1 η 2 C 2ω 1 2ω 1

(7.72)

Then, choosing ∆ 1 to eliminate ηN 1 from (7.72) yields ∆1 D

1 4ω 21

(7.73)

Substituting for ζ2 from (7.69) into (7.39c) and using (7.73), we have u 2 D η 2 C ηN 2

1 (η 1 C ηN 1 ) C O() 4ω 21

(7.74)

Substituting (7.69) into (7.66)–(7.68) yields ηP 1 D i ω 1 η 1 µ 1 (η 1 ηN 1 ) C

i f 12 (z C z)(η N 1 C ηN 1 ) 8ω 31

ηP 2 D i ω 1 η 2 C

(7.75)

3 X i i (z C z) N η 1 µ 2 (η 2 ηN 2 ) C f 2n (η n C ηN n ) 2ω 1 2ω 1 nD1

i f 22 (z C z)(η N 1 C ηN 1 ) 8ω 31

ηP 3 D i ω 3 η 3 µ 3 (η 3 ηN 3 ) C

3 X i (z C zN ) f 1n (η n C ηN n ) 2ω 1 nD1

(7.76)

3 X i (z C zN ) f 3n (η n C ηN n ) 2ω 3 nD1

i f 32 (z C zN )(η 1 C ηN 1 ) 8ω 21 ω 3

(7.77)

To simplify (7.75)–(7.77), we introduce the near-identity transformation η m D ξm C h m ξn , ξN n , z, zN

(7.78)

and choose the h m to eliminate the nonresonance terms. The result depends on the resonance conditions. Three of these conditions, namely Ω 2ω 1 , Ω ω 3 C ω 1 , and Ω ω 3 ω 1 , are discussed next. In Section 7.3.4, we carry out the transformation to second order for the case Ω ω 1 .

197

198

7 Parametrically Excited Systems

7.3.1 The Case of Ω Near 2ω 1

When Ω 2ω 1 and there are no other resonances, the terms z ηN 1 and z ηN 2 are near-resonance terms in both (7.75) and (7.76). Hence, the normal form of (7.75)– (7.77) is i i ξP1 D i ω 1 ξ1 µ 1 ξ1 C f 11 z ξN1 C f 12 z ξN2 f 12 z ξN1 2ω 1 8ω 31 i i ξP2 D i ω 1 ξ2 C f 21 z ξN1 C f 22 z ξN2 ξ1 µ 2 ξ2 C 2ω 1 2ω 1 i f 22 z ξN1 8ω 31 ξP3 D i ω 3 ξ3 µ 3 ξ3

(7.79)

(7.80) (7.81)

As noted earlier, (7.79) and (7.80) can be further simpliﬁed because u 2 is much larger than u 1 and hence ξ2 is much larger than ξ1 . To accomplish this simpliﬁcation, we scale the damping coefﬁcients µ 1 and µ 2 and the variables ξ1 and ξ2 as ξ1 D χ 1 ,

ξ2 D λ 2 χ 2 ,

µ n D λ 1 µO n

where χ 1 and χ 2 are O(1) and λ 1 and λ 2 are positive constants to be determined from the analysis. Because (7.79) and (7.80) are linear and homogeneous, we ordered ξ1 at O() without loss of generality. Using these scalings, we rewrite (7.79) and (7.80) as i χP 1 D i ω 1 χ 1 1λ 1 µO 1 χ 1 C f 11 z χN 1 C 1λ 2 f 12 z χN 2 2ω 1 i f 12 z χN 1 (7.82) 8ω 31 i λ2 i 1Cλ 2 χP 2 D i ω 1 χ 2 C χ 1 1λ 1 µO 2 χ 2 C f 21 z χN 1 C f 22 z χN 2 2ω 1 2ω 1 i 1Cλ 2 f 22 z χN 1 (7.83) 8ω 31 As ! 0, we note that f 11 z χO 1 is small compared with 1λ 2 f 12 z χN 2 because λ 2 > 0. Then, requiring the damping term in (7.82) be the same order as the resonance term i(2ω 1 )1 1λ 2 f 12 z χN 2 , we have 1 λ1 D 1 λ2

or

λ1 D λ2

(7.84)

Similarly, as ! 0, the resonance terms (the terms inside the parenthesis) in (7.83) are small compared with the damping term 1λ 1 µO 2 χ 2 because λ 1 and λ 2 are positive. Then, requiring i(2ω 1 )1 λ 2 χ 1 to be the same order as 1λ 1 µO 2 χ 2 , we have λ2 D 1 λ1

(7.85)

7.3 Linear Systems Having Repeated Frequencies

Hence, λ 1 D λ 2 D 1/2. Consequently, keeping terms up to O( 1/2 ) in (7.82) and (7.83), we obtain the simpliﬁed normal form χP 1 D i ω 1 χ 1 1/2 µO 1 χ 1 C

i 1/2 f 12 z χN 2 C 2ω 1

(7.86)

χP 2 D i ω 2 χ 2 1/2 µO 2 χ 2 C

i 1/2 χ1 C 2ω 1

(7.87)

Equations 7.86 and 7.87 are also called the distinguished limit or least degenerate form of (7.82) and (7.83). We note that (7.86) and (7.87) agree with those obtained by Nayfeh and Mook (1979) by using the method of multiple scales. 7.3.2 The Case of Ω Near ω 3 C ω 1

In this case, the term z ηN 3 is near-resonance in (7.75) and (7.76), while the terms z ηN 1 and z ηN 2 are near-resonance in (7.77). Hence, the normal form of (7.75)–(7.77) is i ξP1 D i ω 1 ξ1 µ 1 ξ1 C f 13 z ξN3 2ω 1 i i ξP2 D i ω 1 ξ2 C ξ1 µ 2 ξ2 C f 23 z ξN3 2ω 1 2ω 1 i i ξP3 D i ω 3 ξ3 µ 3 ξ3 C f 31 z ξN1 C f 32 z ξN2 f 32 z ξN1 2ω 3 8ω 21 ω 3

(7.88) (7.89) (7.90)

Again, using the fact that ξ2 is much larger than ξ1 and ξ3 , we further simplify (7.88)–(7.90). To accomplish this, we scale the ξn and µ n as ξ1 D χ 1 ,

ξ2 D λ 2 χ 2 ,

ξ3 D λ 3 χ 3 ,

µ n D λ 1 µO n

(7.91)

where the χ n and µO n are O(1) and the λ n are positive constants. Substituting (7.91) into (7.88)–(7.90), we obtain i 1λ 3 f 13 z χN 3 2ω 1 i λ2 i 1Cλ 2λ 3 χ 1 1λ 1 µO 2 χ 2 C f 23 z χN 3 χP 2 D i ω 1 χ 2 C 2ω 1 2ω 1

χP 1 D i ω 1 χ 1 1λ 1 µO 1 χ 1 C

χ 3 D i ω 3 χ 3 1λ 1 µO 3 χ 3 C

As ! 0, the distinguished limit of (7.92)–(7.94) corresponds to 1 λ1 D λ2 1 λ1 D 1 C λ3 λ2

(7.93)

i 1Cλ 3 i 1Cλ 3λ 2 f 31 z χN 1 C f 32 z χN 2 2ω 3 2ω 3

i 1Cλ 3 f 32 z χN 1 8ω 21 ω 3

1 λ1 D 1 λ3

(7.92)

(7.94)

199

200

7 Parametrically Excited Systems

or λ2 D

2 3

and

λ1 D λ3 D

1 3

Thus, to the ﬁrst approximation, (7.92)–(7.94) simplify to the normal form χP 1 D i ω 1 χ 1 2/3 µO 1 χ 1 C

i 2/3 f 13 z χN 3 C 2ω 1

(7.95)

i 2/3 χ1 C 2ω 1 i 2/3 f 32 z χN 2 C χP 3 D i ω 3 χ 3 2/3 µO 3 χ 3 C 2ω 3 χP 2 D i ω 1 χ 2 2/3 µO 2 χ 2 C

(7.96) (7.97)

Equations 7.95–7.97 agree with those obtained by Nayfeh and Mook (1979) by using the method of multiple scales. 7.3.3 The Case of Ω Near ω 3 ω 1

In this case, the term zN ξ3 is near-resonance in (7.75) and (7.76), while the terms z ξ1 and z ξ2 are near-resonance in (7.77). Hence, the normal form of (7.75)–(7.77) is i ξP1 D i ω 1 ξ1 µ 1 ξ1 C f 13 zN ξ3 (7.98) 2ω 1 i i ξP2 D i ω 1 ξ2 C ξ1 µ 2 ξ2 C f 23 zN ξ3 (7.99) 2ω 1 2ω 1 i i ( f 31 z ξ1 C f 32 z ξ2 ) ξP3 D i ω 3 ξ3 µ 3 ξ3 C f 32 z ξ1 (7.100) 2ω 3 8ω 21 ω 3 As in Section 7.3.2, (7.98)–(7.100) can be further simpliﬁed by using the scaling ξ1 D χ 1 ,

ξ2 D 2/3 χ 2 ,

ξ3 D 1/3 χ 3 ,

µ n D 1/3 µO n

Substituting these scaled expressions into (7.98)–(7.100) and keeping the leading terms in (i.e., up to O( 2/3 )), we obtain the simpliﬁed normal form χP 1 D i ω 1 χ 1 2/3 µO 1 χ 1 C

i 2/3 f 13 zO χ 3 C 2ω 1

(7.101)

χP 2 D i ω 1 χ 2 2/3 µO 2 χ 2 C

i 2/3 χ1 C 2ω 1

(7.102)

χP 3 D i ω 3 χ 3 2/3 µO 3 χ 3 C

i 2/3 f 32 z χ 2 C 2ω 3

(7.103)

7.3.4 The Case of Ω Near ω 1

In this case, instead of ﬁrst determining the normal form of (7.66)–(7.68) and then scaling the dependent variables to simplify the obtained normal forms, we ﬁrst

7.3 Linear Systems Having Repeated Frequencies

scale the dependent variables and then obtain the simpliﬁed normal form, thereby reducing the algebra involved. Because ω 3 is away from ω 1 and hence Ω ω 1 is away from ω 3 , we assume that ζ3 is O(1), which is the same order as ζ1 . However, because ζ2 is large compared with ζ1 , we assume that ζ2 D O( λ 2 ), where λ 2 is positive. Therefore, we let ζ1 D η 1 ,

ζ2 D λ 2 η 2 ,

and

ζ3 D η 3

(7.104)

and rewrite (7.66)–(7.68) as i (z C zN ) ηP 1 D i ω 1 η 1 µ 1 (η 1 ηN 1 ) C 2ω 1 h i f 11 (η 1 C ηN 1 ) C 1λ 2 f 12 (η 2 C ηN 2 ) C f 13 (η 3 C ηN 3 )

(7.105)

i i λ2 (z C z) ηP 2 D i ω 1 η 2 C N (η 1 C ηN 1 ) µ 2 (η 2 ηN 2 ) C 2ω 1 2ω 1 h i 1Cλ 2 f 21 (η 1 C ηN 1 ) C f 22 (η 2 C ηN 2 ) C 1Cλ 2 f 23 (η 3 C ηN 3 ) (7.106) i (z C zN ) ηP 3 D i ω 3 η 3 µ 3 (η 3 ηN 3 ) C 2ω 3 h i f 31 (η 1 C ηN 1 ) C 1λ 2 f 32 (η 2 C ηN 2 ) C f 33 (η 3 C ηN 3 )

(7.107)

Because Ω ω 1 and is away from ω 3 , the term (z C z)(η N 2 C ηN 2 ) is a nonresonance term and hence we choose λ 2 D 1, making the terms i(2ω 1 )1 (z C z) N f 12 (η 2 C ηN 2 ) and i(2ω 3 )1 (z C zN ) f 32 (η 2 C ηN 2 ) in (7.105) and (7.107), respectively, of O(1). Hence, to simplify (7.105)–(7.107), we let N ξ2 , ξN2 C h 11 z, z, N ξn , ξN n C η 1 D ξ1 C h 10 z, z,

(7.108)

η 2 D ξ2 C h 21 z, zN , ξn , ξN n C

(7.109)

η 3 D ξ3 C h 30 z, z, N ξ2 , ξN2 C h 31 z, z, N ξn , ξN n C

(7.110)

N ξn , ξN n C ξP1 D i ω 1 ξ1 C g 1 z, z,

(7.111)

ξP2 D i ω 1 ξ2 C g 2 z, z, N ξn , ξN n C

(7.112)

ξP3 D i ω 3 ξ3 C g 3 z, z, N ξn , ξN n C

(7.113)

where

We have included the terms h 10 and h 30 at O(1) to eliminate the terms of O(1) in (7.105) and (7.107). Substituting (7.108)–(7.113) into (7.105)–(7.107) using (7.40),

201

202

7 Parametrically Excited Systems

and equating the coefﬁcients of 0 and on both sides, we obtain @h 10 @h 10 @h 10 N @h 10 ξ2 h 10 iΩ ξ2 z zN C i ω 1 @z @ zN @ξ2 @ ξN2 i (7.114) D N ξ2 C ξN2 f 12 (z C z) 2ω 1 @h 30 @h 30 @h 30 N @h 30 ξ2 i ω 3 h 30 iΩ ξ2 z zN C i ω 1 @z @ zN @ξ2 @ ξN2 i (7.115) N ξ2 C ξN2 D f 32 (z C z) 2ω 3 X 3 @h 11 @h 11 @h 11 N @h 11 ξn i ω 1 h 11 g1 C i Ω i ωn ξn z zN C @z @ zN @ξn @ ξN n nD1

@h 10 @h 10 D g2 gN 2 µ 1 ξ1 ξN1 C h 10 hN 10 @ξ2 @ ξN2 i f 11 (z C zN ) ξ1 C ξN1 C h 10 C hN 10 C 2ω 1 i C f 12 (z C zN ) h 21 C hN 21 2ω 1 i C f 13 (z C zN ) ξ3 C ξN3 C h 30 C hN 30 2ω 1

(7.116)

X 3 @h 21 @h 21 @h 21 N @h 21 ξn i ω 1 h 21 g2 C i Ω i ωn ξn z zN C @z @ zN @ξn @ ξN n nD1 i i D f 22 (z C zN ) ξ2 C ξN2 ξ1 C ξN1 C h 10 C hN 10 µ 2 ξ2 ξN2 C 2ω 1 2ω 1 (7.117)

X 3 @h 31 @h 31 @h 31 N @h 31 ξn i ω 3 h 31 i ωn ξn z zN C @z @ zN @ξn @ ξN n nD1 @h 30 @h 30 D g2 gN 2 µ 3 ξ3 ξN3 C h 30 hN 30 @ξ2 @ zN i C f 31 (z C zN ) ξ1 C ξN1 C h 10 C hN 10 2ω 3 i C f 32 (z C zN ) h 21 C hN 21 2ω 3 i C f 33 (z C zN ) ξ3 C ξN3 C h 30 C hN 30 (7.118) 2ω 3

g3 C i Ω

where ω 2 D ω 1 . The form of the right-hand sides of (7.114) and (7.115) suggests the following forms for h 10 and h 30 : h n0 D Γn1 z ξ2 C Γn2 z ξN2 C Γn3 zN ξ2 C Γn4 zN ξN2

(7.119)

7.3 Linear Systems Having Repeated Frequencies

for n D 1 and 3. Substituting (7.119) into (7.114) and (7.115) and equating each of the coefﬁcients of z ξ2 , z ξN2 , zN ξ2 , and zN ξN2 on both sides, we obtain f 12 f 12 , Γ12 D , 2ω 1 Ω 2ω 1 (Ω 2ω 1 ) f 12 f 12 , Γ14 D , Γ13 D 2ω 1 Ω 2ω 1 (Ω C 2ω 1 ) f 32 f 32 , Γ32 D , Γ31 D 2ω 3 (Ω C ω 1 ω 3 ) 2ω 3 (Ω ω 1 ω 3 ) f 32 f 32 , Γ34 D Γ33 D 2ω 3 (Ω ω 1 C ω 3 ) 2ω 3 (Ω C ω 1 C ω 3 ) Γ11 D

(7.120)

(7.121)

Substituting (7.119) into (7.117) yields g2 C i Ω D

X 3 @h 21 @h 21 @h 21 N @h 21 ξn i ω 1 h 21 i ωn ξn z zN C @z @ zN @ξn @ ξN n nD1

i i ξ1 C ξN1 µ 2 ξ2 ξN2 C N ξ2 C ξN2 f 22 (z C z) 2ω 1 2ω 1

i (Γ11 C Γ14 ) z ξ2 C zN ξN2 C (Γ12 C Γ13 ) z ξN2 C zN ξ2 C 2ω 1

(7.122)

Equating g 2 to the resonance terms on the right-hand side of (7.122), we have g2 D

i ξ1 µ 2 ξ2 2ω 1

(7.123)

Then, the form of the remaining terms in (7.122) suggests the following form for h 21 : h 21 D Γ21 z ξ2 C Γ22 z ξN2 C Γ23 zN ξ2 C Γ24 zN ξN2 C Γ25 ξN2 C Γ26 ξN1

(7.124)

Substituting (7.124) into (7.122), using (7.123), and equating each of the coefﬁcients of z ξ2 , z ξN2 , zN ξ2 , zN ξN2 , ξN2 , and ξN1 on both sides, we obtain Γ11 C Γ14 C f 22 Γ12 C Γ13 C f 22 , Γ22 D , 2ω 1 Ω 2ω 1 (Ω 2ω 1 ) Γ12 C Γ13 C f 22 , Γ23 D 2ω 1 Ω Γ11 C Γ14 C f 22 i µ2 1 , Γ26 D 2 , Γ25 D Γ24 D 2ω 1 (Ω C 2ω 1 ) 2ω 1 4ω 1 Γ21 D

(7.125)

We note that we needed to explicitly determine h 21 because it is needed to determine g 1 from (7.116).

203

204

7 Parametrically Excited Systems

Substituting (7.119) and (7.124) into (7.116) and (7.118) yields X 3 @h 11 @h 11 @h 11 N @h 11 ξn i ω 1 h 11 i ωn ξn z zN C @z @ zN @ξn @ ξN n nD1

D µ 1 ξ1 ξN1 C (Γ11 Γ14 ) z ξ2 zN ξN2 C (Γ12 Γ13 ) z ξN2 zN ξ2

g1 C i Ω

(Γ11 z C Γ13 z) N g 2 (Γ12 z C Γ14 z) N gN 2 C

i f 11 (z C zN ) Λ 1 2ω 1

i i f 12 (z C zN ) Λ 2 C f 13 (z C zN ) Λ 3 (7.126) 2ω 1 2ω 1 X 3 @h 31 @h 31 N @h 31 @h 31 ξn i ω 3 h 31 i ωn ξn g3 C i Ω z zN C @z @ zN @ξn @ ξN n nD1

D µ 3 ξ3 ξN3 C (Γ31 Γ34 ) z ξ2 zN ξN2 C (Γ32 Γ33 ) z ξN2 zN ξ2 C

(Γ31 z C Γ33 z) N g 2 (Γ32 z C Γ34 z) N gN 2 C C

i f 31 (z C zN ) Λ 1 2ω 3

i i f 32 (z C zN ) Λ 2 C f 33 (z C zN ) Λ 3 2ω 3 2ω 3

(7.127)

where Λ 1 D ξ1 C ξN1 C (Γ11 C Γ14 ) z ξ2 C zN ξN2 C (Γ12 C Γ13 ) z ξN2 C zN ξ2 Λ 2 D (Γ21 C Γ24 ) z ξ2 C zN ξN2 C (Γ22 C Γ23 ) z ξN2 C zN ξ2 1 i µ2 N ξ2 ξ2 ξ1 C ξN1 C 2ω 1 4ω 21 Λ 3 D ξ3 C ξN3 C (Γ31 C Γ34 ) z ξ2 C zN ξN2 C (Γ32 C Γ33 ) z ξN2 C zN ξ2 If we do not want to continue the expansion beyond the terms in (7.108)–(7.110), we do not need to explicitly calculate h 11 and h 31 , and all what we need is to determine g 1 and g 3 by equating them to the resonance and near-resonance terms on the right-hand sides of (7.126) and (7.127). The result is g 1 D µ 1 ξ1 C

i i α 1 z zN ξ2 C α 2 z 2 ξN2 2ω 1 2ω 1

g 3 D µ 3 ξ3

(7.128) (7.129)

where α 1 D f 11 (Γ11 C Γ14 C Γ12 C Γ13 ) C f 12 (Γ21 C Γ24 C Γ22 C Γ23 ) C f 13 (Γ31 C Γ34 C Γ32 C Γ33 ) α 2 D f 11 (Γ12 C Γ13 ) C f 12 (Γ22 C Γ23 ) C f 13 (Γ32 C Γ33 )

(7.130) (7.131)

7.4 Gyroscopic Systems

Substituting for the Γm n from (7.120), (7.121), and (7.125) into (7.130) and (7.131), we obtain 2 1 2 f 12 ( f 11 C f 22 ) f 12 1 α1 D C C Ω 2 (Ω C 2ω 1 )2 (Ω 2ω 1 )2 Ω 2 4ω 21 1 1 (7.132) C f 13 f 32 C (Ω C ω 1 )2 ω 23 (Ω ω 1 )2 ω 23 2 f 12 ( f 11 C f 22 ) f 13 f 32 f 12 α2 D C (7.133) C 2 Ω (Ω 2ω 1 ) Ω (Ω 2ω 1 )2 (Ω ω 1 )2 ω 23 Substituting for g 1 , g 2 , and g 3 from (7.128), (7.123), and (7.129) into (7.111)–(7.113), we obtain i ξP1 D i ω 1 ξ1 µ 1 ξ1 C α 1 z zN ξ2 C α 2 z 2 ξN2 C (7.134) 2ω 1 i ξP2 D i ω 1 ξ2 µ 2 ξ2 C ξ1 C (7.135) 2ω 1 ξP3 D i ω 3 ξ3 µ 3 ξ3

(7.136)

Equations 7.132–7.136 agree with those obtained by Nayfeh and Mook (1979) by using the method of multiple scales.

7.4 Gyroscopic Systems

For simplicity, we consider a two-degree-of-freedom system to illustrate the method of solution. Speciﬁcally, we consider uR 1 C λ 1 uP 2 C α 1 u 1 D F1 D 2 ( f 11 u 1 C f 12 u 2 ) cos Ω t

(7.137)

uR 2 λ 2 uP 1 C α 2 u 2 D F2 D 2 ( f 21 u 1 C f 22 u 2 ) cos Ω t

(7.138)

where , Ω , λ m , α m , and f m n are constants. As a ﬁrst step, we cast (7.137) and (7.138) in complex-valued form. To accomplish this, we ﬁrst write down the solution of the unperturbed problem; that is, u 1 D A 1 e i ω 1 t C AN 1 e i ω 1 t C A 2 e i ω 2 t C AN 2 e i ω 2 t

(7.139)

uP 1 D i ω 1 A 1 e i ω 1 t AN 1 e i ω 1 t C i ω 2 A 2 e i ω 2 t AN 2 e i ω 2 t

(7.140)

u2 D

i(α 1 ω 22 ) i(α 1 ω 21 ) A 1 e i ω 1 t AN 1 e i ω 1 t C A 2 e i ω 2 t AN 2 e i ω 2 t λ1 ω1 λ1 ω2 (7.141)

uP 2 D

α 1 ω 22 α 1 ω 21 A 1 e i ω 1 t C AN 1 e i ω 1 t A 2 e i ω 2 t C AN 2 e i ω 2 t λ1 λ1 (7.142)

205

206

7 Parametrically Excited Systems

where ω 21 and ω 22 are solutions of ω 4 (α 1 C α 2 C λ 1 λ 2 ) ω 2 C α 1 α 2 D 0

(7.143)

Here, we assume that ω 21 and ω 22 are positive and that ω 2 > ω 1 . We associate A 1 e i ω 1 t with ζ1 and A 2 e i ω 2 t with ζ2 in (7.139)–(7.142) to deﬁne the transformation u 1 D ζ1 C ζN 1 C ζ2 C ζN 2

(7.144)

uP 1 D i ω 1 ζ1 ζN 1 C i ω 2 ζ2 ζN 2

(7.145)

i(α 1 ω 22 ) i(α 1 ω 21 ) ζ1 ζN 1 C ζ2 ζN 2 λ1 ω1 λ1 ω2 α 1 ω 22 α 1 ω 21 ζ1 C ζN 1 ζ2 C ζN 2 uP 2 D λ1 λ1

u2 D

(7.146) (7.147)

It follows from (7.144)–(7.147) that

ζ1 C ζN 1 D ω 22 α 1 u 1 λ 1 uP 2

(7.148)

ζ2 C ζN 2 D α 1 ω 21 u 1 C λ 1 uP 2

(7.149)

α 1 ω 22 ω 21 ζ1 ζN 1 D i ω 1 α 1 ω 22 uP 1 λ 1 ω 22 u 2

(7.150)

α 1 ω 22 ω 21 ζ2 ζN 2 D i ω 2 α 1 ω 21 uP 1 λ 1 ω 21 u 2

(7.151)

ω 22 ω 21 ω 22 ω 21

Solving (7.148) and (7.150) for ζ1 yields 2α 1 ω 22 ω 21 ζ1 D i ω 1 α 1 ω 22 uP 1 α 1 λ 1 uP 2 α 1 α 1 ω 22 u 1 i λ 1 ω 1 ω 22 u 2

(7.152)

Solving (7.149) and (7.151) for ζ2 yields 2α 1 ω 22 ω 21 ζ2 D i ω 2 α 1 ω 21 uP 1 C α 1 λ 1 uP 2 C α 1 α 1 ω 21 u 1 C i λ 1 ω 21 ω 2 u 2

(7.153)

Differentiating (7.152) with respect to t and using (7.137) and (7.138) to eliminate uR 1 and uR 2 , we obtain

2α 1 ω 22 ω 21 ζP 1 D i ω 1 α 1 λ 1 uP 2 C α 1 ω 22 u 1

α 1 α 1 C λ 1 λ 2 ω 22 uP 1 λ 1 α 2 u 2 C iω 1 α 1 ω 22 F1 α 1 λ 1 F2 (7.154) It follows from (7.143) that α 1 C λ 1 λ 2 ω 22 D ω 21 α 2

and

α 1 α 2 D ω 21 ω 22

(7.155)

7.4 Gyroscopic Systems

Using (7.155), we rewrite (7.154) as

2α 1 ω 22 ω 21 ζP 1 D i ω 1 α 1 λ 1 uP 2 C α 1 ω 22 u 1

C ω 21 ω 22 α 1 ω 21 uP 1 C λ 1 ω 21 ω 22 u 2 C iω 1 α 1 ω 22 F1 α 1 λ 1 F2

(7.156)

Using (7.144)–(7.147) to eliminate the u m and uP m from (7.156) and rearranging, we obtain iω 1 (α 1 ω 22 ) λ 1 ζP 1 D i ω 1 ζ1 C F1 F2 2α 1 (ω 22 ω 21 ) 2(ω 22 ω 21 )

(7.157)

Differentiating (7.153) and using a procedure similar to that used to determine ζP 1 , we obtain iω 2 (α 1 ω 21 ) λ 1 ζP 2 D i ω 2 ζ2 F1 C F2 2 2 2 2α 1 (ω 2 ω 1 ) 2(ω 2 ω 21 )

(7.158)

Finally, we substitute for F1 and F2 from (7.137) and (7.138) into (7.157) and (7.158), use (7.3), (7.144), and (7.146), and obtain iω 1 (α 1 ω 22 ) λ 1 (z C z) (z C z) ζP 1 D i ω 1 ζ1 C N Λ1 N Λ 2 (7.159) 2 2 2α 1 (ω 2 ω 1 ) 2(ω 22 ω 21 ) iω 2 (α 1 ω 21 ) λ 1 (z C zN ) Λ 1 C (z C z) ζP 2 D i ω 2 ζ2 N Λ 2 (7.160) 2α 1 (ω 22 ω 21 ) 2(ω 22 ω 21 ) where

Λ 1 D f 11 ζ1 C ζN 1 C ζ2 C ζN 2 α 1 ω 22 α 1 ω 21 ζ1 ζN 1 C C i f 12 λ1 ω1 λ1 ω2 Λ 2 D f 21 ζ1 C ζN 1 C ζ2 C ζN 2 α 1 ω 22 α 1 ω 21 ζ1 ζN 1 C C i f 22 λ1 ω1 λ1 ω2

ζ2 ζN 2

ζ2 ζN 2

To simplify (7.159) and (7.160), we introduce the near-identity transformation (7.44) and choose the h m to eliminate the nonresonance terms, thereby leaving the resonance and near-resonance terms. To O(), there are several resonance combinations of Ω , ω 1 , and ω 2 . These include:

Ω Ω Ω Ω

2ω 1 : Principal parametric resonance of the ﬁrst mode, 2ω 2 : Principal parametric resonance of the second mode, ω 2 C ω 1 : Combination parametric resonance of the additive type, ω 2 ω 1 : Combination parametric resonance of the difference type.

Next, we discuss the cases Ω 2ω 1 and Ω ω 2 ω 1 treated by Nayfeh and Mook (1979) using the method of multiple scales.

207

208

7 Parametrically Excited Systems

7.4.1 The Case of Ω Near 2ω 1

When Ω 2ω 1 and there are no other resonances, there are no resonance terms in (7.160) and z ζN 1 is a near-resonance term in (7.159). Therefore, substituting (7.44) into (7.159) and (7.160) and choosing the h m to eliminate the nonresonance terms, we obtain the normal form λ 1 f 21 C λ 2 f 12 ηP 1 D i ω 1 η 1 2(ω 22 ω 21 ) ω 1 (α 1 ω 22 ) f 11 α 1 ω 21 i C f 22 ηN 1 z (7.161) α1 ω1 ηP 2 D i ω 2 η 2

(7.162)

Putting η n D A n e ω n t in (7.161) and (7.162) yields the same equations obtained by using the method of multiple scales (Nayfeh and Mook, 1979). 7.4.2 The Case of Ω Near ω 2 ω 1

In this case, the resonance terms are zN ζ2 and z ζ1 in (7.159) and (7.160), respectively. Therefore, substituting (7.44) into (7.159) and (7.160) and choosing the h m to eliminate the nonresonance terms, we obtain the normal forms ω 1 (α 1 ω 22 )2 f 12 λ 1 f 21 C ηP 1 D i ω 1 η 1 2 2 λ1 α1 ω2 2(ω 2 ω 1 ) f 22 ω 1 f 11 η 2 zN C i α 1 ω 22 (7.163) ω2 α1 ω 2 (α 1 ω 21 )2 f 12 ηP 2 D i ω 2 η 2 C λ 1 f 21 C 2 2 λ1 α1 ω1 2(ω 2 ω 1 ) ω 2 f 11 f 22 η1 z i α 1 ω 21 (7.164) α1 ω1 Putting η m D A m e i ω m t in (7.163) and (7.164) yields the same equations obtained by Nayfeh and Mook (1979) by using the method of multiple scales.

7.5 A Nonlinear Single-Degree-of-Freedom System

Finally, we illustrate the method of normal forms using a parametrically excited single-degree-of-freedom system with quadratic and cubic nonlinearities. Speciﬁcally, we conside uR C ω 2 u C 2µ uP C 2F u cos Ω t C δ u2 C 2 α u3 D 0

(7.165)

7.5 A Nonlinear Single-Degree-of-Freedom System

Using the transformation (7.2) and (7.3), we rewrite (7.165) as iF ζ C ζN (z C zN ) ζP D i ωζ µ ζ ζN C 2ω 2 3 iδ i 2 α N ζCζ C ζ C ζN C (7.166) 2ω 2ω We seek a second-order uniform expansion of the solution of (7.166) in the form (7.6) and (7.7), equate coefﬁcients of like powers of , and obtain iF iδ (η C η) (η C η) (7.167) N (z C zN ) C N 2 2ω 2ω @h 1 @h 1 iF g 2 C L(h 2 ) D g 1 gN 1 µ h 1 hN 1 C h 1 C hN 1 (z C zN ) @η @ ηN 2ω iα iδ (7.168) (η C η) (η C η) N h 1 C hN 1 C N 3 C ω 2ω where the operator L is deﬁned in (7.10). There are two cases, depending on whether Ω 2ω (i.e., principal parametric resonance) or Ω is away from 2ω. We ﬁrst consider the latter case. N C g 1 C L(h 1 ) D µ (η η)

7.5.1 The Case of Ω Away from 2ω

In this case, to ﬁrst order, µ η is a resonance term and there are no near-resonance terms. Thus, we choose g 1 to eliminate the resonance term in (7.167); that is, g 1 D µ η

(7.169)

Then, the form of the remaining terms on the right-hand side of (7.167) suggests choosing h 1 in the form h 1 D Γ1 ηN C Γ2 η z C Γ3 ηN zN C Γ4 η zN C Γ5 ηz N C Γ6 η 2 C Γ7 η ηN C Γ8 ηN 2 (7.170) Substituting (7.170) into (7.167) and using (7.169) and (7.10) yields F F (2i ωΓ1 C µ) ηN i Ω Γ2 η z C i (2ω C Ω ) Γ3 C ηN zN 2ω 2ω F F η zN C i (2ω Ω ) Γ5 C ηz N C i Ω Γ4 C 2ω 2ω δ δ δ 2 i ω Γ6 η η η N C i ω 3Γ ηN 2 C i ω Γ C C 7 8 2ω 2 ω2 2ω 2 D0

(7.171)

Choosing the Γm to eliminate the nonresonance terms, we have Γ1 D

iµ , 2ω

Γ5 D

F , 2ω(Ω 2ω)

Γ2 D

F , 2ωΩ

F F , Γ4 D , 2ω(Ω C 2ω) 2ωΩ δ δ δ , Γ7 D 2 , Γ8 D 2 (7.172) Γ6 D 2 2ω ω 6ω Γ3 D

209

210

7 Parametrically Excited Systems

Substituting (7.172) into (7.170) yields h1 D

i µ ηN Fz η F η zN Fz ηN F zN ηN C 2ω 2ωΩ 2ωΩ 2ω(2ω Ω ) 2ω(2ω C Ω ) C

δ η ηN δ ηN 2 δ η2 2ω 2 ω2 6ω 2

(7.173)

Substituting (7.173) and (7.6) into (7.2), we have iµ F(z η C ηN z) N δ 2 (η 3η ηN C ηN 2 ) C (η η) N C 2ω 3ω 2 Ω (Ω C 2ω) F(η zN C ηz) N C C (7.174) Ω (Ω 2ω)

u D η C ηN

Substituting (7.169) and (7.173) into (7.168) yields 2F 2 η z zN i µ2 η F 2 ηN z 2 i C C 2ω 2ω Ω 2 4ω 2 Ω (Ω 2ω) 10δ 2 2 4 N C 3α η 2 ηN 2 F δ η ηz C Ω 2 4ω 2 ω 3ω 2 2 2 1 2 F δ η C z N C η N z C NRT . C 3ω 2 Ω (Ω 2ω)

g 2 C L(h 2 ) D

(7.175) There are two possible resonances at this order: those corresponding to Ω ω (fundamental parametric resonance) and Ω 3ω (subharmonic resonance of order one-third). When Ω is away from ω and 3ω, choosing g 2 to eliminate the resonance terms in (7.175), we obtain g2 D

i µ2 η i C 2ω 2ω

2F 2 η z zN 10δ 2 2 η C 3α η N Ω 2 4ω 2 3ω 2

(7.176)

Substituting (7.169) and (7.176) into (7.7), we obtain the normal form i 2 F 2 η z zN i 2 µ 2 ηC C i 2 ηP D i ωη µ η 2ω ω(Ω 2 4ω 2 )

3α 5δ 2 2ω 3ω 3

η 2 ηN (7.177)

When Ω ω, choosing g 2 to eliminate the resonance and near-resonance terms from (7.175) and approximating Ω with ω, we obtain 2 2F i µ2 η F 2 2 10F δ 5F δ 2 i η z z N C ηN z C η ηz N C η zN 2 2 2 2ω 2ω 3ω ω 3ω 3ω 2 10δ 2 η 2 ηN 3α (7.178) 3ω 2

g2 D

7.5 A Nonlinear Single-Degree-of-Freedom System

Substituting (7.169) and (7.178) into (7.7), we obtain the normal form F2 10F δ i 2 2F 2 i 2 µ 2 η z zN C 2 ηN z 2 C η ηz N η 2 2ω 2ω 3ω ω 3ω 2 10δ 2 5F δ 2 2 η η z N 3α η N C (7.179) 3ω 2 3ω 2

ηP D i ωη µ η

Putting η D Ae i ω t

and

z D eiΩ t

(7.180)

in (7.179), we obtain the equation found by using the method of multiple scales (Exercise 7.6.2). When Ω 3ω, choosing g 2 to eliminate resonance and near-resonance terms in (7.175) and approximating Ω with 3ω, we obtain g2 D

i µ2 η i C 2ω 2ω

2F 2 10δ 2 Fδ 2 2 η η z z N C 3α η N C η N z 5ω 2 3ω 2 ω2

(7.181)

Substituting (7.169) and (7.181) into (7.7), we obtain the normal form ηP D i ωη

10δ 2 Fδ 2 i 2 2F 2 2 µ 2 η 2 η η z z N C 3α η N C η z C 2ω 2ω 5ω 2 3ω 2 ω2 (7.182)

Substituting (7.180) into (7.182), we obtain the equation found by using the method of multiple scales (Exercise 7.6.2). 7.5.2 The Case of Ω Near 2ω

In this case, it follows from (7.172) that Γ5 has a small-divisor term and hence z ηN is near-resonance in (7.167). Choosing g 1 to eliminate the resonance and nearresonance terms in (7.167), we obtain g 1 D µ η C

iF ηz N 2ω

(7.183)

Then, we can choose all of the Γm , m D 1, 2, . . . , 8, except Γ5 , as in (7.172) to eliminate the nonresonance terms. Consequently, h1 D

i µ ηN δ ηN 2 Fz η F η zN F zN ηN δ η 2 δ η ηN 2 (7.184) C C 2 2ω 2ωΩ 2ωΩ 2ω(2ω C Ω ) 2ω ω 6ω 2

211

212

7 Parametrically Excited Systems

Substituting (7.184) into (7.2) yields F(z η C ηN z) N δ 2 iµ (η 3η ηN C ηN 2 ) C (η η) N C 2 2ω 3ω Ω (Ω C 2ω) F(η zN C ηN z) C (7.185) 2ωΩ Substituting (7.183) and (7.184) into (7.168) and choosing g 2 to eliminate the resonance and near-resonance terms, we obtain u D η C ηN

i µ2 (Ω C 2ω)µ F iF2 η z η N η z zN 2ω 4ω 2 Ω 4ω 2 (Ω C 2ω) 10δ 2 i 3α η 2 ηN C 2ω 3ω 2

g2 D

(7.186)

Substituting (7.183) and (7.186) into (7.7), we obtain the normal form iF i 2 µ 2 (Ω C 2ω) 2 µ F ηN z η z ηN 2ω 2ω 4ω 2 Ω 10δ 2 i 2 F 2 i 2 3α η 2 ηN η z zN C 2 4ω (Ω C 2ω) 2ω 3ω 2

ηP D i ωη µ η C

(7.187)

7.6 Exercises

7.6.1 Use the method of multiple scales to determine the normal form of (7.1) when Ω ω and Ω 2ω. Use the complex-valued form (7.5) and seek a second-order approximate solution in the form ζ D ζ1 (T0 , T1 , T2 ) C ζ2 (T0 , T1 , T2 ) C C 3 ζ2 (T0 , T1 , T2 ) C and obtain the system of equations D0 ζ1 i ωζ1 D 0 i i T0 Ω C e i T0 Ω (ζ1 C ζN 1 ) e D0 ζ2 i ωζ2 D D1 ζ1 µ(ζ1 ζN 1 ) C 2ω D0 ζ3 i ωζ3 D D1 ζ2 D2 ζ1 µ(ζ2 ζN 2 ) i i T0 Ω C e i T0 Ω (ζ2 C ζN 2 ) e C 2ω Express the solution of the ﬁrst-order problem as ζ1 D A(T1 , T2 )e i ω T0 Then, show that the second-order problem becomes N i ω T0 D0 ζ2 i ωζ2 D D1 Ae i ω T0 µ Ae i ω T0 C µ Ae i i T0 Ω N i ω T0 e C C e i T0 Ω Ae i ω T0 C Ae 2ω

7.6 Exercises

The Case Ω ω Let Ω D ω C 2 σ and show that the solvability condition of the second-order problem is D1 A D µ A Then, show that ζ2 D

i µ N i ω T0 1 1 Ae C Ae i(Ω Cω)T0 Ae i(ωΩ )T0 2ω 2ωΩ 2ωΩ 1 1 N i(Ω Cω)T0 C N i(Ω ω)T0 Ae Ae 2ω(Ω C 2ω) 2ω(Ω 2ω)

Substitute for ζ1 and ζ2 into the third-order equation, eliminate the terms that produce secular terms, and obtain D2 A D

i µ2 i i N 2i σ T2 Ae AC AC 2ω ω(Ω 2 4ω 2 ) 2ωΩ (Ω 2ω)

Finally, use the method of reconstitution to obtain the normal form 2 iµ i i N 2i σ T2 A A Ae AP D µ A 2 2ω ω(Ω 2 4ω 2 ) 2ωΩ (Ω 2ω) Compare these results with those obtained in Section 7.1.1 by using the method of normal forms. The Case Ω 2ω Let Ω D 2ω C σ and show that the solvability condition of the second-order problem is D1 A D µ A C

i N i σ T1 Ae 2ω

Then, show that ζ2 D

i µ N i ω T0 1 1 Ae C e i(Ω Cω)T0 e i(ωΩ )T0 2ω 2ωΩ 2ωΩ 1 e i(Ω Cω)T0 2ω(Ω C 2ω)

Substitute for ζ1 and ζ2 into the third-order equation, eliminate the terms that produce secular terms, and obtain D2 A D

i µ2 i (Ω C 2ω)µ N i σ T1 Ae A A 2 2ω 4ω Ω (Ω C 2ω) 4ω 2 Ω

Finally, use the method of reconstitution to obtain the normal form 2 iµ i i N i σ T1 2 Ae AC A AP D µ A 2ω 2ω 4ω 2 Ω (Ω C 2ω) (Ω C 2ω)µ N i σ T1 Ae C 4ω 2 Ω

213

214

7 Parametrically Excited Systems

Compare these results with those obtained in Section 7.1.2 by using the method of normal forms. 7.6.2 Use the method of multiple scales to determine the normal form of (7.165) when (a) Ω is away from ω, 2ω, and 3ω; (b) Ω ω; (c) Ω 3ω; and (d) Ω 2ω. Seek an approximate solution of (7.166) in the form ζ D ζ1 (T0 , T1 , T2 ) C ζ2 (T0 , T1 , T2 ) C C 3 ζ2 (T0 , T1 , T2 ) C and obtain the system of equations D0 ζ1 i ωζ1 D 0 iδ (ζ1 C ζN 1 )2 D0 ζ2 i ωζ2 D D1 ζ1 µ(ζ1 ζN 1 ) C 2ω i F i Ω T0 C C e i Ω T0 (ζ1 C ζN 1 ) e 2ω iδ D0 ζ3 i ωζ3 D D1 ζ2 D2 ζ1 µ(ζ2 ζN 2 ) C (ζ1 C ζN 1 )(ζ2 C ζN 2 ) ω iα i F i Ω T0 C C e i Ω T0 (ζ2 C ζN 2 ) (ζ1 C ζN 1 )3 C e 2ω 2ω Express the solution of the ﬁrst-order problem as ζ1 D A(T1 , T2 )e i ω T0 Then, show that the second-order problem becomes i δ i ω T0 N i ω T0 C Ae Ae 2ω N i ω T0 N i ω T0 C i F e i Ω T0 C e i Ω T0 Ae i ω T0 C Ae C µ Ae 2ω

D0 ζ2 i ωζ2 D D1 Ae i ω T0 µ Ae i ω T0 C

The Case Ω Away from 2ω When Ω is away from 2ω, there are no nearresonances at second order. Eliminate the secular term and obtain D1 A D µ A Then, show that ζ2 D

i µ N i ω T0 δ 2 2i ω T0 δ δ N2 2i ω T0 Ae A e C A e 2 A AN 2ω 2ω 2 ω 6ω 2 F F F N i(Ω Cω)T0 C Ae i(Ω Cω)T0 Ae i(ωΩ )T0 Ae 2ωΩ 2ωΩ 2ω(Ω C 2ω) F N i(Ω ω)T0 Ae C 2ω(Ω 2ω)

Substitute ζ1 and ζ2 into the third-order equation, use the expression for D1 A, eliminate the terms that lead to secular and small-divisor terms, and obtain i µ2 iF2 3α 5δ 2 D2 A D A2 AN AC A C i 2ω ω(Ω 2 4ω 2 ) 2ω 3ω 3

7.6 Exercises

when Ω is away from ω and 3ω, i µ2 iF2 3α 5δ 2 A2 AN AC A C i 2ω ω(Ω 2 4ω 2 ) 2ω 3ω 3 5i F δ 2 i σ T2 i F δ N i σ T2 C A e A Ae 6ω 3 3ω 3

D2 A D

when Ω ω, and i µ2 iF2 ACi AC D2 A D 2ω 5ω 3

3α 5δ 2 2ω 3ω 3

i F δ N2 i σ T2 A e A2 AN C 2ω 3

when Ω 3ω. Compare the results obtained with the method of multiple scales with those obtained in Section 7.5.1 with the method of normal forms. The Case Ω 2ω Eliminate the terms that produce secular and small-divisor terms from the second-order equation and obtain D1 A D µ A C

i F N i σ T1 Ae 2ω

Then, show that ζ2 D

i µ N i ω T0 δ 2 2i ω T0 δ δ N2 2i ω T0 Ae A e C A e 2 A AN 2 2ω 2ω ω 6ω 2 F F F N i(Ω Cω)T0 Ae C Ae i(Ω Cω)T0 Ae i(ωΩ )T0 2ωΩ 2ωΩ 2ω(Ω C 2ω)

Substitute ζ1 and ζ2 into the third-order equation, use the expression for D1 A, eliminate the terms that lead to secular and small-divisor terms, and obtain i µ2 iF2 3α 5δ 2 A2 AN D2 A D A A C i 2ω 4ω 2 (ΩC 2ω) 2ω 3ω 3 F µ(Ω C 2ω) N i σ T1 Ae 4ω 2 Ω Compare the results obtained with the method of multiple scales with those obtained in Section 7.5.2 with the method of normal forms. 7.6.3 Consider the equation uR C ω 20 u C 2 u3 cos 2t D 0 1 Use the methods of multiple scales and normal forms to determine the equations describing the amplitude and the phase to ﬁrst order when

215

216

7 Parametrically Excited Systems

a) ω 0 is away from 1 and 1/2, b) ω 0 1 , c) ω 0 1/2 . 7.6.4 The parametric excitation of a two-degree-of-freedom system is governed uR 1 C ω 21 u 1 C cos Ω t ( f 11 u 1 C f 12 u 2 ) D 0 uR 2 C ω 22 u 2 C cos Ω t ( f 21 u 2 C f 22 u 2 ) D 0 Use the methods of multiple scales and normal forms to determine the equations describing the amplitudes and the phases when Ω ω 2 ω 1 .

217

8 MDOF Systems with Quadratic Nonlinearities In this chapter, we treat multiple-degree-of-freedom nongyroscopic and gyroscopic systems with quadratic nonlinearities subject to harmonic excitations.

8.1 Nongyroscopic Systems

A general nonlinear multiple-degree-of-freedom nongyroscopic system can be modeled by P t) D 0 xR C Ax C D xP C N (x, x,

(8.1)

where A and D are n n matrices, x is a column vector of length n, and N is a P and t having n components. nonlinear vector function of x, x, In this section, we treat systems whose linear parts are expressed in normalmode form, and in Section 8.3 we treat two linearly coupled oscillators. The linear part of (8.1) can be put in the following normal-mode form by using the linear transformation x D P u: O (u, uP , t) D 0 uR C J u C DO uP C N

(8.2)

where J D P 1 AP ,

DO D P 1 D P ,

O D P 1 N (P u, P uP , t) N

We choose P so that J has a Jordan canonical form and assume that its diagonal elements are positive. We also assume modal damping so that DO is a diagonal matrix. We limit our discussion to systems for which the frequencies (i.e., eigenvalues of A) are distinct and hence J is diagonal. We can explore the principal features of the analysis and limit the algebra to a minimum by considering a system having three degrees of freedom. Thus, we consider uR m C ω 2m u m C 2µ m uP m D

@V (u 1 , u 2 , u 3 ) C 2 f m cos (Ω t C τ m ) @u m

(8.3)

The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

218

8 MDOF Systems with Quadratic Nonlinearities

where V D α 1 u31 C α 2 u21 u 2 C α 3 u 1 u22 C α 4 u21 u 3 C α 5 u 1 u23 C α 6 u32 C α 7 u22 u 3 C α 8 u 2 u23 C α 9 u33 C α 10 u 1 u 2 u 3

(8.4)

and the ω n , α n , f n , τ n , µ n , and Ω are constants independent of . Here, is a small nondimensional parameter used for bookkeeping purposes. Again, as a ﬁrst step, we cast (8.3) and (8.4) in complex-valued form using the transformation u m D ζ m C ζN m , zP m D i Ωm z m ,

uP m D i ω m ζ m ζN m z m D f m e i (Ω tCτ m )

(8.5) (8.6)

The result is i @V ζ1 C ζN 1 , ζ2 C ζN 2 , ζ3 C ζN 3 ζP m D i ω m ζ m µ m ζ m ζN m 2ω m @u m i (z m C zN m ) (8.7) 2ω m To simplify the O(1) terms in (8.7), we introduce the transformation ζ m D η m C ∆ m1 z m C ∆ m2 zN m

(8.8)

use (8.6), and obtain 1 zm ηP m D i ω m η m C i (ω m Ω ) ∆ m1 2ω m 1 zN m C O() C i (ω m C Ω ) ∆ m2 2ω m

(8.9)

There are two possibilities: (a) Ω ω m for m D 1 or 2 or 3; and (b) Ω is away from all of the ω m . The ﬁrst possibility corresponds to primary resonance of the mth mode. It is treated in Section 8.1.4. When Ω is away from all of the ω m , we choose the ∆ m n to eliminate z m and zN m from (8.9); that is, ∆ m1 D

1 2ω m (ω m Ω )

and

∆ m2 D

1 2ω m (ω m C Ω )

(8.10)

8.1 Nongyroscopic Systems

We note that there are no small divisors in (8.10) because we assumed that Ω is away from every ω m . Substituting (8.8) into (8.7) and using (8.10), we obtain " i z1 C zN 1 2 ηP 1 D i ω 1 η 1 µ 1 (η 1 ηN 1 ) 3α 1 η 1 C ηN 1 C 2 2ω 1 ω1 Ω 2 z1 C zN 1 z2 C zN 2 η C 2α 2 η 1 C ηN 1 C 2 C η N C 2 2 ω1 Ω 2 ω 22 Ω 2 z1 C zN 1 z3 C zN 3 η 3 C ηN 3 C 2 C 2α 4 η 1 C ηN 1 C 2 ω1 Ω 2 ω3 Ω 2 2 z2 C zN 2 z3 C zN 3 2 C α 3 η 2 C ηN 2 C 2 η C α C η N C 5 3 3 ω2 Ω 2 ω 23 Ω 2 # z2 C zN 2 z3 C zN 3 η 3 C ηN 3 C 2 Cα 10 η 2 C ηN 2 C 2 (8.11) ω2 Ω 2 ω3 Ω 2 " i z1 C zN 1 2 α 2 η 1 C ηN 1 C 2 ηP 2 D i ω 2 η 2 µ 2 (η 2 ηN 2 ) 2ω 2 ω1 Ω 2 z1 C zN 1 z2 C zN 2 η C 2α 3 η 1 C ηN 1 C 2 C η N C 2 2 ω1 Ω 2 ω 22 Ω 2 z2 C zN 2 z3 C zN 3 η C 2α 7 η 2 C ηN 2 C 2 C η N C 3 3 ω2 Ω 2 ω 23 Ω 2 2 z2 C zN 2 z3 C zN 3 2 C 3α 6 η 2 C ηN 2 C 2 η C α C η N C 8 3 3 ω2 Ω 2 ω 23 Ω 2 # z1 C zN 1 z3 C zN 3 η 3 C ηN 3 C 2 Cα 10 η 1 C ηN 1 C 2 (8.12) ω1 Ω 2 ω3 Ω 2 " i z1 C zN 1 2 α 4 η 1 C ηN 1 C 2 ηP 3 D i ω 3 η 3 µ 3 (η 3 ηN 3 ) 2ω 3 ω1 Ω 2 z1 C zN 1 z3 C zN 3 η 3 C ηN 3 C 2 C 2α 5 η 1 C ηN 1 C 2 ω1 Ω 2 ω3 Ω 2 z2 C zN 2 z3 C zN 3 η C 2α 8 η 2 C ηN 2 C 2 C η N C 3 3 ω2 Ω 2 ω 23 Ω 2 2 z2 C zN 2 z3 C zN 3 2 C α 7 η 2 C ηN 2 C 2 η C 3α C η N C 9 3 3 ω2 Ω 2 ω 23 Ω 2 # z1 C zN 1 z2 C zN 2 η 2 C ηN 2 C 2 Cα 10 η 1 C ηN 1 C 2 (8.13) ω1 Ω 2 ω2 Ω 2 Because zP n D i Ω z n , resonance occurs when ω s 2ω m : Two-to-one autoparametric resonance, ω s ω n C ω m : Combination autoparametric resonance,

219

220

8 MDOF Systems with Quadratic Nonlinearities

Ω 2ω m : Subharmonic resonance of order one-half, Ω 1/2ω m : Superharmonic resonance of order two, Ω ω n ˙ ω m : Combination resonance. The ﬁrst two resonances are called internal or autoparametric resonances, whereas the last three are called external resonances because they are produced by the excitation. We note that some of these resonances might occur simultaneously. Next, we treat the three autoparametric resonances ω 2 2ω 1 , ω 3 ω 2 C ω 1 , and ω 2 2ω 1 and ω 3 2ω 2 in conjunction with a few of the external resonances. 8.1.1 Two-to-One Autoparametric Resonance

We consider the case ω 2 2ω 1 and no other autoparametric resonances. To simplify (8.11)–(8.13), we introduce the near-identity transformation η m D ξm C h m ξn , ξN n , z n , zN n

(8.14)

and choose the h m to eliminate the nonresonance terms. The autoparametric resonance produces the near-resonance terms ξ2 ξN1 and ξ12 in (8.11) and (8.12), respectively. The near-resonance terms produced by the excitation depend on the resonances being considered. Next, normal forms corresponding to several of these resonances are presented. No External Resonance

iα 2 N ξP1 D i ω 1 ξ1 µ 1 ξ1 ξ2 ξ1 ω1 iα 2 2 ξP2 D i ω 2 ξ2 µ 2 ξ2 ξ 2ω 2 1 ξP3 D i ω 3 ξ3 µ 3 ξ3

(8.15) (8.16) (8.17)

When expressed in the polar coordinates ξn D 1/2r n e i β n , (8.15)–(8.17) become α 2 r1 r2 sin (β 2 2β 1 ) 2ω 1 α 2 2 rP2 D µ 2 r2 r sin (β 2 2β 1 ) 4ω 2 1 rP1 D µ 1 r1 C

rP3 D µ 3 r3 α 2 r1 r2 cos (β 2 2β 1 ) 2ω 1 α 2 2 rP2 βP 2 D ω 2 r2 r cos (β 2 2β 2 ) 4ω 1 1

rP1 βP 1 D ω 1 r1

r3 βP 3 D ω 3 r3

(8.18) (8.19) (8.20) (8.21) (8.22) (8.23)

8.1 Nongyroscopic Systems

Ω 1/2ω 1

iα 2 N i ξP1 D i ω 1 ξ1 µ 1 ξ1 ξ2 ξ1 ω1 2ω 1

3α 1 z12 α 3 z22 C 2 2 2 2 (ω 1 Ω ) (ω 2 Ω 2 )2

α 5 z32 2α 2 z1 z2 2α 4 z1 z3 C 2 C 2 2 2 2 2 2 2 (ω 3 Ω ) (ω 1 Ω )(ω 2 Ω ) (ω 1 Ω 2 )(ω 23 Ω 2 ) α 10 z2 z3 C 2 (8.24) (ω 2 Ω 2 )(ω 23 Ω 2 )

C

iα 2 2 ξP2 D i ω 2 ξ2 µ 2 ξ2 ξ 2ω 2 1

(8.25)

ξP3 D i ω 3 ξ3 µ 3 ξ3

(8.26)

Ω 2ω 2

iα 2 N ξP1 D i ω 1 ξ1 µ 1 ξ1 ξ2 ξ1 ω1 iα 2 2 i ξP2 D i ω 2 ξ2 µ 2 ξ2 ξ 2ω 2 1 ω 2

(8.27)

α 3 z1 3α 6 z2 α 7 z3 ξN2 C C ω 21 Ω 2 ω 22 Ω 2 ω 23 Ω 2 (8.28)

ξP3 D i ω 3 ξ3 µ 3 ξ3

(8.29)

Ω ω3 C ω2

iα 2 N ξP1 D i ω 1 ξ1 µ 1 ξ1 ξ2 ξ1 ω1

(8.30)

α 10 z1 2α 7 z2 2α 8 z3 N ξ3 C C ω 21 Ω 2 ω 22 Ω 2 ω 23 Ω 2 (8.31) α 10 z1 i 2α 7 z2 2α 8 z3 ξP3 D i ω 3 ξ3 µ 3 ξ3 ξN2 (8.32) C C 2ω 3 ω 21 Ω 2 ω 22 Ω 2 ω 23 Ω 2 iα 2 2 i ξP2 D i ω 2 ξ2 µ 2 ξ2 ξ 2ω 2 1 2ω 2

Ω ω3 ω1

ξP1 D i ω 1 ξ1 µ 1 ξ1

2α 4 zN 1 iα 2 N i α 10 zN 2 2α 5 zN 3 ξ3 ξ2 ξ1 C 2 C 2 (8.33) ω1 2ω 1 ω 21 Ω 2 ω2 Ω 2 ω3 Ω 2 iα 2 2 ξP2 D i ω 2 ξ2 µ 2 ξ2 ξ (8.34) 2ω 2 1 2α 4 z1 i α 10 z2 2α 5 z3 ξP3 D i ω 3 ξ3 µ 3 ξ3 ξ1 (8.35) C 2 C 2 2ω 3 ω 21 Ω 2 ω2 Ω 2 ω3 Ω 2

221

222

8 MDOF Systems with Quadratic Nonlinearities

Ω 2ω 2 and Ω ω 3 C ω 1

iα 2 N ξP1 D i ω 1 ξ1 µ 1 ξ1 ξ2 ξ1 ω1 2α 4 z1 i α 10 z2 2α 5 z3 ξN3 C C (8.36) 2ω 1 ω 21 Ω 2 ω 22 Ω 2 ω 23 Ω 2 α 3 z1 iα 2 2 i 3α 6 z2 α 7 z3 ξN2 ξP2 D i ω 2 ξ2 µ 2 ξ2 ξ1 C C 2ω 2 ω 2 ω 21 Ω 2 ω 22 Ω 2 ω 23 Ω 2 (8.37) 2α 4 z1 i α 10 z2 2α 5 z3 ξP3 D i ω 3 ξ3 µ 3 ξ3 ξN1 (8.38) C C 2ω 3 ω 21 Ω 2 ω 22 Ω 2 ω 23 Ω 2 8.1.2 Combination Autoparametric Resonance

We consider the autoparametric-resonance condition ω 3 ω 2 C ω 1 , which produces the near-resonance terms ξ3 ξN2 , ξ3 ξN1 , and ξ1 ξ2 in (8.11)–(8.13), respectively, In the absence of external resonances, the normal form of (8.11)–(8.13) is iα 10 N ξP1 D i ω 1 ξ1 µ 1 ξ1 ξ3 ξ2 2ω 1 iα 10 N ξP2 D i ω 2 ξ2 µ 2 ξ2 ξ3 ξ1 2ω 2 iα 10 ξP3 D i ω 3 ξ3 µ 3 ξ3 ξ1 ξ2 2ω 3

(8.39) (8.40) (8.41)

When expressed in the polar coordinates ξn D 1/2r n e i β n , (8.39)–(8.41) become α 10 r2 r3 sin (β 3 β 2 β 1 ) 4ω 1 α 10 rP2 D µ 2 r2 C r1 r3 sin (β 3 β 2 β 1 ) 4ω 2 α 10 rP3 D µ 3 r3 r1 r2 sin (β 3 β 2 β 1 ) 4ω 3 α 10 rP1 βP 1 D ω 1 r1 r2 r3 cos (β 3 β 2 β 1 ) 4ω 1 α 10 rP2 βP 2 D ω 2 r2 r1 r3 cos (β 3 β 2 β 1 ) 4ω 2 α 10 rP3 βP 3 D ω 3 r3 r1 r2 cos (β 3 β 2 β 1 ) 4ω 3 rP1 D µ 1 r1 C

(8.42) (8.43) (8.44) (8.45) (8.46) (8.47)

The normal forms in cases of external resonances can be obtained by adding appropriate terms to (8.39)–(8.41), as in Section 8.1.1.

8.1 Nongyroscopic Systems

8.1.3 Simultaneous Two-to-One Autoparametric Resonances

We consider the autoparametric resonances ω 3 2ω 2 and ω 2 2ω 1 , which produce the near-resonance terms ξ2 ξN1 , ξ12 and ξ3 ξN2 , and ξ22 in (8.11)–(8.13), respectively. Consequently, in the absence of external resonances, the normal form of (8.11)–(8.13) is iα 2 N ξP1 D i ω 1 ξ1 µ 1 ξ1 ξ2 ξ1 ω1 i ξP2 D i ω 2 ξ2 µ 2 ξ2 α 2 ξ12 C 2α 7 ξ3 ξN2 2ω 2 iα 7 2 ξP3 D i ω 3 ξ3 µ 3 ξ3 ξ 2ω 3 2

(8.48) (8.49) (8.50)

When expressed in the polar coordinates ξn D 1/2r n e i β n , (8.48)–(8.50) become rP1 D µ 1 r1 C rP2 D µ 2 r2 rP3 D µ 3 r3 rP1 βP 1 D ω 1 r1 rP2 βP 2 D ω 2 r2 rP3 βP 3 D ω 3 r3

α 2 r1 r2 sin (β 2 2β 1 ) 2ω 1 α 2 2 r sin (β 2 2β 1 ) C 4ω 2 1 α 7 2 r sin (β 3 2β 2 ) 4ω 3 2 α 2 r1 r2 cos (β 2 2β 1 ) 2ω 1 α 2 2 r cos (β 2 2β 1 ) 4ω 2 1 α 7 2 r cos (β 3 2β 2 ) 4ω 3 2

(8.51) α 7 r2 r3 sin (β 3 2β 2 ) 2ω 2

(8.52) (8.53) (8.54)

α 7 r2 r3 cos (β 3 2β 2 ) 2ω 2

(8.55) (8.56)

This form is in full agreement with that obtained by Nayfeh, Asrar, and Nayfeh (1992) by using the method of multiple scales. Again, the normal forms in the presence of external resonances can be obtained by adding appropriate terms to (8.48)–(8.50), as in Section 8.1.1. 8.1.4 Primary Resonances

To treat this case, we scale the excitation amplitudes f n and hence the z n at O() so that the resonance terms produced by the excitation appear at the same order as those produced by the nonlinearity. Thus, we modify (8.7) into 2 i i n (z1 C zN 1 ) ζP 1 D i ω 1 ζ1 µ 1 ζ1 ζN 1 3α 1 ζ1 C ζN 1 2ω 1 2ω 1 2 C 2α 2 ζ1 C ζN 1 ζ2 C ζN 2 C α 3 ζ2 C ζN 2 C 2α 4 ζ1 C ζN 1 ζ3 C ζN 3 2 o Cα 5 ζ3 C ζN 3 C α 10 ζ2 C ζN 2 ζ3 C ζN 3 (8.57)

223

224

8 MDOF Systems with Quadratic Nonlinearities

2 i i n (z2 C zN 2 ) ζP 2 D i ω 2 ζ2 µ 2 ζ2 ζN 2 α 2 ζ1 C ζN 1 2ω 2 2ω 2 2 C 2α 3 ζ1 C ζN 1 ζ2 C ζN 2 C 3α 6 ζ2 C ζN 2 2 o C2α 7 ζ2 C ζN 2 ζ3 C ζN 3 C α 8 ζ3 C ζN 3 C α 10 ζ1 C ζN 1 ζ3 C ζN 3 (8.58) 2 i i n (z3 C zN 3 ) ζP 3 D i ω 3 ζ3 µ 3 ζ3 ζN 3 α 4 ζ1 C ζN 1 2ω 3 2ω 3 2 N N N C 2α 5 ζ1 C ζ1 ζ3 C ζ3 C α 7 ζ2 C ζ2 C 2α 8 ζ2 C ζN 2 ζ3 C ζN 3 2 o C3α 9 ζ3 C ζN 3 C α 10 ζ1 C ζN 1 ζ2 C ζN 2 (8.59) To simplify (8.57)–(8.59), we introduce the near-identity transformation ζ m D ξm C h m ξn , ξN n , z n , zN n

(8.60)

and choose the h m to eliminate the nonresonance terms. The resulting normal forms depend on the resonance conditions being considered. As discussed in the preceding section, internal or autoparametric resonances occur when ω n 2ω m and/or ω s ω n ˙ ω m . Primary resonances occur when Ω ω m . Consequently, the normal forms of (8.57)–(8.59) for a few of these resonances are: ω 2 2ω 1 and Ω ω n

iα 2 N i ξP1 D i ω 1 ξ1 µ 1 ξ1 ξ2 ξ1 δ 1n z1 ω1 2ω 1 iα 2 2 i ξP2 D i ω 2 ξ2 µ 2 ξ2 ξ δ 2n z2 2ω 2 1 2ω 2 i ξP3 D i ω 3 ξ3 µ 3 ξ3 δ 3n z3 2ω 3

(8.61) (8.62) (8.63)

where δ m n is the Kronecker delta. ω 3 ω 2 C ω 1 and Ω ω n

iα 10 N ξP1 D i ω 1 ξ1 µ 1 ξ1 ξ3 ξ2 2ω 1 iα 10 N ξP2 D i ω 2 ξ2 µ 2 ξ2 ξ3 ξ1 2ω 2 iα 10 ξP3 D i ω 3 ξ3 µ 3 ξ3 ξ1 ξ2 2ω 3

i δ 1n z1 2ω 1 i δ 2n z2 2ω 2 i δ 3n z3 2ω 3

(8.64) (8.65) (8.66)

8.2 Gyroscopic Systems

ω 3 2ω 2 , ω 2 2ω 1 , and Ω ω n

iα 2 N i ξP1 D i ω 1 ξ1 µ 1 ξ1 ξ2 ξ1 δ 1n z1 ω1 2ω 1 i i ξP2 D i ω 2 ξ2 µ 2 ξ2 α 2 ξ12 C 2α 7 ξ3 ξN2 δ 2n z2 2ω 2 2ω 2 iα 7 2 i ξP3 D i ω 3 ξ3 µ 3 ξ3 ξ δ 3n z3 2ω 3 2 2ω 3

(8.67) (8.68) (8.69)

ω 3 ω 2 C ω 1 , ω 2 2ω 1 , and Ω ω n

i i ξP1 D i ω 1 ξ1 µ 1 ξ1 2α 2 ξ2 ξN1 C α 10 ξ3 ξN2 δ 1n z1 2ω 1 2ω 1 i i ξP2 D i ω 2 ξ2 µ 2 ξ2 α 2 ξ12 C α 10 ξ3 ξN1 δ 2n z2 2ω 2 2ω 2 iα 10 i ξP3 D i ω 3 ξ3 µ 3 ξ3 ξ1 ξ2 δ 3n z3 2ω 3 2ω 3

(8.70) (8.71) (8.72)

8.2 Gyroscopic Systems

As in Section 7.4, for simplicity, we consider a two-degree-of-freedom system to illustrate the method of solution. Speciﬁcally, we consider uR 1 C λ 1 uP 2 C α 1 u 1 D 3δ 1 u21 C 2δ 2 u 1 u 2 C δ 3 u22 C 2F1 cos (Ω t C τ 1 ) (8.73) uR 2 λ 2 uP 1 C α 2 u 2 D δ 2 u21 C 2δ 3 u 1 u 2 C 3δ 4 u22 C 2F2 cos (Ω t C τ 2 ) (8.74) Using the transformation (7.144)–(7.147) and following steps similar to those used in Section 7.4, we cast (8.73) and (8.74) in the complex-valued form λ1 i ω 1 (α 1 ω 22 ) (z1 C zN 1 ) (z2 C zN 2 ) ζP 1 D i ω 1 ζ1 C 2α 1 (ω 22 ω 21 ) 2(ω 22 ω 21 ) 2 C χ 11 ζ1 C ζN 1 C ζ2 C ζN 2 C χ 12 ζ1 C ζN 1 C ζ2 C ζN 2 i(α 1 ω 22 ) i(α 1 ω 21 ) N N ζ1 ζ1 C ζ2 ζ2 λ1 ω1 λ1 ω2 i(α 1 ω 22 ) 2 i(α 1 ω 21 ) C χ 13 ζ1 ζN 1 C ζ2 ζN 2 λ1 ω1 λ1 ω2

(8.75)

225

226

8 MDOF Systems with Quadratic Nonlinearities

λ1 i ω 2 (α 1 ω 21 ) (z1 C zN 1 ) C (z2 C zN 2 ) ζP 2 D i ω 2 ζ2 2 2 2 2α 1 (ω 2 ω 1 ) 2(ω 2 ω 21 ) 2 C χ 21 ζ1 C ζN 1 C ζ2 C ζN 2 C χ 22 ζ1 C ζN 1 C ζ2 C ζN 2 i(α 1 ω 22 ) i(α 1 ω 21 ) ζ1 ζN 1 C ζ2 ζN 2 λ1 ω1 λ1 ω2 2 i(α 1 ω 22 ) 2 i(α 1 ω 1 ) C χ 23 ζ1 ζN 1 C ζ2 ζN 2 λ1 ω1 λ1 ω2

(8.76)

on account of (8.6), where χ 11 D

3i δ 1 ω 1 (α 1 ω 22 ) δ 2 α 1 λ 1 , 2α 1 (ω 22 ω 21 )

χ 12 D

i δ 2 ω 1 (α 1 ω 22 ) δ 3 α 1 λ 1 α 1 (ω 22 ω 21 )

χ 13 D

i δ 3 ω 1 (α 1 ω 22 ) 3δ 4 α 1 λ 1 , 2α 1 (ω 22 ω 21 )

χ 21 D

δ 2 α 1 λ 1 3i δ 1 ω 2 (α 1 ω 21 ) 2α 1 (ω 22 ω 21 )

χ 22 D

δ 3 α 1 λ 1 i δ 2 ω 2 (α 1 ω 21 ) , α 1 (ω 22 ω 21 )

χ 33 D

3δ 4 α 1 λ 1 i δ 3 ω 2 (α 1 ω 21 ) 2α 1 (ω 22 ω 21 )

As discussed in Section 8.1, the excitation may produce one of two possible types of resonances: primary and secondary resonances. Primary resonances are treated in the next section, and secondary resonances are treated in Section 8.2.2. 8.2.1 Primary Resonances

As discussed in Section 8.1.4, we scale the f n and hence the z n at O() so that the resonance terms produced by the excitation appear at the same order as those produced by the nonlinearity. To simplify (8.75) and (8.76), we introduce the near-identity transformation (8.60) and choose the h m to eliminate the nonresonance terms. Resonance terms occur when ω 2 2ω 1 (i.e., when there is a two-to-one autoparametric resonance) and Ω ω n (i.e., when there is a primary resonance of the nth mode). When ω 2 2ω 1 and Ω ω n , the normal form of (8.75) and (8.76) is i ω 1 (α 1 ω 2 )z1 λ 1 α 1 z2 ξP1 D i ω 1 ξ1 C Λ 1 ξ2 ξN1 C δ 1n 2α 1 (ω 22 ω 21 ) i ω 2 (α 1 ω 21 )z1 λα 1 z2 ξP2 D i ω 2 ξ2 C Λ 2 ξ12 δ 2n 2α 1 (ω 22 ω 21 ) 2

(8.77) (8.78)

where Λ 1 D 2χ 11 Λ 2 D χ 21 C

i χ 12 (ω 2 ω 1 )(α 1 C ω 1 ω 2 ) 2χ 13 α 1 λ 2 λ1 ω1 ω2 λ1 ω1 ω2 i χ 22 (α 1 ω 21 ) χ 23 (α 1 ω 21 )2 λ1 ω1 λ 21 ω 21

(8.79) (8.80)

8.2 Gyroscopic Systems

8.2.2 Secondary Resonances

To treat this case, we ﬁrst introduce the linear transformation (8.8) and choose ∆ m1 and ∆ m2 to eliminate z n and zN n from (8.75) and (8.76); that is, z1 ω 1 (α 1 ω 22 ) zN 1 C ω1 C Ω 2α 1 (ω 22 ω 21 ) ω 1 Ω z2 i λ1 zN 2 C ω1 C Ω 2(ω 22 ω 21 ) ω 1 Ω z1 ω 2 (α 1 ω 21 ) zN 1 ζ2 D η 2 C C ω2 C Ω 2α 1 (ω 22 ω 21 ) ω 2 Ω i λ1 zN 2 z2 C C ω2 C Ω 2(ω 22 ω 21 ) ω 2 Ω

ζ1 D η 1

(8.81)

(8.82)

Hence, (8.75) and (8.76) become ηP 1 D i ω 1 η 1 C χ 11 [η 1 C ηN 1 C η 2 C ηN 2 C b 1 (z1 C zN 1 ) C b 2 (z2 zN 2 )]2 C χ 12 [η 1 C ηN 1 C η 2 C ηN 2 C b 1 (z1 C zN 1 ) C b 2 (z2 zN 2 )] i(α 1 ω 22 ) i(α 1 ω 21 ) (η 1 ηN 1 ) C (η 2 ηN 2 ) C b 3 (z1 zN 1 ) λ1 ω1 λ1 ω2 i(α 1 ω 21 ) (η 1 ηN 1 ) Cb 4 (z2 C zN 2 ) C χ 13 λ2 ω1 2 i(α 1 ω 22 ) (η 2 ηN 2 ) C b 3 (z1 zN 1 ) C b 4 (z2 C zN 2 ) C (8.83) λ1 ω2 ηP 2 D i ω 2 η 2 C χ 21 [η 1 C ηN 1 C η 2 C ηN 2 C b 1 (z1 C zN 1 ) C b 2 (z2 zN 2 )]2 C χ 22 [η 1 C ηN 1 C η 2 C ηN 2 C b 1 (z1 C zN 1 ) C b 2 (z2 zN 2 )] i(α 1 ω 22 ) i(α 1 ω 21 ) (η 1 ηN 1 ) C (η 2 ηN 2 ) C b 3 (z1 zN 1 ) λ1 ω1 λ1 ω2 i(α 1 ω 21 ) (η 1 ηN 1 ) Cb 4 (z2 C zN 2 ) C χ 23 λ1 ω1 2 i(α 1 ω 22 ) (η 2 ηN 2 ) C b 3 (z1 zN 1 ) C b 4 (z2 C zN 2 ) C (8.84) λ1 ω2 where (b 1 , b 2 , b 3 , b 4 ) D

(α 2 Ω 2 , i λ 1 Ω , i λ 2 Ω , α 1 Ω 2 ) (ω 21 Ω 2 )(ω 22 Ω 2 )

(8.85)

It follows from (8.83) and (8.84) that, to this order, only a two-to-one autoparametric resonance is possible and that the excitation can produce resonances corresponding to

227

228

8 MDOF Systems with Quadratic Nonlinearities

Ω 2ω m : Subharmonic resonance of order one-half of the mth mode, Ω 1/2ω m : Superharmonic resonance of order two of the mth mode, Ω ω 2 ˙ ω 1 : Combination resonance of the additive or difference types. Next, we present the normal forms for several of the external resonances for the case of two-to-one autoparametric resonance. Because (8.83) and (8.84) were derived under the assumption that Ω is away from ω 1 and ω 2 , the resonance conditions Ω 2ω 1 , Ω 1/2ω 2 , and Ω ω 2 ω 1 must be excluded when considering the two-to-one autoparametric resonance. Ω 2ω 2

ξP1 D i ω 1 ξ1 C Λ 1 ξ2 ξN1

(8.86)

ξP2 D i ω 2 ξ2 C Λ 2 ξ12 C Γ ξN2

(8.87)

where Λ 1 and Λ 2 are deﬁned in (8.79) and (8.80) and (α 1 ω 22 ) (b 3 z1 C b 4 z2 ) Γ D 2χ 21 (b 1 z1 C b 2 z2 ) 2i χ 23 λ1 ω2 i(α 1 ω 22 ) (b 1 z1 C b 2 z2 ) C χ 22 b 3 z1 C b 4 z2 λ1 ω2

(8.88)

Ω 1/2ω 1

ξP1 D i ω 1 ξ1 C Λ 1 ξ2 ξN1 C Γ

(8.89)

ξP2 D i ω 2 ξ2 C Λ 2 ξ12

(8.90)

where Λ 1 and Λ 2 are deﬁned in (8.79) and (8.80) and Γ D χ 11 (b 1 z1 C b 2 z2 )2 C χ 12 (b 1 z1 C b 2 z2 ) (b 3 z1 C b 4 z2 ) C χ 13 (b 3 z1 C b 4 z2 )2

(8.91)

Ω ω2 C ω1

ξP1 D i ω 1 ξ1 C Λ 1 ξ2 ξN1 C Γ1 ξN2

(8.92)

ξP2 D i ω 2 ξ2 C Λ 2 ξ12 C Γ2 ξN1

(8.93)

where Λ 1 and Λ 2 are deﬁned in (8.79) and (8.80) and i α 1 ω 22 (b 3 z1 C b 4 z2 ) Γ1 D 2χ 11 (b 1 z1 C b 2 z2 ) 2χ 13 λ1 ω2 # " i α 1 ω 22 (b 1 z1 C b 2 z2 ) C χ 12 b 3 z1 C b 4 z2 λ1 ω2

(8.94)

8.3 Two Linearly Coupled Oscillators

i α 1 ω 21 (b 3 z1 C b 4 z2 ) Γ2 D 2χ 21 (b 1 z1 C b 2 z2 ) 2χ 23 λ1 ω1 # " i α 1 ω 21 (b 1 z1 C b 2 z2 ) C χ 22 b 3 z1 C b 4 z2 λ1 ω1

(8.95)

8.3 Two Linearly Coupled Oscillators

In this section, we consider the two linearly coupled oscillators uR 1 C k11 u 1 C k12 u 2 D 2 d1 uP 1 C α 11 u21 C α 12 u22 C α 13 u 1 u 2

(8.96)

uR 2 C k21 u 1 C k22 u 2 D 2 d2 uP 2 C α 21 u21 C α 22 u22 C α 23 u 1 u 2

(8.97)

To determine the normal form of (8.96) and (8.97), we use two approaches. First, we use these equations as they are. Second, we ﬁrst transform them so that their linear parts are in normal-mode form, as in Section 8.1. The solution of the linearly undamped equations (8.96) and (8.97) can be expressed as u 1 D A 1 e i ω 1 t C A 2 e i ω 2 t C cc

(8.98)

u 2 D Γ1 A 1 e i ω 2 t C Γ2 A 2 e i ω 2 t C cc

(8.99)

when ω 1 ¤ ω 2 , where ω 4 (k11 C k22 ) ω 2 C k11 k22 k12 k21 D 0 Γi D

k11 ω 21 k21 D k12 k22 ω 2i

for i D 1 and 2

(8.100)

Using (8.98) and (8.99), we express (8.96) and (8.97) as a system of two ﬁrst-order complex-valued equations by using the transformation u 1 D ζ1 C ζN 1 C ζ2 C ζN 2

(8.101)

u 2 D Γ1 ζ1 C ζN 1 C Γ2 ζ2 C ζN 2

(8.102)

uP 1 D i ω 1 ζ1 ζN 1 C i ω 2 ζ2 ζN 2

(8.103)

uP 2 D i ω 1 Γ1 ζ1 ζN 1 C i ω 2 Γ2 ζ1 ζN 2

(8.104)

229

230

8 MDOF Systems with Quadratic Nonlinearities

Solving (8.101)–(8.104), we have 1 Γ2 u 1 u 2 C ζ1 D 2(Γ2 Γ1 ) 1 u 2 Γ1 u 1 C ζ2 D 2(Γ2 Γ1 )

i ( uP 2 Γ2 uP 1 ) ω1 i (Γ1 uP 1 uP 2 ) ω2

Differentiating (8.105) and (8.106) with respect to t, we obtain 1 i ( uR 2 Γ2 uR 1 ) ζP 1 D Γ2 uP 1 uP 2 C 2(Γ2 Γ1 ) ω1 1 i (Γ1 uR 1 uR 2 ) ζP 2 D uP 2 Γ1 uP 1 C 2(Γ2 Γ1 ) ω2

(8.105) (8.106)

(8.107) (8.108)

Eliminating uR 1 and uR 2 from (8.107) and (8.108) by using (8.96) and (8.97), we obtain ω 2 Γ2 (d2 d1 ) (d2 Γ1 d1 Γ2 ) ζP 1 D i ω 1 ζ1 C ζ1 ζN 1 C ζ2 ζN 2 Γ2 Γ1 ω 1 (Γ2 Γ1 ) n 2 i (α 21 Γ2 α 11 ) ζ1 C ζN 1 C ζ2 C ζN 2 C 2ω 1 (Γ2 Γ1 ) 2 C (α 22 Γ2 α 12 ) Γ1 ζ1 C ζN 1 C Γ2 ζ2 C ζN 2 o C (α 23 Γ2 α 13 ) ζ1 C ζN 1 C ζ2 C ζN 2 Γ1 ζ1 C ζN 1 C Γ2 ζ2 C ζN 2 (8.109) (d1 Γ1 d2 Γ2 ) ω 1 Γ1 (d1 d2 ) ζP 2 D i ω 2 ζ2 C ζ1 ζN 1 C ζ2 ζN 2 ω 2 (Γ2 Γ1 ) Γ2 Γ1 n 2 i (Γ1 α 11 α 21 ) ζ1 C ζN 1 C ζ2 C ζN 2 C 2ω 2 (Γ2 Γ1 ) 2 C (Γ1 α 12 α 22 ) Γ1 ζ1 C ζN 1 C Γ2 ζ2 C ζN 2 o C (Γ1 α 13 α 23 ) ζ1 C ζN 1 C ζ2 C ζN 2 Γ1 ζ1 C ζN 1 C Γ2 ζ2 C ζN 2 (8.110) To simplify (8.109) and (8.110), we introduce the near-identity transformation (8.111) ζ m D ξm C h m ξ1 , ξN1 , ξ2 , ξN2 and choose the h m to eliminate the nonresonance terms. There are two cases: twoto-one autoparametric resonance (i.e., ω 2 2ω 1 ) and no autoparametric resonance. Next, we present the normal forms for these cases starting with the second. No Autoparametric Resonance

ξP1 D i ω 1 ξ1 µ 1 ξ1

(8.112)

ξP2 D i ω 2 ξ2 µ 2 ξ2

(8.113)

8.3 Two Linearly Coupled Oscillators

where µ1 D

d1 Γ2 d2 Γ1 d2 (k11 ω 21 ) d1 (k11 ω 22 ) D Γ2 Γ1 ω 22 ω 21

(8.114)

µ2 D

d2 Γ2 d1 Γ1 d2 (k11 ω 22 ) d1 (k11 ω 21 ) D Γ2 Γ1 ω 21 ω 22

(8.115)

Two-to-One Autoparametric Resonance

ξP1 D i ω 1 ξ1 µ 1 ξ1 C iΛ 1 ξ2 ξN1

(8.116)

ξP2 D i ω 2 ξ2 µ 2 ξ2 C iΛ 2 ξ12

(8.117)

where 1 [2 (α 21 Γ2 α 11 ) C 2Γ1 Γ2 (α 22 Γ2 α 12 ) 2ω 1 (Γ2 Γ1 )

Λ1 D

Λ2 D

C (Γ1 C Γ2 ) (α 23 Γ2 α 13 )]

(8.118)

1 Γ1 α 11 α 21 C Γ12 (Γ1 α 12 α 22 ) 2ω 2 (Γ2 Γ1 ) CΓ1 (Γ1 α 13 α 23 )

(8.119)

Next, we ﬁrst transform the linear parts of (8.96) and (8.97) into a normal-mode form. To accomplish this, we note from (8.98)–(8.100) that the eigenvectors corresponding to the eigenvalues ω 1 and ω 2 are [1 Γ1 ]T and [1 Γ2 ]T . Hence, we introduce the transformation 1 1 v (8.120) u D Pv D Γ1 Γ2 into (8.96) and (8.97) and obtain

1 vR1 k 1 1 v1 k12 C 11 Γ2 vR2 k21 k22 Γ1 Γ2 v2 d 1 1 vP1 0 D 2 1 0 d2 Γ1 Γ2 vP2 α (v C v2 )2 C α 12 (Γ1 v1 C Γ2 v2 )2 C α 13 (v1 C v2 )(Γ1 v1 C Γ2 v2 ) C 11 1 α 21 (v1 C v2 )2 C α 22 (Γ1 v1 C Γ2 v2 )2 C α 23 (v1 C v2 )(Γ1 v1 C Γ2 v2 ) 1 Γ1

(8.121) Multiplying (8.121) from the left with P 1 D

1 Γ2 Γ1

Γ2 Γ1

1 1

(8.122)

231

232

8 MDOF Systems with Quadratic Nonlinearities

and after some algebraic manipulations, we obtain 2 vR1 ω1 0 v1 µ 1 µ 3 vP1 C D 2 0 ω 22 v2 vR2 µ 4 µ 2 vP2 2 3 (Γ2 α 11 α 21 ) (v1 C v2 )2 C (Γ2 α 12 α 22 ) (Γ1 v1 C Γ2 v2 )2 6 7 C (Γ2 α 13 α 23 ) (v1 C v2 ) (Γ1 v1 C Γ2 v2 ) 6 7 C 2 25 4 (α ) (v ) ) (Γ ) (α Γ2 Γ1 21 Γ1 α 11 1 C v2 C 22 Γ1 α 12 1 v1 C Γ2 v2 C (α 23 Γ1 α 13 ) (v1 C v2 ) (Γ1 v1 C Γ2 v2 ) (8.123) where µ 1 and µ 2 are deﬁned in (8.114) and (8.115) and µ3 D

Γ2 (d1 d2 ) Γ1 (d2 d1 ) and µ 4 D Γ2 Γ1 Γ2 Γ1

(8.124)

Next, we transform (8.123) into two ﬁrst-order complex-valued equations by letting (8.125) v1 D ζ1 C ζN 1 , vP1 D i ω 1 ζ1 ζN 1 v2 D ζ2 C ζN 2 , vP2 D i ω 2 ζ2 ζN 2

(8.126)

Using (8.125) and (8.126), we transform (8.123) into (8.109) and (8.110). Then, we obtain the normal forms (8.112) and (8.113) in the absence of autoparametric resonance and (8.116) and (8.117) in the presence of autoparametric resonance.

8.4 Exercises

8.4.1 Use the methods of multiple scales and normal forms to determine a ﬁrstorder uniform expansion for uR 1 C ω 21 u 1 D α 1 u 1 u 2 uR 2 C ω 22 u 2 D α 2 u21 for small but ﬁnite amplitudes when ω 2 2ω 1 . 8.4.2 Use the methods of multiple scales and normal forms to determine ﬁrst-order uniform expansions for uR 1 C ω 21 u 1 D α 1 u 1 u 2 C k1 cos Ω1 t uR 2 C ω 22 u 2 D α 2 u21 C k2 cos Ω2 t when a) ω 2 2ω 1 and Ω1 ω 1 , b) ω 2 2ω 1 and Ω2 ω 2 .

8.4 Exercises

8.4.3 Determine the normal form of uR 1 C 12 uP 2 C δ u D u 1 u 2 uR 2 12 uP 1 C 12 u 2 D u21 when δ 1/2. 8.4.4 Determine the normal form of uR 1 C ω 21 u 1 D α 1 u 2 u 3 uR 2 C ω 22 u 2 D α 2 u 1 u 3 uR 3 C ω 23 u 3 D α 3 u 1 u 2 when ω 3 ω 2 C ω 1 . 8.4.5 Determine the normal form of uR 1 C 2 uP 2 C 3u 1 C 2 cos Ω t ( f 11 u 1 C f 12 u 2 ) D 0 uR 2 C uP 1 C 12u 2 C 2 cos Ω t ( f 21 u 1 C f 22 u 2 ) D 0 when Ω 5 and Ω 1. 8.4.6 Determine the normal form of uR 1 C 4 uP 2 C 3u 1 C 2 cos Ω t ( f 11 u 1 C f 12 u 2 ) D 0 uR 2 uP 1 C 3u 2 C 2 cos Ω t ( f 21 u 1 C f 22 u 2 ) D 0 when Ω 4 and Ω 2. 8.4.7 Determine the normal form of uR 1 C 5 uP 2 C 6u 1 C 2 cos Ω t ( f 11 u 1 C f 12 u 2 ) D 0 uR 2 C uP 1 C 24u 2 C 2 cos Ω t ( f 21 u 1 C f 22 u 2 ) D 0 when Ω 7 and Ω 1. 8.4.8 Use the methods of multiple scales and normal forms to determine an approximation to the solution of uR 1 C ω 21 u 1 D α 1 uP 1 uP 2 uR 2 C ω 22 u 2 D α 2 uP 21 ω 2 2ω 1 8.4.9 Use the methods of multiple scales and normal forms to determine a ﬁrstorder approximation to the solution of uR 1 C ω 21 u 1 D α 1 u21 C α 2 u 1 u 2 C α 3 u22 uR 2 C ω 22 u 2 D α 4 u21 C α 5 u 1 u 2 C α 6 u32 ω 2 2ω 1

233

234

8 MDOF Systems with Quadratic Nonlinearities

8.4.10 Use the methods of multiple scales and normal forms to determine a ﬁrstorder approximation to uR 1 C ω 21 u 1 C δ 1 u21 C 2δ 2 u 1 u 2 C δ 3 u22 D F1 cos Ω t uR 2 C ω 22 u 2 C δ 2 u21 C 2δ 3 u 1 u 2 C δ 4 u22 D 0 where ω 2 2ω 1

and

Ω 2ω 1

8.4.11 Use the methods of multiple scales and normal forms to determine a ﬁrstorder approximation to uR 1 C ω 21 u 1 C δ 1 u21 C 2δ 2 u 1 u 2 C δ 3 u22 D 2F1 cos Ω t uR 2 C ω 22 u 2 C δ 2 u21 C 2δ 3 u 1 u 2 C δ 4 u22 D 0 where ω 2 2ω 1

and

Ω ω1

235

9 TDOF Systems with Cubic Nonlinearities In this chapter, we consider two-degree-of-freedom nongyroscopic and gyroscopic systems with cubic nonlinearities. Nongyroscopic systems are treated in the following section, and gyroscopic systems are treated in Section 9.2.

9.1 Nongyroscopic Systems

We assume that the system can be modeled by (8.1), where A and D are 2 2 matrices and x and N are column vectors of length 2. Moreover, we assume that a linear transformation x D P u has been introduced so that (8.1) becomes (8.2) and that J and D are diagonal. The case in which J has a generic nonsemisimple structure is treated in Sections 9.1.5 and 9.1.6. Thus, we rewrite (8.2) as uR m C ω 2m u m C 2µ m uP m D

@V (u 1 , u 2 ) C 2 f m cos (Ω t C τ m ) @u m

(9.1)

where V D δ 1 u41 C δ 2 u31 u 2 C δ 3 u21 u22 C δ 4 u 1 u32 C δ 5 u42

(9.2)

Again, as a ﬁrst step, we cast (9.1) in complex-valued form using the transformation u m D ζ m C ζN m , uP m D i ω m ζ m ζN m zP m D i Ω z m ,

z m D f m e i (Ω tCτ m )

and obtain ζP m D i ω m ζ m µ m ζ m ζN m

i (z m C zN m ) 2ω m i @V ζ1 C ζN 1 , ζ2 C ζN 2 2ω m @u m

(9.3)

Before proceeding further, we must distinguish between two cases: Ω ω m (corresponding to the primary resonance of the mth mode) and Ω is away from both The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

236

9 TDOF Systems with Cubic Nonlinearities

ω 1 and ω 2 . We start with the latter case and consider the primary-resonance case in Section 9.1.4. When Ω is away from ω 1 and ω 2 , we use the transformation (8.8) and (8.10) and rewrite (9.3) as " i z1 C zN 1 3 ηP 1 D i ω 1 η 1 µ 1 (η 1 ηN 1 ) 4δ 1 η 1 C ηN 1 C 2 2ω 1 ω1 Ω 2 2 z1 C zN 1 z2 C zN 2 η C 3δ 2 η 1 C ηN 1 C 2 C η N C 2 2 ω1 Ω 2 ω 22 Ω 2 z1 C zN 2 z2 C zN 2 2 C 2δ 3 η 1 C ηN 1 C 2 η 2 C ηN 2 C 2 ω1 Ω 2 ω2 Ω 2 # 3 z2 C zN 2 Cδ 4 η 2 C ηN 2 C 2 (9.4) ω2 Ω 2 " i z1 C zN 1 3 ηP 2 D i ω 2 η 2 µ 2 (η 2 ηN 2 ) δ 2 η 1 C ηN 1 C 2 2ω 2 ω1 Ω 2 z1 C zN 1 2 z2 C zN 2 η C 2δ 3 η 1 C ηN 1 C 2 C η N C 2 2 ω1 Ω 2 ω 22 Ω 2 z1 C zN 2 z2 C zN 2 2 C 3δ 4 η 1 C ηN 1 C 2 η C η N C 2 2 ω1 Ω 2 ω 22 Ω 2 3 # z2 C zN 2 C4δ 5 η 2 C ηN 2 C 2 (9.5) ω2 Ω 2 In addition to the resonance terms η 1 , η 21 ηN 1 , and η 2 ηN 2 η 1 in (9.4) and the resonance terms η 2 , η 22 ηN 2 , and η 1 ηN 1 η 2 in (9.5), near-resonance terms occur when

ω 2 3ω 1 : Three-to-one internal resonance, ω 2 ω 1 : One-to-one internal resonance, Ω 3ω m : Subharmonic resonance of order one-third, Ω 1/3ω m : Superharmonic resonance of order three, Ω ω m ˙ ω n : Combination resonance.

Next, we present the normal forms of (9.4) and (9.5) for several combinations of the internal and external resonances. 9.1.1 The Case of No Internal Resonances

Substituting the near-identity transformation (8.14) into (9.4) and (9.5) and choosing the h m to eliminate the nonresonance terms, we obtain the following normal forms:

9.1 Nongyroscopic Systems

No External Resonance iσ 1 ξP1 D i ω 1 ξ1 ξ1 µ 1 ξ1 2ω 1 iσ 2 ξP2 D i ω 2 ξ2 ξ2 µ 2 ξ2 2ω 2

i 2ω 1 i 2ω 2

12δ 1 ξ12 ξN1 C 4δ 3 ξ2 ξN2 ξ1

(9.6)

4δ 3 ξ1 ξN1 ξ2 C 12δ 5 ξ22 ξN2

(9.7)

where 24δ 1 z1 zN 1 6δ 2 (z1 zN 2 C z2 zN 1 ) 4δ 3 z2 zN 2 C 2 C 2 (ω 21 Ω 2 )2 (ω 1 Ω 2 )(ω 22 Ω 2 ) (ω 2 Ω 2 )2 24δ 5 z2 zN 2 6δ 4 (z1 zN 2 C z2 zN 1 ) 4δ 3 z1 zN 1 C 2 C 2 σ2 D 2 2 2 2 2 2 (ω 2 Ω ) (ω 1 Ω )(ω 2 Ω ) (ω 1 Ω 2 )2 σ1 D

(9.8) (9.9)

Ω 3ω 2

iσ 1 i ξP1 D i ω 1 ξ1 12δ 1 ξ12 ξN1 C 4δ 3 ξ2 ξN2 ξ1 ξ1 µ 1 ξ1 2ω 1 2ω 1

(9.10)

iσ 2 i ξP2 D i ω 2 ξ2 4δ 3 ξ1 ξN1 ξ2 C 12δ 5 ξ22 ξN2 iΓ1 ξN22 ξ2 µ 2 ξ2 2ω 2 2ω 2 (9.11) where σ 1 and σ 2 are deﬁned in (9.8) and (9.9) and 3δ 4 z1 1 12δ 5 z2 Γ1 D C 2ω 2 ω 21 Ω 2 ω 22 Ω 2

(9.12)

Ω 1/3ω 1

iσ 1 i ξP1 D i ω 1 ξ1 12δ 1 ξ12 ξN1 C 4δ 3 ξ2 ξN2 ξ1 iΓ2 (9.13) ξ1 µ 1 ξ1 2ω 1 2ω 1 iσ 2 i ξP2 D i ω 2 ξ2 4δ 3 ξ1 ξN1 ξ2 C 12δ 5 ξ22 ξN2 (9.14) ξ2 µ 2 ξ2 2ω 2 2ω 2 where σ 1 and σ 2 are deﬁned in (9.8) and (9.9) and 4δ 1 z13 1 3δ 2 z12 z2 Γ2 D C 2 2 2 3 2ω 1 (ω 1 Ω ) (ω 1 Ω 2 )2 (ω 22 Ω 2 ) δ 4 z23 2δ 3 z1 z22 C C 2 (ω 1 Ω 2 )(ω 22 Ω 2 )2 (ω 22 Ω 2 )3

(9.15)

Ω ω 2 C 2ω 1

iσ 1 i ξP1 D i ω 1 ξ1 12δ 1 ξ12 ξN1 C 4δ 3 ξ2 ξN2 ξ1 iΓ3 ξN2 ξN1 ξ1 µ 1 ξ1 2ω 1 2ω 1 (9.16) iσ 2 i 4δ 3 ξ1 ξN1 ξ2 C 12δ 5 ξ22 ξN2 iΓ4 ξN12 ξP2 D i ω 2 ξ2 ξ2 µ 2 ξ2 2ω 2 2ω 2 (9.17)

237

238

9 TDOF Systems with Cubic Nonlinearities

where σ 1 and σ 2 are deﬁned in (9.8) and (9.9) and 3δ 2 z1 1 2δ 3 z2 Γ3 D C ω 1 ω 21 Ω 2 ω 22 Ω 2 3δ 2 z1 1 2δ 3 z2 Γ4 D C 2ω 2 ω 21 Ω 2 ω 22 Ω 2

(9.18) (9.19)

9.1.2 Three-to-One Autoparametric Resonance

The three-to-one internal resonance ω 2 3ω 1 produces the near-resonance term η 2 ηN 21 in (9.4) and the near-resonance term η 31 in (9.5). Hence, substituting the nearidentity transformation (8.14) into (9.4) and (9.5) and choosing the h m to eliminate the nonresonance terms, we obtain the following normal forms: No External Resonance

iσ 1 i ξP1 D i ω 1 ξ1 12δ 1 ξ12 ξN1 C 4δ 3 ξ2 ξN2 ξ1 C 3δ 2 ξ2 ξN12 ξ1 µ 1 ξ1 2ω 1 2ω 1 (9.20) iσ 2 i 4δ 3 ξ1 ξN1 ξ2 C 12δ 5 ξ22 ξN2 C δ 2 ξ13 ξ2 µ 2 ξ2 ξP2 D i ω 2 ξ2 2ω 2 2ω 2 (9.21) where σ 1 and σ 2 are deﬁned in (9.8) and (9.9). Ω 2ω 2 ω 1

iσ 1 ξP1 D i ω 1 ξ1 ξ1 µ 1 ξ1 2ω 1 i 12δ 1 ξ12 ξN1 C 4δ 3 ξ2 ξN2 ξ1 C 3δ 2 ξ2 ξN12 iΓ5 ξ22 2ω 1 iσ 2 ξP2 D i ω 2 ξ2 ξ2 µ 2 ξ2 2ω 2 i 4δ 3 ξ1 ξN1 ξ2 C 12δ 5 ξ22 ξN2 C δ 2 ξ13 iΓ6 ξ1 ξN2 2ω 2 where

2δ 3 zN 1 1 3δ 4 zN 2 C 2ω 1 ω 21 Ω 2 ω 22 Ω 2 2δ 3 z1 1 3δ 4 z2 Γ6 D C ω 2 ω 21 Ω 2 ω 22 Ω 2 Γ5 D

(9.22)

(9.23)

(9.24) (9.25)

The normal forms corresponding to the other external resonances can be obtained by simply adding the appropriate Γm terms to the right-hand sides of (9.20) and (9.21), as in Section 9.1.1.

9.1 Nongyroscopic Systems

9.1.3 One-to-One Internal Resonance

The one-to-one internal resonance corresponding to ω 2 ω 1 produces the nearresonance terms η 22 ηN 2 , η 21 ηN 2 , η 1 ηN 1 η 2 , and η 22 ηN 1 in (9.4) and the near-resonance terms η 21 ηN 1 , η 2 ηN 2 η 1 , η 22 ηN 1 , and η 21 ηN 2 in (9.5). Consequently, substituting the nearidentity transformation (8.14) into (9.4) and (9.5) and choosing the h m to eliminate the nonresonance terms, we obtain the following normal form in the absence of external resonances: iσ 1 iσ 3 i 3δ 2 2ξ1 ξN1 ξ2 C ξ12 ξN2 ξP1 D i ω 1 ξ1 ξ1 ξ2 µ 1 ξ1 2ω 1 2ω 1 2ω 1 (9.26) C12δ 1 ξ12 ξN1 C 2δ 3 2ξ2 ξN2 ξ1 C ξ22 ξN1 C 3δ 4 ξ22 ξN2 iσ 2 iσ 3 i 2δ 3 2ξ1 ξN1 ξ2 C ξ12 ξN2 ξP2 D i ω 2 ξ2 ξ2 ξ1 µ 2 ξ2 2ω 2 2ω 2 2ω 2 (9.27) C3δ 2 ξ12 ξN1 C 3δ 4 2ξ2 ξN2 ξ1 C ξ22 ξN1 C 12δ 5 ξ22 ξN2 where σ3 D

6δ 2 z1 zN 1 4δ 3 (z1 zN 2 C zN 1 z2 ) 6δ 4 z2 zN 2 C 2 C 2 (ω 21 Ω 2 )2 (ω 1 Ω 2 )(ω 22 Ω 2 ) (ω 2 Ω 2 )2

(9.28)

The normal forms of (9.4) and (9.5) in the presence of external resonances can be obtained by simply adding the appropriate Γm terms to the right-hand sides of (9.26) and (9.27), as in Sections 9.1.1 and 9.1.2. However, because the analysis in this section is valid only when Ω is away from ω 1 and ω 2 , one needs to make sure that the external resonance does not produce a primary resonance. 9.1.4 Primary Resonances

Primary resonances can easily be treated by scaling the excitation at O() so that its effect balances the effects of the damping and the nonlinearity. Consequently, substituting (8.60) into (9.3), using (9.2), and choosing the h m to eliminate the nonresonance terms, we obtain the following normal forms: No Internal Resonances, Ω ω n

i ξP1 D i ω 1 ξ1 µ 1 ξ1 2ω 1 i ξP2 D i ω 2 ξ2 µ 2 ξ2 2ω 2

i 12δ 1 ξ12 ξN1 C 4δ 3 ξ2 ξN2 ξ1 z1 δ 1n 2ω 1 i 4δ 3 ξ1 ξN1 ξ2 C 12δ 5 ξ22 ξN2 z2 δ 2n 2ω 2

(9.29) (9.30)

239

240

9 TDOF Systems with Cubic Nonlinearities

Three-to-One Internal Resonance, Ω ω n

i i ξP1 D i ω 1 ξ1 µ 1 ξ1 12δ 1 ξ12 ξN1 C 4δ 3 ξ2 ξN2 ξ1 C 3δ 2 ξ2 ξN12 z1 δ 1n 2ω 1 2ω 1 (9.31) i i 4δ 3 ξ1 ξN1 ξ2 C 12δ 5 ξ22 ξN2 C δ 2 ξ13 ξP2 D i ω 2 ξ2 µ 2 ξ2 z2 δ 2n 2ω 2 2ω 2 (9.32) One-to-One Internal Resonance, Ω ω n

i ξP1 D i ω 1 ξ1 µ 1 ξ1 12δ 1 ξ12 ξN1 C 3δ 2 2ξ1 ξN1 ξ2 C ξ12 ξN2 2ω 1 i z1 δ 1n C2δ 3 2ξ2 ξN2 ξ1 C ξ22 ξN1 C 3δ 4 ξ22 ξN2 2ω 1 i ξP2 D i ω 2 ξ2 µ 2 ξ2 3δ 2 ξ12 ξN1 C 2δ 3 2ξ1 ξN1 ξ2 C ξ12 ξN2 2ω 2 i z2 δ 2n C3δ 4 2ξ2 ξN2 ξ1 C ξ22 ξN1 C 12δ 5 ξ22 ξN2 2ω 2

(9.33)

(9.34)

9.1.5 A Nonsemisimple One-to-One Internal Resonance

In this section, we treat a two-degree-of-freedom system whose linear undamped operator has a generic nonsemisimple structure; that is, we treat the system uR 1 C ω 2 u 1 D 2µ 1 µP 1 α 11 u31 α 12 u21 u 2 α 13 u 1 u22 α 14 u32

(9.35)

uR 2 C ω 2 u 2 D 2µ 2 µP 2 u 1 α 21 u31 α 22 u21 u 2 α 23 u 1 u22 α 24 u32 (9.36) Again, as a ﬁrst step in the application of the method of normal forms, we use the transformation u 1 D ζ1 C ζN 1 ,

uP 1 D i ω ζ1 ζN 1

(9.37)

u 2 D ζ2 C ζN 2 ,

uP 2 D i ω ζ2 ζN 2

(9.38)

to recast (9.35) and (9.36) in the complex-valued form ζP 1 D i ωζ1 µ 1 ζ1 ζN 1 3 2 i h C α 11 ζ1 C ζN 1 C α 12 ζ1 C ζN 1 ζ2 C ζN 2 2ω 2 3 i Cα 13 ζ1 C ζN 1 ζ2 C ζN 2 C α 14 ζ2 C ζN 2

(9.39)

9.1 Nongyroscopic Systems

3 i i h ζP 2 D i ωζ2 µ 2 ζ2 ζN 2 C α 21 ζ1 C ζN 1 ζ1 C ζN 1 C 2ω 2ω 2 2 N N N C α 22 ζ1 C ζ1 ζ2 C ζ2 C α 23 ζ1 C ζ1 ζ2 C ζN 2 i 3 Cα 24 ζ2 C ζN 2

(9.40)

In determining the normal form of (9.39) and (9.40), one can follow one of two approaches. In the ﬁrst approach, which we followed in Sections 7.3.1–7.3.3, we determine the normal form assuming that ζ1 and ζ2 are the same order and then we use the fact that ζ2 is much larger than ζ1 to simplify the resulting forms. In the second approach, we use the fact that ζ2 is much larger than ζ1 to ﬁrst simplify (9.39) and (9.40) and then determine the normal form of the simpliﬁed equations. Following the ﬁrst approach, we note that ζ1 , ζ2 , ζ12 ζN 1 , ζ12 ζN 2 , ζ1 ζN 1 ζ2 , ζ22 ζN 2 , ζ2 ζN 2 ζ1 , and ζ22 ζN 2 are resonance terms. Hence, introducing the near-identity transformation ζ m D η m C h m (η 1 , η 2 , ηN 1 , ηN 2 )

(9.41)

into (9.39) and (9.40) and choosing h 1 and h 2 to eliminate the nonresonance terms, we obtain the normal form i 3α 11 η 21 ηN 1 C 2α 12 η 1 ηN 1 η 2 C α 12 η 21 ηN 2 2ω Cα 13 η 22 ηN 1 C 2α 13 η 1 η 2 ηN 2 C 3α 14 η 22 ηN 2

(9.42)

i i 3α 21 η 21 ηN 1 C 2α 22 η 1 ηN 1 η 2 η1 C 2ω 2ω Cα 22 η 21 ηN 2 C 2α 23 η 2 ηN 2 η 1 C α 23 η 22 ηN 1 C 3α 24 η 22 ηN 2

(9.43)

ηP 1 D i ωη 1 µ 1 η 1 C

ηP 2 D i ωη 2 µ 2 η 2 C

Next, we use the fact that η 2 is much larger than η 1 and introduce new variables deﬁned by η 1 D ξ1

and

η 2 D 1λ 2 ξ2

(9.44)

where ξ1 and ξ2 are O(1) and λ 2 > 0. Moreover, we assume that the damping is weak and hence scale the µ n as µ n D λ 1 µO n

(9.45)

where the µO n are O(1). Substituting (9.44) and (9.45) into (9.42) and (9.43), we have i h 2 ξP1 D i ωξ1 λ 1 µO 1 ξ1 C 3 α 11 ξ12 ξN1 C 2 2λ 2 α 12 ξ1 ξN1 ξ2 2ω C 2λ 2 α 12 ξ12 ξN2 C 22λ 2 α 13 ξ22 ξN1 C 2 22λ 2 α 13 ξ1 ξ2 ξN2 i C3 23λ 2 α 14 ξ 2 ξN2 2

(9.46)

241

242

9 TDOF Systems with Cubic Nonlinearities

i λ2 i h 2Cλ 2 ξP2 D i ωξ2 λ 1 µO 2 ξ2 C α 21 ξ12 ξN1 C 2 2 α 22 ξ1 ξN1 ξ2 ξ1 C 3 2ω 2ω i C 2 α 22 ξ 2 ξN2 C 2 2λ 2 α 23 ξ2 ξN2 ξ1 C 2λ 2 α 23 ξ 2 ξN1 C 3 22λ 2 α 24 ξ 2 ξN2 1

2

2

(9.47) Balancing the damping term and the dominant nonlinear term in (9.46), namely, the term proportional to α 14 , we have λ 1 D 2 3λ 2

(9.48)

Balancing the damping term and the term that causes the instability in (9.47), namely, the term i(2ω)1 λ 2 ξ1 , we have λ1 D λ2

(9.49)

We note that due to (9.48), the dominant nonlinear term in (9.47), namely, the term proportional to α 24 , is small compared with the damping term. Solving (9.48) and (9.49), we obtain λ1 D λ2 D

1 2

(9.50)

and hence (9.46) and (9.47) simplify to 3i 1/2 ξP1 D i ωξ1 1/2 µO 1 ξ1 C α 14 ξ22 ξN2 C 2ω

(9.51)

i 1/2 ξP2 D i ωξ2 1/2 µO 2 ξ2 C ξ1 C 2ω

(9.52)

Equations 9.51 and 9.52 agree with those obtained by Tezak, Nayfeh, and Mook (1982) by using the method of multiple scales. In the second approach, we use the fact that the damping and nonlinearity are weak and that ζ2 is much larger than ζ1 to ﬁrst simplify (9.39) and (9.40) and then determine the normal form of the simpliﬁed equations. If we assume that ζ1 D O(), where is a small nondimensional parameter, then ζ2 D O( 1λ 2 ), where λ 2 > 0. Moreover, for the case of weak damping, µ m D O( λ 1 ), where λ 1 is positive. Using these scalings, we introduce new scaled variables deﬁned by µ m D λ 1 µO n , ζ1 D η 1 , and ζ2 D 1λ 2 η 2 , where the η n and µO n are O(1), in (9.39) and (9.40) and obtain i 2 α 11 (η 1 C ηN 1 )3 ηP 1 D i ωη 1 λ 1 µO 1 (η 1 ηN 1 ) C 2ω i C 23λ 2 α 14 (η 2 C ηN 2 )3 C 2λ 2 α 12 (η 1 C ηN 1 )2 (η 2 C ηN 2 ) C 22λ 2 α 13 (η 1 C ηN 1 ) (η 2 C ηN 2 )2

(9.53)

9.1 Nongyroscopic Systems

i λ2 ηP 2 D i ωη 2 λ 1 µO 2 (η 2 ηN 2 ) C (η 1 C ηN 1 ) 2ω h i C 2Cλ 2 α 21 (η 1 C ηN 1 )3 C 2 α 22 (η 1 C ηN 1 )2 (η 2 C ηN 2 ) 2ω i C 2λ 2 α 23 (η 1 C ηN 1 ) (η 2 C ηN 2 )2 C 22λ 2 α 24 (η 2 C ηN 2 )3

(9.54)

Balancing the damping term and the dominant nonlinear term in (9.53), namely, the term proportional to α 14 , we have λ 1 D 2 3λ 2

(9.55)

Balancing the damping term and the source of the instability in (9.54), namely, the term i(2ω)1 λ 2 (η 1 C ηN 1 ), we have λ1 D λ2

(9.56)

Solving (9.55) and (9.56) yields λ1 D λ2 D

1 2

(9.57)

Hence, (9.53) and (9.54) become ηP 1 D i ωη 1 1/2 µO 1 (η 1 ηN 1 ) C

i 1/2 α 14 (η 2 C ηN 2 )3 C 2ω

(9.58)

ηP 2 D i ωη 2 1/2 µO 2 (η 2 ηN 2 ) C

i 1/2 (η 1 C ηN 1 ) C 2ω

(9.59)

To simplify (9.58) and (9.59), we introduce the near-identity transformation η m D ξm C 1/2 h m ξ1 , ξ2 , ξN1 , ξN2 C

(9.60)

and choose h 1 and h 2 to eliminate the resonance terms. We note that the terms proportional to η 1 and η 22 ηN 2 are resonance terms in (9.58), and hence, we can choose h 1 to eliminate all of the other terms, resulting in the normal form 3i 1/2 ξP1 D i ωξ1 1/2 µO 1 ξ1 C α 14 ξ22 ξN2 C 2ω

(9.61)

Similarly, we note that η 2 and η 1 are resonance terms in (9.59), and hence, we can choose h 2 to eliminate all of the other terms, resulting in the normal form i 1/2 ξP2 D i ωξ2 1/2 µO 2 ξ2 C ξ1 C 2ω

(9.62)

Equations 9.61 and 9.62 agree with (9.51) and (9.52) and with those obtained by Tezak, Nayfeh, and Mook (1982) by using the method of multiple scales.

243

244

9 TDOF Systems with Cubic Nonlinearities

9.1.6 A Parametrically Excited System with a Nonsemisimple Linear Structure

We consider the parametrically excited two-degree-of-freedom system uR 1 C ω 2 u 1 C 2µ 1 uP 1 C α 11 u31 C α 12 u21 u 2 C α 13 u 1 u22 C α 14 u32 C 2 (F11 u 1 C F12 u 2 ) cos Ω t D 0

(9.63)

uR 2 C ω 2 u 2 C 2µ 2 uP 2 C u 1 C α 21 u31 C α 22 u21 u 2 C α 23 u 1 u22 C α 24 u32 C 2 (F21 u 1 C F22 u 2 ) cos Ω t D 0

(9.64)

when the damping and nonlinearities are weak. Again, as a ﬁrst step, we transform (9.63) and (9.64) into two ﬁrst-order complex-valued equations. Using (9.37) and (9.38) and the transformation z D eiΩ t

(9.65)

we rewrite (9.63) and (9.64) as 3 3 i h ζP 1 D i ωζ1 µ 1 ζ1 ζN 1 C α 11 ζ1 C ζN 1 C α 14 ζ2 C ζN 2 2ω 2 2 i Cα 12 ζ1 C ζN 1 ζ2 C ζN 2 C α 13 ζ1 C ζN 1 ζ2 C ζN 2 i F11 ζ1 C ζN 1 C F12 ζ2 C ζN 2 (z C zN ) (9.66) 2ω 2 i h ζP 2 D i ωζ2 µ 2 ζ2 ζN 2 C ζ1 C ζN 1 C α 22 ζ1 C ζN 1 ζ2 C ζN 2 2ω 3 2 3 i Cα 21 ζ1 C ζN 1 C α 23 ζ1 C ζN 1 ζ2 C ζN 2 C α 24 ζ2 C ζN 2 C

i (9.67) F21 ζ1 C ζN 1 C F22 ζ2 C ζN 2 (z C zN ) 2ω Next, we determine the normal forms of (9.66) and (9.67) for weak damping and nonlinearities and two cases of parametric resonance: principal parametric resonance (i.e., Ω 2ω) and fundamental parametric resonance (i.e., Ω ω). C

The Case of Principal Parametric Resonance As a ﬁrst step, we scale the variables. If we assume that ζ1 D O(), where is a small nondimensional parameter, then, due to the nonsemisimple structure of the linear operator, ζ2 D O( 1λ 2 ), where λ 2 > 0. Moreover, because the damping is weak, µ n D O( λ 1 ), where λ 1 > 0. Balancing the damping terms and the dominant nonlinear terms in (9.66) and (9.67), as done in the preceding section, we ﬁnd that λ 1 D λ 2 D 1/2. For the case of principal parametric resonance, Ω 2ω and the terms z ζN 1 and z ζN 2 are near-resonance terms. In order that the damping and nonlinearity balance the dominant parametric resonance terms, F m D O(). Thus, we introduce new variables deﬁned by

ζ1 D η 1 ,

ζ2 D 1/2 η 2 ,

µ n D 1/2 µO n ,

Fm n D f m n

9.1 Nongyroscopic Systems

into (9.66) and (9.67) and obtain i 1/2 α 14 (η 2 C ηN 2 )3 ηP 1 D i ωη 1 1/2 µO 1 (η 1 ηN 1 ) C 2ω N C C f 12 (η 2 C ηN 2 ) (z C z) ηP 2 D i ωη 2 1/2 µO 2 (η 2 ηN 2 ) C

i 1/2 (η 1 C ηN 1 ) C 2ω

(9.68) (9.69)

To simplify (9.68) and (9.69), we use the near-identity transformation (9.60), choose the h m to eliminate the nonresonance terms, and obtain 3i 1/2 i 1/2 ξP1 D i ωξ1 1/2 µO 1 ξ1 C α 14 ξ22 ξN2 C f 12 z ξN2 C 2ω 2ω

(9.70)

i 1/2 ξP2 D i ωξ2 1/2 µO 2 ξ2 C ξ1 C 2ω

(9.71)

Equations 9.70 and 9.71 agree with those obtained by Tezak, Nayfeh, and Mook (1982) by using the method of multiple scales. Moreover, when µO n D 0, (9.70) and (9.71) agree with those obtained by Namachchivaya and Malhotra (1992). In the absence of the parametric resonance, (9.70) and (9.71) reduce to (9.61) and (9.62). Furthermore, in the absence of the nonlinearity, (9.70) and (9.71) reduce to (7.86) and (7.87). Fundamental Parametric Resonance In this case, Ω ω and z zN ζ1 and z zN ζ2 are resonance terms and z 2 ζN 1 and z 2 ζN 2 are near-resonance terms. Hence, we scale F m n to be O( 1/2 ) so that the damping and nonlinear terms balance the parametric resonance terms. Therefore, we introduce new scaled variables deﬁned by

ζ1 D η 1 ,

ζ2 D 1/2 η 2 ,

µ n D 1/2 µO n ,

and

F m n D 1/2 f m n

into (9.66) and (9.67) and obtain i N 1/2 µO 1 (η 1 ηN 1 ) f 12 (η 2 C ηN 2 ) (z C z) 2ω α 14 (η 2 C ηN 2 )3 C f 11 (η 1 C ηN 1 ) (z C zN ) C

ηP 1 D i ωη 1 C C

i 1/2 2ω

(9.72)

ηP 2 D i ωη 2 1/2 µO 2 (η 2 ηN 2 ) C

i 1/2 η 1 C ηN 1 C f 22 (η 2 C ηN 2 ) (z C z) N C 2ω

(9.73)

245

246

9 TDOF Systems with Cubic Nonlinearities

To simplify (9.72) and (9.73), we introduce the transformation η 1 D ξ1 C h 11 ξn , ξN n , z, zN C 1/2 h 12 ξn , ξN n , z, zN C

(9.74)

η 2 D ξ2 C 1/2 h 22 ξn , ξN n , z, zN C

(9.75)

and choose the h m n so that ξP1 D i ωξ1 C 1/2 g 1 ξn , ξN n , z, zN C

(9.76)

ξP2 D i ωξ2 C 1/2 g 2 ξn , ξN n , z, zN C

(9.77)

have the simplest possible form. Substituting (9.74)–(9.77) into (9.72) and (9.73), using (9.65), and equating coefﬁcients of like powers of , we obtain

@h 11 @h 11 @h 11 N @h 11 @h 11 N @h 11 ξ1 C ξ2 h 11 C i Ω ξ1 ξ2 z zN @ξ1 @ξ2 @z @ zN @ ξN1 @ ξN2 i (9.78) f 12 ξ2 C ξN2 (z C zN ) D 2ω @h 12 @h 12 N @h 12 @h 12 N ξ1 C ξ2 h 12 g1 C i ω ξ1 ξ2 @ξ1 @ξ2 @ ξN1 @ ξN2 @h 11 @h 11 @h 11 @h 11 @h 12 @h 12 C iΩ gN 1 g2 gN 2 z zN D g 1 N @z @ zN @ξ1 @ξ2 @ ξ1 @ ξN2

i C f 12 h 22 C hN 22 (z C zN ) µO 1 ξ1 C h 11 ξN1 hN 11 2ω

i 3 i h (9.79) C N α 14 ξ2 C ξN2 C f 11 ξ1 C ξN1 C h 11 C hN 11 (z C z) 2ω @h 22 @h 22 N @h 22 @h 22 N ξ1 C ξ2 h 22 ξ1 ξ2 g2 C i ω @ξ1 @ξ2 @ ξN1 @ ξN2 i h @h 22 @h 22 C iΩ z zN D µO 2 ξ2 ξN2 C ξ1 C ξN1 @z @ zN 2ω i N (9.80) Ch 11 C hN 11 C f 22 ξ2 C ξN2 (z C z)

iω

The right-hand side of (9.78) suggests seeking h 11 in the form h 11 D Γ1 ξ2 z C Γ2 ξ2 zN C Γ3 ξN2 z C Γ4 ξN2 zN

(9.81)

9.1 Nongyroscopic Systems

Substituting (9.81) into (9.78) and equating each of the coefﬁcients of ξ2 z, ξ2 zN , ξN2 z, and ξN2 zN on both sides, we obtain f 12 f 12 , Γ2 D , 2ωΩ 2ωΩ f 12 f 12 Γ3 D , Γ4 D 2ω(Ω 2ω) 2ω(Ω C 2ω)

Γ1 D

Substituting (9.81) into (9.80) and using (9.82) yields @h 22 N @h 22 @h 22 N @h 22 ξ1 C ξ2 h 22 ξ1 ξ2 g2 C i ω @ξ1 @ξ2 @ ξN1 @ ξN2 @h 22 @h 22 C iΩ z zN D µO 2 ξ2 ξN2 @z @ zN i f 12 ξ1 C ξN1 C ξ2 z C ξN2 zN C 2ω Ω (Ω C 2ω) f 12 ξ2 zN C ξN2 z C f 22 ξ2 C ξN2 (z C zN ) C Ω (Ω 2ω)

(9.82)

(9.83)

Equating g 2 to the resonance and near-resonance terms on the right-hand side of (9.83), we have g 2 D µO 2 ξ2 C

i ξ1 2ω

(9.84)

Then, we seek h 22 in the form h 22 D Γ5 ξN2 C Γ6 ξN1 C Γ7 ξ2 z C Γ8 ξ2 zN C Γ9 ξN2 z C Γ10 ξN2 zN

(9.85)

Substituting (9.85) into (9.83), using (9.84), and equating each of the coefﬁcients of ξN1 , ξN2 , ξ2 z, ξ2 zN , ξN2 z, and ξN2 zN on both sides, we obtain i µO 2 1 , Γ6 D 2 , 2ω 4ω f 22 f 12 D C , 2ωΩ 2ωΩ 2 (Ω C 2ω) f 22 f 12 D , 2ωΩ 2ωΩ 2 (Ω 2ω) f 22 f 12 D , C 2ω(Ω 2ω) 2ωΩ (Ω 2ω)2 f 22 f 12 D 2ω(Ω C 2ω) 2ωΩ (Ω C 2ω)2

Γ5 D Γ7 Γ8 Γ9 Γ10

(9.86)

247

248

9 TDOF Systems with Cubic Nonlinearities

Substituting (9.81), (9.82), (9.85), and (9.86) into (9.79) and using (9.84), we obtain h 12 @h 12 N @h 12 @h 12 N ξ1 C ξ2 h 12 ξ1 ξ2 @ξ1 @ξ2 @ ξN1 @ ξN2 @h 12 @h 12 C iΩ zN @z @ zN i i N ξ1 ξ1 C (Γ3 z C Γ4 zN ) µO 2 ξN2 C D (Γ1 z C Γ2 zN ) µO 2 ξ2 2ω 2ω µO 1 f 12 (Ω C ω) ξ2 z ξN2 zN µO 1 ξ1 ξN1 ωΩ (Ω C 2ω) µO 1 f 12 (Ω ω) ξ2 zN ξN2 z C ωΩ (Ω 2ω) f 22 i µO 2 N f 12 i f 12 (z C zN ) ξ2 ξ2 C C 2 C 2ω 2ω Ω (Ω C 2ω) Ω (Ω C 2ω)2 f 22 1 f 12 N N ξ1 C ξ1 C ξ2 z C ξ2 zN C 2 4ω 2 Ω (Ω 2ω) Ω (Ω 2ω)2 i f 11 ξ1 C ξN1 (z C zN ) ξ2 zN C ξN2 z C 2ω f 11 f 12 C N ξ2 z C ξN2 zN (z C z) Ω (Ω C 2ω) 3 f 11 f 12 ξ2 zN C ξN2 z (z C z) N C α 14 ξ2 C ξN2 C Ω (Ω 2ω) (9.87)

g1 C i ω

Because we are stopping at this order, we do not need to explicitly calculate h 12 . Then, equating g 1 to the resonance and near-resonance terms in (9.87), we obtain 3i g 1 D µO 1 ξ1 C α 14 ξ22 ξN2 2ω 2 f 12 1 2 f 12 ( f 11 C f 22 ) i 1 C z zN ξ2 C C 2ω Ω 2 (Ω C 2ω)2 (Ω 2ω)2 Ω 2 4ω 2 2 f 12 ( f 11 C f 22 ) i f 12 z 2 ξN2 C C 2 (9.88) 2ω Ω (Ω 2ω) Ω (Ω 2ω)2 Finally, substituting (9.81) and (9.82) into (9.74) yields " # ξN2 z ξN2 zN f 12 ξ2 z ξ2 zN η 1 D ξ1 C C C 2ω Ω Ω Ω 2ω Ω C 2ω

(9.89)

9.2 Gyroscopic Systems

Then, substituting (9.88) and (9.84) into (9.76) and (9.77), we obtain the normal form 1/2 2 f 12 ( f 11 C f 22 ) f 12 Pξ1 D i ωξ1 1/2 µO 1 ξ1 C i z 2 ξN2 C 2 2ω Ω (Ω 2ω) Ω (Ω 2ω)2 2 2 f 12 f 12 2 f 12 ( f 11 C f 22 ) ξ2 C C C Ω 2 4ω 2 Ω 2 (Ω C 2ω)2 Ω 2 (Ω 2ω)2 C3α 14 ξ22 ξN2 C (9.90) i 1/2 ξP2 D i ωξ2 1/2 µO 2 ξ2 C ξ1 C 2ω

(9.91)

because z zN D 1. In the absence of the parametric excitation (i.e., f m n D 0), (9.90) and (9.91) reduce to (9.61) and (9.62). In the absence of the nonlinearity (i.e., α 14 D 0), (9.90) and (9.91) reduce to (7.134) and (7.135) when f 13 D 0 and f 32 D 0.

9.2 Gyroscopic Systems

Again, for simplicity, we consider a two-degree-of-freedom system to illustrate the method. Speciﬁcally, we consider uR 1 C λ 1 uP 2 C α 1 u 1 D 4δ 1 u31 C 3δ 2 u21 u 2 C 2δ 3 u 1 u22 C δ 4 u32 C 2 f 1 cos (Ω t C τ 1 ) uR 2 λ 2 uP 1 C α 2 u 2 D δ 2 u31 C 2δ 3 u21 u 2 C 3δ 4 u 1 u22 C 4δ 5 u32

(9.92)

C 2 f 2 cos (Ω t C τ 2 )

(9.93)

Using the transformation (7.144)–(7.147) and following steps similar to those used in Section 7.4, we cast (9.92) and (9.93) in the complex-valued forms λ1 i ω 1 (α 1 ω 22 ) (z1 C zN 1 ) (z2 C zN 2 ) ζP 1 D i ω 1 ζ1 C 2 2 2 2α 1 (ω 2 ω 1 ) 2(ω 2 ω 21 ) 3 2 C χ 11 ζ1 C ζN 1 C ζ2 C ζN 2 C χ 12 ζ1 C ζN 1 C ζ2 C ζN 2 i(α 1 ω 22 ) i(α 1 ω 21 ) ζ1 ζN 1 C ζ2 ζN 2 λ1 ω1 λ1 ω2 2 i(α 1 ω 22 ) 3 i(α 1 ω 1 ) C χ 14 ζ1 ζN 1 C ζ2 ζN 2 λ1 ω1 λ1 ω2 i(α 1 ω 21 ) ζ1 ζN 1 C χ 13 ζ1 C ζN 1 C ζ2 C ζN 2 λ1 ω1 2 2 i(α 1 ω 2 ) ζ2 ζN 2 C λ1 ω2

(9.94)

249

250

9 TDOF Systems with Cubic Nonlinearities

λ1 i ω 2 (α 1 ω 21 ) (z1 C zN 1 ) C (z2 C zN 2 ) ζP 2 D i ω 2 ζ2 2α 1 (ω 22 ω 21 ) 2(ω 22 ω 21 ) 3 2 C χ 21 ζ1 C ζN 1 C ζ2 C ζN 2 C χ 22 ζ1 C ζN 1 C ζ2 C ζN 2 i(α 1 ω 22 ) i(α 1 ω 21 ) ζ1 ζN 1 C ζ2 ζN 2 λ1 ω1 λ1 ω2 i(α 1 ω 22 ) 3 i(α 1 ω 21 ) C χ 24 ζ1 ζN 1 C ζ2 ζN 2 λ1 ω1 λ1 ω2 i(α 1 ω 21 ) ζ1 ζN 1 C χ 23 ζ1 C ζN 1 C ζ2 C ζN 2 λ1 ω1 2 i(α 1 ω 22 ) ζ2 ζN 2 C λ1 ω2

(9.95)

where χ 11 D

4 i δ 1 ω 1 (α 1 ω 22 ) α 1 δ 2 λ 1 3i δ 2 ω 1 (α 1 ω 22 ) 2α 1 δ 3 λ 1 , χ 12 D , 2 2 2α 1 (ω 2 ω 1 ) 2α 1 (ω 22 ω 21 )

χ 13 D

2i δ 3 ω 1 (α 1 ω 22 ) 3α 1 δ 4 λ 1 i δ 4 ω 1 (α 1 ω 22 ) 4α 1 δ 5 λ 1 , χ 14 D , 2 2 2α 1 (ω 2 ω 1 ) 2α 1 (ω 22 ω 21 )

χ 21 D

α 1 δ 2 λ 2 4 i δ 1 ω 2 (α 1 ω 21 ) 2α 1 δ 3 λ 1 3i δ 2 ω 2 (α 1 ω 21 ) , χ 22 D , 2 2 2α 1 (ω 2 ω 1 ) 2α 1 (ω 22 ω 21 )

χ 23 D

3α 1 δ 4 λ 1 2i δ 3 ω 2 (α 1 ω 21 ) 4α 1 δ 5 λ 1 i δ 4 ω 2 (α 1 ω 21 ) , χ 24 D 2 2 2α 1 (ω 2 ω 1 ) 2α 1 (ω 22 ω 21 ) (9.96)

There are two cases to consider: Ω ω m (corresponding to the primary resonance of the mth mode) and Ω is away from ω 1 and ω 2 . The case of primary resonance is discussed next and the other case (secondary resonances) is discussed in Section 9.2.2. 9.2.1 Primary Resonances

As in Section 8.2.1, we treat this case by scaling z1 and z2 at O(). Then, substituting the near-identity transformation (8.60) into (9.94) and (9.95) and choosing the h m to eliminate the nonresonance terms, we obtain the following normal forms: No Internal Resonance, Ω ω n

ξP1 D i ω 1 ξ1 C S11 ξ12 ξN1 C S12 ξ2 ξN2 ξ1 C Γ1 δ 1n

(9.97)

ξP2 D i ω 2 ξ2 C S21 ξ1 ξN1 ξ2 C S22 ξ22 ξN2 C Γ2 δ 2n

(9.98)

9.2 Gyroscopic Systems

where S11 D 3χ 11 C S12 D 6χ 11 C

S21 D 6χ 21 C

(9.99)

2i χ 12 (α 1 ω 21 ) 2χ 13 (α 1 ω 22 )2 6i χ 14 (α 1 ω 21 )(α 1 ω 22 )2 C C λ1 ω1 λ 21 ω 22 λ 31 ω 1 ω 22 (9.100) 2i χ 22 (α 1 ω 22 ) 2χ 23 (α 1 ω 21 )2 6i χ 24 (α 1 ω 21 )2 (α 1 ω 22 ) C C λ1 ω2 λ 21 ω 21 λ 31 ω 21 ω 2 (9.101)

S22 D 3χ 21 C Γ1 D

i χ 12 (α 1 ω 21 ) χ 13 (α 1 ω 21 )2 3i χ 14 (α 1 ω 21 )3 C C λ1 ω1 λ 21 ω 21 λ 31 ω 31

i χ 22 (α 1 ω 22 ) χ 23 (α 1 ω 22 )2 3i χ 24 (α 1 ω 22 )3 C C 2 2 λ1 ω2 λ1 ω2 λ 31 ω 32

i ω 1 (α 2 ω 21 )z1 λ 1 α 1 z2 , 2α 1 (ω 22 ω 21 )

Γ2 D

(9.102)

i ω 2 (α 1 ω 21 )z1 λ 1 α 1 z2 2α 1 (ω 22 ω 21 ) (9.103)

Three-to-One Internal Resonance, Ω ω n

ξP1 D i ω 1 ξ1 C S11 ξ12 ξN1 C S12 ξ2 ξN2 ξ1 C S13 ξ2 ξN12 C Γ1 δ 1n

(9.104)

ξP2 D i ω 2 ξ2 C S21 ξ1 ξN1 ξ2 C S22 ξ22 ξN2 C S23 ξ13 C Γ2 δ 2n

(9.105)

where S11 , S12 , S21 , S22 , Γ1 , and Γ2 are deﬁned in (9.99)–(9.103) and α 1 ω 22 3i χ 14 (α 1 ω 21 )2 (α 1 ω 22 ) α 1 ω 21 2ω 2 ω1 λ 31 ω 21 ω 2 2χ 13 (α 1 ω 21 ) α 1 ω 22 α 1 ω 21 C (9.106) 2 λ ω1 ω2 2ω 1

S13 D 3χ 11 C

S23 D χ 21 C

2i χ 12 λ1

i χ 22 (α 1 ω 21 ) χ 23 (α 1 ω 21 )2 i χ 24 (α 1 ω 21 )3 2 2 λ1 ω1 λ1 ω1 λ 31 ω 31

(9.107)

We note that the transformation (7.144)–(7.147) is not valid when ω 2 ω 1 and hence (9.94) and (9.95) do not apply in this case. Consequently, the case of a oneto-one internal resonance needs special treatment. 9.2.2 Secondary Resonances in the Absence of Internal Resonances

In this case, we scale z1 and z2 at O(1). Then, we introduce the transformation (8.81) and (8.82) into (9.94) and (9.95) to eliminate the terms involving z n and zN n

251

252

9 TDOF Systems with Cubic Nonlinearities

at O(1) and obtain ηP 1 D i ω 1 η 1 C χ 11 [η 1 C ηN 1 C η 2 C ηN 2 C b 1 (z1 C zN 1 ) C b 2 (z2 zN 2 )]3 C χ 12 [η 1 C ηN 1 C η 2 C ηN 2 C b 1 (z1 C zN 1 ) C b 2 (z2 zN 2 )]2 i(α 1 ω 22 ) i(α 1 ω 21 ) (η 1 ηN 1 ) C (η 2 ηN 2 ) λ1 ω1 λ1 ω2 Cb 3 (z1 zN 1 ) C b 4 (z2 C zN 2 ) C χ 13 [η 1 C ηN 1 C η 2 C ηN 2 C b 1 (z1 C zN 1 ) C b 2 (z2 zN 2 )] i(α 1 ω 22 ) i(α 1 ω 21 ) (η 1 ηN 1 ) C (η 2 ηN 2 ) λ1 ω1 λ1 ω2 2 Cb 3 (z1 zN 1 ) C b 4 (z2 C zN 2 )

i(α 1 ω 21 ) i(α 1 ω 22 ) (η 1 ηN 1 ) C (η 2 ηN 2 ) λ1 ω1 λ1 ω2 3 Cb 3 (z1 zN 1 ) C b 4 (z2 C zN 2 )

C χ 14

(9.108)

ηP 2 D i ω 2 η 2 C χ 21 [η 1 C ηN 1 C η 2 C ηN 2 C b 1 (z1 C zN 1 ) C b 2 (z2 zN 2 )]3 C χ 22 [η 1 C ηN 1 C η 2 C ηN 2 C b 1 (z1 C zN 1 ) C b 2 (z2 zN 2 )]2 i(α 1 ω 22 ) i(α 1 ω 21 ) (η 1 ηN 1 ) C (η 2 ηN 2 ) λ1 ω1 λ1 ω2 Cb 3 (z1 zN 1 ) C b 4 (z2 C zN 2 ) C χ 23 [η 1 C ηN 1 C η 2 C ηN 2 C b 1 (z1 C zN 1 ) C b 2 (z2 zN 2 )] i(α 1 ω 22 ) i(α 1 ω 21 ) (η 1 ηN 1 ) C (η 2 ηN 2 ) λ1 ω1 λ1 ω2 2 Cb 3 (z1 zN 1 ) C b 4 (z2 C zN 2 )

i(α 1 ω 21 ) i(α 1 ω 22 ) (η 1 ηN 1 ) C (η 2 ηN 2 ) λ1 ω1 λ1 ω2 3 Cb 3 (z1 zN 1 ) C b 4 (z2 C zN 2 )

C χ 24

(9.109)

where the b i are deﬁned in (8.85). It follows from (9.108) and (9.109) that, to O(), only a three-to-one internal resonance is possible because (8.81) and (8.82) are not valid when ω 2 ω 1 (i.e., when there is a one-to-one internal resonance). Moreover, the excitation can produce resonances corresponding to

9.2 Gyroscopic Systems

Ω 3ω m : Subharmonic resonance of order one-third, Ω 1/3ω m : Superharmonic resonance of order three, Ω 2ω n ˙ ω m : Combination resonance. We present the normal forms of (9.108) and (9.109) for several of these external resonances in the absence of internal resonances in this section and in the presence of a three-to-one internal resonance in Section 9.2.3. Substituting the near-identity transformation (8.14) into (9.108) and (9.109) and choosing the h m to eliminate the nonresonance terms, we obtain the following normal forms: No External Resonance

ξP1 D i ω 1 ξ1 C σ 1 ξ1 C S11 ξ12 ξN1 C S12 ξ2 ξN2 ξ1

(9.110)

ξP2 D i ω 2 ξ2 C σ 2 ξ2 C S21 ξ1 ξN1 ξ2 C S22 ξ22 ξN2

(9.111)

where S11 , S12 , S21 , and S22 are deﬁned in (9.99)–(9.102), and σ 1 D 6χ 11 b 21 z1 zN 1 b 22 z2 zN 2 C b 1 b 2 (z2 zN 1 z1 zN 2 ) i(α 1 ω 21 ) 2 b 1 z1 zN 1 b 22 z2 zN 2 C b 1 b 2 (z2 zN 1 z1 zN 2 ) C 2χ 12 λ1 ω1 ) (z ) (b C 1 b 4 b 2 b 3 1 zN 2 C zN 1 z2 C 2χ 13 b 24 z2 zN 2 b 23 z1 zN 1 C b 3 b 4 (z1 zN 2 zN 1 z2 ) i(α 1 ω 21 ) (b 1 b 4 b 2 b 3 ) (z1 zN 2 C zN 1 z2 ) λ1 ω1 6i χ 14 (α 1 ω 21 ) 2 b 4 z2 zN 2 b 23 z1 zN 1 C b 3 b 4 (z1 zN 2 zN 1 z2 ) (9.112) C λ1 ω1 σ 2 D 6χ 21 b 21 z1 zN 1 b 22 z2 zN 2 C b 1 b 2 (z2 zN 1 z1 zN 2 ) i(α 1 ω 21 ) 2 b 1 z1 zN 1 b 22 z2 zN 2 C b 1 b 2 (z2 zN 1 z1 zN 2 ) C 2χ 32 λ1 ω1 C (b 1 b 4 b 2 b 3 ) (z1 zN 2 C zN 1 z2 ) C 2χ 23 b 24 z2 zN 2 b 23 z1 zN 1 C b 3 b 4 (z1 zN 2 zN 1 z2 ) C

i(α 1 ω 21 ) (b 1 b 4 b 2 b 4 ) (z1 zN 2 C zN 1 z2 ) λ1 ω2 6i χ 24 (α 1 ω 21 ) 2 b 4 z2 zN 2 b 23 z1 zN 1 C b 3 b 4 (z1 zN 2 zN 1 z2 ) C λ1 ω2 C

(9.113)

253

254

9 TDOF Systems with Cubic Nonlinearities

Ω 3ω 2

ξP1 D i ω 1 ξ1 C σ 1 ξ1 C S11 ξ12 ξN1 C S12 ξ2 ξN2 ξ1

(9.114)

ξP2 D i ω 2 ξ2 C σ 2 ξ2 C S21 ξ1 ξN1 ξ2 C S22 ξ22 ξN2 C Γ1 ξN22

(9.115)

where σ 1 and σ 2 are deﬁned in (9.112) and (9.113), the S m n are deﬁned in (9.99)– (9.102), and 2i χ 22 (α 1 ω 22 ) χ 23 (α 1 ω 22 )2 Γ1 D (b 1 z1 C b 2 z2 ) 3χ 21 λ1 ω2 λ 21 ω 22 2i χ 23 (α 1 ω 22 ) 3χ 24 (α 1 ω 22 )2 C (b 3 z1 C b 4 z4 ) χ 22 λ1 ω2 λ 21 ω 22

(9.116)

Ω 1/3ω 1

ξP1 D i ω 1 ξ1 C σ 1 ξ1 C S11 ξ12 ξN1 C S12 ξ2 ξN2 ξ1 C Γ2

(9.117)

ξP2 D i ω 2 ξ2 C σ 2 ξ2 C S21 ξ1 ξN1 ξ2 C S22 ξ22 ξN2

(9.118)

where σ 1 and σ 2 are deﬁned in (9.112) and (9.113), the S m n are deﬁned in (9.99)– (9.102), and Γ2 D χ 11 (b 1 z1 C b 2 z2 )3 C χ 12 (b 1 z1 C b 2 z2 )2 (b 3 z1 C b 4 z2 ) C χ 13 (b 1 z1 C b 2 z2 ) (b 3 z1 C b 4 z2 )2 C χ 14 (b 3 z1 C b 4 z2 )3

(9.119)

Ω ω 2 C 2ω 1

ξP1 D i ω 1 ξ1 C σ 1 ξ1 C S11 ξ12 ξN1 C S12 ξ2 ξN2 ξ1 C Γ3 ξN2 ξN1

(9.120)

ξP2 D i ω 2 ξ2 C σ 2 ξ2 C S21 ξ1 ξN1 ξ2 C S22 ξ22 ξN2 C Γ4 ξN12

(9.121)

where σ 1 and σ 2 are deﬁned in (9.112) and (9.113), the S m n are deﬁned in (9.99)– (9.102), and 6α 1 λ 2 (b 3 z1 C b 4 z2 ) Γ3 D 6χ 11 (b 1 z1 C b 2 z2 ) C λ1 ω1 ω2 α 1 ω 21 α 1 ω 22 ) (b C C 2χ 12 b 3 z1 C b 4 z2 i 1 z1 C b 2 z2 λ1 ω1 λ1 ω2 2 2 α1 ω1 α1 ω2 2χ 13 i (b 3 z1 C b 4 z2 ) C λ1 ω1 λ1 ω2 α1 λ2 (b 1 z1 C b 2 z2 ) (9.122) λ1 ω1 ω2

9.2 Gyroscopic Systems

(α 1 ω 21 )2 (b 3 z1 C b 4 z2 ) λ 21 ω 21 2i(α 1 ω 21 ) (b 1 z1 C b 2 z2 ) C χ 22 b 3 z1 C b 4 z2 λ1 ω1 (α 1 ω 21 )2 2i(α 1 ω 21 ) (b ) (b ) C z C b z z C b z χ 23 1 1 2 2 3 1 4 2 λ1 ω1 λ 21 ω 21 (9.123)

Γ4 D 3χ 21 (b 1 z1 C b 2 z2 ) 3χ 24

9.2.3 Three-to-One Internal Resonance

Substituting the near-identity transformation (8.14) into (9.108) and (9.109) and choosing the h m to eliminate the nonresonance terms, we obtain the following normal form in the absence of external resonances: ξP1 D i ω 1 ξ1 C σ 1 ξ1 C S11 ξ12 ξN1 C S12 ξ2 ξN2 ξ1 C S13 ξ2 ξN12

(9.124)

ξP2 D i ω 2 ξ2 C σ 2 ξ2 C S21 ξ1 ξN1 ξ2 C S22 ξ22 ξN2 C S23 ξ13

(9.125)

where σ 1 and σ 2 are deﬁned in (9.112) and (9.113) and the S m n are deﬁned in (9.99)–(9.102), (9.106), and (9.107). The normal forms of (9.108) and (9.109) in the presence of external resonances can be obtained by simply adding the appropriate Γm terms to (9.124) and (9.125), as in Section 9.2.2.

255

257

10 Systems with Quadratic and Cubic Nonlinearities We consider the response of two-degree-of-freedom damped systems with quadratic and cubic nonlinearities to simultaneous principal and combination parametric resonances in the presence and absence of internal resonances. The case of no internal resonance is treated in Section 10.2, the case of three-to-one internal resonance is treated in Section 10.3, and the case of one-to-one internal resonance is treated in Section 10.4. The case of two-to-one internal resonance is treated in Section 10.5 using the method of normal forms, in Section 10.6 using the method of multiple scales, and in Section 10.7 using the generalized method of averaging. Finally, the case of a nonsemisimple one-to-one internal resonance is treated in Section 10.8.

10.1 Introduction

In this chapter, we consider the response of two-degree-of-freedom damped systems with quadratic and cubic nonlinearities to multifrequency parametric excitations. Speciﬁcally, we consider systems modeled by uR 1 C ω 21 u 1 C 2µ 1 uP 1 C C 2u 2

m X

M X @V (u 1 , u 2 ) C 2u 1 f 1m cos (Ωm t C τ 1m ) @u 1 mD1

f 2m cos (Ωm t C τ 2m ) D 0

mD1

uR 2 C ω 22 u 2 C 2µ 2 uP 2 C C 2u 2

m X

(10.1)

M X @V (u 1 , u 2 ) C 2u 1 g 1m cos (Ωm t C ν 1m ) @u 2 mD1

g 2m cos (Ωm t C ν 2m ) D 0

mD1

(10.2)

V D 13 δ 1 u31 C δ 2 u21 u 2 C δ 3 u 1 u22 C 13 δ 4 u32 C 14 α 1 u41 C α 2 u31 u 2 C 12 α 3 u21 u22 C α 4 u 1 u32 C 14 α 5 u42

(10.3)

The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

258

10 Systems with Quadratic and Cubic Nonlinearities

Clearly, the undamped and unforced system is derivable from the Lagrangian LD

1 2

uP 21 C uP 22 C ω 21 u21 C ω 22 u22 V(u 1 , u 2 )

Thus the undamped unforced normal form should be derivable from a Lagrangian, which can be used to check the accuracy of the calculated normal forms. Again, as a ﬁrst step, it is convenient to cast (10.1) and (10.2) in complex-valued form using the transformation u m D ζ m C ζN m , uP m D i ω m ζ m ζN m

(10.4)

z m D e i Ωm t , zP m D i Ωm z m

(10.5)

To keep track of the different orders of magnitude, we introduce a small nondimensional parameter . Then, if u m D O(), u2m D O( 2 ) and u3m D O( 3 ), and we order the damping and excitation terms such that their effects balance each other as well as the effect of the nonlinearity. To reduce the amount of algebra, we assume that the µ m , f m n , and g m n are O( 2 ). Using this ordering and the transformation (10.4) and (10.5), we rewrite (10.1)–(10.3) as i ζP 1 D i ω 1 ζ1 2 µ 1 ζ1 ζN 1 C δ 1 (ζ1 C ζN 1 )2 2ω 1 i 2 α 1 (ζ1 C ζN 1 )3 C2δ 2 (ζ1 C ζN 1 )(ζ2 C ζN 2 ) C δ 3 (ζ2 C ζN 2 )2 C 2ω 1 C3α 2 (ζ1 C ζN 1 )2 (ζ2 C ζN 2 ) C α 3 (ζ1 C ζN 1 )(ζ2 C ζN 2 )2 C α 4 (ζ2 C ζN 2 )3 C

M X i 2 (ζ1 C ζN 1 ) f 1m z m e i τ 1m C zN m e i τ 1m 2ω 1 mD1

C

M X i 2 (ζ2 C ζN 2 ) f 2m z m e i τ 2m C zN m e i τ 2m 2ω 1 mD1

(10.6)

i ζP 2 D i ω 2 ζ2 2 µ 2 ζ2 ζN 2 C δ 2 (ζ1 C ζN 1 )2 2ω 2 2 N 2 C i α 2 (ζ1 C ζN 1 )3 C2δ 3 (ζ1 C ζN 1 )(ζ2 C ζN 2 ) C δ 4 (ζ2 C ζ) 2ω 2 Cα 3 (ζ1 C ζN 1 )2 (ζ2 C ζN 2 ) C 3α 4 (ζ1 C ζN 1 )(ζ2 C ζN 2 )2 C α 5 (ζ2 C ζN 2 )3 C

M X i 2 (ζ1 C ζN 1 ) g 1m z m e i ν 1m C zN m e i ν 1m 2ω 2 mD1

C

M X i 2 (ζ2 C ζN 2 ) g 2m z m e i ν 2m C zN m e i ν 2m 2ω 1 mD1

(10.7)

10.1 Introduction

Next, we introduce the near-identity transformation ζ m D η m C h m1(η n , ηN n ) C 2 h m2(η n , ηN n , z n , zN n ) C

(10.8)

into (10.6) and (10.7) and choose the h m n to eliminate the nonresonance terms so that the resulting equations have the simplest possible form ηP m D i ω m η m C F m1(η n , ηN n ) C 2 F m2 (η n , ηN n , z n , zN n ) C

(10.9)

where the F m n are chosen to eliminate the resonance and near-resonance terms. Substituting (10.8) and (10.9) into (10.6) and (10.7), using (10.5), and equating coefﬁcients of like powers of , we obtain the following:

Order ()

F11 C L(h 11 ) i ω 1 h 11 D

F21 C L(h 21 ) i ω 2 h 21 D

i δ 1 (η 1 C ηN 1 )2 C 2δ 2 (η 1 C ηN 1 )(η 2 C ηN 2 ) 2ω 1 (10.10) Cδ 3 (η 2 C ηN 2 )2 i δ 2 (η 1 C ηN 1 )2 C 2δ 3 (η 1 C ηN 1 )(η 2 C ηN 2 ) 2ω 2 (10.11) Cδ 4 (η 2 C ηN 2 )2

Order (2 )

F12 C L(h 12 ) i ω 1 h 12 D

@h 11 @h 11 N @h 11 @h 11 N F11 F21 F11 F21 @η 1 @ ηN 1 @η 2 @ ηN 2

i h 2δ 1 (η 1 C ηN 1 )(h 11 C hN 11 ) C 2δ 2 (η 1 C ηN 1 )(h 21 C hN 21 ) 2ω 1 i C2δ 2 (h 11 C hN 11 )(η 2 C ηN 2 ) C 2δ 3 (η 2 C ηN 2 )(h 21 C hN 21 )

C

i α 1 (η 1 C ηN 1 )3 C 3α 2 (η 1 C ηN 1 )2 (η 2 C ηN 2 ) 2ω 1 Cα 3 (η 1 C ηN 1 )(η 2 C ηN 2 )2 C α 4 (η 2 C ηN 2 )3 µ 1 (η 1 ηN 1 )

C

C

M X i (η 1 C ηN 1 ) f 1m z m e i τ 1m C zN m e i τ 1m 2ω 1 mD1

C

M X i (η 2 C ηN 2 ) f 2m z m e i τ 2m C zN m e i τ 2m 2ω 1 mD1

(10.12)

259

260

10 Systems with Quadratic and Cubic Nonlinearities

F22 C L(h 22 ) i ω 2 h 22 D

@h 21 @h 21 N @h 21 @h 21 N F11 F21 F11 F21 @η 1 @ ηN 1 @η 2 @ ηN 2

i h 2δ 2 (η 1 C ηN 1 )(h 11 C hN 11 ) C 2δ 3 (η 1 C ηN 1 )(h 21 C hN 21 ) 2ω 2 i C2δ 3 (η 2 C ηN 2 )(h 11 C hN 11 ) C 2δ 4 (η 2 C ηN 2 )(h 21 C hN 21 )

C

i α 2 (η 1 C ηN 1 )3 C α 3 (η 1 C ηN 1 )2 (η 2 C ηN 2 ) 2ω 2 C3α 4 (η 1 C ηN 1 )(η 2 C ηN 2 )2 C α 5 (η 2 C ηN 2 )3

C

µ 2 (η 2 ηN 2 ) C C where

M X i (η 1 C ηN 1 ) g 1m z m e i ν 1m C zN m e i ν 1m 2ω 2 mD1

M X i (η 2 C ηN 2 ) g 2m z m e i ν 2m C zN m e i ν 2m 2ω 2 mD1

(10.13)

@h @h @h @h η1 ηN 1 C i ω 2 η2 ηN 2 @η 1 @ ηN 1 @η 2 @ ηN 2 M X @h @h Ωm zm zN m Ci @z @ zN m m mD1

L(h) D i ω 1

Next, we choose h 11 and h 21 to eliminate the nonresonance terms and F11 and F21 to eliminate the resonance and near-resonance terms. We start by assuming that there are no resonance and near-resonance terms and hence set F11 and F21 equal to zero. Then, we determine the conditions under which this assumption is violated. The forms of the terms in (10.10) and (10.11) suggest seeking h 11 and h 21 in the forms h 11 D Γ1 η 21 C Γ2 η 1 ηN 1 C Γ3 ηN 21 C Γ4 η 22 C Γ5 η 2 ηN 2 C Γ6 ηN 22 C Γ7 η 1 η 2 C Γ8 η 1 ηN 2 C Γ9 ηN 1 η 2 C Γ10 ηN 1 ηN 2

(10.14)

h 21 D Λ 1 η 21 C Λ 2 η 1 ηN 1 C Λ 3 ηN 21 C Λ 4 η 22 C Λ 5 η 2 ηN 2 C Λ 6 ηN 22 C Λ 7 η 1 η 2 (10.15) C Λ 8 η 1 ηN 2 C Λ 9 ηN 1 η 2 C Λ 10 ηN 1 ηN 2 Substituting (10.14) and (10.5) into (10.10) and (10.11), putting F11 and F21 equal to zero, and choosing the Γm and Λ m to eliminate all of the terms, we obtain δ1 δ1 δ1 δ3 , Γ2 D 2 , Γ3 D 2 , Γ4 D , 2 2ω 1 (2ω 2 ω 1 ) 2ω 1 ω1 6ω 1 δ3 δ3 δ2 δ2 , Γ8 D , , Γ7 D Γ5 D 2 , Γ6 D 2ω 1 (2ω 2 C ω 1 ) ω1 ω2 ω1 ω2 ω1 δ2 δ2 Γ9 D , Γ10 D (10.16) ω 1 (ω 2 2ω 1 ) ω 1 (ω 2 C 2ω 1 ) Γ1 D

10.1 Introduction

δ2 δ2 δ2 , Λ2 D 2 , Λ3 D , 2ω 2 (2ω 1 ω 2 ) 2ω 2 (ω 2 C 2ω 1 ) ω2 δ4 δ4 δ4 δ3 , Λ5 D 2 , Λ6 D 2 , Λ7 D , Λ4 D ω1 ω2 2ω 22 ω2 6ω 2 δ3 δ3 δ3 , Λ 10 D Λ8 D , Λ9 D ω 2 (ω 1 2ω 2 ) ω1 ω2 ω 2 (2ω 2 C ω 1 ) Λ1 D

(10.17) It follows from (10.16) and (10.17) that the transformation (10.8) breaks down when ω 2 2ω 1 or ω 1 2ω 2 and hence (10.10) and (10.11) contain near-resonance terms in these cases. These cases correspond to two-to-one internal resonances, which need to be treated separately. This is done in Section 10.5. It follows from (10.14)–(10.17) that 2δ 1 2 δ1 2 δ3 η 1 C ηN 21 2 η 1 ηN 1 C η 2 C ηN 22 h 11 C hN 11 D 2 2 2 3ω 1 ω1 4ω 2 ω 1 2δ 3 2δ 2 (η 1 η 2 C ηN 1 ηN 2 ) 2 η 2 ηN 2 C ω 2 (ω 2 C 2ω 1 ) ω1 2δ 2 (η 2 ηN 1 C η 1 ηN 2 ) C ω 2 (ω 2 2ω 1 ) 2 2δ 2 δ2 δ4 2 η C ηN 21 2 η 1 ηN 1 C η C ηN 22 h 21 C hN 21 D 4ω 21 ω 22 1 ω2 3ω 22 2 2δ 4 2δ 3 (η 2 η 1 C ηN 1 ηN 2 ) 2 η 2 ηN 2 C ω 1 (ω 1 C 2ω 2 ) ω2 2δ 3 (η 2 ηN 1 C η 1 ηN 2 ) C ω 1 (ω 1 2ω 2 )

(10.18)

(10.19)

Next, we substitute (10.18) and (10.19) into (10.12) and (10.13) and choose h 12 and h 22 to eliminate all of the nonresonance terms, thereby leaving F12 and F22 with all of the resonance and near-resonance terms. If we stop at O( 2 ), then we do not need to determine h 12 and h 22 , and hence all that we need to do is to determine F12 and F22 . Inspecting the right-hand sides of (10.12) and (10.13), we ﬁnd that the following resonances are possible:

ω 2 ω 1 : One-to-one internal resonance, ω 2 3ω 1 : Three-to-one internal resonance, Ωn 2ω n : Principal parametric resonance, Ωm ω 1 ˙ ω 1 : Combination parametric resonance.

In what follows, we consider the different internal resonance cases in conjunction with the parametric resonances: Ωm ω 2 ω 1 ,

Ωn ω 2 C ω 1 ,

Ω p 2ω 1 ,

and

Ωq 2ω 2

261

262

10 Systems with Quadratic and Cubic Nonlinearities

10.2 The Case of No Internal Resonance

Choosing F12 and F22 to eliminate all of the resonance and near-resonance terms in (10.12) and (10.13), we obtain F12 D µ 1 η 1 C

4i S11 η 21 ηN 1 C S12 η 2 ηN 2 η 1 ω1

i f 1p ηN 1 z p e i τ 1p C f 2m η 2 zN m e i τ 2m C f 2n ηN 2 z n e i τ 2n 2ω 1 4i S12 η 1 ηN 1 η 2 C S22 η 22 ηN 2 D µ 2 η 2 C ω2 i g 1m η 1 z m e i ν 1m C g 1n ηN 1 z n e i ν 1n C g 2q ηN 2 z q e i ν 2q C 2ω 2 C

F22

(10.20)

(10.21)

where 10δ 21 4δ 22 2δ 22 2 C 2 3ω 1 ω2 4ω 21 ω 22 4δ 1 δ 3 4δ 2 δ 4 8δ 22 8δ 23 D 2α 3 C C ω 21 ω 22 ω 22 4ω 21 ω 21 4ω 22 2 2 2 10δ 4 4δ 3 2δ 3 D 3α 5 2 C 3ω 22 ω1 4ω 22 ω 21

8S11 D 3α 1

(10.22)

8S12

(10.23)

8S22

Substituting (10.8), (10.18), and (10.19) into (10.4) yields 2δ 1 δ1 2 δ3 u 1 D η 1 C ηN 1 C η C ηN 21 2 η 1 ηN 1 C 3ω 21 1 ω1 4ω 22 ω 21 2δ 3 2δ 2 (η 1 η 2 C ηN 1 ηN 2 ) 2 η 2 ηN 2 C ω 2 (ω 2 C 2ω 1 ) ω1 2δ 2 (η 2 ηN 1 C η 1 ηN 2 ) C C ω 2 (ω 2 2ω 1 ) 2 2δ 2 δ2 δ4 η 1 C ηN 21 2 η 1 ηN 1 C u 2 D η 2 C ηN 2 C 2 2 4ω 1 ω 2 ω2 3ω 22 2δ 4 2δ 3 (η 2 η 1 C ηN 2 ηN 1 ) 2 η 2 ηN 2 C ω 1 (ω 1 C 2ω 2 ) ω2 2δ 3 (η 2 ηN 1 C η 1 ηN 2 ) C C ω 1 (ω 1 2ω 2 )

(10.24)

η 22 C ηN 22

(10.25)

η 22 C ηN 22

(10.26)

Substituting for the F m n into (10.9), we obtain the normal form ηP 1 D i ω 1 η 1 2 µ 1 η 1 C C

4 i 2 S11 η 21 ηN 1 C S12 η 2 ηN 2 η 1 ω1

i 2 f 1p ηN 1 z p e i τ 1p C f 2m η 2 zN m e i τ 2m C f 2n ηN 2 z n e i τ 2n C 2ω 1 (10.27)

10.3 The Case of Three-to-One Internal Resonance

ηP 2 D i ω 2 η 2 2 µ 2 η 2 C C

4 i 2 S12 η 1 ηN 1 η 2 C S22 η 22 ηN 2 ω2

i 2 g 1m η 1 z m e i ν 1m C g 1n ηN 1 z n e i ν 1n C g 2q ηN 2 z q e i ν 2q C (10.28) 2ω 2

Replacing the η n in (10.25)–(10.28) with A n e i ω n t , we obtain the same results found by using the method of multiple scales (Nayfeh and Jebril, 1987). Moreover, in the absence of damping and forcing, (10.27) and (10.28) are derivable from the Lagrangian L D i ω 1 η 1 ηPN 1 ηP 1 ηN 1 C i ω 2 η 2 ηPN 2 ηP 2 ηN 2 2ω 21 η 1 ηN 1 2ω 22 η 2 ηN 2 4 2 S11 η 21 ηN 21 C 2S12 η 1 ηN1 η 2 ηN2 C S22 η 22 ηN 22

10.3 The Case of Three-to-One Internal Resonance

The three-to-one internal resonance produces the additional near-resonance terms 4i S13 η 2 ηN 21 ω1

and

4i S23 η 31 ω2

in (10.12) and (10.13), respectively, and hence F12 and F22 are chosen to contain these additional terms, respectively, where 2δ 1 δ 2 4δ 1 δ 2 4δ 2 δ 3 2δ 2 δ 3 C (10.29) C C ω 2 (ω 2 2ω 1 ) ω 1 (ω 1 2ω 2 ) 4ω 21 ω 22 3ω 21 2δ 1 δ 2 2δ 2 δ 3 D α2 C C (10.30) 2 3ω 1 4ω 21 ω 22

8S13 D 3α 2 C 8S23

Consequently, to the second approximation, u 1 and u 2 are given by (10.25) and (10.26), where η 1 and η 2 are given by the normal form 4 i 2 S11 η 21 ηN 1 C S12 η 2 ηN 2 η 1 C S13 η 2 ηN 21 ω1 i 2 f 1p ηN 1 z p e i τ 1p C f 2m η 2 zN m e i τ 2m C f 2n ηN 2 z n e i τ 2n C C 2ω 1 (10.31)

ηP 1 D i ω 1 η 1 2 µ 1 η 1 C

4 i 2 S12 η 1 ηN 1 η 2 C S22 η 22 ηN 2 C S23 η 31 ω2 i 2 g 1m η 1 z m e i ν 1m C g 1n ηN 1 zN n e i ν 1n C g 2q ηN 2 z q e i ν 2q C (10.32) C 2ω 2

ηP 2 D i ω 2 η 2 2 µ 2 η 2 C

As ω 2 ! 3ω 1 , (10.29) and (10.30) reduce to 2δ 1 δ 2 6δ 2 δ 3 ω 21 5ω 21 2δ 1 δ 2 2δ 2 δ 3 D α2 C 3ω 21 5ω 21

8S13 D 3α 2 C 8S23

263

264

10 Systems with Quadratic and Cubic Nonlinearities

and thus S13 D 3S23 . Therefore, in the absence of damping and forcing, (10.31) and (10.32) are derivable from the Lagrangian L D i ω 1 (η 1 ηPN 1 ηP 1 ηN 1 ) C i ω 2 (η 2 ηPN 2 ηP 2 ηN 2 ) 2ω 21 η 1 ηN 1 2ω 22 η 2 ηN 2 4 2 S11 η 21 ηN 21 C 2S12 η 1 ηN1 η 2 ηN2 C S22 η 22 ηN 22 C S23 η 2 ηN 31 C ηN 2 η 31

10.4 The Case of One-to-One Internal Resonance

Choosing F12 and F22 to eliminate the resonance near-resonance terms in (10.12) and (10.13) and then substituting the results into (10.9), we obtain the following normal form: 4 i 2 S11 η 21 ηN 1 C S12 η 2 ηN 2 η 1 C S13 η 21 ηN 2 ω1 CS14 η 22 ηN 1 C S15 η 1 ηN 1 η 2 C S16 η 22 ηN 2

ηP 1 D i ω 1 η 1 2 µ 1 η 1 C

i 2 f 1p ηN 1 z p e i τ 1p C f 2m η 2 zN m e i τ 2m C f 2n ηN 2 z n e i τ 2n 2ω 1 4 i 2 S12 η 2 η 1 ηN 1 C S22 η 22 ηN 2 C S23 η 21 ηN 1 ηP 2 D i ω 2 η 2 2 µ 2 η 2 C ω2 CS24 η 22 ηN 1 C S25 η 21 ηN 2 C S26 η 2 ηN 2 η 1 C

C

i 2 g 1m η 1 z m e i ν 1m C g 1n ηN 1 z n e i ν 1n C g 2q ηN 2 z q e i ν 2q 2ω 2

(10.33)

(10.34)

where 2δ 1 δ 2 4δ 2 δ 3 4δ 1 δ 2 2δ 2 δ 3 C C C ω 1 (ω 1 2ω 2 ) ω 2 (ω 2 2ω 1 ) 4ω 21 ω 22 3ω 21 2δ 1 δ 3 2δ 2 δ 4 4δ 22 4δ 23 D α3 C C C C 2 2 2 ω 2 (ω 2 2ω 1 ) ω 1 (ω 1 2ω 2 ) 4ω 2 ω 1 3ω 2 8δ 1 δ 2 8δ 2 δ 3 4δ 1 δ 2 4δ 2 δ 3 D 6α 2 C 2 C 2 2 2 ω 2 4ω 1 ω 1 4ω 2 ω 21 ω 22 10δ 3 δ 4 4δ 2 δ 3 2δ 2 δ 3 D 3α 4 C 3ω 22 ω 21 4ω 22 ω 21 10δ 1 δ 2 4δ 2 δ 3 2δ 2 δ 3 D 3α 2 C 3ω 21 ω 22 4ω 21 ω 22

S13 D 3α 2 C

(10.35)

S14

(10.36)

S15 S16 S23

(10.37) (10.38) (10.39)

10.4 The Case of One-to-One Internal Resonance

2δ 2 δ 3 2δ 3 δ 4 4δ 2 δ 3 4δ 3 δ 4 C C C (10.40) 2 2 2 ω 2 (ω 2 2ω 1 ) ω 1 (ω 1 2ω 2 ) 4ω 2 ω 1 3ω 2 2δ 1 δ 3 2δ 2 δ 4 4δ 22 4δ 23 D α3 C C C C (10.41) ω 2 (ω 2 2ω 1 ) ω 1 (ω 1 2ω 2 ) 3ω 21 4ω 21 ω 22 4δ 1 δ 3 4δ 2 δ 4 8δ 2 8δ 2 D 6α 4 C 2 2 2 C 2 3 2 (10.42) 2 2 ω1 ω2 ω 2 4ω 1 ω 1 4ω 2

S24 D 3α 4 C S25 S26

Therefore, to the second approximation, u 1 and u 2 are given by (10.25) and (10.26), where η 1 and η 2 are given by (10.33) and (10.34). As ω 2 ! ω 1 , (10.35)–(10.42) reduce to 8S13 D 3α 2

10δ 1 δ 2 10δ 2 δ 3 3ω 21 3ω 21

4δ 22 2δ 1 δ 3 4δ 23 2δ 2 δ 4 C 2 C ω 21 3ω 21 ω1 3ω 21 20δ 1 δ 2 20δ 2 δ 3 D 6α 2 3ω 21 3ω 21 10δ 2 δ 3 10δ 3 δ 4 D 3α 4 3ω 21 3ω 21 10δ 1 δ 2 10δ 2 δ 3 D 3α 2 3ω 21 3ω 21 10δ 2 δ 3 10δ 3 δ 4 D 3α 4 3ω 21 3ω 21

8S14 D α 3 8S15 8S16 8S23 8S24

4δ 22 2δ 1 δ 3 4δ 23 2δ 2 δ 4 C 2 C 2 ω1 3ω 21 ω1 3ω 21 20δ 2 δ 3 20δ 3 δ 4 D 6α 4 3ω 21 3ω 21

8S25 D α 3 8S26 Thus,

S13 D S23 D 12 S15 ,

S16 D S24 D 12 S26 ,

and

S14 D S25

Therefore, in the absence of damping and forcing, (10.33) and (10.34) are derivable from the Lagrangian L D i ω 1 (η 1 ηPN 1 ηP 1 ηN 1 ) C i ω 2 (η 2 ηPN 2 ηP 2 ηN 2 ) 2ω 21 η 1 ηN 1 2ω 22 η 2 ηN 2 4 2 S11 η 21 ηN 21 C S22 η 22 ηN 22 C 2S12 η 1 ηN1 η 2 ηN2 C2S13 η 21 ηN 1 ηN 2 C ηN 21 η 1 η 2 CS14 η 22 ηN 21 C η 21 ηN 22 C 2S16 η 22 ηN 2 ηN 1 C ηN 22 η 2 η 1

265

266

10 Systems with Quadratic and Cubic Nonlinearities

10.5 The Case of Two-to-One Internal Resonance

We consider the case of ω 2 2ω 1 . Speciﬁcally, we let ω 2 D 2ω 1 C σ, where σ is a detuning parameter. The case of ω 1 2ω 2 can be treated in a similar fashion. It follows from (10.16) and (10.17) that Γ9 and Λ 1 have small divisors and hence the term η 2 ηN 1 in (10.10) and the term η 21 in (10.11) are near-resonance terms. For a uniform expansion, we choose F11 and F21 to eliminate these terms; that is, F11 D

i δ2 η 2 ηN 1 ω1

and

F21 D

i δ2 2 η 2ω 2 1

(10.43)

Then, substituting (10.14) and (10.15) into (10.10) and (10.11) and eliminating the nonresonance terms, we ﬁnd that the Γm and Λ n , except Γ9 and Λ 1 , are given by (10.16) and (10.17). Substituting (10.14)–(10.17) into (10.4), we obtain

2δ 1 2 δ1 2 δ3 η C ηN 21 2 η 1 ηN 1 C η C ηN 22 3ω 21 1 ω1 4ω 22 ω 21 2 2δ 3 2δ 2 (η 1 η 2 C ηN 1 ηN 2 ) 2 η 2 ηN 2 C ω 2 (ω 2 C 2ω 1 ) ω1 δ2 (η 1 ηN 2 C ηN 1 η 2 ) C (10.44) ω1 ω2 2δ 2 2 δ2 u 2 D η 2 C ηN 2 η C ηN 21 C 2 η 1 ηN 1 2ω 2 (ω 2 C 2ω 1 ) 1 ω2 2δ 4 δ4 2δ 3 (η 2 η 1 ηN 2 ηN 1 ) η 2 C ηN 22 C 2 η 2 ηN 2 ω 1 (ω 1 C 2ω 2 ) 3ω 22 2 ω2 2δ 3 (η 2 ηN 1 C η 1 ηN 2 ) C (10.45) ω 2 (ω 1 2ω 2 ) u 1 D η 1 C ηN 1 C

Substituting (10.14)–(10.17) into (10.12) and (10.13) and choosing F12 and F22 to eliminate the resonance and near-resonance terms, we obtain F12 D µ 1 η 1 C

i f 1p ηN 1 z p e i τ 1p C f 2m η 2 zN m e i τ 2m C f 2n ηN 2 z n e i τ 2n 2ω 1 4i S21 η 1 ηN 1 η 2 C S22 η 22 ηN 2 D µ 2 η 2 C ω2 i g 1m η 1 z m e i ν 1m C g 1n ηN 2 zN n e i ν 1n C g 2q ηN 2 z q e i ν 2q C 2ω 2 C

F22

4i S11 η 21 ηN 1 C S12 η 2 ηN 2 η 1 ω1

(10.46)

(10.47)

10.6 Method of Multiple Scales

where 10δ 21 4δ 22 δ 22 (10.48) ω 2 (ω 2 C 2ω 1 ) 3ω 21 ω 22 4δ 1 δ 3 4δ 2 δ 4 8δ 2 2δ 22 D 8S21 D 2α 3 C 2 3 2 (10.49) 2 2 ω1 ω2 ω 1 4ω 2 ω 1 (ω 2 C 2ω 1 ) 10δ 24 4δ 23 2δ 23 D 3α 5 2 C (10.50) 2 3ω 2 ω1 4ω 22 ω 21

8S11 D 3α 1 8S12 8S22

Therefore, to the second approximation, u 1 and u 2 are given by (10.44) and (10.45) where η 1 and η 2 are given by the normal form iδ 2 4 i 2 S11 η 21 ηN 1 C S12 η 2 ηN 2 η 1 η 2 ηN 1 2 µ 1 η 1 C ω1 ω1 i 2 f 1p ηN 1 z p e i τ 1p C f 2m η 2 zN m e i τ 2m C f 2n ηN 2 z n e i τ 2n C (10.51) 2ω 1 iδ 2 2 4 i 2 S21 η 1 ηN 1 η 2 C S22 η 22 ηN 2 ηP 2 D i ω 2 η 2 C η1 2 µ2 η2 C 2ω 2 ω2 i 2 g 1m η 1 z m e i ν 1m C g 1n ηN 2 z n e i ν 1n C g 2q ηN 2 z q e i ν 2q (10.52) C 2ω 2

ηP 1 D i ω 1 η 1 C

In the absence of damping and forcing, (10.51) and (10.52) are derivable from the Lagrangian L D i ω 1 η 1 ηPN 1 ηP 1 ηN 1 C i ω 2 η 2 ηPN 2 ηP 2 ηN 2 2ω 21 η 1 ηN 1 2ω 22 η 2 ηN 2 δ 2 η 2 ηN 21 C ηN 2 η 21 4 2 S11 η 21 ηN 21 C 2S12 η 1 ηN1 η 2 ηN2 C S22 η 22 ηN 22

10.6 Method of Multiple Scales

To describe the method without algebraic complication, we consider the following special case of (10.1) and (10.2): uR 1 C ω 21 u 1 D 2δ 2 u 1 u 2

(10.53)

uR 2 C ω 22 u 2 D δ 2 u21

(10.54)

where ω 2 2ω 1 . We describe three implementations of the method of multiple scales: (a) treatment of the governing equations in second-order form, (b) treatment of the governing equations in state-space form, and (a) treatment of the governing equations in complex-valued form.

267

268

10 Systems with Quadratic and Cubic Nonlinearities

10.6.1 Second-Order Form

We seek a second-order approximation of the solution of (10.53) and (10.54) in the form u n (tI ) D

2 X

m u n m (T0 , T1 , T2 ) C

(10.55)

mD0

Substituting (10.55) into (10.53) and (10.54) and equating coefﬁcients of like powers of , we obtain:

Order (0 )

D02 u 10 C ω 21 u 10 D 0

(10.56)

D02 u 20 C ω 22 u 20 D 0

(10.57)

Order ()

D02 u 11 C ω 21 u 11 D 2D0 D1 u 10 2δ 2 u 10 u 20

(10.58)

D02 u 21 C ω 22 u 21 D 2D0 D1 u 20 δ 2 u210

(10.59)

Order (2 )

D02 u 12 C ω 21 u 12 D 2D0 D1 u 11 D12 u 10 2δ 2 u 11 u 20 2D0 D2 u 10 2δ 2 u 10 u 21

(10.60)

D02 u 22 C ω 22 u 22 D 2D0 D1 u 21 D12 u 20 2D0 D2 u 20 2δ 2 u 10 u 11 (10.61) The general solutions of (10.56) and (10.57) can be expressed as u m0 D A m (T1 , T2 ) e i ω m T0 C cc

(10.62)

Then, (10.58) and (10.59) become D02 u 11 C ω 21 u 11 D 2i ω 1 D1 A 1 e i ω 1 T0 2δ 2 A 2 A 1 e i(ω 2Cω 1 )T0 2δ 2 A 2 AN 1 e i(ω 2ω 1 )T0 C cc D02 u 21 C ω 22 u 21 D 2i ω 2 D1 A 2 e i ω 2 T0 δ 2 A21 e 2i ω 1 T0 δ 2 A 1 AN 1 C cc

(10.63)

(10.64)

Any particular solution of (10.63) and (10.64) contains secular and small-divisor terms when ω 2 2ω 1 . To quantify the nearness of ω 2 to 2ω 1 , we introduce the detuning parameter σ deﬁned by ω 2 D 2ω 1 C σ

(10.65)

10.6 Method of Multiple Scales

Using (10.65) in eliminating the terms that lead to secular terms from (10.63) and (10.64), we have i ω 1 D1 A 1 D δ 2 A 2 AN 1 e i σ T1

(10.66)

2i ω 2 D1 A 2 D δ 2 A21 e i σ T1

(10.67)

Then, the general solutions of (10.63) and (10.64) can be expressed as 2δ 2 A 2 A 1 e i(ω 2Cω 1 )T0 C cc ω 2 (ω 2 C 2ω 1 ) δ2 2 A 1 AN 1 C cc ω2

u 11 D B1 (T1 , T2 ) e i ω 1 T0 C

(10.68)

u 21 D B2 (T1 , T2 ) e i ω 2 T0

(10.69)

Substituting (10.62) and (10.66)–(10.69) into (10.60) and (10.61) and eliminating the terms that lead to secular terms, we obtain 2i ω 1 D2 A 1 2i ω 1 D1 B1 D12 A 1 D 2δ 2 A 2 BN 1 C B2 AN 1 e i σ T1

4δ 22 2 N 4δ 22 A1A1 C A 1 A 2 AN 2 2 ω 2 (ω 2 C 2ω 1 ) ω2

(10.70)

2i ω 2 D2 A 2 2i ω 2 D1 B2 D12 A 2 D 2δ 2 A 1 B1 e i σ T1 C

4δ 22 A 2 A 1 AN 1 ω 2 (ω 2 C 2ω 1 )

(10.71)

Using (10.66) and (10.67) to eliminate D12 A 1 and D12 A 2 from (10.70) and (10.71) and using the method of reconstitution, we obtain the modulation equations 2i ω 1 AP 1 D 2δ 2 A 2 AN 1 e iσ t C 2 2i ω 1 D1 B1 2δ 2 A 2 BN 1 C B2 AN 1 e iσ t ω2 σ δ2 4δ 22 iσ t N A21 AN 1 C A2A1e C 2 1C ω1 8ω 1 ω2

4ω 21 δ2 A 1 A 2 AN 2 C 22 1 C ω 2 (ω 2 C 2ω 1 ) ω1 2i ω 2 AP 2 D δ 2 A21 e iσ t C 2 2i ω 2 D1 B2 2δ 2 A 1 B1 e iσ t

σ δ 2 2 iσ t δ 22 A1e C 2ω 2 ω1 ω2

1

4ω 1 ω 2 C 2ω 1

(10.72)

A 2 A 1 AN 1 C (10.73)

When B1 D 0 and B2 D 0, (10.72) and (10.73) are not derivable from a Lagrangian because the coefﬁcient of A 1 A 2 AN 2 in (10.72) is different from the coefﬁcient of A 2 A 1 AN 1 in (10.73). The ﬁrst is (3δ 22 )/(2ω 21 ), whereas the second is 0 when ω 2 D 2ω 1 . Moreover, the coefﬁcient of A 2 AN 1 in (10.72) is not equal to twice

269

270

10 Systems with Quadratic and Cubic Nonlinearities

the coefﬁcient of A21 in (10.73). To make these coefﬁcients the same, we choose B 1 and B2 in (10.70) and (10.71) such that 2i ω 1 D1 B1 C D12 A 1 D 0 and 2i ω 2 D1 B2 C D12 A 2 D 0

(10.74)

or B1 D

δ2 A 2 AN 1 e i σ T1 2ω 21

and

B2 D

δ 2 2 i σ T1 A e 4ω 22 1

(10.75)

Substituting (10.75) into (10.70) and (10.71), letting ω 2 D 2ω 1 , and using the method of reconstitution yields 2 9δ 2 2 N δ 22 N 2 C (10.76) A A A C A A 2i ω 1 AP 1 D 2δ 2 A 2 AN 1 e iσ t C 2 1 1 2 1 8ω 21 2ω 21 2 δ 2i ω 2 AP 2 D δ 2 A21 e iσ t C 22 2 A 2 A 1 AN 1 (10.77) 2ω 1 which are derivable from the Lagrangian 9δ 22 2 2 N2 A1A1 L D i ω 1 A 1 APN 1 AN 1 AP 1 C i ω 2 A 2 APN 2 AN 2 AP 2 C 16ω 21 δ2 δ 2 A 2 AN21 e iσ t C A21 AN 2 e iσ t C 22 2 A 1 AN 1 A 2 AN 2 2ω 1 (10.78) Letting η m (t) D A m (t)e i ω m t in (10.51) and (10.52) and setting f 1m D 0, g 1m D 0, δ 1 D 0, δ 3 D 0, and δ 4 D 0, we obtain (10.76) and (10.77). To compare the results obtained in this section with those obtained in Section 10.7, we express them in real-variable form by introducing the polar transformation A n D 12 a n e i β n into (10.76) and (10.77) and separating real and imaginary parts. The result is δ 2 a 1 a 2 sin (β 2 2β 1 C σ t) 2ω 1 δ 2 2 a sin (β 2 2β 1 C σ t) aP 2 D 4ω 2 1

aP 1 D

δ 2 9 2 δ 22 3 2 δ 22 a 1 a 2 cos (β 2 2β 1 C σ t) a a 1 a 22 a 1 βP 1 D 2ω 1 64ω 31 1 16ω 31 δ 2 2 2 δ 22 2 a 1 cos (β 2 2β 1 C σ t) a a2 a 2 βP 2 D 4ω 2 32ω 31 1

(10.79) (10.80) (10.81) (10.82)

10.6 Method of Multiple Scales

Substituting (10.62), (10.68), and (10.69) into (10.55) and using (10.75) yields δ 2 2δ 2 N 1 e i σ T1 e i ω 1 T0 C A A A 2 A 1 e i(ω 2Cω 1 )T0 Ccc 2 2 ω (ω 2ω 1 2 2 C 2ω 1 ) (10.83) δ 2 2 i σ T1 i ω 2 T0 δ 2 e A e 2 A 1 AN 1 C cc (10.84) u2 D A 2 4ω 22 1 ω2

u1 D

A1

Then, using the polar transformation, we rewrite (10.83) and (10.84) as δ 2 a 1 a 2 cos [(ω 2 C ω 1 ) t C β 2 C β 1 ] ω 2 (ω 2 C 2ω 1 )

ω 2 (ω 2 C 2ω 1 ) ) ] [(ω t C β C cos ω β 2 1 2 1 (10.85) 4ω 21

u 1 D a 1 cos (ω 1 t C β 1 ) C

u 2 D a 2 cos (ω 2 t C β 2 )

δ 2 a 21 fcos (2ω 1 t C 2β 1 ) C 4g C 8ω 22

(10.86)

10.6.2 State-Space Form

We start with writing (10.53) and (10.54) in the state-space form uP 1 v1 D 0

(10.87)

vP1 C ω 21 u 1 D 2δ 2 u 1 u 2

(10.88)

uP 2 v2 D 0

(10.89)

vP2 C ω 22 u 2 D δ 2 u21

(10.90)

Then, we seek a second-order approximate solution of (10.87)–(10.90) in the form un D vn D

2 X

m u n m (T0 , T1 , T2 ) C

(10.91)

m v n m (T0 , T1 , T2 ) C

(10.92)

mD0 2 X mD0

Substituting (10.91) into (10.87)–(10.90) and equating coefﬁcients of like powers of yields.

271

272

10 Systems with Quadratic and Cubic Nonlinearities

Order (0 )

D0 u 10 v10 D 0

(10.93)

D0 v10 C ω 21 u 10 D 0

(10.94)

D0 u 20 v20 D 0

(10.95)

D0 v20 C ω 22 u 20 D 0

(10.96)

Order ()

D0 u 11 v11 D D1 u 10

(10.97)

D0 v11 C ω 21 u 11 D D1 v10 2δ 2 u 10 u 20

(10.98)

D0 u 21 v21 D D1 u 20

(10.99)

D0 v21 C ω 22 u 21 D D1 v20 δ 2 u210

(10.100)

Order (2 )

D0 u 12 v12 D D1 u 11 D2 u 10

(10.101)

D0 v12 C ω 21 u 12 D D1 v11 D2 v10 2δ 2 u 10 u 21 2δ 2 u 11 u 20

(10.102)

D0 u 22 v22 D D1 u 21 D2 u 20

(10.103)

D0 v22 C ω 22 u 22 D D1 v21 D2 v20 2δ 2 u 10 u 11

(10.104)

The solutions of (10.93)–(10.96) can be expressed as u 10 D A 1 (T1 , T2 ) e i ω 1 T0 C cc ,

v10 D i ω 1 A 1 (T1 , T2 ) e i ω 1 T0 C cc (10.105)

u 20 D A 2 (T1 , T2 ) e i ω 2 T0 C cc ,

v20 D i ω 2 A 2 (T1 , T2 ) e i ω 2 T0 C cc (10.106)

Substituting (10.105) and (10.106) into (10.97)–(10.100) yields D0 u 11 v11 D D1 A 1 e i ω 1 T0 C cc

(10.107)

D0 v11 C ω 21 u 11 D i ω 1 D1 A 1 e i ω 1 T0 2δ 2 A 2 A 1 e i(ω 2Cω 1 )T0 2δ A 2 AN 1 e i(ω 2ω 1 )T0 C cc D0 u 21 v21 D D1 A 2 e i ω 2 T0 C cc

(10.108) (10.109)

D0 v21 C ω 22 u 21 D i ω 2 D1 A 2 e i ω 2 T0 δ 2 A21 e 2i ω 1 T0 δ 2 A 1 AN 1 C cc (10.110)

10.6 Method of Multiple Scales

Because the homogeneous equations (10.107)–(10.110) have nontrivial solutions, the nonhomogeneous equations have solutions only if the nonhomogeneous terms are orthogonal to every solution of the corresponding adjoint equations. In this case, the solution of the adjoint of (10.107) and (10.108) is [i ω 1

1]T e i ω 1 T0

Therefore, using (10.65) and imposing the solvability condition on (10.107) and (10.108), one obtains 2i ω 1 D1 A 1 D 2δ 2 A 2 AN 1 e i σ T1

(10.111)

Then, although (10.107) and (10.108) are solvable, their solution is not unique. A unique solution can be obtained by requiring it to be orthogonal to the above adjoint solution. Hence, the unique solution of (10.107) and (10.108) is 2δ 2 δ2 A 2 A 1 e i(ω 2Cω 1 )T0 2 A 2 AN 1 e i(ω 2ω 1 )T0 Ccc (10.112) ω 2 (ω 2 C 2ω 1 ) 2ω 1 2i δ 2 (ω 2 C ω 1 ) i δ2 D A 2 AN 1 e i(ω 2ω 1 )T0 Ccc (10.113) A 2 A 1 e i(ω 2Cω 1 )T0 C ω 2 (ω 2 C 2ω 1 ) ω1

u 11 D v11

Following steps similar to those used above, one ﬁnds that the solvability condition of (10.109) and (10.110) is 2i ω 2 D1 A 2 D δ 2 A21 e i σ T1

(10.114)

and their unique solution is δ2 δ 2 2 2i ω 1 T0 A 1 AN 1 A e C cc ω 22 4ω 22 1 i δ 2 2 2i ω 1 T0 D A e C cc 4ω 2 1

u 21 D

(10.115)

v21

(10.116)

Substituting (10.112), (10.113), (10.115), and (10.116) into (10.101)–(10.104) and imposing the solvability conditions yields δ 22 (8ω 1 C 5ω 2 ) 2 N 2δ 22 A 1 A 2 AN 2 A1A1 C 2 ω 1 (ω 2 C 2ω 1 ) ω 2 (ω 2 C 2ω 1 ) 2δ 22 2i ω 2 D2 A 2 D A 2 A 1 AN 1 ω 1 (ω 2 C 2ω 1 ) 2i ω 1 D2 A 1 D

(10.117) (10.118)

273

274

10 Systems with Quadratic and Cubic Nonlinearities

Using the method of reconstitution, we combine (10.111), (10.114), (10.117), and (10.118) into

C

2δ 22 ω 1 (ω 2 C 2ω 1 )

δ 22 (8ω 1 C 5ω 2 ) 2 N A A1 ω 22 (ω 2 C 2ω 1 ) 1 A 1 A 2 AN 2 C

2i ω 1 AP 1 D 2δ 2 A 2 AN 1 e iσ t C 2

2i ω 2 AP 2 D δ 2 A21 e iσ t C

2 2 δ 22 A 2 A 1 AN 1 C ω 1 (ω 2 C 2ω 1 )

(10.119) (10.120)

Equations 10.119 and 10.120 are derivable from the Lagrangian L D i ω 1 A 1 APN 1 AN 1 AP 1 C i ω 2 A 2 APN 2 AN 2 AP 2 2 δ 22 (8ω 1 C 5ω 2 ) 2 2 δ 2 A 2 AN21 e iσ t C A21 AN 2 e iσ t C A AN 2ω 22 (ω 2 C 2ω 1 ) 1 1 C

2δ 22 A 1 AN 1 A 2 AN 2 ω 1 (ω 2 C 2ω 1 )

(10.121)

Equations 10.119–10.121 reduce to (10.76)–(10.78) when ω 2 is replaced with 2ω 1 . 10.6.3 Complex-Valued Form

Using the transformation (10.4) and (10.5), we rewrite (10.53) and (10.54) in the complex-valued form iδ 2 ζP 1 D i ω 1 ζ1 C ω1 iδ 2 ζP 2 D i ω 2 ζ2 C 2ω 2

ζ1 C ζN 1 ζ1 C ζN 1

ζ2 C ζN 2

2

(10.122) (10.123)

We seek a second-order uniform expansion of (10.122) and (10.123) in the form ζ n (tI ) D

2 X

m δ n m (T0 , T1 , T2 ) C

(10.124)

mD0

Substituting (10.124) into (10.122) and (10.123) and equating coefﬁcients of like powers of , we obtain

Order (0 )

D0 ζ11 i ω 1 ζ11 D 0

(10.125)

D0 ζ21 i ω 2 ζ21 D 0

(10.126)

10.6 Method of Multiple Scales

Order ()

i δ2 ζ11 C ζN 11 ζ21 C ζN 21 ω1 2 i δ2 ζ11 C ζN 11 D D1 ζ21 C 2ω 2

D0 ζ12 i ω 1 ζ12 D D1 ζ11 C

(10.127)

D0 ζ22 i ω 2 ζ22

(10.128)

Order (2 )

i δ2 ζ11 C ζN 11 ζ22 C ζN 22 ω1 i δ2 ζ12 C ζN 12 ζ21 C ζN 21 C (10.129) ω1 i δ2 ζ11 C ζN 11 ζ12 C ζN 12 (10.130) D D2 ζ21 D1 ζ22 C ω2

D0 ζ13 i ω 1 ζ13 D D2 ζ11 D1 ζ12 C

D0 ζ23 i ω 2 ζ23

The solutions of (10.125) and (10.126) can be expressed as ζ n1 D A n (T1 , T2 ) e i ω n T0

(10.131)

Substituting (10.131) into (10.127) and (10.128) yields i δ2 h A 2 AN 1 e i(ω 2ω 1 )T0 ω1 CA 2 A 1 e i(ω 2Cω 1 )T0 C A 1 AN 2 e i(ω 1ω 2 )T0 i C AN 2 AN 1 e i(ω 2Cω 1 )T0

D0 ζ12 i ω 1 ζ12 D D1 A 1 e i ω 1 T0 C

i δ 2 2 2i ω 1 T0 A1e C 2A 1 AN 1 D0 ζ22 i ω 2 ζ22 D D1 A 2 e i ω 2 T0 C 2ω 2 C AN21 e 2i ω 1 T0

(10.132)

(10.133)

Using (10.65) in (10.132) and (10.133) and eliminating the terms that produce secular terms, we obtain i δ2 A 2 AN 1 e i σ T1 ω1 i δ 2 2 i σ T1 A e D1 A 2 D 2ω 2 1

D1 A 1 D

(10.134) (10.135)

Then, the solutions of (10.132) and (10.133) can be expressed as δ2 δ2 A 2 A 1 e i(ω 2Cω 1 )T0 A 1 AN 2 e i(ω 1ω 2 )T0 ω1 ω2 ω1 ω2 δ2 AN 2 AN 1 e i(ω 2Cω 1 )T0 ω 1 (ω 2 C 2ω 1 ) δ2 δ2 D 2 A 1 AN 1 AN2 e i(ω 2Cω 1 )T0 2ω 2 (ω 2 C 2ω 1 ) 1 ω2

ζ12 D

ζ22

(10.136) (10.137)

275

276

10 Systems with Quadratic and Cubic Nonlinearities

Substituting (10.131), (10.136), and (10.137) into (10.129) and (10.130), using (10.134) and (10.135), and eliminating the terms that lead to secular terms, we obtain i δ 22 (5ω 2 C 8ω 1 ) 2 N i δ 22 A1A1 2 2 2ω 1 ω 2 (ω 2 C 2ω 1 ) ω 1 (ω 2 C 2ω 1 ) 2 i δ2 A 2 A 1 AN 1 D2 A 2 D ω 1 ω 2 (ω 2 C 2ω 1 ) D2 A 1 D

(10.138) (10.139)

Using the method of reconstitution, we combine (10.134), (10.135), (10.138), and (10.139) into iδ 2 i 2 δ 22 (5ω 2 C 8ω 1 ) 2 N i 2 δ 22 AP 1 D A 2 AN 1 e i σ T1 A1A1 2 A 1 A 2 AN 2 2 ω1 2ω 1 ω 2 (ω 2 C 2ω 1 ) ω 1 (ω 2 C 2ω 1 ) (10.140) iδ 2 2 i σ T1 i 2 δ 22 A1e AP 2 D A 2 A 1 AN 1 2ω 2 ω 1 ω 2 (ω 2 C 2ω 1 )

(10.141)

Equations 10.140 and 10.141 reduce to (10.76) and (10.77) when ω 2 D 2ω 1 .

10.7 Generalized Method of Averaging

Using the method of variation of parameters, we transform (10.53) and (10.54) into u m D a m (t) cos φ m (t) ,

uP m D ω m a m (t) sin φ m (t)

(10.142)

where δ 2 a 1 a 2 [sin (φ 2 2φ 1 ) sin (φ 2 C 2φ 1 )] 2ω 1 δ 2 φP 1 D ω 1 C a 2 [cos (φ 2 2φ 1 ) C cos (φ 2 C 2φ 1 ) C 2 cos φ 2 ] 2ω 1 δ 2 2 a [sin (φ 2 2φ 1 ) C sin (φ 2 C 2φ 1 ) C 2 sin φ 2 ] aP 2 D 4ω 2 1

(10.144)

δ 2 a 21 [cos (φ 2 2φ 1 ) C cos (φ 2 C 2φ 1 ) C 2 cos φ 2 ] φP 2 D ω 2 C 4ω 2 a 2

(10.146)

aP 1 D

(10.143)

(10.145)

We seek a second-order approximate solution of (10.143)–(10.146) in the form a m (t) D η m (t) C

2 X

n a m n (ψ1 , ψ2 , η 1 , η 2 ) C

nD1 2 X

φ m (t) D ψ m (t) C

nD1

n φ m n (ψ1 , ψ2 , η 1 , η 2 ) C

(10.147) (10.148)

10.7 Generalized Method of Averaging

where the a m n and φ m n are fast-varying functions of time, ηP m D A m1 (η 1 , η 2 , t) C 2 A m2 (η 1 , η 2 , t) C

(10.149)

P m D ω m C Φm1 (η 1 , η 2 , t) C 2 Φm2 (η 1 , η 2 , t) C ψ

(10.150)

and the A m n and Φm n are slowly varying functions of time. Substituting (10.147)– (10.150) into (10.143)–(10.146) and equating the coefﬁcients of to zero yields A 11 C ω 1

@a 11 @a 11 δ2 C ω2 D η 1 η 2 [sin (ψ2 2ψ1 ) sin (ψ2 C 2ψ1 )] @ψ1 @ψ2 2ω 1 (10.151)

Φ11 C ω 1

A 21 C ω 1

@φ 11 @φ 11 δ2 C ω2 D η 2 [cos (ψ2 2ψ1 ) @ψ1 @ψ2 2ω 1 C cos (ψ2 C 2ψ1 ) C 2 cos ψ2 ] @a 21 @a 21 δ2 2 C ω2 D η [sin (ψ2 2ψ1 ) @ψ1 @ψ2 4ω 2 1 C sin (ψ2 C 2ψ1 ) C 2 sin ψ2 ]

Φ21 C ω 1

(10.152)

@φ 21 @φ 21 δ 2 η 21 [cos (ψ2 2ψ1 ) C ω2 D @ψ1 @ψ2 4ω 2 η 2 C cos (ψ2 C 2ψ1 ) C 2 cos ψ2 ]

(10.153)

(10.154)

Choosing the A n1 and Φn1 to eliminate the slowly varying terms in (10.151)– (10.154) and using (10.65), we obtain δ2 η 1 η 2 sin (ψ2 2ψ1 ) 2ω 1 δ2 η 2 cos (ψ2 2ψ1 ) Φ11 D 2ω 1 δ2 2 η sin (ψ2 2ψ1 ) A 21 D 4ω 2 1

A 11 D

Φ21 D

δ 2 η 21 cos (ψ2 2ψ1 ) 4ω 2 η 2

(10.155) (10.156) (10.157) (10.158)

Then, the solutions of (10.151)–(10.154) are given uniquely by δ2 η1 η2 cos (ψ2 C 2ψ1 ) 2ω 1 (ω 2 C 2ω 1 ) δ2 η2 δ2 η2 D sin ψ2 sin (ψ2 C 2ψ1 ) C 2ω 1 (ω 2 C 2ω 1 ) ω1 ω2

a 11 D

(10.159)

φ 11

(10.160)

δ 2 η 21 δ 2 η 21 cos ψ2 cos (ψ2 C 2ψ1 ) 4ω 2 (ω 2 C 2ω 1 ) 2ω 22 δ 2 η 21 δ 2 η 21 D sin (ψ2 C 2ψ1 ) C sin ψ2 4ω 2 (ω 2 C 2ω 1 )η 2 2ω 22 η 2

a 21 D

(10.161)

φ 21

(10.162)

277

278

10 Systems with Quadratic and Cubic Nonlinearities

Substituting (10.155)–(10.162) into the second-order problem and choosing the A n2 and Φn2 to eliminate the slowly varying terms in the resulting equations, we obtain A 12 D 0 and

A 22 D 0

(10.163)

δ 22 (8ω 21 C 5ω 1 ω 2 )η 21 C 2ω 22 η 22 8ω 21 ω 22 (ω 2 C 2ω 1 ) δ 22 D η2 4ω 1 ω 2 (ω 2 C 2ω 1 ) 1

Φ12 D

(10.164)

Φ22

(10.165)

Substituting (10.147) and (10.148) into (10.142), expanding the result for small , and using (10.159)–(10.162) yields u 1 D η 1 cos (ω 1 t C β 1 ) C

δ 2 η 1 η 2 cos [(ω 2 ω 1 ) t C β 2 β 1 ] 2ω 1 ω 2

δ 2 η 1 η 2 cos [(ω 2 C ω 1 ) t C β 2 C β 1 ] C ω 2 (ω 2 C 2ω 1 )

u 2 D η 2 cos (ω 2 t C β 2 )

(10.166)

δ 2 η 21 δ 2 2 η C cos (2ω 1 t C 2β 1 ) 4ω 2 (ω 2 C 2ω 1 ) 2ω 22 1 (10.167)

Substituting (10.155)–(10.158) and (10.163)–(10.165) into (10.149) and (10.150), we obtain the modulation equations ηP 1 D

δ 2 η 1 η 1 sin [(ω 2 2ω 1 ) t C β 2 2β 1 ] C O( 3 ) 2ω 1

(10.168)

δ 2 η 2 cos [(ω 2 2ω 1 ) t C β 2 2β 1 ] 2ω 1 2 δ 22 (8ω 21 C 5ω 1 ω 2 )η 21 C 2ω 22 η 22 C O( 3 ) 2 2 8ω 1 ω 2 (ω 2 C 2ω 1 )

P 1 D ω1 C ψ

(10.169) ηP 2 D

δ 2 2 η sin [(ω 2 2ω 1 ) t C β 2 2β 1 ] C O( 3 ) 4ω 2 1 δ 2 η 21 cos [(ω 2 2ω 1 ) t C β 2 2β 1 ] 4ω 2 η 2 2 δ 22 η 2 C O( 3 ) 4ω 1 ω 2 (ω 2 C 2ω 1 ) 1

(10.170)

P 2 D ω2 C ψ

(10.171)

Equations 10.166–10.171 are in full agreement with (10.85), (10.86), and (10.79)– (10.82) obtained with the method of multiple scales because ω 2 2ω 1 , η n D a n , and ψ n D ω n t C β n .

10.8 A Nonsemisimple One-to-One Internal Resonance

10.8 A Nonsemisimple One-to-One Internal Resonance

In this section, we use the methods of normal form and multiple scales to treat a two-degree-of-freedom system with quadratic and cubic nonlinearities whose linear part has a generic nonsemisimple structure. Speciﬁcally, we treat the system uR 1 C ω 2 u 1 C 2µ 1 uP 1 C δ 11 u21 C δ 12 u 1 u 2 C δ 13 u22 C α 11 u31 C α 12 u21 u 2 C α 13 u 1 u22 C α 14 u32 D 0

(10.172)

uR 2 C ω 2 u 2 C 2µ 2 uP 2 C u 1 C δ 21 u21 C δ 22 u 1 u 2 C δ 23 u22 C α 21 u31 C α 22 u21 u 2 C α 23 u 1 u22 C α 24 u32 D 0

(10.173)

10.8.1 The Method of Normal Forms

Again, as a ﬁrst step, we use the transformation (9.37) and (9.38) to recast (10.172) and (10.173) in the complex-valued form 2 i h ζP 1 D i ωζ1 µ 1 ζ1 ζN 1 C δ 11 ζ1 C ζN 1 C δ 12 ζ1 C ζN 1 ζ2 C ζN 2 2ω 2 i 3 2 i h N C α 11 ζ1 C ζN 1 C α 12 ζ1 C ζN 1 ζ2 C ζN 2 Cδ 13 ζ2 C ζ2 2ω 2 3 i Cα 13 ζ1 C ζN 1 ζ2 C ζN 2 C α 14 ζ2 C ζN 2 (10.174) 2 i i h ζP 2 D i ωζ2 µ 2 ζ2 ζN 2 C ζ1 C ζN 1 C δ 21 ζ1 C ζN 1 2ω 2ω 2 i 3 i h Cδ 22 ζ1 C ζN 1 ζ2 C ζN 2 C δ 23 ζ2 C ζN 2 C α 21 ζ1 C ζN 1 2ω 2 2 Cα 22 ζ1 C ζN 1 ζ2 C ζN 2 C α 23 ζ1 C ζN 1 ζ2 C ζN 2 3 i (10.175) Cα 24 ζ2 C ζN 2 To simplify the algebra, we scale the variables in (10.174) and (10.175) before determining their normal form. We treat the case of weak nonlinearities and damping. Thus, we assume that ζ1 D O(), where is a small nondimensional parameter that is used as a bookkeeping device. Because ζ2 is much larger than ζ1 due to the nonsemisimple structure of the linear undamped operator in (10.172) and (10.173), ζ2 D O( 1λ 2 ), where λ 2 > 0. Because the damping is weak, µ n D O( λ 1 ), where λ 1 > 0. To make these different orderings explicit, we introduce new scaled variables deﬁned by ζ1 D η 1 ,

ζ2 D 1λ 2 η 2 ,

µ n D λ 1 µO n

279

280

10 Systems with Quadratic and Cubic Nonlinearities

where the η n and µO n are O(1) and rewrite (10.174) and (10.175) as ηP 1 D i ωη 1 λ 1 µO 1 (η 1 ηN 1 ) C

i h δ 11 (η 1 C ηN 1 )2 2ω

i C 1λ 2 δ 12 (η 1 C ηN 1 ) (η 2 C ηN 2 ) C 12λ 2 δ 13 (η 2 C ηN 2 )2 i h 2 C α 11 (η 2 C ηN 1 )3 C 2λ 2 α 12 (η 1 C ηN 1 )2 (η 2 C ηN 2 ) 2ω i C 22λ 2 α 13 (η 1 C ηN 1 ) (η 2 C ηN 2 )2 C 23λ 2 α 14 (η 2 C ηN 2 )3

(10.176)

ηP 2 D i ωη 2 λ 1 µO 2 (η 2 ηN 2 )

i h 1Cλ 2 i λ2 δ 21 (η 1 C ηN 1 )2 (η 1 C ηN 1 ) C 2ω 2ω i C δ 22 (η 1 C ηN 1 ) (η 2 C ηN 2 ) C 1λ 2 δ 23 (η 2 C ηN 2 )2 i h 2Cλ 2 C α 21 (η 1 C ηN 1 )3 C 2 α 22 (η 1 C ηN 1 )2 (η 2 C ηN 2 ) 2ω i C 2λ 2 α 23 (η 1 C ηN 1 ) (η 2 C ηN 2 )2 C 22λ 2 α 24 (η 2 C ηN 2 )3 C

(10.177)

Because of the one-to-one internal resonance, η 1 ηN 1 η 2 , η 2 ηN 2 η 1 , η 21 ηN 1 , η 22 ηN 2 , η 21 ηN 2 , η 22 ηN 1 , η 1 , and η 2 are resonance terms in (10.176) and (10.177). Balancing the damping term and the dominant resonance term in (10.176), namely, the term proportional to α 14 , we have λ 1 D 2 3λ 2

(10.178)

Balancing the damping term and the dominant resonance term in (10.177), namely, the term i(2ω)1 η 1 , we have λ1 D λ2

(10.179)

Solving (10.178) and (10.179) yields λ1 D λ2 D

1 2

(10.180)

Hence, (10.176) and (10.177) become ηP 1 D i ωη 1 C C

i δ 13 (η 2 C ηN 2 )2 1/2 µO 1 (η 1 ηN 1 ) 2ω

i 1/2 δ 12 (η 1 C ηN 1 ) (η 2 C ηN 2 ) C α 14 (η 2 C ηN 2 )3 C 2ω

ηP 2 D i ωη 2 1/2 µO 2 (η 2 ηN 2 ) C

(10.181)

i 1/2 η 1 C ηN 1 C δ 23 (η 2 C ηN 2 )3 C 2ω (10.182)

10.8 A Nonsemisimple One-to-One Internal Resonance

To simplify (10.181) and (10.182), we introduce the transformation η 1 D ξ1 C h 11 ξn , ξN n C 1/2 h 12 ξn , ξN n C

(10.183)

η 2 D ξ2 C 1/2 h 22 ξn , ξN n C

(10.184)

and choose the h m n to eliminate the nonresonance terms so that the resulting equations have the simplest possible form ξP n D i ωξn C 1/2 g n ξm , ξN m C

(10.185)

Substituting (10.183)–(10.185) into (10.181) and (10.182) and equating coefﬁcients of like powers of , we obtain 2 i (10.186) δ 13 ξ2 C ξN2 2ω @h 11 @h 11 @h 11 @h 11 gN 1 g2 gN 2 g 1 C L(h 12 ) D g 1 @ξ1 @ξ2 @ ξN1 @ ξN2 i µO 1 ξ1 ξN1 C h 11 hN 11 C δ 13 ξ2 C ξN2 h 22 C hN 22 ω 3 i i h N N C (10.187) δ 12 ξ1 C ξ1 C h 11 C h 11 ξ2 C ξN2 C α 14 ξ2 C ξN2 2ω i i h ξ1 C ξN1 C h 11 C hN 11 g 2 C L(h 22 ) D µO 2 ξ2 ξN2 C 2ω 2 i C (10.188) δ 23 ξ2 C ξN2 2ω

L(h 11 ) D

where

@h @h N @h @h N ξ1 C ξ2 h L(h) D i ω ξ1 ξ2 @ξ1 @ξ2 @ ξN1 @ ξN2

The right-hand side of (10.186) suggests seeking h 11 in the form h 11 D Γ1 ξ22 C Γ2 ξ2 ξN2 C Γ3 ξN22

(10.189)

Substituting (10.189) into (10.186) and equating each of the coefﬁcients of ξ22 , ξ2 ξN2 , and ξN22 on both sides, we obtain Γ1 D

δ 13 δ 13 δ 13 , Γ2 D 2 , Γ3 D 2 2 2ω ω 6ω

(10.190)

It follows from (10.189) and (10.190) that δ 13 2 ξ C ξN22 6ξ2 ξN2 h 11 C hN 11 D 3ω 2 2

(10.191)

281

282

10 Systems with Quadratic and Cubic Nonlinearities

Substituting (10.191) into (10.188) yields @h 22 @h 22 N @h 22 @h 22 N ξ1 C ξ2 h 22 D µO 2 ξ2 ξN2 g2 C i ω ξ1 ξ2 @ξ1 @ξ2 @ ξN1 @ ξN2 2 i δ 13 2 N 2 N N ξ1 C ξN1 C ξ ξ C δ ξ C C ξ 6ξ C ξ (10.192) 2 2 23 2 2 2 2ω 3ω 2 2

Equating g 2 to the resonance terms in (10.192), we have g 2 D µO 2 ξ2 C

i ξ1 2ω

(10.193)

Then, substituting h 22 D Γ4 ξN2 C Γ5 ξN1 C Γ6 ξ22 C Γ7 ξ2 ξN2 C Γ8 ξN22

(10.194)

into (10.192), using (10.193), and equating each of the coefﬁcients of ξN2 , ξN1 , ξ22 , ξ2 ξN2 , and ξN22 on both sides, we obtain i µO 2 1 δ 23 δ 13 C , , Γ5 D 2 , Γ6 D 2 2ω 4ω 2ω 6ω 4 δ 23 δ 13 δ 23 δ 13 Γ7 D 2 C 4 , Γ8 D 2 ω ω 6ω 18ω 4

Γ4 D

(10.195)

Substituting (10.189)–(10.191) and (10.193)–(10.195) into (10.187) yields @h 12 @h 12 N @h 12 @h 12 N g1 C i ω ξ1 C ξ2 h 12 ξ1 ξ2 @ξ1 @ξ2 @ ξN1 @ ξN2 i D 2Γ1 ξ2 C Γ2 ξN2 µO 2 ξ2 C ξ1 2ω i N ξ1 Γ2 ξ2 C 2Γ3 ξN2 µO 2 ξN2 2ω 2 2δ 13 2 N N ξ ξ2 µO 1 ξ1 ξ1 C 3ω 2 2 i µO 2 1 i ξ1 C ξN1 C δ 13 ξ2 C ξN2 ξ2 ξN2 2 ω 2ω 4ω 2 δ δ 13 2 N 2 23 2 N N N ξ C ξ2 6ξ2 ξ2 C 4 ξ2 C ξ2 C 18ξ2 ξ2 C 9ω 3ω 2 2 3 i δ 12 ξ1 C ξN1 ξ2 C ξN2 C α 14 ξ2 C ξN2 C 2ω δ 12 δ 13 2 N 2 N N ξ2 C ξ2 6ξ2 ξ2 ξ2 C ξ2 (10.196) C 3ω 2 Since we are stopping at this order, we do not need to determine h 12 explicitly, and hence all that we need to do is to determine g 1 . Choosing g 1 to eliminate the resonance terms from (10.196), we have g 1 D µO 1 ξ1 C i α e ξ22 ξN2

(10.197)

10.8 A Nonsemisimple One-to-One Internal Resonance

where 1 10 5 38 2 3α 14 αe D δ 13 δ 23 δ 12 δ 13 C δ 2ω 3ω 2 3ω 2 9ω 4 13

(10.198)

Substituting for the g n and h m n in (10.183)–(10.185), we obtain η 1 D ξ1 C

δ 13 2 3ξ2 6ξ2 ξN2 ξN22 C 6ω 2

(10.199)

η 2 D ξ2 C

(10.200)

ξP1 D i ωξ1 1/2 µO 1 ξ1 C i 1/2 α e ξ22 ξN2 C

(10.201)

i 1/2 ξP2 D i ωξ2 1/2 µO 2 ξ2 C ξ1 C 2ω

(10.202)

10.8.2 The Method of Multiple Scales

Alternatively, we use the method of multiple scales to determine an approximation to the solution of (10.172) and (10.173). Following a procedure similar to that used above, we note that if u 1 D O(), then u 2 D O( 1/2 ) and µ n D O( 1/2 ). Hence, we introduce scaled variables deﬁned by u 1 D ν 1 ,

u 2 D 1/2 ν 2 ,

µ n D 1/2 µO n

(10.203)

in (10.172) and (10.173) and obtain νR 1 C ω 2 ν 1 C δ 13 ν 22 C 2 1/2 µO 1 νP 1 C 1/2 δ 12 ν 1 ν 2 C α 14 ν 32 C D 0 (10.204) νR 2 C ω 2 ν 2 C 2 1/2 µO 2 νP 2 C 1/2 ν 1 C δ 23 ν 22 C D 0

(10.205)

We seek an expansion of the solution of (10.204) and (10.205) valid up to O( 1/2 ) in the form ν 1 (tI ) D ν 10 (T0 , T1 ) C 1/2 ν 11 (T0 , T1 ) C

(10.206)

ν 2 (tI ) D ν 20 (T0 , T1 ) C 1/2 ν 21 (T0 , T1 ) C

(10.207)

where T0 D t and T1 D 1/2 t. In terms of T0 and T1 , the time derivatives become d D D0 C 1/2 D1 C dt

and

d2 D D02 C 2 1/2 D0 D1 C d t2

where D n D @/@Tn . Substituting (10.206) and (10.207) into (10.204) and (10.205) and equating coefﬁcients of like powers of , we obtain

283

284

10 Systems with Quadratic and Cubic Nonlinearities

Order (0 )

D02 ν 10 C ω 2 ν 10 D δ 13 ν 220

(10.208)

D02 ν 20 C ω 2 ν 20 D 0

(10.209)

Order ()

D02 ν 11 C ω 2 ν 11 D 2D0 D1 ν 10 2 µO 1 D0 ν 10 2δ 13 ν 20 ν 21 δ 12 ν 10 ν 20 α 14 ν 320

(10.210)

D02 ν 21 C ω 2 ν 21 D 2D0 D1 ν 20 2 µO 2 D0 ν 20 ν 10 δ 23 ν 220

(10.211)

The solution of (10.209) can be expressed as ν 20 D A 2 (T1 )e i ω T0 C cc

(10.212)

Then, (10.208) becomes D02 ν 10 C ω 2 ν 10 D δ 13 A22 e 2i ω T0 C A 2 AN 2 C cc whose solution can be expressed as ν 10 D A 1 (T1 )e i ω T0 C

δ 13 2 2i ω T0 δ 13 A e 2 A 2 AN 2 C cc 3ω 2 2 ω

(10.213)

Substituting (10.212) and (10.213) into (10.210) and (10.211) yields D02 ν 11 C ω 2 ν 11 D 2i ω A01 C µO 1 A 1 e i ω T0 5 A22 AN 2 e i ω T0 3α 14 δ δ 12 13 3ω 2 C cc C NST 2δ 13 ν 20 ν 21

(10.214)

δ 13 2 2i ω T0 D02 ν 21 C ω 2 ν 21 D 2i ω A02 C µO 2 A 2 e i ω T0 A 1 e i ω T0 A e 3ω 2 2 δ 13 C 2 A 2 AN 2 δ 23 A22 e 2i ω T0 C A 2 AN 2 C cc (10.215) ω Eliminating the term that produces secular terms from (10.215), we have 2i ω A02 C µO 2 A 2 C A 1 D 0 (10.216) Then, the solution of (10.215) can be expressed as δ 13 δ 23 δ 13 δ 23 2 2i ω T0 ν 21 D A A 2 AN 2 C cc C e C 2 9ω 4 3ω 2 ω4 ω2

(10.217)

Substituting (10.217) into (10.214) and eliminating the terms that produce secular terms, we obtain 2i ω A01 C µO 1 A 1 C 2ωα e A22 AN 2 D 0 (10.218) where α e is given by (10.198).

10.9 Exercises

Letting ξn D A n e i ω n t in (10.201) and (10.202) yields (10.216) and (10.218). Hence, the results obtained by using the method of normal forms are in full agreement with those found by using the method of multiple scales.

10.9 Exercises

10.9.1 Consider the system uR 1 C ω 21 u 1 D δ 1 uP 1 uP 2 uR 2 C ω 22 u 2 D δ 2 uP 21 Use the method of normal forms, the method of multiple scales, and the generalized method of averaging to determine a second-order approximation to the solution of this system when ω 2 2ω 1 . 10.9.2 Consider the system uR 1 C ω 21 u 1 D δ 1 u 1 uR 2 uR 2 C ω 22 u 2 D δ 2 u 1 uR 1 Use the method of normal forms, the method of multiple scales, and the generalized method of averaging to determine a second-order approximation to the solution of this system when ω 2 2ω 1 .

285

287

11 Retarded Systems Several approaches have been proposed in the literature to analyze the nature of Hopf bifurcations of retarded systems, including integral averaging, the Fredholm alternative, the implicit function theorem, the method of multiple scales, and center-manifold reduction. In this chapter, we follow Nayfeh (2008) and use two of these approaches, the method of multiple scales and center-manifold reduction, to analyze the nature of Hopf bifurcations in retarded systems modeled by nonlinear homogeneous ordinary-differential equations with discrete time delay. Such equations model the behavior of many physical systems arising in physiology, biology, population dynamics, machine-tool chatter, neural networks, and time-delayed feedback controlled systems. To describe the approaches without getting involved in the algebra, we use three simple systems, namely a scalar equation, a single-degree-of-freedom system, and a three-neuron model. We show that the normal forms obtained with all the approaches are the same. However, the method of multiple scales seems to be simpler. In fact, the method of multiple scales is directly applied to the retarded differential equations. In contrast, in the center-manifold approach, one needs to convert the retarded differential equations into operator equations in a Banach space, introduce a device that acts like an inner product because the Banach space does not have a natural inner product associated with its norm, deﬁne the adjoint associated with the linear part of the differential equations, perform the projection onto the center manifold, and calculate the normal form of the subsystem describing the dynamics on the center manifold. Finally, we consider a problem in which the retarded term appears as an acceleration and treat it using the method of multiple scales only.

11.1 A Scalar Equation

We start by a scalar equation with discrete time delay, which models a single neuron with a general activation function; that is, d x(t) O D x(t) O C g x(t) O β x(t O τ) dt

(11.1)

The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

288

11 Retarded Systems

where g( x) O is a general three-times continuously differentiable function, τ > 0, and β > 0. We let x denote an equilibrium of (11.1); that is, x D g[(1 β)x ] Then, we let x(t) O D x C x (t) in (11.1), expand the result in a Taylor series in x, keep up to cubic terms, and obtain x(t) P D (α 1 1)x (t) α 1 β x (t τ) α 2 [x (t) β x (t τ)]2 3 α 3 x (t) β x (t τ)

(11.2)

where α 1 D g 0 [(1 β)x ] ,

α 2 D 12 g 00 [(1 β)x ] ,

α 3 D 61 g 000 [(1 β)x ]

Linearizing (11.2) yields x(t) P D (α 1 1)x (t) α 1 β x (t τ)

(11.3)

which has solutions of the form x D x0 e (σCi ω)t

(11.4)

where σ is the growth or decay rate, ω is the frequency of oscillations, and x0 depends on the initial conditions. Substituting (11.4) into (11.3) yields the characteristic equation σ C i ω D α 1 1 α 1 β e (σCi ω)τ

(11.5)

For a given α 1 , β, and τ, the complex-valued (11.5) has inﬁnitely many solutions σ and ω. When σ > 0 the trivial solution is unstable; when σ < 0 the trivial solution is asymptotically stable; and σ D 0 deﬁnes the stability boundary. To locate this boundary, we let σ D 0 in (11.5) and obtain the complex-valued characteristic equation i ω D α 1 1 α 1 β e i ω τ which, upon separating its real and imaginary parts, yields ω D α 1 β sin(ωτ) α 1 1 D α 1 β cos(ωτ) Hence, ωD

q

α 21 β 2 (α 1 1)2

(11.6)

11.1 A Scalar Equation

For real frequencies, the coefﬁcient under the radical must be positive; that is, α 21 β 2 (α 1 1)2 > 0. In what follows, we take β to be the bifurcation parameter and denote its critical value that separates stable from unstable trivial solutions by β c and the corresponding value of ω by ω c . It follows from (11.5) that α 1 [cos(ω c τ) α 1 β τ] d(σ C i ω) D 6D 0 Real dβ [cos(ω c τ) α 1 β τ]2 C sin2 (ω c τ) at (β, ω) D (β c , ω c ) and hence the bifurcation is a Hopf bifurcation. Next, we construct the normal form of this bifurcation by using the method of multiple scales in the next section and center-manifold reduction in Section 11.1.2. 11.1.1 The Method of Multiple Scales

We seek a uniform second-order approximate solution of (11.2) in the form x (tI ) D x1 (T0 , T2 ) C 2 x2 (T0 , T2 ) C 3 x3 (T0 , T2 ) C

(11.7)

where T0 D t, T2 D 2 t, and is a nondimensional bookkeeping parameter. The solution does not depend on the slow scale T1 D t because secular terms ﬁrst appear at O( 3 ). The derivative with respect to t is transformed into d @ @ C 2 C D D0 C 2 D2 C D dt d T0 @T2

(11.8)

Moreover, we express x (t τ) in terms of the scales T0 and T2 as 3 X

x (t τI ) D

m x m T0 τ, T2 2 τ C

mD1

which upon expansion for small becomes 3 X

x (t τI ) D

m x m (T0 τ, T2 ) 3 τ D2 x1 (T0 τ, T2 ) C

(11.9)

mD1

Next, we introduce the detuning parameter δ to describe the nearness of β to the critical value β c deﬁned by β D βc C 2 δ Substituting (11.7)–(11.9) into (11.2) and equating coefﬁcients of like powers of , we obtain D0 x1 C (1 α 1 )x1 C α 1 β c x1τ D 0

(11.10)

D0 x2 C (1 α 1 )x2 C α 1 β c x2τ D α 2 (x1 β c x1τ )2

(11.11)

289

290

11 Retarded Systems

D0 x3 C (1 α 1 )x3 C α 1 β c x3τ D D2 x1 α 1 δ x1τ C α 1 β c τ D2 x1τ 2α 2 (x1 β c x1τ ) (x2 β c x2τ ) α 3 (x1 β c x1τ )3

(11.12)

where x i D x i (T0 , T2 ) and x i τ D x i (T0 τ, T2 ). The general solution of (11.10) can be expressed as N 2 )e i ω c T0 C x1 D A(T2 )e i ω c T0 C A(T C AN m (T2 )e (σ m i ω m )T0

i

1 h X

A m (T2 )e (σ m Ci ω m )T0

mD1

(11.13)

where ω c is the critical frequency corresponding to σ D 0 on the stability boundary and it is given by (11.6); the σ ˙ i ω m are the remaining roots of (11.5); and the function A(T2 ) is determined by eliminating the secular terms at O( 3 ). Near the stability boundary, all of the eigenvalues have negative real parts except the eigenvalue corresponding to ω c whose real part changes sign as the stability boundary is crossed. Hence, as time increases all of the terms under the summation in (11.13) decay with time, leaving only the ﬁrst two terms. Therefore, the long-time behavior of the system is given by N 2 )e i ω c T0 x1 D A(T2 )e i ω c T0 C A(T

(11.14)

Substituting (11.14) into (11.11) yields D0 x2 C (1 α 1 )x2 C α 1 β c x2τ D α 2 γ 2 A2 e 2i ω c T0 α 2 γ γN A AN C cc (11.15) whose solution can be expressed as x2 D α 2 γ 2 Γ e 2i ω c T0 α 2 γ γN Γ0 A AN C cc

(11.16)

and γ D β c e i ω c τ 1 , Γ D

1 , 2i ω c C 1 α 1 C α 1 β c e 2i ω c τ

Γ0 D

1 1 α1 C α1 βc

(11.17)

Substituting (11.14) and (11.16) into (11.12) and eliminating the terms that lead to secular terms, we obtain the complex-valued form of the normal form of the bifurcation α 1 δ e i ω c τ A 1 α 1 β c τ e i ω c τ γ 2 γN 3α 3 2α 22 Γ 1 e 2ω c τ C 2(1 β c )Γ0 A2 AN C 1 α 1 β c τ e i ω c τ

A0 D

(11.18)

11.1 A Scalar Equation

11.1.2 Center-Manifold Reduction

We start by converting (11.2) into an operator differential equation. We note that, whereas the solution of (11.2) when τ D 0 depends on a single value, the solution of (11.2) with τ ¤ 0 depends on an entire set of values of x in the interval τ t 0. Hence, the initial space is a function space. Consequently, the delay differential equation 11.2 can be expressed as an abstract evolution equation (Campbell et al., 1995; Hale, 1977; Kuang, 1993) on the Banach space B of continuously differentiable complex functions from [τ, 0] to C 2 of complex functions with the norm jjp jj D maxτθ 0 jp (θ )j; that is, xP t D A x t C F(x t )

(11.19)

where x t (θ ) B is deﬁned by the shift operator x t (θ ) D x (t C θ ) for

τθ 0

The linear operator A is deﬁned by 8 < d p (θ ) for A p (θ ) D d θ :(α 1)p (0) α β p (τ) for 1 1 c

(11.20)

τθ 0 θ D0

and the operator F can be written as ( 0 for τ θ 0 FD f for θ D 0

(11.21)

(11.22)

where 2 3 f D 2 α 1 δ x (t τ) α 2 x (t) β c x (t τ) α 3 x (t) β c x (t τ) (11.23) To carry out the analysis, we need an inner product and the adjoint operator associated with (11.21). In contrast with C 2 , the space B does not have a natural inner product associated with its norm. However, following Hale (1977), we introduce the following bilinear form that acts like an “inner product”: Z0 hq, p i D qN (0)p (0) C

qN (ξ C τ) (α 1 β c )p (ξ )d ξ

(11.24)

τ

where q 2 B , the space of continuously differentiable functions from [0, τ] to C 2 with the norm jjqjj D max0θ τ jq(θ )j. With this bilinear form, one can construct the following formal adjoint operator (Hale, 1977) associated with (11.21): 8 0, the trivial solution is unstable; when σ < 0, the trivial solution is asymptotically stable; and σ D 0 separates stable from unstable trivial solutions. Putting σ D 0 in (11.133) and separating real and imaginary parts, we obtain ω D α β sin(ωτ) α 1 D α β cos(ωτ)

305

306

11 Retarded Systems

Hence, for given α and τ, the critical value ω c of ω is given by ω c D (α 1 1) tan(ω c τ)

(11.135)

Differentiating (11.133) with respect to σ C i ω and β and evaluating the result at the critical values yields d(σ C i ω) i ωc C 1 α D dβ β c (i ω c C 1 α) τ whose real part is different from zero. Therefore, the trivial solution loses stability as a result of a pair of complex conjugate eigenvalues transversely crossing the imaginary axis as β exceeds the critical value β c , and hence the bifurcation is a Hopf bifurcation. Next, we construct the normal form of this bifurcation by using the method of multiple scales in the following section and by using center-manifold reduction in Section 11.3.2. 11.3.1 The Method of Multiple Scales

Because the nonlinearity is cubic, we seek a uniform second-order approximate solution of (11.128)–(11.130) in powers of 1/2 instead of powers of as done above for cases with quadratic and cubic nonlinearities. Thus, we let x(tI ) D 1/2 x 1 (T0 , T1 ) C 3/2 x 2 (T0 , T1 ) C

(11.136)

where T0 D t, T1 D t, and is a nondimensional bookkeeping parameter. The derivative with respect to t, in this case, is transformed into @ d @ C C D D0 C D1 C D dt d T0 @T1

(11.137)

Moreover, we express x(t τ) in terms of the scales T0 and T1 as x(t τI ) D 1/2 x 1 (T0 τ, T1 τ) C 3/2 x 2 (T0 τ, T1 τ) C which upon expansion for small becomes x(t τI ) D 1/2 x 1 (T0 τ, T1 ) C 3/2 x 2 (T0 τ, T1 ) 3/2 τ D1 x 1 (T0 τ, T1 ) C

(11.138)

Next, we introduce the detuning parameter δ to describe the nearness of β to the critical value β c deﬁned by β D β c C δ

(11.139)

11.3 A Three-Dimensional System

Substituting (11.136)–(11.139) into (11.128) and equating coefﬁcients of like powers of yields D0 x 1 (T0 , T1 ) Lx 1 (T0 , T1 ) C β c R x 1 (T0 τ, T1 ) D 0

(11.140)

D0 x 2 (T0 , T1 ) Lx 2 (T0 , T1 ) C β c R x 2 (T0 τ, T1 ) D δ R x 1 (T0 τ, T1 ) D1 x 1 (T0 , T1 ) C β c τ R D1 x 1 (T0 τ, T1 ) C f x 1 (T0 , T1 ), x 1 (T0 τ, T1 )

(11.141)

The nondecaying solution of (11.140) can be expressed as N 1 )Nc e i ω c T0 x 1 (T0 , T1 ) D A(T1 )c e i ω c T0 C A(T

(11.142)

where c is given by 2 3 1 c D 4c25 , c3

α 2 1 β c e i ω c τ c2 D , 1 C i ωc

and

c3 D

α1

1 C i ωc 1 β c e i ω c τ (11.143)

Substituting (11.142) into (11.141) yields D0 x 2 (T0 , T1 ) Lx 2 (T0 , T1 ) C β c R x 2 (T0 τ, T1 ) D δ R cAe i ω c τ e i ω c T0 I β c τ R e i ω c τ c A0 e i ω c T0 3γ 2 γN A2 AN Of e i ω c T0 C cc C NST (11.144) where 3 α 1 c 23 cN3 Of D 4 α 2 5 α 3 c 22 cN2 2

γ D β c e i ω c τ 1 ,

(11.145)

Because the homogeneous part of (11.144) has nontrivial solutions, the nonhomogeneous equation has a solution only if a solvability condition is satisﬁed. To determine this solvability condition, we seek a particular solution of (11.144) in the form x 2 (T0 , T1 ) D φ(T1 )e i ω c T0 and obtain L β c R e i ω c τ i ω c I φ

D I β c τ R e i ω c τ c A0 C δ R cAe i ω c τ C 3γ 2 γN Of A2 AN

(11.146)

We note that the problem of ﬁnding the solvability condition for the system of differential equations (11.144) has been transformed into ﬁnding the solvability

307

308

11 Retarded Systems

condition of the system of algebraic equations (11.146). Again, because i ω c is an eigenvalue of the homogeneous part of (11.146), the nonhomogeneous equation has solutions if and only if a solvability condition is satisﬁed. This condition demands that the right-hand side of (11.146) be orthogonal to every solution of the adjoint homogeneous problem. In this case, the adjoint is governed by L β c R e i ωc τ C i ωc I b D 0 (11.147) We note that b is not unique, and to make it unique we impose the condition b .c D 1

(11.148)

Solving (11.147) and (11.148) and using (11.145), we obtain the unique adjoint

bD

1 3

2 3 1 4 b 25 , b3

b2 D

1 i ωc , α 2 (1 β c e i ω c τ )

and

b3 D

α 1 (1 β c e i ω c τ ) 1 i ωc (11.149)

Taking the “inner product” of the right-hand side of (11.146) with b yields the solvability condition, normal form, A0 D δ Λ 1 A C Λ 2 A2 AN where

(11.150)

α 1 c 2 c 23 C α 2 c 3 C α 3 c 22 e i ω c τ 3c 2 c 3 β c τ α 1 c 2 c 23 C α 2 c 3 C α 3 c 22 3 α 1 c 2 c 33 cN3 C α 2 c 3 C α 3 c 32 cN 2 γ 2 γN Λ2 D 3c 2 c 3 β c τ α 1 c 2 c 23 C α 2 c 3 C α 3 c 22

Λ1 D

(11.151)

11.3.2 Center-Manifold Reduction

Again, we start by representing (11.128) as an operator differential equation; that is, xP t D A x t C F(x t )

(11.152)

where x t (θ ) B is deﬁned by the shift operator x t (θ ) D x(t C θ ) for

τθ 0

the linear operator A is deﬁned by 8 < d p (θ ) for A p (θ ) D d θ : L p(0) β R p (τ) for c

τθ 0 θ D0

(11.153)

11.3 A Three-Dimensional System

and the operator F can be written as ( 0 for τ θ 0 FD f δ R x(t τ) for θ D 0 The adjoint operator associated with (11.153) is deﬁned by 8 0. Determine the range of gain k and time delay τ for which the limit cycles are quenched. 11.5.3 Consider the system uR C ω 2 u C 2µ uP C α u3 C k u(t τ) D f cos Ω t Determine the resonance frequency for the linear system. Then, determine an approximation to the nonlinear system response for values of Ω near the resonance frequency.

313

315

References Abed, E.H. (1994) Bifurcation–theoretic issues in the control of voltage collapse, in Proceedings of the IMA Workshop on System and Control Theory for Power Systems (eds J.H. Chow, P.V. Kokotovic, and R.J. Thomas), Springer-Verlag, New York. Andronov, A.A., Vitt, E.A., and Khaiken, S.E. (1966) Theory of Oscillations, Pergamon Press, Oxford. Arnold, V.I. (1978) Mathematical Methods of Classical Mechanics, Springer-Verlag, Berlin. Arnold, V.I. (1983) Geometrical Methods in the Theory of Ordinary Differential Equations, Springer-Verlag, New York. Arnold, V.I. (1987) Ordinary Differential Equations, MIT Press, Cambridge, Massachusetts. Arnold, V.I. (1988) Dynamical Systems, Springer-Verlag, New York. Babcock, K.L. and Westervelt, R.M. (1987) Dynamics of simple electronic neural networks. Physica D, 28, 305–316. Bogoliubov, N.N. and Mitropolsky, Y.A. (1961) Asymptotic Methods in the Theory of Nonlinear Oscillations, Gordon & Breach, New York. Campbell, S.A., Bélair, J., Ohira, T., and Milton, J. (1995) Limit cycles, tori and complex dynamics in a second order differential equation with delayed negative feedback. J. Dyn. Diff. Eqs., 7, 213–236. Carr, J. (1981) Applications of Centre Manifold Theory, Springer-Verlag, New York. Coddington, E.A. and Levinson, N. (1955) Theory of Ordinary Differential Equations, McGraw-Hill, New York.

Cole, D. (1951) On a quasilinear parabolic equation occurring in aerodynamics. Q. Appl. Math., 9, 225–236. Drazin, P.G. (1992) Nonlinear Systems, Cambridge University Press, Cambridge, England. Dumortier, F. (1977) Singularities of vector ﬁelds on the plane, J. Diff. Eqs., 23, 53–106. Fallside, F. and Patel, M.R. (1965) Stepresponse behaviour of a speed-control system with a back-e.m.f. nonlinearity. Proc. IEE, 112, 1979–1984. Gilsinn, D. (2002) Estimating critical Hopf bifurcation parameters for a second-order delay differential equation with application to machine tool. Nonlinear Dyn., 30, 103– 154. Golubitsky, M. and Schaeffer, D.G. (1985) Singularities and Groups in Bifurcation Theory, Vol. 1, Springer-Verlag, New York. Gopalsamy, K. and Leung, I. (1996) Delay induced periodicity in a neural network of excitation and inhibition. Physica D, 89, 395–426. Guckenheimer, J. and Holmes, P.J. (1983) Nonlinear Oscillations, Dynamical Systems and Bifurcation of Vectorﬁelds, SpringerVerlag, New York. Hale, J.K. (1977) Theory of Functional Differential Equations, Applied Mathematical Sciences, Vol. 3, Springer-Verlag, New York. Hale, J. (1980) Ordinary Differential Equations, Krieger, Malabar, Florida. Hanna, N.H. and Tobias, S.A. (1974) A theory of nonlinear regenerative chatter. ASME J. Eng. Indust., 96, 247–255. Hirsch, M. W. and Smale, S. (1974) Differential Equations, Dynamical Systems, and Linear Algebra, Academic Press, New York.

The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

316

References Hopf, E. (1950) The partial differential equation u t C uu x D µu x x . Comm. Pure Appl. Math., 3, 201–230. Hopﬁeld, J. (1984) Neurons with graded response have collective computational properties like those of two-state neurons. Proc. Nat. Acad. Sci. USA, 81, 3088–3092. Iooss, G. (1979) Bifurcation of Maps and Applications, North Holland, Amsterdam. Kalmár-Nagy, T., Stépán, G., and Moon, F.C. (2001) Subcritical Hopf bifurcation in the delay equation model for machine tool vibrations. Nonlinear Dyn., 26, 121–142. Kamel, A.A. (1970) Perturbation method in the theory of nonlinear oscillations. Celestial Mech. Dyn. Astron., 3, 90–106. Kuang, Y. (1993) Delay Differential Equations with Applications in Population Dynamics, Academic Press, Boston, Massachusetts. Liao, X., Guo, S., and Li, C. (2007) Stability and bifurcation analysis in tri-neuron model with time delay. Nonlinear Dyn., 49, 319–345. Liao, X., Wong, K.W., and Wu, Z. (2001) Bifurcation analysis on a two-neuron system with distributed delays. Physica D, 149, 123–141. Lorenz, E.N. (1963) Deterministic nonperiodic ﬂow. J. Atmos. Sci., 20, 130–141. Marsden, J.E. and McCracken, M. (1976) The Hopf Bifurcation and Its Applications, Springer-Verlag, New York. Mingori, D.L. and Harrison, J.A. (1974) Circularly constrained particle motion in spinning and coning bodies. AIAA Journal, 12, 1553–1558. Namachchivaya, N.S. and Malhotra, N. (1992) Parametrically excited Hopf bifurcation with non-semisimple 1 : 1 resonance, in Nonlinear Vibrations (eds R.A. Ibrahim, N.S. Namachchivaya, and A.K. Bajaj), ASME DE, 50, 29–46. Nayfeh, A.H. (1968) Forced oscillations of van der Pol oscillator with delayed amplitude limiting. IEEE Trans., CT–15, 192–200. Nayfeh, A.H. (1973) Perturbation Methods, Wiley-Interscience, New York. Nayfeh, A.H. (1981) Introduction to Perturbation Techniques, Wiley-Interscience, New York. Nayfeh, A.H. (1984) The response of single degree of freedom systems with quadratic

and cubic non-linearities to a subharmonic excitation. J. Sound Vib., 89, 457–470. Nayfeh, A.H. (1985) Problems in Perturbation, Wiley-Interscience, New York. Nayfeh, A.H. (1986) Perturbation methods in nonlinear dynamics, in Nonlinear Dynamics Aspects of Particle Accelerators, Lecture Notes in Physics, No. 247, Springer-Verlag, New York, 238–314. Nayfeh, A.H. (2008) Order reduction of retarded nonlinear systems – the method of multiple scales versus center-manifold reduction. Nonlinear Dyn., 51, 483–500. Nayfeh, A.H. and Balachandran, B. (1995) Applied Nonlinear Dynamics, WileyInterscience, New York. Nayfeh, A.H., Chin, C.M., and Pratt, J. (1997) Perturbation methods in nonlinear dynamics – Applications to machining dynamics. J. Manuf. Sci. Eng., 119, 485–493. Nayfeh, A.H. and Jebril, A.E.S. (1987) The response of two-degree-of-freedom systems with quadratic and cubic non-linearities to multifrequency parametric excitations. J. Sound Vib., 115, 83–101. Nayfeh, A.H. and Mook, D.T. (1979) Nonlinear Oscillations, Wiley-Interscience, New York. Nayfeh, N.A. (2006) Local and Global Stability and Dynamics of a Class of Nonlinear TimeDelayed One-Degree-of-Freedom Systems, PhD Dissertation, Virginia Polytechnic Institute and State University, Blacksburg, Virginia. Nayfeh, T.A., Asrar, W., and Nayfeh, A.H. (1992) Three-mode interactions in harmonically excited systems with quadratic nonlinearities, Nonlinear Dyn., 3, 385–410. Poincaré, H. (1899) Les Méthodes Nouvelles de la Mécanique Céleste, 3 Vols., Gauthier-Villars, Paris. Rössler, O.E. (1976a) An equation for continuous chaos. Phys. Lett. A, 57, 397–398. Shaw, S.W. and Rand, R.H. (1989) The transition to chaos in a simple mechanical system. Int. J. Nonlinear Mech., 24, 41–56. Tezak, E.G., Nayfeh, A.H., and Mook, D.T. (1982) Parametrically excited nonlinear multi-degree-of-freedom systems with repeated frequencies. J. Sound Vib., 85, 459– 472. Thom, R. (1975) Structural Stability and Morphogenesis, Benjamin/Cummings, Menlo Park, California.

References Van Dyke, M. (1964) Perturbation Methods in Fluid Mechanics, Academic Press, New York. Annotated version (1975), Parabolic Press, Palo Alto, California. Walter, W. (1976) Gewohnliche Differentialgleichungen, Springer-Verlag, New York. Whitham, G.B. (1974) Linear and Nonlinear Waves, Wiley-Interscience, New York.

Wiggins, S. (1988) Global Bifurcations and Chaos: Analytical Methods, Springer-Verlag, New York. Wiggins, S. (1990) Introduction to Applied Nonlinear Dynamical Systems and Chaos, Springer-Verlag, New York.

317

319

Further Reading Abraham, R.H. and Shaw, C.D. (1982) Dynamics – The Geometry of Behavior, Vols. 1–4, Aerial Press, Santa Cruz, California. Ariaratnam, S.T. and Namachchivaya, S.N. (1984) Degenerate Hopf bifurcation, in Proceedings of the IEEE International Symposium on Circuits and Systems, Vol. 3, Montreal, Canada, pp. 1343–1348. Arrowsmith, D.K. and Place, C.M. (1982) Ordinary Differential Equations, Chapman and Hall, New York. Arrowsmith, D.K. and Place, C.M. (1990) An Introduction to Dynamical Systems, Cambridge University Press, Cambridge, United Kingdom. Ashkenazi, M. and Chow, S.-N. (1988) Normal forms near critical points for differential equations and maps. IEEE Trans. Circuits and Sys., 35, 850–862. Baider, A. (1989) Unique normal forms for vector ﬁelds and Hamiltonians. J. Diff. Eqs. 78, 33–52. Baider, A. and Churchill, R.C. (1988) Uniqueness and non-uniqueness of normal forms for vector ﬁelds. Proc. R. Soc. Edinburgh Sec. A, 108, 27–33. Bajaj, A.K. (1986) Resonant parametric perturbations of the Hopf bifurcation. J. Math. Anal. Appl., 115, 214–224. Baker, G.L. and Gollub, J.P. (1990) Chaotic Dynamics, Cambridge University Press, Cambridge, United Kingdom. van der Beek, C.G.A. (1989) Normal forms and periodic solutions in the theory of nonlinear oscillations. Existence and asymptotic theory. Int. J. Nonlinear Mech., 24, 263– 279. van der Beek, C.G.A. and van der Burgh, A.H.P. (1987) On the periodic windinduced vibrations of an oscillator with two

degrees of freedom. Nieuw Arch. Wisk, 5, 207–225. Belitskii, G.R. (1978) Equivalence and normal forms of germs of smooth mappings. Russ. Math. Surv., 33, 107–177. Belitskii, G.R. (1979) Normal forms with respect to the ﬁltering action of a group. Russ. Math. Surv., 40, 3–46. Belitskii, G.R. (1986) The smooth equivalence of germs of vector ﬁelds with one zero or a pair of pure imaginary eigenvalues. Funct. Anal. Appl., 20, 253–259. Bensoussan, A. (1988) Perturbation Methods in Optimal Control, Wiley-Interscience, New York. Bibikov, Y.N. (1971) On the reducibility of a system of two different equations to normal form. J. Diff. Eqs., 7, 1438–1441. Birkhoff, G.D. (1927) Dynamical Systems, American Mathematics Society Publications, Vol. 9, Providence, Rhode Island. Birkhoff, G.D. (1935) Nouvelles Recherches sur les systémes dynamiques. Mem. Pont. Acad. Sci. Novi. Lynacaei, 1, 85–216. Birkhoff, G.D. (1975) Versal deformations of a singular point on the plane in the case of zero eigenvalues. Funct. Anal. Appl., 9, 144–145. Bogdanov, R.I. (1976) Reduction to orbital normal form of a vector ﬁeld on the plane. Funct. Anal. Appl., 10, 61–62. Bogdanov, R.I. (1981) Versal deformation of a singularity of a vector ﬁeld on the plane in the case of zero eigenvalues. Sel. Math. Sov., 1, 389–421. Braun, M. (1983) Differential Equations and Their Applications, Springer-Verlag, New York.

The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

320

Further Reading Bruno, A.D. (1971) Analytical form of differential equations, I. Trans. Moscow Math. Soc., 25, 132–198. Bruno, A.D. (1972a) Analytical forms of differential equations, II. Trans. Moscow Math. Soc., 25, 199–299. Bruno, A.D. (1972b) Local methods in nonlinear resonances. Arch. Math. (Brno), 4, 177– 178 (Russian). Bruno, A.D. (1973) Local invariants of differential equations. Math. Notes, 14, 844–848. Bruno, A.D. (1974) The normal form of differential equations with a small parameter. Math. Notes, 16, 832–836. Bruno, A.D. (1975) The normal form of real differential equations. Math. Notes, 18, 722–731. Bruno, A.D. (1976) The normal form and averaging methods. Sov. Math. Dokl., 17, 1268– 1273. Bruno, A.D. (1979) Normal forms in perturbation theory, in Proc. VIII Int. Conf. Nonlinear Oscil., 1, 177–182, Prague, Czechoslovakia. Bruno, A.D. (1989) Local Methods in Nonlinear Differential Equations, Springer-Verlag, New York. Caprino, S., Maffei, C., and Negrini, P. (1984) Hopf bifurcation with 1 : 1 resonance. Nonlinear Anal. Theor. Math. Appl., 8, 1011– 1032. Chen, K.T. (1963) Equivalence and decomposition of vector ﬁelds about an elementary critical point. Am. J. Math., 85, 693–722. Chen, K.T. (1965) Local diffeomorphisms: C 1 realizations of formal properties. Am. J. Math., 87, 140–157. Chow, S.-N. and Hale, J.K. (1982) Methods of Bifurcation Theory, Springer-Verlag, New York. Chow, S.-N. and Wang, D. (1986) Normal forms of bifurcating periodic orbits, in Multi-Parameter Bifurcation Theory (eds M. Golubitsky and J. Guckenheimer), 9–18. Coullet, P. and Spiegel, E.A. (1983) Amplitude equations for systems with competing instabilities. SIAM J. Appl. Math., 43, 774– 819. Chua, L.O. and Lin, P.M. (1975) ComputerAided Analysis of Electronic Circuits, Prentice-Hall, Englewood Cliffs, New Jersey. Cushman, R. (1984) Normal form for Hamiltonian vector ﬁelds with periodic ﬂows,

in Differential Geometric Methods in Mathematical Physics (ed. S. Sternberg), Reidel, Dordrecht, The Netherlands, pp. 125–144. Cushman, R., Deprit, A., and Mosak, R. (1983) Normal form and representation theory. J. Math. Phys., 24, 2103–2116. Cushman, R. and Sanders, J.A. (1986) Nilpotent normal forms and representation theory of sl(2,R), in Multi-Parameter Bifurcation Theory (eds M. Golubitsky and J. Guckenheimer), pp. 31–51. Cvitanovic, P. (1984) Universality in Chaos, Adam Hilger, New York. Deprit, A. (1969) Canonical transformations depending on a small parameter. Celestial Mech. Dyn. Astron., 1, 12–32. Deprit, A. (1982) Delaunay normalizations. Celestial Mech. Dyn. Astron., 26, 9–21. Deprit, A., Henrard, J., Price, F., and Rom, A. (1969) Birkhoff’s normalization. Celestial Mech. Dyn. Astron., 1, 225–251. Devaney, R.L. (1985) An Introduction to Chaotic Dynamical Systems, Benjamin/Cummings, Menlo Park, California. Dowell, E. (1989) A Modern Course in Aeroelasticity, Kluwer, Dordrecht, The Netherlands. Dowell, E.H. and Ilgamova, M. (1988) Studies in Nonlinear Aeroelasticity, Springer-Verlag, New York. Duistermaat, J.J. (1984) Bifurcation of periodic solutions near equilibrium points of Hamiltonian systems, in Bifurcation Theory and Applications, Lecture Notes in Mathematics, Springer-Verlag, New York, Vol. 1057, pp. 57–105. Elphick, C., Tirapegui, E., Brachet, M.E., Coullet, P., and Iooss, G. (1987) A simple global characterization for normal forms of singular vector ﬁelds. Physica D, 29, 95–127. Evan-Iwanowski, R.M. (1976) Resonance Oscillations in Mechanical Systems, Elsevier, New York. Feng, Z.C. and Sethna, P.R. (1990) Global bifurcation and chaos in parametrically forced systems with one-one resonance. Dyn. Stab. Sys., 5, 201–225. Franks, J.M. (1982) Homology and dynamical systems, in CBMS Regional Conference Series in Mathematics, American Mathematics Society, Providence, Rhode Island, 49. Galin, D.M. (1982) Versal deformations of linear Hamiltonian systems. Am. Math. Soc. Trans., 118, 1–12.

Further Reading Gamero, E., Freire, E., and Ponce, E. (1991) Normal forms for planar systems with nilpotent linear part. Int. Series Num. Math., 97, 123–127. Gaponov-Grekhov, A.V. and Rabinovich, M.I. (1991) Nonlinear Physics, Oscillations, Chaos, Structures, Springer-Verlag, New York. van Gils, S.A., Krupa, M., and Langford, W. (1990) Hopf bifurcation with nonsemisimple resonance. Nonlinearity, 3, 825–850. Giorgilli, A. and Galgani, L. (1978) Formal integrals for an autonomous Hamiltonian system near an equilibrium. Celestial Mech. Dyn. Astron., 17, 267–280. Golubitsky, M. and Langford, W.F. (1981) Classiﬁcation and unfoldings of degenerate Hopf bifurcations. J. Diff. Eqs., 41, 375– 415. Golubitsky, M., Stewart, I., and Schaeffer, D.G. (1988) Singularities and Groups in Bifurcation Theory, Vol. 2, Springer-Verlag, New York. Hagedorn, P. (1981) Non-Linear Oscillations, Clarendon, Oxford, United Kingdom. Haken, H. (1983) Advanced Synergetics, Springer-Verlag, New York. Hale, J.K. (1963) Oscillations in Nonlinear Systems, McGraw-Hill, New York. Hartman, P. (1963) On the local linearization of differential equations. Proc. Am. Math. Soc., 14, 568–573. Hassard, B.D., Kazarinoff, N.D., and Wan, Y.-H. (1980) Theory and Applications of the Hopf Bifurcation, Cambridge University Press, Cambridge, United Kingdom. Hassard, B. and Wan, Y.H. (1978) Bifurcation formulae derived from center manifold theory. J. Math. Anal. Appl., 63, 297– 312. Hayachi, C. (1964) Nonlinear Oscillations in Physical Systems, McGraw-Hill, New York. Hinch, E.J. (1991) Perturbation Methods, Cambridge University Press, Cambridge, United Kingdom. Hirsch, M.W. (1976) Differential Topology, Springer-Verlag, New York. Hirsch, M.W., Pugh, C.C., and Shub, M. (1977) Invariant manifolds, in Springer Lecture Notes in Mathematics, Springer-Verlag, New York, p. 583. Holmes, P.J. (1981) Center manifolds, normal forms, and bifurcation of vector ﬁelds with

application to coupling between periodic and steady motions. Physica D, 2, 449–481. Hsu, L. and Favretto, F. (1984) Recursive bifurcation formulae for normal form and center manifold theory. J. Math. Anal. Appl., 101, 562–574. Hubbard, J. and West, B. (1991) Differential Equations – A Dynamical Systems Approach Part I: Ordinary Differential Equations, Springer-Verlag, New York. Ibrahim, R.A. (1985) Parametric Random Vibration, Wiley-Interscience, New York. Iooss, G. and Joseph, D.D. (1980) Elementary Bifurcation and Stability Theory, SpringerVerlag, New York. Jackson, E.A. (1989) Perspectives of Nonlinear Dynamics, Vols. I and II, Cambridge University Press, Cambridge, United Kingdom. Jezequel, L. and Lamarque, C.H. (1994) Analysis of non-linear dynamical systems by the normal form theory. J. Sound Vib., 149, 429–459. Jordan, D.W. and Smith, P. (1977) Nonlinear Ordinary Differential Equations, Oxford University Press, Oxford, United Kingdom. Kahn, P.B. (1990) Mathematical Methods for Scientists and Engineers: Linear and Nonlinear Systems, Wiley-Interscience, New York. Kevorkian, J. and Cole, J.D. (1981) Perturbation Methods in Applied Mechanics, SpringerVerlag, New York. Kim, J.H. and Stringer, J. (1992) Applied Chaos, Wiley-Interscience, New York. Kocak, H. (1984) Normal forms and versal deformations of linear Hamiltonian systems. J. Diff. Eqs., 51, 359–407. Kubicek, M. and Hlavacek, V. (1983) Numerical Solution of Nonlinear Boundary Value Problems with Applications, Prentice-Hall, Englewood Cliffs, New Jersey. Kubicek, M. and Marek, M. (1983) Computational Methods in Bifurcation Theory and Dissipative Structures, Springer-Verlag, New York. Kukulin, V.L., Krasnopolsky, V.M., and Horacek, J. (1989) Theory of Resonances: Principles and Applications, Kluwer, Dordrecht, The Netherlands. Kummer, M. (1971) How to avoid “secular” terms in classical and quantum mechanics. Nuovo Cimento B, 123–148.

321

322

Further Reading Lam, L. and Morris, H.C. (1990) Nonlinear Structures in Physical Systems Pattern Formulations, Chaos and Waves, SpringerVerlag, New York. Langford, W.F. (1979) Periodic and steadystate mode interactions leading to tori. SIAM J. Appl. Math., 37, 22–48. Lictenberg, A.J. and Lieberman, M.A. (1983) Regular and Stochastic Motion, SpringerVerlag, New York. Marchenko, V.A. (1987) Nonlinear Equations and Operator Algebras, Kluwer, Dordrecht, The Netherlands. Marek, M. and Schreiber, I. (1991) Chaotic Behavior of Deterministic Dissipative Systems, Cambridge University Press, Cambridge, United Kingdom. van der Meer, J.C. (1982) Nonsemisimple 1 : 1 resonance at an equilibrium. Celestial Mech. Dyn. Astron., 27, 131–149. van der Meer, J.C. (1986) The Hamiltonian Hopf Bifurcation, Springer-Verlag, New York. Meyer, K.R. (1974) Normal forms for Hamiltonian systems. Celestial Mech. Dyn. Astron., 9, 517–522. Meyer, K.R. and Schmidt, D.S. (1977) Entrainment domans. Funkcialaj Ekvacioj, 20, 171– 192. Mikhailov, A.S. (1994) Foundations of Synergetics I Distributed Active Systems, SpringerVerlag, New York. Minorsky, N. (1947) Introduction to NonLinear Mechanics, J.W. Edwards, Ann Arbor, Michigan. Moon, F.C. (1987) Chaotic Vibrations: An Introduction for Applied Scientists and Engineers, Wiley-Interscience, New York. Moon, F.C. (1992) Chaotic and Fractal Dynamics, Wiley-Interscience, New York. Morse, M. and Hedlund, G.A. (1938) Symbolic dynamics. Am. J. Math., 60, 815–866. Moser, J. (1973) Stable and Random Motion in Dynamical Systems, Hermann Weyl Lectures, Princeton, New Jersey. Murdock, J.A. (1991) Perturbations Theory and Methods, Wiley-Interscience, New York. Murdock, J. and Robinson, C. (1980) Qualitative dynamics from asymptotic expansions: Local theory. J. Diff. Eqs., 36, 425–441. Namachchivaya, N.S. (1989) Non-linear dynamics of supported pipe conveying pulsating ﬂuid-I. Subharmonic resonance. Int. J. Nonlinear Mech., 24, 185–196.

Namachchivaya, N.S. and Ariaratnam, S.T. (1986) Periodically perturbed non-linear gyroscopic systems, J. Struct. Mech., 14, 153–175. Namachchivaya, N.S. and Ariaratnam, S.T. (1987) Periodically perturbed Hopf bifurcation. SIAM J. Appl. Math., 47, 15–39. Namachchivaya, N.S. and Van Roessel, H.J. (1986) Unfolding of degenerate Hopf bifurcation for supersonic ﬂow past a pitching wedge. J. Guid. Cont. Dyn., 9, 413–418. Nayfeh, A.H. and Chin, C.-M. (1999a) Perturbation Methods with Mathematica, Dynamics Press, Virginia. http://www.esm.vt.edu/~anayfeh/ Nayfeh, A.H. and Chin, C.-M. (1999b) Perturbation Methods with Maple, Dynamics Press, Virginia. http://www.esm.vt.edu/~anayfeh/ Nitecki, Z. (1971) Differentiable Dynamics, MIT Press, Cambridge, Massachusetts. Palis, J. and de Melo, W. (1982) Geometric Theory of Dynamical Systems: An Introduction, Springer-Verlag, New York. Parker, T.S. and Chua, L.O. (1989) Practical Numerical Algorithms for Chaotic Systems, Springer-Verlag, New York. Peixoto, M. (1973) Dynamical Systems, Academic Press, New York. Perko, L.M. (1991) Differential Equations and Dynamical Systems, Springer-Verlag, New York. Poincaré, H. (1892) Les Méthodes Nouvelles de la Mécanique Céleste, Vol. 1, Gauthier-Villars, Paris. Poincaré, H. (1929) Sur les propriétés des functions déﬁnies par les équations aux différences partielles, in Oeuvres, GauthierVillars, Paris. Poston, T. and Stewart, I. (1978) Catastrophe Theory and Its Applications, Pitman, Aulander, North Carolina. Rabinowitz, M.I. (1987) Periodic Solutions of Hamiltonian Systems and Related Topics, Kluwer, Dordrecht, The Netherlands. Rabinowitz, M.I. and Trubetskov, D.I. (1989) Oscillations and Waves in Linear and Nonlinear Systems, Kluwer, Dordrecht, The Netherlands. Rand, R.H. (1984) Computer Algebra in Applied Mathematics: An Introduction to Macsyma, Research Notes in Mathematics, Pitman, Aulander, North Carolina.

Further Reading Rand, R.H. and Armbruster, D. (1987) Perturbation Methods, Bifurcation Theory, and Computer Algebra, Springer-Verlag, New York. Rasband, S.N. (1990) Chaotic Dynamics of Nonlinear Systems, Wiley-Interscience, New York. Rössler, O.E. (1976b) Different types of chaos in two simple differential equations. Z. Naturforsch., 31a, 1664–1670. Ruelle, D. (1989) Elements of Differential Dynamics and Bifurcation Theory, Academic Press, New York. Samovol, V.S. (1972) Linearization of a system of differential equations in the neighborhood of a singular point. Sov. Math. Dokl., 13, 1255–1259. Samovol, V.S. (1979) Linearization of systems of differential equations in a neighborhood of a singular point. Trans. Moscow Math. Soc., 38, 190–225. Samovol, V.S. (1982) Equivalence of systems of differential equations in a neighborhood of a singular point. Trans. Moscow Math. Soc., 44, 217–237. Sanders, J. and Verhulst, F. (1985) Averaging Methods in Nonlinear Dynamical Systems, Springer-Verlag, New York. Schmidt, G. and Tondl, A. (1986) Non-Linear Vibrations, Cambridge University Press, Cambridge, United Kingdom. Schuster, H. (1984) Deterministic Chaos, Physik Velag, Weinheim, Germany. Sell, G.R. (1971) Topological Dynamics and Differential Equations, Van Nostrand Reinhold, London, United Kingdom. Sell, G.R. (1978) The structure of a ﬂow in the vicinity of an almost periodic motion. J. Diff. Eqs., 27, 359–393. Sell, G.R. (1979) Bifurcation of higher dimensional tori. Arch. Rat. Mech. Anal., 69, 199– 230. Seydel, R. (1988) From Equilibrium to Chaos – Practical Bifurcation and Stability Analysis, Elsevier, New York. Shaw, S.W. and Pierre, C. (1992) On nonlinear normal mode, in Nonlinear Vibrations (eds R.A. Ibrahim, N.S. Namachchivaya, and A.K. Bajaj), ASME DE, Vol. 50, pp. 1– 5. Shub, M. (1987) Global Stability of Dynamical Systems, Springer-Verlag, New York. Siegel, C.L. (1952) Über die Normalform analytischer Differentialgleichungen in der

Nähe einer Gleichgewichtslösung. Nachr. Akad. Wiss. Goettingen Math. Phys. Kl. Math. Phys. Chem. Abt, 21–30. Sparrow, C. (1982) The Lorenz Equations: Bifurcations, Chaos, and Strange Attractors, Springer-Verlag, New York. Starzhinskii, V.M. (1972) Normal forms of the fourth-order for nonlinear oscillations. Mech. Solids, 7, 1–7. Starzhinskii, V.M. (1973) A practical method for computing normal forms in nonlinear oscillation problems. J. Appl. Math. Mech., 37, 389–396. Starzhinskii, V.M. (1974) On oscillations in third-order systems when the linear approximation is neutral. Mech. Solids, 9, 18– 24. Starzhinskii, V.M. (1980) Applied Methods in the Theory of Nonlinear Oscillations, MIR Publishers, Moscow, Russia. Stauffer, D. and Stanley, H.E. (1990) From Newton to Mandelbrot – A Primer in Modern Theoretical Physics, Springer-Verlag, New York. Sternberg, S. (1958) On the structure of local homeomorphisms of Euclidean n-space. Am. J. Math., 80, 623–631. Sternberg, S. (1959) The structure of local homeomorphisms. Am. J. Math., 81, 578– 604. Sternberg, S. (1961) Finite Lie groups and the formal aspects of dynamical systems. J. Appl. Math. Mech., 10, 451–478. Szemplinska-Stupnicka, W. (1990) The Behavior of Nonlinear Vibrating Systems: Vol. I: Fundamental Concepts and Methods; Applications to Single Degree-of-Freedom Systems, Kluwer, Dordrecht, The Netherlands. Szemplinska-Stupnicka, W. (1990) The Behavior of Nonlinear Vibrating Systems: Vol. II: Advanced Concepts and Applications to Multi-Degree-of-Freedom Systems, Kluwer, Dordrecht, The Netherlands. Tabor, M. (1989) Chaos and Integrability in Nonlinear Dynamics: An Introduction, Wiley-Interscience, New York. Takens, F. (1974) Singularities of vector ﬁelds. Publ. Math. I.H.E.S., 43, 47–100. Takens, F. (1979) Forced oscillations and bifurcations. Comm. Math. Inst., 3, 1–59. Thompson, J.M.T. and Stewart, H.B. (1986) Nonlinear Dynamics and Chaos: Geometrical Methods for Engineers and Scientists, WileyInterscience, New York.

323

324

Further Reading Tokarev, S.P. (1977) Smooth equivalence of differential equation systems in the plane in the case of a noncoarse focus. J. Diff. Eqs., 613–617. Troger, H. and Steindl, A. (1991) Nonlinear Stability and Bifurcation Theory, SpringerVerlag, New York. Urabe, U. (1967) Nonlinear Autonomous Oscillations, Academic Press, New York. Ushiki, S. (1984) Normal forms for singularities of vector ﬁelds. Japan J. Appl. Math., 1, 1–37.

Verhulst, F. (1990) Nonlinear Differential Equations and Dynamical Systems, SpringerVerlag, New York. Weinstein, A. (1973) Normal modes for nonlinear Hamiltonian system. Inv. Math., 20, 47–57. Yoshizawa, T. (1975) Stability Theory and Existence of Periodic Solutions and Almost Periodic Solutions, Springer-Verlag, New York.

325

Index a activation function 287 adjoint 46, 273 Airy equation 5 amplitude, unique deﬁnition of 20 Andronov 115 aperiodic 87 attractive subsystem 292 attractor 101 autoparametric resonance 219, 220, see also combination internal resonance, one-to-one internal resonance, three-to-one internal resonance, two-to-one internal resonance – simultaneous 194, 220, 223 averaging, method of – generalized 6, 26, 276 b Banach space 287 basis, natural 36 Bessel’s equation 1 bifurcation 76, 78, 107 – degenerate 115 – diagram 78 – nondegenerate 82 Blasius problem 3 bookkeeping device 12 boundary layer 3 bracket, Lie 6, 33, 41, 45 Burger’s equation 3 c canonical form see Jordan form Cartesian 44 center 102 center manifold reduction 85, 142 – for continous systems 103–107 – for Hopf bifurcation 141–144 – for maps 72–75

– for retarded 3-D systems 308–310 – for retarded scalar equation 291–295 – for retarded SDoF equation 299–304 – for static bifurcation 126–132 center-manifold theorem 73, 103, 108 characteristics 3 circle map 87 codimension 33 combination autoparametric resonance 219 – in systems with quadratic nonlinearities 222 combination parametric resonance 194, 207 – in linear gyroscopic systems 208 – in linear nongyroscopic systems 191–192 – in systems with repeated frequencies 198, 199 combination resonance 228, 236, 253 complementary subspace 36, 38, 46 constant solution see also ﬁxed point control, speed 151 – time-delayed feedback 287 crane 311 critical point see ﬁxed point cutting force 295 cycle see limit cycle d degeneracy 33 degenerate – bifurcation 150 – form 199 detuning parameter 162 diagonal matrix 191 distinguished limit 199 divisor see small divisor dominant 4 Dufﬁng equation 9, 15 – forced oscillations of 161–172

The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

326

Index – free oscillation 9–13 e eigenvalues – distinct 62 – purely imaginary 42, 52 – repeated 64 – zero 32, 48, 54 – zero and two purely imaginary 52 eigenvector 61 – generalized 61, 62, 65, 66, 97, 99 – of matrix 32 equilibrium solution see ﬁxed point evolution equation 291 excitation see also multiplicative, parametric – harmonic 161 – multifrequency 257 f fast variation 11 ﬁxed point 6, 31, 66, 93, 97, 100 – hyperbolic 32, 72, 101, 107 – neutrally stable 102 – nonhyperbolic 32, 72, 101, 107 – nonstable 102 ﬂip bifurcation see period-doubling bifurcation ﬂutter 115, 196 focus 101 fold 76, see also saddle-node bifurcation form see Jordan form function space 291 fundamental parametric resonance 188, 245 – in Mathieu equation 187 – in nonlinear SDoF systems 208 – in systems with repeated frequencies 245–249 g galloping 115 generalized eigenvector 32, 61, 62, 65 generating vector 6 generic bifurcation 109, 117 gyroscopic systems – externally excited 225–228, 249–255 – parametrically excited 205–208 h harmonic excitation 161 Hartman–Grobman theorem 72, 102 heat transfer equation 3 Hénon map 93 heteroclinic connection 150, 151 homeomorphism 103

homology equations 21, 67 hoop 152 Hopf bifurcation 2, 108, 115, 116 – conditions for 76, 115 – in maps 85–88 – in retarded systems 287–312 – normal form of 2, 115–117, 137–146 hyperbolic ﬁxed point 32, 72, 108 i image 36 incompressible ﬂow 3 inevitable resonance 71 inﬂection point 119 inner product 291 internal resonance 220, see also combination autoparametric resonance, one-to-one autoparametric resonance, three-to-one autoparametric resonance, two-to-one autoparametric resonance, see also autoparametric resonance – absence of 232, 236, 250, 262 – in gyroscopic systems 225–228, 249–255 – in systems with cubic nonlinearities 238–243 – in systems with quadratic and cubic nonlinearities 263–277 – in systems with quadratic nonlinearities 220–225 – in systems with repeated frequencies 238–243, 279–285 invariant – circle 87 – set 66, 100 – time 11, 14 inverted pendulum 151 irrational number 87 j Jacobian 33, 101 – matrix 77, 86, 101, 102 Jordan canonical form 32, 98 k Krylov–Bogoliubov–Mitropolsky technique 170 l Lagrangian 263–265, 267, 270, 274 lathe 295–304 Lie 6, 33, 41 – bracket 41, 45 – transform 6

Index limit cycle 6 linear normal form 35 Liouville equation 4 logistic map 83 long-period terms 6 Lorenz equations 158 lunar orbital dynamics 115 m machine tool 295–304 main, resonance see primary manifold 73, see also center manifold reduction maps 61–95 Mathieu Equation 54, 187 – treated by method of multiple scales 56–57 – treated by method of normal forms 54–56, 187–188 modal-damping 191 monomials 41 multifrequency excitation 257 multiple scales, method of – applied to 2-DoF systems with quadratic nonlinearities 267–276 – applied to Dufﬁng equation 12 – applied to Hopf bifurcation 138–141 – applied to Mathieu equations 56 – applied to Rayleigh equation 15 – applied to retarded 3-DoF systems 306–308 – applied to retarded scalar equation 289–290 – applied to retarded SDoF systems 296–299 – applied to static bifurcation 117–126 – applied to systems with purely imaginary eigenvalues 42, 47 – applied to systems with quadratic and cubic nonlinearities 17 – applied to systems with repeated frequencies 283–285 – applied to time-delay crane control 311–312 – applied to van der Pol equation 19 multiplicity 32, 39 n natural basis 36 near-identity transformation 8 near-resonance 21, 35, 67–70 Neimark–Sacker bifurcation 76, 85 neuron 304 node 101

nonbifurcation 108, 110 nongyroscopic 191, 217 nonhyperbolic ﬁxed point 32, 72, 101 nonsemisimple 196, 240, 244, 279 nonsingular 21 – matrix 31 – transformation 4 nonstable ﬁxed point 102 nonuniform expansion 10 normalization 8, 13, 31 NST 48 null – matrix 33 – space 46, 49 o one-to-one autoparametric resonance 251 – in systems with cubic nonlinearities 239 – in systems with quadratic and cubic nonlinearities 264, 279 – in systems with repeated frequencies 195–205, 240–249, 279–285 orbit – of maps 66 – period-two 83 – quasiperiodic 88 ordering scheme 12 orthogonal 46 p parametric excitation see also Mathieu equation, 187–216 – in gyroscopic systems 205–208 – in single-degree-of-freedom nonlinear systems 208–212 – in systems with distinct frequencies 194 – in systems with repeated frequencies 196–205 – multifrequency 257 parametric resonance – simultaneous 194, see combination parametric resonance, fundamental parametric resonance, principal parametric resonance partial differential equation 1–3 pendulum 29, 151 period-doubling bifurcation 81, 85 pitchfork bifurcation 80, 108, 114 – in continuous systems 113–114 – in maps 80 Poincaré 115 Poisson bracket 33, 41

327

328

Index polar 44 – coordinates 44 – form 11, 19, 24 potential 148 primary resonance 161 – in 2-DOF systems with quadratic nonlinearities 176 – in Dufﬁng equation 161 – in gyroscopic systems 226, 250 – in systems with cubic nonlinearities 239 – in systems with quadratic and cubic nonlinearities 176, 223 principal parametric resonance 188, 190, 244 – in 2-DoF systems 244 – in gyroscopic systems 208 – in Mathieu equation 190 – in nongyroscopic systems 194, 197 – in nonlinear SDoF systems 210 – in systems with repeated frequencies 197 projection method – for Hopf bifurcation 144 – for static bifurcation 132 purely imaginary eigenvalues 7, 33, 42, 52, 107, 115 q quasiperiodic orbit 88, 90 r range 46 rational number 87, 90 Rayleigh equation 13, 15 reconstitution, method of 27, 213, 269 regenerative model 295 regular 22 repeated frequency 195 repellor 102 resonance, deﬁnition of – terms see also autoparametric resonance, combination resonance, fundamental parametric resonance, principal parametric resonance, subharmonic resonance, superharmonic resonance resonance, deﬁnition of 10 – condition for continous systems 37 – condition for maps 68, 69 – conditions of continous systems 35 – inevitable 71 – strong 71 – terms 10, 21, 33, 35, 42, 67–69 retarded systems 287–313

reverse pitchfork bifurcation 80, 114 Reynolds number 3 Rössler equation 157 rotating – circular hoop 152 – particle 29 – phase 6 s saddle 101, 102 saddle-node bifurcation 76, 77, see also static bifurcation, 108, 109 – in continuous systems 108–109 – in maps 76–78 secondary resonance see combination resonance, subharmonic resonance, superharmonic resonance secular terms 10, 20, 47 semisimple 39 sensitivity to initial conditions 88 set, invariant 66, 100 shift operator 291 Shoshitaishvili theorem 103 similarity transformation 3 simultaneous resonance 194, 220, 223 singular 68 – point 100 sink 101 slow variation 11 small divisor 21, 35, 68 solvability conditions 118, 273 source 101, 102 spindle 295 spiral 87, 90 stability – interchange of 79 – local 101 state-space form 271 static bifurcation see pitchfork bifurcation, saddle– node bifurcation, transcritical bifurcation, see ﬁxed point – normal form of 117–137 stationary solution see ﬁxed point straightforward expansion 10, 43 stream function 3 strong resonance 71, 86 subcritical 117 subcritical bifurcation 85, 114, 116 – Hopf bifurcation 116, 117 – period-doubling bifurcation 83–84 – pitchfork bifurcation 83, 85, 114, 120 subharmonic resonance 161, 164, 178 – in Dufﬁng equation 164–167, 171 – in gyroscopic systems 228, 253

Index – in nonlinear SDoF systems 178–180, 182–185 – in systems with quadratic nonlinearities 220 subspace 36, 38, 41 successive transformations 17, 66 supercritical bifurcation 114 – Hopf bifurcation 116, 117 – period-doubling bifurcation 83–84 – pitchfork bifurcation 80, 114, 120 superharmonic resonance 161, 180 – in Dufﬁng equation 167, 172 – in gyroscopic systems 253 – in MDoF systems 220, 228 – in SDoF systems 180–184 t tangent bifurcation 76, see also saddle-node bifurcation three-to-one autoparametric resonance 236 – in gyroscopic systems 251, 255 – in systems with cubic nonlinearities 238, 251 – in systems with quadratic and cubic nonlinearities 263

time delay 287–313 time invariant 11 transcritical bifurcation 79, 108, 112 – in continuous systems 111 – in maps 79 transversality condition 115 troublesome terms 21 turning point 5 two-to-one autoparametric resonance 219 – in gyroscopic systems 225–228 – in systems with quadratic and cubic nonlinearities 266–276 – in systems with quadratic nonlinearities 220, 226–228, 231 – simultaneous 223 – treated with generalized method of averaging 276–278 v van der Pol equation 15, 24, 152 variation of parameters 5, 8 w wanders 88

329

Related Titles Lalanne, C.

Mechanical Vibration and Shock Volume 1: Sinusoidal Vibration 2009 ISBN 978-1-84821-122-3

Gossett, E.

Discrete Mathematics with Proof 2009 ISBN 978-0-470-45793-1

Talman, R.

Geometric Mechanics Toward a Uniﬁcation of Classical Physics 2007 ISBN 978-3-527-40683-8

Zauderer, E.

Partial Differential Equations of Applied Mathematics 2006 ISBN 978-0-471-69073-3

Kahn, P. B., Zarmi, Y.

Nonlinear Dynamics Exploration Through Normal Forms 1998 ISBN 978-0-471-17682-4

Nayfeh, A. H.

Introduction to Perturbation Techniques 1993 ISBN 978-0-471-31013-6

Ali Hasan Nayfeh

The Method of Normal Forms Second, Updated and Enlarged Edition

WILEY-VCH Verlag GmbH & Co. KGaA

The Author Prof. Ali Hasan Nayfeh Virginia Polytechnic Institute and State University Department of Engineering Science and Mechanics Blacksburg, VA 24061 USA [email protected]

All books published by Wiley-VCH are carefully produced. Nevertheless, authors, editors, and publisher do not warrant the information contained in these books, including this book, to be free of errors. Readers are advised to keep in mind that statements, data, illustrations, procedural details or other items may inadvertently be inaccurate. Library of Congress Card No.: applied for British Library Cataloguing-in-Publication Data: A catalogue record for this book is available from the British Library. Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliograﬁe; detailed bibliographic data are available on the Internet at http://dnb.d-nb.de. © 2011 WILEY-VCH Verlag GmbH & Co. KGaA, Boschstr. 12, 69469 Weinheim, Germany All rights reserved (including those of translation into other languages). No part of this book may be reproduced in any form – by photoprinting, microﬁlm, or any other means – nor transmitted or translated into a machine language without written permission from the publishers. Registered names, trademarks, etc. used in this book, even when not speciﬁcally marked as such, are not to be considered unprotected by law. Typesetting le-tex publishing services GmbH, Leipzig Printing Binding Cover Design Adam-Design, Weinheim Printed in Singapore Printed on acid-free paper ISBN Print 978-3-527-41097-2 ISBN ePDF 978-3-527-63578-8 ISBN oBook 978-3-527-63580-1 ISBN ePub 978-3-527-63577-1

V

To my youngest son Nader

VII

Contents Preface XI Introduction 1 1 1.1 1.2 1.3 1.4 1.5 1.5.1 1.5.2 1.5.3 1.6 1.7 1.7.1 1.7.2 1.8

SDOF Autonomous Systems 7 Introduction 7 Dufﬁng Equation 9 Rayleigh Equation 13 Dufﬁng–Rayleigh–van der Pol Equation 15 An Oscillator with Quadratic and Cubic Nonlinearities 17 Successive Transformations 17 The Method of Multiple Scales 19 A Single Transformation 21 A General System with Quadratic and Cubic Nonlinearities 22 The van der Pol Oscillator 24 The Method of Normal Forms 25 The Method of Multiple Scales 26 Exercises 27

2 2.1 2.2 2.3 2.4 2.5 2.5.1 2.5.2 2.6 2.7

Systems of First-Order Equations 31 Introduction 31 A Two-Dimensional System with Diagonal Linear Part 34 A Two-Dimensional System with a Nonsemisimple Linear Form 39 An n-Dimensional System with Diagonal Linear Part 40 A Two-Dimensional System with Purely Imaginary Eigenvalues 42 The Method of Normal Forms 43 The Method of Multiple Scales 47 A Two-Dimensional System with Zero Eigenvalues 48 A Three-Dimensional System with Zero and Two Purely Imaginary Eigenvalues 52 The Mathieu Equation 54 Exercises 57

2.8 2.9

VIII

Contents

3 3.1 3.1.1 3.1.2 3.2 3.3 3.4 3.4.1 3.4.2 3.4.3 3.4.4 3.4.5 3.5

Maps 61 Linear Maps 61 Case of Distinct Eigenvalues 62 Case of Repeated Eigenvalues 64 Nonlinear Maps 66 Center-Manifold Reduction 72 Local Bifurcations 76 Fold or Tangent or Saddle-Node Bifurcation 76 Transcritical Bifurcation 79 Pitchfork Bifurcation 80 Flip or Period-Doubling Bifurcation 81 Hopf or Neimark–Sacker Bifurcation 85 Exercises 91

4 4.1 4.1.1 4.1.2 4.2 4.2.1 4.2.2 4.2.3 4.3 4.4 4.4.1 4.4.2 4.4.3 4.4.4 4.4.5 4.5 4.5.1 4.5.2 4.5.3 4.6 4.6.1 4.6.2 4.6.3 4.7

Bifurcations of Continuous Systems 97 Linear Systems 97 Case of Distinct Eigenvalues 98 Case of Repeated Eigenvalues 99 Fixed Points of Nonlinear Systems 100 Stability of Fixed Points 100 Classiﬁcation of Fixed Points 101 Hartman–Grobman and Shoshitaishvili Theorems Center-Manifold Reduction 103 Local Bifurcations of Fixed Points 107 Saddle-Node Bifurcation 108 Nonbifurcation Point 110 Transcritical Bifurcation 111 Pitchfork Bifurcation 113 Hopf Bifurcations 114 Normal Forms of Static Bifurcations 117 The Method of Multiple Scales 117 Center-Manifold Reduction 126 A Projection Method 132 Normal Form of Hopf Bifurcation 137 The Method of Multiple Scales 138 Center-Manifold Reduction 141 Projection Method 144 Exercises 146

5 5.1 5.2 5.3 5.4 5.4.1 5.4.2

Forced Oscillations of the Dufﬁng Oscillator 161 Primary Resonance 161 Subharmonic Resonance of Order One-Third 164 Superharmonic Resonance of Order Three 167 An Alternate Approach 169 Subharmonic Case 171 Superharmonic Case 172

102

Contents

5.5

Exercises

6 6.1 6.2 6.3 6.4 6.5

Forced Oscillations of SDOF Systems 175 Introduction 175 Primary Resonance 176 Subharmonic Resonance of Order One-Half 178 Superharmonic Resonance of Order Two 180 Subharmonic Resonance of Order One-Third 182

7 7.1 7.1.1 7.1.2 7.2 7.2.1 7.2.2 7.2.3 7.2.4 7.3 7.3.1 7.3.2 7.3.3 7.3.4 7.4 7.4.1 7.4.2 7.5 7.5.1 7.5.2 7.6

Parametrically Excited Systems 187 The Mathieu Equation 187 Fundamental Parametric Resonance 188 Principal Parametric Resonance 190 Multiple-Degree-of-Freedom Systems 191 The Case of Ω Near ω 2 C ω 1 194 The Case of Ω Near ω 2 ω 1 194 The Case of Ω Near ω 2 C ω 1 and ω 3 ω 2 194 The Case of Ω Near 2ω 3 and ω 2 C ω 1 195 Linear Systems Having Repeated Frequencies 195 The Case of Ω Near 2ω 1 198 The Case of Ω Near ω 3 C ω 1 199 The Case of Ω Near ω 3 ω 1 200 The Case of Ω Near ω 1 200 Gyroscopic Systems 205 The Case of Ω Near 2ω 1 208 The Case of Ω Near ω 2 ω 1 208 A Nonlinear Single-Degree-of-Freedom System 208 The Case of Ω Away from 2ω 209 The Case of Ω Near 2ω 211 Exercises 212

8 8.1 8.1.1 8.1.2 8.1.3 8.1.4 8.2 8.2.1 8.2.2 8.3 8.4

MDOF Systems with Quadratic Nonlinearities 217 Nongyroscopic Systems 217 Two-to-One Autoparametric Resonance 220 Combination Autoparametric Resonance 222 Simultaneous Two-to-One Autoparametric Resonances Primary Resonances 223 Gyroscopic Systems 225 Primary Resonances 226 Secondary Resonances 227 Two Linearly Coupled Oscillators 229 Exercises 232

9 9.1 9.1.1 9.1.2

TDOF Systems with Cubic Nonlinearities 235 Nongyroscopic Systems 235 The Case of No Internal Resonances 236 Three-to-One Autoparametric Resonance 238

172

223

IX

X

Contents

9.1.3 9.1.4 9.1.5 9.1.6 9.2 9.2.1 9.2.2 9.2.3

One-to-One Internal Resonance 239 Primary Resonances 239 A Nonsemisimple One-to-One Internal Resonance 240 A Parametrically Excited System with a Nonsemisimple Linear Structure 244 Gyroscopic Systems 249 Primary Resonances 250 Secondary Resonances in the Absence of Internal Resonances Three-to-One Internal Resonance 255

10 10.1 10.2 10.3 10.4 10.5 10.6 10.6.1 10.6.2 10.6.3 10.7 10.8 10.8.1 10.8.2 10.9

Systems with Quadratic and Cubic Nonlinearities 257 Introduction 257 The Case of No Internal Resonance 262 The Case of Three-to-One Internal Resonance 263 The Case of One-to-One Internal Resonance 264 The Case of Two-to-One Internal Resonance 266 Method of Multiple Scales 267 Second-Order Form 268 State-Space Form 271 Complex-Valued Form 274 Generalized Method of Averaging 276 A Nonsemisimple One-to-One Internal Resonance 279 The Method of Normal Forms 279 The Method of Multiple Scales 283 Exercises 285

11 11.1 11.1.1 11.1.2 11.2 11.2.1 11.2.2 11.3 11.3.1 11.3.2 11.4 11.5

Retarded Systems 287 A Scalar Equation 287 The Method of Multiple Scales 289 Center-Manifold Reduction 291 A Single-Degree-of-Freedom System 295 The Method of Multiple Scales 296 Center-Manifold Reduction 299 A Three-Dimensional System 304 The Method of Multiple Scales 306 Center-Manifold Reduction 308 Crane Control with Time-Delayed Feedback Exercises 313 References 315 Further Reading 319 Index 325

311

251

XI

Preface This book gives an introductory treatment of the method of normal forms. This technique has its application in many branches of engineering, physics, and applied mathematics. Approximation techniques such as these are important for people working with dynamical problems and are a valuable tool they should have in their tool box. The exposition is largely by means of examples. The readers need not understand the physical bases of the examples used to describe the techniques. However, it is assumed that they have a knowledge of basic calculus as well as the elementary properties of ordinary differential equations. For most of the examples, the results obtained with the method of normal forms are shown to be equivalent to those obtained with other perturbation methods, such as the methods of multiple scales and averaging. As such, new sections are added treating some of the examples with these methods. Moreover, exercises are added to most chapters. Because the normal forms of maps and differential equations are very useful in bifurcation analysis, I added in this edition three chapters dealing with the normal forms and bifurcations of maps, continuous systems, and retarded systems. The normal forms of continuous systems are constructed using the method of multiple scales, a combination of center-manifold reduction and the method of normal forms, and the new method of projection, which is developed ﬁrst in this edition. Also, the normal forms of retarded systems are constructed using center-manifold reduction and the method of multiple scales. In the center-manifold reduction, we represent the retarded equations as operator differential equations, decompose the solution space of their linearized form into stable and center subspaces, deﬁne an inner product, determine the adjoint of the operator equations, calculate the center manifold, carry out details of the projection using the adjoint of the center subspace, and ﬁnally calculate the normal form on the center manifold.

XII

Preface

I am very much indebted to my late parents, Hasan and Khadrah, who in spite of their lack of formal education insisted that all their sons obtain the highest degrees. If it were not for their incredible foresight on the value of an education even under the most severe conditions, I would not have ﬁnished secondary school. This book and its second edition would not have been written without the patience and continuous encouragement of my wife, Samirah. Blacksburg, VA, December 2010

Ali Hasan Nayfeh

1

Introduction The method of normal forms dates back to the days of Euler, Delaunay, Poincaré, Dulac, and Birkhoff. Moreover, the concept of using coordinate transformations to simplify mathematical problems involving algebraic, ordinary differential, partial differential, integral, and integro-differential equations has been used for a long time, as illustrated by the following examples. As a ﬁrst example, we consider Bessel’s equation of order one-half; that is, x 2 u00 C x u0 C x 2 14 u D 0 Using the transformation x 1/2 v (x), we transform this equation into the simple equation v 00 C v D 0 whose solution is v D c 1 cos x C c 2 sin x where c 1 and c 2 are arbitrary constants. Hence, Bessel’s function of order one-half J1/2 (x) is given by J1/2 (x) D x 1/2 (c 1 cos x C c 2 sin x) As a second example, we consider the vibrations of an n degree-of-freedom system governed by the following set of n coupled, linear equations of motion: xR C K x D 0 where x is a column vector of length n and K is an nn constant symmetric matrix. Using the transformation x D P v, we obtain vR C P 1 K P v D 0 Assuming the eigenvalues λ 1 , λ 2 , . . . , and λ n of K to be distinct and choosing the columns of P to be the orthonormal eigenvectors of K, we ﬁnd that P 1 K P is a The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

2

Introduction

diagonal matrix Λ with entries λ 1 , λ 2 , . . . , λ n . Hence, the system of equations can be written as vR C Λ v D 0 or in the decoupled form vR i C λ i v i D 0

and

i D 1, 2, . . . , n

which is called the normal-modal form of xR C K x D 0 As a third example, we consider the system aP D µ a 12 a sin 2β a βP D 12 σ a 12 a cos 2β where µ and σ are constants, which describes the time variation of the amplitude and phase of a parametrically excited linear oscillator in the case of a principal parametric resonance (Nayfeh and Mook, 1979). This system is nonlinear and its solution is not apparent. However, using the nonlinear transformation x D a cos β and y D a sin β, we transform the nonlinear system into the following linear system: xP D µ x C yP D µ y

1 (σ 1) y 2 1 (σ C 1) x 2

whose closed-form solution is readily obtainable. As a fourth example, we consider the nonlinear system xP D y C (ax C b y ) x 2 C y 2 yP D x C (a y b x) x 2 C y 2 where a and b are constants, which describes the motion near a Hopf bifurcation point (Marsden and McCracken, 1976; Wiggins, 1990), as described in Section 4.4.5. Again the solution of this system is not apparent. However, using the nonlinear transformation x D r cos β and y D r sin β, we transform the system into rP D ar 3 βP D 1 C b r 2 whose closed-form solution is readily obtainable. As a ﬁfth example, we consider the linear partial differential equation u t t c 2 u x x D f 0 (x c t) f 00 (x c t)

Introduction

where f is a known twice differential function, the prime denotes the derivative with respect to the argument (x c t), and subscripts denote partial derivatives. The general solution of this equation can be readily obtained if we express the independent variables x and t in terms of the characteristics ξ D x ct

and

η D x C ct

Thus, this partial differential equation is transformed into 4c 2 u ξ η D f 0 (ξ ) f 00 (ξ ) whose general solution is uD

1 02 f (ξ )η C g(ξ ) C h(η) 8c 2

where g(ξ ) and h(η) are general functions of ξ and η. As a sixth example, we consider the nonlinear partial differential equation u t C uu x D v u x x where v is a constant, which is known as Burger’s equation (Whitham, 1974). Replacing u with ψ x and integrating the result once yields ψ t C 12 ψ x2 D v ψ x x Then, using the nonlinear transformation ψ D 2v ln(φ), Hopf (1950) and Cole (1951) transformed the nonlinear equation into the linear heat transfer equation φt D v φxx which can be solved much more easily than the original nonlinear equation. As a seventh example, we consider the steady, incompressible, high-Reynolds number ﬂow over a ﬂat plate aligned with the oncoming uniform stream. The boundary layer approximation to the stream function ψ(x, y ) is governed by Van Dyke (1964) ψy y y C ψx ψy y ψy ψx y D 0 ψ(x, 0) D 0 ψ y (x, 0) D 0 and 0 < x < 1 ψ y (x, 1) D 1 This nonlinear partial differential equation can be reduced to an ordinary differential equation by using the similarity transformation p p ψ(x, y ) D 2x f (η) , η D y / 2x With this transformation, the boundary layer problem becomes f 000 C f f 00 D 0 ,

f (0) D 0 ,

which is the Blasius problem.

f 0 (0) D 0 ,

f 0 (1) D 1

3

4

Introduction

In the preceding examples, transformations were introduced to transform a difﬁcult problem into a more readily solvable problem. Next, we consider cases in which a transformation is used to transform the problem into a new “approximate” problem for which the exact solution can be readily obtained. Speciﬁcally, we consider the Liouville equation y 00 C λ 2 q(x)y D 0 when

λ1

where λ is a constant, q(x) is a known function, and the prime denotes the derivative with respect to x. To determine an approximate solution of this equation when λ 1, we transform both of the dependent and independent variables as z D φ(x) and

v (z) D ψ(x)y (x)

With this transformation, the Liouville equation becomes 2 1 2φ 0 ψ 0 d v ψ 00 2ψ 02 λ q d2 v 00 φ v D0 C C C d z2 φ 02 ψ dz φ 02 ψ φ 02 ψ 2 φ 02 We choose φ and ψ so that the dominant part of the transformed equation has the simplest possible form and, at the same time, has solutions that have qualitatively the same behavior as the solutions of the original equations. In other words, we have to insist on the transformation being regular everywhere in the interval of interest. To this end, we force the coefﬁcient of d v /d z to be zero; that is, 2φ 0 ψ 0 D0 ψ p Hence, ψ D φ 0 . In order that the transformation be regular, ψpmust be regular and have no zeros in the interval of interest. Then, because ψ D φ 0 , φ 0 must be regular and have no zeros in the interval of interest. Consequently, we set φ 00

λ 2 q D φ 02 ξ (z) so that the transformed equation becomes d2 v C ξ (z)v D δ v d z2 and choose the simplest possible function ξ (z) that yields a nonsingular transformation. In order that φ 0 be regular and have no zeros in the interval of interest, ξ (z) must have the same number, type, and order of singularities and zeros as q. For example, when q > 0 everywhere in the interval of interest, the solutions of the original equation are oscillatory, and hence φ and ψ must be chosen so that the dominant part of the transformed equation is d2 v Cv D0 d z2

Introduction

which is the simplest possible equation with oscillatory solutions. When q < 0 everywhere in the interval of interest, one of the solutions of the original equation grows exponentially with x and the other decays exponentially with x. Hence, φ and ψ must be chosen so that the dominant part of the transformed equation is d2 v v D0 d z2 which is the simplest possible equation with exponentially growing and decaying solutions. When q changes sign once in the interval of interest, the solutions of the original equation are oscillatory on one side of the sign change and exponentially growing and decaying on the other side. For example, if q D 1 x 3 , the solutions of the original equation are oscillatory for x < 1 and exponential for x > 1. Hence, φ and ψ must be chosen so that the dominant part of the transformed equation has solutions whose behavior changes from oscillatory to exponentially growing and decaying at a given point. The simplest possible equation with these properties is the Airy equation d2 v zv D 0 d z2 When z > 0 the solutions of this equation are growing and decaying with z, and when z < 0 they are oscillatory. In other words, if q(x) is regular and has only a simple zero (simple turning point) such as 1 x 3 , then ξ (z) must be chosen to be regular and have only a simple zero. The simplest possible function that satisﬁes these requirements is ξ (z) D z. If q(x) is regular and has only a double zero at a point in the interval of interest (i.e., turning point of order 2), ξ (z) must be chosen to be regular and have only a double zero. The simplest possible function satisfying these requirements is ξ (z) D z 2 . If q(x) is regular and has only a zero of order n (i.e., turning point of order n), ξ (z) must be chosen to be z n . If q(x) has two zeros at x D a and b, where b > 1, of order m and n, then one uses ξ (z) D z m (1 z) n In analyzing oscillations of a weakly nonlinear system, the method of variation of parameters is usually used to transform the equations governing these oscillations into the standard form xP D f (xI ) D where

1 X m f (x) m! m mD0

ˇ @ m f ˇˇ f m (x) D @ m ˇD0

Here x and f are vectors with N components. The vector x may represent, for example, the amplitudes and phases of the system. If we denote the components

5

6

Introduction

of the vector f m by f m n , then a component x k of the vector x is said to be a rapidly rotating phase if f 0k ¤ 0. To analyze this standard system, we introduce a near-identity transformation x D X (yI ) D y C X 1 (y) C 2 X 2 (y) C from x to y such that the system is transformed into yP D g(yI ) D

1 X n g (y) n! n nD0

where the g n contain long-period terms only. Using the generalized method of averaging (Nayfeh, 1973), one determines the X n and g n by substituting the transformation into the standard system and separating the short- and long-period terms assuming that the X n contain short-period terms only. Alternatively, we can deﬁne the transformation x D X (yI ) as the solution of the N differential equations dx D W (xI ) , d

x( D 0) D y

The vector W is called the generating vector. This equation generates the so-called Lie transforms (Kamel, 1970), which are invertible because they are close to the identity. It seems at ﬁrst that we are going in circles because we are proposing to simplify the original system of differential equations by solving a system of N differential equations. This is not the case, because we are interested in the solution of the original system for large t, whereas we need the solution of the transformation for small , which is a signiﬁcant simpliﬁcation. These examples clearly show that linear and nonlinear coordinate transformations can be used to simplify linear and nonlinear problems. A powerful method for systematically constructing these transformations is the method of normal forms. The basic idea underlying the method of normal forms is the use of “local” coordinate transformations to “simplify” the equations describing the dynamics of the system under consideration. In other words, with the method of normal forms, one seeks a near-identity coordinate transformation in which the dynamical system takes the “simplest” or so-called normal form. The transformations are generated in a neighborhood of a known solution, such as a ﬁxed point (constant, stationary, or equilibrium solution) or a periodic orbit (limit cycle) of a system. In this text, the normalization is usually carried out with respect to a perturbation parameter.

7

1 SDOF Autonomous Systems 1.1 Introduction

In this chapter, we describe the method of normal forms using single-degreeof-freedom (SDOF) autonomous systems that can be modeled by the following second-order nonlinear ordinary differential equation: P uR C ω 2 u D f (u, u)

(1.1)

where f (u, u) P can be developed in a power series in terms of u and u. P In what follows, we will refer to uP C ω 2 u D 0 as the unperturbed system and (1.1) as the perturbed system. We assume that (1.1) has an equilibrium at u D 0 and uP D 0. Equation 1.1 can be cast as a system of two ﬁrst-order equations by letting x1 D u

and

x2 D uP

(1.2)

The result is xP 1 D x2

(1.3)

xP 2 D ω 2 x1 C f (x1 , x2 )

(1.4)

It is clear that the unperturbed system xP 1 D x2

and

xP2 D ω 2 x1

has a simple pair of purely imaginary eigenvalues ˙i ω. The main idea underlying the method of normal forms is to introduce a nearidentify transformation x1 D y 1 C h 1 (y 1 , y 2 )

(1.5a)

x2 D y 2 C h 2 (y 1 , y 2 )

(1.5b)

from (x1 , x2 ) to (y 1 , y 2 ) into (1.3) and (1.4) to produce the simplest possible equations (the so-called normal form). We call the transformation (1.5) near-identity The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

8

1 SDOF Autonomous Systems

because x1 (t) y 1 (t) and x2 (t) y 2 (t) are small; that is, o(x1 (t), x2 (t)). This procedure is also called normalization. To this end, we substitute (1.5) into (1.3) and (1.4) and obtain yP1 D y 2 C h 2

@h 1 @h 1 yP1 yP 2 @y 1 @y 2

yP2 D ω 2 y 1 ω 2 h 1 C f (y 1 C h 1 , y 2 C h 2 )

(1.6a) @h 2 @h 2 yP1 yP2 @y 1 @y 2

(1.6b)

Then, we choose h 1 and h 2 such that (1.6) assume their simplest form. This task is accomplished in steps. If one decomposes f (x1 , x2 ) as f (x1 , x2 ) D

N X

f n (x1 , x2 )

(1.7)

nD1

where f n is a polynomial of degree n in x1 and x2 , then one chooses h 1 and h 2 to simplify the terms resulting from the lowest-order polynomial f m (x1 , x2 ), where m 2, in f (x1 , x2 ). In the next step, one chooses a second near-identity transformation to simplify the polynomial terms of degree m C 1, and so on. It turns out that, because the unperturbed system (1.3) and (1.4) represents an oscillator, the governing equations can conveniently be expressed as a single complexvalued equation. To this end, we follow steps similar to those used in the method of variation of parameters (Nayfeh, 1981). When f 0, the solution of (1.1) can be expressed as u D B e i ω t C BN e i ω t where B is a constant and BN is the complex conjugate of B. Hence, uP D i ω B e i ω t BN e i ω t

(1.8)

(1.9)

When f ¤ 0, we continue to represent the solution of (1.1) as in (1.8) subject to the constraint (1.9) but with time-varying rather than constant B. Next, we replace B e i ω t with ζ(t) and rewrite (1.8) and (1.9) as N N u D ζ(t) C ζ(t) and uP D i ω ζ(t) ζ(t) (1.10) N we obtain Hence, solving for ζ and ζ, i i 1 1 u uP u C uP and ζN D ζD 2 ω 2 ω Differentiating (1.11) with respect to t yields i i 1 1 uP uR D uP C i ωu ζP D f 2 ω 2 ω on account of (1.1). Hence, 1 i i ζP D i ω u uP f (u, u) P 2 ω 2ω

(1.11)

(1.12)

(1.13)

1.2 Dufﬁng Equation

which, upon using (1.10), becomes i N i ω ζ ζN ζP D i ωζ f ζ C ζ, 2ω

(1.14)

Next, we consider different polynomial forms for f.

1.2 Dufﬁng Equation

The Dufﬁng equation is uR C ω 2 u D α u3 so that, in this case, f D α u3 and (1.14) becomes 3 iα ζP D i ωζ ζ C ζN 2ω

(1.15)

We introduce a near-identity transformation from ζ to η in the form ζ D η C h (η, η) N

(1.16)

and obtain ηP D i ωη C i ωh

3 iα @h @h P ηN ηP η C h C ηN C hN @η @ ηN 2ω

(1.17)

Because the nonlinearity is cubic, we assume that h is third order in η and η; N that is, h D Λ 1 η 3 C Λ 2 η 2 ηN C Λ 3 η ηN 2 C Λ 4 ηN 3

(1.18)

and choose the Λ i so that (1.17) takes the simplest possible (normal) form. In the ﬁrst step, we eliminate ηP and ηPN from the right-hand side of (1.17). This task is accomplished by iteration. To the ﬁrst approximation, it follows from (1.17) that ηP D i ωη

ηPN D i ω ηN

and

(1.19)

Next, we replace ηP and ηNP on the right-hand side of (1.17) using (1.19), use (1.18), keep up to third-order terms, and obtain

α ηP D i ωη i ω 2Λ 1 C 2ω 2 α ηN 3 C i ω 4Λ 4 2ω 2

3i α 2 3α η η ηN 2 η ηN C i ω 2Λ 3 2ω 2ω 2 3

(1.20)

9

10

1 SDOF Autonomous Systems

Next, we choose Λ 1 , Λ 3 , and Λ 4 to eliminate the terms involving η 3 , η ηN 2 , and ηN 3 ; that is, Λ1 D

α , 4ω 2

Λ3 D

3α , 4ω 2

Λ4 D

α 8ω 2

(1.21)

However, because Λ 2 does not appear in (1.20), the term involving η 2 ηN cannot be eliminated; it is called a resonance term. Consequently, to the second approximation, the simplest possible form for ηP is ηP D i ωη

3i α 2 η ηN 2ω

(1.22)

To show that η 2 ηN is a resonance term, we ﬁnd a solution for (1.22) by iteration. To the ﬁrst approximation, η D Ae i ω t , where A is a constant. Then, (1.22) becomes ηP D i ωη

3i α 2 N i ω t A Ae 2ω

whose solution can be written as η D Ae i ω t

3α 2 N i ω t A At e 2ω

(1.23a)

It is clear that this expansion, which is also a straightforward expansion, is nonuniform for large t because of the presence of a secular term created by η 2 η. N Alternatively, we can demonstrate that the term ζ 2 ζN is a resonance term in the original equation (1.15). To the ﬁrst approximation, we neglect the nonlinear term in (1.15) and ﬁnd that ζ D Ae i ω t . Then, to the second approximation, (1.15) becomes i α 3 3i ω t N i ω t C 3A AN2 e i ω t C AN3 e 3i ω t ζP D i ωζ C 3A2 Ae A e 2ω whose solution can be written as ζ D Ae i ω t C

α 3 3i ω t 3i α 2 N i ω t 3α A e A AN2 e i ω t A At e C 2 4ω 2ω 4ω 2

α N3 3i ω t A e 8ω 2

(1.23b)

It is clear that this expansion is nonuniform because of the presence of a secular N The other three terms proportional to A3 e 3i ω t , A AN2 e i ω t , and term created by ζ 2 ζ. 3 3i ω t N A e created by ζ 3 , ζ ζN 2 , and ζN 3 do not produce secular terms and hence they are nonresonance. Consequently, one can choose a near-identity transformation to eliminate them. As a second alternative, starting with the original equation (1.15), we break the N into two parts as nonlinear part f (ζ, ζ) N D f 1 (ζ, ζ) N C f 2 (ζ, ζ) N f (ζ, ζ) where e i ω t f 1 e i ω t , e i ω t

1.2 Dufﬁng Equation

is time invariant, whereas e i ω t f 2 e i ω t , e i ω t is not time invariant. In the present case, 3 f D ζ C ζN , f 1 D 3ζ 2 ζN , f 2 D ζ 3 C 3ζ ζN 2 C ζN 3 Thus, e i ω t f 1 e i ω t , e i ω t D e i ω t 3e 2i ω t e i ω t D 3 which is time invariant, whereas e i ω t f 2 e i ω t , e i ω t D e 2i ω t C 3e 2i ω t C e 4 i ω t which does not contain any time-invariant terms. Substituting (1.16) and (1.18) into (1.10), using (1.21), and setting Λ 2 D 0 because it is arbitrary yields u D η C ηN

α 3 3α 2 η C ηN 3 C η ηN C η 2 ηN 2 2 8ω 4ω

(1.24)

where η is given by (1.22). Next, we separate the fast from the slow variations in η by introducing the transformation η D A(t)e i ω t where ω is the natural frequency of the system and A is a function of time, into (1.22) and (1.24) and obtain 3i α 2 N AP D A A 2ω

(1.25) α 3 3i ω t A e C AN3 e 3i ω t 2 8ω C AN2 Ae i ω t C

N i ω t u D Ae i ω t C Ae C

3α 2 N i ω t A Ae 4ω 2

(1.26)

Expressing A in the polar form A D 12 ae i β where a and β are functions of t, we rewrite (1.26) as 3α 3 α a3 uD aC cos(ωt C β) a cos(3ωt C 3β) C 2 16ω 32ω 2

(1.27)

(1.28)

Substituting (1.27) into (1.25) and separating real and imaginary parts, we have aP D 0 a βP D

(1.29) 3α 3 a 8ω

(1.30)

11

12

1 SDOF Autonomous Systems

In determining the normal form (1.22), we had to use an ordering scheme to indicate the relative magnitudes of the different terms in (1.15). We based the orN ζ ζN 2 , and ζN 3 dering scheme on the fact that ζ and ζN are small and hence ζ 3 , ζ 2 ζ, N In other words, we based the ordering scheme on are much smaller than ζ and ζ. the degree of the terms. This worked well in this example, but there are many physical systems where the ordering does not follow from the degree of the polynomial but from the presence of certain parameters in their models. We consider such an example in the next section. Next, we treat (1.15) by using the method of multiple scales. To this end, we introduce a small nondimensional parameter as a bookkeeping device and rewrite (1.15) as 3 iα ζP D i ωζ ζ C ζN 2ω

(1.31)

Then, we seek an approximate solution of (1.31) in the form ζ(tI ) D ζ0 (T0 , T1 ) C ζ1 (T0 , T1 ) C

(1.32)

where Tn D n t and d @ @ C C D D0 C D1 C D dt @T0 @T1

(1.33)

Substituting (1.32) and (1.33) into (1.31) and equating coefﬁcients of like powers of yields Order (0 )

D0 ζ0 i ωζ0 D 0

(1.34)

Order ()

D0 ζ1 i ωζ1 D D1 ζ0

3 iα ζ0 C ζN 0 2ω

(1.35)

The solution of (1.34) can be expressed as ζ0 D A(T1 )e i ω T0

(1.36)

Then, (1.35) becomes i α 3 3i ω T0 N i ω T0 A e C 3A2 Ae 2ω C 3A AN2 e i ω T0 C AN3 e 3i ω T0

D0 ζ1 i ωζ1 D A0 e i ω T0

(1.37)

Eliminating the terms that lead to secular terms from (1.37), we have A0 D

3i α 2 N A A 2ω

(1.38)

1.3 Rayleigh Equation

Then, a particular solution of (1.37) can be expressed as ζ1 D

α 3 3i ω T0 3α α N3 3i ω T0 A e A e C A AN2 e i ω T0 C 4ω 2 4ω 2 8ω 2

(1.39)

Substituting (1.36) and (1.39) into (1.10), we obtain N i ω t α A3 e 3i ω t C AN3 e 3i ω t u D Ae i ω t C Ae 8ω 2 3α 2 N i ω t A Ae C A AN2 e i ω t C C 2 4ω

(1.40)

Equations 1.38–1.40 are in full agreement with (1.25) and (1.26) obtained with the method of normal forms because T1 D t and can be set equal to unity.

1.3 Rayleigh Equation

The Rayleigh equation is uR C ω 2 u D uP 13 uP 3

(1.41)

where is a small, positive nondimensional parameter. Here f D uP 13 uP 3 and (1.14) becomes h 3 i ζP D i ωζ C 12 ζ ζN C 13 ω 2 ζ ζN

(1.42)

In this example, the ordering is not based on the degree of the polynomial, but on the small nondimensional parameter . Normalization is carried out in terms of the small parameter . In fact, the perturbation contains linear as well as cubic terms. Using the transformation (1.16), we rewrite (1.42) as @h @h P 1 ηP D i ωη C i ωh ηP ηN C η ηN C h hN @η @ ηN 2 3

1 2 (1.43) C ω η ηN C h hN 3 Because the perturbation in (1.43) involves linear and cubic terms, we express h in the form (1.44) h D ∆ 1 η C ∆ 2 ηN C Λ 1 η 3 C Λ 2 η 2 ηN C Λ 3 η ηN 2 C Λ 4 ηN 3 Moreover, to the ﬁrst approximation, ηP and ηPN are given by (1.19). Then, substituting (1.19) and (1.44) into the right-hand side of (1.43) and keeping terms up to O(),

13

14

1 SDOF Autonomous Systems

we obtain 1 i 1 ηN C η iω 2Λ 1 C i ω η 3 ηP D i ωη C 2iω ∆ 2 C 4ω 2 6 1 2 2 1 1 ω η ηN C iω 2Λ 3 i ω η ηN 2 C iω 4Λ 4 C i ω ηN 3 2 2 6 (1.45) We note that (1.45) is independent of ∆ 1 and Λ 2 and hence they are arbitrary. Moreover, the terms proportional to η and η 2 ηN are resonance terms and hence cannot be eliminated from (1.45). Next, we choose ∆ 2 , Λ 1 , Λ 3 , and Λ 4 to eliminate the terms involving η, N η 3 , η ηN 2 , and ηN 3 , thereby producing the simplest possible equation for η. Thus, we have ∆2 D

i , 4ω

Λ1 D

1 iω , 12

Λ3 D

1 iω , 4

Λ4 D

1 iω 24

(1.46)

With this choice, (1.45) takes the normal form ηP D i ωη C 12 η 12 ω 2 η 2 ηN

(1.47)

Again, in this case, we could have identiﬁed the resonance terms in (1.42) by one of the procedures described in Section 1.2. Because the solution of the unperturbed problem is proportional to e i ω t , the resonance terms in N 3 N D ζ ζN C 1 ω 2 (ζ ζ) f (ζ, ζ) 3 are the terms proportional to e i ω t or the time-invariant terms in e i ω t f e i ω t e i ω t , i ω e i ω t e i ω t N is the only resonance A simple calculation shows that the term 1/2(ζ ω 2 ζ 2 ζ) term. Hence, keeping only the resonance terms in (1.42), we have ζP D i ωζ C 12 ζ ω 2 ζ 2 ζN C which is formally equivalent to (1.47). Next, we treat (1.42) with the method of multiple scales. To this end, we substitute (1.32) and (1.33) into (1.42), equate coefﬁcients of equal powers of , and obtain Order (0 )

D0 ζ0 i ωζ0 D 0

(1.48)

Order ()

D0 ζ1 i ωζ1 D D1 ζ0 C

1 2

h

3 i ζ0 ζN 0 C 13 ω 2 ζ0 ζN 0

(1.49)

1.4 Dufﬁng–Rayleigh–van der Pol Equation

The solution of (1.48) can be expressed as ζ0 D A(T1 )e i ω T0

(1.50)

Then, (1.49) becomes N i ω T0 C 1 ω 2 A3 e 3i ω T0 D0 ζ1 i ωζ1 D A0 e i ω T0 C 12 Ae i ω T0 12 Ae 6 N i ω T0 C 1 ω 2 A AN2 e i ω T0 1 ω 2 AN3 e 3i ω T0 12 ω 2 A2 Ae 2 6 (1.51) Eliminating the terms that lead to secular terms from (1.51), we have A0 D 12 A 12 ω 2 A2 AN

(1.52)

Letting η D Ae i ω t in (1.47), we obtain (1.52) because T1 D t.

1.4 Dufﬁng–Rayleigh–van der Pol Equation

The Dufﬁng, Rayleigh, and van der Pol equations are special cases of uR C ω 2 u D µ uP C α 1 u3 C α 2 u2 uP C α 3 u uP 2 C α 4 uP 3

(1.53)

so that f D µ uP C α 1 u3 C α 2 u2 uP C α 3 u uP 2 C α 4 uP 3 and (1.14) becomes 3 2 i h ζP D i ωζ ζ ζN i µ ω ζ ζN C α 1 ζ C ζN C i ωα 2 ζ C ζN 2ω 2 3 i (1.54) ω 2 α 3 ζ C ζN ζ ζN i ω 3 α 4 ζ ζN Using the transformation (1.16), where h D O(), we rewrite (1.54) as ηP D i ωη C i ωh

@h @h P ηP ηN @η @ ηN

i N 2 (η η) N i µ ω (η η) N C i ωα 2 (η C η) 2ω N 3 ω 2 α 3 (η C η) N (η η) N 2 i ω 3 α 4 (η η) N 3 Cα 1 (η C η)

(1.55)

where terms of O( 2 ) and higher have been neglected. Again, because the perturbation contains linear as well as third-order terms, h has the form (1.44). Moreover, to the ﬁrst approximation, ηP and ηPN are given by

15

16

1 SDOF Autonomous Systems

(1.19). Hence, substituting (1.19) and (1.44) into (1.55) yields iµ ηN ηP D i ωη C 2iω ∆ 2 C 4ω i 3α 1 C i ωα 2 C ω 2 α 3 C 3i ω 3 α 4 η 2 ηN 2ω

3 1 1 2 3 α η C µ η C iω 2Λ 1 C i ωα ω α i ω α 1 2 3 4 2 2ω 2

1 2 3 3α η ηN 2 C iω 2Λ 3 i ωα C ω α 3i ω α 1 2 3 4 2ω 2

3 1 2 3 α ηN C iω 4Λ 4 i ωα ω α C i ω α (1.56) 1 2 3 4 2ω 2 We note that ∆ 1 and Λ 2 do not appear in (1.56) and hence they are arbitrary and the terms η and η 2 ηN are resonance terms. To produce the simplest form for (1.56), we choose ∆ 2 , Λ 1 , Λ 3 , and Λ 4 to eliminate the terms involving η, N η 3 , η ηN 2 , and ηN 3 ; that is, iµ 4ω 1 Λ 1 D 2 α 1 C i ωα 2 ω 2 α 3 i ω 3 α 4 4ω 1 3α 1 i ωα 2 C ω 2 α 3 3i ω 3 α 4 Λ3 D 2 4ω 1 α 1 i ωα 2 ω 2 α 3 C i ω 3 α 4 Λ4 D 8ω 2

∆2 D

(1.57) (1.58) (1.59) (1.60)

With these choices, (1.56) assumes the simple form 1 i 3α 1 C i ωα 2 C ω 2 α 3 C 3i ω 3 α 4 η 2 ηN ηP D i ωη C µ η 2 2ω

(1.61)

Again, we did not have to go through the lengthy algebra to arrive at the normal form (1.61). Because the solution of the unperturbed problem (1.54) is proportional to e i ω t , we could have replaced ζ with e i ω t in the perturbation and identiﬁed the terms proportional to e i ω t . In this case, they are i 1 3α 1 C i ωα 2 C ω 2 α 3 C 3i ω 3 α 4 ζ 2 ζN µ ζ 2 2ω Hence, keeping only the resonance terms in (1.54), we obtain the normal form 1 i ζP D i ωζ C µ ζ 3α 1 C i ωα 2 C ω 2 α 3 C 3i ω 3 α 4 ζ 2 ζN 2 2ω which is formally equivalent to (1.61).

1.5 An Oscillator with Quadratic and Cubic Nonlinearities

1.5 An Oscillator with Quadratic and Cubic Nonlinearities

We consider free oscillations of a single-degree-of-freedom system governed by uR C ω 2 u C δ u2 C α u3 D 0

(1.62)

To keep track of the different orders of magnitude, we use a nondimensional parameter that is the order of the amplitude of oscillations and hence rewrite (1.62) as uR C ω 2 u C δ u2 C 2 α u3 D 0

(1.63)

Thus, f D δ u2 2 α u3 and (1.14) becomes 2 3 iδ i 2 α ζP D i ωζ C ζ C ζN C ζ C ζN 2ω 2ω

(1.64)

In the next section, we use two successive transformations to produce the normal form of (1.64). In Section 1.5.3, we use a single transformation to produce the same normal form, and in Section 1.5.2, we use the method of multiple scales to determine a second-order expansion of (1.64). 1.5.1 Successive Transformations

To simplify the O() terms in (1.64), we introduce the near-identity transformation ζ D η C h 1 (η, η) N

(1.65)

and rewrite (1.64) as ηP D i ωη C iωh 1

2 @h 1 @h 1 P iδ ηP ηN C η C ηN C h 1 C hN 1 @η @ ηN 2ω

i 2 α (η C η) N 3 C 2ω The form of the O() terms suggests choosing h 1 in the form C

h 1 D Γ1 η 2 C Γ2 η ηN C Γ3 ηN 2

(1.66)

(1.67)

It follows from (1.66) that ηP D i ωη C O() and

ηPN D i ω ηN C O()

so that to O() (1.66) becomes δ δ 2 η η ηN C iω Γ C ηP D i ωη C iω Γ1 C 2 2ω 2 ω2 δ ηN 2 C O( 2 ) C iω 3Γ3 C 2ω 2

(1.68)

17

18

1 SDOF Autonomous Systems

The simplest possible form for (1.68) corresponds to the vanishing of the terms involving η 2 , η η, N and ηN 2 ; that is, choosing Γ1 , Γ2 , and Γ3 to be Γ1 D

δ , 2ω 2

Γ2 D

δ , ω2

Γ3 D

δ 6ω 2

(1.69)

Then, (1.68) reduces to ηP D i ωη C O( 2 )

(1.70)

PN we obtain Substituting (1.67) into (1.66) and using (1.70) to eliminate ηP and η, ηP D i ωη C

2δ 2 3 i 2 3 2 2 η C (1.71) C η N 5η η N 5η η N α (η C η) N 3C 2ω 3ω 2

Next, we introduce a near-identity transformation from η to ξ in the form η D ξ C 2 h 2 ξ , ξN

(1.72)

and obtain @h 2 P @h 2 NP ξ 2 ξ ξP D i ωξ C i 2 ωh 2 2 @ξ @ ξN

3 i 2 2δ 2 3 N 3 2N N2 C C ξ 5ξ ξ ξ C ξ 5ξ α ξ C ξN C 2ω 3ω 2

(1.73)

The form of the O( 2 ) terms suggests choosing h 2 in the form h 2 D Λ 1 ξ 3 C Λ 2 ξ 2 ξN C Λ 3 ξ ξN 2 C Λ 4 ξN 3

(1.74)

It follows from (1.73) that ξP D i ωξ C O( 2 )

and

ξPN D i ω ξN C O( 2 )

(1.75)

Therefore, substituting (1.74) and (1.75) into the right-hand side of (1.73) and keeping terms up to O( 2 ), we have α δ2 ξP D i ωξ C i 2 ω 2Λ 1 C ξ3 C 2ω 2 3ω 4 α δ2 i 2 10δ 2 2 3 N ξ C ξ 2 ξN C i ω 4Λ 4 C C 3α 2ω 2 3ω 4 2ω 3ω 2 3α 5δ 2 ξ ξN 2 C C i 2 ω 2Λ 3 C 2ω 2 3ω 4

(1.76)

We note that (1.76) is independent of Λ 2 and hence it is arbitrary and the term ξ 2 ξN is a resonance term. Equation 1.76 takes the simplest possible form if Λ1 D

α δ2 C , 4ω 2 6ω 4

Λ3 D

3α 5δ 2 C , 4ω 2 6ω 4

Λ4 D

α δ2 (1.77) 8ω 2 12ω 4

1.5 An Oscillator with Quadratic and Cubic Nonlinearities

Then, (1.76) becomes i 2 ξP D i ωξ C 2ω

3α

10δ 2 3ω 2

ξ 2 ξN C

(1.78)

Substituting (1.65) into (1.10), we have u D ζ C ζN D η C ηN C h 1 (η, η) N C hN 1 (η, η) N Then, substituting (1.72) into (1.79) yields u D ξ C ξN C h 1 ξ , ξN C hN 1 ξ , ξN C

(1.79)

(1.80)

Substituting for h 1 from (1.67) into (1.80) and using (1.69), we obtain δ 2 u D ξ C ξN C ξ 6ξ ξN C ξN 2 C 2 3ω

(1.81)

Substituting the polar form ξ D 12 ae i(ω tCβ) into (1.81), we ﬁnd that u D a cos (ωt C β) C

δ a 2 [cos (2ωt C 2β) 3] C 6ω 2

(1.82)

Substituting the polar form into (1.78) and separating real and imaginary parts, we have aP D 0

(1.83)

a βP D 2 a 3

3α 5δ 2 8ω 12ω 3

(1.84)

Equations 1.82–1.84 are in full agreement with those obtained by using the method of multiple scales, as shown in the next section. 1.5.2 The Method of Multiple Scales

Using the method of multiple scales, we seek a second-order uniform expansion of the solution of (1.64) in the form ζ(tI ) D

2 X

n ζ n (T0 , T1 , T2 ) C

(1.85)

nD0

where Tn D n t. In terms of these scales, the time derivative becomes d D D0 C D1 C 2 D2 C , dt

Dn D

@ @Tn

(1.86)

19

20

1 SDOF Autonomous Systems

Substituting (1.85) and (1.86) into (1.64) and equating coefﬁcients of like powers of , we obtain Order (0 )

D0 ζ0 i ωζ0 D 0

(1.87)

Order ()

D0 ζ1 i ωζ1 D D1 ζ0 C

2 iδ ζ0 C ζN 0 2ω

(1.88)

Order (2 )

D0 ζ2 i ωζ2 D D2 ζ0 D1 ζ1 C

3 iα iδ ζ0 C ζN 0 ζ1 C ζN 1 C ζ0 C ζN 0 ω 2ω (1.89)

The general solution of (1.87) can be expressed as ζ0 D A (T1 , T2 ) e i ω T0

(1.90)

where A is an undetermined function of T1 and T2 at this order; it is determined by eliminating the secular terms at the next orders of approximation. Substituting (1.90) into (1.88) yields D0 ζ1 i ωζ1 D D1 Ae i ω T0 C

i δ 2 2i ω T0 A e C 2A AN C AN2 e 2i ω T0 2ω

(1.91)

Eliminating the terms that produce secular terms in (1.91) demands that D1 A D 0 or A D A(T2 ). Then, the solution of (1.91) can be expressed as ζ1 D

δ A2 2i ω T0 δ A AN δ AN2 2i ω T0 e e 2ω 2 ω2 6ω 2

(1.92)

where the solution of the homogeneous equation has not been included so that the amplitude of the term at the frequency of oscillation is uniquely deﬁned by the zeroth-order solution (1.90). We note that the coefﬁcients in (1.92) are the same as the Γi deﬁned in (1.69). Substituting (1.90) and (1.92) into (1.89) and using the fact that D1 A D 0, we have 10δ 2 i N i ω T0 C NST (1.93) 3α D0 ζ2 i ωζ2 D D2 Ae i ω T0 C A2 Ae 2ω 3ω 2 where NST stands for the terms that do not produce secular terms. Eliminating the terms that produce secular terms from (1.93), we obtain 10δ 2 i 3α (1.94) A2 AN D2 A D 2ω 3ω 2

1.5 An Oscillator with Quadratic and Cubic Nonlinearities

Putting ξ D Ae i ω t in (1.78) and using the fact that D2 A D 2 d A/d t, we obtain exactly (1.94). We note that, for a uniform second approximation, we do not need to solve for ζ2 , but we only need to inspect (1.93) and eliminate the terms that produce secular terms. Similarly, to determine a uniform second approximation by using the method of normal forms, we do not need to determine h 2 in (1.72), but we need only keep the resonance terms in (1.73). 1.5.3 A Single Transformation

Instead of using successive transformations to produce the normal form of (1.64), one can formulate the process as a perturbation method. Thus, we expand ζ in a power series of in terms of a new variable η in the form ζ D η C h 1 (η, η) N C 2 h 2 (η, η) N C

(1.95)

ηP D i ωη C g 1 (η, η) N C 2 g 2 (η, η) N C

(1.96)

where h 1 and h 2 are smooth functions of η and ηN and g 1 and g 2 contain all of the resonance and near-resonance terms. Substituting (1.95) and (1.96) into (1.64) and equating coefﬁcients of like powers of , we obtain iδ @h 1 @h 1 (η C η) g1 C i ω η (1.97) ηN h1 D N 2 @η @ ηN 2ω @h 1 @h 1 @h 2 @h 2 iα (η C η) g2 C i ω η ηN h 2 D g 1 gN 1 C N 3 @η @ ηN @η @ ηN 2ω iδ (1.98) (η C η) C N h 1 C hN 1 ω Equations 1.97 and 1.98 are the so-called homology equations for h 1 and h 2 . Next, we need to determine g 1 and h 1 from (1.97). In order that h 1 be nonsingular (smooth), we choose g 1 to eliminate all of the resonance and near-resonance terms; otherwise, h 1 will be singular (i.e., have secular terms) if there are resonance terms and near singular (i.e., have small divisors) if there are near-resonance terms. In the present case, there are no resonance terms in (1.97). To see this, we note from (1.96) that η D B e i ω t and hence the perturbation terms on the right-hand side of (1.97) contain terms proportional to e ˙2i ω t and a constant. Because none of these terms is proportional to e i ω t , which is the solution of the ﬁrst-order problem in (1.96), there are no resonance terms. If we are in doubt, we seek a function h 1 that can be used to eliminate all of the perturbation terms. If we are successful in ﬁnding a smooth function h 1 that eliminates all of the perturbation terms, then g 1 D 0. Otherwise, we choose g 1 to eliminate all terms that produced the troublesome terms (i.e., singular and near-singular terms) in h 1 . Because the perturbation terms are of second degree, we let h 1 D Γ1 η 2 C Γ2 η ηN C Γ3 ηN 2

(1.99)

21

22

1 SDOF Autonomous Systems

choose Γ1 , Γ2 , and Γ3 to eliminate i δ(η C η) N 2 /2ω and obtain (1.69). Because the obtained Γm are regular, there are no resonance terms and g 1 D 0. Substituting (1.99) and g 1 D 0 into (1.98) and using (1.69), we obtain i δ2 @h 2 @h 2 (η C η) N η 2 C ηN 2 6η ηN g2 C i ω η ηN h2 D 3 @η @ ηN 3ω iα (η C η) N 3 C 2ω

(1.100)

As stated earlier, we do not need to determine h 2 and all that we need is to inspect (1.100) and choose g 2 to eliminate all of the resonance and near-resonance terms. Again, because η / e i ω t , only the term proportional to η 2 ηN is a resonance term. Hence, choosing g 2 to eliminate this term, we obtain i g2 D 2ω

10δ 2 3α 3ω 2

η 2 ηN

(1.101)

Substituting (1.95) and (1.99) into (1.10) and using (1.69), we obtain u D η C ηN C

δ 2 η C ηN 2 6η ηN C 2 3ω

(1.102)

Substituting for g 2 from (1.101) into (1.96) and using the fact that g 1 D 0, we have ηP D i ωη C

i 2 2ω

10δ 2 3α η 2 ηN C 3ω 2

(1.103)

in agreement through second order with the expansions obtained in Sections 1.5.1 and 1.5.2.

1.6 A General System with Quadratic and Cubic Nonlinearities

We consider free oscillations of a single-degree-of-freedom system governed by uR C ω 2 u C δ 1 u2 C δ 2 uP 2 C 2 2µ uP C α 1 u3 C α 2 u2 uP C α 3 u uP 2 C α 4 uP 3 D 0

(1.104)

so that f D δ 1 u2 C δ 2 uP 2 2 2µ uP C α 1 u3 C α 2 u2 uP C α 3 u uP 2 C α 4 uP 3 (1.105) We note that a small nondimensional parameter has been introduced as a bookkeeping device. The quadratic terms have been ordered as O(), whereas the cubic terms and the linear damping term have been ordered as O( 2 ). Then, (1.14) be-

1.6 A General System with Quadratic and Cubic Nonlinearities

comes

2 2 i i h ζP D i ωζ C 2 µ ζ ζN δ 1 ζ C ζN δ 2 ω 2 ζ ζN 2ω 3 2 i 2 h ζ ζN α 1 ζ C ζN C i ωα 2 ζ C ζN C 2ω 2 3 i ω 2 α 3 ζ C ζN ζ ζN i ω 3 α 4 ζ ζN

(1.106)

As in Section 1.5.3, we seek an expansion for (1.106) in the form (1.95) and (1.96), equate coefﬁcients of like powers of , and obtain i @h 1 @h 1 N 2 δ 2 ω 2 (η η) N 2 ηN h1 D δ 1 (η C η) g1 C i ω η @η @ ηN 2ω (1.107) @h 1 @h 1 @h 2 @h 2 g2 C i ω η ηN h 2 D g 1 gN 1 µ (η η) N @η @ ηN @η @ ηN i i i h N h 1 C hN 1 δ 2 ω 2 (η η) N h 1 hN 1 C N 3 δ 1 (η C η) α 1 (η C η) C ω 2ω i i ωα 2 (η C η) N 2 (η η) N i ω 3 α 4 (η η) N 3 ω 2 α 3 (η C η) N (η η) N 2 C 2ω (1.108) The right-hand side of (1.107) does not contain resonance or near-resonance terms, and hence we put g 1 D 0, seek h 1 in the form (1.99), and obtain δ1 1 δ1 δ1 1 δ 2 , Γ2 D 2 δ 2 , Γ3 D 2 C δ 2 (1.109) 2ω 2 2 ω 6ω 6 Because all of the Γm are regular, our conclusion that there are no resonance or near-resonance terms in (1.107) and hence g 1 D 0 is justiﬁed a posteriori. Substituting (1.99) and (1.109) into (1.108) and using the fact that g 1 D 0, we obtain 2 @h 2 δ1 @h 2 (η η) N η 2 ηN 2 g2 C i ω η δ ηN h2 D i δ2 ω 2 2 @η @ ηN 3 ω

2 δ1 i δ1 δ1 2 (η (η µ η) N C δ 2 η C ηN 6 C δ 2 η ηN C η) N 3ω ω2 ω2 i α 1 (η C η) N 3 C i ωα 2 (η C η) N 2 (η η) N ω 2 α 3 (η C η) N (η η) N 2 C 2ω (1.110) N 3 C i ω 3 α 4 (η η) Inspecting the right-hand side of (1.110), we conclude that the terms proportional to η and η 2 ηN are the only resonance terms and there are no near-resonance terms. Consequently, choosing g 2 to eliminate the resonance terms, we obtain i 2 5δ 21 2 2 3α 1 C ω 2 α 3 g 2 D µ η C C 5δ δ C 2δ ω 1 2 2 2ω 3 ω2

Ci ω α 2 C 3ω 2 α 4 η 2 ηN (1.111) Γ1 D

23

24

1 SDOF Autonomous Systems

Substituting (1.95) and (1.99) into (1.10) and using (1.109), we obtain

2 2δ 1 1 δ1 2 η η ηN C (1.112) C η N C 2δ δ u D ηC ηC N 2 2 3ω 2 3 ω2 Substituting for g 2 from (1.111) into (1.96) and using the fact that g 1 D 0, we obtain i 2 2 5δ 21 2 2 2 3α 1 C ω α 3 C 5δ 1 δ 2 C 2δ 2 ω ηP D i ωη µ η C 2ω 3 ω2

Ci ω α 2 C 3ω 2 α 4 η 2 ηN (1.113) 2

To compare the expansion (1.112) and (1.113) with that obtained by using the method of multiple scales (Nayfeh, 1984), we substitute the polar form η D 12 ae i(ω tCβ)

(1.114)

in (1.112) and (1.113) and obtain 1 u D a cos (ωt C β) C a 2 6

δ1 δ1 (2ωt cos C 2β) 3 δ C δ 2 2 ω2 ω2

C

(1.115)

where aP D 2 µ a 18 2 α 2 C 3ω 2 α 4 a 3 C a βP D

2 5δ 1 2 2 2 3α 1 C ω 2 α 3 a3 C C 5δ δ C 2δ ω 1 2 2 8ω 3 ω2

(1.116) (1.117)

which is formally equivalent to that obtained by using the method of multiple scales.

1.7 The van der Pol Oscillator

In this section, we construct a second-order approximation of the normal form of the van der Pol oscillator uR C ω 2 u D (1 u2 ) uP

(1.118)

Using the transformation (1.10), we rewrite (1.118) as N 1 (ζ C ζ) N 2 ζP D i ωζ C 12 (ζ ζ)

(1.119)

1.7 The van der Pol Oscillator

1.7.1 The Method of Normal Forms

As in the preceding two sections, we seek a second-order expansion of (1.119) in the form (1.95) and (1.96), equate coefﬁcients of like powers of , and obtain 1 @h 1 @h 1 1 3 η C η 2 ηN ηN 2 η ηN 3 ηN h 1 D (η η) N g1 C i ω η @η @ ηN 2 2 (1.120) @h 1 @h 1 @h 2 @h 2 1 ηN h 2 D g 1 gN 1 C h 1 hN 1 g2 C i ω η @η @ ηN @η @ ηN 2 3 2 1 1 3 2 2 2 η h 1 η η(h N 1 hN 1 ) η hN 1 C ηN h 1 C ηN hN 1 (1.121) 2 2 2 2 Choosing g 1 to eliminate the resonance terms in (1.120), we have g 1 D 12 η 12 η 2 ηN

(1.122)

Then, we seek h 1 in the form h 1 D ∆ 1 ηN C Λ 1 η 3 C Λ 2 η ηN 2 C Λ 3 ηN 3

(1.123)

Substituting (1.123) and (1.122) into (1.120) yields 2i ω∆ 1 12 ηN 2i ωΛ 1 C 12 η 3 C 2i ωΛ 2 C 12 η ηN 2 C 4 i ωΛ 3 C 12 ηN 3 D 0

(1.124)

Hence, ∆1 D Therefore, h1 D

i , 4ω

Λ1 D

i , 4ω

Λ2 D

i , 4ω

1 1 i ηN η 3 η ηN 2 ηN 3 4ω 2

Λ3 D

i 8ω

(1.125)

(1.126)

Substituting (1.122) and (1.126) into (1.121) yields i @h 2 @h 2 g2 C i ω η ηN h2 D 2η C 5η 3 C 5η 5 12η 2 ηN @η @ ηN 16ω (1.127) 2η 4 ηN 4η ηN 2 C 11η 3 ηN 2 C 2 ηN 3 C 5η 2 ηN 3 C η ηN 4 C 5 ηN 5 Choosing g 2 to eliminate the resonance terms (terms proportional to η, η 2 η, N and η 3 ηN 2 ) from (1.127), we have 1 (1.128) i 2η 12η 2 ηN C 11η 3 ηN 2 16ω Substituting (1.122) and (1.128) into (1.96), we obtain, to the second approximation, the normal form 1 1 ηP D i ωη C η η 2 ηN (1.129) i 2 2η 12η 2 ηN C 11η 3 ηN 2 2 16ω g2 D

25

26

1 SDOF Autonomous Systems

Substituting the polar form (1.27) into (1.129) and separating real and imaginary parts, we obtain aP D 12 a 14 a 3 1 2 3 11 4 βP D 1 a2 C a 8ω 2 32

(1.130) (1.131)

in agreement with those obtained with the generalized method of averaging (Nayfeh, 1973). 1.7.2 The Method of Multiple Scales

We seek a second-order expansion of the solution of (1.119) in the form (1.85). Substituting (1.85) and (1.86) into (1.119) and equating coefﬁcients of like powers of , we obtain Order (0 )

D0 ζ0 i ωζ0 D 0

(1.132)

Order ()

D0 ζ1 i ωζ1 D D1 ζ0 C

1 2

ζ0 ζN 0 12 ζ03 C ζ02 ζN 0 ζ0 ζN 02 ζN 03 (1.133)

Order (2 )

D0 ζ2 i ωζ2 D D2 ζ0 D1 ζ1 C 12 ζ1 ζN 1 12 3ζ02 C 2ζ0 ζN 0 ζN 02 ζ1 1 ζ 2 2ζ0 ζN 0 3 ζN 2 ζN 1 2

0

0

(1.134) The general solution of (1.132) can be expressed as in (1.90). Then, (1.133) becomes N i ω T0 1 A3 e 3i ω T0 D0 ζ1 i ωζ1 D D1 Ae i ω T0 C 12 Ae i ω T0 12 Ae 2 N i ω T0 C 1 A AN2 e i ω T0 C 1 AN3 e 3i ω T0 12 A2 Ae 2 2

(1.135)

Eliminating the terms that lead to secular terms from (1.135) yields D1 A D

1 2

A A2 AN

(1.136)

Then, the solution of (1.135) can be expressed as ζ1 D

i N i ω T0 C 2A3 e 3i ω T0 C 2A AN2 e i ω T0 C AN3 e 3i ω T0 2 Ae 8ω

(1.137)

1.8 Exercises

Substituting (1.90), (1.136), and (1.137) into (1.134), we have D0 ζ2 i ωζ2 D D2 Ae i ω T0

i 2A 12A2 AN C 11A3 AN2 e i ω T0 16ω

C NST

(1.138)

Eliminating the terms that lead to secular terms from (1.138) yields D2 A D

i 2A 12A2 AN C 11A3 AN2 16ω

(1.139)

Using the method of reconstitution, we obtain from (1.136) and (1.139) that 1 i 2 AP D A A2 AN 2A 12A2 AN C 11A3 AN2 2 16ω

(1.140)

Letting ζ D Ae i ω t in (1.129), we obtain (1.140), which means that the results obtained with the methods of normal forms and multiple scales are the same.

1.8 Exercises

1.8.1 Use the methods of normal forms and multiple scales to determine the normal forms of a) b) c) d) e)

uR C ω 2 u C α uP 3 D 0 , uR C ω 2 u C α u2 uP D 0 , uR C ω 2 u C α u5 D 0 , uR C ω 2 u C α u3 uP 2 D 0 , uR C ω 2 u C uP 5 D 0 .

1.8.2 Use the methods of multiple scales and normal forms to construct a secondorder approximation to the normal form of uR C ω 2 u C α u3 C 2µ u2 uP D 0 1.8.3 Use the methods of multiple scales and normal forms to construct a ﬁrstorder approximation to the normal form of uR C ω 2 u C α u2 uR D 0 1.8.4 Use the methods of normal forms and multiple scales to construct a secondorder approximation to the normal form of uR C ω 2 u C α uP 13 uP 3 D 0

27

28

1 SDOF Autonomous Systems

1.8.5 Consider the equation xR C x C 3x 2 C 2x 3 D 0 Determine the equilibrium points. Determine, to second order, the normal form of the system near each of these equilibrium points. 1.8.6 Consider xR C x C ax 2 C 2x 3 D 0 p Show that there is only one equilibrium p point when a < 2 2 and that there are three equilibrium points when a > 2 2. Determine, to second order, the normal forms near the equilibrium points. 1.8.7 Consider xR C x

a D0 1x

Show that there is only one equilibrium point when a 1/4. Determine, to second order, the normal form near this equilibrium point. 1.8.8 Consider xR 3x C x 3 D 2 Determine the equilibrium points and the normal forms near them. 1.8.9 Consider uR u C u4 D 0 Determine the equilibrium points and the normal forms near them. 1.8.10 Consider xR 2x x 2 C x 3 D 0 Determine the equilibrium points and the normal forms near them. 1.8.11 Consider uR C u

3 D0 16(1 u)

Determine the equilibrium points and the normal forms near them. 1.8.12 Use the methods of multiple scales and normal forms to determine a ﬁrstorder uniform expansion for the general solution of θR C ω 2 sin θ C for small but ﬁnite θ .

4 sin2 θ θP D 0 1 C 4(1 cos θ )

1.8 Exercises

1.8.13 Consider the equation uR C ω 20 u C

µ uP D0 1 u2

Use the methods of multiple scales and normal forms to determine a ﬁrst-order uniform expansion for small u. 1.8.14 Consider the equation

2 l 2 C r 2 2r l cos θ θR C r l sin θ θP C g l sin θ D 0

where g, r, and l are constants. Determine a ﬁrst-order expansion for small but ﬁnite θ by using the methods of multiple scales and normal forms. 1.8.15 Consider the equation 1 12

2 l 2 C r 2 θ 2 θR C r 2 θ θP C g r θ cos θ D 0

where r, l, and g are constants. Determine a ﬁrst-order uniform expansion for small but ﬁnite θ by using the methods of multiple scales and normal forms. 1.8.16 The motion of a simple pendulum is governed by g θR C sin θ D 0 l Use the methods of multiple scales and normal forms to determine a ﬁrst-order uniform expansion for small but ﬁnite θ . 1.8.17 Consider the equation g θR D Ω 2 sin θ cos θ sin θ R Use the methods of multiple scales and normal forms to determine a ﬁrst-order uniform expansion for small but ﬁnite θ . 1.8.18 The motion of a particle on a rotating parabola is governed by 1 C 4p 2 x 2 xR C Λ x C 4p 2 xP 2 D 0 where p and Λ are constants. Use the methods of multiple scales and normal forms to determine a ﬁrst-order expansion for small but ﬁnite x. 1.8.19 Consider the equation u uP 2 u2 g u uR C 1C C ω 20 u C p D0 2 1u (1 u2 )2 l 1 u2 Use the methods of multiple scales and normal forms to determine a ﬁrst-order expansion for small u.

29

31

2 Systems of First-Order Equations 2.1 Introduction

In this chapter, we describe in detail the problem of calculating the normal form for a system of ﬁrst-order equations having the form uP D Au C F 1 (u) C 2 F 2 (u) C

(2.1)

where u and the F m are column vectors of length n, A is an n n constant matrix, and is a small nondimensional parameter, which may represent a physical quantity, or it may be just a bookkeeping device and set equal to unity in the ﬁnal result. In this book, the normalization is carried out in terms of , whereas in the literature the normalization is usually carried out in terms of the degree of the polynomials in the nonlinear terms. In the latter case, the components of F m (u) are homogeneous polynomials of degree m C 1 in u. We assume that the F m are smooth vector ﬁelds satisfying F m (0) D 0 so that u D 0 is a ﬁxed point of (2.1). As a ﬁrst step, we introduce a linear transformation u D P x, where P is a nonsingular or invertible matrix, in (2.1) and obtain P xP D AP x C F 1 (P x) C 2 F 2 (P x) C

(2.2)

Multiplying (2.2) from the left by P 1 , the inverse of P, we obtain xP D J x C f 1 (x) C 2 f 2 (x) C

(2.3)

J D P 1 AP

(2.4)

where and

f m (x) D P 1 F m (P x)

We choose P so that J has a simple real form. The simplest possible real form for J depends on the eigenvalues λ i and eigenvectors p i of A. There are three possibilities (Coddington and Levinson, 1955; Hale, 1980; Hirsch and Smale, 1974; Walter, 1976): The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

32

2 Systems of First-Order Equations

a) All eigenvalues are real and distinct. b) Some eigenvalues occur in complex conjugate pairs. c) Some eigenvalues are repeated. In the ﬁrst case, there are n linearly independent real eigenvectors p i of A. Then, forming the matrix P by having its columns be the eigenvectors p 1 , p 2 , . . . , p n produces a diagonal matrix J with real entries λ 1 , λ 2 , . . . , λ n . For two-dimensional systems, λ1 0 JD 0 λ2 When the eigenvalues are real and some of them are repeated, there are two possibilities. First, there are n linearly independent real eigenvectors p i . Then, choosing P as in the ﬁrst case produces a diagonal matrix with real entries λ 1 , λ 2 , . . . , λ n , where some of them are repeated. For two-dimensional systems, λ 0 JD 0 λ Second, there are less than n linearly independent eigenvectors. In this case, we form the matrix P by having its columns p i be the generalized eigenvectors of A; that is, the eigenvectors of (A λ j I ) m , where m is the multiplicity of the eigenvalue λ j . With this choice for P, J takes the so-called Jordan form. For twodimensional systems, λ 1 λ 0 JD or J D 0 λ 1 λ When some of the eigenvalues occur in complex conjugate pairs, some of the eigenvectors are complex. Forming the matrix P by having its columns be the generalized real eigenvectors of A produces a matrix J that is in the so-called real Jordan form. For two-dimensional systems, α ω α ω JD or J D ω α ω α where α and ω are real constants. The ﬁxed point u D 0 is called hyperbolic if none of the eigenvalues of A has a zero real part and nonhyperbolic if at least one of the eigenvalues of A has a zero real part. The zero real parts imply the existence of constraints on the elements of A. For two-dimensional systems, there are four possible constraints: 1. A has real eigenvalues and one of them is zero; that is, jAj D 0 and Tr(A) ¤ 0 where Tr(A) is the trace of A.

2.1 Introduction

2. A has purely imaginary eigenvalues; that is, jAj ¤ 0

and Tr(A) D 0

3. Both eigenvalues of A are zero but A is not the null matrix; that is, A¤0,

jAj D 0 ,

and Tr(A) D 0

4. A D 0. The equality constraints are called degeneracy conditions. The number of degeneracy conditions indicates the level of degeneracy or codimension of the ﬁxed point (Dumortier, 1977; Guckenheimer and Holmes, 1983). Our aim is to construct a sequence of transformations that successively remove the perturbation terms f m , starting from f 1 . Ideally, we would like to be able to remove all of the f m , especially, if they are nonlinear, thereby reducing (2.3) to a linear problem. In general, this is not possible, as discussed below. Instead of using a sequence of transformations, we introduce the single nearidentity transformation x D y C h 1 (y) C 2 h 2 (y) C

(2.5)

into (2.3) and choose the h m so that it takes the simplest possible form, the so-called normal form yP D J y C g 1 (y) C 2 g 2 (y) C

(2.6)

As discussed later, the g n are referred to as resonance and near-resonance terms. Substituting (2.5) into (2.3) yields yP C D h 1 (y) yP C 2 D h 2 (y) yP C

D J y C J h 1 (y) C 2 J h 2 (y) C C f 1 y C h 1 (y) C 2 h 2 (y) C C 2 f 2 y C h 1 (y) C 2 h 2 (y) C C (2.7)

where D h is the Jacobian of h. Using (2.6) to eliminate yP from (2.7) and equating coefﬁcients of like powers of , we obtain g 1 (y) C D h 1 (y) J y J h 1 (y) D f 1 (y)

(2.8)

g 2 (y) C D h 2 (y) J y J h 2 (y) D f 2 (y) C D f 1 (y)h 1 (y) D h 1 (y)g 1 (y) (2.9) The operator L(h 1 (y)) D D h 1 (y) J y J h 1 (y) D [h 1 , J y] is called the Lie or Poisson bracket. We note that although the h n (y), which describe the transformation (2.5), are in general nonlinear functions of y, they are found by solving a sequence of linear problems, as described below. Next, we carry out the details of determining the h m and g m for speciﬁc problems.

33

34

2 Systems of First-Order Equations

2.2 A Two-Dimensional System with Diagonal Linear Part

As a ﬁrst example, we consider the two-dimensional system xP D

α x 2 C α 2 x1 x2 C α 3 x22 0 x C 1 12 λ2 α 4 x1 C α 5 x1 x2 C α 6 x22

λ1 0

(2.10)

where the λ i and α i may be real or complex, but both of λ 1 and λ 2 are not zero. Using (2.10), we rewrite (2.8) as 2

g 11 g 12

D

@h 11 6 @y 1 6 C6 4 @h 12 @y 1

α 1 y 12 α 4 y 12

3 @h 11 @y 2 7 7 λ1 7 @h 12 5 0 @y 2

C α 2 y 1 y 2 C α 3 y 22 C α 5 y 1 y 2 C α 6 y 22

0 λ2

y1 λ 1 0 y2

0 λ2

h 11 h 12

(2.11)

where (h 11 , h 12 ) and (g 11 , g 12 ) are the components of h 1 and g 1 . The right-hand side of (2.11) suggests seeking the g 1m and h 1m in the form h 11 D Γ1 y 12 C Γ2 y 1 y 2 C Γ3 y 22

(2.12)

h 12 D Γ4 y 12 C Γ5 y 1 y 2 C Γ6 y 22

(2.13)

g 11 D Λ 1 y 12 C Λ 2 y 1 y 2 C Λ 3 y 22

(2.14)

g 12 D Λ 4 y 12 C Λ 5 y 1 y 2 C Λ 6 y 22

(2.15)

Substituting (2.12)–(2.15) into (2.11) yields λ 1 y 1 (2Γ1 y 1 C Γ2 y 2 ) C λ 2 y 2 (Γ2 y 1 C 2Γ3 y 2 ) λ 1 Γ1 y 12 C Γ2 y 1 y 2 C Γ3 y 22 D (α 1 Λ 1 ) y 12 C (α 2 Λ 2 ) y 1 y 2 C (α 3 Λ 3 ) y 22

(2.16)

λ 1 y 1 (2Γ4 y 1 C Γ5 y 2 ) C λ 2 y 2 (Γ5 y 1 C 2Γ6 y 2 ) λ 2 Γ4 y 12 C Γ5 y 1 y 2 C Γ6 y 22 D (α 4 Λ 4 ) y 12 C (α 5 Λ 5 ) y 1 y 2 C (α 6 Λ 6 ) y 22

(2.17)

Equating the coefﬁcients of y 12 , y 1 y 2 , and y 22 on both sides of (2.16) and (2.17) yields BΓ D α Λ

(2.18)

2.2 A Two-Dimensional System with Diagonal Linear Part

where Γ , α, and Λ are column vectors having the components Γm , α m , and Λ m , respectively, and 2 3 0 0 0 0 λ1 0 60 λ 0 0 0 07 2 6 7 6 7 0 2λ 2 λ 1 0 0 07 60 BD6 (2.19) 7 60 0 0 2λ 1 λ 2 0 07 6 7 40 0 0 0 λ1 0 5 0 0 0 0 0 λ2 Thus, the matrix B is diagonal having the eigenvalues λ 1 , λ 2 , 2λ 2 λ 1 , and 2λ 1 λ 2 . As shown in Section 2.4, these eigenvalues are related to the eigenvalues λ 1 and λ 2 of J and hence A by λ m,i D m 1 λ 1 C m 2 λ 2 λ i

(2.20)

where m D (m 1 , m 2 ), m 1 and m 2 are positive integers such that m 1 C m 2 D 2 and m 1 , m 2 D 0, 1, and 2. Clearly, the matrix B has an inverse and hence (2.18) has a solution for any α if none of its eigenvalues vanishes. Consequently, we put Λ D 0, explicitly solve (2.18) for the Γm , and obtain α1 α2 α3 , Γ2 D , Γ3 D , λ1 λ2 2λ 2 λ 1 α4 α5 α6 Γ4 D , Γ5 D , Γ6 D 2λ 1 λ 2 λ1 λ2 Γ1 D

(2.21)

With this choice, all of the nonlinear terms are eliminated from (2.10), resulting in the linear normal form yP D J y. When any of the eigenvalues λ 1 , λ 2 , 2λ 2 λ 1 , and 2λ 1 λ 2 vanishes (i.e., m 1 λ 1 C m 2 λ 2 D λ i , for some choice of m 1 and m 2 ), the matrix B is singular and one or more of the Γm are unbounded, implying that one cannot eliminate all of the nonlinear terms in (2.10) and hence obtain a linear normal form. The terms that cannot be eliminated are called resonance terms. Moreover, the condition m 1 λ 1 Cm 2 λ 2 D λ i is called a resonance condition of order 2 because the nonlinearity is quadratic (i.e., m 1 C m 2 D 2). When any of the eigenvalues of B is small, it follows from (2.21) that at least one of the Γm has a small divisor and therefore the transformation breaks down. The terms that produce small divisors are called near-resonance terms. For example, let us consider the resonance condition 2λ 2 λ 1 0. In this case, it follows from (2.21) that all Γm except Γ3 are regular and that Γ3 is either singular or has a small divisor. Thus, we choose Λ 3 D α 3 to avoid this singularity, put the remaining Λ m equal to zero, and choose Γ1 , Γ2 , Γ4 , Γ5 , and Γ6 as in (2.21). Consequently, the normal form of (2.10) is λ α y2 0 yP D 1 y C 3 2 (2.22) 0 λ2 0 When 2λ 1 λ 2 0, we choose Λ 4 D α 4 to avoid the singularity, set the remaining Λ m equal to zero, and choose Γ1 , Γ2 , Γ3 , Γ5 , and Γ6 as in (2.21). In this case, the

35

36

2 Systems of First-Order Equations

normal form of (2.10) is yP D

λ1 0

0 0 y C α 4 y 12 λ2

(2.23)

When λ 1 0, we choose Λ 1 D α 1 and Λ 5 D α 5 to avoid the singularities, set the remaining Λ m equal to zero, and choose Γ2 , Γ3 , Γ4 , and Γ6 as in (2.21). The resulting normal form of (2.10) is yP D

λ1 0

α 1 y 12 0 y C λ2 α5 y1 y2

(2.24)

When λ 2 0, we choose Λ 2 D α 2 and Λ 6 D α 6 , set the remaining Λ m equal to zero, and choose Γ1 , Γ3 , Γ4 , and Γ5 as in (2.21). Consequently, the normal form of (2.10) becomes yP D

λ1 0

α y y 0 y C 2 1 22 λ2 α6 y2

(2.25)

Recapping, we note that the 6 6 matrix B maps six-dimensional vectors in the space 0 and for µ > 0 if f µ f x x < 0. Then, it follows from (3.80) that the upper branch is stable and the lower branch is unstable if f x x < 0 and that the upper branch is unstable and the lower branch is stable if f x x > 0. This bifurcation of the nonhyperbolic ﬁxed point at the origin as µ passes through zero is called fold or tangent or saddle-node bifurcation. Example 3.9 We consider the one-dimensional map x kC1 D x k C µ x k2

(3.81)

where µ is a scalar control parameter. When µ > 0, it is clear from Figure 3.1a that this map intersects the identity map x kC1 D x k in two points. In other words, for µ > 0, (3.81) has the nontrivial ﬁxed points x1 D

p

µ

and

p x2 D µ

The Jacobian matrix associated with the ﬁxed point x j 0 has the single eigenvalue D 1 2x j The ﬁxed point x2 is an unstable node for all µ > 0 because jj > 1. On the other hand, the ﬁxed point x1 is a stable node for 0 < µ < 1 because jj < 1. When µ is decreased to zero, the two ﬁxed points approach each other and coalesce at the single ﬁxed point x D 0, as shown in Figure 3.1b. The map (3.81) is tangent to the identity map and, hence, the associated bifurcation is called tangent bifurcation. When µ < 0, the map (3.81) does not intersect the identity map, as shown in Figure 3.1c, and hence for µ < 0, (3.81) does not have any ﬁxed points. In Figure 3.2, we show the different ﬁxed points of (3.81) and their stability in the vicinity of the origin of the x µ space. Broken and solid lines are used to represent branches

77

78

3 Maps

0 f(x) –0.5

(a)

–1 –0.5

0 x 0.5

0.5

–0.5

0 x

(b)

0.5

0 f(x) –0.5

(c)

–1 –0.5

0 x

0.5

Figure 3.1 The functions f (x) D µ C x x 2 and x for (a) µ D 0.1, (b) µ D 0.0, and (c) µ D 0.1.

x 0

0 μ Figure 3.2 Scenario in the vicinity of a saddle-node bifurcation.

of unstable and stable ﬁxed points, respectively. A saddle-node or tangent bifurcation occurs at (x, µ) D (0, 0). Diagrams such as Figure 3.2 in which the variation of solutions and their stability are displayed in the state-control space are called bifurcation diagrams. In the bifurcation diagram, a branch of stable solutions is called a stable branch and a branch of unstable solutions is called an unstable branch. In most situations, a branch of solutions either ends or begins at a bifurcation point.

3.4 Local Bifurcations

3.4.2 Transcritical Bifurcation

When f µ D 0 and f x x D 6 0, the limit of (3.77) as x ! 0 and µ ! 0 is x kC1 D x k C 12 f µ µ µ 2 C 2 f x µ x k µ C f x x x k2

(3.82)

whose ﬁxed points are q q f x µ C f x2µ f µ µ f x x f x µ f x2µ f µ µ f x x x1 D µ and x2 D µ fxx fxx (3.83) provided that f x2µ f µ µ f x x > 0. It follows from (3.83) that there are two curves of ﬁxed points in the neighborhood of (x, µ) D (0, 0), which intersect transversely at (0, 0). The multipliers associated with these ﬁxed points are q q (x1 ) D 1 C f x2µ f µ µ f x x µ and (x2 ) D 1 f x2µ f µ µ f x x µ (3.84) Therefore, x1 is stable when µ < 0 and unstable when µ > 0. On the other hand, x2 is unstable when µ < 0 and stable when µ > 0. In this case, we have two branches of ﬁxed points that interchange stability at (0, 0). The bifurcation of the nonhyperbolic ﬁxed point at the origin as µ passes through zero is called transcritical bifurcation. Example 3.10 We consider the one-dimensional map x kC1 D x k C µ x k x k2

(3.85)

where µ is again a scalar control parameter. This map has the two ﬁxed points x1 D 0W

trivial ﬁxed point

x2

nontrivial ﬁxed point .

D µW

For the ﬁxed point x j , the Jacobian matrix has the single eigenvalue D 1 C µ 2x j Hence, it follows that the trivial ﬁxed point is stable for 2 < µ < 0 and unstable for all µ > 0. On the other hand, the nontrivial ﬁxed point is unstable for all µ < 0 and stable for 0 < µ < 2. The scenario in the vicinity of (x, µ) D (0, 0) is illustrated in Figure 3.3. A transcritical bifurcation occurs at the origin.

79

80

3 Maps

x 0

0 μ Figure 3.3 Scenario in the vicinity of a transcritical bifurcation.

3.4.3 Pitchfork Bifurcation

When f µ D 0, f x x D 0, f x µ ¤ 0, and f x x x ¤ 0, the limit of (3.77) as x ! 0 and µ ! 0 is x kC1 D x k C f x µ x k µ C

1 6

f x x x x k3

(3.86)

whose ﬁxed points are s x1

D0,

x2

D

6 f x µ µ , fxxx

s and

x3

D

6 f x µ µ fxxx

(3.87)

It follows from (3.87) that there are three branches of solutions. The ﬁrst branch exists for all values of µ and the other two branches exist for µ > 0 if f x µ f x x x < 0 and for µ < 0 if f x µ f x x x > 0. The multipliers associated with these ﬁxed points are (x1 ) D 1 C f x µ µ ,

(x2 ) D 1 2 f x µ µ ,

and

(x3 ) D 1 2 f x µ µ (3.88)

When f x µ > 0, the trivial ﬁxed point is stable when µ < 0 and unstable when µ > 0. When f x x x < 0, the two nontrivial ﬁxed points exist and are stable when µ > 0. This bifurcation of the nonhyperbolic ﬁxed point at the origin as µ passes through zero is called supercritical pitchfork bifurcation. On the other hand, when f x x x > 0, the two nontrivial ﬁxed points exist and are unstable when µ < 0. This bifurcation of the nonhyperbolic ﬁxed point at the origin as µ passes through zero is called subcritical or reverse pitchfork bifurcation. Example 3.11 We consider the one-dimensional map x kC1 D x k C µ x k C α x k3

(3.89)

3.4 Local Bifurcations

x0

0 μ

(a)

(b)

0 μ

Figure 3.4 Local scenarios: (a) supercritical pitchfork bifurcation and (b) subcritical pitchfork bifurcation.

where µ is a scalar control parameter. This map has the ﬁxed points x1 D 0W

trivial ﬁxed point

x2,3

nontrivial ﬁxed points.

p D ˙ µ/αW

The Jacobian matrix associated with the ﬁxed point x j has the single eigenvalue D 1 C µ C 3α x 2 j Therefore, the trivial ﬁxed point is stable for 2 < µ < 0 and unstable for all µ > 0. For α < 0, nontrivial ﬁxed points exist only for µ > 0, and they are stable for 0 < µ < 1. For α > 0, nontrivial ﬁxed points exist only for µ < 0, and they are unstable. The scenarios for α D 1 and α D 1 near the origin (x, µ) D (0, 0) are shown in Figure 3.4a and b, respectively. There is a supercritical pitchfork bifurcation at the origin in Figure 3.4a and a subcritical pitchfork bifurcation at the origin in Figure 3.4b.

3.4.4 Flip or Period-Doubling Bifurcation

As in the cases of fold, transcritical, and pitchfork bifurcations, analysis of the dynamics of an n-dimensional map in the neighborhood of a nonhyperbolic ﬁxed point with one eigenvalue being equal to 1 and the remaining eigenvalues being inside the unit circle can be reduced, by using the center-manifold theorem, to the analysis of the dynamics of the following one-dimensional map on the center manifold: x kC1 D f (x k I µ)

(3.90)

We assume that the ﬁxed point is at x D 0 when µ D 0 and that f (0, 0) D 0 and f x (0, 0) D 1.

81

82

3 Maps

Again, we expand f (xI µ) in a Taylor series for small x and µ and obtain x kC1 D x k C f µ µ C 12 f µ µ µ 2 C 2 f x µ x k µ C f x x x k2 C 16 f µ µ µ µ 3 C 3 f x µ µ x k µ 2 C 3 f x x µ x k2 µ C f x x x x k3 C

(3.91)

Therefore, the ﬁxed points of the map (3.91) are solutions of 2x C f µ µ C 12 f µ µ µ 2 C 2 f x µ x µ C f x x x 2 C 16 f µ µ µ µ 3 C 3 f x µ µ x µ 2 C 3 f x x µ x 2 µ C f x x x x 3 C D 0

(3.92)

For (x, µ) near (0, 0), (3.92) has a single solution x given by x D

1 2

f µ µ C O(µ 2 )

(3.93)

To ascertain the stability of this ﬁxed point, we differentiate (3.91) with respect to x and obtain D x f (xI µ) D 1C f x µ µC f x x x C 12 f x µ µ µ 2 C f x x µ x µC 12 f x x x x 2 C (3.94) Substituting (3.93) into (3.94) yields the multiplier (x ) D 1 C f x µ C

1 2

f µ f x x µ C O(µ 2 )

(3.95)

Hence, if f x µ C 1/2 f µ f x x > 0, the ﬁxed point x is stable when µ > 0 and unstable when µ < 0. On the other hand, if f x µ C 1/2 f µ f x x < 0, the ﬁxed point x is stable when µ < 0 and unstable when µ > 0. For nondegenerate bifurcation, 1 d (x ) D f x µ C f µ f x x ¤ 0 dµ 2

(3.96)

In other words, the multiplier of the map at the ﬁxed point should exit the unit circle through 1 with nonzero speed. To investigate the bifurcating solutions, we ﬁrst introduce a transformation to reduce (3.91) to its normal form by shifting the ﬁxed point to x D 0 and eliminating the quadratic term. To this end, we let xD

1 2

fµ µ C y C by2

(3.97)

Substituting (3.97) into (3.91) and choosing b to eliminate the quadratic term in y, we obtain the normal form y kC1 D (1 C ν)y k C α y k3

(3.98)

where ν D fxµ C αD

1 6

fxxx C

1 µC f f 2 µ xx 3 2 f C O(µ) 8 xx

O(µ 2 ) ,

bD

1 4

f x x C O(µ) ,

3.4 Local Bifurcations

The ﬁxed point y D 0 of (3.98) ceases to be hyperbolic when ν D 0 and the map undergoes a ﬂip bifurcation there. To determine the bifurcating period-two orbit, we investigate the ﬁxed points of the second iterate of the map (3.98); that is, y kC2 D (1 C 2ν)y k 2α y k3 C

(3.99)

It follows from (3.99) that the ﬁxed points of the second iterate map are r r ν ν y D 0 , , α α The ﬁrst ﬁxed point is also a ﬁxed point of the map. The associated multipliers are D 1 C 2ν , 1 4ν , 1 4ν Therefore, the ﬁxed point y D 0 is stable when ν < 0 and unstable when ν > 0. If α > 0, the nontrivial ﬁxed points exist and are stable for ν > 0. Consequently, the period-two bifurcating orbit is stable, and the bifurcation of the nonhyperbolic ﬁxed point at the origin as µ passes through zero is called supercritical period-doubling or ﬂip bifurcation. If α < 0, the nontrivial ﬁxed points exist and are unstable for ν < 0. Consequently, the period-two bifurcating orbit is unstable, and the bifurcation of the nonhyperbolic ﬁxed point at the origin as µ passes through zero is called subcritical period-doubling or ﬂip bifurcation. Example 3.12 We consider the logistic map x kC1 D F(x k ) D 4α x k (1 x k )

(3.100)

for 0 x 1 and positive α. The ﬁxed points of this map are x1 D 0 and

x2 D 1

1 4α

The multiplier associated with the ﬁxed point x j is given by j D 4α(1 2x j ) 1

1

F(x)

F2(x)

0

(a)

0

x

1

0

(b)

0

x

Figure 3.5 Graphs to determine solutions of the logistic equation at α D 0.8: (a) ﬁxed points and (b) period-two points.

1

83

84

3 Maps

The trivial ﬁxed point x1 exists for all α. Because 1 D 4α, it is stable when α < 1/4. The nontrivial ﬁxed point x2 exists for α > 1/4. Because 2 D 2 4α, this nontrivial ﬁxed point is stable for 1/4 < α < 3/4. At α D 3/4, x2 is nonhyperbolic because 2 D 1 and it is unstable for α > 3/4. In Figure 3.5a, we plot F(x) versus x when α D 0.8. The intersections of the line y D x with the curve F(x) give the ﬁxed points of F(x). They occur at x1 D 0 and x2 D 0.6875. The ﬁxed point x2 D 0.6875 is unstable because jF 0 (x2 D 0.6875)j D 1.2 > 1. Next, we investigate the ﬁxed points of F (2) (x), which are given by F (2) (x) D F [F(x)] D F [4α x (1 x)] or F (2) (x) D 16α 2 x (1 x) 1 4α x (1 x) Hence, the ﬁxed points of F (2) (x) are given by the solutions of 16α 2 x (1 x) 1 4α x (1 x) D x which are 1 x D 0 ,1 , 4α

and

2 3 s 1 2 1 1 41 2α 15 C ˙ 2 4α 2 2

The ﬁrst two ﬁxed points are also ﬁxed points of F(x). In Figure 3.5b, we plot F (2) (x) versus x when α D 0.8. The intersections of the curve F (2) (x) with the curve y D x give the ﬁxed points of F (2) (x). We note that there are four intersections. The dot corresponds to the ﬁxed point x D 1 1/(4α) D 0.6875 of F (2) (x), which is also a ﬁxed point of F(x). The other three ﬁxed points are x D 0, x D 0.5130, and x D 0.7995. We note that F(0.5130) D 0.7995

and

F (2) (0.5130) D 0.5130

and

F (2) (0.7995) D 0.7995

and that F(0.7995) D 0.5130

Hence, x D 0.5130 and x D 0.7995 are period-two points of F(x). To determine the stability of the ﬁxed points x j of F (2) (x), we calculate its Jacobian; that is, j D

h i d h (2) i F (x j ) D 16α 2 (1 2x j ) 1 8α x j (1 x j ) dx

For the ﬁxed points x D 0 and x D 0.6875, D 10.24 and D 1.44, respectively. Hence, these ﬁxed points are unstable. This is expected because these ﬁxed points are unstable ﬁxed points of F(x); F 0 (x D 0) D 3.2 and F 0 (x D 0.6875) D 1.2. For the ﬁxed points x D 0.5130 and x D 0.7995 of F (2) (x), D 0.16, and hence both of the period-two points of F(x) are stable.

3.4 Local Bifurcations

0.8 x 0.6

0.4 0.65

0.7

0.75 α

0.8

0.85

Figure 3.6 Scenario in the vicinity of a period-doubling bifurcation of the ﬁxed point of the logistic map.

We numerically determined that the ﬁxed points x3 and x4 of F (2) (x) or equivalently the period-two points of F(x) are stable for α < 0.85. In Figure 3.6, we show the different solutions of F(x) and their stability for 0.65 < α < 0.85. The branches of stable and unstable solutions are depicted by solid and broken lines, respectively. The ﬁxed point x2 of F(x) experiences a supercritical period-doubling bifurcation at α D 0.75. As a consequence, two branches of period-two points emerge from this bifurcation point. We note that an iterate of F(x) initiated at either of the two period-two points ﬂips back and forth between them because F(x3 ) D x4

and

F(x4 ) D x3

For this reason, the period-doubling bifurcation of a ﬁxed point of a map is also called a ﬂip bifurcation. The ﬁxed point x2 and the period-two points x3 and x4 of the map F(x) are all ﬁxed points of the map F (2) (x). At α D 3/4, the ﬁxed point x2 D 2/3 of F (2) (x) is nonhyperbolic because 2 D 1. Hence, from Figure 3.6, we infer that this ﬁxed point of F (2) (x) experiences a pitchfork bifurcation at α D 3/4. In this case, the pitchfork bifurcation is supercritical. In other cases, it is possible that a perioddoubling bifurcation of F(x) can correspond to a subcritical pitchfork bifurcation of the map F (2) (x). The associated period-two points that arise due to this bifurcation will be unstable.

3.4.5 Hopf or Neimark–Sacker Bifurcation

When a ﬁxed point of (3.55) is nonhyperbolic with a pair of complex conjugate eigenvalues on the unit circle, the ﬁxed point of the map can experience what is called the Neimark–Sacker bifurcation or Hopf bifurcation (Iooss, 1979; Wiggins, 1990). This bifurcation can occur in two- and higher-dimensional maps. Again, center-manifold reduction analysis can be used to reduce the analysis of the dynamics of the n-dimensional map in the vicinity of the nonhyperbolic ﬁxed point

85

86

3 Maps

into the analysis of the dynamics of a two-dimensional map on the center manifold (Carr, 1981); that is, x kC1 D f (x k , y k I µ)

(3.101)

y kC1 D g(x k , y k I µ)

(3.102)

We assume that coordinate and parameter transformations have been used so that the ﬁxed point is at (x, y ) D (0, 0) when µ D 0. Hence, f (0, 0I 0) D 0 , and the Jacobian matrix 2 @f (0, 0I 0) 6 @x 6 AD4 @g (0, 0I 0) @x

g(0, 0I 0) D 0 3 @f (0, 0I 0) 7 @y 7 5 @g (0, 0I 0) @y

(3.103)

has a pair of complex-conjugate eigenvalues D e i β and N D e i β lying on the unit circle. In the absence of strong resonances (i.e., n ¤ 1 for n D 1, 2, 3, 4), one can follow steps similar to those in Examples 3.4 and 3.5 and the next example to obtain the following normal form of the map (3.101) and (3.102):

z kC1 D e i β C νµ e i τ z k C α z 2 zN (3.104) where ν and τ are real-valued parameters and z is a complex-valued quantity. We assume that ν cos(β τ) ¤ 0 so that the pair of complex-conjugate multipliers transversely exit the unit circle as µ varies past zero. Substituting the polar form z D r eiθ

(3.105)

into (3.104) and separating real and imaginary parts, we obtain r kC1 D r k C νµ cos(β τ)r k C Λ r r k3 C

(3.106)

θkC1 D θk C β νµ sin(β τ) C Λ i r k2 C

(3.107)

for small r and µ, where Λ r C i Λ i D α e i β . We note that, because the r component is independent of θ , the problem is reduced to studying the stability of the ﬁxed points of the one-dimensional map (3.106). We assume that Λ r ¤ 0; otherwise higher-order terms need to be included in (3.104). There are three ﬁxed points: s s νµ cos(β τ) νµ cos(β τ) r D0, r D , and (3.108) Λr Λr The origin is asymptotically stable for νµ cos(β τ) < 0, unstable for νµ cos(β τ) > 0, unstable for νµ cos(β τ) D 0 and Λ r > 0, and asymp-

3.4 Local Bifurcations

totically stable for νµ cos(β τ) D 0 and Λ r < 0. On the other hand, the nontrivial ﬁxed points exist when νµ Λ r cos(β τ) > 0 and correspond to invariant circles. They are stable for νµ cos(βτ) > 0 and Λ r < 0 and unstable for νµ cos(βτ) < 0 and Λ r > 0. Substituting for the nontrivial ﬁxed points from (3.108) into (3.107) leads to the circle map θkC1 D θk C β νµ sin(β τ)

νµ Λ i cos(β τ) Λr

(3.109)

which describes the dynamics on the invariant circle. The dynamics on the invariant circle is periodic or aperiodic (densely ﬁlls the invariant circle) depending on whether χ D β νµ sin(β τ)

νµ Λ i cos(β τ) Λr

is rational or irrational. A rational number is a number that can be written as the ratio of two integers; that is, χ D p /q where p and q are integers. An irrational number cannot be written as the ratio of two integer. p For example, 0.2 is a rational number because it can be written as 1/5, whereas 5 is an irrational number. When χ D p /q where p and q are integers, the orbit of (3.109) starting at θ0 consists of q points fθ0 , θ0 C 2π χ, θ0 C 4π χ, , θ0 C 2(q 2)π χ, θ0 C 2(q 1)π χg We note that the next point in the orbit is θ0 C 2q π χ D θ0 C 2p π D θ0 . Consequently, the orbit would repeat itself. This orbit can be represented by q points on a circle. For example, when χ D 3/5, the orbit consists of the ﬁve points ˚

θ0 , θ0 C 65 π, θ0 C

12 π, θ0 5

C

18 π, 5

θ0 C

24 π 5

which can be represented as ﬁve points on a circle, as shown in Figure 3.7a.

(a)

(b)

Figure 3.7 Dynamics of a linear p circle map: (a) periodic motion when χ D 3/5 and (b) quasiperiodic motion when χ D 1/ 7.

87

88

3 Maps

When χ is irrational, starting from any point θ0 on a circle, we ﬁnd that none of the iterates F (m) returns to the initial point θ0 for any ﬁnite value of m. Thus, the orbit wanders on the circle, ﬁlling it up without becoming periodic. In other words, the orbit for irrational χ is aperiodic and is p an example of a quasiperiodic orbit. An example is shown in Figure 3.7b for χ D 1/ 7. We note that two quasiperiodic orbits starting from two points a small distance apart remain close for all subsequent iterations; that is, there is no sensitivity to initial conditions. Example 3.13 We consider the map x kC1 D (1 C µ)y k

(3.110)

y kC1 D y k x k 2x k y k

(3.111)

which has the ﬁxed points (x, y ) D (0, 0) and (x, y ) D (1/2 1/2µ, 1/2). The Jacobian of this map is 0 1Cµ JD 1 2y 1 2x Consequently, the multipliers associated with the nontrivial ﬁxed point are D 0 and D 2 C µ and hence it is unstable for values of µ near zero. On the other hand, the multipliers associated with the trivial ﬁxed point are D

1 2

p

p 2 3 p p i tan1 3C 3 µ ˙ 12 i 3 C 4µ D 1 C µe

These complex-conjugate multipliers transversely exit the unit circle away from the real axis as µ increases past zero, and hence the trivial ﬁxed point undergoes a Neimark–Sacker bifurcation as µ increases past zero. To analyze the dynamics of the map near this bifurcation, we construct the normal form of the map (3.110) and (3.111) for small x, y, and µ. To this end, we rewrite it as x kC1 0 1 xk 0 µ xk 0 D C C (3.112) 1 1 y k 0 0 yk 2x k y k y kC1 The eigenvalues of the linear part of (3.112) when µ D 0 are p p 1 1 1 D 12 (1 C i 3) D e 3 i π and 2 D N1 D 12 (1 i 3) D e 3 i π and the associated eigenvectors are the columns of the matrix p p 1 (1 i 3) 12 (1 C i 3) 2 PD 1 1 Next, we introduce the transformation x z DP y zN

(3.113)

3.4 Local Bifurcations

Hence, p p x D 12 (1 i 3)z C 12 (1 C i 3) zN

and

y D z C zN

(3.114)

Substituting (3.113) and (3.114) into (3.112) and multiplying the outcome from the left with P 1 , we obtain p 1p 2 3 2 1 3i µ(z k C zN k ) C z kC1 D e 3 i π z k C i zk 3 3 p ! p ! 3 3 1 (3.115) i z k zN k 1 C i zN k2 3 3 with the second equation being the complex-conjugate of this equation. Next, we determine the normal form of (3.115) for small z and µ. Instead of using two successive transformations to simplify the quadratic terms and then the cubic terms, we use a single transformation to accomplish simpliﬁcation of both nonlinearities. To this end, we introduce a small nondimensional bookkeeping parameter , scale z as z and µ as 2 µ, let z D ξ C h 1 (ξ , ξN ) C 2 h 2 (ξ , ξN )

(3.116)

and choose h 1 (ξ , ξN ) and h 2 (ξ , ξN ) so that (3.115) takes on the simplest form 1 ξkC1 D e 3 i π ξk C g 1 (ξk , ξN k ) C 2 g 2 (ξk , ξN k )

(3.117)

Substituting (3.116) and (3.117) into (3.115) and equating coefﬁcients of like powers of , we obtain Order ()

1 1 1 g 1 (ξ , ξN ) C h 1 e 3 i π ξ , e 3 i π ξN e 3 i π h 1 (ξ , ξN ) p ! p ! p 3 3 2 3 2 N iξ 1 i ξξ 1C i ξN 2 D 3 3 3

(3.118)

Order (2 )

1 1 1 g 2 (ξ , ξN ) C h 2 e 3 i π ξ , e 3 i π ξN e 3 i π h 2 (ξ , ξN ) p 4 3 1p 3i µ(ξ C ξN ) C i ξ h 1 (ξ , ξN ) D 3 3 p !

1 3 @ i π e 3 g 1 (ξ , ξN ) h 1 (ξ , ξN ) 1 i ξ hN 1 (ξ , ξN ) C ξN h 1 (ξ , ξN ) @ξ 3 p ! 3 @ N 13 i π N N gN 1 (ξ , ξ ) h 1 (ξ , ξ ) 2 1 C i ξN hN 1 (ξ , ξN ) e (3.119) N 3 @ξ

89

90

3 Maps

Because e 1/3i n π ¤ 1 for n D 1 and 3, there are no strong resonances, and hence h 1 (ξ , ξN ) can be chosen to eliminate all of the quadratic terms in (3.118); that is, p ! p ! p 3 3 3 2 1 2 (3.120) h 1 (ξ , ξN ) D iξ 1 C i ξ ξN C 1 i ξN 2 3 3 2 3 Consequently, g 1 (ξ , ξN ) 0 and (3.119) becomes

1 1 1 g 2 (ξ , ξN ) C h 2 e 3 i π ξ , e 3 i π ξN e 3 i π h 2 (ξ , ξN ) p p p D 13 3i µ(ξ C ξN ) C 2ξ 3 C 2(1 i 3)ξ 2 ξN C 4ξ ξN 2 C (1 i 3) ξN 3 (3.121) One can choose h 2 (ξ , ξN ) to eliminate the nonresonance terms in (3.122), leaving g 2 (ξ , ξN ) with the resonance terms; that is, p p (3.122) g 2 (ξ , ξN ) D 13 3i µ ξ C 2(1 i 3)ξ 2 ξN Therefore, the normal form of the map is p p 1 ξkC1 D e 3 i π ξk C 13 3i µ ξk C 2(1 i 3)ξk2 ξN k

(3.123)

where the bookkeeping parameter has been set equal to unity. Equation 3.123 can be rewritten as h i p p 1 1 ξkC1 D 1 C 12 µ e 3 i πC 6 3i µ ξk (2 C 2i 3)ξk2 ξN k C

(3.124)

Letting ξ D r e i θ in (3.124), we obtain r kC1 D 1 C 12 µ r k 2r k3

(3.125)

θkC1 D θk C 13 π C

1 6

p

p 3µ 2 3r k2

(3.126)

p It follows from (3.125) that the attracting invariant circle r D 1/2 µ bifurcates from the origin as µ increases past zero. The dynamics on this p invariant circle is periodic or quasiperiodic, depending on whether 1/3(π 3µ) is rational or irrational. We note that higher-order terms in µ and r have been neglected in arriving at (3.124) and (3.125). Including these higher-order terms, one ﬁnds that the attracting smooth invariant curve is a circle only for values of µ close to zero. Moreover, for large values of µ, the attractor may be complicated. In Figure 3.8, we show phase portraits obtained numerically for the map (3.110) and (3.111) for µ D 0.01 and µ D 0.02. When µ < 0, the origin is asymptotically stable and all iterates starting within its domain of attraction spiral to it, as shown in Figure 3.8a. The six-fold rotational symmetry is the result of β being 1/3π. When µ > 0, the origin is unstable and all iterates starting close to the origin spiral out to a smooth closed invariant curve enclosing the origin, as shown in Figure 3.8b. For the chosen value of µ, the long-time iterates densely ﬁll the invariant curve. Again, the six-fold rotational symmetry is the result of β being 1/3π.

3.5 Exercises

0.2

0.2

0.1

0.1

0

0

−0.1

−0.1

−0.1

(a)

0

0.1

0.2

(b)

−0.1

0

0.1

0.2

Figure 3.8 Phase portrait of the map (3.110) and (3.111): (a) µ < 0 and the origin is stable and iterates spiral to it and (b) µ > 0 and the origin is unstable and iterates spiral away from it and onto a smooth invariant closed curve encircling it.

3.5 Exercises

3.5.1 Consider the map x kC1 D x k C µ x k2 Examine the bifurcation that occurs at (x k , µ) D (1, 1). 3.5.2 Consider the map x kC1 D x k C µ x k x k2 Study the bifurcation that takes place at (x, µ) D (0, 2). 3.5.3 Consider the following map: x kC1 D µ C x k2 Examine the bifurcation that takes place at (x, µ) D (1/2, 1/4). 3.5.4 Consider the following map: x kC1 D µ x k C x k2 Examine the bifurcation that takes place at (x, µ) D (0, 1). 3.5.5 Consider the map x kC1 D ax k2 C 1 Determine its ﬁxed points and their stability. Examine the bifurcation that takes place as a is varied. 3.5.6 Find the ﬁxed points of x nC1 D ax n x n3 Determine their stability and the bifurcations, which they undergo as a is varied.

91

92

3 Maps

3.5.7 Find the ﬁxed points of x nC1 D ax n C x n3 Determine their stability and the bifurcations, which they undergo as a is varied. 3.5.8 Consider the following map: x kC1 D µ x k C x k2 Determine whether the period-doubled orbit bifurcating from the point (x, µ) D (0, 0) is supercritical or subcritical. 3.5.9 Consider the map x nC1 D (1 C a)x n C b x n2 x n3 What is the type of bifurcation which occurs at x D 0 and a D 0? Is it supercritical or subcritical? 3.5.10 Show that a ﬁxed point of the map f (xI λ) D (1 C λ)x C x 2 undergoes a transcritical bifurcation at λ D 0. 3.5.11 Show that the map f (xI λ) D e x λ undergoes a saddle-node bifurcation at λ D 1. 3.5.12 Determine the ﬁxed points of the following cubic map and discuss their stability: x nC1 D (1 λ)x n C λx n3 For what value of λ does the ﬁrst period-doubling bifurcation occur? 3.5.13 Consider the map x kC1 D x k C ax k2 C b x k3 Compute the second iterate of this map and hence determine whether the perioddoubled orbit is stable or unstable. 3.5.14 Determine the normal form of the map x kC1 D x k C ax k2 C b x k3 3.5.15 Consider the two one-dimensional maps x nC1 D e x n λ y nC1 D 12 λ tan1 y n Find the ﬁxed points and their stability. Show that x undergoes a saddle-node bifurcation at λ D 1, whereas y undergoes a period-doubling bifurcation at λ D 3.

3.5 Exercises

3.5.16 Consider the Hénon map x kC1 D 1 C y k α x k2 y kC1 D β x k Assume that β ¤ 0 and α > 0. a) Verify that this map has a stable ﬁxed point and an unstable ﬁxed point when α < 34 (1 β)2 b) Is this map dissipative for β D 0.3? For this case, determine the period-one and period-two points of this map for α D 0.1, α D 0.5, and α D 1.3. Discuss their stability. c) For the above values of α, plot the iterates of this map. d) Examine the bifurcation that takes place at (1 β) β(1 β) 3 (x k , y k , α) D , , (1 β)2 . 2α 2α 4

3.5.17 Consider the two-dimensional map x nC1 D λ C x n C λ y n C x n2 y nC1 D 12 y n C λx n C x n2 Determine its ﬁxed points and then show that the origin undergoes a saddle-node bifurcation at λ D 0. 3.5.18 Consider the two-dimensional map: x nC1 D y n y nC1 D 12 x n C λ y n y n3 Determine its ﬁxed points. Show that the origin undergoes a pitchfork bifurcation at λ D 3/2. Analyze the bifurcation at λ D 3. 3.5.19 Consider the following map: x kC1 D 2x k y kC1 D 12 y k C 7x k2 Show that the stable manifold of its ﬁxed point is the y-axis and the unstable manifold is y D 2x 2 .

93

94

3 Maps

3.5.20 Consider the map x kC1 D y k y kC1 D α y k (1 x k ) Show that the origin is a saddle. Determine analytical expressions for the stable and unstable manifolds. 3.5.21 Consider the map x kC1 D y k y kC1 D α y k (1 x k ) Examine the bifurcation that occurs at α D 2 and determine the normal form of the map near this bifurcation. 3.5.22 Consider the map x kC1 D y k y kC1 D α y k (1 x k ) Examine the bifurcation that occurs at α D 1 and determine the normal form of the map near this bifurcation. 3.5.23 Consider the following map near the origin: x kC1 D x k C α x k y k y kC1 D 12 y k C x k2 Compute the center manifold, describe the dynamics on the center manifold, and then determine the stability of the origin. 3.5.24 Consider the map x nC1 D y n y nC1 D b x n y n C x n y n What type of bifurcation occurs at (x, y, b) D (0, 0, 0). 3.5.25 Consider the map x kC1 D 58 x k C 38 y k C x y y kC1 D 38 x k C 58 y k C x 2 y 2 Compute the center manifold near the origin, describe the dynamics on the center manifold, and then determine the stability of the origin.

3.5 Exercises

3.5.26 Consider the two-dimensional map x nC1 D (1 C µ)y n y nC1 D y n x n 2x n y n Determine the ﬁxed points of this map and their stability. Show that the origin undergoes a Hopf bifurcation at µ D 0. Calculate and plot the phase portraits for a) µ D 0.01 using the initial condition x D 0.4, y D 0.4. b) µ D 0.05 using the initial condition x D 0.01, y D 0.01. 3.5.27 Consider the map x kC1 D y k y kC1 D 12 µ C (1 C µ) (y k x k 2x k y k ) a) Determine the ﬁxed points and their stability. b) Compute the normal form of this map near the origin when µ 0. c) Use this normal form to calculate the bifurcating invariant circle from the origin. d) Calculate the phase portraits for µ D 0.01 and µ D 0.02. 3.5.28 Consider the map z kC1 D z k C b 11 z 2 C b 12 z zN C b 13 zN 2 C a 11 z 3 C a 12 z 2 zN C a 13 z zN 2 C a 14 zN 3 Show that its normal form near the origin has the form z kC1 D z k C α 1 zN 2 C α 2 z 2 zN when 3 D 1 and the form z kC1 D z k C α 1 z 2 zN C α 2 zN 3 when 4 D 1. 3.5.29 Consider the map p z nC1 D 12 ( 3 C i C 2i µ)z n C z n2 zN n What type of bifurcation occurs at (z, zN , µ) D (0, 0, 0). Determine the bifurcating solutions.

95

97

4 Bifurcations of Continuous Systems In this chapter, we construct normal forms of smooth continuous systems depending on a scalar control parameter µ near their ﬁxed or equilibrium points. Speciﬁcally, we consider the following system: xP D F(xI µ) where x 2 U R n , F 2 U R n , and µ 2 V R. The ﬁxed points or equilibrium solutions of this system are solutions of the algebraic system of equations F(xI µ) D 0 In Section 4.1, we consider linear systems; in Section 4.2, we consider the ﬁxed points of nonlinear systems and their stability; in Section 4.3, we discuss centermanifold reduction; in Section 4.4, we consider local bifurcations of ﬁxed points; and in Sections 4.5 and 4.6, we illustrate how the method of multiple scales, a combination of center-manifold reduction and the method of normal forms, and a projection method can be used to construct the normal forms of static and Hopf bifurcations in the neighborhood of a ﬁxed point.

4.1 Linear Systems

We consider the linear system xP D Ax

(4.1)

where A is an n n constant matrix. In this case, the trivial solution x D 0 is a ﬁxed point of this linear system. We denote the eigenvalues of A by λ i , i D 1, 2, . . . , n, and the corresponding eigenvectors (generalized eigenvectors) by p i , i D 1, 2, . . . , n. The eigenvalues are the roots of the characteristic equation det(A λ I ) D 0

(4.2)

The eigenvector p i corresponding to a distinct eigenvalue λ i is given by Ap i D λ i p i

(4.3)

The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

98

4 Bifurcations of Continuous Systems

and the generalized eigenvectors corresponding to an eigenvalue λ m with multiplicity n m are the nontrivial solutions of (A λ m I )p D 0 ,

(A λ m I )2 p D 0 ,

... ,

(A λ m I ) n m p D 0

(4.4)

If an eigenvalue is complex-valued, then its corresponding eigenvector and generalized eigenvectors are also complex-valued. Introducing the transformation x D Py

(4.5)

where the matrix P D [p 1 p 2 , . . . , p n ], into (4.1) yields P yP D AP y

(4.6)

Multiplying (4.6) from the left with the inverse P 1 of P, we obtain the normal of (4.1) as yP D J y

(4.7)

1

where J D P AP is called the Jordan canonical form of A. Next, we discuss two cases: systems with distinct and nondistinct eigenvalues. 4.1.1 Case of Distinct Eigenvalues

If the eigenvalues of A are distinct, then J is a diagonal matrix D with entries λ i , i D 1, 2, . . . , n; that is, 2 3 λ1 0 0 6 0 λ2 0 7 6 7 D D6 (4.8) 7 6 7 4 5 0 0 λn Then, (4.7) can be rewritten as yP (m) D λ m y (m) ,

m D 1, 2, . . . , n

(4.9)

where y (m) is the mth component of y. Therefore, y (m) D c m e λ m t ,

m D 1, 2, . . . , n

(4.10)

where c m is a constant. It follows from (4.10) that y (m) ! 0 as t ! 1 when λ m lies in the left-half of the complex plane, y (m) ! 1 as t ! 1 when λ m lies in the right-half of the complex plane, and y (m) D c m for all time when λ m lies on the imaginary axis. Therefore, the origin of (4.1) is (a) asymptotically stable if all of the eigenvalues λ i of the matrix A lie in the left-half of the complex plane, (b) unstable if one or more λ i lie in the right-half of the complex plane, and (c) neutrally or marginally stable if one or more λ i lie on the imaginary axis with the rest of the eigenvalues being in the left-half of the complex plane.

4.1 Linear Systems

4.1.2 Case of Repeated Eigenvalues

If the number of distinct eigenvalues of A is k < n, then J is diagonal if all of the p i are eigenvectors; otherwise, it has the form 3 2 J1 φ φ 6φ J φ7 2 7 6 7 6 7 6 (4.11) J D6 7 7 6 7 6 4 5 φ φ Jk where φ represents a matrix with zero entries and 3 2 1 0 0 λm 60 λm 1 07 7 6 6 0 λm 7 Jm D 6 0 7 4 5 0 0 0 λm

(4.12)

Example 4.1 We consider a system with repeated roots and a nondiagonal J; that is, 1 (a b) a 2 AD 12 (a b) b where b ¤ a. The eigenvalues of this matrix are D 1/2(a C b) with a multiplicity of two. It follows from Example 3.3. that the corresponding eigenvector and generalized eigenvector are 1 p1 D 1 and

2

3

1

p2 D 4

1C

Then,

2 5 ba 2

1 6 2 (a C b) 1 P AP D J D 4 0

3 1

7 5 1 (a C b) 2

and (4.7) yields yP (1) D λ y (1) C y (2)

(4.13)

yP (2) D λ y (2)

(4.14)

99

100

4 Bifurcations of Continuous Systems

The general solutions of (4.13) and (4.14) can be expressed as y (1) D c 1 e λ t C t c 2 e λ t

(4.15)

y (2) D c 2 e λ t

(4.16)

where c 1 and c 2 are arbitrary constants. Therefore, the origin of (4.1) is (a) asymptotically stable if Real(λ) < 0 and (b) unstable if Real(λ) 0.

4.2 Fixed Points of Nonlinear Systems

The ﬁxed points of the autonomous system xP D F(xI µ)

(4.17)

are deﬁned by the vanishing of the vector ﬁeld; that is, F(xI µ) D 0

(4.18)

A location in the state space where this condition is satisﬁed is called a singular point. At such a point, the integral curve of the vector ﬁeld F corresponds to the point itself. Also, an orbit of a ﬁxed point is the ﬁxed point itself. Fixed points are also called stationary solutions, critical points, constant solutions, and sometimes steady-state solutions. Physically, a ﬁxed point corresponds to an equilibrium position of a system. Further, ﬁxed points are examples of invariant sets of (4.17). 4.2.1 Stability of Fixed Points

To investigate the stability of a ﬁxed point x (µ ), where x 2 R n and µ 2 R, we superimpose on it a small disturbance y(t) and obtain x(t) D x C y(t)

(4.19)

Substituting (4.19) into (4.17) yields yP D F(x C yI µ )

(4.20)

We note that the ﬁxed point x D x of (4.17) has been transformed into the ﬁxed point y D 0 of (4.20). Assuming that F is at least twice continuously differentiable (i.e., C 2 ), expanding (4.20) in a Taylor series about x , and retaining only linear terms in the disturbance leads to yP D F(x I µ ) C D x F(x I µ )y C O(jjyjj2 )

4.2 Fixed Points of Nonlinear Systems

or yP D x F (x I µ )y Ay

(4.21)

where A, the matrix of ﬁrst partial derivatives, is called the Jacobian matrix. If the components of F are F1 (x1 , x2 , . . . , x n ), F2 (x1 , x2 , . . . , x n ), . . . , F n (x1 , x2 , . . . , x n ) then

2 @F

1

@x 6 @F1 6 2 6 @x1

6 6 6 AD6 6 6 6 6 4

@F n @x1

@F1 @x2 @F2 @x2

@F n @x2

3

@F1 @x n 7 @F2 7 @x n 7

7 7 7 7 7 7 7 7 5

@F n @x n

We have transformed the problem of determining the local stability of the ﬁxed point x of (4.17) into that of determining the stability of the trivial solution of the linear system (4.21). We say local stability because we have considered a small disturbance and linearized the vector ﬁeld. It follows from the preceding section that the trivial solution of (4.21) and hence the ﬁxed point x of (4.17) is (a) asymptotically stable if all of the eigenvalues λ i of the matrix A lie in the left-half of the complex plane and (b) unstable if one or more λ i lie in the right-half of the complex plane. In the case of repeated eigenvalues, the trivial solution of (4.21) and hence the ﬁxed point x of (4.17) is unstable if one or more eigenvalues lie on the imaginary axis and the Jordan form is not diagonal. 4.2.2 Classiﬁcation of Fixed Points

When all of the eigenvalues of A have nonzero real parts, the corresponding ﬁxed point is called a hyperbolic ﬁxed point, irrespective of the values of the imaginary parts; otherwise, it is called a nonhyperbolic ﬁxed point. There are three types of hyperbolic ﬁxed points: sinks, sources, and saddles. If all of the eigenvalues of A have negative real parts, then all of the components of the disturbance y decay in time, and hence x approaches the ﬁxed point x of (4.17) as t ! 1. Therefore, the ﬁxed point x of (4.17) is asymptotically stable. An asymptotically stable ﬁxed point is called a sink. If the matrix A associated with a sink has complex eigenvalues, the sink is also called a stable focus. On the other hand, if all of the eigenvalues of the matrix A associated with a sink are real, the sink is also called a stable node. A sink is stable in forward time (i.e., t ! 1) but unstable in reverse time (i.e., t ! 1). Further, all sinks qualify as attractors. If one or more of the eigenvalues of A have positive real parts, some of the components of y grow in time, and x moves away from the ﬁxed point x of (4.17) as

101

102

4 Bifurcations of Continuous Systems

t increases. In this case, the ﬁxed point x is said to be unstable. When all of the eigenvalues of A have positive real parts, x is said to be a source. If the matrix A associated with a source has complex eigenvalues, the source is also called an unstable focus. On the other hand, if all of the eigenvalues of the matrix A associated with a source are real, the source is also called an unstable node. A source is unstable in forward time but stable in reverse time. Because trajectories move away from a source in forward time, the source is an example of a repellor. When some, but not all, of the eigenvalues have positive real parts while the rest of the eigenvalues have negative real parts, the associated ﬁxed point is called a saddle point. Because a saddle point is unstable in both forward and reverse times, it is called a nonstable ﬁxed point. A nonhyperbolic ﬁxed point is unstable if one or more of the eigenvalues of A have positive real parts. If some of the eigenvalues of A have negative real parts while the rest of the eigenvalues are distinct and have zero real parts, the ﬁxed point x D x of (4.17) is said to be neutrally or marginally stable. If all of the eigenvalues of A are distinct, nonzero, and purely imaginary, the corresponding ﬁxed point is called a center. Example 4.2 We consider the system xP D x (3 x 2y )

(4.22)

yP D y (2 x y )

(4.23)

Its ﬁxed points are (0, 0), (0, 2), (3, 0), and (1, 1). The Jacobian matrix of the system (4.22) and (4.23) is AD

3 2x 2y y

2x 2 x 2y

The eigenvalues of A corresponding to the ﬁxed point (0, 0) are λ 1 D 2 and λ 2 D 3; hence, it is an unstable node. The eigenvalues of A corresponding to the ﬁxed point (0, 2) are λ 1 D 1 and λ 2 D 2; hence, it is a stable node. The eigenvalues of A corresponding to the ﬁxed point (3, 0) are λ 1 D 1 and λ 2 D 3; hence, it is a stable node. p The eigenvalues p of A corresponding to the ﬁxed point (1, 1) are λ 1 D 1 C 2 and λ 2 D 1 2; hence, it is a saddle. In this example, all of the ﬁxed points are hyperbolic.

4.2.3 Hartman–Grobman and Shoshitaishvili Theorems

Many theorems provide precise statements on what the stability of ﬁxed-point solutions of the linearized system (4.21) imply for the stability of ﬁxed-point solutions

4.3 Center-Manifold Reduction

of the full nonlinear system (4.17). The Hartman–Grobman theorem (e.g., Arnold, 1988, Chapter 3; Wiggins, 1990, Chapter 2) is applicable to hyperbolic ﬁxed points, whereas the Shoshitaishvili theorem (e.g., Arnold, 1988, Chapter 6) is applicable to nonhyperbolic ﬁxed points. From these theorems, it follows that (a) the ﬁxed point x D x of the nonlinear system (4.17) is stable when the ﬁxed point y D 0 of the linear system (4.21) is asymptotically stable; (b) the ﬁxed point x D x of the nonlinear system (4.17) is unstable when the ﬁxed point y D 0 of the linear system (4.21) is unstable; and (c) linearization cannot determine the stability of neutrally stable ﬁxed points (including centers) of (4.17). In the case of neutrally stable ﬁxed points, a nonlinear analysis is necessary to determine the stability of x . It will be necessary to retain quadratic and, sometimes, higher-order terms in the disturbance y in the Taylor-series expansion of (4.20). In a topological setting, the Hartman–Grobman theorem implies that the trajectories in the vicinity of a hyperbolic ﬁxed point x D x of (4.17) are qualitatively similar to those in the vicinity of the hyperbolic ﬁxed point y D 0 of (4.21). In other words, the local nonlinear dynamics near x D x is qualitatively similar to the linear dynamics near y D 0, and a qualitative change in the local nonlinear dynamics can be detected by examining the associated linear dynamics. According to the Hartman–Grobman theorem, there exists a continuous coordinate transformation (i.e., a homeomorphism) that transforms the nonlinear ﬂow into the linear ﬂow in the vicinity of a hyperbolic ﬁxed point. In the absence of resonances or near resonances, the method of normal forms (e.g., Sections 2.2 and 2.4) may be used to generate a coordinate transformation to transform the nonlinear ﬂow into linear ﬂow (e.g., Arnold, 1988; Guckenheimer and Holmes, 1983). Further, such a coordinate transformation would be a differentiable one because the method of normal forms yields transformations in the form of power-series expansions.

4.3 Center-Manifold Reduction

We consider the local dynamics near a nonhyperbolic ﬁxed point x of the nonlinear system (4.17), where F is an analytic vector function of x. According to the center-manifold theorem (Carr, 1981), there exists a C r local center manifold for the nonlinear system (4.17) near x . Furthermore, if none of the eigenvalues of this ﬁxed point lies in the right-half of the complex plane, the long-time dynamics of (4.17) can be reduced to determining the dynamics on the center manifold. Next, we describe how to construct the center manifold in the neighborhood of a nonhyperbolic ﬁxed point with one eigenvalue being zero and the real parts of all of the other eigenvalues being negative. We assume that the ﬁxed point has been shifted to the origin and that the linear part has been transformed into a Jordan canonical form; that is, we consider the system xP kC1 D J x k C F(x)

(4.24)

103

104

4 Bifurcations of Continuous Systems

We arrange the system (4.24) and rewrite it as xP D f (x, y)

(4.25)

yP D B y C G(x, y)

(4.26)

where B is a constant matrix with the real parts of all of its eigenvalues being negative and f and G are scalar and vector-valued nonlinear functions of x and y. According to the center-manifold theorem, there exists a center manifold y D h(x)

(4.27)

Moreover, the dynamics of the system (4.25) and (4.26) is qualitatively similar to the dynamics on this manifold; that is, xP D f x, h(x) (4.28) Substituting (4.27) into (4.26) yields P h(x) D B h(x) C G x, h(x)

(4.29)

which upon using (4.28) becomes h 0 (x) f x, h(x) D B h(x) C G x, h(x)

(4.30)

Using three examples, we describe how to construct approximate solutions of (4.30). Example 4.3 We consider the system xP D α 1 x y

(4.31)

yP D y C α 2 x 2

(4.32)

where α 1 and α 2 are constants. Clearly, the origin is a ﬁxed point of (4.31) and (4.32). Because its associated eigenvalues are λ D 0 and λ D 1, the origin is nonhyperbolic and the center and stable subspaces are one-dimensional. Consequently, the center manifold is one-dimensional. To calculate the center manifold and the dynamics on this manifold, we note that B D 1, f (x, y ) D α 1 x y , and G(x, y ) D α 2 x 2 . Hence, (4.30) becomes h 0 (x)α 1 x h(x) D h(x) C α 2 x 2

(4.33)

We seek an approximate solution of (4.33) in the form h(x) D ax 2 C

(4.34)

and obtain 2a 2 α 1 x 4 C ax 2 α 2 x 2 C D 0

(4.35)

4.3 Center-Manifold Reduction

Equating to zero the coefﬁcient of x 2 , we obtain a D α 2 . Hence, the center manifold is given by h(x) D α 2 x 2 C

(4.36)

and it follows from (4.31) that the long-time dynamics on this center manifold is given by xP D α 1 α 2 x 3 C

(4.37)

We note that linearization is not sufﬁcient for determining the stability of the origin because its associated eigenvalues are λ D 0 and λ D 1. However, including the nonlinear terms, we ﬁnd from (4.37) that the origin is unstable when α 1 α 2 > 0 and stable when α 1 α 2 < 0. Example 4.4 We consider the system xP D α 1 x y C α 2 x z

(4.38)

yP D µ y C z C α 3 x 2

(4.39)

zP D µ z y C α 4 x 2

(4.40)

where µ > 0. Clearly, the origin is a ﬁxed point of (4.38)–(4.40) and its associated eigenvalues are λ D 0 and λ D µ ˙ i. Hence, the origin is a nonhyperbolic ﬁxed point and the center and stable subspaces are one-dimensional and twodimensional, respectively. Consequently, the center manifold is two-dimensional. Thus, we seek the center manifold of (4.38)–(4.40) emanating from the origin in the form y (x) D ax 2 C

and

z(x) D b x 2 C

(4.41)

Substituting (4.41) into (4.38) leads to the following equation describing the dynamics on the manifold: xP D α 1 ax 3 C α 2 b x 3 C

(4.42)

Substituting (4.41) and (4.42) into (4.39) and (4.40) yields 2ax (α 1 ax 3 C α 2 b x 3 ) C µ ax 2 b x 2 α 3 x 2 C D 0

(4.43)

2b x (α 1 ax 3 C α 2 b x 3 ) C µ b x 2 C ax 2 α 4 x 2 C D 0

(4.44)

Equating the coefﬁcients of x 2 to zero in (4.43) and (4.44), we obtain µ a b D α3

(4.45)

a C µ b D α4

(4.46)

105

106

4 Bifurcations of Continuous Systems

Hence, µ α3 C α4 1 C µ2

aD

and

bD

µ α4 α3 1 C µ2

(4.47)

and it follows from (4.42) that the dynamics on the center manifold is given by xP D Γ x 3 C

(4.48)

where µ(α 1 α 3 C α 2 α 4 ) C α 1 α 4 α 2 α 3 1 C µ2

Γ D

We note again that linearization is not sufﬁcient for determining the stability of the origin because its associated eigenvalues are λ D 0 and λ D µ ˙ i. However, including the nonlinear terms, we ﬁnd from (4.48) that the origin is unstable when Γ > 0 and stable when Γ < 0. Example 4.5 We consider the system xP D y C a 1 x 2 C a 2 y 2 C a 3 z 2 C a 4 x y C a 5 x z C a 6 y z

(4.49)

yP D x C b 1 x 2 C b 2 y 2 C b 3 z 2 C b 4 x y C b 5 x z C b 6 y z

(4.50)

zP D z C c 1 x 2 C c 2 y 2 C c 3 z 2 C c 4 x y C c 5 x z C c 6 y z

(4.51)

Clearly, the origin x D y D z D 0 is a ﬁxed point of (4.49)–(4.51). Its associated eigenvalues are λ D i, i, 1, and hence the center subspace is two-dimensional and the stable subspace is one-dimensional. Consequently, the center manifold is one-dimensional. We seek the center manifold emanating from the origin in the form z(x, y ) D α 1 x 2 C α 2 x y C α 3 y 2

(4.52)

Substituting (4.52) into (4.49) and (4.50) yields the following two-dimensional system describing the dynamics on the center manifold: xP D y C a 1 x 2 C a 2 y 2 C a 4 x y C (a 5 x C a 6 y )(α 1 x 2 C α 2 x y C α 3 y 2 ) C

(4.53)

yP D x C b 1 x 2 C b 2 y 2 C b 4 x y C (b 5 x C b 6 y )(α 1 x 2 C α 2 x y C α 3 y 2 ) C

(4.54)

4.4 Local Bifurcations of Fixed Points

Substituting (4.52) into (4.51) and using (4.53) and (4.54), we obtain 2α 1 x y C α 2 y 2 α 2 x 2 2α 3 x y D α 1 x 2 α 2 x y α 3 y 2 C c1 x 2 C c2 y 2 C c4 x y C

(4.55)

Equating the coefﬁcients of x 2 , x y , and y 2 on both sides of (4.55) yields α1 α2 D c1

(4.56)

2α 1 2α 3 C α 2 D c 4

(4.57)

α2 C α3 D c2

(4.58)

Therefore, α 1 D 15 (3c 1 C 2c 2 C c 4 )

(4.59)

α 2 D 51 (2c 1 2c 2 c 4 )

(4.60)

α 3 D 15 (2c 1 C 3c 2 c 4 )

(4.61)

Substituting for the α i from (4.59)–(4.61) into (4.53) and (4.54), we obtain a twodimensional dynamical system describing the dynamics on the center manifold. Using a methodology similar to that in Section 2.5, we reduce this dynamical system into its normal form.

4.4 Local Bifurcations of Fixed Points

From Section 4.2, we know that the matrix A in (4.21) and the associated eigenvalues are functions of the control parameter µ. Let us suppose that, as µ is slowly varied, a ﬁxed point becomes nonhyperbolic at a certain location in the state-control space. Then, if the state-space portraits before and after this location are qualitatively different, this location is called a bifurcation point, and the accompanying qualitative change is called a bifurcation. There are two cases in which a ﬁxed point x of the continuous system (4.17) ceases to be hyperbolic at a critical value µ c of the control parameter. These cases are 1. D x F (x I µ c ) has one eigenvalue equal to zero with the remaining eigenvalues being in the left-half of the complex plane, 2. D x F (x I µ c ) has a pair of purely imaginary eigenvalues with the remaining eigenvalues being in the left-half of the complex plane.

107

108

4 Bifurcations of Continuous Systems

According to the center-manifold theorem, analysis of the the dynamics of an ndimensional continuous system near one of its ﬁxed points can be reduced to the analysis of the dynamics on its center manifold. In the ﬁrst case, the center manifold is (n 1)-dimensional and the analysis of the dynamics of the system can be reduced to that of analyzing the dynamics of a one-dimensional dynamical system. On the other hand, in the second case, the center manifold is (n 2)-dimensional and the analysis of the dynamics of the system can be reduced to that of analyzing a two-dimensional dynamical system. In Sections 4.5 and 4.6, we show how one can obtain such reduced dynamical systems. In Case 2, Hopf bifurcation can occur; and in Case 1, three types of static bifurcation can occur: saddle-node, transcritical, and pitchfork bifurcations. In Section 4.4.5, we consider Hopf bifurcation; and in Sections 4.4.1–4.4.4 we consider bifurcations of the one-dimensional dynamical system xP D f (xI µ)

(4.62)

where x and f are scalars and f (xI µ) is a smooth function of x and µ. We assume that (x D 0, µ D 0) is a nonhyperbolic ﬁxed point of (4.62); that is, f (0I 0) D 0 and f x (0, 0) D 0. Expanding f (xI µ) in a Taylor series for small x and µ, we obtain xP D f µ µ C 12 ( f µ µ µ 2 C 2 f x µ x µ C f x x x 2 ) C 16 ( f µ µ µ µ 3 C 3 f x µ µ x µ 2 C 3 f x x µ x 2 µ C f x x x x 3 ) C

(4.63)

Next, we consider the following four cases: (a) f µ ¤ 0 and f x x ¤ 0; (b) f µ ¤ 0, f x x D 0, and f x x x ¤ 0; (c) f µ D 0 and f x x ¤ 0; and (d) f µ D 0 and f x x D 0. We show below that the ﬁrst case corresponds to saddle-node bifurcation, the second case corresponds to a nonbifurcation, the third case corresponds to transcritical bifurcation, and the fourth case corresponds to pitchfork bifurcation. 4.4.1 Saddle-Node Bifurcation

We consider the case f µ ¤ 0 and f x x ¤ 0. As x ! 0 and µ ! 0, (4.63) tends to xP D f µ µ C

1 2

f xx x2 C

When f µ µ ¤ 0, the ﬁxed points of (4.64) are s 2 f µ µ x D˙ fxx

(4.64)

(4.65)

which exist only when f µ f x x µ < 0. The eigenvalues associated with these two ﬁxed points are s 2 f µ µ λ D ˙ fxx (4.66) fxx

4.4 Local Bifurcations of Fixed Points

It follows from (4.65) that there are two branches of ﬁxed points in the neighborhood of (x, µ) D (0, 0) for f µ µ < 0 if f x x > 0 and for f µ µ > 0 if f x x < 0. Then, it follows from (4.66) that the upper branch is stable and the lower branch is unstable if f x x < 0 and that the upper branch is unstable and the lower branch is stable if f x x > 0. This bifurcation of the nonhyperbolic ﬁxed point at the origin as µ passes through zero is called fold or tangent or saddle-node bifurcation. Example 4.6 We let f µ D 1 and f x x D α D ˙1 and consider the system xP D f (xI µ) D µ C α x 2

(4.67)

When α D 1, (4.67) does not have any ﬁxed points for µ < 0 but has the two nontrivial ﬁxed points p p x D µ and x D µ for µ > 0. The Jacobian matrix has a single eigenvalue given by λ D 2x p Thus, the ﬁxed point x D µ is a stable node because λ < 0, and the ﬁxed point p x D µ is an unstable node because λ > 0. In Figure 4.1a, we display the different ﬁxed-point solutions of (4.67) and their stability in the x µ space. We use broken and solid lines to depict branches of unstable and stable ﬁxed points, respectively. On the other hand, when α D 1, (4.67) does not have any ﬁxed points for µ > 0 but has the two nontrivial ﬁxed points p p x D µ and x D µ for µ < 0. The Jacobian matrix has a single eigenvalue given by λ D 2x p Thus, the ﬁxed point x D µ is an unstable node because λ > 0, and the ﬁxed p point x D µ is a stable node because λ < 0. In Figure 4.1b, we display the different ﬁxed-point solutions of (4.67) and their stability in the x µ space. We note that f (xI µ) D 0 and f x (xI µ) D 0 at (0, 0), and hence there is a nonhyperbolic ﬁxed point at µ D 0. Moreover, we note that there is a change in the number of ﬁxed points from zero to two as µ passes through zero. Therefore, the origin of the x µ space is a static bifurcation. It is clear from Figure 4.1 that the stable and unstable branches meet at the bifurcation point and have the same tangent. Therefore, this bifurcation is called a tangent bifurcation. Although branches of stable and unstable nodes meet at this bifurcation point, the tangent bifurcation is also called saddle-node bifurcation because, in higher-dimensional systems, branches of saddle points and stable nodes meet at such static bifurcation points. Equation 4.67 is the normal form for a generic saddle-node bifurcation of a ﬁxed point of a continuous system.

109

110

4 Bifurcations of Continuous Systems 0.4

0.4

0.2

0.2

x 0.0

x 0.0

–0.2

–0.2

–0.4 0.00

0.05

(a)

0.10 μ

0.15

0.20

–0.4 –0.20

(b)

–0.15

–0.10 μ

–0.05

0.00

Figure 4.1 Scenario in the vicinity of a saddle-node bifurcation: (a) xP D µ x 2 and (b) xP D µ C x 2.

4.4.2 Nonbifurcation Point

We consider the case f µ ¤ 0, f x x D 0, and f x x x ¤ 0. As µ ! 0 and x ! 0, (4.63) tends to xP D f µ µ C

1 6

f xxx x3 C

(4.68)

It has only one nontrivial ﬁxed point when f µ µ ¤ 0 given by

x D

6 f µ µ fxxx

1/3 (4.69)

The eigenvalue associated with this ﬁxed point is 1 λ D fxxx 2

6 f µ µ fxxx

2/3 (4.70)

It follows from (4.69) that this ﬁxed point exists on both sides of f µ µ and it follows from (4.70) that it is stable when f x x x < 0 and unstable when f x x x > 0. Although the ﬁxed point (4.69) is nonhyperbolic because f (xI µ) D 0 and λ D 0 at µ D 0, this ﬁxed point is not a bifurcation point because there is no qualitative change either in the number of ﬁxed-point solutions or in the stability of the ﬁxed-point solutions as µ passes through zero in the state-control space. Example 4.7 We let f µ D 1 and f x x x D 6 in (4.68) and consider xP D f (x, µ) D µ x 3 We have only one ﬁxed point, namely, x D µ 1/3

(4.71)

4.4 Local Bifurcations of Fixed Points

1.0 0.5 x 0.0 –0.5 –1.0 –1.0

–0.5

0.0 μ

0.5

1.0

Figure 4.2 Fixed-point solutions of (4.71).

This solution is depicted in Figure 4.2. At the origin of the x µ space, f (x, µ) D 0 and f x has a zero eigenvalue, implying that x D 0 is a nonhyperbolic ﬁxed point at µ D 0. However, (0, 0) is not a bifurcation point because there is no qualitative change either in the number of ﬁxed-point solutions or in the stability of the ﬁxedpoint solutions as µ passes through zero in the state-control space.

4.4.3 Transcritical Bifurcation

We consider the case f µ D 0 and f x x ¤ 0. As µ ! 0 and x ! 0, (4.63) tends to xP D 12 ( f µ µ µ 2 C 2 f x µ x µ C f x x x 2 ) C

(4.72)

whose ﬁxed points are x1 D x2

D

fxµ C fxµ

q q

f x2µ f µ µ f x x fxx f x2µ f µ µ f x x fxx

µ

and

µ

(4.73)

They exist when f x2µ f µ µ f x x > 0. Their associated eigenvalues are λ(x1 ) D

q

f x2µ f µ µ f x x µ

and

λ(x2 ) D

q

f x2µ f µ µ f x x µ

(4.74)

Therefore, x1 is stable when µ < 0 and unstable when µ > 0. On the other hand, x2 is unstable when µ < 0 and stable when µ > 0. In this case, we have two branches of ﬁxed points that interchange stability at (0, 0). Therefore, the bifurcation of the nonhyperbolic ﬁxed point at the origin as µ passes through zero is transcritical bifurcation.

111

112

4 Bifurcations of Continuous Systems

Example 4.8 We let f µ µ D 0, f µ x D 1, and f x x D 2 and consider the system xP D f (xI µ) D µ x x 2

(4.75)

There are two ﬁxed points: x D0I

trivial ﬁxed point

xDµI

nontrivial ﬁxed point

The Jacobian matrix f x D µ 2x has the single eigenvalue λDµ

at x D 0

λ D µ

at x D µ

In the corresponding bifurcation diagram shown in Figure 4.3, the ﬁxed point x D 0 is a nonhyperbolic ﬁxed point at µ D 0. At this point, a static bifurcation occurs because there is an exchange of stability between the trivial and nontrivial branches. The bifurcation point in Figure 4.3 is a transcritical bifurcation point. We point out that all of the branches that meet at this bifurcation point do not have the same tangent. Equation 4.75 is the normal form for a generic transcritical bifurcation of a ﬁxed point of a continuous system.

0.2 0.1 x 0.0 0.1 0.2

0.2

0.1

0.0 μ

0.1

0.2

Figure 4.3 Scenario in the vicinity of a transcritical bifurcation.

4.4 Local Bifurcations of Fixed Points

4.4.4 Pitchfork Bifurcation

We consider the case f µ D 0 and f x x D 0. As µ ! 0 and x ! 0, (4.63) tends to xP D f x µ x µ C

1 6

f xxx x3 C

(4.76)

whose ﬁxed points are s x1 D 0 ,

x2 D

6 f x µ µ , fxxx

s and

x3 D

6 f x µ µ fxxx

The second and third ﬁxed points exist only when f x µ f x x x µ < 0. The Jacobian matrix in this case f x D f µx µ C

1 2

f xxx x2

has the single eigenvalue λ(x1 ) D f µ x µ ,

λ(x2 ) D 2 f µ x µ ,

λ(x3 ) D 2 f µ x µ

Consequently, the trivial ﬁxed point is stable when f µ x µ < 0 and unstable when f µ x µ > 0. On the other hand, the nontrivial ﬁxed points exist and are stable when f x x x < 0 and f µ x µ > 0 and exist and are unstable when f x x x > 0 and f µ x µ < 0. Example 4.9 We let f µ µ D 0, f µ x D 1, and f x x x /6 D α D ˙1 and consider the system xP D f (xI µ) D µ x C α x 3

(4.77)

where µ is again the scalar control parameter. There are three ﬁxed points: x D 0 I trivial ﬁxed point p x D ˙ µ/α I nontrivial ﬁxed points In this case, the Jacobian matrix f x D µ C 3α x 2 has the single eigenvalue λDµ

at x D 0

λ D 2µ

p at x D ˙ µ/α

Consequently, the trivial ﬁxed point is stable when µ < 0 and unstable when µ > 0. On the other hand, when α < 0, nontrivial ﬁxed points exist only when µ > 0 and they are stable. However, when α > 0, nontrivial ﬁxed points exist only when

113

114

4 Bifurcations of Continuous Systems 0.4

0.4

0.2

0.2

x 0.0

x 0.0

–0.2

–0.2

–0.4 –0.2

–0.1

(a)

0.0 μ

0.1

0.2

–0.4 –0.2

(b)

–0.1

0.0 μ

0.1

0.2

Figure 4.4 Scenario in the vicinity of a pitchfork bifurcation: (a) supercritical pitchfork bifurcation (α D 1) and (b) subcritical pitchfork bifurcation (α D 1).

µ < 0 and they are unstable. The bifurcation diagrams of Figure 4.4a,b correspond to α D 1 and α D 1, respectively. In both cases, we note the following at (0, 0): (a) f (x, µ) D 0, (b) f x has a zero eigenvalue, (c) the number of ﬁxed-point solutions for µ < 0 is different from that for µ > 0, and (d) there is a change in the stability of the trivial ﬁxed point as we pass through µ D 0. Hence, the origin of the statecontrol space is a bifurcation point. p p When α D 1, two stable branches of ﬁxed points x D µ and x D µ bifurcate from the bifurcation point, as shown in Figure 4.4a. When α D 1, two p p unstable branches of ﬁxed points x D µ and x D µ bifurcate from the bifurcation point, as shown in Figure 4.4b. The bifurcations observed in Figure 4.4a,b are called pitchfork bifurcations because the bifurcating nontrivial branches have the geometry of a pitchfork at (0, 0). Speciﬁcally, the bifurcation in Figure 4.4a is called a supercritical pitchfork bifurcation, and the bifurcation in Figure 4.4b is called a subcritical or reverse pitchfork bifurcation. In the case of a supercritical pitchfork bifurcation, locally we have a branch of stable ﬁxed points on one side of the bifurcation point and two branches of stable ﬁxed points and a branch of unstable ﬁxed points on the other side of the bifurcation point. In the case of a subcritical pitchfork bifurcation, locally we have two branches of unstable ﬁxed points and a branch of stable ﬁxed points on one side of the bifurcation point and a branch of unstable ﬁxed points on the other side of the bifurcation point. Equation 4.77 is the normal form for a generic pitchfork bifurcation of a ﬁxed point of a continuous system.

4.4.5 Hopf Bifurcations

When a scalar control parameter µ is varied, a Hopf bifurcation of a ﬁxed point of a system such as (4.17) is said to occur at µ D µ c if the following conditions (Marsden and McCracken, 1976) are satisﬁed:

4.4 Local Bifurcations of Fixed Points

1. F(x I µ c ) D 0, 2. The matrix D x F has a pair of purely imaginary eigenvalues ˙i ω while all of its other eigenvalues have nonzero negative real parts at (x I µ c ), 3. For µ ' µ c , let the analytic continuation of the pair of purely imaginary eigenvalues be λ r ˙ i ω. Then, Real(d λ r /d µ) ¤ 0 at µ D µ c . This condition implies a transversal or nonzero speed crossing of the imaginary axis and hence is called a transversality condition. Again, the ﬁrst two conditions imply that the ﬁxed point undergoing the bifurcation is a nonhyperbolic ﬁxed point. When all of the above three conditions are satisﬁed, a periodic solution of period 2π/ω is born at (x I µ c ); bifurcating periodic solutions can also occur when the transversality condition is not satisﬁed (e.g., Marsden and McCracken, 1976). It is to be noted that bifurcating periodic solutions can also occur under certain other degenerate conditions (e.g., Golubitsky and Schaeffer, 1985). In such cases, we have degenerate Hopf bifurcations. The Hopf bifurcation is also called the Poincaré–Andronov–Hopf bifurcation (e.g., Wiggins, 1990) to give credit to the works of Poincaré and Andronov that preceded the work of Hopf. As pointed out in the literature (e.g., Abed, 1994; Arnold, 1988), Poincaré (1899) was aware of the conditions for this bifurcation to occur. (Poincaré studied such bifurcations in the context of lunar orbital dynamics.) Andronov and his coworkers studied Hopf bifurcations in planar systems before Hopf studied such bifurcations in general n-dimensional systems (Andronov et al., 1966; Arnold, 1988). In aeroelasticity, the consequence of a Hopf bifurcation is known as galloping or ﬂutter. Example 4.10 We consider the planar system xP D µ x ω y C (α x β y )(x 2 C y 2 )

(4.78)

yP D ωx C µ y C (β x C α y )(x 2 C y 2 )

(4.79)

where x and y are the states and µ is the scalar control parameter. The ﬁxed point (0, 0) is a solution of (4.78) and (4.79) for all values of µ. The eigenvalues of the corresponding Jacobian matrix there are λ1 D µ i ω

and

λ2 D µ C i ω

From these eigenvalues, we note that (0, 0) is a nonhyperbolic ﬁxed point of (4.78) and (4.79) when µ D 0. Further, at (x, y, µ) D (0, 0, 0), we note that d λ1 D 1 and dµ

d λ2 D1 dµ

115

116

4 Bifurcations of Continuous Systems

Hence, the three conditions required for a Hopf bifurcation are satisﬁed, and a Hopf bifurcation of the ﬁxed point (0, 0) of (4.78) and (4.79) occurs at µ D 0. The period of the bifurcating periodic solution at (0, 0, 0) is 2π/ω. By using the transformation x D r cos θ

and

y D r sin θ

we transform (4.78) and (4.79) into rP D µ r C α r 3

(4.80)

θP D ω C β r 2

(4.81)

The trivial ﬁxed point of (4.80) corresponds to the ﬁxed point (0, 0) of (4.78) and (4.79), and a nontrivial ﬁxed point (i.e., r ¤ 0) of (4.80) corresponds to a periodic solution of (4.78) and (4.79). In the latter case, r is the amplitude and θP is the frequency of the periodic solution that is created due to the Hopf bifurcation. A stable nontrivial ﬁxed point of (4.80) corresponds to a stable periodic solution of (4.78) and (4.79). Likewise, an unstable nontrivial ﬁxed point of (4.80) corresponds to an unstable periodic solution of (4.78) and (4.79). We note that (4.80) is identical to (4.77), so the Hopf bifurcation at (0, 0, 0) in the x y µ space is equivalent to a pitchfork bifurcation at (0, 0) in the r µ space. When α D 1, we have a supercritical pitchfork bifurcation in the r µ space and, hence, a supercritical Hopf bifurcation in the x y µ space. When α D 1, we have a subcritical pitchfork bifurcation in the r µ space and, hence, a subcritical Hopf bifurcation in the x y µ space. The bifurcation diagrams for α D 1 are shown in Figure 4.5a,c and α D 1 are shown in Figure 4.5b,d, respectively. In y

y x

x

0

μ

(a)

μ

(b)

r 0

(c)

0

μ

(d)

0

μ

Figure 4.5 Local scenarios: (a,c) supercritical Hopf bifurcation and (b,d) subcritical Hopf bifurcation.

4.5 Normal Forms of Static Bifurcations

Figure 4.5a,b, the bifurcating periodic solutions in the x y µ space are depicted as parabolic surfaces. In the case of a supercritical Hopf bifurcation, locally we have a branch of stable ﬁxed points on one side of the bifurcation point and a branch of unstable ﬁxed points and a branch of stable periodic solutions on the other side of the bifurcation point. In the case of a subcritical Hopf bifurcation, locally we have a branch of unstable periodic solutions and a branch of stable ﬁxed points on one side of the bifurcation point and a branch of unstable ﬁxed points on the other side of the bifurcation point. When α ¤ 0, (4.78) and (4.79) are the normal form for a generic Hopf bifurcation of a ﬁxed point of a continuous system. On the other hand, when α D 0 in (4.78) and (4.79), although the conditions for a Hopf bifurcation are satisﬁed, there are no periodic orbits in the vicinity of the bifurcation point to third order. This case is degenerate.

4.5 Normal Forms of Static Bifurcations

In this section, we consider reduction of the nonlinear continuous system (4.17) near a static bifurcation ﬁxed point to its normal form. We assume that the ﬁxed point has been shifted to x D 0 and its corresponding control parameter has been shifted to µ D 0. Moreover, we expand (4.17) in a Taylor series for small x and µ and obtain xP D Ax C b µ C B µ x C Q(x, x) C C (x, x, x) C

(4.82)

where A D D x F, B D D µ x F, and b D D µ F at (x, µ) D (0, µ) and Q(x, x) and C (x, x, x) are bilinear and trilinear column vectors involving quadratic and cubic terms, respectively. Because the ﬁxed point (x, µ) D (0, 0) is a static bifurcation point, one of the eigenvalues of the matrix A is zero and all of its other eigenvalues are in the left-half of the complex plane. Next, we demonstrate how one can use the method of multiple scales, a combination of center-manifold reduction and the method of normal forms, and a projection method to compute the normal forms of saddle-node, transcritical, and pitchfork bifurcations. 4.5.1 The Method of Multiple Scales

To compute the normal form of the static bifurcation of (4.82) at the origin, we introduce a small nondimensional parameter as a bookkeeping parameter and seek a third-order approximate solution of (4.82) in the form x(tI µ) D x 1 (T0 , T1 , T2 ) C 2 x 2 (T0 , T1 , T2 ) C 3 x 3 (T0 , T1 , T2 ) C

(4.83)

where the time scales Tm D m t. The time derivative can be expressed in terms of these scales as d (4.84) D D0 C D1 C 2 D2 C dt

117

118

4 Bifurcations of Continuous Systems

where D m D @/@Tm . Moreover, because µ is small, we expand it in terms of as µ D µ 1 C 2 µ 2 C 3 µ 3 C

(4.85)

Substituting (4.83)–(4.85) into (4.82) and equating coefﬁcients of like powers of , we obtain Order ()

D0 x 1 Ax 1 D bµ 1

(4.86)

Order (2 )

D0 x 2 Ax 2 D bµ 2 D1 x 1 C B x 1 µ 1 C Q(x 1 , x 1 )

(4.87)

Order (3 )

D0 x 3 Ax 3 D b µ 3 D1 x 2 D2 x 1 C B x 2 µ 1 C B x 1 µ 2 C 2Q(x 1 , x 2 ) C C (x 1 , x 1 , x 1 )

(4.88)

The general solution of (4.86) is the superposition of a homogeneous solution x 1h and a particular solution x 1p . For the homogeneous solution, we denote the eigenvalues of the matrix A by λ m , m D 1, 2, . . . , n and order them so that λ 1 D 0. We let p m be the eigenvector (generalized eigenvector) corresponding to λ m . Then, all of the terms corresponding to λ m and p m for m > 2 decay with time so that the long-time dynamics is given by x 1h D u(T1 , T2 )p

(4.89)

where the subscript in p 1 has been suppressed and the function u(T1 , T2 ) is determined by imposing solvability conditions at the higher-order approximations. We note that A is singular because one of its eigenvalues is zero. Because the term bµ 1 on the right-hand side of (4.86) is independent of T0 , the particular solution of (4.86) is given by Ax 1p D b µ 1

(4.90)

Because A is singular, (4.90) has a solution if and only if its right-hand side bµ 1 is orthogonal to every solution of the adjoint homogeneous equation; that is solutions of qT A D 0 which has a nontrivial solution because A is singular. This condition demands that q T b µ 1 D 0. There are two possibilities. First, q T b ¤ 0 and hence µ 1 D 0. Second, q T b D 0 and hence µ 1 is arbitrary. In the latter case, the solution is not unique. To make it unique, we require it to be orthogonal to the left eigenvector q and write the particular solution as x 1p D c µ 1

where

Ac D b

(4.91)

and q T c D 1. We consider these two cases separately, starting with the ﬁrst case.

4.5 Normal Forms of Static Bifurcations

The Case q T b ¤ 0 In this case, the solution of the ﬁrst-order problem, (4.86), can be expressed as

x 1 D u(T1 , T2 )p

(4.92)

Substituting (4.92) into (4.87) yields D0 x 2 Ax 2 D b µ 2 D1 u p C Q(p , p )u2

(4.93)

Again, the general solution of (4.93) is the superposition of a homogeneous term and a particular term. The homogeneous solution is the same as that given in (4.89) and hence is neglected. A particular solution of (4.93) is given by Ax 2 D bµ 2 C D1 u p Q(p , p )u2

(4.94)

Equation 4.94 has a solution if and only if its right-hand side is orthogonal to q; that is, D1 u D q T bµ 2 C q T Q(p , p )u2

(4.95)

Hence, there is a generic saddle-node bifurcation at the origin of (4.82) if q T Q(p, p ) ¤ 0. When q T Q(p , p ) D 0, it follows from (4.95) that µ 2 D 0 and D1 u D 0 or u D u(T2 ). Then, a unique particular solution of (4.94) is given by x 2 D z u2

where

Az D Q(p , p )

and

qT z D 1

(4.96)

Substituting (4.92) and (4.96) into (4.88) and using the fact that µ 1 D 0, µ 2 D 0, and D1 u D 0, we obtain D0 x 3 Ax 3 D b µ 3 D2 u p C 2Q(z, p ) C C (p , p , p ) u3

(4.97)

Equation 4.97 has a solution if and only if its right-hand side is orthogonal to q; that is, D2 u D q T bµ 3 C 2q T Q(z, p ) C q T C (p , p , p ) u3

(4.98)

Therefore, the origin of (4.82) is an inﬂection point and there is no bifurcation there if the coefﬁcient of u3 in (4.98) is different from zero. The Case q T b D 0

In this case, the general solution of (4.86) can be expressed as

x 1 D u(T1 , T2 )p C c µ 1

(4.99)

Substituting (4.99) into (4.87) yields D0 x 2 Ax 2 D b µ 2 p D1 u C B p C 2Q(c, p ) µ 1 u C B c C Q(c, c) µ 21 C Q(p, p )u2

(4.100)

119

120

4 Bifurcations of Continuous Systems

The solvability condition of (4.100) demands that its right-hand side be orthogonal to q, which yields D1 u D q T B c C q T Q(c, c) µ 21 C q T B p C 2q T Q(c, p ) µ 1 u C q T Q(p , p )u2 (4.101) Therefore, there is a generic transcritical bifurcation of (4.82) at the origin when q T Q(p , p ) ¤ 0. When q T Q(p, p ) D 0, it follows from (4.101) that µ 1 D 0 and D1 u D 0 and hence u D u(T2 ). Then, the solution of (4.100) can be expressed as x 2 D c µ 2 C z u2

(4.102)

where c and z are deﬁned in (4.91) and (4.96), respectively. Then, substituting (4.99) and (4.102) into (4.88) and using the fact that µ 1 D 0 and D1 u D 0, we obtain D0 x 3 Ax 3 D b µ 3 D2 u p C B p C 2Q(c, p ) µ 2 u (4.103) C 2Q(z, p ) C C (p, p , p ) u3 Imposing the solvability condition in (4.103) yields D2 u D q T B p C 2q T Q(c, p ) µ 2 u C 2q T Q(z, p ) C q T C (p , p , p ) u3 (4.104) Therefore, there is a generic supercritical pitchfork bifurcation at the origin of (4.82) when 2q T Q(z, p ) C q T C(p , p , p ) < 0 and a generic subcritical pitchfork bifurcation at the origin of (4.82) when 2q T Q(z, p ) C q T C (p , p , p ) > 0. Example 4.11 We construct the normal form of the system xP1 D b 1 µ C 2x1 2x2 C µ x1 C a 11 x12 C a 12 x1 x2 C a 13 x22

(4.105)

xP2 D b 2 µ C 4x1 4x2 4µ x2 C a 21 x12 C a 22 x1 x2 C a 23 x22

(4.106)

near the ﬁxed point (x1 , x2 , µ) D (0, 0, 0). We seek a third-order approximate solution of (4.105) and (4.106) in the form x n (tI µ) D x n1 (T0 , T1 , T2 ) C 2 x n2 (T0 , T1 , T2 ) C 3 x n3 (T0 , T1 , T2 ) C (4.107) Substituting (4.107), (4.84), and (4.85) into (4.105) and (4.106) and equating coefﬁcients of equal powers of , we obtain

4.5 Normal Forms of Static Bifurcations

Order ()

D0 x11 2x11 C 2x21 D µ 1 b 1

(4.108)

D0 x21 4x11 C 4x21 D µ 1 b 2

(4.109)

Order (2 ) 2 D0 x12 2x12 C 2x22 D µ 2 b 1 D1 x11 C µ 1 x11 C a 11 x11 2 C a 12 x11 x21 C a 13 x21

(4.110)

2 D0 x22 4x12 C 4x22 D µ 2 b 2 D1 x21 4µ 1 x21 C a 21 x11 2 C a 22 x11 x21 C a 23 x21

(4.111)

Order (3 )

D0 x13 2x13 C 2x23 D µ 3 b 1 D1 x12 D2 x11 C µ 1 x12 C µ 2 x11 C 2a 11 x11 x12 C a 12 x11 x22 C a 12 x12 x21 C 2a 13 x21 x22

(4.112)

D0 x23 4x13 C 4x23 D µ 3 b 2 D1 x23 D2 x21 4µ 1 x23 4µ 2 x21 C 2a 21 x11 x13 C a 23 x11 x23 C a 23 x13 x21 C 2a 23 x21 x23

(4.113)

In this case, the coefﬁcient matrix of (4.108) and (4.109) is AD

2 4

2 4

(4.114)

whose eigenvalues are λ D 0 and λ D 2. Hence, the origin of (4.105) and (4.106) may be a static bifurcation point. The right and left eigenvectors of A corresponding to λ D 0 are 1 2 pD and q D 1 1 where q T p D 1. Because A is singular, (4.108) and (4.109) have a solution if and only if their right-hand sides are orthogonal to q T ; that is, if and only if (2b 1 b 2 )µ 1 D 0. There are two possibilities: (a) (2b 1 b 2 ) ¤ 0 and hence µ 1 D 0 and (a) (2b 1 b 2 ) D 0 and hence µ 1 is arbitrary. We treat both cases separately, starting with the ﬁrst case. When (2b 1 b 2 ) ¤ 0, the solution of (4.108) and (4.109) can be expressed as x11 D u(T1 , T2 )

and

x21 D u(T1 , T2 )

(4.115)

121

122

4 Bifurcations of Continuous Systems

Substituting (4.115) into (4.110) and (4.111) yields D0 x12 2x12 C 2x22 D µ 2 b 1 D1 u C (a 11 C a 12 C a 13 )u2

(4.116)

D0 x22 4x12 C 4x22 D µ 2 b 2 D1 u C (a 21 C a 22 C a 23 )u2

(4.117)

Demanding that the right-hand side of (4.116) and (4.117) be orthogonal to q yields the normal form D1 u D (2b 1 b 2 )µ 2 C (2a 11 C 2a 12 C 2a 13 a 21 a 22 a 23 )u2

(4.118)

which indicates that the origin of (4.105) and (4.106) undergoes a generic saddlenode bifurcation as µ passes through zero if the coefﬁcient of u2 is different from zero. When this coefﬁcient is zero, one needs to carry out the expansion to higher order. When (2b 1 b 2 ) D 0, the solution of (4.108) and (4.109) that is orthogonal to q can be expressed as x11 D u(T1 , T2 ) C 12 b 1 µ 1 and x21 D u(T1 , T2 ) C b 1 µ 1

(4.119)

Substituting (4.119) into (4.110) and (4.111) yields D0 x12 2x12 C 2x22 D µ 2 b 1 D1 u C 12 C 14 a 11 b 1 C 12 a 12 b 1 C a 13 b 1 b 1 µ 21 C 1 C a 11 b 1 C 32 a 12 b 1 C 2a 13 b 1 µ 1 u C (a 11 C a 12 C a 13 ) u2

(4.120)

D0 x22 4x12 C 4x22 D µ 2 b 2 D1 u C 4 C 14 a 21 b 1 C 12 a 22 b 1 C a 23 b 1 b 1 µ 21 C 4 C a 21 b 1 C 32 a 22 b 1 C 2a 23 b 1 µ 1 u C (a 21 C a 22 C a 23 )u2 (4.121) The solvability condition of (4.120) and (4.121) yields the following normal form of (4.105) and (4.106): D1 u D Γ0 µ 21 C Γ1 µ 1 u C Γ2 u2

(4.122)

where Γ0 D 14 b 1 (20 C 2a 11 b 1 C 4a 12 b 1 C 8a 13 b 1 a 21 b 1 2a 22 b 1 4a 23 b 1 ) (4.123) Γ1 D

1 2

(12 C 4a 11 b 1 C 6a 12 b 1 C 8a 13 b 1 2a 21 b 1 3a 22 b 1 4a 23 b 1 ) (4.124)

Γ2 D 2a 11 C 2a 12 C 2a 13 a 21 a 22 a 23

(4.125)

Equation 4.122 indicates that the origin of (4.105) and (4.106) undergoes a generic transcritical bifurcation as µ passes through zero if Γ2 ¤ 0.

4.5 Normal Forms of Static Bifurcations

When Γ2 D 0, it follows from (4.122) that µ 1 D 0 and D1 u D 0 or u D u(T2 ). Then, the particular solution of (4.120) and (4.121) that is orthogonal to q T can be expressed as x12 D 12 b 1 µ 2 C 12 (a 11 C a 12 C a 13 )u2

and

x22 D b 1 µ 2 C(a 11 C a 12 C a 13 )u2 (4.126)

Substituting (4.119) and (4.126) into (4.112) and (4.113) and recalling that µ 1 D 0 and D1 u D 0, we obtain D0 x13 2x13 C 2x23 D µ 3 b 1 D2 u C C

1 4

1 2

(2 C 2a 11 b 1 C 3a 12 b 1 C 4a 13 b 1 ) µ 2 u

(2a 11 C 3a 12 C 4a 13 ) (a 21 C a 22 C a 23 ) u3

(4.127)

D0 x23 4x13 C 4x23 D µ 3 b 2 D2 u C

1 4

1 2

(8 2a 21 b 1 3a 23 b 1 4a 23 b 1 ) µ 2 u

(a 21 C a 23 C a 23 ) (2a 21 C 3a 23 C 4a 23 ) u3

(4.128)

Imposing the solvability condition for (4.127) and (4.128) yields D2 u D νµ 2 u C α u3

(4.129)

where νD αD

1 2 1 4

(12 C 4a 11 b 1 C 6a 12 b 1 C 8a 13 b 1 2a 21 b 1 3a 22 b 1 4a 23 b 1 ) (4a 11 C 6a 12 C 8a 13 2a 21 3a 22 4a 23 ) (a 21 C a 22 C a 23 )

It follows from (4.139) that, as µ passes through zero, the origin of (4.105) and (4.106) undergoes a generic supercritical pitchfork bifurcation when α < 0 and a generic subcritical pitchfork bifurcation when α > 0. Example 4.12 We consider the three-dimensional dynamic system xP 1 D b 1 µ x1 C x2 2x3 C µ x1 C α 1 (x1 C 2x2 C x3 )2

(4.130)

xP 2 D b 2 µ 2x1 x2 x3 C µ x2 C α 2 (x1 x2 C x3 )2

(4.131)

xP 3 D b 3 µ C x1 C 2x2 x3 2µ x3 C α 3 (2x1 C x2 x3 )2

(4.132)

Substituting (4.107), (4.84), and (4.85) into (4.130)–(4.132) and equating coefﬁcients of like powers of yields.

123

124

4 Bifurcations of Continuous Systems

Order ()

D0 x11 C x11 x21 C 2x31 D b 1 µ 1

(4.133)

D0 x21 C 2x11 C x21 C x31 D b 2 µ 1

(4.134)

D0 x31 x11 2x21 C x31 D b 3 µ 1

(4.135)

Order (2 )

D0 x12 C x12 x22 C 2x32 D D1 x11 C b 1 µ 2 C µ 1 x11 C α 1 (x11 C 2x21 C x31 )2

(4.136)

D0 x22 C 2x12 C x22 C x32 D D1 x21 C b 2 µ 2 C µ 1 x21 C α 2 (x11 x21 C x31 )2

(4.137)

D0 x32 x12 2x22 C x32 D D1 x31 C b 3 µ 2 2µ 1 x31 C α 3 (2x11 C x21 x31 )2

(4.138)

Order (3 )

D0 x13 C x13 x23 C 2x33 D D1 x12 D2 x11 C b 1 µ 3 C µ 2 x11 C µ 1 x12 C 2α 1 (x11 C 2x21 C x31 ) (x12 C 2x22 C x32 ) (4.139) D0 x23 C 2x13 C x23 C x33 D D1 x22 D2 x21 C b 2 µ 3 C µ 2 x21 C µ 1 x22 C 2α 2 (x11 x21 C x31 ) (x12 x22 C x32 ) (4.140) D0 x33 x13 2x23 C x33 D D1 x32 D2 x31 C b 3 µ 3 2µ 2 x31 2µ 1 x32 C 2α 3 (2x11 C x21 x31 ) (2x12 C x22 x32 ) (4.141) In this case, the coefﬁcient matrix of (4.133)–(4.135) is 1 0 1 1 2 @2 1 1A 1

2

(4.142)

1

p whose eigenvalues are λ D 0 and λ D 3/2(1˙ i 3). Hence, the origin of (4.130)– (4.133) may be a static bifurcation point. The right and left eigenvectors of A corresponding to λ D 0 are 2 3 2 3 1 1 1 p D 4 1 5 and q D 4 1 5 3 1 1

4.5 Normal Forms of Static Bifurcations

where q T p D 1. Because A is singular, (4.133)–(4.135) have a solution if and only if their right-hand sides are orthogonal to q T ; that is, if and only if (b 3 Cb 2 b 1 )µ 1 D 0. There are two possibilities: (a) (b 3 C b 2 b 1 ) ¤ 0 and hence µ 1 D 0 and (a) (b 3 C b 2 b 1) D 0 and hence µ 1 is arbitrary. We treat both cases separately, starting with the ﬁrst case. When (b 3 C b 2 b 1 ) ¤ 0, the nondecaying solution of (4.133)–(4.135) can be expressed as x11 D u(T1 , T2 ) ,

x21 D u(T1 , T2 ) ,

and

x31 D u(T1 , T2 )

(4.143)

Substituting (4.143) into (4.136)–(4.138) yields D0 x12 C x12 x22 C 2x32 D D1 u C 4u2 α 1 C b 1 µ 2

(4.144)

D0 x22 C 2x12 C x22 C x32 D D1 u C u2 α 2 C b 2 µ 2

(4.145)

D0 x32 x12 2x22 C x32 D D1 u C 4u2 α 3 C b 3 µ 2

(4.146)

Demanding that the right-hand side of (4.144)–(4.146) be orthogonal to q yields the normal form D1 u D

1 3

(b 3 C b 2 b 1 ) µ 2 C

1 3

(4α 3 C α 2 4α 1 ) u2

(4.147)

which indicates that the origin of (4.130)–(4.132) undergoes a generic saddle-node bifurcation as µ passes through zero if the coefﬁcient of u2 is different from zero. When this coefﬁcient is zero, one needs to carry out the expansion to higher order. When (b 3 C b 2 b 1 ) D 0, the solution of (4.133)–(4.135) that is orthogonal to q can be expressed as x11 D u(T1 , T2 ) C 13 b 2 µ 1 x21 D u(T1 , T2 ) 13 b 3 µ 1 x31 D u(T1 , T2 ) C 13 (b 3 C b 2 )µ 1

(4.148)

To simplify the algebra, we let b 1 D 2b, b 2 D b, and b 3 D b. Substituting (4.148) into (4.136)–(4.138) yields D0 x12 C x12 x22 C 2x32 D D1 u C 19 b (3 C b α 1 ) µ 21 C 2b µ 2 13 (3 4b α 1 ) µ 1 u C 4α 1 u2

(4.149)

D0 x22 C 2x12 C x22 C x32 D D1 u 19 b (3 16b α 2 ) µ 21 C b µ 2 C

1 3

(3 8b α 2 ) µ 1 u C α 2 u2

(4.150)

D0 x32 x12 2x22 C x32 D D1 u 19 b (12 b α 3 ) µ 21 C b µ 2 23 (3 2b α 3 ) µ 1 u C 4α 3 u2

(4.151)

The solvability condition of (4.149)–(4.151) yields the normal form of (4.130)– (4.132) as D1 u D Γ0 µ 21 C Γ1 µ 1 u C Γ2 u2

(4.152)

125

126

4 Bifurcations of Continuous Systems

where 1 Γ0 D 27 b (18 C b α 1 16b α 2 b α 3 )

Γ1 D 49 b (α 3 2α 2 α 1 ) Γ2 D

1 3

(4α 3 C α 2 4α 1 )

Equation 4.152 indicates that the origin of (4.130)–(4.132) undergoes a generic transcritical bifurcation as µ passes through zero if the coefﬁcient of u2 is different from zero; that is, (4α 3 C α 2 4α 1 ) ¤ 0. When Γ2 D 0, it follows from (4.152) that µ 1 D 0 and D1 u D 0 or u D u(T2 ). Then, the particular solution of (4.144)–(4.146) that is orthogonal to q T can be expressed as x12 D 13 b µ 2 C 13 α 2 u2 x22 D 31 b µ 2 43 α 3 u2 x32 D 23 b µ 2 C 13 α 2 C 43 α 3 u2

(4.153)

Substituting (4.148) and (4.153) into (4.139)–(4.141) and recalling that µ 1 D 0 and D1 u D 0, we obtain D0 x13 C x13 x23 C 2x33 D D2 u C 2b µ 3 C

1 3

(4b α 3 C b α 2 3) µ 2 u (4.154)

C 23 (α 2 2α 3 )(α 2 C 4α 3 )u3

D0 x23 C 2x13 C x23 C x33 D D2 u C b µ 3 13 (8b α 2 3) µ 2 u 43 α 2 (α 2 C 4α 3 ) u3 D0 x32 x12 2x22 C x32 D D2 u C b µ 3 C

2 3

(4.155)

(2b α 3 3) µ 2 u

C (8α 3 α 2 ) α 3 u3 4 3

(4.156)

Demanding that the right-hand side of (4.154)–(4.156) be orthogonal to q and using the fact that α 1 D 1/4α 2 C α 3 yields the normal form (4.157) D2 u D b α 2 µ 2 u C 13 16α 23 8α 2 α 3 2α 22 u3 It follows from (4.157) that the origin of (4.130)–(4.132) undergoes a supercritical pitchfork bifurcation as µ passes through zero when 16α 23 8α 2 α 3 2α 22 < 0 and a subcritical pitchfork bifurcation when (16α 23 8α 2 α 3 2α 22 ) > 0. 4.5.2 Center-Manifold Reduction

To compute the normal form of the static bifurcation of (4.82) at the origin by using center-manifold reduction, we ﬁrst calculate the eigenvalues λ 1 , λ 2 , . . . , λ n

4.5 Normal Forms of Static Bifurcations

and eigenvectors (generalized eigenvectors) p 1 , p 2 , . . . , p n of the matrix A. Second, we arrange the eigenvalues of A so that λ 1 D 0 and let p 1 be be its corresponding eigenvector. Third, we introduce the transformation x D P y and obtain yP D J y C P 1 b µ C P 1 B P y µ C P 1 Q(P y, P y) C P 1 C(P y, P y, P y) (4.158) where J D P 1 AP . We note that J can be rewritten as JD

0 0

0 Js

(4.159)

where J s is an (n 1) (n 1) matrix whose eigenvalues are λ 2 , λ 3 , . . . , λ n . Fourth, we let y s be the (n 1)-dimensional vector with the components y 2 , y 3 , . . . , y n and rewrite (4.158) as yP 1 D bO 1 µ C

n X

ξi y i µ C f 2 (y 1 , y s ) C f 3 (y 1 , y s )

(4.160)

iD1

and yP s D J s y s C bO s µ C ζ y 1 µ C BO y µ C F 2 (y 1 , y s ) C F 3 (y 1 , y s )

(4.161)

where bO D P 1 b, f 2 and F 2 are the ﬁrst and last (n 1) components of P 1 Q(P y, P y), and f 3 and F 3 are the ﬁrst and last (n 1) components of P 1 C (P y, P y, P y). To determine the dependence of the center manifold on µ, we augment (4.160) and (4.161) with the additional equation µP D 0

(4.162)

Because the f i and F i are polynomials and hence inﬁnitely differentiable, there exists a local center manifold of the form (Carr, 1981) y s D h(y 1 I µ)

(4.163)

where h is a polynomial function of y 1 and µ such that h(0, 0) D 0,

D y 1 h i (0I 0) D 0 ,

and

D µ h i (0I 0) D 0

(4.164)

where the h i are the scalar components of h. Substituting (4.163) into (4.161) yields " # n X 0 O b h (y 1 ) 1 µ C ξ1 y 1 µ C ξi h i (y 1 )µ C f 2 (y 1 , h(y 1 ) iD2

D J s h(y 1 ) C bO s µ C ζ y 1 µ C BO h(y 1 )µ C F 2 [y 1 , h(y 1 )] C

(4.165)

127

128

4 Bifurcations of Continuous Systems

To solve (4.165), one approximates the components of h(y 1 I µ) with polynomials. The polynomial approximations are usually taken to be quadratic to the ﬁrst approximation and do not contain constant and linear terms so that the conditions (4.164) are satisﬁed. Substituting the assumed quadratic polynomial approximations into (4.165) and equating the coefﬁcients of the different terms in the polynomials on both sides, one obtains a system of algebraic equations for the coefﬁcients of the polynomials. Solving these equations, we obtain a ﬁrst approximation to the center manifold y s D h(y 1 I µ). Finally, substituting this approximation into (4.160), we obtain the one-dimensional dynamical system yP1 D bO 1 µ C ξ1 y 1 µ C

n X

ξi y i µ C f 2 [y 1 , h(y 1 )] C f 3 [y 1 , h(y 1 )]

(4.166)

iD2

describing the dynamics of the system (4.82) on the center manifold. Next, we consider two examples. Example 4.13 To reduce the algebra, we consider a special case of Example 4.11, namely xP1 D b 1 µ C 2x1 2x2 C µ x1 C α 1 x12

(4.167)

xP2 D b 2 µ C 4x1 4x2 4µ x2 C α 2 x1 x2

(4.168)

The coefﬁcient matrix of this system at (0, 0, 0) is given by (4.114). Its eigenvalues are λ 1 D 0 and λ 2 D 2 and their corresponding eigenvectors are p1 D

1 1

and

p2 D

1 2

Hence, PD

1 1

1 2

and

P 1 D

2 1

1 1

Next, we introduce the transformation x1 1 1 y1 D 1 2 y2 x2 into (4.167) and (4.168) and obtain yP1 D (2b 1 b 2 )µ C 6y 1 µ C 10y 2 µ C (2α 1 α 2 )y 12 C (4α 1 3α 2 )y 1 y 2 C 2(α 1 α 2 )y 22

(4.169)

yP2 D 2y 2 C (b 2 b 1 )µ 5y 1 µ 9y 2 µ (α 1 α 2 )y 12 (2α 1 3α 2 )y 1 y 2 (α 1 2α 2 )y 22

(4.170)

4.5 Normal Forms of Static Bifurcations

We seek the center manifold of (4.169) and (4.170) in the form y 2 (tI µ) D h(y 1I µ) and rewrite (4.169) as yP1 D (2b 1 b 2 )µ C 6y 1 µ C 10h µ C (2α 1 α 2 )y 12 C (4α 1 3α 2 )y 1 h C 2(α 1 α 2 )h 2

(4.171)

Substituting (4.171) and the expression for the center manifold into (4.170) yields h 0 (2b 1 b 2 )µ C 6y 1 µ C 10h µ C (2α 1 α 2 )y 12 D 2h C (b 2 b 1 )µ 5y 1 µ 9h µ (α 1 α 2 )y 12 (2α 1 3α 2 )y 1 h (α 1 2α 2 )h 2 C (4.172) whose approximate solution can be expressed as h D 12 (b 2 b 1 )µ 12 (α 1 α 2 )y 12 C

(4.173)

Substituting (4.173) into (4.171) yields the following one-dimensional equation describing the dynamics on the center manifold: yP1 D (2b 1 b 2 )µ C 5(b 2 b 1 ) C 14 (α 1 α 2 )(b 2 b 1 )2 µ 2 C (2α 1 α 2 )y 12 C 6 C 12 (b 2 b 1 )(4α 1 3α 2 ) y 1 µ 12 (α 1 α 2 )(4α 1 3α 2 )y 13 (4.174) There are two possibilities: (a) b 2 ¤ 2b 1 and (b) b 2 D 2b 1 . In the ﬁrst case, as y 1 ! 0 and µ ! 0, (4.174) tends to yP1 D (2b 1 b 2 )µ C (2α 1 α 2 )y 12

(4.175)

Therefore, the origin of (4.167) and (4.168) undergoes a saddle-node bifurcation as µ passes through zero when α 2 ¤ 2α 1 . When α 2 D 2α 1 , the origin of (4.167) and (4.168) is not a bifurcation point. When b 2 D 2b 1 , as y 1 ! 0 and µ ! 0, (4.174) tends to 1 1 yP1 D 5b 1 C (α 1 α 2 )b 21 µ 2 C 6 C b 1 (4α 1 3α 2 ) y 1 µC(2α 1 α 2 )y 12 4 2 (4.176) and therefore the origin of (4.167) and (4.168) undergoes a transcritical bifurcation as µ passes through zero when α 2 ¤ 2α 1 . When α 2 D 2α 1 , as y 1 ! 0 and µ ! 0, (4.174) tends to yP1 D (6 α 1 b 1 ) y 1 µ α 21 y 13

(4.177)

and therefore the origin of (4.167) and (4.168) undergoes a supercritical pitchfork bifurcation as µ passes through zero.

129

130

4 Bifurcations of Continuous Systems

Example 4.14 We reconsider the system (4.130)–(4.132) discussed in Example 4.12. In this case, the coefﬁcient matrix of the linear at (0, 0, 0I 0) is given by (4.142). Its eigen system p

values are λ D 0 and λ D 3/2 1 ˙ i 3 and their corresponding eigenvectors can be expressed as 2 2 p 3 p 3 2 3 1 1 1Ci 3 1i 3 1 2 2 6 6 p 7 p 7 7 7 6 1 1 p1 D 4 1 5 p2 D 6 4 2 i i C 3 5 , and p 3 D 4 2 i i C 3 5 1 1 1 We note that p 3 D pN 2 . Then, 2 p

1 12 1 C i 3 6 p

1 P D6 i iC 3 41 2 1 1 and

2

P 1

1 3 p

61 1i 3 D6 46 p

1 1Ci 3 6

p 3 1i 3 p 7 7 12 i i C 3 5 1 2

1

1

3 p

16 i i C 3 p

1 i iC 3 6

3

1 3 17 37 5 1 3

Next, we introduce the transformation x D P y into (4.130)–(4.132) and after some algebraic manipulations obtain yP1 D 13 (b 1 b 2 b 3 ) µ 13 (4α 1 α 2 4α 3 ) y 12 µ(y 2 C yN2 ) h

p p 23 α 1 C 3i 3α 1 C 2α 2 α 3 C 3i 3α 3 y 1 y 2

p p 16 13α 1 3i 3α 1 C 8α 2 13α 3 3i 3α 3 y 22 C 13 (7α 1 4α 2 7α 3 ) y 2 yN2 C cc

(4.178)

p

p p yP2 D 32 1 i 3 y 2 C 16 b 1 i 3b 1 b 2 i 3b 2 C 2b 3 µ

p p (y 1 C yN2 )µ C 16 4α 1 4 i 3α 1 α 2 i 3α 2 C 8α 3 y 12

p p p C 23 5α 1 C i 3α 1 C α 2 C i 3α 2 C α 3 3i 3α 3 y 1 y 2

p p p C 23 4α 1 2i 3α 1 C α 2 C i 3α 2 C α 3 C 3i 3α 3 y 1 yN2

p p p C 16 i 2i α 1 C 8 3α 1 C 4 i α 2 4 3α 2 C 13i α 3 3 3α 3 y 22

p p C 13 7α 1 7i 3α 1 4α 2 4 i 3α 2 C 14α 3 y 2 yN 2

p p p C 16 i 11i α 1 C 5 3α 1 C 4 i α 2 4 3α 2 C 13i α 3 C 3 3α 3 yN22 (4.179)

4.5 Normal Forms of Static Bifurcations

where y 3 D yN2 . We note that the equation governing y 3 is the complex conjugate of (4.179). We seek the center manifold of (4.178) and (4.179) in the form y 2 (tI µ) D h(y 1 I µ), where h is a complex-valued function, and rewrite (4.178) as yP 1 D 31 (b 1 b 2 b 3 ) µ 13 (4α 1 α 2 4α 3 ) y 12 µ h µ hN h

p p 23 α 1 C 3i 3α 1 C 2α 2 α 3 C 3i 3α 3 y 1 h C 13 (7α 1 4α 2 7α 3 ) h hN i

p p 16 13α 1 3i 3α 1 C 8α 2 13α 3 3i 3α 3 h 2 C cc

(4.180)

Substituting y 2 (tI µ) D h(y 1I µ) into (4.179) and using (4.180), we obtain i h h 0 13 (b 1 b 2 b 3 ) µ 13 (4α 1 α 2 4α 3 ) y 12 µ h µ hN

p

p p D 32 1 i 3 h C 16 b 1 i 3b 1 b 2 i 3b 2 C 2b 3 µ

p p C 16 4α 1 4 i 3α 1 α 2 i 3α 2 C 8α 3 y 12 C

(4.181)

An approximate solution of (4.181) can be expressed as p p

1 h D 18 2b 1 C b 2 i 3b 2 C b 3 C i 3b 3 µ

p p 1 8α 1 C α 2 i 3α 2 C 4α 3 C 4 i 3α 3 y 12 C C 18

(4.182)

Substituting (4.182) into (4.180) yields the following one-dimensional equation describing the dynamics on the center manifold: yP 1 D 31 (b 1 b 2 b 3 ) µ 13 (4α 1 α 2 4α 3 ) y 12

4 27

[b 3 (4α 1 C α 2 5α 3 )

Cb 1 (α 1 C 2α 2 α 3 ) C b 2 (5α 1 C α 2 C 4α 3 )] µ y 1 C Γ0 µ 2 2 4 4α 1 C α 22 C α 1 (13α 2 20α 3 ) C 8α 2 α 3 20α 23 y 13 27

(4.183)

where Γ0 D

1 243

b 2 (27 C 8b 3 (5α 1 C α 2 5α 3 )) C b 21 (α 1 C 16α 2 C α 3 )

2b 1 (27 C b 2 (5α 1 8α 2 C 4α 3 ) b 3 (4α 1 C 8α 2 C 5α 3 )) Cb 22 (25α 1 C 4 (α 2 C 4α 3 )) Cb 3 (27 C b 3 (16α 1 C 4α 2 C 25α 3 ))

There are two possibilities: (a) b 1 ¤ b 2 C b 3 and (b) b 1 D b 2 C b 3 . In the ﬁrst case, as y 1 ! 0 and µ ! 0, (4.183) tends to yP 1 D 31 (b 1 b 2 b 3 ) µ 13 (4α 1 α 2 4α 3 ) y 12

(4.184)

131

132

4 Bifurcations of Continuous Systems

Therefore, the origin of (4.130)–(4.132) undergoes a saddle-node bifurcation as µ passes through zero when 4α 1 ¤ α 2 C 4α 3 . When 4α 1 D α 2 C 4α 3 , the origin of (4.130)–(4.132) is not a bifurcation point. Equation 4.184 is in full agreement with (4.147) obtained with the method of multiple scales. When b 1 D b 2 C b 3 , as y 1 ! 0 and µ ! 0, (4.183) tends to yP 1 D Γ0 µ 2

4 9

[b 3 (α 1 C α 2 2α 3 ) C b 2 (2α 1 C α 2 C α 3 )] µ y 1

(4α 1 α 2 4α 3 ) y 12 1 3

(4.185)

where Γ0 D

1 b [9 C 4b 3 (α 1 C 27 2 1 b 3 [9 C b 3 (α 1 C 27

2α 2 α 3 )] C

1 2 b (4α 1 27 2

C 4α 2 C α 3 )

C 4 (α 2 C α 3 ))]

Therefore, the origin of (4.130)–(4.132) undergoes a transcritical bifurcation as µ passes through zero when 4α 1 ¤ α 2 C 4α 3 . Equation 4.185 is in full agreement with (4.152) obtained with the method of multiple scales. When 4α 1 D α 2 C 4α 3 , as y 1 ! 0 and µ ! 0, (4.183) tends to yP1 D

1 3

[b 3 (4α 3 α 2 ) 2b 2 (α 2 C 2α 3 )] µ y 1

2 3

α 22 C 4α 2 α 3 8α 23 y 13 (4.186)

Therefore, the origin of (4.130)–(4.132) undergoes a supercritical or subcritical pitchfork bifurcation as µ passes through zero, depending on whether α 22 C 4α 2 α 3 8α 23 is positive or negative, respectively. Equation 4.186 is in full agreement with (4.157) obtained with the method of multiple scales. 4.5.3 A Projection Method

In this section, we propose a method of reducing (4.82) directly without calculating the center manifold. We decompose its solution as follows: x D p u(t) C y (t)

(4.187)

where p is the right eigenvector of A, q T y(t) D 0

(4.188)

and q is the left eigenvector of A normalized so that q T p D 1. Substituting (4.187) into (4.82) yields p uP C yP D Ay C bµ C B p µ u C B y µ C Q(p , p )u 2 C 2Q(p , y)u C Q(y, y) C C (p , p , p )u3 C

(4.189)

4.5 Normal Forms of Static Bifurcations

Multiplying (4.189) from the left with q and using (4.188) leads to uP D q T bµ C q T B p µ u C q T B y µ C q T Q(p, p )u2 C 2q T Q(p , y)u C q T Q(y, y) C q T C (p , p , p )u3 C Substituting for uP from (4.190) and (4.189), we obtain yP D Ay C b µ (q T b)p µ C Q(p , p )u2 q T Q(p , p ) p u2 C

(4.190)

(4.191)

There are two possibilities: q T b ¤ 0 and q T b D 0. In the ﬁrst case, as u ! 0 and µ ! 0, (4.190) tends to uP D q T bµ C q T Q(p , p )u2

(4.192)

and therefore the origin of (4.82) undergoes a generic saddle-node bifurcation as µ passes through zero when q T Q(p , p ) ¤ 0. When q T b D 0, we express the solution of (4.191) and (4.188) as y D cµ C z u2

(4.193)

where Ac D b

and

Az D Q(p , p ) C q T Q(p, p ) p

(4.194)

and q T c D 0 and

qT z D 0

(4.195)

Substituting (4.193) into (4.190), we ﬁnd that, as u ! 0 and µ ! 0, (4.190) tends to uP D q T B c C q T Q(c, c) µ 2 C q T B p µ u C q T Q(p , p )u2 (4.196) when q T Q(p , p ) ¤ 0. Therefore, the origin of (4.82) undergoes a generic transcritical bifurcation as µ passes through zero. When q T Q(p , p ) D 0, as u ! 0 and µ ! 0, (4.190) tends to uP D q T B p µ u C 2q T Q(p , z) C q T C (p , p , p ) u3 (4.197) Therefore, the origin of (4.82) undergoes a generic supercritical or subcritical pitchfork bifurcation as µ passes through zero depending on whether [2q T Q(p , z) C q T C (p , p , p )] is negative or positive. Example 4.15 We reconsider (4.167) and (4.168). Their Jacobian matrix evaluated at (0, 0, 0) is given in (4.114) and its eigenvalues are λ D 0 and λ D 2. The right and left eigenvectors of A are 1 pD 1

and

2 qD 1

133

134

4 Bifurcations of Continuous Systems

Therefore, we express the solution of (4.167) and (4.168) as x D p u(t) C y(t) or x1 D u(t) C y 1 (t) and x2 D u(t) C y 2 (t)

(4.198)

and impose the condition q T y D 0. Substituting (4.198) into (4.167) and (4.168) yields uP C yP1 D b 1 µ C 2y 1 2y 2 C µ(u C y 1 ) C α 1 (u C y 1 )2

(4.199)

uP C yP2 D b 2 µ C 4y 1 4y 2 4(u C y 2 )µ C (u C y 1 )(u C y 2 )

(4.200)

The condition q T y D 0 yields 2y 1 y 2 D 0

(4.201)

Multiplying (4.199) and (4.200) from the left with q T and using the condition q T y D 0, we obtain uP D (2b 1 b 2 )µ C (2α 1 α 2 )u2 C 6µ u C 2µ y 1 C 4µ y 2 C (4α 1 α 2 )u y 1 α 2 u y 2 C 2α 1 y 12 α 2 y 1 y 2

(4.202)

Substituting (4.202) into (4.199) and (4.200) yields yP1 D (b 2 b 1 )µ C 2y 1 2y 2 C (α 2 α 1 )u2 C

(4.203)

yP2 D 2(b 2 b 1 )µ C 4y 1 4y 2 C 2(α 2 α 1 )u2 C

(4.204)

Because u(t) varies slowly with time, the solution of (4.204) and (4.201) can be expressed as y 1 D 12 (b 2 b 1 )µ C 12 (α 2 α 1 )u2 C

and

y 2 D 2y 1

(4.205)

It follows from (4.202) that there are two possibilities: b 2 ¤ 2b 1 and b 2 D 2b 1 . In the ﬁrst case, as u ! 0 and µ ! 0, (4.202) tends to uP D (2b 1 b 2 )µ C (2α 1 α 2 )u2

(4.206)

Therefore, the origin of (4.167) and (4.168) undergoes a saddle-node bifurcation as µ passes through zero if α 2 ¤ 2α 1 . We note that (4.206) is in full agreement with (4.118) and (4.175) obtained with the methods of multiple scales and centermanifold reduction, respectively.

4.5 Normal Forms of Static Bifurcations

When b 2 D 2b 1 , substituting (4.205) into (4.202) yields uP D 12 b 1 (10 C b 1 α 1 b 1 α 2 ) µ 2 C C (2α 1 α 2 )u 2

1 (4α 1 2

1 2

(12 C 4b 1 α 1 3b 1 α 2 ) µ u

3α 2 )(α 1 α 2 )u3 C

(4.207)

As u ! 0 and µ ! 0, (4.207) tends to uP D 12 b 1 (10 C b 1 α 1 b 1 α 2 ) µ 2 C C (2α 1 α 2 )u

1 2

(12 C 4b 1 α 1 3b 1 α 2 ) µ u

2

(4.208)

Therefore, the origin of (4.167) and (4.168) undergoes a transcritical bifurcation as µ passes through zero if α 2 ¤ 2α 1 . We note that (4.208) is in full agreement with (4.122) and (4.176) obtained with the methods of multiple scales and centermanifold reduction, respectively. When α 2 D 2α 1 , as u ! 0 and µ ! 0, (4.207) tends to uP D (6 b 1 α 1 )µ u α 21 u3

(4.209)

Therefore, the origin of (4.167) and (4.168) undergoes a supercritical pitchfork bifurcation as µ passes through zero if α 2 D 2α 1 . We note that (4.209) is in full agreement with (4.129) and (4.177) obtained with the methods of multiple scales and center-manifold reduction, respectively. Example 4.16 We reconsider the system (4.130)–(4.132) treated in Examples 4.12 and 4.14. The coefﬁcient matrix of this system evaluated p at (0, 0, 0I 0) is given by (4.142). Its eigenvalues are λ D 0 and λ D 3/2(1 ˙ i 3). Hence, the origin of (4.130)–(4.133) may be a static bifurcation point. The right and left eigenvectors of A corresponding to λ D 0 are 2 3 2 3 1 1 1 p D 4 1 5 and q D 4 1 5 3 1 1 where q is normalized so that q T p D 1. We express the solution of (4.130)–(4.132) as x D p u(t) C y(t) where

qT y D 0

or x1 D u C y 1 , y1 y2 y3 D 0

x2 D u C y 2 ,

x3 D u C y 3

(4.210) (4.211)

135

136

4 Bifurcations of Continuous Systems

Substituting (4.210) into (4.130)–(4.132), we have yP1 uP D b 1 µ y 1 C y 2 2y 3 µ u C µ y 1 C α 1 (2u C y 1 C 2y 2 C y 3 )2

(4.212)

yP2 C uP D b 2 µ 2y 1 y 2 y 3 C µ u C µ y 2 C α 2 (u C y 1 y 2 C y 3 )2

(4.213)

yP3 C uP D b 3 µ C y 1 C 2y 2 y 3 2µ u 2µ y 3 C α 3 (2u C 2y 1 C y 2 y 3 )2

(4.214)

Multiplying (4.212)–(4.213) from the left with q T yields uP D 31 (b 1 b 2 b 3 )µ C 13 (4α 3 C α 2 4α 1 )u2 (y 1 y 2 C 2y 3 )µ 23 (4α 3 C α 2 C 2α 1 )u y 1 23 (2α 3 α 2 C 4α 1 )u y 2 C 23 (2α 3 α 2 2α 1 )u y 3 C 13 (4α 3 C α 2 α 1 )y 12 C 13 (α 3 C α 2 4α 1 )y 22 C 13 (α 3 C α 2 4α 1 )y 22 C 13 (α 3 C α 2 α 1 )y 32 C 23 (2α 3 α 2 2α 1 )y 1 y 2 23 (2α 3 α 2 C α 1 )y 1 y 3 23 (α 3 C α 2 C 2α 1 )y 2 y 3

(4.215)

Substituting (4.215) into (4.212)–(4.213) yields yP 1 D 13 (4b 1 b 2 b 3 )µ y 1 C y 2 2y 3 C 23 (2α 1 C α 2 2α 3 )u2 C

(4.216)

yP 2 D 13 (b 1 C 2b 2 b 3 )µ 2y 1 y 2 y 3 C 13 (4α 1 α 2 C 8α 3 )u2 C

(4.217)

yP 3 D 13 (b 1 b 2 C 2b 3 )µ C y 1 C 2y 2 y 3 C 13 (4α 1 α 2 C 8α 3 )u2 C

(4.218)

Because u(t) is a slowly varying function of time, the particular solution of (4.216)– (4.218) orthogonal to q can be expressed as y 1 D 19 (b 1 C 2b 2 b 3 )µ C 29 (2α 1 C α 2 2α 3 )u2 C

(4.219)

y 2 D 19 (b 2 b 1 2b 3 )µ 19 (4α 1 α 2 C 8α 3 )u2 C

(4.220)

y 3 D 19 (2b 1 C b 2 C b 3 )µ C 19 (8α 1 C α 2 C 4α 3 )u2 C

(4.221)

There are two possibilities: b 1 b 2 b 3 ¤ 0 and b 1 b 2 b 3 D 0. In the ﬁrst case, as u ! 0 and µ ! 0, (4.215) tends to uP D 13 (b 1 b 2 b 3 )µ C 13 (4α 3 C α 2 4α 1 )u2

(4.222)

4.6 Normal Form of Hopf Bifurcation

Therefore, the origin of (4.130)–(4.132) undergoes a saddle-node bifurcation as µ passes through zero when 4α 1 ¤ α 2 C 4α 3 . When 4α 1 D α 2 C 4α 3 , the origin of (4.130)–(4.132) is not a bifurcation point. Equation 4.222 is in full agreement with (4.147) and (4.184) obtained with the methods of multiple scales and centermanifold reduction, respectively. When b 1 D b 2 C b 3 , substituting (4.219)–(4.221) into (4.215), we have 1 9b 2 C 9b 3 C 4b 22 α 1 4b 2 b 3 α 1 C b 23 α 1 4b 22 α 2 8b 2 b 3 α 2 uP D 27 2 4b 3 α 2 b 22 α 3 C 4b 2 b 3 α 3 4b 23 α 3 µ 2

4 27

(b 1 α 1 C 5b 2 α 1 4b 3 α 1 C 2b 1 α 2 C b 2 α 2

Cb 2 α 2 C b 3 α 2 b 1 α 3 C 4b 2 α 3 5b 3 α 3 ) µ u C 13 (4α 3 C α 2 4α 1 )u2 2 4 (4.223) 4α 1 C 13α 1 α 2 C α 22 20α 1 α 3 C 8α 2 α 3 20α 23 u3 27 When 4α 1 ¤ α 2 C 4α 3 , as u ! 0 and µ ! 0, (4.223) tends to 1 uP D 27 9b 2 C 9b 3 C 4b 22 α 1 4b 2 b 3 α 1 C b 23 α 1 4b 22 α 2 8b 2 b 3 α 2 4b 23 α 2 b 22 α 3 C 4b 2 b 3 α 3 4b 23 α 3 µ 2

4 27

(b 1 α 1 C 5b 2 α 1 4b 3 α 1 C 2b 1 α 2 C b 2 α 2

Cb 2 α 2 C b 3 α 2 b 1 α 3 C 4b 2 α 3 5b 3 α 3 ) µ u C 13 (4α 3 C α 2 4α 1 )u2

(4.224)

Therefore, the origin of (4.130)–(4.132) undergoes a transcritical bifurcation as µ passes through zero when 4α 1 ¤ α 2 C 4α 3 . Equation 4.224 is in full agreement with (4.152) and (4.185) obtained with the methods of multiple scales and centermanifold reduction, respectively. When α 1 D α 3 C 1/4α 2 , as u ! 0 and µ ! 0, (4.223) tends to uP D

1 3

[b 3 (4α 3 α 2 ) 2b 2 (α 2 C 2α 3 )] µ u

2 3

α 22 C 4α 2 α 3 8α 23 u3 (4.225)

Therefore, the origin of (4.130)–(4.132) undergoes a supercritical or subcritical pitchfork bifurcation as µ passes through zero, depending on whether α 22 C 4α 2 α 3 8α 23 is positive or negative, respectively. Equation 4.225 is in full agreement with (4.157) and (4.186) obtained with the methods of multiple scales and center-manifold reduction, respectively.

4.6 Normal Form of Hopf Bifurcation

In this section, we describe methods for simplifying the dynamical system (4.17) near a Hopf bifurcation point. We assume that this point has been shifted to x D 0 and that the control parameter has been shifted to µ D 0. Moreover, we expand

137

138

4 Bifurcations of Continuous Systems

(4.17) for small x and µ and obtain (4.82). Furthermore, because A is nonsingular, we can shift the ﬁxed point by A1 bµ and hence rewrite (4.82) as xP D Ax C µ B x C Q(x, x) C C (x, x, x) C

(4.226)

Because the origin is a Hopf bifurcation, two of the eigenvalues are purely imaginary complex conjugates (i.e., ˙i ω) and the remaining eigenvalues are in the lefthalf of the complex plane. Next, we demonstrate how one can use the methods of multiple scales, projection method, and center-manifold reduction to compute the normal form of (4.226) near x D 0 and small µ. 4.6.1 The Method of Multiple Scales

To compute the normal form of the Hopf bifurcation of (4.226) at the origin, we introduce a small nondimensional parameter as a bookkeeping parameter and seek a third-order approximate solution of (4.226) in the form x(tI µ) D x 1 (T0 , T2 ) C 2 x 2 (T0 , T2 ) C 3 x 3 (T0 , T2 ) C

(4.227)

where the time scales Tm D m t. The time derivative can be expressed in terms of these scales as d D D0 C 2 D2 C dt

(4.228)

where D m D @/@Tm . As shown below, there is no dependence on T1 because the second-order problem is solvable. Moreover, because µ is small, we scale it as 2 µ. Substituting (4.227) and (4.228) into (4.226) and equating coefﬁcients of like powers of , we obtain Order ()

D0 x 1 Ax 1 D 0

(4.229)

Order (2 )

D0 x 2 Ax 2 D Q(x 1 , x 1 )

(4.230)

Order (3 )

D0 x 3 Ax 3 D D2 x 1 C µ B x 1 C 2Q(x 1 , x 2 ) C C (x 1 , x 1 , x 1 )

(4.231)

The general solution of (4.229) is the superposition of n linearly independent solutions corresponding to the n eigenvalues of A: n 2 of these eigenvalues have negative real parts and two are purely imaginary. The n 2 solutions corresponding

4.6 Normal Form of Hopf Bifurcation

to the eigenvalues with negative real parts decay with time and hence the longtime dynamics of the system depends on the center space corresponding to the two purely imaginary eigenvalues. Therefore, we express the solution of (4.229) as x 1 D z(T2 )p e i ω T0 C z(T N 2 ) pN e i ω T0

(4.232)

where the function z(T2 ) is determined by imposing the solvability condition at third order and p is the eigenvector of A corresponding to the eigenvalue i ω; that is, Ap D i ω p

(4.233)

Substituting (4.232) into (4.230) yields D0 x 2 Ax 2 D Q(p , p )z 2 e 2i ω T0 C 2Q(p , pN )z zN C Q( pN , pN ) zN 2 e 2i ω T0 (4.234) The solution of (4.234) can be expressed as x 2 D ζ 2 z 2 e 2i ω T0 C 2ζ 0 z zN C ζN 2 zN 2 e 2i ω T0

(4.235)

where [2i ω I A] ζ 2 D Q(p , p )

and

Aζ 0 D Q(p, pN )

(4.236)

Substituting (4.232) and (4.235) into (4.231), we have D0 x 3 Ax 3 D D2 z p µ B p z 4Q(p , ζ 0 )z 2 zN 2Q( pN , ζ 2 )z 2 zN 3C (p , p , pN )z 2 zN e i ω T0 C cc C NST . (4.237) Because p e i ω T0 is a solution of the homogeneous equation 4.237, the nonhomogeneous equation has a solution only if a solvability condition is satisﬁed. We let q be the left eigenvector of A corresponding to the eigenvalue i ω; that is, AT q D i ωq

(4.238)

We normalize it so that q T p D 1. Then, the solvability condition demands that the term proportional to e i ω T0 in (4.237) is orthogonal to q. Imposing this condition, we obtain the following normal form of the Hopf bifurcation: D2 z D q T B p µ z C 4q T Q(p , ζ 0 ) C 2q T Q( pN , ζ 2 ) C 3q T C(p , p , pN ) z 2 zN (4.239)

139

140

4 Bifurcations of Continuous Systems

Example 4.17 We consider the system xP1 D µ x1 x2 C x12 α 1 x1 x3

(4.240)

xP2 D µ x2 C x1 C α 2 x2 x3

(4.241)

xP3 D x3 C x12 C 2x22

(4.242)

Substituting (4.227) and (4.228) into (4.240)–(4.242), scaling µ as 2 µ, and equating coefﬁcients of like powers of on both sides, we obtain Order () D0 x11 C x21 D 0

(4.243)

D0 x21 x11 D 0

(4.244)

D0 x31 C x31 D 0

(4.245)

Order ( 2 ) 2 α 1 x11 x31 D0 x12 C x22 D x11

(4.246)

D0 x22 x12 D α 2 x21 x31

(4.247)

2 2 D0 x32 C x32 D x11 C 2x21

(4.248)

Order ( 3 ) D0 x13 C x23 D D2 x11 C µ x11 C 2x11 x12 α 1 x11 x32 α 1 x12 x31

(4.249)

D0 x23 x13 D D2 x21 C µ x21 C α 2 x21 x32 C α 2 x22 x31

(4.250)

D0 x33 C x33 D D2 x31 C 2x11 x12 C 4x21 x22

(4.251)

The nondecaying solutions of (4.243)–(4.245) can be expressed as x11 D i z(T2 )e i T0 i zN (T2 )e i T0 , N 2 )e i T0 , x21 D z(T2 )e i T0 C z(T x31 D 0

(4.252)

4.6 Normal Form of Hopf Bifurcation

Substituting (4.252) into (4.246)–(4.248) yields D0 x12 C x22 D z 2 e 2i T0 2z zN C zN 2 e 2i T0

(4.253)

D0 x22 x12 D 0

(4.254)

D0 x32 C x32 D z 2 e 2i T0 C 6z zN C zN 2 e 2i T0

(4.255)

whose solutions can be expressed as x12 D 32 i z 2 e 2i T0 C 23 i zN 2 e 2i T0

(4.256)

x22 D 31 z 2 e 2i T0 2z zN 13 zN 2 e 2i T0

(4.257)

x32 D 15 (1 2i)z 2 e 2i T0 C 6z zN C 15 (1 C 2i) zN 2 e 2i T0

(4.258)

Substituting (4.252) and (4.256)–(4.258) into (4.249) and (4.250), we have D0 x13 C x23 D i D2 z C i µ z C 43 C 25 α 1 (2 29i) z 2 zN e i T0 C cc C NST

(4.259)

D0 x23 x13 D D2 z C µ z C 15 α 2 (31 2i)z 2 zN e i T0 C cc C NST (4.260) The solvability condition of (4.259) and (4.260) demands that their right-hand sides be orthogonal to q, where q T D 1/2(i, 1). Imposing this condition, we obtain the following normal form of the system (4.240)–(4.242) near its origin: D2 z D µ z C

1 30

93α 2 87α 1 (20 C 6α 1 C 6α 2 )i z 2 zN

(4.261)

Letting z D 1/2e i θ in (4.261) and separating real and imaginary parts, we obtain the following alternate normal form of the Hopf bifurcation: D2 r D 12 µ r C

1 (31α 2 40

29α 1 )r 3

1 D2 θ D 60 (10 C 3α 1 C 3α 2 )r 2

(4.262) (4.263)

4.6.2 Center-Manifold Reduction

To determine the normal form of the Hopf bifurcation of the system (4.226) at the origin, we ﬁrst calculate the eigenvalues λ 1 , λ 2 , . . . , λ n and eigenvectors (generalized eigenvectors) p 1 , p 2 , . . . , p n of the matrix A. Second, we arrange the eigenvalues of A so that λ 1 D i ω and λ 2 D i ω and let p 1 and pN 1 be their corresponding eigenvectors. Third, we introduce the transformation x D P y and obtain yP D J y C P 1 B P y µ C P 1 Q(P y, P y) C P 1 C (P y, P y, P y)

(4.264)

141

142

4 Bifurcations of Continuous Systems

where J D P 1 AP . We note that J can be rewritten as Jc 0 iω 0 JD and J c D 0 Js 0 i ω

(4.265)

and J s is an (n 2) (n 2) matrix whose eigenvalues are λ 3 , λ 4 , . . . , λ n . Fourth, we let y c be the two-dimensional vector with the components y 1 and y 2 and let y s be the (n 2)-dimensional vector with the components y 3 , y 4 , . . . , y n and rewrite (4.264) as yP c D J c y c C µ B1c y c C µ B1s y s C G 2 (y c , y s ) C G 3 (y c , y s )

(4.266)

yP s D J s y s C µ B2c y c C µ B2s y s C F 2 (y c , y s ) C F 3 (y c , y s )

(4.267)

and

We note that y c and y s are linearly uncoupled but nonlinearly coupled. Further, F j (0, 0) D 0, G j (0, 0) D 0, and the Jacobian matrices D F j (0, 0) and D G j (0, 0) are matrices with zero entries. Because F j and G j are polynomials and hence inﬁnitely differentiable, there exists a local center manifold of the form y s D h(y c ) where h is a polynomial function of y c and h(0) D 0

and

D y c h(0) D 0

(4.268)

Fifth, we determine the (n2)-dimensional function h by constraining the center manifold to be two-dimensional in the n-dimensional space. Substituting for y s into (4.266) yields yP c D J c y c C µ B1c y c C µ B1s h(y c ) C G 2 y c , h(y c ) C G 3 y c , h(y c ) (4.269) Substituting for y s into (4.267) and using (4.268), we have ˚ D y c h(y c ) J c y c C G 2 y c , h(y c ) C G 3 y c , h(y c ) D J s h(y c ) C F 2 [y c , h(y c )] C F 3 [y c , h(y c )] C

(4.270)

To solve (4.270), one approximates the components of h(y c ) with polynomials. The polynomial approximations are usually taken to be quadratic to the ﬁrst approximation and do not contain constant and linear terms so that the conditions on h are satisﬁed. Substituting the assumed quadratic polynomial approximations into (4.270) and equating the coefﬁcients of the different powers in the polynomials on both sides, one obtains a system of algebraic equations for the coefﬁcients of the polynomials. Solving these equations, we obtain a ﬁrst approximation to the center manifold y s D h(y c ). Finally, substituting this approximation into (4.269), we obtain a two-dimensional dynamical system describing the dynamics on the center manifold.

4.6 Normal Form of Hopf Bifurcation

Example 4.18 We use a combination of center-manifold reduction and the method of normal forms to compute the normal form of the Hopf bifurcation near the origin of (4.240)–(4.242). In this case, the local manifold is one-dimensional and tangent to the x1 x2 plane at the origin. We approximate this manifold by a quadratic function in the form x3 (x1 , x2 ) D h(x1 , x2 ) D a 1 x12 C a 2 x22 C a 3 x1 x2 C

(4.271)

Substituting (4.271) into (4.242) yields 2a 1 x1 xP 1 C 2a 2 x2 xP 2 C a 3 x1 xP 2 C a 3 x2 xP 1 D a 1 x12 C a 2 x22 C a 3 x1 x2 C x12 C 2x22 C

(4.272)

Substituting (4.240) and (4.241) into (4.272) and using (4.271), we have 2(a 2 a 1 )x1 x2 C a 3 x12 a 3 x22 D a 1 x12 C a 2 x22 C a 3 x1 x2 C x12 C2x22 C (4.273) Equating the coefﬁcients of each of x12 , x22 , and x1 x2 on both sides of (4.273), we obtain a3 C a1 D 1 ,

a2 a3 D 2 ,

a 3 C 2a 2 2a 1 D 0

whose solution is a1 D

7 5

,

a2 D

8 5

,

a 3 D 25

(4.274)

Substituting (4.274) into (4.271) and then substituting the result into (4.240) and (4.241), we obtain the following two-dimensional system describing the dynamics of (4.240)–(4.242) on the center manifold: xP 1 D µ x1 x2 C x12 75 x13 α 1 C 25 x12 x2 α 1 85 x1 x22 α 1

(4.275)

xP 2 D µ x2 C x1 C 75 x12 x2 α 2 25 x1 x22 α 2 C 85 x23 α 2

(4.276)

Next, we use the method of normal forms to simplify (4.275) and (4.276). First, we calculate the Jacobian matrix of (4.275) and (4.276) evaluated at the origin. The result is 0 1 1 0

143

144

4 Bifurcations of Continuous Systems

Its eigenvalues are λ D ˙i. The right and left eigenvectors of this matrix corresponding to λ D i are pD

i 1

qD

and

1 2

i 1

Second, we introduce the transformation x1 D p u(t) C pN u(t) N x2 and obtain x1 (t) D i u(t) i u(t) N

and

x2 (t) D u(t) C u(t) N

(4.277)

Substituting (4.277) into (4.275) and (4.276) and multiplying the result from the left with q yields (31 2i)α 2 (29 C 2i)α 1 u2 uN (4.278) C nonresonance cubic and higher-order terms

uP D i u C µ u C 12 i(u2 2u uN C uN 2 ) C

1 10

Third, we introduce a near-identity transformation of the form u(t) D v (t) C b 1 v 2 (t) C b 2 v (t) vN (t) C b 3 vN 2

(4.279)

into (4.278), approximate vP and vPN with i v and i vN , respectively, and obtain vP D i v C µ v C 12 i(1 2b 1 )v 2 i(1 b 2 )v vN C 12 i(1 C 6b 3 ) vN 2 1 (31 2i)α 2 (29 C 2i)α 1 C 10i(b 3 b 1 ) v 2 vN C 10

(4.280)

Fourth, we choose the b i to eliminate the quadratic terms and obtain b1 D

1 2

,

b2 D 1 ,

b 3 D 61

(4.281)

Finally, we substitute for the b i in (4.280) and obtain the normal form vP D i v C µ v C

1 30

93α 2 87α 1 (20 C 6α 1 C 6α 2 )i v 2 vN

(4.282)

which is in full agreement with (4.261) obtained with the method of multiple scales. 4.6.3 Projection Method

To determine the normal form of the Hopf bifurcation at the origin of the system (4.226) as µ increases past zero with the projection method, we assume that x(t) D p u(t) C pN u(t) N C y (t)

(4.283)

4.6 Normal Form of Hopf Bifurcation

where p and pN are the right eigenvectors of A corresponding to the eigenvalues ˙i ω. We assume that all of the other eigenvalues of A are in the left-half of the complex plane. Moreover, we constrain y(t) to be orthogonal to the left eigenvectors q and qN corresponding to the eigenvalues ˙i ω normalized so that q T p D 1 and qN T pN D 1; that is, we require qT y D 0

and

qN T y D 0

(4.284)

Substituting (4.283) into (4.226) yields p uP C pN uPN C yP D i ω p u i ω pN uN C Ay C µ uB p C Q(p , p )u2 C 2Q(p , pN )u uN C Q( pN , pN ) uN 2 C 3C (p, p , pN )u2 uN C 2Q(p , y)u C Q( pN , y) uN C

(4.285)

where the three dots stand for nonresonance and higher-order terms. Multiplying (4.285) from the left with q T , we have uP D i ω p u C µ uB p C q T Q(p, p )u2 C 2q T Q(p , pN )u uN C q T Q( pN , pN ) uN 2 (4.286) C 3q T C (p , p , pN )u2 uN C 2q T Q(p , y)u C q T Q( pN , y) uN C Substituting (4.286) and its complex conjugate into (4.285), we obtain yP D Ay C Q(p , p ) q T Q(p , p )p qN T Q(p, p ) pN u2 C 2 Q(p , pN ) q T Q(p , pN )p qN T Q(p , pN ) pN u uN C Q( pN , pN ) q T Q( pN , pN )p qN T Q( pN , pN ) pN uN 2 C A particular solution of (4.287) can be expressed as i i T qN Q(p , p ) pN u2 y D C η 2 C q T Q(p , p )p C ω 3ω i T i T C 2 η 0 q Q(p , pN )p C qN Q(p , pN ) pN u uN ω ω i T i q Q( pN , pN )p qN T Q( pN , pN ) pN uN 2 C C ηN 2 3ω ω

(4.287)

(4.288)

Next, we substitute (4.288) into (4.284) and determine constraint conditions on η 2 and η 0 . Finally, we substitute (4.288) into (4.286) and introduce a near-identity transformation to eliminate the quadratic terms, thereby obtaining the normal form of the bifurcation. We note that, out of the three methods, the method of multiple scales seems to be the best for simplifying high-dimensional nonlinear systems near a Hopf bifurcation.

145

146

4 Bifurcations of Continuous Systems

Example 4.19 We use the projection method to compute the normal form of the Hopf bifurcation at the origin of the system (4.240)–(4.242) as µ passes through zero. In this case, 0 1 i p D @1 A 0

and

0 1 i [email protected] A qD 1 2 0

(4.289)

and (4.283) and (4.284) lead to x1 (t) D i u(t) i u(t) N C y 1 (t) , N C y 2 (t) , x2 (t) D u(t) C u(t)

and

x3 (t) D y 3 (t) y 1 (t) D 0 and

(4.290) y 2 (t) D 0

(4.291)

Substituting (4.290) and (4.291) into (4.240)–(4.242) yields i uP i uNP D u uN C i µ(u uN ) u2 C 2u uN uN 2 i α 1 u y 3 C i α 1 uN y 3 (4.292) uP C uNP D i uN C µ(u u) N C α 2 u y 3 C α 2 uy N 3

(4.293)

yP3 D y 3 C u2 C 6u uN C uN 2

(4.294)

Multiplying (4.292)–(4.294) from the left with q, we have uP D i u C µ u C 12 i u2 i u uN C 12 i u2 12 u y 3 α 1 C 12 uN y 3 α 1 C 12 u y 3 α 2 C 12 uN y 3 α 2

(4.295)

Substituting (4.295) and its complex conjugate into (4.292)–(4.294), we ﬁnd that the ﬁrst two equations reduce to zero. Then, solving (4.294) yields y 3 D 15 (1 2i)u2 C 6u uN C 15 (1 C 2i) uN 2

(4.296)

Substituting (4.296) into (4.295), we obtain (4.282) obtained with center-manifold reduction.

4.7 Exercises

4.7.1 Consider the following one-dimensional systems: a) xP D µ x C x 2 b) xP D µ C x 2

4.7 Exercises

c) xP D µ x C x 3 d) xP D µ x x 3 e) xP D µ x 3 . In each case, x is the state variable and µ is the control parameter. Construct the bifurcation diagrams for all cases and discuss them. 4.7.2 Consider the following one-dimensional systems: a) b) c) d)

xP xP xP xP

D 2µ 2 C µ x x 2 D xP D 2µ 2 C µ x x 3 D µ C µx x3 D µ C µx x2 C x3 .

What type of bifurcation does each of these systems undergo with increasing µ at µ D 0? Determine the bifurcating solutions. 4.7.3 Determine the ﬁxed points of xP D x 2 4x C 3 Determine their stability. 4.7.4 Find the equilibrium solutions of xP D x 2ax 2 C x 3 Determine their stability and the bifurcations, which they undergo as a is varied. 4.7.5 Sketch the bifurcation diagram of xP D x 3 C a 3 3ax in the (a, x) plane, indicating which equilibrium solutions are stable and identifying a turning point and a pitchfork bifurcation. 4.7.6 Sketch the bifurcation diagram of xP D 2(a 2 x 2 ) (a 2 C x 2 )2 in the (a, x) plane, indicating which equilibrium solutions are stable and identifying two turning points and a transcritical bifurcation. 4.7.7 Sketch the bifurcation diagram of xP D x (x 2 2b x a) in the (a, x) plane for a given positive b, indicating which solutions are stable. 4.7.8 Consider the one-dimensional system xP D ax C b x 3 C c x 5 Determine the ﬁxed points and their stability.

147

148

4 Bifurcations of Continuous Systems

4.7.9 Consider the system xP D x 4 C ax 2 C b x C c Sketch the intersection of the bifurcation set and three planes a D constant for a < 0, a D 0, and a > 0 (Thom, 1975). 4.7.10 Consider the one-dimensional system (Drazin, 1992) xP D x 3 2ax 2 (b 3)x C c for real a, b, and c. a) Show that, if c D 0, then there is a transcritical bifurcation, but if c ¤ 0, there are two (nonbifurcating) branches of equilibria. b) Show that the loci of the bifurcation points is given by the curve (27c 18b C 38)2 D 4(3b 5)3 which has a cusp at b D 3/5 and c D 8/27. 4.7.11 Determine the ﬁxed points and their types for the following systems, and for each case sketch the trajectories and the separatrices in the phase plane: a) b) c) d) e)

xR xR xR xR xR

C 2 xP C x C x 3 D 0 C 2 xP C x x 3 D 0 C 2 xP x C x 3 D 0 C 2 xP x x 3 D 0 a C x 2 D 0 for a > 0, a D 0, and a < 0 .

4.7.12 Consider the following system (Drazin, 1992): xP D x 3 C δ x 2 µ x Determine the ﬁxed points of this system and study the bifurcations in the (x, µ) plane for zero and nonzero values of δ. Show that the pitchfork bifurcation at (0, 0) for δ D 0 becomes a transcritical bifurcation for small δ and that there is a turning point at (1/2δ, 1/4δ 2 ). Sketch the bifurcation diagram in the (x, µ) plane for δ > 0. 4.7.13 Consider the following single-degree-of-freedom system with quadratic and cubic nonlinearities: xR C x C δ x 2 C α x 3 D 0 Sketch the potential energy V(x) for the system and the associated phase portrait for each of the following cases: (a) δ D 3 and α D 4, (b) δ D α D 4, and (c) δ D 5 and α D 4.

4.7 Exercises

It is common to refer to the ﬁrst case as a single-well potential system because there is a well in the graph of V(x) versus x. The third case is referred to as a twowell potential system. From the phase portraits, one can discern a qualitative change as one goes from the ﬁrst case to the third case. 4.7.14 Consider the system xP D x (3 x 2y ) yP D y (2 x y ) where x > 0 and y > 0. Determine all of the ﬁxed points of this system and determine their types. 4.7.15 Consider the system xP D x 2 µ 2 yP D y C x 2 µ 2 Determine the equilibrium solutions and their stability. 4.7.16 Consider the system xP 1 D x2 13 λ(x13 3x1 ) xP 2 D x1 Show that the origin is the only equilibrium point. Determine its stability as a function of λ. 4.7.17 Show that the origin is an unstable equilibrium point for each of the two systems a) xP 1 D x2 , xP 2 D x1 C 2x23 . b) xP 1 D x1 C 5x2 C x12 x2 , xP 2 D 5x1 C x2 x23 . 4.7.18 Show that the origin is a saddle point for each of the two systems a) xP 1 D x2 , xP 2 D x2 C x13 . b) xP 1 D x2 , xP 2 D x1 C x13 . Determine the stable and unstable manifolds of the origin for the linearized as well as the nonlinear systems. 4.7.19 Determine an approximation to the stable and unstable manifolds of the saddles of the system xP 1 D 1 x1 x2 xP 2 D x1 x23

149

150

4 Bifurcations of Continuous Systems

4.7.20 Consider the system xP 1 D λ C 2x1 x2 xP 2 D 1 C x12 x22 Show that there are two saddles when λ D 0 and that the x2 -axis is invariant. Hence, show that there is a heteroclinic connection. Sketch the phase plane. Show that, when jλj ¤ 0 and is small, there are still two saddle points but the saddle connection is no longer present. Sketch the phase plane for 1 λ > 0 and 1 λ < 0. 4.7.21 Show that the trivial solution is the only equilibrium solution of xP D x y 2 C x 2 y C x 3

yP D y 3 x 3

and

and that it is unstable. 4.7.22 Show that the trivial solution of the system xP D 2x y 2 x 3 and yP D

2 2 x y y3 5

is asymptotically stable. 4.7.23 Show that the origin is a stable equilibrium point of the system xP D y x 3

yP D x 2

and

4.7.24 Show that the origin is an asymptotically stable equilibrium point of the system xP D y x (x 4 C y 4 )

and

yP D x y (x 4 C y 4 )

4.7.25 The origin of each of the following systems is a degenerate saddle point: a) xP D x 2 , yP D y b) xP D x 2 y 2 , yP D 2x y . Sketch the phase portrait for each case. 4.7.26 Show that the origin is an unstable equilibrium point of the system xP D 2x 2 y

and

yP D 2x y 2

Hint: Show that x y is constant on each orbit. 4.7.27 Consider the system xP D x y

and

yP D 2 x y

Find the ﬁxed points and determine their stability.

4.7 Exercises

4.7.28 Consider the planar system xP D 12 (x C x 3 ) 2y C yP D 1 2x 2 where 1. Show that there are three saddle points. For D 0 and ¤ 0, sketch the phase portrait and indicate any heteroclinic connections. 4.7.29 Consider the system xR C ω 2 x C δ x 2 cos(Ω t) D 0 where is a small nondimensional parameter. Use the method of multiple scales to determine the modulation equations when Ω 3ω. 4.7.30 Consider the system xR C ω 2 x C δ x 2 cos(Ω t) D 0 where is a small nondimensional parameter. Use the method of averaging to determine the modulation equations when Ω 3ω. 4.7.31 Consider the following speed-control system investigated by Fallside and Patel (1965): xP1 D x2

x2 xP 2 D K d x2 x1 G x12 C x1 C 1 Kd a) For K d D 1 and G D 6, determine the ﬁxed points and their stability. b) Plot the stable manifolds of the unstable ﬁxed points and a few other trajectories in the (x1 , x2 ) space. c) Discuss the phase portrait. 4.7.32 Calculate the ﬁxed points of the system xP D α x y

and

yP D y C x 2

Determine their stability. 4.7.33 The free oscillation about the upright position of an inverted pendulum constrained to oscillate between two closely spaced rigid barriers is described by (e.g., Shaw and Rand, 1989) xR C 2µ xP x D 0 jxj < 1 xP ! r xP

jxj D 1

151

152

4 Bifurcations of Continuous Systems

where x describes the position of the pendulum; the locations of the rigid barriers are x D 1 and x D 1; µ is a measure of the friction; and r 1 is a reﬂection coefﬁcient representing energy loss during impact with either of the rigid barriers. Assuming elastic impact (i.e., r D 1), construct and discuss the phase portraits for the following two cases: (a) µ D 0 and (b) µ > 0. 4.7.34 In studying the forced response of a van der Pol oscillator with delayed amplitude limiting, Nayfeh (1968) encountered the following system of equations: xP 1 D x1 (1 x12 ) C F cos x2 xP 2 D σ C νx12

F sin x2 x1

a) Show that the ﬁxed points (x10 , x20 ) of this system satisfy (1 )2 C (σ C ν)2 D F 2 2 where D x10 . b) For ν D 0.15, plot the loci of the ﬁxed points in the σ plane for F 2 D 1, 1/3, 4/27, and 1/10. What is the signiﬁcance of the value 4/27? c) Show that the interior points of the ellipse deﬁned by

(1 )(1 3) C (σ C ν)(σ C 3ν) D 0 are saddle points and hence unstable. Also, show that the exterior points are nodes if D 0 and foci if D < 0, where D D 4 (1 3ν 2 )2 4νσ σ 2 . d) Finally, show that the exterior points are stable if > 1/2 and unstable if < 1/2. 4.7.35 A bead of mass m sliding on a rotating circular hoop of radius R is described by g θR C 2µ θP C sin θ ω 2 sin θ cos θ D 0 R Here, θ describes the angular position of the bead on the hoop, g is the acceleration due to gravity, µ is a measure of the friction experienced by the bead, and ω is the angular velocity of the hoop. a) For µ D 0, determine the ﬁxed points (equilibrium positions) of the system and sketch the phase portrait in each of the following cases: (i) ω 2 < g/R, (ii) ω 2 D g/R, and (iii) ω 2 > g/R. b) For µ > 0, choose ω as a control parameter and examine the different local bifurcations of the ﬁxed points that occur as ω is increased from zero. Construct appropriate bifurcation diagrams.

4.7 Exercises

4.7.36 Mingori and Harrison (1974) studied the following system for analyzing the motion of a particle constrained to move on a circular path that is spinning and coning: uP D v vP D µ 1 (v 1) C µ 2 µ 3 sin u C 12 µ 23 sin 2u Let µ 1 D 0.1 and µ 2 D 2.0. Then, as µ 3 is varied from zero, bifurcations take place at 0.0502, 0.3, and 2.265. Examine the qualitative changes that take place in the (u, v ) space due to these bifurcations. 4.7.37 Consider the nonlinear oscillator uR C u C 2µ 1 uP C µ 2 uj P uj P C α u3 C 2K u cos(Ω t) D 0 where is a small, positive parameter. Further, the parameters µ 1 , µ 2 , and K are all independent of while the parameter Ω is such that Ω D 2 C σ A ﬁrst approximation obtained for this system has the form u D p cos 12 Ω t C q sin 12 Ω t C O() where

q 1 4µ 2 3α p 0 D µ 1 p (σ C K )q C q(p 2 C q 2 ) p p 2 C q2 2 8 3π q 1 4µ 2 3α 0 2 2 q D µ 1 q C (σ K )p p (p C q ) q p 2 C q2 2 8 3π

In the above equations, the prime denotes the derivative with respect to the time scale τ D t. a) Simplify the dynamical system governing p and q to its normal form for a transcritical bifurcation in the vicinity of q (p, q, K c ) D 0, 0, 4µ 21 C σ 2 . b) What happens to the bifurcation at the above-mentioned bifurcation point when µ 2 D 0? c) Construct the frequency-response curves when µ 2 D 0 and discuss them. 4.7.38 Consider the system x xP D y C p (1 x 2 y 2 ) x2 C y2 y (1 x 2 y 2 ) yP D x C p 2 x C y2 Determine its limit cycle and indicate whether it is stable or unstable.

153

154

4 Bifurcations of Continuous Systems

4.7.39 Consider the system xP D µ x C y C x y yP D 2µ C x y y 2 where µ is the control parameter. Determine the normal form of this system in the vicinity of the bifurcation point (x, y, µ) D (0, 0, 0). What is the type of this bifurcation? 4.7.40 Consider the system xP 1 D µ C x1 x2 C 2µ x2 C x12 xP 2 D b µ C 2x1 2x2 C µ x1 C α x22 where µ is the control parameter. Determine the normal form of this system in the vicinity of the bifurcation point (x1 , x2 , µ) D (0, 0, 0). a) Show that this system undergoes a generic saddle-node bifurcation as µ increases past zero when b ¤ 2 and α ¤ 2. b) Is there a bifurcation at (x1 , x2 , µ) D (0, 0, 0) when b ¤ 2 and α D 2? c) Show that this system undergoes a generic transcritical bifurcation as µ increases past zero when b D 2 and α ¤ 2. d) Show that this system undergoes a generic pitchfork bifurcation as µ increases past zero when b D 2 and α D 2. 4.7.41 Consider the system xP 1 D x1 2x2 C µ b C α 1 (x1 C 2x2 )2 xP 2 D 2x1 4x2 C µ b C α 2 (x1 x2 )2 Determine the type and normal form of the bifurcation that takes place at (x1 , x2 , µ) D (0, 0, 0) 4.7.42 Consider the system xP 1 D x1 2x2 C µ b C α(x1 C 2x2 )2 xP 2 D 2x1 4x2 C 2µ b C 29α(x1 x2 )2 Determine the type and normal form of the bifurcation that takes place at (x1 , x2 , µ) D (0, 0, 0) 4.7.43 Consider the system xP 1 D x1 2x2 C µ b C µ x1 C α(x1 C 2x2 )2 xP 2 D 2x1 4x2 C 2µ b C µ x2 C 32α(x1 x2 )2

4.7 Exercises

Determine the type and normal form of the bifurcation that takes place at (x1 , x2 , µ) D (0, 0, 0) 4.7.44 Consider the system xP D y C µ x C α x (x 2 C y 2 ) yP D x C µ y C α y (x 2 C y 2 ) What type of bifurcation occurs at (x, y, µ) D (0, 0, 0). Determine the bifurcating solutions. 4.7.45 Consider the system xP1 D µ C x1 x2 C 2µ x2 C x12 xP 2 D 2µ C 2x1 2x2 C µ x1 C x22 where µ is the control parameter. Determine the normal form of this system in the vicinity of the bifurcation point (x1 , x2 , µ) D (0, 0, 0). What is the type of this bifurcation? 4.7.46 Consider the system xP 1 D µ C x1 x2 C 2µ x2 C x12 xP 2 D µ C 2x1 2x2 C µ x1 C x22 where µ is the control parameter. Determine the normal form of this system in the vicinity of the bifurcation point (x1 , x2 , µ) D (0, 0, 0). What is the type of this bifurcation? 4.7.47 Consider the system xP 1 D µ C x1 x2 C 2µ x2 C x12 xP 2 D µ C 2x1 2x2 C µ x1 C 2x22 where µ is the control parameter. Determine the normal form of this system in the vicinity of the bifurcation point (x1 , x2 , µ) D (0, 0, 0). What is the type of this bifurcation? 4.7.48 Consider the system xP 1 D µ C x1 x2 C 2µ x2 C x12 xP 2 D b µ C 2x1 2x2 C µ x1 C α x22 where µ is the control parameter. Determine the normal form of this system in the vicinity of the bifurcation point (x1 , x2 , µ) D (0, 0, 0) for each of the following cases:

155

156

4 Bifurcations of Continuous Systems

a) b) c) d)

b b b b

¤ 2 and ¤ 2 and D 2 and D 2 and

α α α α

¤2 D2 ¤2 D2.

4.7.49 Consider the system xP 1 D x1 2x2 C µ b C α(x1 C 2x2 )2 xP 2 D 2x1 4x2 C 2µ b C 29α(x1 x2 )2 Determine the type and normal form of the bifurcation that takes place at (x1 , x2 , µ) D (0, 0, 0) 4.7.50 Consider the system xP 1 D x1 2x2 C µ b C µ x1 C α(x1 C 2x2 )2 xP 2 D 2x1 4x2 C 2µ b C µ x2 C 32α(x1 x2 )2 Determine the type and normal form of the bifurcation that takes place at (x1 , x2 , µ) D (0, 0, 0) 4.7.51 Consider the system xP 1 D x1 2x2 C µ b C α 1 (x1 C 2x2 )2 xP 2 D 2x1 4x2 C µ b C α 2 (x1 x2 )2 Determine the type and normal form of the bifurcation that takes place at (x1 , x2 , µ) D (0, 0, 0) 4.7.52 Consider the dynamical system xP D 32 µ C x 2y C x 2 yP D 2x 4y C x y Examine the bifurcation that takes place when (x, y, µ) D (0, 0, 0) and determine the normal form of the system near this bifurcation. 4.7.53 Consider the system xP 1 D µ C x1 x2 C 2µ x2 C x12 xP 2 D 2µ C 2x1 2x2 C µ x1 C 2x22

4.7 Exercises

where µ is the control parameter. Determine the normal form of this system in the vicinity of the bifurcation point (x1 , x2 , µ) D (0, 0, 0). What is the type of this bifurcation? 4.7.54 Consider the system xP D µ x y yP D x C µ y C α x 3 Examine the bifurcation that takes place at (x, y, µ) D (0, 0, 0) Is the bifurcation supercritical or subcritical? 4.7.55 Consider the system xP D µ x y C x y 2 yP D µ y C x C 2x 2 y where µ is the control parameter. Determine the normal form of this system in the vicinity of the bifurcation point (x, y, µ) D (0, 0, 0). What is the type of this bifurcation? 4.7.56 Consider the system xP 1 D x2 xP 2 D x1 C λx2 x12 Discuss the bifurcations of the ﬁxed points of this system as a function of the control parameter λ. Sketch the state space for λ D 0, λ > 0, and λ < 0. 4.7.57 Consider the system xP D µ x y C x y 2 yP D µ y C x where µ is the control parameter. Determine the normal form of this system in the vicinity of the bifurcation point (x, y, µ) D (0, 0, 0). What is the type of this bifurcation? 4.7.58 Consider the system xP D 2y C µ x C α x (x 2 C y 2 ) yP D 2x C µ y C α y (x 2 C y 2 ) What type of bifurcation occurs at (x, y, µ) D (0, 0, 0). Determine the bifurcating solutions. 4.7.59 Consider the Rössler equations (Rössler, 1976a): xP D (y C z) yP D x C a y zP D b C (x c)z

157

158

4 Bifurcations of Continuous Systems

Assume that the parameters a, b, and c are positive. a) When a is used as a control parameter, verify that a ﬁxed point of this system experiences a saddle-node bifurcation at c c c c2 . (x, y, z, a) D , , , 2 2a 2a 4b b) Simplify the three-dimensional system to the normal form for a saddle-node bifurcation in the vicinity of the bifurcation point. 4.7.60 Consider the Lorenz equations (Lorenz, 1963): xP D σ(y x) yP D x y x z zP D β z C x y Assume that the parameters σ, β, and are positive. a) Choose as the control parameter and examine the different local bifurcations experienced by the different ﬁxed points. Verify that a Hopf bifurcation of a ﬁxed point occurs at c D

σ(σ C β C 3) . σβ1

b) Construct the bifurcation diagram for 0 < c . c) Simplify the three-dimensional system to its normal form for a pitchfork bifurcation in the vicinity of (x, y, z, ) D (0, 0, 0, 1) and obtain uP D

σ σ(1 ) u u3 . σC1 β(σ C 1)

4.7.61 Consider the three-dimensional dynamical system xP 1 D b 1 µ x1 C x2 2x3 C Cb 11 µ x1 C α 1 (x1 C 2x2 C x3 )2

(4.297)

xP 2 D b 2 µ 2x1 x2 x3 C b 22 µ x2 C α 2 (x1 x2 C x3 )2

(4.298)

xP 3 D b 3 µ C x1 C 2x2 x3 C b 33 µ x3 C α 3 (2x1 C x2 x3 )2

(4.299)

Show that the origin of this system undergoes a saddle-node bifurcation as µ increases past zero when (b 3 C b 2 b 1 ) ¤ 0. Show that this system can be simpliﬁed to uP D

1 3

(b 3 C b 2 b 1 ) µ 2 C

1 3

(4α 3 C α 2 4α 1 ) u2

(4.300)

4.7 Exercises

Show that the origin of this system undergoes a transcritical bifurcation as µ increases past zero when (b 3 C b 2 b 1 ) D 0. Show that this system can be simpliﬁed to uP D Γ0 µ 21 C Γ1 µ 1 u C Γ2 u2

(4.301)

where Γ0 D Γ1 D

3b 2 b 11 3b 3 b 22 C 3b 2 b 33 C 3b 3 b 33 4b 22 α 1 C 4b 2 b 3 α 1 b 23 α 1 C4b 22 α 2 C 8b 2 b 3 α 2 C 4b 23 α 2 C b 22 α 3 4b 2 b 3 α 3 C 4b 23 α 3 1 27

1 9

(3b 11 C 3b 22 C 3b 33 8b 2 α 1 C 4b 3 α 1

4b 2 α 2 4b 3 α 2 4b 2 α 3 C 8b 3 α 3 ) Γ2 D

1 3

(4α 3 C α 2 4α 1 )

4.7.62 Consider the system xP D µ x 2y yP D µ y C 2x C λ 1 x z C λ 2 y z zP D z C α x 2 where α, µ, λ 1 , and τ are constants. Use a combination of center-manifold reduction and the method of normal forms to construct periodic solutions of this system for small µ. 4.7.63 Consider the system xP 1 D x1 C h(x3 ) xP 2 D h(x3 ) xP 3 D ax1 C b x2 c h(x3 ) where a, b, and c are positive constants and h(0) D 0 and y h(y ) > 0

for 0 < jy j < k

for some

k>0

a) Show that the origin is an isolated equilibrium point. b) Is the origin an asymptotically stable equilibrium point? c) Suppose that y h(y ) > 0. Is the origin globally asymptotically stable? 4.7.64 Consider the system xP D y yP D x C µ y α y z zP D z C x 2

159

160

4 Bifurcations of Continuous Systems

Show that the center manifold of the origin (x D 0, y D 0, µ D 0) is given by zD

1 5

3x 2 2x y C 2y 2

Then, determine the equations governing the dynamics on this manifold. 4.7.65 Consider the system xP D µ x y C x 2 α 1 x z yP D µ y C x C α 2 y z zP D z C x 2 C 2y 2 a) Show that the origin of this system undergoes a Hopf bifurcation as µ increases through zero. b) Determine the normal form of this bifurcation. c) Is the bifurcation supercritical or subcritical? d) Calculate the amplitude and frequency of the bifurcating limit cycle. 4.7.66 Consider the system xP 1 D µ x1 x2 xP 2 D µ x2 C x1 C 2x1 x3 xP 3 D x3 C x12 where µ is a small parameter. Design a controller to convert the bifurcation at (x1 , x2 , x3 , µ) D (0, 0, 0, 0) from subcritical to supercritical. 4.7.67 Consider the system uR C ω 2 u D µ uP α 1 uv α 2 u vP vP C v D δ u2 Determine the normal form of this system near (u, v ) D (0, 0) as µ passes through zero.

161

5 Forced Oscillations of the Dufﬁng Oscillator In this chapter, we consider the response of a single-degree-of-freedom system to a harmonic excitation modeled by uR C ω 2 u C 2µ uP C α u3 D F cos Ω t where is a small nondimensional parameter that is used as a bookkeeping device. Carrying out a straightforward expansion, one ﬁnds that up to O(), resonances occur if Ω ω: Primary or main resonance Ω 3ω: Subharmonic resonance of order one-third Ω 13 ω: Superharmonic resonance of order three. Next, we use the method of normal forms to determine second-order uniform approximations to the solutions of this equation for these resonances.

5.1 Primary Resonance

In the case of primary resonance, the linear theory shows that a small excitation leads to a large response. Hence, we need to determine the orders of F and u that will explicitly display this observation. One way of accomplishing this is to order the excitation at O() and rewrite the governing equation as uR C ω 2 u D 2µ uP C α u3 F cos Ω t

(5.1)

To apply the method of normal forms, we let z D F eiΩ t

(5.2)

N F cos Ω t D 12 (z C z)

(5.3)

so that

The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

162

5 Forced Oscillations of the Dufﬁng Oscillator

and zP D i Ω z

(5.4)

Hence, we transform the two-dimensional nonautonomous problem into a threedimensional autonomous problem. Next, to quantitatively describe the nearness of the primary resonance, we introduce a detuning parameter σ deﬁned according to Ω 2 D ω 2 C σ

(5.5)

Using (5.5) and (5.2) to eliminate ω 2 and F cos Ω t, respectively, from (5.1), we have uR C Ω 2 u D 2µ uP σ u C α u3 12 (z C z) N

(5.6)

Equation 5.6 can conveniently be represented as a ﬁrst-order complex-valued equation. To this end, we note that when, D 0, the solution of (5.6) can be expressed as N i Ω t u D Ae i Ω t C Ae where A is a constant. Then, N i Ω t uP D i Ω Ae i Ω t Ae When ¤ 0, A will be time-varying rather than constant. To represent (5.6) as a single complex-valued equation, we identify Ae i Ω t as ζ and hence introduce the transformation u D ζ C ζN

uP D i Ω ζ ζN

and

(5.7)

so that ζD

1 2

u

i uP Ω

and

1 ζN D 2

uC

i uP Ω

(5.8)

With this transformation, (5.6) becomes 3 1 i 2i Ω µ ζ ζN σ ζ C ζN C α ζ C ζN (z C z) ζP D i Ω ζ C N 2Ω 2 (5.9) We note that the unperturbed problem is transformed into the simple equation ζP D i Ω ζ under the transformation (5.7). Other transformations may lead to an N equation involving ζ and ζ. Next, we use the near-identity transformation ζ D η C h (η, η, N z, z) N C

(5.10)

5.1 Primary Resonance

so that (5.9) takes the simple form ηP D i Ω η C g (η, η, N z, zN ) C

(5.11)

Substituting (5.10) and (5.11) into (5.9), using (5.4), and equating the coefﬁcients of on both sides, we obtain @h @h @h @h g C iΩ η ηN C z zN h @η @ ηN @z @ zN 1 i 3 (η (η (z (η σ C η) N C α C η) N C z) N (5.12) D µ η) N C 2Ω 2 Next, we choose h to eliminate all of the nonresonance terms in (5.12), leaving g with the resonance and near-resonance terms. The right-hand side of (5.12) suggests choosing h in the form h D Λ 1 η 3 C Λ 2 η 2 ηN C Λ 3 η ηN 2 C Λ 4 ηN 3 C ∆ 1 η C ∆ 2 ηN C ∆ 3 z C ∆ 4 zN (5.13) Substituting (5.13) into (5.12) and rearranging yields iσ i α 3i α 2 η3 g C µη C η η ηN C z i 2Ω Λ 1 C 2Ω 2Ω 4Ω 2Ω 3α α 2 η ηN i 4Ω Λ 4 C ηN 3 i 2Ω Λ 3 C 2Ω 2Ω iσ 1 ηN i 2Ω ∆ 4 zN D 0 2i Ω ∆ 2 C µ 2Ω 4Ω

(5.14)

We note that (5.14) does not depend on ∆ 1 , ∆ 3 , and Λ 2 , and hence they are arbitrary, and the terms proportional to η, η 2 η, N and z are resonance terms. Choosing Λ 1 , Λ 3 , Λ 4 , ∆ 2 , and ∆ 4 to eliminate the nonresonance terms, we have iµ 1 σ , ∆4 D C 2Ω 4Ω 2 8Ω 2 α 3α α , Λ3 D , Λ4 D Λ1 D 4Ω 2 4Ω 2 8Ω 2

∆2 D

(5.15) (5.16)

With the choices (5.15) and (5.16), (5.14) yields g as g D µ η

i iσ 3i α 2 ηC η ηN z 2Ω 2Ω 4Ω

(5.17)

which, upon substitution into (5.11), yields the normal form ηP D i Ω η µ η

i 3iα 2 iσ ηC η ηN z 2Ω 2Ω 4Ω

(5.18)

Substituting (5.13) into (5.10) and then substituting the result into (5.7), we obtain N 2 C Λ 3 η ηN 2 u D η C ηN C (Λ 1 C Λ 4 ) η 3 C ηN 3 C (Λ 2 C Λ 3 ) η 2 ηN C Λ N 3 C ∆ 4 zN C N 2 η C ∆2 C ∆ N 1 ηN C (∆ 3 C ∆ 4 ) z C ∆ C ∆1 C ∆ (5.19)

163

164

5 Forced Oscillations of the Dufﬁng Oscillator

If we express η in the polar form η D 12 ae i (Ω tCγ )

(5.20)

we ﬁnd that, to O(1), the amplitude of the fundamental frequency Ω of oscillations N η ηN 2 , η, η, N z, is a. However, some of the ﬁrst-order terms in (5.19), namely, η 2 η, and zN , also have the frequency Ω , and hence modify a and make it nonunique. In order to uniquely deﬁne the amplitude of the term at the fundamental frequency Ω of oscillations by a, we choose Λ 2 , ∆ 1 , and ∆ 3 to eliminate the terms involving z, zN , η, η, N η 2 η, N and η ηN 2 in (5.19). With this choice, (5.19) becomes u D η C ηN C

α 3 η C ηN 3 C 8Ω 2

(5.21)

on account of (5.16). In terms of the polar coordinates (5.20), (5.21) becomes u D a cos (Ω t C γ ) C

α a 3 cos (3Ω t C 3γ ) C 32Ω 2

(5.22)

Substituting (5.2) and (5.20) into (5.18) and separating real and imaginary parts, we obtain F sin γ 2Ω

aP D µ a a γP D

3α a 3 F σ aC cos γ 2Ω 8Ω 2Ω

(5.23) (5.24)

5.2 Subharmonic Resonance of Order One-Third

To express the nearness of Ω to 3ω, we introduce the detuning parameter σ deﬁned as 1 Ω2 9

D ω 2 C σ

(5.25)

Moreover, because this resonance is called secondary resonance, we need to order F cos Ω t at O(1) so that its inﬂuence is accounted for in the transformed equation. Consequently, using (5.3) to replace F cos Ω t with 1/2(z C zN ) and using (5.25) to replace ω 2 with 1/9Ω 2 σ, we rewrite (5.1) as uR C 19 Ω 2 u D

1 2

(z C z) N 2µ uP σ u C α u3

(5.26)

To express (5.26) in complex-valued form, we let u D ζ C ζN

and

uP D 13 i Ω ζ ζN

(5.27)

5.2 Subharmonic Resonance of Order One-Third

so that ζD

1 2

3i u uP Ω

and

ζN D

1 2

3i uC uP Ω

(5.28)

Substituting (5.27) and (5.28) into (5.26) yields 1 3i (z C z) ζP D i Ω ζ N 3 4Ω 3 3i 2 C i µ Ω ζ ζN σ ζ C ζN C a ζ C ζN 2Ω 3

(5.29)

Because the frequency of the response is 1/3Ω and because z and zN have the frequencies Ω and Ω , z C zN is not a resonance term in this case and because it appears at O(1), we ﬁrst introduce a linear transformation to remove it from (5.29) at O(1). To this end, we let ζ D η C Γ1 z C Γ2 zN

(5.30)

in (5.29) and obtain ηP D

3i 1 1 (z C z) i Ω η C i Ω (Γ1 z C Γ2 zN ) Γ1 zP Γ2 zPN N C O() (5.31) 3 3 4Ω

Using (5.4) to eliminate zP and zPN from (5.31) yields ηP D

3i 1 1 (z C zN ) C O() i Ω η C i Ω (Γ1 z C Γ2 zN ) i Ω Γ1 z C i Ω Γ2 zN 3 3 4Ω (5.32)

Next, we choose Γ1 and Γ2 to eliminate z and zN from (5.32) and obtain Γ1 D

9 8Ω 2

Γ2 D

and

9 16Ω 2

(5.33)

Substituting (5.30) into (5.29) and using (5.4) and (5.33), we obtain 1 3i ηP D i Ω η C 3 2Ω σ η C ηN

"

2 27 zN 27z C i µ Ω η ηN 3 16Ω 2 16Ω 2 3 # 9z 9z 9 zN 9 zN C α η C ηN 16Ω 2 16Ω 2 16Ω 2 16Ω 2 (5.34)

To simplify (5.34), we introduce the near-identity transformation η D ξ C h ξ , ξN , z, zN

(5.35)

165

166

5 Forced Oscillations of the Dufﬁng Oscillator

and obtain @h PN @h 1 1 @h P @h ξ ξ zP zPN ξP D i Ω ξ C iΩ h N 3 3 @ξ @z @ zN @ξ " 3 3i 9 zN 9z C α ξ C ξN 2Ω 16Ω 2 16Ω 2 9z 9 zN N σ ξ C ξ 16Ω 2 16Ω 2 # 2 27 zN 27z N C iµΩ ξ ξ C C 3 16Ω 2 16Ω 2

(5.36)

Substituting (5.35) into (5.30) and then substituting the result into (5.27), we have

9 N C (z u D ξ C ξN C z) N C h C h (5.37) 16Ω 2 on account of (5.33). To be strictly consistent, we should not include the term (h C N in (5.37) if we are not going to keep the resonance terms in (5.36) that arise h) from h. The form of the O() terms in (5.36) suggests choosing h in the form h D ∆ 1 ξ C ∆ 2 ξN C ∆ 3 z C ∆ 4 zN C Λ 1 ξ 3 C Λ 2 ξ 2 ξN C Λ 3 ξ ξN 2 C Λ 4 ξN 3 C Λ 5 z 3 C Λ 6 z 2 zN C Λ 7 z zN 2 C Λ 8 zN 3 C Λ 9 ξ 2 z C Λ 10 ξ 2 zN C Λ 11 ξN 2 z C Λ 12 ξN 2 zN C Λ 13 ξ z 2 C Λ 14 ξN z 2 C Λ 15 ξ zN 2 C Λ 16 ξN zN 2 C Λ 17 ξ z zN (5.38) C Λ 18 ξN z zN C Λ 19 ξ ξN z C Λ 20 ξ ξN zN Because zP D i Ω z and, to the ﬁrst approximation, ξP D 1/3i Ω ξ , out of all of N and ξN 2 z are resonance the terms in (5.36), only the terms involving ξ , ξ 2 ξN , ξ z z, terms and hence the coefﬁcients ∆ 1 , Λ 2 , Λ 11 , and Λ 17 , are arbitrary. Moreover, as N in (5.37) if we are going aforementioned, we are not justiﬁed in including (h C h) to eliminate only the resonance terms from (5.36). Substituting (5.38) into (5.36) and using (5.4), we can choose the Λ m to eliminate all of the terms except the resonance terms. Consequently, we obtain the simple equation 243α 1 27α N 2 3iσ 3i 3α ξ 2 ξN C C ξP D i Ω ξ µ ξ ξ z z N z ξ ξC 3 2Ω 2Ω 128Ω 4 16Ω 2 (5.39) Substituting (5.2) and the polar form ξ D 12 ae i ( 3 Ω tCγ ) 1

(5.40)

into (5.37) and (5.39) and separating real and imaginary parts, we obtain to the second approximation 1 9F cos Ω t C (5.41) u D a cos Ωt C γ 3 8Ω 2

5.3 Superharmonic Resonance of Order Three

where aP D µ a a γP D

81α F 2 a sin 3γ 64Ω 3

(5.42)

9α 3 81α F 2 3σ 729F 2 aC a cos 3γ aC a 2Ω 256Ω 5 8Ω 64Ω 3

(5.43)

5.3 Superharmonic Resonance of Order Three

To express the nearness of ω to 3Ω , we introduce the detuning parameter σ deﬁned according to 9Ω 2 D ω 2 C σ

(5.44)

As in the case of subharmonic resonance of order one-third, we need to order the excitation at O(1) so that the effect of this resonance will occur at the same order as the effects of the damping and nonlinearity. Substituting (5.44) into (5.1), using (5.3), and assuming that z D O(1), we obtain uR C 9Ω 2 u D

1 2

(z C zN ) 2µ uP σ u C α u3

(5.45)

To express (5.45) in complex-valued form, we let u D ζ C ζN

and

uP D 3i Ω ζ ζN

(5.46)

so that ζD

1 2

u

i uP 3Ω

and

1 ζN D 2

uC

i uP 3Ω

(5.47)

With this transformation, (5.45) becomes i (z C zN ) ζP D 3i Ω ζ 12Ω 3 i i h C 6i µ Ω ζ ζN σ ζ C ζN C α ζ C ζN 6Ω

(5.48)

The ﬁrst step is to introduce a linear transformation to eliminate the term proportional to z C zN at O(1). To this end, we let ζ D η C Γ1 z C Γ2 zN

(5.49)

in (5.48) and obtain N Γ1 zP Γ2 zPN ηP D 3i Ω η C 3i Ω (Γ1 z C Γ2 z)

i (z C z) N C O() (5.50) 12Ω

167

168

5 Forced Oscillations of the Dufﬁng Oscillator

Using (5.4), we express zP as i Ω z and zNP as i Ω zN in (5.50) and obtain i i z C 4 i Ω Γ2 zN C O() ηP D 3i Ω η C 2i Ω Γ1 12Ω 12Ω

(5.51)

We choose Γ1 and Γ2 to eliminate the terms involving z and zN in (5.51); that is, we put Γ1 D

1 24Ω 2

and

Γ2 D

1 48Ω 2

(5.52)

Substituting (5.49) into (5.48) and using (5.4) and (5.52), we obtain " i zN z ηP D 3i Ω η C 6i µ Ω η ηN C 6Ω 48Ω 2 48Ω 2 3 # z z zN zN C α η C ηN C σ η C ηN C C C 16Ω 2 16Ω 2 16Ω 2 16Ω 2 (5.53) To simplify (5.53), we introduce the near-identity transformation η D ξ C h ξ , ξN , z, zN

(5.54)

and obtain @h P @h PN @h @h ξP D 3i Ω ξ C 3iΩ h ξ ξ zP zPN @ξ @z @ zN @ ξN " i zN z C 6i µ Ω ξ ξN C 6Ω 48Ω 2 48Ω 2 zN z C σ ξ C ξN C 16Ω 2 16Ω 2 3 # z N z C C Cα ξ C ξN C 16Ω 2 16Ω 2

(5.55)

Next, we choose h to eliminate the nonresonance terms. As discussed in the preceding section, if we want to stop at second order, we do not need to determine h and all that we need is to keep the resonance terms in (5.55). The result is z3 Pξ D 3i Ω ξ µ ξ iσ ξ C iα 3ξ 2 ξN C 3z zN ξ C (5.56) 6Ω 6Ω 128Ω 4 4096Ω 6 Substituting (5.54) into (5.49) and then substituting the result into (5.46), we have 1 (z C zN ) C O() u D ξ C ξN C (5.57) 16Ω 2 Expressing ξ in the polar form ξ D 12 ae i (3Ω tCγ )

(5.58)

5.4 An Alternate Approach

and using (5.2) to replace z with F e i Ω t , we rewrite (5.57) as u D a cos (3Ω t C γ ) C

F cos Ω t C 8Ω 2

(5.59)

Substituting (5.58) and (5.2) into (5.56) and separating real and imaginary parts, we obtain aP D µ a C a γP D

α F 3 sin γ 12 288Ω 7

(5.60)

α 3 α F 3 σ α F 2 a C C cos γ aC a 6Ω 256Ω 5 8Ω 12 288Ω 7

(5.61)

5.4 An Alternate Approach

In the preceding sections, we used the detuning relationships (5.5), (5.25), and (5.44) to replace ω 2 in terms of Ω 2 , 1/9Ω 2 , and 9Ω 2 in (5.1). Alternatively, we can have ω as it is in the governing equation and watch for near-resonance terms. We describe this approach for the cases of subharmonic and superharmonic resonances of order one-third and three, respectively. In these cases, F is assumed to be O(1). To express (5.1) in complex-valued form we let u D ζ C ζN and uP D i ω ζ ζN (5.62) so that ζD

1 2

u

i uP ω

and

1 ζN D 2

uC

i uP ω

(5.63)

Using (5.3), (5.62), and (5.63) and the fact that F D O(1), we rewrite (5.1) as 3 i i i h (z C zN ) C ζP D i ωζ (5.64) 2i ωµ ζ ζN C α ζ C ζN 4ω 2ω Again, the ﬁrst step is to introduce a linear transformation to eliminate the term proportional to z C zN . To this end, we substitute (5.49) into (5.64) and obtain i (z C zN ) C O() ηP D i ωη C i ω (Γ1 z C Γ2 z) N Γ1 zP Γ2 zPN 4ω

(5.65)

Using (5.4) to replace zP and zPN with i Ω z and i Ω zN , we rewrite (5.65) as 1 1 z C i ωΓ2 C Ω Γ2 zN C O() ηP D i ωη C i ωΓ1 Ω Γ1 4ω 4ω (5.66) Next, we choose Γ1 and Γ2 to eliminate the terms involving z and zN in (5.66); that is, Γ1 D

1 4ω(ω Ω )

and

Γ2 D

1 4ω(ω C Ω )

(5.67)

169

170

5 Forced Oscillations of the Dufﬁng Oscillator

Substituting (5.49) into (5.64) and using (5.67), we obtain ( i Ω (z zN ) ηP D i ωη C 2i ωµ η ηN C 2ω 2ω(ω 2 Ω 2 ) ) 3 z C zN Cα η C ηN C 2(ω 2 Ω 2 )

(5.68)

To simplify (5.68), we use the near-identity transformation (5.54) and obtain @h PN @h @h P @h ξ ξ zP zPN ξP D i ωξ C iωh @ξ @z @ zN @ ξN ( i Ω (z zN ) C 2i ωµ ξ ξN C 2ω 2ω(ω 2 Ω 2 ) 3 ) z C zN N Cα ξ C ξ C C 2(ω 2 Ω 2 )

(5.69)

Then, we choose h to eliminate all of the terms in (5.69) except the resonance and near-resonance terms. The resonance terms correspond to the terms that produce secular terms, whereas the near-resonance terms correspond to terms that produce small-divisor terms in the applications of the method of multiple scales (Nayfeh, 1973, 1981) and the Krylov–Bogoliubov–Mitropolsky technique (Bogoliubov and Mitropolsky, 1961). The resonance terms in (5.69) are the terms proportional to ξ , ξ 2 ξN , and ξ z z. N The near-resonance terms depend on the resonance being considered. In the subharmonic-resonance case of order one-third, the term proportional to z ξN 2 is the near-resonance term, whereas in the superharmonic-resonance case of order three, the term proportional to z 3 is the near-resonance term. Therefore, keeping the resonance and near-resonance terms in (5.69), we obtain iα 3z zN 3z N 2 (5.70) 3ξ 2 ξN C ξ ξP D i ωξ µ ξ C ξ C 2ω 2(ω 2 Ω 2 )2 2(ω 2 Ω 2 ) when Ω 3ω (i.e., when there is a subharmonic resonance of order one-third), and 3z zN iα z3 3ξ 2 ξN C (5.71) ξP D i ωξ µ ξ C ξ C 2ω 2(ω 2 Ω 2 )2 8(ω 2 Ω 2 )3 when Ω 1/3ω (i.e., when there is a superharmonic resonance of order three). Again, to second order, we do not need to determine h. Then, substituting (5.54) and (5.49) into (5.62) and using (5.67), we obtain u D ξ C ξN C

1 (z C zN ) C 2(ω 2 Ω 2 )

(5.72)

Next, substituting the polar form ξD

1 i (ω tCβ ) ae 2

(5.73)

5.4 An Alternate Approach

into (5.72) and using the fact that z D F e i Ω t , we have u D a cos (ωt C β) C

ω2

F cos Ω t C Ω2

(5.74)

5.4.1 Subharmonic Case

For the subharmonic-resonance case, substituting (5.73) into (5.70), separating real and imaginary parts, and using the fact that z D F e i Ω t , we obtain aP D µ a C a βP D

3α Fa 2 sin 3γ 8ω(ω 2 Ω 2 )

3α F 2 a 3α Fa 2 3α 3 C a C cos 3γ 2 2 2 8ω 4ω(ω Ω ) 8ω(ω 2 Ω 2 )

(5.75) (5.76)

where 3γ D 3β (Ω 3ω) t

(5.77)

Substituting for β from (5.77) into (5.74) yields F 1 cos Ω t C Ωt C γ C 2 u D a cos 3 ω Ω2

(5.78)

while substituting for β from (5.77) into (5.76) yields 3α F 2 a 3α Fa 2 3α 3 1 C a C cos 3γ a γP D (Ω 3ω) a C 2 2 2 3 8ω 4ω(ω Ω ) 8ω(ω 2 Ω 2 ) (5.79) Therefore, to the second approximation, u is given by (5.78), where a and γ are given by the autonomous equations (5.75) and (5.79). These equations are in full agreement with those obtained by using the methods of multiple scales and averaging. To compare the expansion obtained in this section with that obtained in Section 5.2, we solve for ω in terms of Ω from (5.25) and obtain ω2 D

1 2 Ω σ 9

or

ωD

1 3σ Ω C 3 2Ω

(5.80)

Substituting for ω from (5.80) into (5.78), (5.75), and (5.79), we obtain (5.41)–(5.43) to O().

171

172

5 Forced Oscillations of the Dufﬁng Oscillator

5.4.2 Superharmonic Case

In this case, substituting (5.73) and (5.2) into (5.71) and separating real and imaginary parts, we obtain aP D µ a C a βP D

α F 3 sin γ 8ω(ω 2 Ω 2 )3

(5.81)

3α F 2 a α F 3 3α 3 C cos γ a C 2 2 2 8ω 4ω(ω Ω ) 8ω(ω 2 Ω 2 )3

(5.82)

where γ D β (3Ω ω) t

(5.83)

Using (5.83) to eliminate β from (5.74) yields u D a cos (3Ω t C γ ) C

F cos Ω t C ω2 Ω 2

(5.84)

while using (5.83) to eliminate β from (5.82) yields a γP D (3Ω ω) a C

3α F 2 a α F 3 3α 3 C cos γ a C 8ω 4ω(ω 2 Ω 2 )2 8ω(ω 2 Ω 2 )3 (5.85)

Therefore, to the second approximation, u is given by (5.84) where a and γ are given by the autonomous equations (5.81) and (5.85). This approximation is in full agreement with that obtained by using the methods of multiple scales and averaging. To compare the expansion obtained in this section with that obtained in Section 5.3, we use (5.44) to solve for ω in terms of Ω and obtain ω 2 D 9Ω 2 σ

or

ω D 3Ω

σ C 6Ω

(5.86)

Using (5.86) to eliminate ω from (5.84), (5.81), and (5.85), we obtain (5.59)–(5.61) to O().

5.5 Exercises

5.5.1 Consider the system uR C ω 2 u D uP 13 uP 3 C F cos Ω t

5.5 Exercises

Use the method of normal forms to determine an approximation to this system when Ω ω, 3ω, and 1/3ω. 5.5.2 Use the method of normal forms to determine a second-order approximation to the solution of uR C ω 2 C 2µ uP C α u2 D F cos Ω t when Ω 2ω.

173

175

6 Forced Oscillations of SDOF Systems 6.1 Introduction

In this chapter, we consider small- but ﬁnite-amplitude responses of a singledegree-of-freedom system with quadratic and cubic nonlinearities to a harmonic excitation; speciﬁcally, we consider solutions of uR C ω 2 u C 2µ uP C δ u2 C α u3 D F cos Ω t

(6.1)

The results of Section 1.5 show that, to the second approximation, the free undamped solution of (6.1) can be expressed as u D a cos ( ωt O C β) C

δ a2 [cos (2 ωt O C 2β) 3] C 6ω 2

where a and β are constants and the frequency ω O of free oscillations is given by 2 3α 5δ ω O DωC a2 C 8ω 12ω 3 Consequently, as the oscillation amplitude a increases, the undamped frequency ω O of oscillation is shifted from the undamped linear frequency ω of oscillation by 3α 5δ 2 a2 8ω 12ω 3 To keep track of the different orders of magnitude, we introduce a small nondimensional parameter as a bookkeeping device and scale the quadratic terms at O() and the cubic term at O( 2 ). The resonances produced by the excitation try to make the response very large. But as the response grows, the nonlinearity, which modiﬁes the natural frequency of the system, and the damping, which dissipates part of the input energy, are activated, thereby limiting the response at small but ﬁnite amplitude. Because the quadratic and cubic nonlinearities produce a resonance term ﬁrst at O( 2 ) according to Section 1.5, we scale the damping at O( 2 ). With these scalings, we rewrite (6.1) as uR C ω 2 u D F cos Ω t δ u2 2 2µ uP C α u3 (6.2) The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

176

6 Forced Oscillations of SDOF Systems

Because the appearance of resonance terms produced by the excitation depends on the excitation magnitude F and the type of resonance, we scale F differently, depending on the resonance being considered. Carrying out a straightforward expansion, one ﬁnds the following resonances:

Ω Ω Ω Ω Ω

ω: Primary or main resonance 2ω: Subharmonic resonance of order one-half 12 ω: Superharmonic resonance of order two 3ω: Subharmonic resonance of order one-third 13 ω: Superharmonic resonance of order three.

To treat these cases and determine the appropriate scaling of F for each resonance, we cast (6.2) in complex-valued form by letting u D ζ C ζN and uP D i ω ζ ζN (6.3) and z D F eiΩ t

(6.4)

zP D i Ω z

(6.5)

so that

Using (6.3)–(6.5), we rewrite (6.2) as 2 i 2 h 3 i iδ i (z C z)C ζ C ζN C N 2i µ ω ζ ζN Cα ζ C ζN ζP D i ωζ 4ω 2ω 2ω (6.6)

6.2 Primary Resonance

In this case Ω ω and we need to scale F at O( 2 ) so that the resonance term produced by the excitation appears at the same order as those produced by the damping and the nonlinearity. With this scaling, we rewrite (6.6) as 2 Pζ D i ωζ C iδ ζ C ζN 2 2 µ ζ ζN C i α ζ C ζN 3 1 (z C z) N 2ω 2ω 2 (6.7) To determine a second-order uniform expansion of (6.7), we let N C 2 h 2 (η, η, N z, zN ) C ζ D η C h 1 (η, η)

(6.8)

N C 2 g 2 (η, η, N z, zN ) C ηP D i ωη C g 1 (η, η)

(6.9)

where

6.2 Primary Resonance

and the g i represent resonance and near-resonance terms that cannot be eliminated by nonsingular choices of the h i . Substituting (6.8) and (6.9) into (6.7) and equating coefﬁcients of like powers of , we obtain iδ @h 1 @h 1 (η C η) (6.10) η ηN h 1 D N 2 g1 C i ω @η @ ηN 2ω @h 2 @h 2 @h 2 @h 2 η ηN h 2 C i Ω z zN g2 C i ω @η @ ηN @z @ zN @h 1 iδ @h 1 (η C η) N C g1 gN 1 µ (η η) N h 1 C hN 1 D @η @ ηN ω 1 i α (η C η) N 3 (z C zN ) (6.11) C 2ω 2 Because we will proceed to the next order, we need to explicitly determine h 1 . The right-hand side of (6.10) suggests seeking h 1 in the form h 1 D Γ1 η 2 C Γ2 η ηN C Γ3 ηN 2

(6.12)

Substituting (6.12) into (6.10) yields δ δ δ η 2 C ωΓ2 C η ηN C 3ωΓ3 C ηN 2 D 0 (6.13) g 1 ωΓ1 2ω ω 2ω N and ηN 2 can be eliminated from (6.13) if Hence, the terms proportional to η 2 , η η, Γ1 D

δ , 2ω 2

Γ2 D

δ , ω2

Γ3 D

δ 6ω 2

(6.14)

Then, (6.13) reduces to g 1 D 0. Substituting (6.12) and (6.14) into (6.11) and using the fact that g 1 D 0, we obtain @h 2 @h 2 @h 2 @h 2 η ηN h 2 C i Ω z zN D µ (η η) N @η @ ηN @z @ zN 2 i 1 i δ2 3 2 (η (η (z ) α C η) N C η) N η C η N 6η η N C C z N C 3ω 3 2ω 2 (6.15)

g2 C i ω

The usual approach for determining the normal form is to choose h 2 to eliminate as many terms as possible from (6.15), thereby leaving g 2 with the remaining terms. Alternatively, we choose g 2 to eliminate the resonance and near-resonance terms in (6.15) knowing that we can always eliminate the nonresonance terms by a proper choice of h 2 . Because we are stopping at this order (seeking a second-order approximation), we do not need to solve explicitly for h 2 and all that we need to do is choose g 2 to eliminate the resonance and near-resonance terms in (6.15); that is, we put 10δ 2 i i 3α η 2 ηN g 2 D µ η C z 2ω 3ω 2 4ω

177

178

6 Forced Oscillations of SDOF Systems

Substituting for g 1 and g 2 into (6.9), we obtain the normal form ηP D i ωη 2 µ η C

i 2 2ω

10δ 2 i 2 3α η 2 ηN z 2 3ω 4ω

(6.16)

Substituting (6.12) and (6.14) into (6.8) and then substituting the result into (6.3), we obtain to the second approximation u D η C ηN C

δ 2 η C ηN 2 6η ηN C 3ω 2

(6.17)

where η is given by (6.16). Finally, we introduce the polar representation η D 12 ae i (ω tCβ )

(6.18)

into (6.17) and obtain u D a cos(ωt C β) C

δ a 2 cos(2ωt C 2β) 3 C 6ω 2

(6.19)

Substituting (6.18) and (6.4) into (6.16) and separating real and imaginary parts, we obtain 2 F sin [(Ω ω) t β] 2ω 3α 2 F 5δ 2 3 a cos [(Ω ω) t β] a βP D 2 8ω 12ω 3 2ω aP D 2 µ a C

(6.20) (6.21)

Equations 6.19–6.21 are the same as those obtained by using the methods of multiple scales and averaging (Nayfeh and Mook, 1979).

6.3 Subharmonic Resonance of Order One-Half

Because z and zN are nonresonance terms when Ω is away from ω, we ﬁrst introduce the linear transformation ζ D η C ∆ 1 z C ∆ 2 zN

(6.22)

into (6.6) and obtain ηP D i ωη C i ω (∆ 1 z C ∆ 2 zN ) ∆ 1 zP ∆ 2 zPN

i (z C zN ) C O() 4ω

(6.23)

6.3 Subharmonic Resonance of Order One-Half

Using (6.5) to express zP and zPN as i Ω z and i Ω zN , we rewrite (6.23) as 1 1 z C i (ω C Ω ) ∆ 2 zN C O() ηP D i ωη C i (ω Ω ) ∆ 1 4ω 4ω (6.24) We choose ∆ 1 and ∆ 2 to eliminate the terms involving z and zN ; that is, ∆1 D

1 4ω(ω Ω )

and

∆2 D

1 4ω(ω C Ω )

(6.25)

It follows from (6.25) that the transformation (6.22) is singular or near singular when Ω ω; that is, z and zN would be resonance terms in this case. This is the reason we scaled F and hence z and zN at O( 2 ) in the preceding section. Substituting (6.22) and (6.25) into (6.6) yields 2 z C zN iδ η C ηN C 2 µ (η η) N ηP D i ωη C 2ω 2(ω 2 Ω 2 ) " 3 # i 2 2i µ Ω (z zN ) z C zN C (6.26) C α η C ηN C 2ω (ω 2 Ω 2 ) 2(ω 2 Ω 2 ) When Ω 2ω, the excitation produces the near-resonance term z η. N In order to have this term appear at the same order at which the nonlinear resonance term η 2 ηN appears, we rescale F and hence z as O() and rewrite (6.26) as ηP D i ωη C

iδ i 2 δ (η C η) (η C η) N N 2C N (z C z) N 2 µ (η η) 2ω 2ω(ω 2 Ω 2 )

i 2 α (6.27) (η C η) N 3 C 2ω Because there are no resonance or near-resonance terms at O() in (6.27), we can eliminate the O() terms, as in the preceding section, by using the near-identity transformation C

ηDξC

δ 2 N 2 3ξ ξ 6ξ ξN 6ω 2

(6.28)

Then, (6.27) becomes i 2 δ ξ C ξN (z C zN ) 2 µ ξ ξN 2ω(ω 2 Ω 2 ) 3 i 2 δ 2 i 2 α ξ C ξN ξ 2 C ξN 2 6ξ ξN C ξ C ξN C 3 2ω 3ω

ξP D i ωξ C

(6.29)

To simplify (6.29), we can introduce a near-identify transformation to eliminate the nonresonance terms. Because we are stopping at this order, we need to keep only the resonance terms ξ and ξ 2 ξN and near-resonance term z ξN . The result is 10δ 2 i 2 δ i 2 ξP D i ωξ 2 µ ξ C 3α ξ 2 ξN C z ξN (6.30) 2 2ω 3ω 2ω(ω 2 Ω 2 )

179

180

6 Forced Oscillations of SDOF Systems

Substituting (6.22) and (6.28) into (6.3), using (6.25), and recalling that F is scaled at O(), we obtain to the second approximation u D ξ C ξN C

(z C z) N δ 2 ξ C ξN 2 6ξ ξN C C 2 2 2 2(ω Ω ) 3ω

(6.31)

where ξ is given by (6.30). Substituting the polar forms (6.4) and (6.18) into (6.31) yields u D a cos (ωt C β) C

F δ a 2 [cos (2ωt C 2β) 3] C cos Ω t C ω2 Ω 2 6ω 2 (6.32)

while substituting (6.4) and (6.18) into (6.30) and separating real and imaginary parts yields 2 δ F sin [2β (Ω 2ω) t] 2ω(ω 2 Ω 2 ) 3α 2 δ F 5δ 2 3 a C cos [2β (Ω 2ω) t] a βP D 2 8ω 12ω 3 2ω(ω 2 Ω 2 ) aP D 2 µ a C

(6.33) (6.34)

Equations 6.32–6.34 are in full agreement with those obtained by using the method of multiple scales (Nayfeh, 1986).

6.4 Superharmonic Resonance of Order Two

In this case, Ω 1/2ω. Inspection of (6.26) reveals that the resonance term arising from the excitation is z 2 . In order that its inﬂuence balance the resonance terms arising from the damping and the nonlinearity, we scale F and hence z at O( 1/2 ) and then rewrite (6.26) as ηP D i ωη C C

iδ i 3/2 δ (η C η) (η C η) N N 2C N (z C z) N 2 µ (η η) 2ω 2ω(ω 2 Ω 2 )

i 2 α i 2 δ(z C zN )2 (η C η) C N 3 C 2 2 2 8ω(ω Ω ) 2ω

(6.35)

As in the preceding section, we introduce the near-identity transformation (6.28) to eliminate the O() terms in (6.35) and hence obtain i 3/2 δ ξ C ξN (z C zN ) 2 µ ξ ξN 2 2 2ω(ω Ω ) N 2 i 2 δ 2 i 2 δ(z C z) ξ C ξN ξ 2 C ξN 2 6ξ ξN C C 2 2 2 3 8ω(ω Ω ) 3ω 3 i 2 α ξ C ξN C C 2ω

ξP D i ωξ C

(6.36)

6.4 Superharmonic Resonance of Order Two

Because there are no resonance terms at O( 3/2 ), we can introduce a near-identity transformation ξ D χ C 3/2 h (χ, χN , z, zN )

(6.37)

to eliminate these terms and rewrite (6.36) as i 2 δ 2 (χ C χN ) χ 2 C χN 2 6χ χN 2 µ (χ χN ) 3ω 3 i 2 δ i 2 α (z C z) (χ C χN )3 C N 2 C C 2ω 8ω(ω 2 Ω 2 )2

χP D i ωχ C

(6.38)

Because our aim is to obtain an expression that is valid to O( 2 ), we do not need to determine h. Moreover, we also know that we can choose a near-identity transformation N z, z) N χ D w C 2 g (w, w,

(6.39)

to eliminate the nonresonance terms from (6.38) and obtain 10δ 2 i 2 i 2 δ 2 2 3α w w N C z 2 C (6.40) wP D i ωw µ w C 2ω 3ω 2 8ω(ω 2 Ω 2 )2 Again, for an expression that is valid to O( 2 ), we do not need to determine g. Substituting (6.22), (6.25), (6.28), (6.37), and (6.39) into (6.3), we have u D w C wN C

δ 2 1/2 (z C zN ) w C wN 2 6w wN C C 2 2 2 2(ω Ω ) 3ω

(6.41)

where use has been made of the assumption that F and hence z are O( 1/2 ). Expressing w in the polar form wD

1 i (ω tCβ ) ae 2

(6.42)

we rewrite (6.41) as u D a cos (ωt C β) C

δ a 2 1/2 F [cos (2ωt C 2β) 3] C cos Ω t C 2 2 ω Ω 6ω 2 (6.43)

Substituting (6.4) and (6.42) into (6.40) and separating real and imaginary parts, we obtain 2 δ F 2 sin [β (2Ω ω) t] 4ω(ω 2 Ω 2 )2 3α 2 δ F 2 5δ 2 3 a C cos [β (2Ω ω) t] a βP D 2 8ω 12ω 3 4ω(ω 2 Ω 2 )2

aP D 2 µ a C

(6.44) (6.45)

Equations 6.43–6.45 are in full agreement with those obtained by using the method of multiple scales (Nayfeh, 1986).

181

182

6 Forced Oscillations of SDOF Systems

6.5 Subharmonic Resonance of Order One-Third

In this case, Ω 3ω. Inspection of (6.26) shows that the resonance term produced by the excitation is z ηN 2 . Because this resonance term occurs at O( 2 ), F and hence z are O(1). Because none of the O() terms is resonance in this case, one can introduce a near-identity transformation in the form η D ξ C g ξ , ξN , z, zN

(6.46)

and eliminate them. The form of the O() terms suggests seeking g in the form g D Λ 1 ξ 2 C Λ 2 ξ ξN C Λ 3 ξN 2 C Λ 4 z 2 C Λ 5 z zN C Λ 6 zN 2 C Λ 7 ξ z C Λ 8 ξ zN (6.47) C Λ 9 ξN z C Λ 10 ξN zN Substituting (6.46) into (6.26) shows that ξP D i ωξ C O()

(6.48)

Consequently, substituting (6.46) and (6.47) into (6.26) and replacing zPN and zP with i Ω zN and i Ω z and ξPN and ξP with i ω ξN and i ωξ on the right-hand side of the resulting equation, we obtain to O() δ δ ξP D i ωξ C i ωΛ 1 C ξ 2 C i ωΛ 2 C ξ ξN 2ω ω δ N2 δ C i 3ωΛ 3 C ξ C i ωΛ 5 C z zN 2ω 4ω(ω 2 Ω 2 )2 δ C i (ω 2Ω ) Λ 4 C z2 8ω(ω 2 Ω 2 )2 δ zN 2 C i (ω C 2Ω ) Λ 6 C 8ω(ω 2 Ω 2 )2 δ ξz i Ω Λ 7 2ω(ω 2 Ω 2 ) δ C i Ω Λ 8 C ξ zN 2ω(ω 2 Ω 2 ) δ ξN z C i (2ω Ω ) Λ 9 C 2ω(ω 2 Ω 2 ) δ C i (2ω C Ω ) Λ 10 C ξN zN C O( 2 ) 2ω(ω 2 Ω 2 )

(6.49)

6.5 Subharmonic Resonance of Order One-Third

Choosing the Λ m to eliminate all of the O() terms in (6.49), we have Λ1 D

δ , 2ω 2

Λ2 D

δ , ω2

Λ3 D

δ , 6ω 2

δ , 8ω(ω 2 Ω 2 )2 (ω 2Ω ) δ δ , Λ6 D , Λ5 D 2 2 4ω (ω Ω 2 )2 8ω(ω 2 Ω 2 )2 (ω C 2Ω ) δ δ , Λ8 D , Λ7 D 2 2 2ωΩ (ω Ω ) 2ωΩ (ω 2 Ω 2 ) δ δ , Λ 10 D Λ9 D 2ω(ω 2 Ω 2 )(2ω Ω ) 2ω(ω 2 Ω 2 )(2ω C Ω ) (6.50)

Λ4 D

It is clear from (6.50) that, in addition to the small-divisor term that occurs when Ω ω (primary resonance), there are small-divisor terms when Ω 2ω (subharmonic resonance of order one-half) and Ω 1/2ω (superharmonic resonance of order two). Thus, the terms involving z ξN and z 2 are near-resonance terms. These are the cases treated in the preceding two sections. Substituting (6.46) into (6.26) and using (6.47) and (6.50) yields " 2 2 1 2 N2 z C zN Pξ D i ωξ C i δ ξ C ξN C ξ C ξ 6ξ ξN ω 2(ω 2 Ω 2 ) 3ω 2 z zN z 2 C zN 2 Ω 2 )2 (ω 2 4Ω 2 ) 2ω 2 (ω 2 Ω 2 )2 # ξ z C ξN zN ξ zN C ξN z C Ω (2ω C Ω )(ω 2 Ω 2 ) Ω (2ω Ω )(ω 2 Ω 2 ) " Ω (z z) N i 2 2i µ ω ξ ξN C C 2ω 2ω(ω 2 Ω 2 ) 3 # z C z N C α ξ C ξN C C 2(ω 2 Ω 2 )

4(ω 2

(6.51) Next, we introduce the near-identity transformation ξ D w C 2 h(w, wN , z, zN )

(6.52)

to eliminate the nonresonance terms from (6.51). The remainder depends on the type of resonance being considered. Because we are stopping at this order, we do not need to determine h. In the case of subharmonic resonance of order one-third, the excitation produces the near-resonance term z wN 2 . Keeping this term as well as

183

184

6 Forced Oscillations of SDOF Systems

the resonance terms in the transformed equation (6.51), we obtain 10δ 2 i 2 3α w 2 wN 2ω 3ω 2 3 1 1 i 2 2 z zN w C C α δ ω(ω 2 Ω 2 )2 4 2ω 2 4ω 2 Ω 2 3 1 i 2 1 2 z wN 2 C α C δ ω(ω 2 Ω 2 ) 4 6ω 2 Ω (2ω Ω )

wP D i ωw 2 µ w C

(6.53)

In the case of superharmonic resonance of order three, the excitation produces the near-resonance term z 3 . Then, keeping this term as well as the resonance terms in the transformed equation (6.51), we obtain i 2 10δ 2 w 2 wN 3α 2ω 3ω 2 3 1 1 i 2 2 z zN w C C α δ ω(ω 2 Ω 2 )2 4 2ω 2 4ω 2 Ω 2 2δ 2 i 2 α z3 C 16ω(ω 2 Ω 2 )3 ω 2 4Ω 2

wP D i ωw 2 µ w C

(6.54)

Substituting the transformations (6.52), (6.46), and (6.22) into (6.3), using the expressions (6.25), (6.47), and (6.50), and recalling that z D O(1), we obtain δ 2 z C zN w C wN 2 6w wN C u D w C wN C 2 2 2 2(ω Ω ) 3ω z 2 C zN 2 z zN 4(ω 2 Ω 2 )2 (ω 2 4Ω 2 ) 2ω 2 (ω 2 Ω 2 )2 w z C wN zN w zN C wN z C C Ω (2ω C Ω )(ω 2 Ω 2 ) Ω (2ω Ω )(ω 2 Ω 2 )

(6.55)

Substituting (6.4) and the polar form (6.42) into (6.55) yields ( F F2 u D a cos (ωt C β) C 2 cos Ω t C 2 2 ω Ω 2ω (ω 2 Ω 2 )2 F 2 cos 2Ω t δ a2 [cos (2ωt C 2β) 3] C 2(ω 2 Ω 2 )2 (ω 2 4Ω 2 ) 6ω 2

) Fa cos (Ω C ω)t C β Fa cos (Ω ω)t β C C Ω (2ω C Ω )(ω 2 Ω 2 ) Ω (2ω Ω )(ω 2 Ω 2 )

(6.56)

6.5 Subharmonic Resonance of Order One-Third

Substituting (6.4) and (6.42) into (6.53) and separating real and imaginary parts, we have 3 1 2 1 2 aP D 2 µ a C α C δ 2ω(ω 2 Ω 2 ) 4 6ω 2 Ω (2ω Ω ) Fa 2 sin [3β (Ω 3ω) t] a βP D

(6.57)

2 F 2 3 1 1 2 a C αδ ω(ω 2 Ω 2 )2 4 2ω 2 4ω 2 Ω 2 3α 2 5δ 2 a3 C C 2 3 8ω 12ω 2ω(ω 2 Ω 2 ) 3 1 1 Fa 2 cos [3β (Ω 3ω) t] α C δ2 4 6ω 2 Ω (2ω Ω ) (6.58)

for the subharmonic case. Equations 6.56–6.58 are in full agreement with those obtained by using the method of multiple scales (Nayfeh, 1986). In the case of superharmonic resonance of order three, substituting (6.4) and (6.42) into (6.54) and separating real and imaginary parts, we obtain aP D 2 µ a C

2 F 3 8ω(ω 2 Ω 2 )3

α

2δ 2 ω 2 4Ω 2

sin [β (3Ω ω) t] (6.59)

a βP D

3 1 1 2 F 2 a C α δ2 ω(ω 2 Ω 2 )2 4 2ω 2 4ω 2 Ω 2 3α 5δ 2 a3 C 2 8ω 12ω 3 2δ 2 2 F 3 α 2 cos [β (3Ω ω) t] C 8ω(ω 2 Ω 2 )3 ω 4Ω 2

(6.60)

Equations 6.56, 6.59, and 6.60 are in full agreement with those obtained by using the method of multiple scales (Nayfeh, 1986).

185

187

7 Parametrically Excited Systems In the preceding two chapters, the excitation was taken to be external or sometimes called additive. In this chapter, we treat the case in which the excitation is parametric or sometimes called multiplicative. We start with the Mathieu equation and then progress to linear multiple-degree-of-freedom systems and ﬁnally include the effect of nonlinearities.

7.1 The Mathieu Equation

In this section, we determine approximations to the solutions of the Mathieu equation uR C ω 2 u C 2µ uP C 2 u cos Ω t D 0

(7.1)

where is a small nondimensional parameter and µ, ω, and Ω are constants. As in the preceding chapters, we ﬁrst cast (7.1) into a complex-valued form by introducing the transformation u D ζ C ζN and uP D i ω ζ ζN (7.2) and 2 cos Ω t D z C zN ,

z D eiΩ t

(7.3)

so that zP D i Ω z

(7.4)

With this transformation, (7.1) becomes i ζP D i ωζ µ ζ ζN C ζ C ζN (z C z) N 2ω

(7.5)

We note that the transformation (7.2) is not valid when ω 0. This case is treated in Section 2.8. The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

188

7 Parametrically Excited Systems

To determine a second-order uniform expansion of and hence determine a normal form of (7.5), we let ζ D η C h 1 (η, η, N z, z) N C 2 h 2 (η, η, N z, zN ) C

(7.6)

N z, zN ) C 2 g 2 (η, η, N z, zN ) C ηP D i ωη C g 1 (η, η,

(7.7)

where

Substituting (7.6) and (7.7) into (7.5) and equating coefﬁcients of like powers of , we obtain i (η C η) N (z C zN ) (7.8) 2ω @h 1 @h 1 i g1 gN 1 µ h 1 hN 1 C h 1 C hN 1 (z C zN ) (7.9) g 2 C L(h 2 ) D @η @ ηN 2ω g 1 C L(h 1 ) D µ (η η) N C

where the operator L is deﬁned by @ @ @ @ L D iω η ηN 1 C iΩ z zN @η @ ηN @z @ zN

(7.10)

Next, we consider two cases: fundamental parametric resonance (i.e., Ω ω) and principal parametric resonance (i.e., Ω 2ω). 7.1.1 Fundamental Parametric Resonance

We start by choosing g 1 to eliminate the resonance and near-resonance terms in (7.8). Because Ω ω, only the term µ η is a resonance term and there are no near-resonance terms. Therefore, we choose g 1 to eliminate this resonance term; that is, g 1 D µ η

(7.11)

Then, we choose h 1 to eliminate all of the nonresonance terms in (7.8). The righthand side of (7.8) suggests seeking h 1 in the form h 1 D Γ1 ηN C Γ2 η z C Γ3 ηN zN C Γ4 η zN C Γ5 ηN z

(7.12)

Substituting (7.12) into (7.8) and using (7.11) yields 1 1 (2i ωΓ1 C µ) ηN C i Ω Γ2 C η z C i Ω Γ4 C η zN 2ω 2ω 1 1 ηN zN C i (2ω Ω ) Γ5 C ηN z D 0 C i (2ω C Ω ) Γ3 C 2ω 2ω (7.13)

7.1 The Mathieu Equation

Equating each of the coefﬁcients of η, N η z, ηN z, N η z, N and ηz N to zero, we obtain iµ 1 1 , Γ2 D , Γ3 D , 2ω 2ωΩ 2ω(Ω C 2ω) 1 1 , Γ5 D Γ4 D 2ωΩ 2ω(Ω 2ω) Γ1 D

(7.14)

Substituting (7.11) and (7.12) into (7.9) and using (7.14), we have

iµ Ω ηz (Ω 2ω) ηN zN Ω η zN η C 2 2 2 2ω 4ω (Ω C 2ω) 4ω Ω 4ω (Ω 2ω) i η z C ηN zN (Ω C 2ω) ηN z η zN C ηz N (z C z) C N C C (7.15) 4ω 2 Ω 2ω Ω (Ω C 2ω) Ω (Ω 2ω)

g 2 C L(h 2 ) D µ

We note that the terms proportional to η and η z zN are resonance terms and the term proportional to ηz N 2 is a near-resonance term because Ω ω. Hence, choosing g 2 to eliminate these resonance and near-resonance terms, we have 2η z zN i µ2 z 2 ηN i (7.16) C ηC g2 D 2ω 2ω Ω 2 4ω 2 Ω (Ω 2ω) Substituting for g 1 and g 2 from (7.11) and (7.16) into (7.7), we obtain the normal form 2η z zN i 2 µ 2 z 2 ηN i 2 (7.17) ηP D i ωη µ η C ηC 2ω 2ω Ω 2 4ω 2 Ω (Ω 2ω) Substituting (7.6) and (7.12) into (7.2) and using (7.14), we obtain η z C ηN zN η zN C ηz N iµ η ηN C C C (7.18) u D η C ηN C 2ω Ω (Ω C 2ω) Ω (Ω 2ω) Expressing η in the polar form ηD

1 i (Ω tCβ ) a 2

(7.19)

and using (7.3), we rewrite (7.18) as a cos(2Ω t C β) µ a sin(Ω t C β) C u D a cos (Ω t C β) C 2ω Ω (Ω C 2ω) a cos(β) C C Ω (Ω 2ω)

(7.20)

Substituting (7.3) and (7.19) into (7.17) and separating real and imaginary parts, we obtain aP D µ a C

2 a sin(2β) 2ωΩ (Ω 2ω)

a βP D (Ω ω) a

2 a 2 a cos(2β) 2 µ2 aC C 2ω ω(Ω 2 4ω 2 ) 2ωΩ (Ω 2ω)

(7.21) (7.22)

189

190

7 Parametrically Excited Systems

7.1.2 Principal Parametric Resonance

Because Ω 2ω in this case, z ηN is a near-resonance term. Consequently, choosing g 1 to eliminate the resonance and near-resonance terms in (7.8), we obtain g 1 D µ η C

i ηz N 2ω

(7.23)

Then, we seek the solution of (7.8) in the form h 1 D Γ1 ηN C Γ2 η z C Γ3 ηN zN C Γ4 η zN

(7.24)

Following steps similar to those used in the preceding section, we ﬁnd that the Γi for i D 1, 2, 3, 4 are given by (7.14). Substituting (7.23) and (7.24) into (7.9), using (7.14), and choosing g 2 to eliminate the resonance and near-resonance terms, we obtain g2 D

i µ2 i µ(Ω C 2ω) η η z zN z ηN 2ω 4ω 2 (Ω C 2ω) 4ω 2 Ω

(7.25)

Substituting (7.23) and (7.25) into (7.7) yields the normal form 2 iµ η i z zN η µ(Ω C2ω)z ηN i z ηN 2 C C ηP D i ωη µ η 2ω 2ω 4ω 2 (Ω C 2ω) 4ω 2 Ω (7.26) Substituting (7.6) into (7.12) and using (7.24) and (7.14), we have iµ η z C ηN zN η zN C ηz N (η η) u D η C ηN C N C C 2ω Ω (Ω C 2ω) 2Ω ω

(7.27)

Because z D e i Ω t , the form of the near-resonance term in (7.26) suggests the following polar form for η: ηD

1 i ( 1 Ω tCβ ) ae 2 2

(7.28)

We note that this choice produces a set of autonomous equations for a and β as shown below. Substituting (7.3) and (7.28) into (7.27) yields " cos 32 Ω t C β µ sin 12 Ω t C β 1 u D a cos Ω t C β C a C 2 Ω (Ω C 2ω) 2ω 1 # cos 2 Ω t β C (7.29) 2ωΩ

7.2 Multiple-Degree-of-Freedom Systems

Substituting (7.3) and (7.28) into (7.26) and separating real and imaginary parts, we obtain 2 µ(Ω C 2ω) a cos(2β) a sin 2β 2ω 4Ω ω 2 1 βP D (2ω Ω ) C cos 2β 2 2ω 1 µ(Ω C 2ω) 2 2 µ C sin(2β) 2ω 2ω(Ω C 2ω) 2ωΩ

aP D µ a C

(7.30)

(7.31)

7.2 Multiple-Degree-of-Freedom Systems

In this section, we consider parametrically excited, nongyroscopic, multiple-degreeof-freedom systems governed by xR C Ax C D xP C (2 cos Ω t) F x D 0

(7.32)

where A, D, and F are n n constant matrices and x is a column vector of length n. We assume that none of the eigenvalues of A is zero; this case is considered in Chapter 2. Then, we introduce the nonsingular linear transformation x D P u into (7.32) and obtain P uR C AP u C D P uP C (2 cos Ω t) F P u D 0

(7.33)

Multiplication of (7.33) from the left by P 1 , the inverse of P, yields uR C J u C DO uP C (2 cos Ω t) FO u D 0

(7.34)

where J D P 1 AP ,

DO D P 1 D P ,

and

FO D P 1 F P

(7.35)

One can always choose P so that J is a Jordan canonical form. In this section, we treat the case in which the eigenvalues of A are distinct and positive so that J is a diagonal matrix. We assume that DO is a diagonal matrix (the so-called modaldamping assumption). Moreover, we limit our discussion to ﬁrst-order expansions of the solutions of three-degree-of-freedom systems modeled by uR 1 C ω 21 u 1 C 2µ 1 uP 1 C (2 cos Ω t) uR 2 C ω 22 u 2 C 2µ 2 uP 2 C (2 cos Ω t) uR 3 C ω 23 u 3 C 2µ 3 uP 3 C (2 cos Ω t)

3 X nD1 3 X nD1 3 X

f 1n u n D 0

(7.36)

f 2n u n D 0

(7.37)

f 3n u n D 0

(7.38)

nD1

where the ω m have been arranged so that ω 3 > ω 2 > ω 1 .

191

192

7 Parametrically Excited Systems

As a ﬁrst step in determining uniform expansions of the solutions of (7.36)– (7.38), we recast them into a system of three ﬁrst-order complex-valued equations. To accomplish this, we note that when D 0 the solutions of (7.36)–(7.38) can be expressed as u n D A n e i ω n t C AN n e i ω n t

(7.39a)

where the A n are complex. Hence, uP n D i ω n A n e i ω n t AN n e i ω n t

(7.39b)

When ¤ 0, we continue to represent u n and uP n as in (7.39a) and (7.39b) but with time-varying rather than constant A n . Then, we identify A n e i ω n t with ζ n and rewrite (7.39a) and (7.39b) as u m D ζ m C ζN m , uP m D i ω m ζ m ζN m (7.39c) Moreover, we let z D eiΩ t

and

zP D i Ω z

(7.40)

With this transformation, (7.36)–(7.38) become 3 X i (z C z) ζP 1 D i ω 1 ζ1 µ 1 ζ1 ζN 1 C N f 1n ζ n C ζN n 2ω 1 nD1

(7.41)

3 X i (z C z) ζP 2 D i ω 2 ζ2 µ 2 ζ2 ζN 2 C N f 2n ζ n C ζN n 2ω 2 nD1

(7.42)

3 X i (z C z) ζP 3 D i ω 3 ζ3 µ 3 ζ3 ζN 3 C N f 3n ζ n C ζN n 2ω 3 nD1

(7.43)

To simplify (7.41)–(7.43), we introduce the near-identity transformation ζ m D η m C h m (η n , ηN n , z, z) N

(7.44)

and obtain ηP m D i ω m η m C iω m h m µ m (η m ηN m )

3 X @h m nD1

@η n

ηP n C

@h m P ηN n @ ηN n

@h m P i @h m (z C z) N f m1 (η 1 C ηN 1 ) C f m2 (η 2 C ηN 2 ) zP zN C @z @ zN 2ω m

(7.45) C f m3 (η 3 C ηN 3 ) C

for m D 1, 2, and 3. The form of the O() terms suggests the following form for the h m : h m D ∆ m1 η m C ∆ m2 ηN m C Γm1 z η 1 C Γm2 z ηN 1 C Γm3 z η 2 C Γm4 z ηN 2 C Γm5 z η 3 C Γm6 z ηN 3 C Γm7 zN η 1 C Γm8 zN ηN 1 C Γm9 zN η 2 C Γm10 zN ηN 2 C Γm11 zN η 3 C Γm12 zN ηN 3

(7.46)

7.2 Multiple-Degree-of-Freedom Systems

It follows from (7.45) that ηP m D i ω m η m C O()

(7.47)

Hence, substituting (7.40), (7.46), and (7.47) into the right-hand side of (7.45), we obtain ηP m D i ω m η m µ m η m C (2i ω m ∆ m2 C µ m ) ηN m f m1 z η1 i (Ω C ω 1 ω m ) Γm1 2ω m f m1 z ηN 1 i (Ω ω 1 ω m ) Γm2 2ω m f m2 z η2 i (Ω C ω 2 ω m ) Γm3 2ω m f m2 z ηN 2 i (Ω ω 2 ω m ) Γm4 2ω m f m3 z η3 i (Ω C ω 3 ω m ) Γm5 2ω m f m3 z ηN 3 i (Ω ω 3 ω m ) Γm6 2ω m f m1 zN η 1 C i (Ω ω 1 C ω m ) Γm7 2ω m f m1 zN ηN 1 C i (Ω C ω 1 C ω m ) Γm8 C 2ω m f m2 zN η 2 C i (Ω ω 2 C ω m ) Γm9 C 2ω m f m2 zN ηN 2 C i (Ω C ω 2 C ω m ) Γm10 C 2ω m f m3 zN η 2 C i (Ω ω 3 C ω m ) Γm11 C 2ω m f m3 zN ηN 3 C i (Ω C ω 3 C ω m ) Γm12 C 2ω m C

(7.48)

for m D 1, 2, and 3. Substituting (7.44) into (7.39c), we obtain to the ﬁrst approximation u m D η m C ηN m C O()

(7.49)

We note that (7.48) does not depend on the ∆ m1, and hence they are arbitrary. Choosing the Γm n to eliminate the terms involving z η m , z ηN m , zN η m , and zN ηN m , we ﬁnd that some of the Γm n have small-divisor terms when Ω ωm C ωn

for

m, n D 1, 2, and 3

Ω ωm ωn

for

m, n D 1, 2, and 3 but

m¤n

193

194

7 Parametrically Excited Systems

The case Ω 2ω m , m D 1, 2, or 3, is called principal parametric resonance of the mth mode, which is treated in the preceding section. The case Ω ω m C ω n , for m ¤ n, is called combination parametric resonance of the additive type and the case Ω ω m ω n , for m ¤ n, is called combination parametric resonance of the difference type. The cases (a) Ω 2ω s and Ω ω n ˙ ω m and (b) Ω ω s and Ω ω n ˙ ω m are called simultaneous parametric resonances. Next, we treat some of these cases. 7.2.1 The Case of Ω Near ω 2 C ω 1

When Ω ω 2 C ω 1 and there are no other resonances, the term involving z ηN 2 is near-resonance when m D 1, while the term involving z ηN 1 is near-resonance when m D 2, and there are no near-resonance terms when m D 3. Consequently, choosing the ∆ m2 and Γm n to eliminate all nonresonance terms in (7.48), we obtain i f 12 z ηN 2 C O 2 2ω 1 i f 21 ηP 2 D i ω 2 η 2 µ 2 η 2 C z ηN 1 C O 2 2ω 2 ηP 3 D i ω 3 η 3 µ 3 η 3 C O 2

ηP 1 D i ω 1 η 1 µ 1 η 1 C

(7.50) (7.51) (7.52)

We note that Γ14 and Γ22 are arbitrary. Moreover, (7.50)–(7.52) show that, although the ﬁrst and second modes are coupled, the third mode is uncoupled from the ﬁrst two modes. The polar form of (7.50)–(7.52) is in full agreement with that obtained by using the method of multiple scales (Nayfeh and Mook, 1979). 7.2.2 The Case of Ω Near ω 2 ω 1

In this case, η 2 zN and η 1 z are near-resonance terms when m D 1 and 2, respectively, and there are no resonance terms when m D 3. Consequently, to O(), the normal form of (7.48) is i f 12 zN η 2 2ω 1 i f 21 z η1 ηP 2 D i ω 2 η 2 µ 2 η 1 C 2ω 2

ηP 1 D i ω 1 η 1 µ 1 η 1 C

ηP 3 D i ω 3 η 3 µ 3 η 3

(7.53) (7.54) (7.55)

7.2.3 The Case of Ω Near ω 2 C ω 1 and ω 3 ω 2

In this case, z ηN 2 is a near-resonance term when m D 1, z ηN 1 and η 3 zN are nearresonance terms when m D 2, and z η 2 is a near-resonance term when m D 3.

7.3 Linear Systems Having Repeated Frequencies

Consequently, to O(), the normal form of (7.48) is i f 12 z ηN 2 2ω 1 i f 21 i f 23 z ηN 1 C zN η 3 ηP 2 D i ω 2 η 2 µ 2 η 2 C 2ω 2 2ω 2 i f 32 z η2 ηP 3 D i ω 3 η 3 µ 3 η 3 C 2ω 3

ηP 1 D i ω 1 η 1 µ 1 η 1 C

(7.56) (7.57) (7.58)

7.2.4 The Case of Ω Near 2ω 3 and ω 2 C ω 1

In this case, the near-resonance terms are z ηN 2 when m D 1, z ηN 1 when m D 2, and z ηN 3 when m D 3. Consequently, the normal form of (7.48) is i f 12 z ηN 2 2ω 1 i f 21 z ηN 1 ηP 2 D i ω 2 η 2 µ 2 η 2 C 2ω 2 i f 33 z ηN 3 ηP 3 D i ω 3 η 3 µ 3 η 3 C 2ω 3

ηP 1 D i ω 1 η 1 µ 1 η 1 C

(7.59) (7.60) (7.61)

We note that η 3 is uncoupled from η 1 and η 2 .

7.3 Linear Systems Having Repeated Frequencies

In contrast with the preceding section, here we consider a three-degree-of-freedom system having two repeated frequencies and the Jordan form 2

ω 21 4 JD 1 0

0 ω 21 0

3 0 05 ω 23

(7.62)

Thus, we consider uR 1 C ω 21 u 1 C 2µ 1 uP 1 C (2 cos Ω t)

3 X

f 1n u n D 0

(7.63)

nD1

uR 2 C ω 21 u 2 C u 1 C 2µ 2 uP 2 C (2 cos Ω t)

3 X

f 2n u n D 0

(7.64)

nD1

uR 3 C ω 23 u 3 C 2µ 3 uP 3 C (2 cos Ω t)

3 X nD1

We assume that ω 3 > ω 1 .

f 3n u n D 0

(7.65)

195

196

7 Parametrically Excited Systems

In the absence of damping and the parametric excitation, the system is unstable (the system is said to be in ﬂutter). To show this, we note that when µ n D 0 and f m n D 0, u 1 D a 1 cos (ω 1 t C β 1 ) u 3 D a 3 cos (ω 3 t C β 3 ) where a 1 , a 3 , β 1 and β 3 are constants. Then (7.64) becomes uR 2 C ω 21 u 2 D u 1 D a 1 cos (ω 1 t C β 1 ) Hence, u 2 D a 2 cos (ω 2 t C β 2 )

a1 t sin (ω 1 t C β 1 ) 2ω 1

which contains a secular term or resonance term. Consequently, one refers to this one-to-one resonance as a nonsemisimple one-to-one resonance and the linear operator J is said to have a generic nonsemisimple structure. We wish to determine the normal forms of (7.63)–(7.65) in the presence of damping and the parametric excitation so that one can use these forms to ascertain if the damping and parametric excitation can stabilize the system. In the stable case, one assumes that all of the three variables u 1 , u 2 , and u 3 are bounded and, if possible, determine the values of the parameters which are consistent with this assumption. Although the damping and parametric excitation might stabilize the system, we still expect the amplitude of u 2 to be much larger than those of u 1 and u 3 . We use this observation to simplify the obtained normal forms. As a ﬁrst step in the application of the method of normal forms, we recast (7.63)– (7.65) in complex-valued form by using the transformation (7.39c) and (7.40). The result is 3 X i (z C z) ζP 1 D i ω 1 ζ1 µ 1 ζ1 ζN 1 C N f 1n ζ n C ζN n 2ω 1 nD1 i ζ1 C ζN 1 µ 2 ζ2 ζN 2 ζP 2 D i ω 1 ζ2 C 2ω 1 3 X i (z C zN ) f 2n ζ n C ζN n C 2ω 1 nD1 3 X i (z C z) ζP 3 D i ω 3 ζ3 µ 3 ζ3 ζN 3 C N f 3n ζ n C ζN n 2ω 3 nD1

(7.66)

(7.67) (7.68)

First, we introduce a transformation to simplify the ﬁrst-order problem. We note that (i ζ1 )/(2ω 1 ) is a resonance term in (7.67) and hence it cannot be eliminated by any transformation. This leaves the term (i ζN 1 )/(2ω 1 ), which can be eliminated by the transformation ζ1 D η 1 ,

ζ2 D η 2 C ∆ 1 ηN 1 ,

ζ3 D η 3

(7.69)

7.3 Linear Systems Having Repeated Frequencies

Substituting (7.69) into the O(1) terms in (7.66) and (7.67) yields ηP 1 D i ω 1 η 1 C O()

(7.70)

ηP 2 D i ω 1 η 2 C i ω 1 ∆ 1 ηN 1 ∆ 1 ηPN 1 C

i (η 1 C ηN 1 ) C O() 2ω 1

(7.71)

Using (7.70) in (7.71), we have i i ηN 1 C O() η 1 C 2i ω 1 ∆ 1 C ηP 2 D i ω 1 η 2 C 2ω 1 2ω 1

(7.72)

Then, choosing ∆ 1 to eliminate ηN 1 from (7.72) yields ∆1 D

1 4ω 21

(7.73)

Substituting for ζ2 from (7.69) into (7.39c) and using (7.73), we have u 2 D η 2 C ηN 2

1 (η 1 C ηN 1 ) C O() 4ω 21

(7.74)

Substituting (7.69) into (7.66)–(7.68) yields ηP 1 D i ω 1 η 1 µ 1 (η 1 ηN 1 ) C

i f 12 (z C z)(η N 1 C ηN 1 ) 8ω 31

ηP 2 D i ω 1 η 2 C

(7.75)

3 X i i (z C z) N η 1 µ 2 (η 2 ηN 2 ) C f 2n (η n C ηN n ) 2ω 1 2ω 1 nD1

i f 22 (z C z)(η N 1 C ηN 1 ) 8ω 31

ηP 3 D i ω 3 η 3 µ 3 (η 3 ηN 3 ) C

3 X i (z C zN ) f 1n (η n C ηN n ) 2ω 1 nD1

(7.76)

3 X i (z C zN ) f 3n (η n C ηN n ) 2ω 3 nD1

i f 32 (z C zN )(η 1 C ηN 1 ) 8ω 21 ω 3

(7.77)

To simplify (7.75)–(7.77), we introduce the near-identity transformation η m D ξm C h m ξn , ξN n , z, zN

(7.78)

and choose the h m to eliminate the nonresonance terms. The result depends on the resonance conditions. Three of these conditions, namely Ω 2ω 1 , Ω ω 3 C ω 1 , and Ω ω 3 ω 1 , are discussed next. In Section 7.3.4, we carry out the transformation to second order for the case Ω ω 1 .

197

198

7 Parametrically Excited Systems

7.3.1 The Case of Ω Near 2ω 1

When Ω 2ω 1 and there are no other resonances, the terms z ηN 1 and z ηN 2 are near-resonance terms in both (7.75) and (7.76). Hence, the normal form of (7.75)– (7.77) is i i ξP1 D i ω 1 ξ1 µ 1 ξ1 C f 11 z ξN1 C f 12 z ξN2 f 12 z ξN1 2ω 1 8ω 31 i i ξP2 D i ω 1 ξ2 C f 21 z ξN1 C f 22 z ξN2 ξ1 µ 2 ξ2 C 2ω 1 2ω 1 i f 22 z ξN1 8ω 31 ξP3 D i ω 3 ξ3 µ 3 ξ3

(7.79)

(7.80) (7.81)

As noted earlier, (7.79) and (7.80) can be further simpliﬁed because u 2 is much larger than u 1 and hence ξ2 is much larger than ξ1 . To accomplish this simpliﬁcation, we scale the damping coefﬁcients µ 1 and µ 2 and the variables ξ1 and ξ2 as ξ1 D χ 1 ,

ξ2 D λ 2 χ 2 ,

µ n D λ 1 µO n

where χ 1 and χ 2 are O(1) and λ 1 and λ 2 are positive constants to be determined from the analysis. Because (7.79) and (7.80) are linear and homogeneous, we ordered ξ1 at O() without loss of generality. Using these scalings, we rewrite (7.79) and (7.80) as i χP 1 D i ω 1 χ 1 1λ 1 µO 1 χ 1 C f 11 z χN 1 C 1λ 2 f 12 z χN 2 2ω 1 i f 12 z χN 1 (7.82) 8ω 31 i λ2 i 1Cλ 2 χP 2 D i ω 1 χ 2 C χ 1 1λ 1 µO 2 χ 2 C f 21 z χN 1 C f 22 z χN 2 2ω 1 2ω 1 i 1Cλ 2 f 22 z χN 1 (7.83) 8ω 31 As ! 0, we note that f 11 z χO 1 is small compared with 1λ 2 f 12 z χN 2 because λ 2 > 0. Then, requiring the damping term in (7.82) be the same order as the resonance term i(2ω 1 )1 1λ 2 f 12 z χN 2 , we have 1 λ1 D 1 λ2

or

λ1 D λ2

(7.84)

Similarly, as ! 0, the resonance terms (the terms inside the parenthesis) in (7.83) are small compared with the damping term 1λ 1 µO 2 χ 2 because λ 1 and λ 2 are positive. Then, requiring i(2ω 1 )1 λ 2 χ 1 to be the same order as 1λ 1 µO 2 χ 2 , we have λ2 D 1 λ1

(7.85)

7.3 Linear Systems Having Repeated Frequencies

Hence, λ 1 D λ 2 D 1/2. Consequently, keeping terms up to O( 1/2 ) in (7.82) and (7.83), we obtain the simpliﬁed normal form χP 1 D i ω 1 χ 1 1/2 µO 1 χ 1 C

i 1/2 f 12 z χN 2 C 2ω 1

(7.86)

χP 2 D i ω 2 χ 2 1/2 µO 2 χ 2 C

i 1/2 χ1 C 2ω 1

(7.87)

Equations 7.86 and 7.87 are also called the distinguished limit or least degenerate form of (7.82) and (7.83). We note that (7.86) and (7.87) agree with those obtained by Nayfeh and Mook (1979) by using the method of multiple scales. 7.3.2 The Case of Ω Near ω 3 C ω 1

In this case, the term z ηN 3 is near-resonance in (7.75) and (7.76), while the terms z ηN 1 and z ηN 2 are near-resonance in (7.77). Hence, the normal form of (7.75)–(7.77) is i ξP1 D i ω 1 ξ1 µ 1 ξ1 C f 13 z ξN3 2ω 1 i i ξP2 D i ω 1 ξ2 C ξ1 µ 2 ξ2 C f 23 z ξN3 2ω 1 2ω 1 i i ξP3 D i ω 3 ξ3 µ 3 ξ3 C f 31 z ξN1 C f 32 z ξN2 f 32 z ξN1 2ω 3 8ω 21 ω 3

(7.88) (7.89) (7.90)

Again, using the fact that ξ2 is much larger than ξ1 and ξ3 , we further simplify (7.88)–(7.90). To accomplish this, we scale the ξn and µ n as ξ1 D χ 1 ,

ξ2 D λ 2 χ 2 ,

ξ3 D λ 3 χ 3 ,

µ n D λ 1 µO n

(7.91)

where the χ n and µO n are O(1) and the λ n are positive constants. Substituting (7.91) into (7.88)–(7.90), we obtain i 1λ 3 f 13 z χN 3 2ω 1 i λ2 i 1Cλ 2λ 3 χ 1 1λ 1 µO 2 χ 2 C f 23 z χN 3 χP 2 D i ω 1 χ 2 C 2ω 1 2ω 1

χP 1 D i ω 1 χ 1 1λ 1 µO 1 χ 1 C

χ 3 D i ω 3 χ 3 1λ 1 µO 3 χ 3 C

As ! 0, the distinguished limit of (7.92)–(7.94) corresponds to 1 λ1 D λ2 1 λ1 D 1 C λ3 λ2

(7.93)

i 1Cλ 3 i 1Cλ 3λ 2 f 31 z χN 1 C f 32 z χN 2 2ω 3 2ω 3

i 1Cλ 3 f 32 z χN 1 8ω 21 ω 3

1 λ1 D 1 λ3

(7.92)

(7.94)

199

200

7 Parametrically Excited Systems

or λ2 D

2 3

and

λ1 D λ3 D

1 3

Thus, to the ﬁrst approximation, (7.92)–(7.94) simplify to the normal form χP 1 D i ω 1 χ 1 2/3 µO 1 χ 1 C

i 2/3 f 13 z χN 3 C 2ω 1

(7.95)

i 2/3 χ1 C 2ω 1 i 2/3 f 32 z χN 2 C χP 3 D i ω 3 χ 3 2/3 µO 3 χ 3 C 2ω 3 χP 2 D i ω 1 χ 2 2/3 µO 2 χ 2 C

(7.96) (7.97)

Equations 7.95–7.97 agree with those obtained by Nayfeh and Mook (1979) by using the method of multiple scales. 7.3.3 The Case of Ω Near ω 3 ω 1

In this case, the term zN ξ3 is near-resonance in (7.75) and (7.76), while the terms z ξ1 and z ξ2 are near-resonance in (7.77). Hence, the normal form of (7.75)–(7.77) is i ξP1 D i ω 1 ξ1 µ 1 ξ1 C f 13 zN ξ3 (7.98) 2ω 1 i i ξP2 D i ω 1 ξ2 C ξ1 µ 2 ξ2 C f 23 zN ξ3 (7.99) 2ω 1 2ω 1 i i ( f 31 z ξ1 C f 32 z ξ2 ) ξP3 D i ω 3 ξ3 µ 3 ξ3 C f 32 z ξ1 (7.100) 2ω 3 8ω 21 ω 3 As in Section 7.3.2, (7.98)–(7.100) can be further simpliﬁed by using the scaling ξ1 D χ 1 ,

ξ2 D 2/3 χ 2 ,

ξ3 D 1/3 χ 3 ,

µ n D 1/3 µO n

Substituting these scaled expressions into (7.98)–(7.100) and keeping the leading terms in (i.e., up to O( 2/3 )), we obtain the simpliﬁed normal form χP 1 D i ω 1 χ 1 2/3 µO 1 χ 1 C

i 2/3 f 13 zO χ 3 C 2ω 1

(7.101)

χP 2 D i ω 1 χ 2 2/3 µO 2 χ 2 C

i 2/3 χ1 C 2ω 1

(7.102)

χP 3 D i ω 3 χ 3 2/3 µO 3 χ 3 C

i 2/3 f 32 z χ 2 C 2ω 3

(7.103)

7.3.4 The Case of Ω Near ω 1

In this case, instead of ﬁrst determining the normal form of (7.66)–(7.68) and then scaling the dependent variables to simplify the obtained normal forms, we ﬁrst

7.3 Linear Systems Having Repeated Frequencies

scale the dependent variables and then obtain the simpliﬁed normal form, thereby reducing the algebra involved. Because ω 3 is away from ω 1 and hence Ω ω 1 is away from ω 3 , we assume that ζ3 is O(1), which is the same order as ζ1 . However, because ζ2 is large compared with ζ1 , we assume that ζ2 D O( λ 2 ), where λ 2 is positive. Therefore, we let ζ1 D η 1 ,

ζ2 D λ 2 η 2 ,

and

ζ3 D η 3

(7.104)

and rewrite (7.66)–(7.68) as i (z C zN ) ηP 1 D i ω 1 η 1 µ 1 (η 1 ηN 1 ) C 2ω 1 h i f 11 (η 1 C ηN 1 ) C 1λ 2 f 12 (η 2 C ηN 2 ) C f 13 (η 3 C ηN 3 )

(7.105)

i i λ2 (z C z) ηP 2 D i ω 1 η 2 C N (η 1 C ηN 1 ) µ 2 (η 2 ηN 2 ) C 2ω 1 2ω 1 h i 1Cλ 2 f 21 (η 1 C ηN 1 ) C f 22 (η 2 C ηN 2 ) C 1Cλ 2 f 23 (η 3 C ηN 3 ) (7.106) i (z C zN ) ηP 3 D i ω 3 η 3 µ 3 (η 3 ηN 3 ) C 2ω 3 h i f 31 (η 1 C ηN 1 ) C 1λ 2 f 32 (η 2 C ηN 2 ) C f 33 (η 3 C ηN 3 )

(7.107)

Because Ω ω 1 and is away from ω 3 , the term (z C z)(η N 2 C ηN 2 ) is a nonresonance term and hence we choose λ 2 D 1, making the terms i(2ω 1 )1 (z C z) N f 12 (η 2 C ηN 2 ) and i(2ω 3 )1 (z C zN ) f 32 (η 2 C ηN 2 ) in (7.105) and (7.107), respectively, of O(1). Hence, to simplify (7.105)–(7.107), we let N ξ2 , ξN2 C h 11 z, z, N ξn , ξN n C η 1 D ξ1 C h 10 z, z,

(7.108)

η 2 D ξ2 C h 21 z, zN , ξn , ξN n C

(7.109)

η 3 D ξ3 C h 30 z, z, N ξ2 , ξN2 C h 31 z, z, N ξn , ξN n C

(7.110)

N ξn , ξN n C ξP1 D i ω 1 ξ1 C g 1 z, z,

(7.111)

ξP2 D i ω 1 ξ2 C g 2 z, z, N ξn , ξN n C

(7.112)

ξP3 D i ω 3 ξ3 C g 3 z, z, N ξn , ξN n C

(7.113)

where

We have included the terms h 10 and h 30 at O(1) to eliminate the terms of O(1) in (7.105) and (7.107). Substituting (7.108)–(7.113) into (7.105)–(7.107) using (7.40),

201

202

7 Parametrically Excited Systems

and equating the coefﬁcients of 0 and on both sides, we obtain @h 10 @h 10 @h 10 N @h 10 ξ2 h 10 iΩ ξ2 z zN C i ω 1 @z @ zN @ξ2 @ ξN2 i (7.114) D N ξ2 C ξN2 f 12 (z C z) 2ω 1 @h 30 @h 30 @h 30 N @h 30 ξ2 i ω 3 h 30 iΩ ξ2 z zN C i ω 1 @z @ zN @ξ2 @ ξN2 i (7.115) N ξ2 C ξN2 D f 32 (z C z) 2ω 3 X 3 @h 11 @h 11 @h 11 N @h 11 ξn i ω 1 h 11 g1 C i Ω i ωn ξn z zN C @z @ zN @ξn @ ξN n nD1

@h 10 @h 10 D g2 gN 2 µ 1 ξ1 ξN1 C h 10 hN 10 @ξ2 @ ξN2 i f 11 (z C zN ) ξ1 C ξN1 C h 10 C hN 10 C 2ω 1 i C f 12 (z C zN ) h 21 C hN 21 2ω 1 i C f 13 (z C zN ) ξ3 C ξN3 C h 30 C hN 30 2ω 1

(7.116)

X 3 @h 21 @h 21 @h 21 N @h 21 ξn i ω 1 h 21 g2 C i Ω i ωn ξn z zN C @z @ zN @ξn @ ξN n nD1 i i D f 22 (z C zN ) ξ2 C ξN2 ξ1 C ξN1 C h 10 C hN 10 µ 2 ξ2 ξN2 C 2ω 1 2ω 1 (7.117)

X 3 @h 31 @h 31 @h 31 N @h 31 ξn i ω 3 h 31 i ωn ξn z zN C @z @ zN @ξn @ ξN n nD1 @h 30 @h 30 D g2 gN 2 µ 3 ξ3 ξN3 C h 30 hN 30 @ξ2 @ zN i C f 31 (z C zN ) ξ1 C ξN1 C h 10 C hN 10 2ω 3 i C f 32 (z C zN ) h 21 C hN 21 2ω 3 i C f 33 (z C zN ) ξ3 C ξN3 C h 30 C hN 30 (7.118) 2ω 3

g3 C i Ω

where ω 2 D ω 1 . The form of the right-hand sides of (7.114) and (7.115) suggests the following forms for h 10 and h 30 : h n0 D Γn1 z ξ2 C Γn2 z ξN2 C Γn3 zN ξ2 C Γn4 zN ξN2

(7.119)

7.3 Linear Systems Having Repeated Frequencies

for n D 1 and 3. Substituting (7.119) into (7.114) and (7.115) and equating each of the coefﬁcients of z ξ2 , z ξN2 , zN ξ2 , and zN ξN2 on both sides, we obtain f 12 f 12 , Γ12 D , 2ω 1 Ω 2ω 1 (Ω 2ω 1 ) f 12 f 12 , Γ14 D , Γ13 D 2ω 1 Ω 2ω 1 (Ω C 2ω 1 ) f 32 f 32 , Γ32 D , Γ31 D 2ω 3 (Ω C ω 1 ω 3 ) 2ω 3 (Ω ω 1 ω 3 ) f 32 f 32 , Γ34 D Γ33 D 2ω 3 (Ω ω 1 C ω 3 ) 2ω 3 (Ω C ω 1 C ω 3 ) Γ11 D

(7.120)

(7.121)

Substituting (7.119) into (7.117) yields g2 C i Ω D

X 3 @h 21 @h 21 @h 21 N @h 21 ξn i ω 1 h 21 i ωn ξn z zN C @z @ zN @ξn @ ξN n nD1

i i ξ1 C ξN1 µ 2 ξ2 ξN2 C N ξ2 C ξN2 f 22 (z C z) 2ω 1 2ω 1

i (Γ11 C Γ14 ) z ξ2 C zN ξN2 C (Γ12 C Γ13 ) z ξN2 C zN ξ2 C 2ω 1

(7.122)

Equating g 2 to the resonance terms on the right-hand side of (7.122), we have g2 D

i ξ1 µ 2 ξ2 2ω 1

(7.123)

Then, the form of the remaining terms in (7.122) suggests the following form for h 21 : h 21 D Γ21 z ξ2 C Γ22 z ξN2 C Γ23 zN ξ2 C Γ24 zN ξN2 C Γ25 ξN2 C Γ26 ξN1

(7.124)

Substituting (7.124) into (7.122), using (7.123), and equating each of the coefﬁcients of z ξ2 , z ξN2 , zN ξ2 , zN ξN2 , ξN2 , and ξN1 on both sides, we obtain Γ11 C Γ14 C f 22 Γ12 C Γ13 C f 22 , Γ22 D , 2ω 1 Ω 2ω 1 (Ω 2ω 1 ) Γ12 C Γ13 C f 22 , Γ23 D 2ω 1 Ω Γ11 C Γ14 C f 22 i µ2 1 , Γ26 D 2 , Γ25 D Γ24 D 2ω 1 (Ω C 2ω 1 ) 2ω 1 4ω 1 Γ21 D

(7.125)

We note that we needed to explicitly determine h 21 because it is needed to determine g 1 from (7.116).

203

204

7 Parametrically Excited Systems

Substituting (7.119) and (7.124) into (7.116) and (7.118) yields X 3 @h 11 @h 11 @h 11 N @h 11 ξn i ω 1 h 11 i ωn ξn z zN C @z @ zN @ξn @ ξN n nD1

D µ 1 ξ1 ξN1 C (Γ11 Γ14 ) z ξ2 zN ξN2 C (Γ12 Γ13 ) z ξN2 zN ξ2

g1 C i Ω

(Γ11 z C Γ13 z) N g 2 (Γ12 z C Γ14 z) N gN 2 C

i f 11 (z C zN ) Λ 1 2ω 1

i i f 12 (z C zN ) Λ 2 C f 13 (z C zN ) Λ 3 (7.126) 2ω 1 2ω 1 X 3 @h 31 @h 31 N @h 31 @h 31 ξn i ω 3 h 31 i ωn ξn g3 C i Ω z zN C @z @ zN @ξn @ ξN n nD1

D µ 3 ξ3 ξN3 C (Γ31 Γ34 ) z ξ2 zN ξN2 C (Γ32 Γ33 ) z ξN2 zN ξ2 C

(Γ31 z C Γ33 z) N g 2 (Γ32 z C Γ34 z) N gN 2 C C

i f 31 (z C zN ) Λ 1 2ω 3

i i f 32 (z C zN ) Λ 2 C f 33 (z C zN ) Λ 3 2ω 3 2ω 3

(7.127)

where Λ 1 D ξ1 C ξN1 C (Γ11 C Γ14 ) z ξ2 C zN ξN2 C (Γ12 C Γ13 ) z ξN2 C zN ξ2 Λ 2 D (Γ21 C Γ24 ) z ξ2 C zN ξN2 C (Γ22 C Γ23 ) z ξN2 C zN ξ2 1 i µ2 N ξ2 ξ2 ξ1 C ξN1 C 2ω 1 4ω 21 Λ 3 D ξ3 C ξN3 C (Γ31 C Γ34 ) z ξ2 C zN ξN2 C (Γ32 C Γ33 ) z ξN2 C zN ξ2 If we do not want to continue the expansion beyond the terms in (7.108)–(7.110), we do not need to explicitly calculate h 11 and h 31 , and all what we need is to determine g 1 and g 3 by equating them to the resonance and near-resonance terms on the right-hand sides of (7.126) and (7.127). The result is g 1 D µ 1 ξ1 C

i i α 1 z zN ξ2 C α 2 z 2 ξN2 2ω 1 2ω 1

g 3 D µ 3 ξ3

(7.128) (7.129)

where α 1 D f 11 (Γ11 C Γ14 C Γ12 C Γ13 ) C f 12 (Γ21 C Γ24 C Γ22 C Γ23 ) C f 13 (Γ31 C Γ34 C Γ32 C Γ33 ) α 2 D f 11 (Γ12 C Γ13 ) C f 12 (Γ22 C Γ23 ) C f 13 (Γ32 C Γ33 )

(7.130) (7.131)

7.4 Gyroscopic Systems

Substituting for the Γm n from (7.120), (7.121), and (7.125) into (7.130) and (7.131), we obtain 2 1 2 f 12 ( f 11 C f 22 ) f 12 1 α1 D C C Ω 2 (Ω C 2ω 1 )2 (Ω 2ω 1 )2 Ω 2 4ω 21 1 1 (7.132) C f 13 f 32 C (Ω C ω 1 )2 ω 23 (Ω ω 1 )2 ω 23 2 f 12 ( f 11 C f 22 ) f 13 f 32 f 12 α2 D C (7.133) C 2 Ω (Ω 2ω 1 ) Ω (Ω 2ω 1 )2 (Ω ω 1 )2 ω 23 Substituting for g 1 , g 2 , and g 3 from (7.128), (7.123), and (7.129) into (7.111)–(7.113), we obtain i ξP1 D i ω 1 ξ1 µ 1 ξ1 C α 1 z zN ξ2 C α 2 z 2 ξN2 C (7.134) 2ω 1 i ξP2 D i ω 1 ξ2 µ 2 ξ2 C ξ1 C (7.135) 2ω 1 ξP3 D i ω 3 ξ3 µ 3 ξ3

(7.136)

Equations 7.132–7.136 agree with those obtained by Nayfeh and Mook (1979) by using the method of multiple scales.

7.4 Gyroscopic Systems

For simplicity, we consider a two-degree-of-freedom system to illustrate the method of solution. Speciﬁcally, we consider uR 1 C λ 1 uP 2 C α 1 u 1 D F1 D 2 ( f 11 u 1 C f 12 u 2 ) cos Ω t

(7.137)

uR 2 λ 2 uP 1 C α 2 u 2 D F2 D 2 ( f 21 u 1 C f 22 u 2 ) cos Ω t

(7.138)

where , Ω , λ m , α m , and f m n are constants. As a ﬁrst step, we cast (7.137) and (7.138) in complex-valued form. To accomplish this, we ﬁrst write down the solution of the unperturbed problem; that is, u 1 D A 1 e i ω 1 t C AN 1 e i ω 1 t C A 2 e i ω 2 t C AN 2 e i ω 2 t

(7.139)

uP 1 D i ω 1 A 1 e i ω 1 t AN 1 e i ω 1 t C i ω 2 A 2 e i ω 2 t AN 2 e i ω 2 t

(7.140)

u2 D

i(α 1 ω 22 ) i(α 1 ω 21 ) A 1 e i ω 1 t AN 1 e i ω 1 t C A 2 e i ω 2 t AN 2 e i ω 2 t λ1 ω1 λ1 ω2 (7.141)

uP 2 D

α 1 ω 22 α 1 ω 21 A 1 e i ω 1 t C AN 1 e i ω 1 t A 2 e i ω 2 t C AN 2 e i ω 2 t λ1 λ1 (7.142)

205

206

7 Parametrically Excited Systems

where ω 21 and ω 22 are solutions of ω 4 (α 1 C α 2 C λ 1 λ 2 ) ω 2 C α 1 α 2 D 0

(7.143)

Here, we assume that ω 21 and ω 22 are positive and that ω 2 > ω 1 . We associate A 1 e i ω 1 t with ζ1 and A 2 e i ω 2 t with ζ2 in (7.139)–(7.142) to deﬁne the transformation u 1 D ζ1 C ζN 1 C ζ2 C ζN 2

(7.144)

uP 1 D i ω 1 ζ1 ζN 1 C i ω 2 ζ2 ζN 2

(7.145)

i(α 1 ω 22 ) i(α 1 ω 21 ) ζ1 ζN 1 C ζ2 ζN 2 λ1 ω1 λ1 ω2 α 1 ω 22 α 1 ω 21 ζ1 C ζN 1 ζ2 C ζN 2 uP 2 D λ1 λ1

u2 D

(7.146) (7.147)

It follows from (7.144)–(7.147) that

ζ1 C ζN 1 D ω 22 α 1 u 1 λ 1 uP 2

(7.148)

ζ2 C ζN 2 D α 1 ω 21 u 1 C λ 1 uP 2

(7.149)

α 1 ω 22 ω 21 ζ1 ζN 1 D i ω 1 α 1 ω 22 uP 1 λ 1 ω 22 u 2

(7.150)

α 1 ω 22 ω 21 ζ2 ζN 2 D i ω 2 α 1 ω 21 uP 1 λ 1 ω 21 u 2

(7.151)

ω 22 ω 21 ω 22 ω 21

Solving (7.148) and (7.150) for ζ1 yields 2α 1 ω 22 ω 21 ζ1 D i ω 1 α 1 ω 22 uP 1 α 1 λ 1 uP 2 α 1 α 1 ω 22 u 1 i λ 1 ω 1 ω 22 u 2

(7.152)

Solving (7.149) and (7.151) for ζ2 yields 2α 1 ω 22 ω 21 ζ2 D i ω 2 α 1 ω 21 uP 1 C α 1 λ 1 uP 2 C α 1 α 1 ω 21 u 1 C i λ 1 ω 21 ω 2 u 2

(7.153)

Differentiating (7.152) with respect to t and using (7.137) and (7.138) to eliminate uR 1 and uR 2 , we obtain

2α 1 ω 22 ω 21 ζP 1 D i ω 1 α 1 λ 1 uP 2 C α 1 ω 22 u 1

α 1 α 1 C λ 1 λ 2 ω 22 uP 1 λ 1 α 2 u 2 C iω 1 α 1 ω 22 F1 α 1 λ 1 F2 (7.154) It follows from (7.143) that α 1 C λ 1 λ 2 ω 22 D ω 21 α 2

and

α 1 α 2 D ω 21 ω 22

(7.155)

7.4 Gyroscopic Systems

Using (7.155), we rewrite (7.154) as

2α 1 ω 22 ω 21 ζP 1 D i ω 1 α 1 λ 1 uP 2 C α 1 ω 22 u 1

C ω 21 ω 22 α 1 ω 21 uP 1 C λ 1 ω 21 ω 22 u 2 C iω 1 α 1 ω 22 F1 α 1 λ 1 F2

(7.156)

Using (7.144)–(7.147) to eliminate the u m and uP m from (7.156) and rearranging, we obtain iω 1 (α 1 ω 22 ) λ 1 ζP 1 D i ω 1 ζ1 C F1 F2 2α 1 (ω 22 ω 21 ) 2(ω 22 ω 21 )

(7.157)

Differentiating (7.153) and using a procedure similar to that used to determine ζP 1 , we obtain iω 2 (α 1 ω 21 ) λ 1 ζP 2 D i ω 2 ζ2 F1 C F2 2 2 2 2α 1 (ω 2 ω 1 ) 2(ω 2 ω 21 )

(7.158)

Finally, we substitute for F1 and F2 from (7.137) and (7.138) into (7.157) and (7.158), use (7.3), (7.144), and (7.146), and obtain iω 1 (α 1 ω 22 ) λ 1 (z C z) (z C z) ζP 1 D i ω 1 ζ1 C N Λ1 N Λ 2 (7.159) 2 2 2α 1 (ω 2 ω 1 ) 2(ω 22 ω 21 ) iω 2 (α 1 ω 21 ) λ 1 (z C zN ) Λ 1 C (z C z) ζP 2 D i ω 2 ζ2 N Λ 2 (7.160) 2α 1 (ω 22 ω 21 ) 2(ω 22 ω 21 ) where

Λ 1 D f 11 ζ1 C ζN 1 C ζ2 C ζN 2 α 1 ω 22 α 1 ω 21 ζ1 ζN 1 C C i f 12 λ1 ω1 λ1 ω2 Λ 2 D f 21 ζ1 C ζN 1 C ζ2 C ζN 2 α 1 ω 22 α 1 ω 21 ζ1 ζN 1 C C i f 22 λ1 ω1 λ1 ω2

ζ2 ζN 2

ζ2 ζN 2

To simplify (7.159) and (7.160), we introduce the near-identity transformation (7.44) and choose the h m to eliminate the nonresonance terms, thereby leaving the resonance and near-resonance terms. To O(), there are several resonance combinations of Ω , ω 1 , and ω 2 . These include:

Ω Ω Ω Ω

2ω 1 : Principal parametric resonance of the ﬁrst mode, 2ω 2 : Principal parametric resonance of the second mode, ω 2 C ω 1 : Combination parametric resonance of the additive type, ω 2 ω 1 : Combination parametric resonance of the difference type.

Next, we discuss the cases Ω 2ω 1 and Ω ω 2 ω 1 treated by Nayfeh and Mook (1979) using the method of multiple scales.

207

208

7 Parametrically Excited Systems

7.4.1 The Case of Ω Near 2ω 1

When Ω 2ω 1 and there are no other resonances, there are no resonance terms in (7.160) and z ζN 1 is a near-resonance term in (7.159). Therefore, substituting (7.44) into (7.159) and (7.160) and choosing the h m to eliminate the nonresonance terms, we obtain the normal form λ 1 f 21 C λ 2 f 12 ηP 1 D i ω 1 η 1 2(ω 22 ω 21 ) ω 1 (α 1 ω 22 ) f 11 α 1 ω 21 i C f 22 ηN 1 z (7.161) α1 ω1 ηP 2 D i ω 2 η 2

(7.162)

Putting η n D A n e ω n t in (7.161) and (7.162) yields the same equations obtained by using the method of multiple scales (Nayfeh and Mook, 1979). 7.4.2 The Case of Ω Near ω 2 ω 1

In this case, the resonance terms are zN ζ2 and z ζ1 in (7.159) and (7.160), respectively. Therefore, substituting (7.44) into (7.159) and (7.160) and choosing the h m to eliminate the nonresonance terms, we obtain the normal forms ω 1 (α 1 ω 22 )2 f 12 λ 1 f 21 C ηP 1 D i ω 1 η 1 2 2 λ1 α1 ω2 2(ω 2 ω 1 ) f 22 ω 1 f 11 η 2 zN C i α 1 ω 22 (7.163) ω2 α1 ω 2 (α 1 ω 21 )2 f 12 ηP 2 D i ω 2 η 2 C λ 1 f 21 C 2 2 λ1 α1 ω1 2(ω 2 ω 1 ) ω 2 f 11 f 22 η1 z i α 1 ω 21 (7.164) α1 ω1 Putting η m D A m e i ω m t in (7.163) and (7.164) yields the same equations obtained by Nayfeh and Mook (1979) by using the method of multiple scales.

7.5 A Nonlinear Single-Degree-of-Freedom System

Finally, we illustrate the method of normal forms using a parametrically excited single-degree-of-freedom system with quadratic and cubic nonlinearities. Speciﬁcally, we conside uR C ω 2 u C 2µ uP C 2F u cos Ω t C δ u2 C 2 α u3 D 0

(7.165)

7.5 A Nonlinear Single-Degree-of-Freedom System

Using the transformation (7.2) and (7.3), we rewrite (7.165) as iF ζ C ζN (z C zN ) ζP D i ωζ µ ζ ζN C 2ω 2 3 iδ i 2 α N ζCζ C ζ C ζN C (7.166) 2ω 2ω We seek a second-order uniform expansion of the solution of (7.166) in the form (7.6) and (7.7), equate coefﬁcients of like powers of , and obtain iF iδ (η C η) (η C η) (7.167) N (z C zN ) C N 2 2ω 2ω @h 1 @h 1 iF g 2 C L(h 2 ) D g 1 gN 1 µ h 1 hN 1 C h 1 C hN 1 (z C zN ) @η @ ηN 2ω iα iδ (7.168) (η C η) (η C η) N h 1 C hN 1 C N 3 C ω 2ω where the operator L is deﬁned in (7.10). There are two cases, depending on whether Ω 2ω (i.e., principal parametric resonance) or Ω is away from 2ω. We ﬁrst consider the latter case. N C g 1 C L(h 1 ) D µ (η η)

7.5.1 The Case of Ω Away from 2ω

In this case, to ﬁrst order, µ η is a resonance term and there are no near-resonance terms. Thus, we choose g 1 to eliminate the resonance term in (7.167); that is, g 1 D µ η

(7.169)

Then, the form of the remaining terms on the right-hand side of (7.167) suggests choosing h 1 in the form h 1 D Γ1 ηN C Γ2 η z C Γ3 ηN zN C Γ4 η zN C Γ5 ηz N C Γ6 η 2 C Γ7 η ηN C Γ8 ηN 2 (7.170) Substituting (7.170) into (7.167) and using (7.169) and (7.10) yields F F (2i ωΓ1 C µ) ηN i Ω Γ2 η z C i (2ω C Ω ) Γ3 C ηN zN 2ω 2ω F F η zN C i (2ω Ω ) Γ5 C ηz N C i Ω Γ4 C 2ω 2ω δ δ δ 2 i ω Γ6 η η η N C i ω 3Γ ηN 2 C i ω Γ C C 7 8 2ω 2 ω2 2ω 2 D0

(7.171)

Choosing the Γm to eliminate the nonresonance terms, we have Γ1 D

iµ , 2ω

Γ5 D

F , 2ω(Ω 2ω)

Γ2 D

F , 2ωΩ

F F , Γ4 D , 2ω(Ω C 2ω) 2ωΩ δ δ δ , Γ7 D 2 , Γ8 D 2 (7.172) Γ6 D 2 2ω ω 6ω Γ3 D

209

210

7 Parametrically Excited Systems

Substituting (7.172) into (7.170) yields h1 D

i µ ηN Fz η F η zN Fz ηN F zN ηN C 2ω 2ωΩ 2ωΩ 2ω(2ω Ω ) 2ω(2ω C Ω ) C

δ η ηN δ ηN 2 δ η2 2ω 2 ω2 6ω 2

(7.173)

Substituting (7.173) and (7.6) into (7.2), we have iµ F(z η C ηN z) N δ 2 (η 3η ηN C ηN 2 ) C (η η) N C 2ω 3ω 2 Ω (Ω C 2ω) F(η zN C ηz) N C C (7.174) Ω (Ω 2ω)

u D η C ηN

Substituting (7.169) and (7.173) into (7.168) yields 2F 2 η z zN i µ2 η F 2 ηN z 2 i C C 2ω 2ω Ω 2 4ω 2 Ω (Ω 2ω) 10δ 2 2 4 N C 3α η 2 ηN 2 F δ η ηz C Ω 2 4ω 2 ω 3ω 2 2 2 1 2 F δ η C z N C η N z C NRT . C 3ω 2 Ω (Ω 2ω)

g 2 C L(h 2 ) D

(7.175) There are two possible resonances at this order: those corresponding to Ω ω (fundamental parametric resonance) and Ω 3ω (subharmonic resonance of order one-third). When Ω is away from ω and 3ω, choosing g 2 to eliminate the resonance terms in (7.175), we obtain g2 D

i µ2 η i C 2ω 2ω

2F 2 η z zN 10δ 2 2 η C 3α η N Ω 2 4ω 2 3ω 2

(7.176)

Substituting (7.169) and (7.176) into (7.7), we obtain the normal form i 2 F 2 η z zN i 2 µ 2 ηC C i 2 ηP D i ωη µ η 2ω ω(Ω 2 4ω 2 )

3α 5δ 2 2ω 3ω 3

η 2 ηN (7.177)

When Ω ω, choosing g 2 to eliminate the resonance and near-resonance terms from (7.175) and approximating Ω with ω, we obtain 2 2F i µ2 η F 2 2 10F δ 5F δ 2 i η z z N C ηN z C η ηz N C η zN 2 2 2 2ω 2ω 3ω ω 3ω 3ω 2 10δ 2 η 2 ηN 3α (7.178) 3ω 2

g2 D

7.5 A Nonlinear Single-Degree-of-Freedom System

Substituting (7.169) and (7.178) into (7.7), we obtain the normal form F2 10F δ i 2 2F 2 i 2 µ 2 η z zN C 2 ηN z 2 C η ηz N η 2 2ω 2ω 3ω ω 3ω 2 10δ 2 5F δ 2 2 η η z N 3α η N C (7.179) 3ω 2 3ω 2

ηP D i ωη µ η

Putting η D Ae i ω t

and

z D eiΩ t

(7.180)

in (7.179), we obtain the equation found by using the method of multiple scales (Exercise 7.6.2). When Ω 3ω, choosing g 2 to eliminate resonance and near-resonance terms in (7.175) and approximating Ω with 3ω, we obtain g2 D

i µ2 η i C 2ω 2ω

2F 2 10δ 2 Fδ 2 2 η η z z N C 3α η N C η N z 5ω 2 3ω 2 ω2

(7.181)

Substituting (7.169) and (7.181) into (7.7), we obtain the normal form ηP D i ωη

10δ 2 Fδ 2 i 2 2F 2 2 µ 2 η 2 η η z z N C 3α η N C η z C 2ω 2ω 5ω 2 3ω 2 ω2 (7.182)

Substituting (7.180) into (7.182), we obtain the equation found by using the method of multiple scales (Exercise 7.6.2). 7.5.2 The Case of Ω Near 2ω

In this case, it follows from (7.172) that Γ5 has a small-divisor term and hence z ηN is near-resonance in (7.167). Choosing g 1 to eliminate the resonance and nearresonance terms in (7.167), we obtain g 1 D µ η C

iF ηz N 2ω

(7.183)

Then, we can choose all of the Γm , m D 1, 2, . . . , 8, except Γ5 , as in (7.172) to eliminate the nonresonance terms. Consequently, h1 D

i µ ηN δ ηN 2 Fz η F η zN F zN ηN δ η 2 δ η ηN 2 (7.184) C C 2 2ω 2ωΩ 2ωΩ 2ω(2ω C Ω ) 2ω ω 6ω 2

211

212

7 Parametrically Excited Systems

Substituting (7.184) into (7.2) yields F(z η C ηN z) N δ 2 iµ (η 3η ηN C ηN 2 ) C (η η) N C 2 2ω 3ω Ω (Ω C 2ω) F(η zN C ηN z) C (7.185) 2ωΩ Substituting (7.183) and (7.184) into (7.168) and choosing g 2 to eliminate the resonance and near-resonance terms, we obtain u D η C ηN

i µ2 (Ω C 2ω)µ F iF2 η z η N η z zN 2ω 4ω 2 Ω 4ω 2 (Ω C 2ω) 10δ 2 i 3α η 2 ηN C 2ω 3ω 2

g2 D

(7.186)

Substituting (7.183) and (7.186) into (7.7), we obtain the normal form iF i 2 µ 2 (Ω C 2ω) 2 µ F ηN z η z ηN 2ω 2ω 4ω 2 Ω 10δ 2 i 2 F 2 i 2 3α η 2 ηN η z zN C 2 4ω (Ω C 2ω) 2ω 3ω 2

ηP D i ωη µ η C

(7.187)

7.6 Exercises

7.6.1 Use the method of multiple scales to determine the normal form of (7.1) when Ω ω and Ω 2ω. Use the complex-valued form (7.5) and seek a second-order approximate solution in the form ζ D ζ1 (T0 , T1 , T2 ) C ζ2 (T0 , T1 , T2 ) C C 3 ζ2 (T0 , T1 , T2 ) C and obtain the system of equations D0 ζ1 i ωζ1 D 0 i i T0 Ω C e i T0 Ω (ζ1 C ζN 1 ) e D0 ζ2 i ωζ2 D D1 ζ1 µ(ζ1 ζN 1 ) C 2ω D0 ζ3 i ωζ3 D D1 ζ2 D2 ζ1 µ(ζ2 ζN 2 ) i i T0 Ω C e i T0 Ω (ζ2 C ζN 2 ) e C 2ω Express the solution of the ﬁrst-order problem as ζ1 D A(T1 , T2 )e i ω T0 Then, show that the second-order problem becomes N i ω T0 D0 ζ2 i ωζ2 D D1 Ae i ω T0 µ Ae i ω T0 C µ Ae i i T0 Ω N i ω T0 e C C e i T0 Ω Ae i ω T0 C Ae 2ω

7.6 Exercises

The Case Ω ω Let Ω D ω C 2 σ and show that the solvability condition of the second-order problem is D1 A D µ A Then, show that ζ2 D

i µ N i ω T0 1 1 Ae C Ae i(Ω Cω)T0 Ae i(ωΩ )T0 2ω 2ωΩ 2ωΩ 1 1 N i(Ω Cω)T0 C N i(Ω ω)T0 Ae Ae 2ω(Ω C 2ω) 2ω(Ω 2ω)

Substitute for ζ1 and ζ2 into the third-order equation, eliminate the terms that produce secular terms, and obtain D2 A D

i µ2 i i N 2i σ T2 Ae AC AC 2ω ω(Ω 2 4ω 2 ) 2ωΩ (Ω 2ω)

Finally, use the method of reconstitution to obtain the normal form 2 iµ i i N 2i σ T2 A A Ae AP D µ A 2 2ω ω(Ω 2 4ω 2 ) 2ωΩ (Ω 2ω) Compare these results with those obtained in Section 7.1.1 by using the method of normal forms. The Case Ω 2ω Let Ω D 2ω C σ and show that the solvability condition of the second-order problem is D1 A D µ A C

i N i σ T1 Ae 2ω

Then, show that ζ2 D

i µ N i ω T0 1 1 Ae C e i(Ω Cω)T0 e i(ωΩ )T0 2ω 2ωΩ 2ωΩ 1 e i(Ω Cω)T0 2ω(Ω C 2ω)

Substitute for ζ1 and ζ2 into the third-order equation, eliminate the terms that produce secular terms, and obtain D2 A D

i µ2 i (Ω C 2ω)µ N i σ T1 Ae A A 2 2ω 4ω Ω (Ω C 2ω) 4ω 2 Ω

Finally, use the method of reconstitution to obtain the normal form 2 iµ i i N i σ T1 2 Ae AC A AP D µ A 2ω 2ω 4ω 2 Ω (Ω C 2ω) (Ω C 2ω)µ N i σ T1 Ae C 4ω 2 Ω

213

214

7 Parametrically Excited Systems

Compare these results with those obtained in Section 7.1.2 by using the method of normal forms. 7.6.2 Use the method of multiple scales to determine the normal form of (7.165) when (a) Ω is away from ω, 2ω, and 3ω; (b) Ω ω; (c) Ω 3ω; and (d) Ω 2ω. Seek an approximate solution of (7.166) in the form ζ D ζ1 (T0 , T1 , T2 ) C ζ2 (T0 , T1 , T2 ) C C 3 ζ2 (T0 , T1 , T2 ) C and obtain the system of equations D0 ζ1 i ωζ1 D 0 iδ (ζ1 C ζN 1 )2 D0 ζ2 i ωζ2 D D1 ζ1 µ(ζ1 ζN 1 ) C 2ω i F i Ω T0 C C e i Ω T0 (ζ1 C ζN 1 ) e 2ω iδ D0 ζ3 i ωζ3 D D1 ζ2 D2 ζ1 µ(ζ2 ζN 2 ) C (ζ1 C ζN 1 )(ζ2 C ζN 2 ) ω iα i F i Ω T0 C C e i Ω T0 (ζ2 C ζN 2 ) (ζ1 C ζN 1 )3 C e 2ω 2ω Express the solution of the ﬁrst-order problem as ζ1 D A(T1 , T2 )e i ω T0 Then, show that the second-order problem becomes i δ i ω T0 N i ω T0 C Ae Ae 2ω N i ω T0 N i ω T0 C i F e i Ω T0 C e i Ω T0 Ae i ω T0 C Ae C µ Ae 2ω

D0 ζ2 i ωζ2 D D1 Ae i ω T0 µ Ae i ω T0 C

The Case Ω Away from 2ω When Ω is away from 2ω, there are no nearresonances at second order. Eliminate the secular term and obtain D1 A D µ A Then, show that ζ2 D

i µ N i ω T0 δ 2 2i ω T0 δ δ N2 2i ω T0 Ae A e C A e 2 A AN 2ω 2ω 2 ω 6ω 2 F F F N i(Ω Cω)T0 C Ae i(Ω Cω)T0 Ae i(ωΩ )T0 Ae 2ωΩ 2ωΩ 2ω(Ω C 2ω) F N i(Ω ω)T0 Ae C 2ω(Ω 2ω)

Substitute ζ1 and ζ2 into the third-order equation, use the expression for D1 A, eliminate the terms that lead to secular and small-divisor terms, and obtain i µ2 iF2 3α 5δ 2 D2 A D A2 AN AC A C i 2ω ω(Ω 2 4ω 2 ) 2ω 3ω 3

7.6 Exercises

when Ω is away from ω and 3ω, i µ2 iF2 3α 5δ 2 A2 AN AC A C i 2ω ω(Ω 2 4ω 2 ) 2ω 3ω 3 5i F δ 2 i σ T2 i F δ N i σ T2 C A e A Ae 6ω 3 3ω 3

D2 A D

when Ω ω, and i µ2 iF2 ACi AC D2 A D 2ω 5ω 3

3α 5δ 2 2ω 3ω 3

i F δ N2 i σ T2 A e A2 AN C 2ω 3

when Ω 3ω. Compare the results obtained with the method of multiple scales with those obtained in Section 7.5.1 with the method of normal forms. The Case Ω 2ω Eliminate the terms that produce secular and small-divisor terms from the second-order equation and obtain D1 A D µ A C

i F N i σ T1 Ae 2ω

Then, show that ζ2 D

i µ N i ω T0 δ 2 2i ω T0 δ δ N2 2i ω T0 Ae A e C A e 2 A AN 2 2ω 2ω ω 6ω 2 F F F N i(Ω Cω)T0 Ae C Ae i(Ω Cω)T0 Ae i(ωΩ )T0 2ωΩ 2ωΩ 2ω(Ω C 2ω)

Substitute ζ1 and ζ2 into the third-order equation, use the expression for D1 A, eliminate the terms that lead to secular and small-divisor terms, and obtain i µ2 iF2 3α 5δ 2 A2 AN D2 A D A A C i 2ω 4ω 2 (ΩC 2ω) 2ω 3ω 3 F µ(Ω C 2ω) N i σ T1 Ae 4ω 2 Ω Compare the results obtained with the method of multiple scales with those obtained in Section 7.5.2 with the method of normal forms. 7.6.3 Consider the equation uR C ω 20 u C 2 u3 cos 2t D 0 1 Use the methods of multiple scales and normal forms to determine the equations describing the amplitude and the phase to ﬁrst order when

215

216

7 Parametrically Excited Systems

a) ω 0 is away from 1 and 1/2, b) ω 0 1 , c) ω 0 1/2 . 7.6.4 The parametric excitation of a two-degree-of-freedom system is governed uR 1 C ω 21 u 1 C cos Ω t ( f 11 u 1 C f 12 u 2 ) D 0 uR 2 C ω 22 u 2 C cos Ω t ( f 21 u 2 C f 22 u 2 ) D 0 Use the methods of multiple scales and normal forms to determine the equations describing the amplitudes and the phases when Ω ω 2 ω 1 .

217

8 MDOF Systems with Quadratic Nonlinearities In this chapter, we treat multiple-degree-of-freedom nongyroscopic and gyroscopic systems with quadratic nonlinearities subject to harmonic excitations.

8.1 Nongyroscopic Systems

A general nonlinear multiple-degree-of-freedom nongyroscopic system can be modeled by P t) D 0 xR C Ax C D xP C N (x, x,

(8.1)

where A and D are n n matrices, x is a column vector of length n, and N is a P and t having n components. nonlinear vector function of x, x, In this section, we treat systems whose linear parts are expressed in normalmode form, and in Section 8.3 we treat two linearly coupled oscillators. The linear part of (8.1) can be put in the following normal-mode form by using the linear transformation x D P u: O (u, uP , t) D 0 uR C J u C DO uP C N

(8.2)

where J D P 1 AP ,

DO D P 1 D P ,

O D P 1 N (P u, P uP , t) N

We choose P so that J has a Jordan canonical form and assume that its diagonal elements are positive. We also assume modal damping so that DO is a diagonal matrix. We limit our discussion to systems for which the frequencies (i.e., eigenvalues of A) are distinct and hence J is diagonal. We can explore the principal features of the analysis and limit the algebra to a minimum by considering a system having three degrees of freedom. Thus, we consider uR m C ω 2m u m C 2µ m uP m D

@V (u 1 , u 2 , u 3 ) C 2 f m cos (Ω t C τ m ) @u m

(8.3)

The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

218

8 MDOF Systems with Quadratic Nonlinearities

where V D α 1 u31 C α 2 u21 u 2 C α 3 u 1 u22 C α 4 u21 u 3 C α 5 u 1 u23 C α 6 u32 C α 7 u22 u 3 C α 8 u 2 u23 C α 9 u33 C α 10 u 1 u 2 u 3

(8.4)

and the ω n , α n , f n , τ n , µ n , and Ω are constants independent of . Here, is a small nondimensional parameter used for bookkeeping purposes. Again, as a ﬁrst step, we cast (8.3) and (8.4) in complex-valued form using the transformation u m D ζ m C ζN m , zP m D i Ωm z m ,

uP m D i ω m ζ m ζN m z m D f m e i (Ω tCτ m )

(8.5) (8.6)

The result is i @V ζ1 C ζN 1 , ζ2 C ζN 2 , ζ3 C ζN 3 ζP m D i ω m ζ m µ m ζ m ζN m 2ω m @u m i (z m C zN m ) (8.7) 2ω m To simplify the O(1) terms in (8.7), we introduce the transformation ζ m D η m C ∆ m1 z m C ∆ m2 zN m

(8.8)

use (8.6), and obtain 1 zm ηP m D i ω m η m C i (ω m Ω ) ∆ m1 2ω m 1 zN m C O() C i (ω m C Ω ) ∆ m2 2ω m

(8.9)

There are two possibilities: (a) Ω ω m for m D 1 or 2 or 3; and (b) Ω is away from all of the ω m . The ﬁrst possibility corresponds to primary resonance of the mth mode. It is treated in Section 8.1.4. When Ω is away from all of the ω m , we choose the ∆ m n to eliminate z m and zN m from (8.9); that is, ∆ m1 D

1 2ω m (ω m Ω )

and

∆ m2 D

1 2ω m (ω m C Ω )

(8.10)

8.1 Nongyroscopic Systems

We note that there are no small divisors in (8.10) because we assumed that Ω is away from every ω m . Substituting (8.8) into (8.7) and using (8.10), we obtain " i z1 C zN 1 2 ηP 1 D i ω 1 η 1 µ 1 (η 1 ηN 1 ) 3α 1 η 1 C ηN 1 C 2 2ω 1 ω1 Ω 2 z1 C zN 1 z2 C zN 2 η C 2α 2 η 1 C ηN 1 C 2 C η N C 2 2 ω1 Ω 2 ω 22 Ω 2 z1 C zN 1 z3 C zN 3 η 3 C ηN 3 C 2 C 2α 4 η 1 C ηN 1 C 2 ω1 Ω 2 ω3 Ω 2 2 z2 C zN 2 z3 C zN 3 2 C α 3 η 2 C ηN 2 C 2 η C α C η N C 5 3 3 ω2 Ω 2 ω 23 Ω 2 # z2 C zN 2 z3 C zN 3 η 3 C ηN 3 C 2 Cα 10 η 2 C ηN 2 C 2 (8.11) ω2 Ω 2 ω3 Ω 2 " i z1 C zN 1 2 α 2 η 1 C ηN 1 C 2 ηP 2 D i ω 2 η 2 µ 2 (η 2 ηN 2 ) 2ω 2 ω1 Ω 2 z1 C zN 1 z2 C zN 2 η C 2α 3 η 1 C ηN 1 C 2 C η N C 2 2 ω1 Ω 2 ω 22 Ω 2 z2 C zN 2 z3 C zN 3 η C 2α 7 η 2 C ηN 2 C 2 C η N C 3 3 ω2 Ω 2 ω 23 Ω 2 2 z2 C zN 2 z3 C zN 3 2 C 3α 6 η 2 C ηN 2 C 2 η C α C η N C 8 3 3 ω2 Ω 2 ω 23 Ω 2 # z1 C zN 1 z3 C zN 3 η 3 C ηN 3 C 2 Cα 10 η 1 C ηN 1 C 2 (8.12) ω1 Ω 2 ω3 Ω 2 " i z1 C zN 1 2 α 4 η 1 C ηN 1 C 2 ηP 3 D i ω 3 η 3 µ 3 (η 3 ηN 3 ) 2ω 3 ω1 Ω 2 z1 C zN 1 z3 C zN 3 η 3 C ηN 3 C 2 C 2α 5 η 1 C ηN 1 C 2 ω1 Ω 2 ω3 Ω 2 z2 C zN 2 z3 C zN 3 η C 2α 8 η 2 C ηN 2 C 2 C η N C 3 3 ω2 Ω 2 ω 23 Ω 2 2 z2 C zN 2 z3 C zN 3 2 C α 7 η 2 C ηN 2 C 2 η C 3α C η N C 9 3 3 ω2 Ω 2 ω 23 Ω 2 # z1 C zN 1 z2 C zN 2 η 2 C ηN 2 C 2 Cα 10 η 1 C ηN 1 C 2 (8.13) ω1 Ω 2 ω2 Ω 2 Because zP n D i Ω z n , resonance occurs when ω s 2ω m : Two-to-one autoparametric resonance, ω s ω n C ω m : Combination autoparametric resonance,

219

220

8 MDOF Systems with Quadratic Nonlinearities

Ω 2ω m : Subharmonic resonance of order one-half, Ω 1/2ω m : Superharmonic resonance of order two, Ω ω n ˙ ω m : Combination resonance. The ﬁrst two resonances are called internal or autoparametric resonances, whereas the last three are called external resonances because they are produced by the excitation. We note that some of these resonances might occur simultaneously. Next, we treat the three autoparametric resonances ω 2 2ω 1 , ω 3 ω 2 C ω 1 , and ω 2 2ω 1 and ω 3 2ω 2 in conjunction with a few of the external resonances. 8.1.1 Two-to-One Autoparametric Resonance

We consider the case ω 2 2ω 1 and no other autoparametric resonances. To simplify (8.11)–(8.13), we introduce the near-identity transformation η m D ξm C h m ξn , ξN n , z n , zN n

(8.14)

and choose the h m to eliminate the nonresonance terms. The autoparametric resonance produces the near-resonance terms ξ2 ξN1 and ξ12 in (8.11) and (8.12), respectively. The near-resonance terms produced by the excitation depend on the resonances being considered. Next, normal forms corresponding to several of these resonances are presented. No External Resonance

iα 2 N ξP1 D i ω 1 ξ1 µ 1 ξ1 ξ2 ξ1 ω1 iα 2 2 ξP2 D i ω 2 ξ2 µ 2 ξ2 ξ 2ω 2 1 ξP3 D i ω 3 ξ3 µ 3 ξ3

(8.15) (8.16) (8.17)

When expressed in the polar coordinates ξn D 1/2r n e i β n , (8.15)–(8.17) become α 2 r1 r2 sin (β 2 2β 1 ) 2ω 1 α 2 2 rP2 D µ 2 r2 r sin (β 2 2β 1 ) 4ω 2 1 rP1 D µ 1 r1 C

rP3 D µ 3 r3 α 2 r1 r2 cos (β 2 2β 1 ) 2ω 1 α 2 2 rP2 βP 2 D ω 2 r2 r cos (β 2 2β 2 ) 4ω 1 1

rP1 βP 1 D ω 1 r1

r3 βP 3 D ω 3 r3

(8.18) (8.19) (8.20) (8.21) (8.22) (8.23)

8.1 Nongyroscopic Systems

Ω 1/2ω 1

iα 2 N i ξP1 D i ω 1 ξ1 µ 1 ξ1 ξ2 ξ1 ω1 2ω 1

3α 1 z12 α 3 z22 C 2 2 2 2 (ω 1 Ω ) (ω 2 Ω 2 )2

α 5 z32 2α 2 z1 z2 2α 4 z1 z3 C 2 C 2 2 2 2 2 2 2 (ω 3 Ω ) (ω 1 Ω )(ω 2 Ω ) (ω 1 Ω 2 )(ω 23 Ω 2 ) α 10 z2 z3 C 2 (8.24) (ω 2 Ω 2 )(ω 23 Ω 2 )

C

iα 2 2 ξP2 D i ω 2 ξ2 µ 2 ξ2 ξ 2ω 2 1

(8.25)

ξP3 D i ω 3 ξ3 µ 3 ξ3

(8.26)

Ω 2ω 2

iα 2 N ξP1 D i ω 1 ξ1 µ 1 ξ1 ξ2 ξ1 ω1 iα 2 2 i ξP2 D i ω 2 ξ2 µ 2 ξ2 ξ 2ω 2 1 ω 2

(8.27)

α 3 z1 3α 6 z2 α 7 z3 ξN2 C C ω 21 Ω 2 ω 22 Ω 2 ω 23 Ω 2 (8.28)

ξP3 D i ω 3 ξ3 µ 3 ξ3

(8.29)

Ω ω3 C ω2

iα 2 N ξP1 D i ω 1 ξ1 µ 1 ξ1 ξ2 ξ1 ω1

(8.30)

α 10 z1 2α 7 z2 2α 8 z3 N ξ3 C C ω 21 Ω 2 ω 22 Ω 2 ω 23 Ω 2 (8.31) α 10 z1 i 2α 7 z2 2α 8 z3 ξP3 D i ω 3 ξ3 µ 3 ξ3 ξN2 (8.32) C C 2ω 3 ω 21 Ω 2 ω 22 Ω 2 ω 23 Ω 2 iα 2 2 i ξP2 D i ω 2 ξ2 µ 2 ξ2 ξ 2ω 2 1 2ω 2

Ω ω3 ω1

ξP1 D i ω 1 ξ1 µ 1 ξ1

2α 4 zN 1 iα 2 N i α 10 zN 2 2α 5 zN 3 ξ3 ξ2 ξ1 C 2 C 2 (8.33) ω1 2ω 1 ω 21 Ω 2 ω2 Ω 2 ω3 Ω 2 iα 2 2 ξP2 D i ω 2 ξ2 µ 2 ξ2 ξ (8.34) 2ω 2 1 2α 4 z1 i α 10 z2 2α 5 z3 ξP3 D i ω 3 ξ3 µ 3 ξ3 ξ1 (8.35) C 2 C 2 2ω 3 ω 21 Ω 2 ω2 Ω 2 ω3 Ω 2

221

222

8 MDOF Systems with Quadratic Nonlinearities

Ω 2ω 2 and Ω ω 3 C ω 1

iα 2 N ξP1 D i ω 1 ξ1 µ 1 ξ1 ξ2 ξ1 ω1 2α 4 z1 i α 10 z2 2α 5 z3 ξN3 C C (8.36) 2ω 1 ω 21 Ω 2 ω 22 Ω 2 ω 23 Ω 2 α 3 z1 iα 2 2 i 3α 6 z2 α 7 z3 ξN2 ξP2 D i ω 2 ξ2 µ 2 ξ2 ξ1 C C 2ω 2 ω 2 ω 21 Ω 2 ω 22 Ω 2 ω 23 Ω 2 (8.37) 2α 4 z1 i α 10 z2 2α 5 z3 ξP3 D i ω 3 ξ3 µ 3 ξ3 ξN1 (8.38) C C 2ω 3 ω 21 Ω 2 ω 22 Ω 2 ω 23 Ω 2 8.1.2 Combination Autoparametric Resonance

We consider the autoparametric-resonance condition ω 3 ω 2 C ω 1 , which produces the near-resonance terms ξ3 ξN2 , ξ3 ξN1 , and ξ1 ξ2 in (8.11)–(8.13), respectively, In the absence of external resonances, the normal form of (8.11)–(8.13) is iα 10 N ξP1 D i ω 1 ξ1 µ 1 ξ1 ξ3 ξ2 2ω 1 iα 10 N ξP2 D i ω 2 ξ2 µ 2 ξ2 ξ3 ξ1 2ω 2 iα 10 ξP3 D i ω 3 ξ3 µ 3 ξ3 ξ1 ξ2 2ω 3

(8.39) (8.40) (8.41)

When expressed in the polar coordinates ξn D 1/2r n e i β n , (8.39)–(8.41) become α 10 r2 r3 sin (β 3 β 2 β 1 ) 4ω 1 α 10 rP2 D µ 2 r2 C r1 r3 sin (β 3 β 2 β 1 ) 4ω 2 α 10 rP3 D µ 3 r3 r1 r2 sin (β 3 β 2 β 1 ) 4ω 3 α 10 rP1 βP 1 D ω 1 r1 r2 r3 cos (β 3 β 2 β 1 ) 4ω 1 α 10 rP2 βP 2 D ω 2 r2 r1 r3 cos (β 3 β 2 β 1 ) 4ω 2 α 10 rP3 βP 3 D ω 3 r3 r1 r2 cos (β 3 β 2 β 1 ) 4ω 3 rP1 D µ 1 r1 C

(8.42) (8.43) (8.44) (8.45) (8.46) (8.47)

The normal forms in cases of external resonances can be obtained by adding appropriate terms to (8.39)–(8.41), as in Section 8.1.1.

8.1 Nongyroscopic Systems

8.1.3 Simultaneous Two-to-One Autoparametric Resonances

We consider the autoparametric resonances ω 3 2ω 2 and ω 2 2ω 1 , which produce the near-resonance terms ξ2 ξN1 , ξ12 and ξ3 ξN2 , and ξ22 in (8.11)–(8.13), respectively. Consequently, in the absence of external resonances, the normal form of (8.11)–(8.13) is iα 2 N ξP1 D i ω 1 ξ1 µ 1 ξ1 ξ2 ξ1 ω1 i ξP2 D i ω 2 ξ2 µ 2 ξ2 α 2 ξ12 C 2α 7 ξ3 ξN2 2ω 2 iα 7 2 ξP3 D i ω 3 ξ3 µ 3 ξ3 ξ 2ω 3 2

(8.48) (8.49) (8.50)

When expressed in the polar coordinates ξn D 1/2r n e i β n , (8.48)–(8.50) become rP1 D µ 1 r1 C rP2 D µ 2 r2 rP3 D µ 3 r3 rP1 βP 1 D ω 1 r1 rP2 βP 2 D ω 2 r2 rP3 βP 3 D ω 3 r3

α 2 r1 r2 sin (β 2 2β 1 ) 2ω 1 α 2 2 r sin (β 2 2β 1 ) C 4ω 2 1 α 7 2 r sin (β 3 2β 2 ) 4ω 3 2 α 2 r1 r2 cos (β 2 2β 1 ) 2ω 1 α 2 2 r cos (β 2 2β 1 ) 4ω 2 1 α 7 2 r cos (β 3 2β 2 ) 4ω 3 2

(8.51) α 7 r2 r3 sin (β 3 2β 2 ) 2ω 2

(8.52) (8.53) (8.54)

α 7 r2 r3 cos (β 3 2β 2 ) 2ω 2

(8.55) (8.56)

This form is in full agreement with that obtained by Nayfeh, Asrar, and Nayfeh (1992) by using the method of multiple scales. Again, the normal forms in the presence of external resonances can be obtained by adding appropriate terms to (8.48)–(8.50), as in Section 8.1.1. 8.1.4 Primary Resonances

To treat this case, we scale the excitation amplitudes f n and hence the z n at O() so that the resonance terms produced by the excitation appear at the same order as those produced by the nonlinearity. Thus, we modify (8.7) into 2 i i n (z1 C zN 1 ) ζP 1 D i ω 1 ζ1 µ 1 ζ1 ζN 1 3α 1 ζ1 C ζN 1 2ω 1 2ω 1 2 C 2α 2 ζ1 C ζN 1 ζ2 C ζN 2 C α 3 ζ2 C ζN 2 C 2α 4 ζ1 C ζN 1 ζ3 C ζN 3 2 o Cα 5 ζ3 C ζN 3 C α 10 ζ2 C ζN 2 ζ3 C ζN 3 (8.57)

223

224

8 MDOF Systems with Quadratic Nonlinearities

2 i i n (z2 C zN 2 ) ζP 2 D i ω 2 ζ2 µ 2 ζ2 ζN 2 α 2 ζ1 C ζN 1 2ω 2 2ω 2 2 C 2α 3 ζ1 C ζN 1 ζ2 C ζN 2 C 3α 6 ζ2 C ζN 2 2 o C2α 7 ζ2 C ζN 2 ζ3 C ζN 3 C α 8 ζ3 C ζN 3 C α 10 ζ1 C ζN 1 ζ3 C ζN 3 (8.58) 2 i i n (z3 C zN 3 ) ζP 3 D i ω 3 ζ3 µ 3 ζ3 ζN 3 α 4 ζ1 C ζN 1 2ω 3 2ω 3 2 N N N C 2α 5 ζ1 C ζ1 ζ3 C ζ3 C α 7 ζ2 C ζ2 C 2α 8 ζ2 C ζN 2 ζ3 C ζN 3 2 o C3α 9 ζ3 C ζN 3 C α 10 ζ1 C ζN 1 ζ2 C ζN 2 (8.59) To simplify (8.57)–(8.59), we introduce the near-identity transformation ζ m D ξm C h m ξn , ξN n , z n , zN n

(8.60)

and choose the h m to eliminate the nonresonance terms. The resulting normal forms depend on the resonance conditions being considered. As discussed in the preceding section, internal or autoparametric resonances occur when ω n 2ω m and/or ω s ω n ˙ ω m . Primary resonances occur when Ω ω m . Consequently, the normal forms of (8.57)–(8.59) for a few of these resonances are: ω 2 2ω 1 and Ω ω n

iα 2 N i ξP1 D i ω 1 ξ1 µ 1 ξ1 ξ2 ξ1 δ 1n z1 ω1 2ω 1 iα 2 2 i ξP2 D i ω 2 ξ2 µ 2 ξ2 ξ δ 2n z2 2ω 2 1 2ω 2 i ξP3 D i ω 3 ξ3 µ 3 ξ3 δ 3n z3 2ω 3

(8.61) (8.62) (8.63)

where δ m n is the Kronecker delta. ω 3 ω 2 C ω 1 and Ω ω n

iα 10 N ξP1 D i ω 1 ξ1 µ 1 ξ1 ξ3 ξ2 2ω 1 iα 10 N ξP2 D i ω 2 ξ2 µ 2 ξ2 ξ3 ξ1 2ω 2 iα 10 ξP3 D i ω 3 ξ3 µ 3 ξ3 ξ1 ξ2 2ω 3

i δ 1n z1 2ω 1 i δ 2n z2 2ω 2 i δ 3n z3 2ω 3

(8.64) (8.65) (8.66)

8.2 Gyroscopic Systems

ω 3 2ω 2 , ω 2 2ω 1 , and Ω ω n

iα 2 N i ξP1 D i ω 1 ξ1 µ 1 ξ1 ξ2 ξ1 δ 1n z1 ω1 2ω 1 i i ξP2 D i ω 2 ξ2 µ 2 ξ2 α 2 ξ12 C 2α 7 ξ3 ξN2 δ 2n z2 2ω 2 2ω 2 iα 7 2 i ξP3 D i ω 3 ξ3 µ 3 ξ3 ξ δ 3n z3 2ω 3 2 2ω 3

(8.67) (8.68) (8.69)

ω 3 ω 2 C ω 1 , ω 2 2ω 1 , and Ω ω n

i i ξP1 D i ω 1 ξ1 µ 1 ξ1 2α 2 ξ2 ξN1 C α 10 ξ3 ξN2 δ 1n z1 2ω 1 2ω 1 i i ξP2 D i ω 2 ξ2 µ 2 ξ2 α 2 ξ12 C α 10 ξ3 ξN1 δ 2n z2 2ω 2 2ω 2 iα 10 i ξP3 D i ω 3 ξ3 µ 3 ξ3 ξ1 ξ2 δ 3n z3 2ω 3 2ω 3

(8.70) (8.71) (8.72)

8.2 Gyroscopic Systems

As in Section 7.4, for simplicity, we consider a two-degree-of-freedom system to illustrate the method of solution. Speciﬁcally, we consider uR 1 C λ 1 uP 2 C α 1 u 1 D 3δ 1 u21 C 2δ 2 u 1 u 2 C δ 3 u22 C 2F1 cos (Ω t C τ 1 ) (8.73) uR 2 λ 2 uP 1 C α 2 u 2 D δ 2 u21 C 2δ 3 u 1 u 2 C 3δ 4 u22 C 2F2 cos (Ω t C τ 2 ) (8.74) Using the transformation (7.144)–(7.147) and following steps similar to those used in Section 7.4, we cast (8.73) and (8.74) in the complex-valued form λ1 i ω 1 (α 1 ω 22 ) (z1 C zN 1 ) (z2 C zN 2 ) ζP 1 D i ω 1 ζ1 C 2α 1 (ω 22 ω 21 ) 2(ω 22 ω 21 ) 2 C χ 11 ζ1 C ζN 1 C ζ2 C ζN 2 C χ 12 ζ1 C ζN 1 C ζ2 C ζN 2 i(α 1 ω 22 ) i(α 1 ω 21 ) N N ζ1 ζ1 C ζ2 ζ2 λ1 ω1 λ1 ω2 i(α 1 ω 22 ) 2 i(α 1 ω 21 ) C χ 13 ζ1 ζN 1 C ζ2 ζN 2 λ1 ω1 λ1 ω2

(8.75)

225

226

8 MDOF Systems with Quadratic Nonlinearities

λ1 i ω 2 (α 1 ω 21 ) (z1 C zN 1 ) C (z2 C zN 2 ) ζP 2 D i ω 2 ζ2 2 2 2 2α 1 (ω 2 ω 1 ) 2(ω 2 ω 21 ) 2 C χ 21 ζ1 C ζN 1 C ζ2 C ζN 2 C χ 22 ζ1 C ζN 1 C ζ2 C ζN 2 i(α 1 ω 22 ) i(α 1 ω 21 ) ζ1 ζN 1 C ζ2 ζN 2 λ1 ω1 λ1 ω2 2 i(α 1 ω 22 ) 2 i(α 1 ω 1 ) C χ 23 ζ1 ζN 1 C ζ2 ζN 2 λ1 ω1 λ1 ω2

(8.76)

on account of (8.6), where χ 11 D

3i δ 1 ω 1 (α 1 ω 22 ) δ 2 α 1 λ 1 , 2α 1 (ω 22 ω 21 )

χ 12 D

i δ 2 ω 1 (α 1 ω 22 ) δ 3 α 1 λ 1 α 1 (ω 22 ω 21 )

χ 13 D

i δ 3 ω 1 (α 1 ω 22 ) 3δ 4 α 1 λ 1 , 2α 1 (ω 22 ω 21 )

χ 21 D

δ 2 α 1 λ 1 3i δ 1 ω 2 (α 1 ω 21 ) 2α 1 (ω 22 ω 21 )

χ 22 D

δ 3 α 1 λ 1 i δ 2 ω 2 (α 1 ω 21 ) , α 1 (ω 22 ω 21 )

χ 33 D

3δ 4 α 1 λ 1 i δ 3 ω 2 (α 1 ω 21 ) 2α 1 (ω 22 ω 21 )

As discussed in Section 8.1, the excitation may produce one of two possible types of resonances: primary and secondary resonances. Primary resonances are treated in the next section, and secondary resonances are treated in Section 8.2.2. 8.2.1 Primary Resonances

As discussed in Section 8.1.4, we scale the f n and hence the z n at O() so that the resonance terms produced by the excitation appear at the same order as those produced by the nonlinearity. To simplify (8.75) and (8.76), we introduce the near-identity transformation (8.60) and choose the h m to eliminate the nonresonance terms. Resonance terms occur when ω 2 2ω 1 (i.e., when there is a two-to-one autoparametric resonance) and Ω ω n (i.e., when there is a primary resonance of the nth mode). When ω 2 2ω 1 and Ω ω n , the normal form of (8.75) and (8.76) is i ω 1 (α 1 ω 2 )z1 λ 1 α 1 z2 ξP1 D i ω 1 ξ1 C Λ 1 ξ2 ξN1 C δ 1n 2α 1 (ω 22 ω 21 ) i ω 2 (α 1 ω 21 )z1 λα 1 z2 ξP2 D i ω 2 ξ2 C Λ 2 ξ12 δ 2n 2α 1 (ω 22 ω 21 ) 2

(8.77) (8.78)

where Λ 1 D 2χ 11 Λ 2 D χ 21 C

i χ 12 (ω 2 ω 1 )(α 1 C ω 1 ω 2 ) 2χ 13 α 1 λ 2 λ1 ω1 ω2 λ1 ω1 ω2 i χ 22 (α 1 ω 21 ) χ 23 (α 1 ω 21 )2 λ1 ω1 λ 21 ω 21

(8.79) (8.80)

8.2 Gyroscopic Systems

8.2.2 Secondary Resonances

To treat this case, we ﬁrst introduce the linear transformation (8.8) and choose ∆ m1 and ∆ m2 to eliminate z n and zN n from (8.75) and (8.76); that is, z1 ω 1 (α 1 ω 22 ) zN 1 C ω1 C Ω 2α 1 (ω 22 ω 21 ) ω 1 Ω z2 i λ1 zN 2 C ω1 C Ω 2(ω 22 ω 21 ) ω 1 Ω z1 ω 2 (α 1 ω 21 ) zN 1 ζ2 D η 2 C C ω2 C Ω 2α 1 (ω 22 ω 21 ) ω 2 Ω i λ1 zN 2 z2 C C ω2 C Ω 2(ω 22 ω 21 ) ω 2 Ω

ζ1 D η 1

(8.81)

(8.82)

Hence, (8.75) and (8.76) become ηP 1 D i ω 1 η 1 C χ 11 [η 1 C ηN 1 C η 2 C ηN 2 C b 1 (z1 C zN 1 ) C b 2 (z2 zN 2 )]2 C χ 12 [η 1 C ηN 1 C η 2 C ηN 2 C b 1 (z1 C zN 1 ) C b 2 (z2 zN 2 )] i(α 1 ω 22 ) i(α 1 ω 21 ) (η 1 ηN 1 ) C (η 2 ηN 2 ) C b 3 (z1 zN 1 ) λ1 ω1 λ1 ω2 i(α 1 ω 21 ) (η 1 ηN 1 ) Cb 4 (z2 C zN 2 ) C χ 13 λ2 ω1 2 i(α 1 ω 22 ) (η 2 ηN 2 ) C b 3 (z1 zN 1 ) C b 4 (z2 C zN 2 ) C (8.83) λ1 ω2 ηP 2 D i ω 2 η 2 C χ 21 [η 1 C ηN 1 C η 2 C ηN 2 C b 1 (z1 C zN 1 ) C b 2 (z2 zN 2 )]2 C χ 22 [η 1 C ηN 1 C η 2 C ηN 2 C b 1 (z1 C zN 1 ) C b 2 (z2 zN 2 )] i(α 1 ω 22 ) i(α 1 ω 21 ) (η 1 ηN 1 ) C (η 2 ηN 2 ) C b 3 (z1 zN 1 ) λ1 ω1 λ1 ω2 i(α 1 ω 21 ) (η 1 ηN 1 ) Cb 4 (z2 C zN 2 ) C χ 23 λ1 ω1 2 i(α 1 ω 22 ) (η 2 ηN 2 ) C b 3 (z1 zN 1 ) C b 4 (z2 C zN 2 ) C (8.84) λ1 ω2 where (b 1 , b 2 , b 3 , b 4 ) D

(α 2 Ω 2 , i λ 1 Ω , i λ 2 Ω , α 1 Ω 2 ) (ω 21 Ω 2 )(ω 22 Ω 2 )

(8.85)

It follows from (8.83) and (8.84) that, to this order, only a two-to-one autoparametric resonance is possible and that the excitation can produce resonances corresponding to

227

228

8 MDOF Systems with Quadratic Nonlinearities

Ω 2ω m : Subharmonic resonance of order one-half of the mth mode, Ω 1/2ω m : Superharmonic resonance of order two of the mth mode, Ω ω 2 ˙ ω 1 : Combination resonance of the additive or difference types. Next, we present the normal forms for several of the external resonances for the case of two-to-one autoparametric resonance. Because (8.83) and (8.84) were derived under the assumption that Ω is away from ω 1 and ω 2 , the resonance conditions Ω 2ω 1 , Ω 1/2ω 2 , and Ω ω 2 ω 1 must be excluded when considering the two-to-one autoparametric resonance. Ω 2ω 2

ξP1 D i ω 1 ξ1 C Λ 1 ξ2 ξN1

(8.86)

ξP2 D i ω 2 ξ2 C Λ 2 ξ12 C Γ ξN2

(8.87)

where Λ 1 and Λ 2 are deﬁned in (8.79) and (8.80) and (α 1 ω 22 ) (b 3 z1 C b 4 z2 ) Γ D 2χ 21 (b 1 z1 C b 2 z2 ) 2i χ 23 λ1 ω2 i(α 1 ω 22 ) (b 1 z1 C b 2 z2 ) C χ 22 b 3 z1 C b 4 z2 λ1 ω2

(8.88)

Ω 1/2ω 1

ξP1 D i ω 1 ξ1 C Λ 1 ξ2 ξN1 C Γ

(8.89)

ξP2 D i ω 2 ξ2 C Λ 2 ξ12

(8.90)

where Λ 1 and Λ 2 are deﬁned in (8.79) and (8.80) and Γ D χ 11 (b 1 z1 C b 2 z2 )2 C χ 12 (b 1 z1 C b 2 z2 ) (b 3 z1 C b 4 z2 ) C χ 13 (b 3 z1 C b 4 z2 )2

(8.91)

Ω ω2 C ω1

ξP1 D i ω 1 ξ1 C Λ 1 ξ2 ξN1 C Γ1 ξN2

(8.92)

ξP2 D i ω 2 ξ2 C Λ 2 ξ12 C Γ2 ξN1

(8.93)

where Λ 1 and Λ 2 are deﬁned in (8.79) and (8.80) and i α 1 ω 22 (b 3 z1 C b 4 z2 ) Γ1 D 2χ 11 (b 1 z1 C b 2 z2 ) 2χ 13 λ1 ω2 # " i α 1 ω 22 (b 1 z1 C b 2 z2 ) C χ 12 b 3 z1 C b 4 z2 λ1 ω2

(8.94)

8.3 Two Linearly Coupled Oscillators

i α 1 ω 21 (b 3 z1 C b 4 z2 ) Γ2 D 2χ 21 (b 1 z1 C b 2 z2 ) 2χ 23 λ1 ω1 # " i α 1 ω 21 (b 1 z1 C b 2 z2 ) C χ 22 b 3 z1 C b 4 z2 λ1 ω1

(8.95)

8.3 Two Linearly Coupled Oscillators

In this section, we consider the two linearly coupled oscillators uR 1 C k11 u 1 C k12 u 2 D 2 d1 uP 1 C α 11 u21 C α 12 u22 C α 13 u 1 u 2

(8.96)

uR 2 C k21 u 1 C k22 u 2 D 2 d2 uP 2 C α 21 u21 C α 22 u22 C α 23 u 1 u 2

(8.97)

To determine the normal form of (8.96) and (8.97), we use two approaches. First, we use these equations as they are. Second, we ﬁrst transform them so that their linear parts are in normal-mode form, as in Section 8.1. The solution of the linearly undamped equations (8.96) and (8.97) can be expressed as u 1 D A 1 e i ω 1 t C A 2 e i ω 2 t C cc

(8.98)

u 2 D Γ1 A 1 e i ω 2 t C Γ2 A 2 e i ω 2 t C cc

(8.99)

when ω 1 ¤ ω 2 , where ω 4 (k11 C k22 ) ω 2 C k11 k22 k12 k21 D 0 Γi D

k11 ω 21 k21 D k12 k22 ω 2i

for i D 1 and 2

(8.100)

Using (8.98) and (8.99), we express (8.96) and (8.97) as a system of two ﬁrst-order complex-valued equations by using the transformation u 1 D ζ1 C ζN 1 C ζ2 C ζN 2

(8.101)

u 2 D Γ1 ζ1 C ζN 1 C Γ2 ζ2 C ζN 2

(8.102)

uP 1 D i ω 1 ζ1 ζN 1 C i ω 2 ζ2 ζN 2

(8.103)

uP 2 D i ω 1 Γ1 ζ1 ζN 1 C i ω 2 Γ2 ζ1 ζN 2

(8.104)

229

230

8 MDOF Systems with Quadratic Nonlinearities

Solving (8.101)–(8.104), we have 1 Γ2 u 1 u 2 C ζ1 D 2(Γ2 Γ1 ) 1 u 2 Γ1 u 1 C ζ2 D 2(Γ2 Γ1 )

i ( uP 2 Γ2 uP 1 ) ω1 i (Γ1 uP 1 uP 2 ) ω2

Differentiating (8.105) and (8.106) with respect to t, we obtain 1 i ( uR 2 Γ2 uR 1 ) ζP 1 D Γ2 uP 1 uP 2 C 2(Γ2 Γ1 ) ω1 1 i (Γ1 uR 1 uR 2 ) ζP 2 D uP 2 Γ1 uP 1 C 2(Γ2 Γ1 ) ω2

(8.105) (8.106)

(8.107) (8.108)

Eliminating uR 1 and uR 2 from (8.107) and (8.108) by using (8.96) and (8.97), we obtain ω 2 Γ2 (d2 d1 ) (d2 Γ1 d1 Γ2 ) ζP 1 D i ω 1 ζ1 C ζ1 ζN 1 C ζ2 ζN 2 Γ2 Γ1 ω 1 (Γ2 Γ1 ) n 2 i (α 21 Γ2 α 11 ) ζ1 C ζN 1 C ζ2 C ζN 2 C 2ω 1 (Γ2 Γ1 ) 2 C (α 22 Γ2 α 12 ) Γ1 ζ1 C ζN 1 C Γ2 ζ2 C ζN 2 o C (α 23 Γ2 α 13 ) ζ1 C ζN 1 C ζ2 C ζN 2 Γ1 ζ1 C ζN 1 C Γ2 ζ2 C ζN 2 (8.109) (d1 Γ1 d2 Γ2 ) ω 1 Γ1 (d1 d2 ) ζP 2 D i ω 2 ζ2 C ζ1 ζN 1 C ζ2 ζN 2 ω 2 (Γ2 Γ1 ) Γ2 Γ1 n 2 i (Γ1 α 11 α 21 ) ζ1 C ζN 1 C ζ2 C ζN 2 C 2ω 2 (Γ2 Γ1 ) 2 C (Γ1 α 12 α 22 ) Γ1 ζ1 C ζN 1 C Γ2 ζ2 C ζN 2 o C (Γ1 α 13 α 23 ) ζ1 C ζN 1 C ζ2 C ζN 2 Γ1 ζ1 C ζN 1 C Γ2 ζ2 C ζN 2 (8.110) To simplify (8.109) and (8.110), we introduce the near-identity transformation (8.111) ζ m D ξm C h m ξ1 , ξN1 , ξ2 , ξN2 and choose the h m to eliminate the nonresonance terms. There are two cases: twoto-one autoparametric resonance (i.e., ω 2 2ω 1 ) and no autoparametric resonance. Next, we present the normal forms for these cases starting with the second. No Autoparametric Resonance

ξP1 D i ω 1 ξ1 µ 1 ξ1

(8.112)

ξP2 D i ω 2 ξ2 µ 2 ξ2

(8.113)

8.3 Two Linearly Coupled Oscillators

where µ1 D

d1 Γ2 d2 Γ1 d2 (k11 ω 21 ) d1 (k11 ω 22 ) D Γ2 Γ1 ω 22 ω 21

(8.114)

µ2 D

d2 Γ2 d1 Γ1 d2 (k11 ω 22 ) d1 (k11 ω 21 ) D Γ2 Γ1 ω 21 ω 22

(8.115)

Two-to-One Autoparametric Resonance

ξP1 D i ω 1 ξ1 µ 1 ξ1 C iΛ 1 ξ2 ξN1

(8.116)

ξP2 D i ω 2 ξ2 µ 2 ξ2 C iΛ 2 ξ12

(8.117)

where 1 [2 (α 21 Γ2 α 11 ) C 2Γ1 Γ2 (α 22 Γ2 α 12 ) 2ω 1 (Γ2 Γ1 )

Λ1 D

Λ2 D

C (Γ1 C Γ2 ) (α 23 Γ2 α 13 )]

(8.118)

1 Γ1 α 11 α 21 C Γ12 (Γ1 α 12 α 22 ) 2ω 2 (Γ2 Γ1 ) CΓ1 (Γ1 α 13 α 23 )

(8.119)

Next, we ﬁrst transform the linear parts of (8.96) and (8.97) into a normal-mode form. To accomplish this, we note from (8.98)–(8.100) that the eigenvectors corresponding to the eigenvalues ω 1 and ω 2 are [1 Γ1 ]T and [1 Γ2 ]T . Hence, we introduce the transformation 1 1 v (8.120) u D Pv D Γ1 Γ2 into (8.96) and (8.97) and obtain

1 vR1 k 1 1 v1 k12 C 11 Γ2 vR2 k21 k22 Γ1 Γ2 v2 d 1 1 vP1 0 D 2 1 0 d2 Γ1 Γ2 vP2 α (v C v2 )2 C α 12 (Γ1 v1 C Γ2 v2 )2 C α 13 (v1 C v2 )(Γ1 v1 C Γ2 v2 ) C 11 1 α 21 (v1 C v2 )2 C α 22 (Γ1 v1 C Γ2 v2 )2 C α 23 (v1 C v2 )(Γ1 v1 C Γ2 v2 ) 1 Γ1

(8.121) Multiplying (8.121) from the left with P 1 D

1 Γ2 Γ1

Γ2 Γ1

1 1

(8.122)

231

232

8 MDOF Systems with Quadratic Nonlinearities

and after some algebraic manipulations, we obtain 2 vR1 ω1 0 v1 µ 1 µ 3 vP1 C D 2 0 ω 22 v2 vR2 µ 4 µ 2 vP2 2 3 (Γ2 α 11 α 21 ) (v1 C v2 )2 C (Γ2 α 12 α 22 ) (Γ1 v1 C Γ2 v2 )2 6 7 C (Γ2 α 13 α 23 ) (v1 C v2 ) (Γ1 v1 C Γ2 v2 ) 6 7 C 2 25 4 (α ) (v ) ) (Γ ) (α Γ2 Γ1 21 Γ1 α 11 1 C v2 C 22 Γ1 α 12 1 v1 C Γ2 v2 C (α 23 Γ1 α 13 ) (v1 C v2 ) (Γ1 v1 C Γ2 v2 ) (8.123) where µ 1 and µ 2 are deﬁned in (8.114) and (8.115) and µ3 D

Γ2 (d1 d2 ) Γ1 (d2 d1 ) and µ 4 D Γ2 Γ1 Γ2 Γ1

(8.124)

Next, we transform (8.123) into two ﬁrst-order complex-valued equations by letting (8.125) v1 D ζ1 C ζN 1 , vP1 D i ω 1 ζ1 ζN 1 v2 D ζ2 C ζN 2 , vP2 D i ω 2 ζ2 ζN 2

(8.126)

Using (8.125) and (8.126), we transform (8.123) into (8.109) and (8.110). Then, we obtain the normal forms (8.112) and (8.113) in the absence of autoparametric resonance and (8.116) and (8.117) in the presence of autoparametric resonance.

8.4 Exercises

8.4.1 Use the methods of multiple scales and normal forms to determine a ﬁrstorder uniform expansion for uR 1 C ω 21 u 1 D α 1 u 1 u 2 uR 2 C ω 22 u 2 D α 2 u21 for small but ﬁnite amplitudes when ω 2 2ω 1 . 8.4.2 Use the methods of multiple scales and normal forms to determine ﬁrst-order uniform expansions for uR 1 C ω 21 u 1 D α 1 u 1 u 2 C k1 cos Ω1 t uR 2 C ω 22 u 2 D α 2 u21 C k2 cos Ω2 t when a) ω 2 2ω 1 and Ω1 ω 1 , b) ω 2 2ω 1 and Ω2 ω 2 .

8.4 Exercises

8.4.3 Determine the normal form of uR 1 C 12 uP 2 C δ u D u 1 u 2 uR 2 12 uP 1 C 12 u 2 D u21 when δ 1/2. 8.4.4 Determine the normal form of uR 1 C ω 21 u 1 D α 1 u 2 u 3 uR 2 C ω 22 u 2 D α 2 u 1 u 3 uR 3 C ω 23 u 3 D α 3 u 1 u 2 when ω 3 ω 2 C ω 1 . 8.4.5 Determine the normal form of uR 1 C 2 uP 2 C 3u 1 C 2 cos Ω t ( f 11 u 1 C f 12 u 2 ) D 0 uR 2 C uP 1 C 12u 2 C 2 cos Ω t ( f 21 u 1 C f 22 u 2 ) D 0 when Ω 5 and Ω 1. 8.4.6 Determine the normal form of uR 1 C 4 uP 2 C 3u 1 C 2 cos Ω t ( f 11 u 1 C f 12 u 2 ) D 0 uR 2 uP 1 C 3u 2 C 2 cos Ω t ( f 21 u 1 C f 22 u 2 ) D 0 when Ω 4 and Ω 2. 8.4.7 Determine the normal form of uR 1 C 5 uP 2 C 6u 1 C 2 cos Ω t ( f 11 u 1 C f 12 u 2 ) D 0 uR 2 C uP 1 C 24u 2 C 2 cos Ω t ( f 21 u 1 C f 22 u 2 ) D 0 when Ω 7 and Ω 1. 8.4.8 Use the methods of multiple scales and normal forms to determine an approximation to the solution of uR 1 C ω 21 u 1 D α 1 uP 1 uP 2 uR 2 C ω 22 u 2 D α 2 uP 21 ω 2 2ω 1 8.4.9 Use the methods of multiple scales and normal forms to determine a ﬁrstorder approximation to the solution of uR 1 C ω 21 u 1 D α 1 u21 C α 2 u 1 u 2 C α 3 u22 uR 2 C ω 22 u 2 D α 4 u21 C α 5 u 1 u 2 C α 6 u32 ω 2 2ω 1

233

234

8 MDOF Systems with Quadratic Nonlinearities

8.4.10 Use the methods of multiple scales and normal forms to determine a ﬁrstorder approximation to uR 1 C ω 21 u 1 C δ 1 u21 C 2δ 2 u 1 u 2 C δ 3 u22 D F1 cos Ω t uR 2 C ω 22 u 2 C δ 2 u21 C 2δ 3 u 1 u 2 C δ 4 u22 D 0 where ω 2 2ω 1

and

Ω 2ω 1

8.4.11 Use the methods of multiple scales and normal forms to determine a ﬁrstorder approximation to uR 1 C ω 21 u 1 C δ 1 u21 C 2δ 2 u 1 u 2 C δ 3 u22 D 2F1 cos Ω t uR 2 C ω 22 u 2 C δ 2 u21 C 2δ 3 u 1 u 2 C δ 4 u22 D 0 where ω 2 2ω 1

and

Ω ω1

235

9 TDOF Systems with Cubic Nonlinearities In this chapter, we consider two-degree-of-freedom nongyroscopic and gyroscopic systems with cubic nonlinearities. Nongyroscopic systems are treated in the following section, and gyroscopic systems are treated in Section 9.2.

9.1 Nongyroscopic Systems

We assume that the system can be modeled by (8.1), where A and D are 2 2 matrices and x and N are column vectors of length 2. Moreover, we assume that a linear transformation x D P u has been introduced so that (8.1) becomes (8.2) and that J and D are diagonal. The case in which J has a generic nonsemisimple structure is treated in Sections 9.1.5 and 9.1.6. Thus, we rewrite (8.2) as uR m C ω 2m u m C 2µ m uP m D

@V (u 1 , u 2 ) C 2 f m cos (Ω t C τ m ) @u m

(9.1)

where V D δ 1 u41 C δ 2 u31 u 2 C δ 3 u21 u22 C δ 4 u 1 u32 C δ 5 u42

(9.2)

Again, as a ﬁrst step, we cast (9.1) in complex-valued form using the transformation u m D ζ m C ζN m , uP m D i ω m ζ m ζN m zP m D i Ω z m ,

z m D f m e i (Ω tCτ m )

and obtain ζP m D i ω m ζ m µ m ζ m ζN m

i (z m C zN m ) 2ω m i @V ζ1 C ζN 1 , ζ2 C ζN 2 2ω m @u m

(9.3)

Before proceeding further, we must distinguish between two cases: Ω ω m (corresponding to the primary resonance of the mth mode) and Ω is away from both The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

236

9 TDOF Systems with Cubic Nonlinearities

ω 1 and ω 2 . We start with the latter case and consider the primary-resonance case in Section 9.1.4. When Ω is away from ω 1 and ω 2 , we use the transformation (8.8) and (8.10) and rewrite (9.3) as " i z1 C zN 1 3 ηP 1 D i ω 1 η 1 µ 1 (η 1 ηN 1 ) 4δ 1 η 1 C ηN 1 C 2 2ω 1 ω1 Ω 2 2 z1 C zN 1 z2 C zN 2 η C 3δ 2 η 1 C ηN 1 C 2 C η N C 2 2 ω1 Ω 2 ω 22 Ω 2 z1 C zN 2 z2 C zN 2 2 C 2δ 3 η 1 C ηN 1 C 2 η 2 C ηN 2 C 2 ω1 Ω 2 ω2 Ω 2 # 3 z2 C zN 2 Cδ 4 η 2 C ηN 2 C 2 (9.4) ω2 Ω 2 " i z1 C zN 1 3 ηP 2 D i ω 2 η 2 µ 2 (η 2 ηN 2 ) δ 2 η 1 C ηN 1 C 2 2ω 2 ω1 Ω 2 z1 C zN 1 2 z2 C zN 2 η C 2δ 3 η 1 C ηN 1 C 2 C η N C 2 2 ω1 Ω 2 ω 22 Ω 2 z1 C zN 2 z2 C zN 2 2 C 3δ 4 η 1 C ηN 1 C 2 η C η N C 2 2 ω1 Ω 2 ω 22 Ω 2 3 # z2 C zN 2 C4δ 5 η 2 C ηN 2 C 2 (9.5) ω2 Ω 2 In addition to the resonance terms η 1 , η 21 ηN 1 , and η 2 ηN 2 η 1 in (9.4) and the resonance terms η 2 , η 22 ηN 2 , and η 1 ηN 1 η 2 in (9.5), near-resonance terms occur when

ω 2 3ω 1 : Three-to-one internal resonance, ω 2 ω 1 : One-to-one internal resonance, Ω 3ω m : Subharmonic resonance of order one-third, Ω 1/3ω m : Superharmonic resonance of order three, Ω ω m ˙ ω n : Combination resonance.

Next, we present the normal forms of (9.4) and (9.5) for several combinations of the internal and external resonances. 9.1.1 The Case of No Internal Resonances

Substituting the near-identity transformation (8.14) into (9.4) and (9.5) and choosing the h m to eliminate the nonresonance terms, we obtain the following normal forms:

9.1 Nongyroscopic Systems

No External Resonance iσ 1 ξP1 D i ω 1 ξ1 ξ1 µ 1 ξ1 2ω 1 iσ 2 ξP2 D i ω 2 ξ2 ξ2 µ 2 ξ2 2ω 2

i 2ω 1 i 2ω 2

12δ 1 ξ12 ξN1 C 4δ 3 ξ2 ξN2 ξ1

(9.6)

4δ 3 ξ1 ξN1 ξ2 C 12δ 5 ξ22 ξN2

(9.7)

where 24δ 1 z1 zN 1 6δ 2 (z1 zN 2 C z2 zN 1 ) 4δ 3 z2 zN 2 C 2 C 2 (ω 21 Ω 2 )2 (ω 1 Ω 2 )(ω 22 Ω 2 ) (ω 2 Ω 2 )2 24δ 5 z2 zN 2 6δ 4 (z1 zN 2 C z2 zN 1 ) 4δ 3 z1 zN 1 C 2 C 2 σ2 D 2 2 2 2 2 2 (ω 2 Ω ) (ω 1 Ω )(ω 2 Ω ) (ω 1 Ω 2 )2 σ1 D

(9.8) (9.9)

Ω 3ω 2

iσ 1 i ξP1 D i ω 1 ξ1 12δ 1 ξ12 ξN1 C 4δ 3 ξ2 ξN2 ξ1 ξ1 µ 1 ξ1 2ω 1 2ω 1

(9.10)

iσ 2 i ξP2 D i ω 2 ξ2 4δ 3 ξ1 ξN1 ξ2 C 12δ 5 ξ22 ξN2 iΓ1 ξN22 ξ2 µ 2 ξ2 2ω 2 2ω 2 (9.11) where σ 1 and σ 2 are deﬁned in (9.8) and (9.9) and 3δ 4 z1 1 12δ 5 z2 Γ1 D C 2ω 2 ω 21 Ω 2 ω 22 Ω 2

(9.12)

Ω 1/3ω 1

iσ 1 i ξP1 D i ω 1 ξ1 12δ 1 ξ12 ξN1 C 4δ 3 ξ2 ξN2 ξ1 iΓ2 (9.13) ξ1 µ 1 ξ1 2ω 1 2ω 1 iσ 2 i ξP2 D i ω 2 ξ2 4δ 3 ξ1 ξN1 ξ2 C 12δ 5 ξ22 ξN2 (9.14) ξ2 µ 2 ξ2 2ω 2 2ω 2 where σ 1 and σ 2 are deﬁned in (9.8) and (9.9) and 4δ 1 z13 1 3δ 2 z12 z2 Γ2 D C 2 2 2 3 2ω 1 (ω 1 Ω ) (ω 1 Ω 2 )2 (ω 22 Ω 2 ) δ 4 z23 2δ 3 z1 z22 C C 2 (ω 1 Ω 2 )(ω 22 Ω 2 )2 (ω 22 Ω 2 )3

(9.15)

Ω ω 2 C 2ω 1

iσ 1 i ξP1 D i ω 1 ξ1 12δ 1 ξ12 ξN1 C 4δ 3 ξ2 ξN2 ξ1 iΓ3 ξN2 ξN1 ξ1 µ 1 ξ1 2ω 1 2ω 1 (9.16) iσ 2 i 4δ 3 ξ1 ξN1 ξ2 C 12δ 5 ξ22 ξN2 iΓ4 ξN12 ξP2 D i ω 2 ξ2 ξ2 µ 2 ξ2 2ω 2 2ω 2 (9.17)

237

238

9 TDOF Systems with Cubic Nonlinearities

where σ 1 and σ 2 are deﬁned in (9.8) and (9.9) and 3δ 2 z1 1 2δ 3 z2 Γ3 D C ω 1 ω 21 Ω 2 ω 22 Ω 2 3δ 2 z1 1 2δ 3 z2 Γ4 D C 2ω 2 ω 21 Ω 2 ω 22 Ω 2

(9.18) (9.19)

9.1.2 Three-to-One Autoparametric Resonance

The three-to-one internal resonance ω 2 3ω 1 produces the near-resonance term η 2 ηN 21 in (9.4) and the near-resonance term η 31 in (9.5). Hence, substituting the nearidentity transformation (8.14) into (9.4) and (9.5) and choosing the h m to eliminate the nonresonance terms, we obtain the following normal forms: No External Resonance

iσ 1 i ξP1 D i ω 1 ξ1 12δ 1 ξ12 ξN1 C 4δ 3 ξ2 ξN2 ξ1 C 3δ 2 ξ2 ξN12 ξ1 µ 1 ξ1 2ω 1 2ω 1 (9.20) iσ 2 i 4δ 3 ξ1 ξN1 ξ2 C 12δ 5 ξ22 ξN2 C δ 2 ξ13 ξ2 µ 2 ξ2 ξP2 D i ω 2 ξ2 2ω 2 2ω 2 (9.21) where σ 1 and σ 2 are deﬁned in (9.8) and (9.9). Ω 2ω 2 ω 1

iσ 1 ξP1 D i ω 1 ξ1 ξ1 µ 1 ξ1 2ω 1 i 12δ 1 ξ12 ξN1 C 4δ 3 ξ2 ξN2 ξ1 C 3δ 2 ξ2 ξN12 iΓ5 ξ22 2ω 1 iσ 2 ξP2 D i ω 2 ξ2 ξ2 µ 2 ξ2 2ω 2 i 4δ 3 ξ1 ξN1 ξ2 C 12δ 5 ξ22 ξN2 C δ 2 ξ13 iΓ6 ξ1 ξN2 2ω 2 where

2δ 3 zN 1 1 3δ 4 zN 2 C 2ω 1 ω 21 Ω 2 ω 22 Ω 2 2δ 3 z1 1 3δ 4 z2 Γ6 D C ω 2 ω 21 Ω 2 ω 22 Ω 2 Γ5 D

(9.22)

(9.23)

(9.24) (9.25)

The normal forms corresponding to the other external resonances can be obtained by simply adding the appropriate Γm terms to the right-hand sides of (9.20) and (9.21), as in Section 9.1.1.

9.1 Nongyroscopic Systems

9.1.3 One-to-One Internal Resonance

The one-to-one internal resonance corresponding to ω 2 ω 1 produces the nearresonance terms η 22 ηN 2 , η 21 ηN 2 , η 1 ηN 1 η 2 , and η 22 ηN 1 in (9.4) and the near-resonance terms η 21 ηN 1 , η 2 ηN 2 η 1 , η 22 ηN 1 , and η 21 ηN 2 in (9.5). Consequently, substituting the nearidentity transformation (8.14) into (9.4) and (9.5) and choosing the h m to eliminate the nonresonance terms, we obtain the following normal form in the absence of external resonances: iσ 1 iσ 3 i 3δ 2 2ξ1 ξN1 ξ2 C ξ12 ξN2 ξP1 D i ω 1 ξ1 ξ1 ξ2 µ 1 ξ1 2ω 1 2ω 1 2ω 1 (9.26) C12δ 1 ξ12 ξN1 C 2δ 3 2ξ2 ξN2 ξ1 C ξ22 ξN1 C 3δ 4 ξ22 ξN2 iσ 2 iσ 3 i 2δ 3 2ξ1 ξN1 ξ2 C ξ12 ξN2 ξP2 D i ω 2 ξ2 ξ2 ξ1 µ 2 ξ2 2ω 2 2ω 2 2ω 2 (9.27) C3δ 2 ξ12 ξN1 C 3δ 4 2ξ2 ξN2 ξ1 C ξ22 ξN1 C 12δ 5 ξ22 ξN2 where σ3 D

6δ 2 z1 zN 1 4δ 3 (z1 zN 2 C zN 1 z2 ) 6δ 4 z2 zN 2 C 2 C 2 (ω 21 Ω 2 )2 (ω 1 Ω 2 )(ω 22 Ω 2 ) (ω 2 Ω 2 )2

(9.28)

The normal forms of (9.4) and (9.5) in the presence of external resonances can be obtained by simply adding the appropriate Γm terms to the right-hand sides of (9.26) and (9.27), as in Sections 9.1.1 and 9.1.2. However, because the analysis in this section is valid only when Ω is away from ω 1 and ω 2 , one needs to make sure that the external resonance does not produce a primary resonance. 9.1.4 Primary Resonances

Primary resonances can easily be treated by scaling the excitation at O() so that its effect balances the effects of the damping and the nonlinearity. Consequently, substituting (8.60) into (9.3), using (9.2), and choosing the h m to eliminate the nonresonance terms, we obtain the following normal forms: No Internal Resonances, Ω ω n

i ξP1 D i ω 1 ξ1 µ 1 ξ1 2ω 1 i ξP2 D i ω 2 ξ2 µ 2 ξ2 2ω 2

i 12δ 1 ξ12 ξN1 C 4δ 3 ξ2 ξN2 ξ1 z1 δ 1n 2ω 1 i 4δ 3 ξ1 ξN1 ξ2 C 12δ 5 ξ22 ξN2 z2 δ 2n 2ω 2

(9.29) (9.30)

239

240

9 TDOF Systems with Cubic Nonlinearities

Three-to-One Internal Resonance, Ω ω n

i i ξP1 D i ω 1 ξ1 µ 1 ξ1 12δ 1 ξ12 ξN1 C 4δ 3 ξ2 ξN2 ξ1 C 3δ 2 ξ2 ξN12 z1 δ 1n 2ω 1 2ω 1 (9.31) i i 4δ 3 ξ1 ξN1 ξ2 C 12δ 5 ξ22 ξN2 C δ 2 ξ13 ξP2 D i ω 2 ξ2 µ 2 ξ2 z2 δ 2n 2ω 2 2ω 2 (9.32) One-to-One Internal Resonance, Ω ω n

i ξP1 D i ω 1 ξ1 µ 1 ξ1 12δ 1 ξ12 ξN1 C 3δ 2 2ξ1 ξN1 ξ2 C ξ12 ξN2 2ω 1 i z1 δ 1n C2δ 3 2ξ2 ξN2 ξ1 C ξ22 ξN1 C 3δ 4 ξ22 ξN2 2ω 1 i ξP2 D i ω 2 ξ2 µ 2 ξ2 3δ 2 ξ12 ξN1 C 2δ 3 2ξ1 ξN1 ξ2 C ξ12 ξN2 2ω 2 i z2 δ 2n C3δ 4 2ξ2 ξN2 ξ1 C ξ22 ξN1 C 12δ 5 ξ22 ξN2 2ω 2

(9.33)

(9.34)

9.1.5 A Nonsemisimple One-to-One Internal Resonance

In this section, we treat a two-degree-of-freedom system whose linear undamped operator has a generic nonsemisimple structure; that is, we treat the system uR 1 C ω 2 u 1 D 2µ 1 µP 1 α 11 u31 α 12 u21 u 2 α 13 u 1 u22 α 14 u32

(9.35)

uR 2 C ω 2 u 2 D 2µ 2 µP 2 u 1 α 21 u31 α 22 u21 u 2 α 23 u 1 u22 α 24 u32 (9.36) Again, as a ﬁrst step in the application of the method of normal forms, we use the transformation u 1 D ζ1 C ζN 1 ,

uP 1 D i ω ζ1 ζN 1

(9.37)

u 2 D ζ2 C ζN 2 ,

uP 2 D i ω ζ2 ζN 2

(9.38)

to recast (9.35) and (9.36) in the complex-valued form ζP 1 D i ωζ1 µ 1 ζ1 ζN 1 3 2 i h C α 11 ζ1 C ζN 1 C α 12 ζ1 C ζN 1 ζ2 C ζN 2 2ω 2 3 i Cα 13 ζ1 C ζN 1 ζ2 C ζN 2 C α 14 ζ2 C ζN 2

(9.39)

9.1 Nongyroscopic Systems

3 i i h ζP 2 D i ωζ2 µ 2 ζ2 ζN 2 C α 21 ζ1 C ζN 1 ζ1 C ζN 1 C 2ω 2ω 2 2 N N N C α 22 ζ1 C ζ1 ζ2 C ζ2 C α 23 ζ1 C ζ1 ζ2 C ζN 2 i 3 Cα 24 ζ2 C ζN 2

(9.40)

In determining the normal form of (9.39) and (9.40), one can follow one of two approaches. In the ﬁrst approach, which we followed in Sections 7.3.1–7.3.3, we determine the normal form assuming that ζ1 and ζ2 are the same order and then we use the fact that ζ2 is much larger than ζ1 to simplify the resulting forms. In the second approach, we use the fact that ζ2 is much larger than ζ1 to ﬁrst simplify (9.39) and (9.40) and then determine the normal form of the simpliﬁed equations. Following the ﬁrst approach, we note that ζ1 , ζ2 , ζ12 ζN 1 , ζ12 ζN 2 , ζ1 ζN 1 ζ2 , ζ22 ζN 2 , ζ2 ζN 2 ζ1 , and ζ22 ζN 2 are resonance terms. Hence, introducing the near-identity transformation ζ m D η m C h m (η 1 , η 2 , ηN 1 , ηN 2 )

(9.41)

into (9.39) and (9.40) and choosing h 1 and h 2 to eliminate the nonresonance terms, we obtain the normal form i 3α 11 η 21 ηN 1 C 2α 12 η 1 ηN 1 η 2 C α 12 η 21 ηN 2 2ω Cα 13 η 22 ηN 1 C 2α 13 η 1 η 2 ηN 2 C 3α 14 η 22 ηN 2

(9.42)

i i 3α 21 η 21 ηN 1 C 2α 22 η 1 ηN 1 η 2 η1 C 2ω 2ω Cα 22 η 21 ηN 2 C 2α 23 η 2 ηN 2 η 1 C α 23 η 22 ηN 1 C 3α 24 η 22 ηN 2

(9.43)

ηP 1 D i ωη 1 µ 1 η 1 C

ηP 2 D i ωη 2 µ 2 η 2 C

Next, we use the fact that η 2 is much larger than η 1 and introduce new variables deﬁned by η 1 D ξ1

and

η 2 D 1λ 2 ξ2

(9.44)

where ξ1 and ξ2 are O(1) and λ 2 > 0. Moreover, we assume that the damping is weak and hence scale the µ n as µ n D λ 1 µO n

(9.45)

where the µO n are O(1). Substituting (9.44) and (9.45) into (9.42) and (9.43), we have i h 2 ξP1 D i ωξ1 λ 1 µO 1 ξ1 C 3 α 11 ξ12 ξN1 C 2 2λ 2 α 12 ξ1 ξN1 ξ2 2ω C 2λ 2 α 12 ξ12 ξN2 C 22λ 2 α 13 ξ22 ξN1 C 2 22λ 2 α 13 ξ1 ξ2 ξN2 i C3 23λ 2 α 14 ξ 2 ξN2 2

(9.46)

241

242

9 TDOF Systems with Cubic Nonlinearities

i λ2 i h 2Cλ 2 ξP2 D i ωξ2 λ 1 µO 2 ξ2 C α 21 ξ12 ξN1 C 2 2 α 22 ξ1 ξN1 ξ2 ξ1 C 3 2ω 2ω i C 2 α 22 ξ 2 ξN2 C 2 2λ 2 α 23 ξ2 ξN2 ξ1 C 2λ 2 α 23 ξ 2 ξN1 C 3 22λ 2 α 24 ξ 2 ξN2 1

2

2

(9.47) Balancing the damping term and the dominant nonlinear term in (9.46), namely, the term proportional to α 14 , we have λ 1 D 2 3λ 2

(9.48)

Balancing the damping term and the term that causes the instability in (9.47), namely, the term i(2ω)1 λ 2 ξ1 , we have λ1 D λ2

(9.49)

We note that due to (9.48), the dominant nonlinear term in (9.47), namely, the term proportional to α 24 , is small compared with the damping term. Solving (9.48) and (9.49), we obtain λ1 D λ2 D

1 2

(9.50)

and hence (9.46) and (9.47) simplify to 3i 1/2 ξP1 D i ωξ1 1/2 µO 1 ξ1 C α 14 ξ22 ξN2 C 2ω

(9.51)

i 1/2 ξP2 D i ωξ2 1/2 µO 2 ξ2 C ξ1 C 2ω

(9.52)

Equations 9.51 and 9.52 agree with those obtained by Tezak, Nayfeh, and Mook (1982) by using the method of multiple scales. In the second approach, we use the fact that the damping and nonlinearity are weak and that ζ2 is much larger than ζ1 to ﬁrst simplify (9.39) and (9.40) and then determine the normal form of the simpliﬁed equations. If we assume that ζ1 D O(), where is a small nondimensional parameter, then ζ2 D O( 1λ 2 ), where λ 2 > 0. Moreover, for the case of weak damping, µ m D O( λ 1 ), where λ 1 is positive. Using these scalings, we introduce new scaled variables deﬁned by µ m D λ 1 µO n , ζ1 D η 1 , and ζ2 D 1λ 2 η 2 , where the η n and µO n are O(1), in (9.39) and (9.40) and obtain i 2 α 11 (η 1 C ηN 1 )3 ηP 1 D i ωη 1 λ 1 µO 1 (η 1 ηN 1 ) C 2ω i C 23λ 2 α 14 (η 2 C ηN 2 )3 C 2λ 2 α 12 (η 1 C ηN 1 )2 (η 2 C ηN 2 ) C 22λ 2 α 13 (η 1 C ηN 1 ) (η 2 C ηN 2 )2

(9.53)

9.1 Nongyroscopic Systems

i λ2 ηP 2 D i ωη 2 λ 1 µO 2 (η 2 ηN 2 ) C (η 1 C ηN 1 ) 2ω h i C 2Cλ 2 α 21 (η 1 C ηN 1 )3 C 2 α 22 (η 1 C ηN 1 )2 (η 2 C ηN 2 ) 2ω i C 2λ 2 α 23 (η 1 C ηN 1 ) (η 2 C ηN 2 )2 C 22λ 2 α 24 (η 2 C ηN 2 )3

(9.54)

Balancing the damping term and the dominant nonlinear term in (9.53), namely, the term proportional to α 14 , we have λ 1 D 2 3λ 2

(9.55)

Balancing the damping term and the source of the instability in (9.54), namely, the term i(2ω)1 λ 2 (η 1 C ηN 1 ), we have λ1 D λ2

(9.56)

Solving (9.55) and (9.56) yields λ1 D λ2 D

1 2

(9.57)

Hence, (9.53) and (9.54) become ηP 1 D i ωη 1 1/2 µO 1 (η 1 ηN 1 ) C

i 1/2 α 14 (η 2 C ηN 2 )3 C 2ω

(9.58)

ηP 2 D i ωη 2 1/2 µO 2 (η 2 ηN 2 ) C

i 1/2 (η 1 C ηN 1 ) C 2ω

(9.59)

To simplify (9.58) and (9.59), we introduce the near-identity transformation η m D ξm C 1/2 h m ξ1 , ξ2 , ξN1 , ξN2 C

(9.60)

and choose h 1 and h 2 to eliminate the resonance terms. We note that the terms proportional to η 1 and η 22 ηN 2 are resonance terms in (9.58), and hence, we can choose h 1 to eliminate all of the other terms, resulting in the normal form 3i 1/2 ξP1 D i ωξ1 1/2 µO 1 ξ1 C α 14 ξ22 ξN2 C 2ω

(9.61)

Similarly, we note that η 2 and η 1 are resonance terms in (9.59), and hence, we can choose h 2 to eliminate all of the other terms, resulting in the normal form i 1/2 ξP2 D i ωξ2 1/2 µO 2 ξ2 C ξ1 C 2ω

(9.62)

Equations 9.61 and 9.62 agree with (9.51) and (9.52) and with those obtained by Tezak, Nayfeh, and Mook (1982) by using the method of multiple scales.

243

244

9 TDOF Systems with Cubic Nonlinearities

9.1.6 A Parametrically Excited System with a Nonsemisimple Linear Structure

We consider the parametrically excited two-degree-of-freedom system uR 1 C ω 2 u 1 C 2µ 1 uP 1 C α 11 u31 C α 12 u21 u 2 C α 13 u 1 u22 C α 14 u32 C 2 (F11 u 1 C F12 u 2 ) cos Ω t D 0

(9.63)

uR 2 C ω 2 u 2 C 2µ 2 uP 2 C u 1 C α 21 u31 C α 22 u21 u 2 C α 23 u 1 u22 C α 24 u32 C 2 (F21 u 1 C F22 u 2 ) cos Ω t D 0

(9.64)

when the damping and nonlinearities are weak. Again, as a ﬁrst step, we transform (9.63) and (9.64) into two ﬁrst-order complex-valued equations. Using (9.37) and (9.38) and the transformation z D eiΩ t

(9.65)

we rewrite (9.63) and (9.64) as 3 3 i h ζP 1 D i ωζ1 µ 1 ζ1 ζN 1 C α 11 ζ1 C ζN 1 C α 14 ζ2 C ζN 2 2ω 2 2 i Cα 12 ζ1 C ζN 1 ζ2 C ζN 2 C α 13 ζ1 C ζN 1 ζ2 C ζN 2 i F11 ζ1 C ζN 1 C F12 ζ2 C ζN 2 (z C zN ) (9.66) 2ω 2 i h ζP 2 D i ωζ2 µ 2 ζ2 ζN 2 C ζ1 C ζN 1 C α 22 ζ1 C ζN 1 ζ2 C ζN 2 2ω 3 2 3 i Cα 21 ζ1 C ζN 1 C α 23 ζ1 C ζN 1 ζ2 C ζN 2 C α 24 ζ2 C ζN 2 C

i (9.67) F21 ζ1 C ζN 1 C F22 ζ2 C ζN 2 (z C zN ) 2ω Next, we determine the normal forms of (9.66) and (9.67) for weak damping and nonlinearities and two cases of parametric resonance: principal parametric resonance (i.e., Ω 2ω) and fundamental parametric resonance (i.e., Ω ω). C

The Case of Principal Parametric Resonance As a ﬁrst step, we scale the variables. If we assume that ζ1 D O(), where is a small nondimensional parameter, then, due to the nonsemisimple structure of the linear operator, ζ2 D O( 1λ 2 ), where λ 2 > 0. Moreover, because the damping is weak, µ n D O( λ 1 ), where λ 1 > 0. Balancing the damping terms and the dominant nonlinear terms in (9.66) and (9.67), as done in the preceding section, we ﬁnd that λ 1 D λ 2 D 1/2. For the case of principal parametric resonance, Ω 2ω and the terms z ζN 1 and z ζN 2 are near-resonance terms. In order that the damping and nonlinearity balance the dominant parametric resonance terms, F m D O(). Thus, we introduce new variables deﬁned by

ζ1 D η 1 ,

ζ2 D 1/2 η 2 ,

µ n D 1/2 µO n ,

Fm n D f m n

9.1 Nongyroscopic Systems

into (9.66) and (9.67) and obtain i 1/2 α 14 (η 2 C ηN 2 )3 ηP 1 D i ωη 1 1/2 µO 1 (η 1 ηN 1 ) C 2ω N C C f 12 (η 2 C ηN 2 ) (z C z) ηP 2 D i ωη 2 1/2 µO 2 (η 2 ηN 2 ) C

i 1/2 (η 1 C ηN 1 ) C 2ω

(9.68) (9.69)

To simplify (9.68) and (9.69), we use the near-identity transformation (9.60), choose the h m to eliminate the nonresonance terms, and obtain 3i 1/2 i 1/2 ξP1 D i ωξ1 1/2 µO 1 ξ1 C α 14 ξ22 ξN2 C f 12 z ξN2 C 2ω 2ω

(9.70)

i 1/2 ξP2 D i ωξ2 1/2 µO 2 ξ2 C ξ1 C 2ω

(9.71)

Equations 9.70 and 9.71 agree with those obtained by Tezak, Nayfeh, and Mook (1982) by using the method of multiple scales. Moreover, when µO n D 0, (9.70) and (9.71) agree with those obtained by Namachchivaya and Malhotra (1992). In the absence of the parametric resonance, (9.70) and (9.71) reduce to (9.61) and (9.62). Furthermore, in the absence of the nonlinearity, (9.70) and (9.71) reduce to (7.86) and (7.87). Fundamental Parametric Resonance In this case, Ω ω and z zN ζ1 and z zN ζ2 are resonance terms and z 2 ζN 1 and z 2 ζN 2 are near-resonance terms. Hence, we scale F m n to be O( 1/2 ) so that the damping and nonlinear terms balance the parametric resonance terms. Therefore, we introduce new scaled variables deﬁned by

ζ1 D η 1 ,

ζ2 D 1/2 η 2 ,

µ n D 1/2 µO n ,

and

F m n D 1/2 f m n

into (9.66) and (9.67) and obtain i N 1/2 µO 1 (η 1 ηN 1 ) f 12 (η 2 C ηN 2 ) (z C z) 2ω α 14 (η 2 C ηN 2 )3 C f 11 (η 1 C ηN 1 ) (z C zN ) C

ηP 1 D i ωη 1 C C

i 1/2 2ω

(9.72)

ηP 2 D i ωη 2 1/2 µO 2 (η 2 ηN 2 ) C

i 1/2 η 1 C ηN 1 C f 22 (η 2 C ηN 2 ) (z C z) N C 2ω

(9.73)

245

246

9 TDOF Systems with Cubic Nonlinearities

To simplify (9.72) and (9.73), we introduce the transformation η 1 D ξ1 C h 11 ξn , ξN n , z, zN C 1/2 h 12 ξn , ξN n , z, zN C

(9.74)

η 2 D ξ2 C 1/2 h 22 ξn , ξN n , z, zN C

(9.75)

and choose the h m n so that ξP1 D i ωξ1 C 1/2 g 1 ξn , ξN n , z, zN C

(9.76)

ξP2 D i ωξ2 C 1/2 g 2 ξn , ξN n , z, zN C

(9.77)

have the simplest possible form. Substituting (9.74)–(9.77) into (9.72) and (9.73), using (9.65), and equating coefﬁcients of like powers of , we obtain

@h 11 @h 11 @h 11 N @h 11 @h 11 N @h 11 ξ1 C ξ2 h 11 C i Ω ξ1 ξ2 z zN @ξ1 @ξ2 @z @ zN @ ξN1 @ ξN2 i (9.78) f 12 ξ2 C ξN2 (z C zN ) D 2ω @h 12 @h 12 N @h 12 @h 12 N ξ1 C ξ2 h 12 g1 C i ω ξ1 ξ2 @ξ1 @ξ2 @ ξN1 @ ξN2 @h 11 @h 11 @h 11 @h 11 @h 12 @h 12 C iΩ gN 1 g2 gN 2 z zN D g 1 N @z @ zN @ξ1 @ξ2 @ ξ1 @ ξN2

i C f 12 h 22 C hN 22 (z C zN ) µO 1 ξ1 C h 11 ξN1 hN 11 2ω

i 3 i h (9.79) C N α 14 ξ2 C ξN2 C f 11 ξ1 C ξN1 C h 11 C hN 11 (z C z) 2ω @h 22 @h 22 N @h 22 @h 22 N ξ1 C ξ2 h 22 ξ1 ξ2 g2 C i ω @ξ1 @ξ2 @ ξN1 @ ξN2 i h @h 22 @h 22 C iΩ z zN D µO 2 ξ2 ξN2 C ξ1 C ξN1 @z @ zN 2ω i N (9.80) Ch 11 C hN 11 C f 22 ξ2 C ξN2 (z C z)

iω

The right-hand side of (9.78) suggests seeking h 11 in the form h 11 D Γ1 ξ2 z C Γ2 ξ2 zN C Γ3 ξN2 z C Γ4 ξN2 zN

(9.81)

9.1 Nongyroscopic Systems

Substituting (9.81) into (9.78) and equating each of the coefﬁcients of ξ2 z, ξ2 zN , ξN2 z, and ξN2 zN on both sides, we obtain f 12 f 12 , Γ2 D , 2ωΩ 2ωΩ f 12 f 12 Γ3 D , Γ4 D 2ω(Ω 2ω) 2ω(Ω C 2ω)

Γ1 D

Substituting (9.81) into (9.80) and using (9.82) yields @h 22 N @h 22 @h 22 N @h 22 ξ1 C ξ2 h 22 ξ1 ξ2 g2 C i ω @ξ1 @ξ2 @ ξN1 @ ξN2 @h 22 @h 22 C iΩ z zN D µO 2 ξ2 ξN2 @z @ zN i f 12 ξ1 C ξN1 C ξ2 z C ξN2 zN C 2ω Ω (Ω C 2ω) f 12 ξ2 zN C ξN2 z C f 22 ξ2 C ξN2 (z C zN ) C Ω (Ω 2ω)

(9.82)

(9.83)

Equating g 2 to the resonance and near-resonance terms on the right-hand side of (9.83), we have g 2 D µO 2 ξ2 C

i ξ1 2ω

(9.84)

Then, we seek h 22 in the form h 22 D Γ5 ξN2 C Γ6 ξN1 C Γ7 ξ2 z C Γ8 ξ2 zN C Γ9 ξN2 z C Γ10 ξN2 zN

(9.85)

Substituting (9.85) into (9.83), using (9.84), and equating each of the coefﬁcients of ξN1 , ξN2 , ξ2 z, ξ2 zN , ξN2 z, and ξN2 zN on both sides, we obtain i µO 2 1 , Γ6 D 2 , 2ω 4ω f 22 f 12 D C , 2ωΩ 2ωΩ 2 (Ω C 2ω) f 22 f 12 D , 2ωΩ 2ωΩ 2 (Ω 2ω) f 22 f 12 D , C 2ω(Ω 2ω) 2ωΩ (Ω 2ω)2 f 22 f 12 D 2ω(Ω C 2ω) 2ωΩ (Ω C 2ω)2

Γ5 D Γ7 Γ8 Γ9 Γ10

(9.86)

247

248

9 TDOF Systems with Cubic Nonlinearities

Substituting (9.81), (9.82), (9.85), and (9.86) into (9.79) and using (9.84), we obtain h 12 @h 12 N @h 12 @h 12 N ξ1 C ξ2 h 12 ξ1 ξ2 @ξ1 @ξ2 @ ξN1 @ ξN2 @h 12 @h 12 C iΩ zN @z @ zN i i N ξ1 ξ1 C (Γ3 z C Γ4 zN ) µO 2 ξN2 C D (Γ1 z C Γ2 zN ) µO 2 ξ2 2ω 2ω µO 1 f 12 (Ω C ω) ξ2 z ξN2 zN µO 1 ξ1 ξN1 ωΩ (Ω C 2ω) µO 1 f 12 (Ω ω) ξ2 zN ξN2 z C ωΩ (Ω 2ω) f 22 i µO 2 N f 12 i f 12 (z C zN ) ξ2 ξ2 C C 2 C 2ω 2ω Ω (Ω C 2ω) Ω (Ω C 2ω)2 f 22 1 f 12 N N ξ1 C ξ1 C ξ2 z C ξ2 zN C 2 4ω 2 Ω (Ω 2ω) Ω (Ω 2ω)2 i f 11 ξ1 C ξN1 (z C zN ) ξ2 zN C ξN2 z C 2ω f 11 f 12 C N ξ2 z C ξN2 zN (z C z) Ω (Ω C 2ω) 3 f 11 f 12 ξ2 zN C ξN2 z (z C z) N C α 14 ξ2 C ξN2 C Ω (Ω 2ω) (9.87)

g1 C i ω

Because we are stopping at this order, we do not need to explicitly calculate h 12 . Then, equating g 1 to the resonance and near-resonance terms in (9.87), we obtain 3i g 1 D µO 1 ξ1 C α 14 ξ22 ξN2 2ω 2 f 12 1 2 f 12 ( f 11 C f 22 ) i 1 C z zN ξ2 C C 2ω Ω 2 (Ω C 2ω)2 (Ω 2ω)2 Ω 2 4ω 2 2 f 12 ( f 11 C f 22 ) i f 12 z 2 ξN2 C C 2 (9.88) 2ω Ω (Ω 2ω) Ω (Ω 2ω)2 Finally, substituting (9.81) and (9.82) into (9.74) yields " # ξN2 z ξN2 zN f 12 ξ2 z ξ2 zN η 1 D ξ1 C C C 2ω Ω Ω Ω 2ω Ω C 2ω

(9.89)

9.2 Gyroscopic Systems

Then, substituting (9.88) and (9.84) into (9.76) and (9.77), we obtain the normal form 1/2 2 f 12 ( f 11 C f 22 ) f 12 Pξ1 D i ωξ1 1/2 µO 1 ξ1 C i z 2 ξN2 C 2 2ω Ω (Ω 2ω) Ω (Ω 2ω)2 2 2 f 12 f 12 2 f 12 ( f 11 C f 22 ) ξ2 C C C Ω 2 4ω 2 Ω 2 (Ω C 2ω)2 Ω 2 (Ω 2ω)2 C3α 14 ξ22 ξN2 C (9.90) i 1/2 ξP2 D i ωξ2 1/2 µO 2 ξ2 C ξ1 C 2ω

(9.91)

because z zN D 1. In the absence of the parametric excitation (i.e., f m n D 0), (9.90) and (9.91) reduce to (9.61) and (9.62). In the absence of the nonlinearity (i.e., α 14 D 0), (9.90) and (9.91) reduce to (7.134) and (7.135) when f 13 D 0 and f 32 D 0.

9.2 Gyroscopic Systems

Again, for simplicity, we consider a two-degree-of-freedom system to illustrate the method. Speciﬁcally, we consider uR 1 C λ 1 uP 2 C α 1 u 1 D 4δ 1 u31 C 3δ 2 u21 u 2 C 2δ 3 u 1 u22 C δ 4 u32 C 2 f 1 cos (Ω t C τ 1 ) uR 2 λ 2 uP 1 C α 2 u 2 D δ 2 u31 C 2δ 3 u21 u 2 C 3δ 4 u 1 u22 C 4δ 5 u32

(9.92)

C 2 f 2 cos (Ω t C τ 2 )

(9.93)

Using the transformation (7.144)–(7.147) and following steps similar to those used in Section 7.4, we cast (9.92) and (9.93) in the complex-valued forms λ1 i ω 1 (α 1 ω 22 ) (z1 C zN 1 ) (z2 C zN 2 ) ζP 1 D i ω 1 ζ1 C 2 2 2 2α 1 (ω 2 ω 1 ) 2(ω 2 ω 21 ) 3 2 C χ 11 ζ1 C ζN 1 C ζ2 C ζN 2 C χ 12 ζ1 C ζN 1 C ζ2 C ζN 2 i(α 1 ω 22 ) i(α 1 ω 21 ) ζ1 ζN 1 C ζ2 ζN 2 λ1 ω1 λ1 ω2 2 i(α 1 ω 22 ) 3 i(α 1 ω 1 ) C χ 14 ζ1 ζN 1 C ζ2 ζN 2 λ1 ω1 λ1 ω2 i(α 1 ω 21 ) ζ1 ζN 1 C χ 13 ζ1 C ζN 1 C ζ2 C ζN 2 λ1 ω1 2 2 i(α 1 ω 2 ) ζ2 ζN 2 C λ1 ω2

(9.94)

249

250

9 TDOF Systems with Cubic Nonlinearities

λ1 i ω 2 (α 1 ω 21 ) (z1 C zN 1 ) C (z2 C zN 2 ) ζP 2 D i ω 2 ζ2 2α 1 (ω 22 ω 21 ) 2(ω 22 ω 21 ) 3 2 C χ 21 ζ1 C ζN 1 C ζ2 C ζN 2 C χ 22 ζ1 C ζN 1 C ζ2 C ζN 2 i(α 1 ω 22 ) i(α 1 ω 21 ) ζ1 ζN 1 C ζ2 ζN 2 λ1 ω1 λ1 ω2 i(α 1 ω 22 ) 3 i(α 1 ω 21 ) C χ 24 ζ1 ζN 1 C ζ2 ζN 2 λ1 ω1 λ1 ω2 i(α 1 ω 21 ) ζ1 ζN 1 C χ 23 ζ1 C ζN 1 C ζ2 C ζN 2 λ1 ω1 2 i(α 1 ω 22 ) ζ2 ζN 2 C λ1 ω2

(9.95)

where χ 11 D

4 i δ 1 ω 1 (α 1 ω 22 ) α 1 δ 2 λ 1 3i δ 2 ω 1 (α 1 ω 22 ) 2α 1 δ 3 λ 1 , χ 12 D , 2 2 2α 1 (ω 2 ω 1 ) 2α 1 (ω 22 ω 21 )

χ 13 D

2i δ 3 ω 1 (α 1 ω 22 ) 3α 1 δ 4 λ 1 i δ 4 ω 1 (α 1 ω 22 ) 4α 1 δ 5 λ 1 , χ 14 D , 2 2 2α 1 (ω 2 ω 1 ) 2α 1 (ω 22 ω 21 )

χ 21 D

α 1 δ 2 λ 2 4 i δ 1 ω 2 (α 1 ω 21 ) 2α 1 δ 3 λ 1 3i δ 2 ω 2 (α 1 ω 21 ) , χ 22 D , 2 2 2α 1 (ω 2 ω 1 ) 2α 1 (ω 22 ω 21 )

χ 23 D

3α 1 δ 4 λ 1 2i δ 3 ω 2 (α 1 ω 21 ) 4α 1 δ 5 λ 1 i δ 4 ω 2 (α 1 ω 21 ) , χ 24 D 2 2 2α 1 (ω 2 ω 1 ) 2α 1 (ω 22 ω 21 ) (9.96)

There are two cases to consider: Ω ω m (corresponding to the primary resonance of the mth mode) and Ω is away from ω 1 and ω 2 . The case of primary resonance is discussed next and the other case (secondary resonances) is discussed in Section 9.2.2. 9.2.1 Primary Resonances

As in Section 8.2.1, we treat this case by scaling z1 and z2 at O(). Then, substituting the near-identity transformation (8.60) into (9.94) and (9.95) and choosing the h m to eliminate the nonresonance terms, we obtain the following normal forms: No Internal Resonance, Ω ω n

ξP1 D i ω 1 ξ1 C S11 ξ12 ξN1 C S12 ξ2 ξN2 ξ1 C Γ1 δ 1n

(9.97)

ξP2 D i ω 2 ξ2 C S21 ξ1 ξN1 ξ2 C S22 ξ22 ξN2 C Γ2 δ 2n

(9.98)

9.2 Gyroscopic Systems

where S11 D 3χ 11 C S12 D 6χ 11 C

S21 D 6χ 21 C

(9.99)

2i χ 12 (α 1 ω 21 ) 2χ 13 (α 1 ω 22 )2 6i χ 14 (α 1 ω 21 )(α 1 ω 22 )2 C C λ1 ω1 λ 21 ω 22 λ 31 ω 1 ω 22 (9.100) 2i χ 22 (α 1 ω 22 ) 2χ 23 (α 1 ω 21 )2 6i χ 24 (α 1 ω 21 )2 (α 1 ω 22 ) C C λ1 ω2 λ 21 ω 21 λ 31 ω 21 ω 2 (9.101)

S22 D 3χ 21 C Γ1 D

i χ 12 (α 1 ω 21 ) χ 13 (α 1 ω 21 )2 3i χ 14 (α 1 ω 21 )3 C C λ1 ω1 λ 21 ω 21 λ 31 ω 31

i χ 22 (α 1 ω 22 ) χ 23 (α 1 ω 22 )2 3i χ 24 (α 1 ω 22 )3 C C 2 2 λ1 ω2 λ1 ω2 λ 31 ω 32

i ω 1 (α 2 ω 21 )z1 λ 1 α 1 z2 , 2α 1 (ω 22 ω 21 )

Γ2 D

(9.102)

i ω 2 (α 1 ω 21 )z1 λ 1 α 1 z2 2α 1 (ω 22 ω 21 ) (9.103)

Three-to-One Internal Resonance, Ω ω n

ξP1 D i ω 1 ξ1 C S11 ξ12 ξN1 C S12 ξ2 ξN2 ξ1 C S13 ξ2 ξN12 C Γ1 δ 1n

(9.104)

ξP2 D i ω 2 ξ2 C S21 ξ1 ξN1 ξ2 C S22 ξ22 ξN2 C S23 ξ13 C Γ2 δ 2n

(9.105)

where S11 , S12 , S21 , S22 , Γ1 , and Γ2 are deﬁned in (9.99)–(9.103) and α 1 ω 22 3i χ 14 (α 1 ω 21 )2 (α 1 ω 22 ) α 1 ω 21 2ω 2 ω1 λ 31 ω 21 ω 2 2χ 13 (α 1 ω 21 ) α 1 ω 22 α 1 ω 21 C (9.106) 2 λ ω1 ω2 2ω 1

S13 D 3χ 11 C

S23 D χ 21 C

2i χ 12 λ1

i χ 22 (α 1 ω 21 ) χ 23 (α 1 ω 21 )2 i χ 24 (α 1 ω 21 )3 2 2 λ1 ω1 λ1 ω1 λ 31 ω 31

(9.107)

We note that the transformation (7.144)–(7.147) is not valid when ω 2 ω 1 and hence (9.94) and (9.95) do not apply in this case. Consequently, the case of a oneto-one internal resonance needs special treatment. 9.2.2 Secondary Resonances in the Absence of Internal Resonances

In this case, we scale z1 and z2 at O(1). Then, we introduce the transformation (8.81) and (8.82) into (9.94) and (9.95) to eliminate the terms involving z n and zN n

251

252

9 TDOF Systems with Cubic Nonlinearities

at O(1) and obtain ηP 1 D i ω 1 η 1 C χ 11 [η 1 C ηN 1 C η 2 C ηN 2 C b 1 (z1 C zN 1 ) C b 2 (z2 zN 2 )]3 C χ 12 [η 1 C ηN 1 C η 2 C ηN 2 C b 1 (z1 C zN 1 ) C b 2 (z2 zN 2 )]2 i(α 1 ω 22 ) i(α 1 ω 21 ) (η 1 ηN 1 ) C (η 2 ηN 2 ) λ1 ω1 λ1 ω2 Cb 3 (z1 zN 1 ) C b 4 (z2 C zN 2 ) C χ 13 [η 1 C ηN 1 C η 2 C ηN 2 C b 1 (z1 C zN 1 ) C b 2 (z2 zN 2 )] i(α 1 ω 22 ) i(α 1 ω 21 ) (η 1 ηN 1 ) C (η 2 ηN 2 ) λ1 ω1 λ1 ω2 2 Cb 3 (z1 zN 1 ) C b 4 (z2 C zN 2 )

i(α 1 ω 21 ) i(α 1 ω 22 ) (η 1 ηN 1 ) C (η 2 ηN 2 ) λ1 ω1 λ1 ω2 3 Cb 3 (z1 zN 1 ) C b 4 (z2 C zN 2 )

C χ 14

(9.108)

ηP 2 D i ω 2 η 2 C χ 21 [η 1 C ηN 1 C η 2 C ηN 2 C b 1 (z1 C zN 1 ) C b 2 (z2 zN 2 )]3 C χ 22 [η 1 C ηN 1 C η 2 C ηN 2 C b 1 (z1 C zN 1 ) C b 2 (z2 zN 2 )]2 i(α 1 ω 22 ) i(α 1 ω 21 ) (η 1 ηN 1 ) C (η 2 ηN 2 ) λ1 ω1 λ1 ω2 Cb 3 (z1 zN 1 ) C b 4 (z2 C zN 2 ) C χ 23 [η 1 C ηN 1 C η 2 C ηN 2 C b 1 (z1 C zN 1 ) C b 2 (z2 zN 2 )] i(α 1 ω 22 ) i(α 1 ω 21 ) (η 1 ηN 1 ) C (η 2 ηN 2 ) λ1 ω1 λ1 ω2 2 Cb 3 (z1 zN 1 ) C b 4 (z2 C zN 2 )

i(α 1 ω 21 ) i(α 1 ω 22 ) (η 1 ηN 1 ) C (η 2 ηN 2 ) λ1 ω1 λ1 ω2 3 Cb 3 (z1 zN 1 ) C b 4 (z2 C zN 2 )

C χ 24

(9.109)

where the b i are deﬁned in (8.85). It follows from (9.108) and (9.109) that, to O(), only a three-to-one internal resonance is possible because (8.81) and (8.82) are not valid when ω 2 ω 1 (i.e., when there is a one-to-one internal resonance). Moreover, the excitation can produce resonances corresponding to

9.2 Gyroscopic Systems

Ω 3ω m : Subharmonic resonance of order one-third, Ω 1/3ω m : Superharmonic resonance of order three, Ω 2ω n ˙ ω m : Combination resonance. We present the normal forms of (9.108) and (9.109) for several of these external resonances in the absence of internal resonances in this section and in the presence of a three-to-one internal resonance in Section 9.2.3. Substituting the near-identity transformation (8.14) into (9.108) and (9.109) and choosing the h m to eliminate the nonresonance terms, we obtain the following normal forms: No External Resonance

ξP1 D i ω 1 ξ1 C σ 1 ξ1 C S11 ξ12 ξN1 C S12 ξ2 ξN2 ξ1

(9.110)

ξP2 D i ω 2 ξ2 C σ 2 ξ2 C S21 ξ1 ξN1 ξ2 C S22 ξ22 ξN2

(9.111)

where S11 , S12 , S21 , and S22 are deﬁned in (9.99)–(9.102), and σ 1 D 6χ 11 b 21 z1 zN 1 b 22 z2 zN 2 C b 1 b 2 (z2 zN 1 z1 zN 2 ) i(α 1 ω 21 ) 2 b 1 z1 zN 1 b 22 z2 zN 2 C b 1 b 2 (z2 zN 1 z1 zN 2 ) C 2χ 12 λ1 ω1 ) (z ) (b C 1 b 4 b 2 b 3 1 zN 2 C zN 1 z2 C 2χ 13 b 24 z2 zN 2 b 23 z1 zN 1 C b 3 b 4 (z1 zN 2 zN 1 z2 ) i(α 1 ω 21 ) (b 1 b 4 b 2 b 3 ) (z1 zN 2 C zN 1 z2 ) λ1 ω1 6i χ 14 (α 1 ω 21 ) 2 b 4 z2 zN 2 b 23 z1 zN 1 C b 3 b 4 (z1 zN 2 zN 1 z2 ) (9.112) C λ1 ω1 σ 2 D 6χ 21 b 21 z1 zN 1 b 22 z2 zN 2 C b 1 b 2 (z2 zN 1 z1 zN 2 ) i(α 1 ω 21 ) 2 b 1 z1 zN 1 b 22 z2 zN 2 C b 1 b 2 (z2 zN 1 z1 zN 2 ) C 2χ 32 λ1 ω1 C (b 1 b 4 b 2 b 3 ) (z1 zN 2 C zN 1 z2 ) C 2χ 23 b 24 z2 zN 2 b 23 z1 zN 1 C b 3 b 4 (z1 zN 2 zN 1 z2 ) C

i(α 1 ω 21 ) (b 1 b 4 b 2 b 4 ) (z1 zN 2 C zN 1 z2 ) λ1 ω2 6i χ 24 (α 1 ω 21 ) 2 b 4 z2 zN 2 b 23 z1 zN 1 C b 3 b 4 (z1 zN 2 zN 1 z2 ) C λ1 ω2 C

(9.113)

253

254

9 TDOF Systems with Cubic Nonlinearities

Ω 3ω 2

ξP1 D i ω 1 ξ1 C σ 1 ξ1 C S11 ξ12 ξN1 C S12 ξ2 ξN2 ξ1

(9.114)

ξP2 D i ω 2 ξ2 C σ 2 ξ2 C S21 ξ1 ξN1 ξ2 C S22 ξ22 ξN2 C Γ1 ξN22

(9.115)

where σ 1 and σ 2 are deﬁned in (9.112) and (9.113), the S m n are deﬁned in (9.99)– (9.102), and 2i χ 22 (α 1 ω 22 ) χ 23 (α 1 ω 22 )2 Γ1 D (b 1 z1 C b 2 z2 ) 3χ 21 λ1 ω2 λ 21 ω 22 2i χ 23 (α 1 ω 22 ) 3χ 24 (α 1 ω 22 )2 C (b 3 z1 C b 4 z4 ) χ 22 λ1 ω2 λ 21 ω 22

(9.116)

Ω 1/3ω 1

ξP1 D i ω 1 ξ1 C σ 1 ξ1 C S11 ξ12 ξN1 C S12 ξ2 ξN2 ξ1 C Γ2

(9.117)

ξP2 D i ω 2 ξ2 C σ 2 ξ2 C S21 ξ1 ξN1 ξ2 C S22 ξ22 ξN2

(9.118)

where σ 1 and σ 2 are deﬁned in (9.112) and (9.113), the S m n are deﬁned in (9.99)– (9.102), and Γ2 D χ 11 (b 1 z1 C b 2 z2 )3 C χ 12 (b 1 z1 C b 2 z2 )2 (b 3 z1 C b 4 z2 ) C χ 13 (b 1 z1 C b 2 z2 ) (b 3 z1 C b 4 z2 )2 C χ 14 (b 3 z1 C b 4 z2 )3

(9.119)

Ω ω 2 C 2ω 1

ξP1 D i ω 1 ξ1 C σ 1 ξ1 C S11 ξ12 ξN1 C S12 ξ2 ξN2 ξ1 C Γ3 ξN2 ξN1

(9.120)

ξP2 D i ω 2 ξ2 C σ 2 ξ2 C S21 ξ1 ξN1 ξ2 C S22 ξ22 ξN2 C Γ4 ξN12

(9.121)

where σ 1 and σ 2 are deﬁned in (9.112) and (9.113), the S m n are deﬁned in (9.99)– (9.102), and 6α 1 λ 2 (b 3 z1 C b 4 z2 ) Γ3 D 6χ 11 (b 1 z1 C b 2 z2 ) C λ1 ω1 ω2 α 1 ω 21 α 1 ω 22 ) (b C C 2χ 12 b 3 z1 C b 4 z2 i 1 z1 C b 2 z2 λ1 ω1 λ1 ω2 2 2 α1 ω1 α1 ω2 2χ 13 i (b 3 z1 C b 4 z2 ) C λ1 ω1 λ1 ω2 α1 λ2 (b 1 z1 C b 2 z2 ) (9.122) λ1 ω1 ω2

9.2 Gyroscopic Systems

(α 1 ω 21 )2 (b 3 z1 C b 4 z2 ) λ 21 ω 21 2i(α 1 ω 21 ) (b 1 z1 C b 2 z2 ) C χ 22 b 3 z1 C b 4 z2 λ1 ω1 (α 1 ω 21 )2 2i(α 1 ω 21 ) (b ) (b ) C z C b z z C b z χ 23 1 1 2 2 3 1 4 2 λ1 ω1 λ 21 ω 21 (9.123)

Γ4 D 3χ 21 (b 1 z1 C b 2 z2 ) 3χ 24

9.2.3 Three-to-One Internal Resonance

Substituting the near-identity transformation (8.14) into (9.108) and (9.109) and choosing the h m to eliminate the nonresonance terms, we obtain the following normal form in the absence of external resonances: ξP1 D i ω 1 ξ1 C σ 1 ξ1 C S11 ξ12 ξN1 C S12 ξ2 ξN2 ξ1 C S13 ξ2 ξN12

(9.124)

ξP2 D i ω 2 ξ2 C σ 2 ξ2 C S21 ξ1 ξN1 ξ2 C S22 ξ22 ξN2 C S23 ξ13

(9.125)

where σ 1 and σ 2 are deﬁned in (9.112) and (9.113) and the S m n are deﬁned in (9.99)–(9.102), (9.106), and (9.107). The normal forms of (9.108) and (9.109) in the presence of external resonances can be obtained by simply adding the appropriate Γm terms to (9.124) and (9.125), as in Section 9.2.2.

255

257

10 Systems with Quadratic and Cubic Nonlinearities We consider the response of two-degree-of-freedom damped systems with quadratic and cubic nonlinearities to simultaneous principal and combination parametric resonances in the presence and absence of internal resonances. The case of no internal resonance is treated in Section 10.2, the case of three-to-one internal resonance is treated in Section 10.3, and the case of one-to-one internal resonance is treated in Section 10.4. The case of two-to-one internal resonance is treated in Section 10.5 using the method of normal forms, in Section 10.6 using the method of multiple scales, and in Section 10.7 using the generalized method of averaging. Finally, the case of a nonsemisimple one-to-one internal resonance is treated in Section 10.8.

10.1 Introduction

In this chapter, we consider the response of two-degree-of-freedom damped systems with quadratic and cubic nonlinearities to multifrequency parametric excitations. Speciﬁcally, we consider systems modeled by uR 1 C ω 21 u 1 C 2µ 1 uP 1 C C 2u 2

m X

M X @V (u 1 , u 2 ) C 2u 1 f 1m cos (Ωm t C τ 1m ) @u 1 mD1

f 2m cos (Ωm t C τ 2m ) D 0

mD1

uR 2 C ω 22 u 2 C 2µ 2 uP 2 C C 2u 2

m X

(10.1)

M X @V (u 1 , u 2 ) C 2u 1 g 1m cos (Ωm t C ν 1m ) @u 2 mD1

g 2m cos (Ωm t C ν 2m ) D 0

mD1

(10.2)

V D 13 δ 1 u31 C δ 2 u21 u 2 C δ 3 u 1 u22 C 13 δ 4 u32 C 14 α 1 u41 C α 2 u31 u 2 C 12 α 3 u21 u22 C α 4 u 1 u32 C 14 α 5 u42

(10.3)

The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

258

10 Systems with Quadratic and Cubic Nonlinearities

Clearly, the undamped and unforced system is derivable from the Lagrangian LD

1 2

uP 21 C uP 22 C ω 21 u21 C ω 22 u22 V(u 1 , u 2 )

Thus the undamped unforced normal form should be derivable from a Lagrangian, which can be used to check the accuracy of the calculated normal forms. Again, as a ﬁrst step, it is convenient to cast (10.1) and (10.2) in complex-valued form using the transformation u m D ζ m C ζN m , uP m D i ω m ζ m ζN m

(10.4)

z m D e i Ωm t , zP m D i Ωm z m

(10.5)

To keep track of the different orders of magnitude, we introduce a small nondimensional parameter . Then, if u m D O(), u2m D O( 2 ) and u3m D O( 3 ), and we order the damping and excitation terms such that their effects balance each other as well as the effect of the nonlinearity. To reduce the amount of algebra, we assume that the µ m , f m n , and g m n are O( 2 ). Using this ordering and the transformation (10.4) and (10.5), we rewrite (10.1)–(10.3) as i ζP 1 D i ω 1 ζ1 2 µ 1 ζ1 ζN 1 C δ 1 (ζ1 C ζN 1 )2 2ω 1 i 2 α 1 (ζ1 C ζN 1 )3 C2δ 2 (ζ1 C ζN 1 )(ζ2 C ζN 2 ) C δ 3 (ζ2 C ζN 2 )2 C 2ω 1 C3α 2 (ζ1 C ζN 1 )2 (ζ2 C ζN 2 ) C α 3 (ζ1 C ζN 1 )(ζ2 C ζN 2 )2 C α 4 (ζ2 C ζN 2 )3 C

M X i 2 (ζ1 C ζN 1 ) f 1m z m e i τ 1m C zN m e i τ 1m 2ω 1 mD1

C

M X i 2 (ζ2 C ζN 2 ) f 2m z m e i τ 2m C zN m e i τ 2m 2ω 1 mD1

(10.6)

i ζP 2 D i ω 2 ζ2 2 µ 2 ζ2 ζN 2 C δ 2 (ζ1 C ζN 1 )2 2ω 2 2 N 2 C i α 2 (ζ1 C ζN 1 )3 C2δ 3 (ζ1 C ζN 1 )(ζ2 C ζN 2 ) C δ 4 (ζ2 C ζ) 2ω 2 Cα 3 (ζ1 C ζN 1 )2 (ζ2 C ζN 2 ) C 3α 4 (ζ1 C ζN 1 )(ζ2 C ζN 2 )2 C α 5 (ζ2 C ζN 2 )3 C

M X i 2 (ζ1 C ζN 1 ) g 1m z m e i ν 1m C zN m e i ν 1m 2ω 2 mD1

C

M X i 2 (ζ2 C ζN 2 ) g 2m z m e i ν 2m C zN m e i ν 2m 2ω 1 mD1

(10.7)

10.1 Introduction

Next, we introduce the near-identity transformation ζ m D η m C h m1(η n , ηN n ) C 2 h m2(η n , ηN n , z n , zN n ) C

(10.8)

into (10.6) and (10.7) and choose the h m n to eliminate the nonresonance terms so that the resulting equations have the simplest possible form ηP m D i ω m η m C F m1(η n , ηN n ) C 2 F m2 (η n , ηN n , z n , zN n ) C

(10.9)

where the F m n are chosen to eliminate the resonance and near-resonance terms. Substituting (10.8) and (10.9) into (10.6) and (10.7), using (10.5), and equating coefﬁcients of like powers of , we obtain the following:

Order ()

F11 C L(h 11 ) i ω 1 h 11 D

F21 C L(h 21 ) i ω 2 h 21 D

i δ 1 (η 1 C ηN 1 )2 C 2δ 2 (η 1 C ηN 1 )(η 2 C ηN 2 ) 2ω 1 (10.10) Cδ 3 (η 2 C ηN 2 )2 i δ 2 (η 1 C ηN 1 )2 C 2δ 3 (η 1 C ηN 1 )(η 2 C ηN 2 ) 2ω 2 (10.11) Cδ 4 (η 2 C ηN 2 )2

Order (2 )

F12 C L(h 12 ) i ω 1 h 12 D

@h 11 @h 11 N @h 11 @h 11 N F11 F21 F11 F21 @η 1 @ ηN 1 @η 2 @ ηN 2

i h 2δ 1 (η 1 C ηN 1 )(h 11 C hN 11 ) C 2δ 2 (η 1 C ηN 1 )(h 21 C hN 21 ) 2ω 1 i C2δ 2 (h 11 C hN 11 )(η 2 C ηN 2 ) C 2δ 3 (η 2 C ηN 2 )(h 21 C hN 21 )

C

i α 1 (η 1 C ηN 1 )3 C 3α 2 (η 1 C ηN 1 )2 (η 2 C ηN 2 ) 2ω 1 Cα 3 (η 1 C ηN 1 )(η 2 C ηN 2 )2 C α 4 (η 2 C ηN 2 )3 µ 1 (η 1 ηN 1 )

C

C

M X i (η 1 C ηN 1 ) f 1m z m e i τ 1m C zN m e i τ 1m 2ω 1 mD1

C

M X i (η 2 C ηN 2 ) f 2m z m e i τ 2m C zN m e i τ 2m 2ω 1 mD1

(10.12)

259

260

10 Systems with Quadratic and Cubic Nonlinearities

F22 C L(h 22 ) i ω 2 h 22 D

@h 21 @h 21 N @h 21 @h 21 N F11 F21 F11 F21 @η 1 @ ηN 1 @η 2 @ ηN 2

i h 2δ 2 (η 1 C ηN 1 )(h 11 C hN 11 ) C 2δ 3 (η 1 C ηN 1 )(h 21 C hN 21 ) 2ω 2 i C2δ 3 (η 2 C ηN 2 )(h 11 C hN 11 ) C 2δ 4 (η 2 C ηN 2 )(h 21 C hN 21 )

C

i α 2 (η 1 C ηN 1 )3 C α 3 (η 1 C ηN 1 )2 (η 2 C ηN 2 ) 2ω 2 C3α 4 (η 1 C ηN 1 )(η 2 C ηN 2 )2 C α 5 (η 2 C ηN 2 )3

C

µ 2 (η 2 ηN 2 ) C C where

M X i (η 1 C ηN 1 ) g 1m z m e i ν 1m C zN m e i ν 1m 2ω 2 mD1

M X i (η 2 C ηN 2 ) g 2m z m e i ν 2m C zN m e i ν 2m 2ω 2 mD1

(10.13)

@h @h @h @h η1 ηN 1 C i ω 2 η2 ηN 2 @η 1 @ ηN 1 @η 2 @ ηN 2 M X @h @h Ωm zm zN m Ci @z @ zN m m mD1

L(h) D i ω 1

Next, we choose h 11 and h 21 to eliminate the nonresonance terms and F11 and F21 to eliminate the resonance and near-resonance terms. We start by assuming that there are no resonance and near-resonance terms and hence set F11 and F21 equal to zero. Then, we determine the conditions under which this assumption is violated. The forms of the terms in (10.10) and (10.11) suggest seeking h 11 and h 21 in the forms h 11 D Γ1 η 21 C Γ2 η 1 ηN 1 C Γ3 ηN 21 C Γ4 η 22 C Γ5 η 2 ηN 2 C Γ6 ηN 22 C Γ7 η 1 η 2 C Γ8 η 1 ηN 2 C Γ9 ηN 1 η 2 C Γ10 ηN 1 ηN 2

(10.14)

h 21 D Λ 1 η 21 C Λ 2 η 1 ηN 1 C Λ 3 ηN 21 C Λ 4 η 22 C Λ 5 η 2 ηN 2 C Λ 6 ηN 22 C Λ 7 η 1 η 2 (10.15) C Λ 8 η 1 ηN 2 C Λ 9 ηN 1 η 2 C Λ 10 ηN 1 ηN 2 Substituting (10.14) and (10.5) into (10.10) and (10.11), putting F11 and F21 equal to zero, and choosing the Γm and Λ m to eliminate all of the terms, we obtain δ1 δ1 δ1 δ3 , Γ2 D 2 , Γ3 D 2 , Γ4 D , 2 2ω 1 (2ω 2 ω 1 ) 2ω 1 ω1 6ω 1 δ3 δ3 δ2 δ2 , Γ8 D , , Γ7 D Γ5 D 2 , Γ6 D 2ω 1 (2ω 2 C ω 1 ) ω1 ω2 ω1 ω2 ω1 δ2 δ2 Γ9 D , Γ10 D (10.16) ω 1 (ω 2 2ω 1 ) ω 1 (ω 2 C 2ω 1 ) Γ1 D

10.1 Introduction

δ2 δ2 δ2 , Λ2 D 2 , Λ3 D , 2ω 2 (2ω 1 ω 2 ) 2ω 2 (ω 2 C 2ω 1 ) ω2 δ4 δ4 δ4 δ3 , Λ5 D 2 , Λ6 D 2 , Λ7 D , Λ4 D ω1 ω2 2ω 22 ω2 6ω 2 δ3 δ3 δ3 , Λ 10 D Λ8 D , Λ9 D ω 2 (ω 1 2ω 2 ) ω1 ω2 ω 2 (2ω 2 C ω 1 ) Λ1 D

(10.17) It follows from (10.16) and (10.17) that the transformation (10.8) breaks down when ω 2 2ω 1 or ω 1 2ω 2 and hence (10.10) and (10.11) contain near-resonance terms in these cases. These cases correspond to two-to-one internal resonances, which need to be treated separately. This is done in Section 10.5. It follows from (10.14)–(10.17) that 2δ 1 2 δ1 2 δ3 η 1 C ηN 21 2 η 1 ηN 1 C η 2 C ηN 22 h 11 C hN 11 D 2 2 2 3ω 1 ω1 4ω 2 ω 1 2δ 3 2δ 2 (η 1 η 2 C ηN 1 ηN 2 ) 2 η 2 ηN 2 C ω 2 (ω 2 C 2ω 1 ) ω1 2δ 2 (η 2 ηN 1 C η 1 ηN 2 ) C ω 2 (ω 2 2ω 1 ) 2 2δ 2 δ2 δ4 2 η C ηN 21 2 η 1 ηN 1 C η C ηN 22 h 21 C hN 21 D 4ω 21 ω 22 1 ω2 3ω 22 2 2δ 4 2δ 3 (η 2 η 1 C ηN 1 ηN 2 ) 2 η 2 ηN 2 C ω 1 (ω 1 C 2ω 2 ) ω2 2δ 3 (η 2 ηN 1 C η 1 ηN 2 ) C ω 1 (ω 1 2ω 2 )

(10.18)

(10.19)

Next, we substitute (10.18) and (10.19) into (10.12) and (10.13) and choose h 12 and h 22 to eliminate all of the nonresonance terms, thereby leaving F12 and F22 with all of the resonance and near-resonance terms. If we stop at O( 2 ), then we do not need to determine h 12 and h 22 , and hence all that we need to do is to determine F12 and F22 . Inspecting the right-hand sides of (10.12) and (10.13), we ﬁnd that the following resonances are possible:

ω 2 ω 1 : One-to-one internal resonance, ω 2 3ω 1 : Three-to-one internal resonance, Ωn 2ω n : Principal parametric resonance, Ωm ω 1 ˙ ω 1 : Combination parametric resonance.

In what follows, we consider the different internal resonance cases in conjunction with the parametric resonances: Ωm ω 2 ω 1 ,

Ωn ω 2 C ω 1 ,

Ω p 2ω 1 ,

and

Ωq 2ω 2

261

262

10 Systems with Quadratic and Cubic Nonlinearities

10.2 The Case of No Internal Resonance

Choosing F12 and F22 to eliminate all of the resonance and near-resonance terms in (10.12) and (10.13), we obtain F12 D µ 1 η 1 C

4i S11 η 21 ηN 1 C S12 η 2 ηN 2 η 1 ω1

i f 1p ηN 1 z p e i τ 1p C f 2m η 2 zN m e i τ 2m C f 2n ηN 2 z n e i τ 2n 2ω 1 4i S12 η 1 ηN 1 η 2 C S22 η 22 ηN 2 D µ 2 η 2 C ω2 i g 1m η 1 z m e i ν 1m C g 1n ηN 1 z n e i ν 1n C g 2q ηN 2 z q e i ν 2q C 2ω 2 C

F22

(10.20)

(10.21)

where 10δ 21 4δ 22 2δ 22 2 C 2 3ω 1 ω2 4ω 21 ω 22 4δ 1 δ 3 4δ 2 δ 4 8δ 22 8δ 23 D 2α 3 C C ω 21 ω 22 ω 22 4ω 21 ω 21 4ω 22 2 2 2 10δ 4 4δ 3 2δ 3 D 3α 5 2 C 3ω 22 ω1 4ω 22 ω 21

8S11 D 3α 1

(10.22)

8S12

(10.23)

8S22

Substituting (10.8), (10.18), and (10.19) into (10.4) yields 2δ 1 δ1 2 δ3 u 1 D η 1 C ηN 1 C η C ηN 21 2 η 1 ηN 1 C 3ω 21 1 ω1 4ω 22 ω 21 2δ 3 2δ 2 (η 1 η 2 C ηN 1 ηN 2 ) 2 η 2 ηN 2 C ω 2 (ω 2 C 2ω 1 ) ω1 2δ 2 (η 2 ηN 1 C η 1 ηN 2 ) C C ω 2 (ω 2 2ω 1 ) 2 2δ 2 δ2 δ4 η 1 C ηN 21 2 η 1 ηN 1 C u 2 D η 2 C ηN 2 C 2 2 4ω 1 ω 2 ω2 3ω 22 2δ 4 2δ 3 (η 2 η 1 C ηN 2 ηN 1 ) 2 η 2 ηN 2 C ω 1 (ω 1 C 2ω 2 ) ω2 2δ 3 (η 2 ηN 1 C η 1 ηN 2 ) C C ω 1 (ω 1 2ω 2 )

(10.24)

η 22 C ηN 22

(10.25)

η 22 C ηN 22

(10.26)

Substituting for the F m n into (10.9), we obtain the normal form ηP 1 D i ω 1 η 1 2 µ 1 η 1 C C

4 i 2 S11 η 21 ηN 1 C S12 η 2 ηN 2 η 1 ω1

i 2 f 1p ηN 1 z p e i τ 1p C f 2m η 2 zN m e i τ 2m C f 2n ηN 2 z n e i τ 2n C 2ω 1 (10.27)

10.3 The Case of Three-to-One Internal Resonance

ηP 2 D i ω 2 η 2 2 µ 2 η 2 C C

4 i 2 S12 η 1 ηN 1 η 2 C S22 η 22 ηN 2 ω2

i 2 g 1m η 1 z m e i ν 1m C g 1n ηN 1 z n e i ν 1n C g 2q ηN 2 z q e i ν 2q C (10.28) 2ω 2

Replacing the η n in (10.25)–(10.28) with A n e i ω n t , we obtain the same results found by using the method of multiple scales (Nayfeh and Jebril, 1987). Moreover, in the absence of damping and forcing, (10.27) and (10.28) are derivable from the Lagrangian L D i ω 1 η 1 ηPN 1 ηP 1 ηN 1 C i ω 2 η 2 ηPN 2 ηP 2 ηN 2 2ω 21 η 1 ηN 1 2ω 22 η 2 ηN 2 4 2 S11 η 21 ηN 21 C 2S12 η 1 ηN1 η 2 ηN2 C S22 η 22 ηN 22

10.3 The Case of Three-to-One Internal Resonance

The three-to-one internal resonance produces the additional near-resonance terms 4i S13 η 2 ηN 21 ω1

and

4i S23 η 31 ω2

in (10.12) and (10.13), respectively, and hence F12 and F22 are chosen to contain these additional terms, respectively, where 2δ 1 δ 2 4δ 1 δ 2 4δ 2 δ 3 2δ 2 δ 3 C (10.29) C C ω 2 (ω 2 2ω 1 ) ω 1 (ω 1 2ω 2 ) 4ω 21 ω 22 3ω 21 2δ 1 δ 2 2δ 2 δ 3 D α2 C C (10.30) 2 3ω 1 4ω 21 ω 22

8S13 D 3α 2 C 8S23

Consequently, to the second approximation, u 1 and u 2 are given by (10.25) and (10.26), where η 1 and η 2 are given by the normal form 4 i 2 S11 η 21 ηN 1 C S12 η 2 ηN 2 η 1 C S13 η 2 ηN 21 ω1 i 2 f 1p ηN 1 z p e i τ 1p C f 2m η 2 zN m e i τ 2m C f 2n ηN 2 z n e i τ 2n C C 2ω 1 (10.31)

ηP 1 D i ω 1 η 1 2 µ 1 η 1 C

4 i 2 S12 η 1 ηN 1 η 2 C S22 η 22 ηN 2 C S23 η 31 ω2 i 2 g 1m η 1 z m e i ν 1m C g 1n ηN 1 zN n e i ν 1n C g 2q ηN 2 z q e i ν 2q C (10.32) C 2ω 2

ηP 2 D i ω 2 η 2 2 µ 2 η 2 C

As ω 2 ! 3ω 1 , (10.29) and (10.30) reduce to 2δ 1 δ 2 6δ 2 δ 3 ω 21 5ω 21 2δ 1 δ 2 2δ 2 δ 3 D α2 C 3ω 21 5ω 21

8S13 D 3α 2 C 8S23

263

264

10 Systems with Quadratic and Cubic Nonlinearities

and thus S13 D 3S23 . Therefore, in the absence of damping and forcing, (10.31) and (10.32) are derivable from the Lagrangian L D i ω 1 (η 1 ηPN 1 ηP 1 ηN 1 ) C i ω 2 (η 2 ηPN 2 ηP 2 ηN 2 ) 2ω 21 η 1 ηN 1 2ω 22 η 2 ηN 2 4 2 S11 η 21 ηN 21 C 2S12 η 1 ηN1 η 2 ηN2 C S22 η 22 ηN 22 C S23 η 2 ηN 31 C ηN 2 η 31

10.4 The Case of One-to-One Internal Resonance

Choosing F12 and F22 to eliminate the resonance near-resonance terms in (10.12) and (10.13) and then substituting the results into (10.9), we obtain the following normal form: 4 i 2 S11 η 21 ηN 1 C S12 η 2 ηN 2 η 1 C S13 η 21 ηN 2 ω1 CS14 η 22 ηN 1 C S15 η 1 ηN 1 η 2 C S16 η 22 ηN 2

ηP 1 D i ω 1 η 1 2 µ 1 η 1 C

i 2 f 1p ηN 1 z p e i τ 1p C f 2m η 2 zN m e i τ 2m C f 2n ηN 2 z n e i τ 2n 2ω 1 4 i 2 S12 η 2 η 1 ηN 1 C S22 η 22 ηN 2 C S23 η 21 ηN 1 ηP 2 D i ω 2 η 2 2 µ 2 η 2 C ω2 CS24 η 22 ηN 1 C S25 η 21 ηN 2 C S26 η 2 ηN 2 η 1 C

C

i 2 g 1m η 1 z m e i ν 1m C g 1n ηN 1 z n e i ν 1n C g 2q ηN 2 z q e i ν 2q 2ω 2

(10.33)

(10.34)

where 2δ 1 δ 2 4δ 2 δ 3 4δ 1 δ 2 2δ 2 δ 3 C C C ω 1 (ω 1 2ω 2 ) ω 2 (ω 2 2ω 1 ) 4ω 21 ω 22 3ω 21 2δ 1 δ 3 2δ 2 δ 4 4δ 22 4δ 23 D α3 C C C C 2 2 2 ω 2 (ω 2 2ω 1 ) ω 1 (ω 1 2ω 2 ) 4ω 2 ω 1 3ω 2 8δ 1 δ 2 8δ 2 δ 3 4δ 1 δ 2 4δ 2 δ 3 D 6α 2 C 2 C 2 2 2 ω 2 4ω 1 ω 1 4ω 2 ω 21 ω 22 10δ 3 δ 4 4δ 2 δ 3 2δ 2 δ 3 D 3α 4 C 3ω 22 ω 21 4ω 22 ω 21 10δ 1 δ 2 4δ 2 δ 3 2δ 2 δ 3 D 3α 2 C 3ω 21 ω 22 4ω 21 ω 22

S13 D 3α 2 C

(10.35)

S14

(10.36)

S15 S16 S23

(10.37) (10.38) (10.39)

10.4 The Case of One-to-One Internal Resonance

2δ 2 δ 3 2δ 3 δ 4 4δ 2 δ 3 4δ 3 δ 4 C C C (10.40) 2 2 2 ω 2 (ω 2 2ω 1 ) ω 1 (ω 1 2ω 2 ) 4ω 2 ω 1 3ω 2 2δ 1 δ 3 2δ 2 δ 4 4δ 22 4δ 23 D α3 C C C C (10.41) ω 2 (ω 2 2ω 1 ) ω 1 (ω 1 2ω 2 ) 3ω 21 4ω 21 ω 22 4δ 1 δ 3 4δ 2 δ 4 8δ 2 8δ 2 D 6α 4 C 2 2 2 C 2 3 2 (10.42) 2 2 ω1 ω2 ω 2 4ω 1 ω 1 4ω 2

S24 D 3α 4 C S25 S26

Therefore, to the second approximation, u 1 and u 2 are given by (10.25) and (10.26), where η 1 and η 2 are given by (10.33) and (10.34). As ω 2 ! ω 1 , (10.35)–(10.42) reduce to 8S13 D 3α 2

10δ 1 δ 2 10δ 2 δ 3 3ω 21 3ω 21

4δ 22 2δ 1 δ 3 4δ 23 2δ 2 δ 4 C 2 C ω 21 3ω 21 ω1 3ω 21 20δ 1 δ 2 20δ 2 δ 3 D 6α 2 3ω 21 3ω 21 10δ 2 δ 3 10δ 3 δ 4 D 3α 4 3ω 21 3ω 21 10δ 1 δ 2 10δ 2 δ 3 D 3α 2 3ω 21 3ω 21 10δ 2 δ 3 10δ 3 δ 4 D 3α 4 3ω 21 3ω 21

8S14 D α 3 8S15 8S16 8S23 8S24

4δ 22 2δ 1 δ 3 4δ 23 2δ 2 δ 4 C 2 C 2 ω1 3ω 21 ω1 3ω 21 20δ 2 δ 3 20δ 3 δ 4 D 6α 4 3ω 21 3ω 21

8S25 D α 3 8S26 Thus,

S13 D S23 D 12 S15 ,

S16 D S24 D 12 S26 ,

and

S14 D S25

Therefore, in the absence of damping and forcing, (10.33) and (10.34) are derivable from the Lagrangian L D i ω 1 (η 1 ηPN 1 ηP 1 ηN 1 ) C i ω 2 (η 2 ηPN 2 ηP 2 ηN 2 ) 2ω 21 η 1 ηN 1 2ω 22 η 2 ηN 2 4 2 S11 η 21 ηN 21 C S22 η 22 ηN 22 C 2S12 η 1 ηN1 η 2 ηN2 C2S13 η 21 ηN 1 ηN 2 C ηN 21 η 1 η 2 CS14 η 22 ηN 21 C η 21 ηN 22 C 2S16 η 22 ηN 2 ηN 1 C ηN 22 η 2 η 1

265

266

10 Systems with Quadratic and Cubic Nonlinearities

10.5 The Case of Two-to-One Internal Resonance

We consider the case of ω 2 2ω 1 . Speciﬁcally, we let ω 2 D 2ω 1 C σ, where σ is a detuning parameter. The case of ω 1 2ω 2 can be treated in a similar fashion. It follows from (10.16) and (10.17) that Γ9 and Λ 1 have small divisors and hence the term η 2 ηN 1 in (10.10) and the term η 21 in (10.11) are near-resonance terms. For a uniform expansion, we choose F11 and F21 to eliminate these terms; that is, F11 D

i δ2 η 2 ηN 1 ω1

and

F21 D

i δ2 2 η 2ω 2 1

(10.43)

Then, substituting (10.14) and (10.15) into (10.10) and (10.11) and eliminating the nonresonance terms, we ﬁnd that the Γm and Λ n , except Γ9 and Λ 1 , are given by (10.16) and (10.17). Substituting (10.14)–(10.17) into (10.4), we obtain

2δ 1 2 δ1 2 δ3 η C ηN 21 2 η 1 ηN 1 C η C ηN 22 3ω 21 1 ω1 4ω 22 ω 21 2 2δ 3 2δ 2 (η 1 η 2 C ηN 1 ηN 2 ) 2 η 2 ηN 2 C ω 2 (ω 2 C 2ω 1 ) ω1 δ2 (η 1 ηN 2 C ηN 1 η 2 ) C (10.44) ω1 ω2 2δ 2 2 δ2 u 2 D η 2 C ηN 2 η C ηN 21 C 2 η 1 ηN 1 2ω 2 (ω 2 C 2ω 1 ) 1 ω2 2δ 4 δ4 2δ 3 (η 2 η 1 ηN 2 ηN 1 ) η 2 C ηN 22 C 2 η 2 ηN 2 ω 1 (ω 1 C 2ω 2 ) 3ω 22 2 ω2 2δ 3 (η 2 ηN 1 C η 1 ηN 2 ) C (10.45) ω 2 (ω 1 2ω 2 ) u 1 D η 1 C ηN 1 C

Substituting (10.14)–(10.17) into (10.12) and (10.13) and choosing F12 and F22 to eliminate the resonance and near-resonance terms, we obtain F12 D µ 1 η 1 C

i f 1p ηN 1 z p e i τ 1p C f 2m η 2 zN m e i τ 2m C f 2n ηN 2 z n e i τ 2n 2ω 1 4i S21 η 1 ηN 1 η 2 C S22 η 22 ηN 2 D µ 2 η 2 C ω2 i g 1m η 1 z m e i ν 1m C g 1n ηN 2 zN n e i ν 1n C g 2q ηN 2 z q e i ν 2q C 2ω 2 C

F22

4i S11 η 21 ηN 1 C S12 η 2 ηN 2 η 1 ω1

(10.46)

(10.47)

10.6 Method of Multiple Scales

where 10δ 21 4δ 22 δ 22 (10.48) ω 2 (ω 2 C 2ω 1 ) 3ω 21 ω 22 4δ 1 δ 3 4δ 2 δ 4 8δ 2 2δ 22 D 8S21 D 2α 3 C 2 3 2 (10.49) 2 2 ω1 ω2 ω 1 4ω 2 ω 1 (ω 2 C 2ω 1 ) 10δ 24 4δ 23 2δ 23 D 3α 5 2 C (10.50) 2 3ω 2 ω1 4ω 22 ω 21

8S11 D 3α 1 8S12 8S22

Therefore, to the second approximation, u 1 and u 2 are given by (10.44) and (10.45) where η 1 and η 2 are given by the normal form iδ 2 4 i 2 S11 η 21 ηN 1 C S12 η 2 ηN 2 η 1 η 2 ηN 1 2 µ 1 η 1 C ω1 ω1 i 2 f 1p ηN 1 z p e i τ 1p C f 2m η 2 zN m e i τ 2m C f 2n ηN 2 z n e i τ 2n C (10.51) 2ω 1 iδ 2 2 4 i 2 S21 η 1 ηN 1 η 2 C S22 η 22 ηN 2 ηP 2 D i ω 2 η 2 C η1 2 µ2 η2 C 2ω 2 ω2 i 2 g 1m η 1 z m e i ν 1m C g 1n ηN 2 z n e i ν 1n C g 2q ηN 2 z q e i ν 2q (10.52) C 2ω 2

ηP 1 D i ω 1 η 1 C

In the absence of damping and forcing, (10.51) and (10.52) are derivable from the Lagrangian L D i ω 1 η 1 ηPN 1 ηP 1 ηN 1 C i ω 2 η 2 ηPN 2 ηP 2 ηN 2 2ω 21 η 1 ηN 1 2ω 22 η 2 ηN 2 δ 2 η 2 ηN 21 C ηN 2 η 21 4 2 S11 η 21 ηN 21 C 2S12 η 1 ηN1 η 2 ηN2 C S22 η 22 ηN 22

10.6 Method of Multiple Scales

To describe the method without algebraic complication, we consider the following special case of (10.1) and (10.2): uR 1 C ω 21 u 1 D 2δ 2 u 1 u 2

(10.53)

uR 2 C ω 22 u 2 D δ 2 u21

(10.54)

where ω 2 2ω 1 . We describe three implementations of the method of multiple scales: (a) treatment of the governing equations in second-order form, (b) treatment of the governing equations in state-space form, and (a) treatment of the governing equations in complex-valued form.

267

268

10 Systems with Quadratic and Cubic Nonlinearities

10.6.1 Second-Order Form

We seek a second-order approximation of the solution of (10.53) and (10.54) in the form u n (tI ) D

2 X

m u n m (T0 , T1 , T2 ) C

(10.55)

mD0

Substituting (10.55) into (10.53) and (10.54) and equating coefﬁcients of like powers of , we obtain:

Order (0 )

D02 u 10 C ω 21 u 10 D 0

(10.56)

D02 u 20 C ω 22 u 20 D 0

(10.57)

Order ()

D02 u 11 C ω 21 u 11 D 2D0 D1 u 10 2δ 2 u 10 u 20

(10.58)

D02 u 21 C ω 22 u 21 D 2D0 D1 u 20 δ 2 u210

(10.59)

Order (2 )

D02 u 12 C ω 21 u 12 D 2D0 D1 u 11 D12 u 10 2δ 2 u 11 u 20 2D0 D2 u 10 2δ 2 u 10 u 21

(10.60)

D02 u 22 C ω 22 u 22 D 2D0 D1 u 21 D12 u 20 2D0 D2 u 20 2δ 2 u 10 u 11 (10.61) The general solutions of (10.56) and (10.57) can be expressed as u m0 D A m (T1 , T2 ) e i ω m T0 C cc

(10.62)

Then, (10.58) and (10.59) become D02 u 11 C ω 21 u 11 D 2i ω 1 D1 A 1 e i ω 1 T0 2δ 2 A 2 A 1 e i(ω 2Cω 1 )T0 2δ 2 A 2 AN 1 e i(ω 2ω 1 )T0 C cc D02 u 21 C ω 22 u 21 D 2i ω 2 D1 A 2 e i ω 2 T0 δ 2 A21 e 2i ω 1 T0 δ 2 A 1 AN 1 C cc

(10.63)

(10.64)

Any particular solution of (10.63) and (10.64) contains secular and small-divisor terms when ω 2 2ω 1 . To quantify the nearness of ω 2 to 2ω 1 , we introduce the detuning parameter σ deﬁned by ω 2 D 2ω 1 C σ

(10.65)

10.6 Method of Multiple Scales

Using (10.65) in eliminating the terms that lead to secular terms from (10.63) and (10.64), we have i ω 1 D1 A 1 D δ 2 A 2 AN 1 e i σ T1

(10.66)

2i ω 2 D1 A 2 D δ 2 A21 e i σ T1

(10.67)

Then, the general solutions of (10.63) and (10.64) can be expressed as 2δ 2 A 2 A 1 e i(ω 2Cω 1 )T0 C cc ω 2 (ω 2 C 2ω 1 ) δ2 2 A 1 AN 1 C cc ω2

u 11 D B1 (T1 , T2 ) e i ω 1 T0 C

(10.68)

u 21 D B2 (T1 , T2 ) e i ω 2 T0

(10.69)

Substituting (10.62) and (10.66)–(10.69) into (10.60) and (10.61) and eliminating the terms that lead to secular terms, we obtain 2i ω 1 D2 A 1 2i ω 1 D1 B1 D12 A 1 D 2δ 2 A 2 BN 1 C B2 AN 1 e i σ T1

4δ 22 2 N 4δ 22 A1A1 C A 1 A 2 AN 2 2 ω 2 (ω 2 C 2ω 1 ) ω2

(10.70)

2i ω 2 D2 A 2 2i ω 2 D1 B2 D12 A 2 D 2δ 2 A 1 B1 e i σ T1 C

4δ 22 A 2 A 1 AN 1 ω 2 (ω 2 C 2ω 1 )

(10.71)

Using (10.66) and (10.67) to eliminate D12 A 1 and D12 A 2 from (10.70) and (10.71) and using the method of reconstitution, we obtain the modulation equations 2i ω 1 AP 1 D 2δ 2 A 2 AN 1 e iσ t C 2 2i ω 1 D1 B1 2δ 2 A 2 BN 1 C B2 AN 1 e iσ t ω2 σ δ2 4δ 22 iσ t N A21 AN 1 C A2A1e C 2 1C ω1 8ω 1 ω2

4ω 21 δ2 A 1 A 2 AN 2 C 22 1 C ω 2 (ω 2 C 2ω 1 ) ω1 2i ω 2 AP 2 D δ 2 A21 e iσ t C 2 2i ω 2 D1 B2 2δ 2 A 1 B1 e iσ t

σ δ 2 2 iσ t δ 22 A1e C 2ω 2 ω1 ω2

1

4ω 1 ω 2 C 2ω 1

(10.72)

A 2 A 1 AN 1 C (10.73)

When B1 D 0 and B2 D 0, (10.72) and (10.73) are not derivable from a Lagrangian because the coefﬁcient of A 1 A 2 AN 2 in (10.72) is different from the coefﬁcient of A 2 A 1 AN 1 in (10.73). The ﬁrst is (3δ 22 )/(2ω 21 ), whereas the second is 0 when ω 2 D 2ω 1 . Moreover, the coefﬁcient of A 2 AN 1 in (10.72) is not equal to twice

269

270

10 Systems with Quadratic and Cubic Nonlinearities

the coefﬁcient of A21 in (10.73). To make these coefﬁcients the same, we choose B 1 and B2 in (10.70) and (10.71) such that 2i ω 1 D1 B1 C D12 A 1 D 0 and 2i ω 2 D1 B2 C D12 A 2 D 0

(10.74)

or B1 D

δ2 A 2 AN 1 e i σ T1 2ω 21

and

B2 D

δ 2 2 i σ T1 A e 4ω 22 1

(10.75)

Substituting (10.75) into (10.70) and (10.71), letting ω 2 D 2ω 1 , and using the method of reconstitution yields 2 9δ 2 2 N δ 22 N 2 C (10.76) A A A C A A 2i ω 1 AP 1 D 2δ 2 A 2 AN 1 e iσ t C 2 1 1 2 1 8ω 21 2ω 21 2 δ 2i ω 2 AP 2 D δ 2 A21 e iσ t C 22 2 A 2 A 1 AN 1 (10.77) 2ω 1 which are derivable from the Lagrangian 9δ 22 2 2 N2 A1A1 L D i ω 1 A 1 APN 1 AN 1 AP 1 C i ω 2 A 2 APN 2 AN 2 AP 2 C 16ω 21 δ2 δ 2 A 2 AN21 e iσ t C A21 AN 2 e iσ t C 22 2 A 1 AN 1 A 2 AN 2 2ω 1 (10.78) Letting η m (t) D A m (t)e i ω m t in (10.51) and (10.52) and setting f 1m D 0, g 1m D 0, δ 1 D 0, δ 3 D 0, and δ 4 D 0, we obtain (10.76) and (10.77). To compare the results obtained in this section with those obtained in Section 10.7, we express them in real-variable form by introducing the polar transformation A n D 12 a n e i β n into (10.76) and (10.77) and separating real and imaginary parts. The result is δ 2 a 1 a 2 sin (β 2 2β 1 C σ t) 2ω 1 δ 2 2 a sin (β 2 2β 1 C σ t) aP 2 D 4ω 2 1

aP 1 D

δ 2 9 2 δ 22 3 2 δ 22 a 1 a 2 cos (β 2 2β 1 C σ t) a a 1 a 22 a 1 βP 1 D 2ω 1 64ω 31 1 16ω 31 δ 2 2 2 δ 22 2 a 1 cos (β 2 2β 1 C σ t) a a2 a 2 βP 2 D 4ω 2 32ω 31 1

(10.79) (10.80) (10.81) (10.82)

10.6 Method of Multiple Scales

Substituting (10.62), (10.68), and (10.69) into (10.55) and using (10.75) yields δ 2 2δ 2 N 1 e i σ T1 e i ω 1 T0 C A A A 2 A 1 e i(ω 2Cω 1 )T0 Ccc 2 2 ω (ω 2ω 1 2 2 C 2ω 1 ) (10.83) δ 2 2 i σ T1 i ω 2 T0 δ 2 e A e 2 A 1 AN 1 C cc (10.84) u2 D A 2 4ω 22 1 ω2

u1 D

A1

Then, using the polar transformation, we rewrite (10.83) and (10.84) as δ 2 a 1 a 2 cos [(ω 2 C ω 1 ) t C β 2 C β 1 ] ω 2 (ω 2 C 2ω 1 )

ω 2 (ω 2 C 2ω 1 ) ) ] [(ω t C β C cos ω β 2 1 2 1 (10.85) 4ω 21

u 1 D a 1 cos (ω 1 t C β 1 ) C

u 2 D a 2 cos (ω 2 t C β 2 )

δ 2 a 21 fcos (2ω 1 t C 2β 1 ) C 4g C 8ω 22

(10.86)

10.6.2 State-Space Form

We start with writing (10.53) and (10.54) in the state-space form uP 1 v1 D 0

(10.87)

vP1 C ω 21 u 1 D 2δ 2 u 1 u 2

(10.88)

uP 2 v2 D 0

(10.89)

vP2 C ω 22 u 2 D δ 2 u21

(10.90)

Then, we seek a second-order approximate solution of (10.87)–(10.90) in the form un D vn D

2 X

m u n m (T0 , T1 , T2 ) C

(10.91)

m v n m (T0 , T1 , T2 ) C

(10.92)

mD0 2 X mD0

Substituting (10.91) into (10.87)–(10.90) and equating coefﬁcients of like powers of yields.

271

272

10 Systems with Quadratic and Cubic Nonlinearities

Order (0 )

D0 u 10 v10 D 0

(10.93)

D0 v10 C ω 21 u 10 D 0

(10.94)

D0 u 20 v20 D 0

(10.95)

D0 v20 C ω 22 u 20 D 0

(10.96)

Order ()

D0 u 11 v11 D D1 u 10

(10.97)

D0 v11 C ω 21 u 11 D D1 v10 2δ 2 u 10 u 20

(10.98)

D0 u 21 v21 D D1 u 20

(10.99)

D0 v21 C ω 22 u 21 D D1 v20 δ 2 u210

(10.100)

Order (2 )

D0 u 12 v12 D D1 u 11 D2 u 10

(10.101)

D0 v12 C ω 21 u 12 D D1 v11 D2 v10 2δ 2 u 10 u 21 2δ 2 u 11 u 20

(10.102)

D0 u 22 v22 D D1 u 21 D2 u 20

(10.103)

D0 v22 C ω 22 u 22 D D1 v21 D2 v20 2δ 2 u 10 u 11

(10.104)

The solutions of (10.93)–(10.96) can be expressed as u 10 D A 1 (T1 , T2 ) e i ω 1 T0 C cc ,

v10 D i ω 1 A 1 (T1 , T2 ) e i ω 1 T0 C cc (10.105)

u 20 D A 2 (T1 , T2 ) e i ω 2 T0 C cc ,

v20 D i ω 2 A 2 (T1 , T2 ) e i ω 2 T0 C cc (10.106)

Substituting (10.105) and (10.106) into (10.97)–(10.100) yields D0 u 11 v11 D D1 A 1 e i ω 1 T0 C cc

(10.107)

D0 v11 C ω 21 u 11 D i ω 1 D1 A 1 e i ω 1 T0 2δ 2 A 2 A 1 e i(ω 2Cω 1 )T0 2δ A 2 AN 1 e i(ω 2ω 1 )T0 C cc D0 u 21 v21 D D1 A 2 e i ω 2 T0 C cc

(10.108) (10.109)

D0 v21 C ω 22 u 21 D i ω 2 D1 A 2 e i ω 2 T0 δ 2 A21 e 2i ω 1 T0 δ 2 A 1 AN 1 C cc (10.110)

10.6 Method of Multiple Scales

Because the homogeneous equations (10.107)–(10.110) have nontrivial solutions, the nonhomogeneous equations have solutions only if the nonhomogeneous terms are orthogonal to every solution of the corresponding adjoint equations. In this case, the solution of the adjoint of (10.107) and (10.108) is [i ω 1

1]T e i ω 1 T0

Therefore, using (10.65) and imposing the solvability condition on (10.107) and (10.108), one obtains 2i ω 1 D1 A 1 D 2δ 2 A 2 AN 1 e i σ T1

(10.111)

Then, although (10.107) and (10.108) are solvable, their solution is not unique. A unique solution can be obtained by requiring it to be orthogonal to the above adjoint solution. Hence, the unique solution of (10.107) and (10.108) is 2δ 2 δ2 A 2 A 1 e i(ω 2Cω 1 )T0 2 A 2 AN 1 e i(ω 2ω 1 )T0 Ccc (10.112) ω 2 (ω 2 C 2ω 1 ) 2ω 1 2i δ 2 (ω 2 C ω 1 ) i δ2 D A 2 AN 1 e i(ω 2ω 1 )T0 Ccc (10.113) A 2 A 1 e i(ω 2Cω 1 )T0 C ω 2 (ω 2 C 2ω 1 ) ω1

u 11 D v11

Following steps similar to those used above, one ﬁnds that the solvability condition of (10.109) and (10.110) is 2i ω 2 D1 A 2 D δ 2 A21 e i σ T1

(10.114)

and their unique solution is δ2 δ 2 2 2i ω 1 T0 A 1 AN 1 A e C cc ω 22 4ω 22 1 i δ 2 2 2i ω 1 T0 D A e C cc 4ω 2 1

u 21 D

(10.115)

v21

(10.116)

Substituting (10.112), (10.113), (10.115), and (10.116) into (10.101)–(10.104) and imposing the solvability conditions yields δ 22 (8ω 1 C 5ω 2 ) 2 N 2δ 22 A 1 A 2 AN 2 A1A1 C 2 ω 1 (ω 2 C 2ω 1 ) ω 2 (ω 2 C 2ω 1 ) 2δ 22 2i ω 2 D2 A 2 D A 2 A 1 AN 1 ω 1 (ω 2 C 2ω 1 ) 2i ω 1 D2 A 1 D

(10.117) (10.118)

273

274

10 Systems with Quadratic and Cubic Nonlinearities

Using the method of reconstitution, we combine (10.111), (10.114), (10.117), and (10.118) into

C

2δ 22 ω 1 (ω 2 C 2ω 1 )

δ 22 (8ω 1 C 5ω 2 ) 2 N A A1 ω 22 (ω 2 C 2ω 1 ) 1 A 1 A 2 AN 2 C

2i ω 1 AP 1 D 2δ 2 A 2 AN 1 e iσ t C 2

2i ω 2 AP 2 D δ 2 A21 e iσ t C

2 2 δ 22 A 2 A 1 AN 1 C ω 1 (ω 2 C 2ω 1 )

(10.119) (10.120)

Equations 10.119 and 10.120 are derivable from the Lagrangian L D i ω 1 A 1 APN 1 AN 1 AP 1 C i ω 2 A 2 APN 2 AN 2 AP 2 2 δ 22 (8ω 1 C 5ω 2 ) 2 2 δ 2 A 2 AN21 e iσ t C A21 AN 2 e iσ t C A AN 2ω 22 (ω 2 C 2ω 1 ) 1 1 C

2δ 22 A 1 AN 1 A 2 AN 2 ω 1 (ω 2 C 2ω 1 )

(10.121)

Equations 10.119–10.121 reduce to (10.76)–(10.78) when ω 2 is replaced with 2ω 1 . 10.6.3 Complex-Valued Form

Using the transformation (10.4) and (10.5), we rewrite (10.53) and (10.54) in the complex-valued form iδ 2 ζP 1 D i ω 1 ζ1 C ω1 iδ 2 ζP 2 D i ω 2 ζ2 C 2ω 2

ζ1 C ζN 1 ζ1 C ζN 1

ζ2 C ζN 2

2

(10.122) (10.123)

We seek a second-order uniform expansion of (10.122) and (10.123) in the form ζ n (tI ) D

2 X

m δ n m (T0 , T1 , T2 ) C

(10.124)

mD0

Substituting (10.124) into (10.122) and (10.123) and equating coefﬁcients of like powers of , we obtain

Order (0 )

D0 ζ11 i ω 1 ζ11 D 0

(10.125)

D0 ζ21 i ω 2 ζ21 D 0

(10.126)

10.6 Method of Multiple Scales

Order ()

i δ2 ζ11 C ζN 11 ζ21 C ζN 21 ω1 2 i δ2 ζ11 C ζN 11 D D1 ζ21 C 2ω 2

D0 ζ12 i ω 1 ζ12 D D1 ζ11 C

(10.127)

D0 ζ22 i ω 2 ζ22

(10.128)

Order (2 )

i δ2 ζ11 C ζN 11 ζ22 C ζN 22 ω1 i δ2 ζ12 C ζN 12 ζ21 C ζN 21 C (10.129) ω1 i δ2 ζ11 C ζN 11 ζ12 C ζN 12 (10.130) D D2 ζ21 D1 ζ22 C ω2

D0 ζ13 i ω 1 ζ13 D D2 ζ11 D1 ζ12 C

D0 ζ23 i ω 2 ζ23

The solutions of (10.125) and (10.126) can be expressed as ζ n1 D A n (T1 , T2 ) e i ω n T0

(10.131)

Substituting (10.131) into (10.127) and (10.128) yields i δ2 h A 2 AN 1 e i(ω 2ω 1 )T0 ω1 CA 2 A 1 e i(ω 2Cω 1 )T0 C A 1 AN 2 e i(ω 1ω 2 )T0 i C AN 2 AN 1 e i(ω 2Cω 1 )T0

D0 ζ12 i ω 1 ζ12 D D1 A 1 e i ω 1 T0 C

i δ 2 2 2i ω 1 T0 A1e C 2A 1 AN 1 D0 ζ22 i ω 2 ζ22 D D1 A 2 e i ω 2 T0 C 2ω 2 C AN21 e 2i ω 1 T0

(10.132)

(10.133)

Using (10.65) in (10.132) and (10.133) and eliminating the terms that produce secular terms, we obtain i δ2 A 2 AN 1 e i σ T1 ω1 i δ 2 2 i σ T1 A e D1 A 2 D 2ω 2 1

D1 A 1 D

(10.134) (10.135)

Then, the solutions of (10.132) and (10.133) can be expressed as δ2 δ2 A 2 A 1 e i(ω 2Cω 1 )T0 A 1 AN 2 e i(ω 1ω 2 )T0 ω1 ω2 ω1 ω2 δ2 AN 2 AN 1 e i(ω 2Cω 1 )T0 ω 1 (ω 2 C 2ω 1 ) δ2 δ2 D 2 A 1 AN 1 AN2 e i(ω 2Cω 1 )T0 2ω 2 (ω 2 C 2ω 1 ) 1 ω2

ζ12 D

ζ22

(10.136) (10.137)

275

276

10 Systems with Quadratic and Cubic Nonlinearities

Substituting (10.131), (10.136), and (10.137) into (10.129) and (10.130), using (10.134) and (10.135), and eliminating the terms that lead to secular terms, we obtain i δ 22 (5ω 2 C 8ω 1 ) 2 N i δ 22 A1A1 2 2 2ω 1 ω 2 (ω 2 C 2ω 1 ) ω 1 (ω 2 C 2ω 1 ) 2 i δ2 A 2 A 1 AN 1 D2 A 2 D ω 1 ω 2 (ω 2 C 2ω 1 ) D2 A 1 D

(10.138) (10.139)

Using the method of reconstitution, we combine (10.134), (10.135), (10.138), and (10.139) into iδ 2 i 2 δ 22 (5ω 2 C 8ω 1 ) 2 N i 2 δ 22 AP 1 D A 2 AN 1 e i σ T1 A1A1 2 A 1 A 2 AN 2 2 ω1 2ω 1 ω 2 (ω 2 C 2ω 1 ) ω 1 (ω 2 C 2ω 1 ) (10.140) iδ 2 2 i σ T1 i 2 δ 22 A1e AP 2 D A 2 A 1 AN 1 2ω 2 ω 1 ω 2 (ω 2 C 2ω 1 )

(10.141)

Equations 10.140 and 10.141 reduce to (10.76) and (10.77) when ω 2 D 2ω 1 .

10.7 Generalized Method of Averaging

Using the method of variation of parameters, we transform (10.53) and (10.54) into u m D a m (t) cos φ m (t) ,

uP m D ω m a m (t) sin φ m (t)

(10.142)

where δ 2 a 1 a 2 [sin (φ 2 2φ 1 ) sin (φ 2 C 2φ 1 )] 2ω 1 δ 2 φP 1 D ω 1 C a 2 [cos (φ 2 2φ 1 ) C cos (φ 2 C 2φ 1 ) C 2 cos φ 2 ] 2ω 1 δ 2 2 a [sin (φ 2 2φ 1 ) C sin (φ 2 C 2φ 1 ) C 2 sin φ 2 ] aP 2 D 4ω 2 1

(10.144)

δ 2 a 21 [cos (φ 2 2φ 1 ) C cos (φ 2 C 2φ 1 ) C 2 cos φ 2 ] φP 2 D ω 2 C 4ω 2 a 2

(10.146)

aP 1 D

(10.143)

(10.145)

We seek a second-order approximate solution of (10.143)–(10.146) in the form a m (t) D η m (t) C

2 X

n a m n (ψ1 , ψ2 , η 1 , η 2 ) C

nD1 2 X

φ m (t) D ψ m (t) C

nD1

n φ m n (ψ1 , ψ2 , η 1 , η 2 ) C

(10.147) (10.148)

10.7 Generalized Method of Averaging

where the a m n and φ m n are fast-varying functions of time, ηP m D A m1 (η 1 , η 2 , t) C 2 A m2 (η 1 , η 2 , t) C

(10.149)

P m D ω m C Φm1 (η 1 , η 2 , t) C 2 Φm2 (η 1 , η 2 , t) C ψ

(10.150)

and the A m n and Φm n are slowly varying functions of time. Substituting (10.147)– (10.150) into (10.143)–(10.146) and equating the coefﬁcients of to zero yields A 11 C ω 1

@a 11 @a 11 δ2 C ω2 D η 1 η 2 [sin (ψ2 2ψ1 ) sin (ψ2 C 2ψ1 )] @ψ1 @ψ2 2ω 1 (10.151)

Φ11 C ω 1

A 21 C ω 1

@φ 11 @φ 11 δ2 C ω2 D η 2 [cos (ψ2 2ψ1 ) @ψ1 @ψ2 2ω 1 C cos (ψ2 C 2ψ1 ) C 2 cos ψ2 ] @a 21 @a 21 δ2 2 C ω2 D η [sin (ψ2 2ψ1 ) @ψ1 @ψ2 4ω 2 1 C sin (ψ2 C 2ψ1 ) C 2 sin ψ2 ]

Φ21 C ω 1

(10.152)

@φ 21 @φ 21 δ 2 η 21 [cos (ψ2 2ψ1 ) C ω2 D @ψ1 @ψ2 4ω 2 η 2 C cos (ψ2 C 2ψ1 ) C 2 cos ψ2 ]

(10.153)

(10.154)

Choosing the A n1 and Φn1 to eliminate the slowly varying terms in (10.151)– (10.154) and using (10.65), we obtain δ2 η 1 η 2 sin (ψ2 2ψ1 ) 2ω 1 δ2 η 2 cos (ψ2 2ψ1 ) Φ11 D 2ω 1 δ2 2 η sin (ψ2 2ψ1 ) A 21 D 4ω 2 1

A 11 D

Φ21 D

δ 2 η 21 cos (ψ2 2ψ1 ) 4ω 2 η 2

(10.155) (10.156) (10.157) (10.158)

Then, the solutions of (10.151)–(10.154) are given uniquely by δ2 η1 η2 cos (ψ2 C 2ψ1 ) 2ω 1 (ω 2 C 2ω 1 ) δ2 η2 δ2 η2 D sin ψ2 sin (ψ2 C 2ψ1 ) C 2ω 1 (ω 2 C 2ω 1 ) ω1 ω2

a 11 D

(10.159)

φ 11

(10.160)

δ 2 η 21 δ 2 η 21 cos ψ2 cos (ψ2 C 2ψ1 ) 4ω 2 (ω 2 C 2ω 1 ) 2ω 22 δ 2 η 21 δ 2 η 21 D sin (ψ2 C 2ψ1 ) C sin ψ2 4ω 2 (ω 2 C 2ω 1 )η 2 2ω 22 η 2

a 21 D

(10.161)

φ 21

(10.162)

277

278

10 Systems with Quadratic and Cubic Nonlinearities

Substituting (10.155)–(10.162) into the second-order problem and choosing the A n2 and Φn2 to eliminate the slowly varying terms in the resulting equations, we obtain A 12 D 0 and

A 22 D 0

(10.163)

δ 22 (8ω 21 C 5ω 1 ω 2 )η 21 C 2ω 22 η 22 8ω 21 ω 22 (ω 2 C 2ω 1 ) δ 22 D η2 4ω 1 ω 2 (ω 2 C 2ω 1 ) 1

Φ12 D

(10.164)

Φ22

(10.165)

Substituting (10.147) and (10.148) into (10.142), expanding the result for small , and using (10.159)–(10.162) yields u 1 D η 1 cos (ω 1 t C β 1 ) C

δ 2 η 1 η 2 cos [(ω 2 ω 1 ) t C β 2 β 1 ] 2ω 1 ω 2

δ 2 η 1 η 2 cos [(ω 2 C ω 1 ) t C β 2 C β 1 ] C ω 2 (ω 2 C 2ω 1 )

u 2 D η 2 cos (ω 2 t C β 2 )

(10.166)

δ 2 η 21 δ 2 2 η C cos (2ω 1 t C 2β 1 ) 4ω 2 (ω 2 C 2ω 1 ) 2ω 22 1 (10.167)

Substituting (10.155)–(10.158) and (10.163)–(10.165) into (10.149) and (10.150), we obtain the modulation equations ηP 1 D

δ 2 η 1 η 1 sin [(ω 2 2ω 1 ) t C β 2 2β 1 ] C O( 3 ) 2ω 1

(10.168)

δ 2 η 2 cos [(ω 2 2ω 1 ) t C β 2 2β 1 ] 2ω 1 2 δ 22 (8ω 21 C 5ω 1 ω 2 )η 21 C 2ω 22 η 22 C O( 3 ) 2 2 8ω 1 ω 2 (ω 2 C 2ω 1 )

P 1 D ω1 C ψ

(10.169) ηP 2 D

δ 2 2 η sin [(ω 2 2ω 1 ) t C β 2 2β 1 ] C O( 3 ) 4ω 2 1 δ 2 η 21 cos [(ω 2 2ω 1 ) t C β 2 2β 1 ] 4ω 2 η 2 2 δ 22 η 2 C O( 3 ) 4ω 1 ω 2 (ω 2 C 2ω 1 ) 1

(10.170)

P 2 D ω2 C ψ

(10.171)

Equations 10.166–10.171 are in full agreement with (10.85), (10.86), and (10.79)– (10.82) obtained with the method of multiple scales because ω 2 2ω 1 , η n D a n , and ψ n D ω n t C β n .

10.8 A Nonsemisimple One-to-One Internal Resonance

10.8 A Nonsemisimple One-to-One Internal Resonance

In this section, we use the methods of normal form and multiple scales to treat a two-degree-of-freedom system with quadratic and cubic nonlinearities whose linear part has a generic nonsemisimple structure. Speciﬁcally, we treat the system uR 1 C ω 2 u 1 C 2µ 1 uP 1 C δ 11 u21 C δ 12 u 1 u 2 C δ 13 u22 C α 11 u31 C α 12 u21 u 2 C α 13 u 1 u22 C α 14 u32 D 0

(10.172)

uR 2 C ω 2 u 2 C 2µ 2 uP 2 C u 1 C δ 21 u21 C δ 22 u 1 u 2 C δ 23 u22 C α 21 u31 C α 22 u21 u 2 C α 23 u 1 u22 C α 24 u32 D 0

(10.173)

10.8.1 The Method of Normal Forms

Again, as a ﬁrst step, we use the transformation (9.37) and (9.38) to recast (10.172) and (10.173) in the complex-valued form 2 i h ζP 1 D i ωζ1 µ 1 ζ1 ζN 1 C δ 11 ζ1 C ζN 1 C δ 12 ζ1 C ζN 1 ζ2 C ζN 2 2ω 2 i 3 2 i h N C α 11 ζ1 C ζN 1 C α 12 ζ1 C ζN 1 ζ2 C ζN 2 Cδ 13 ζ2 C ζ2 2ω 2 3 i Cα 13 ζ1 C ζN 1 ζ2 C ζN 2 C α 14 ζ2 C ζN 2 (10.174) 2 i i h ζP 2 D i ωζ2 µ 2 ζ2 ζN 2 C ζ1 C ζN 1 C δ 21 ζ1 C ζN 1 2ω 2ω 2 i 3 i h Cδ 22 ζ1 C ζN 1 ζ2 C ζN 2 C δ 23 ζ2 C ζN 2 C α 21 ζ1 C ζN 1 2ω 2 2 Cα 22 ζ1 C ζN 1 ζ2 C ζN 2 C α 23 ζ1 C ζN 1 ζ2 C ζN 2 3 i (10.175) Cα 24 ζ2 C ζN 2 To simplify the algebra, we scale the variables in (10.174) and (10.175) before determining their normal form. We treat the case of weak nonlinearities and damping. Thus, we assume that ζ1 D O(), where is a small nondimensional parameter that is used as a bookkeeping device. Because ζ2 is much larger than ζ1 due to the nonsemisimple structure of the linear undamped operator in (10.172) and (10.173), ζ2 D O( 1λ 2 ), where λ 2 > 0. Because the damping is weak, µ n D O( λ 1 ), where λ 1 > 0. To make these different orderings explicit, we introduce new scaled variables deﬁned by ζ1 D η 1 ,

ζ2 D 1λ 2 η 2 ,

µ n D λ 1 µO n

279

280

10 Systems with Quadratic and Cubic Nonlinearities

where the η n and µO n are O(1) and rewrite (10.174) and (10.175) as ηP 1 D i ωη 1 λ 1 µO 1 (η 1 ηN 1 ) C

i h δ 11 (η 1 C ηN 1 )2 2ω

i C 1λ 2 δ 12 (η 1 C ηN 1 ) (η 2 C ηN 2 ) C 12λ 2 δ 13 (η 2 C ηN 2 )2 i h 2 C α 11 (η 2 C ηN 1 )3 C 2λ 2 α 12 (η 1 C ηN 1 )2 (η 2 C ηN 2 ) 2ω i C 22λ 2 α 13 (η 1 C ηN 1 ) (η 2 C ηN 2 )2 C 23λ 2 α 14 (η 2 C ηN 2 )3

(10.176)

ηP 2 D i ωη 2 λ 1 µO 2 (η 2 ηN 2 )

i h 1Cλ 2 i λ2 δ 21 (η 1 C ηN 1 )2 (η 1 C ηN 1 ) C 2ω 2ω i C δ 22 (η 1 C ηN 1 ) (η 2 C ηN 2 ) C 1λ 2 δ 23 (η 2 C ηN 2 )2 i h 2Cλ 2 C α 21 (η 1 C ηN 1 )3 C 2 α 22 (η 1 C ηN 1 )2 (η 2 C ηN 2 ) 2ω i C 2λ 2 α 23 (η 1 C ηN 1 ) (η 2 C ηN 2 )2 C 22λ 2 α 24 (η 2 C ηN 2 )3 C

(10.177)

Because of the one-to-one internal resonance, η 1 ηN 1 η 2 , η 2 ηN 2 η 1 , η 21 ηN 1 , η 22 ηN 2 , η 21 ηN 2 , η 22 ηN 1 , η 1 , and η 2 are resonance terms in (10.176) and (10.177). Balancing the damping term and the dominant resonance term in (10.176), namely, the term proportional to α 14 , we have λ 1 D 2 3λ 2

(10.178)

Balancing the damping term and the dominant resonance term in (10.177), namely, the term i(2ω)1 η 1 , we have λ1 D λ2

(10.179)

Solving (10.178) and (10.179) yields λ1 D λ2 D

1 2

(10.180)

Hence, (10.176) and (10.177) become ηP 1 D i ωη 1 C C

i δ 13 (η 2 C ηN 2 )2 1/2 µO 1 (η 1 ηN 1 ) 2ω

i 1/2 δ 12 (η 1 C ηN 1 ) (η 2 C ηN 2 ) C α 14 (η 2 C ηN 2 )3 C 2ω

ηP 2 D i ωη 2 1/2 µO 2 (η 2 ηN 2 ) C

(10.181)

i 1/2 η 1 C ηN 1 C δ 23 (η 2 C ηN 2 )3 C 2ω (10.182)

10.8 A Nonsemisimple One-to-One Internal Resonance

To simplify (10.181) and (10.182), we introduce the transformation η 1 D ξ1 C h 11 ξn , ξN n C 1/2 h 12 ξn , ξN n C

(10.183)

η 2 D ξ2 C 1/2 h 22 ξn , ξN n C

(10.184)

and choose the h m n to eliminate the nonresonance terms so that the resulting equations have the simplest possible form ξP n D i ωξn C 1/2 g n ξm , ξN m C

(10.185)

Substituting (10.183)–(10.185) into (10.181) and (10.182) and equating coefﬁcients of like powers of , we obtain 2 i (10.186) δ 13 ξ2 C ξN2 2ω @h 11 @h 11 @h 11 @h 11 gN 1 g2 gN 2 g 1 C L(h 12 ) D g 1 @ξ1 @ξ2 @ ξN1 @ ξN2 i µO 1 ξ1 ξN1 C h 11 hN 11 C δ 13 ξ2 C ξN2 h 22 C hN 22 ω 3 i i h N N C (10.187) δ 12 ξ1 C ξ1 C h 11 C h 11 ξ2 C ξN2 C α 14 ξ2 C ξN2 2ω i i h ξ1 C ξN1 C h 11 C hN 11 g 2 C L(h 22 ) D µO 2 ξ2 ξN2 C 2ω 2 i C (10.188) δ 23 ξ2 C ξN2 2ω

L(h 11 ) D

where

@h @h N @h @h N ξ1 C ξ2 h L(h) D i ω ξ1 ξ2 @ξ1 @ξ2 @ ξN1 @ ξN2

The right-hand side of (10.186) suggests seeking h 11 in the form h 11 D Γ1 ξ22 C Γ2 ξ2 ξN2 C Γ3 ξN22

(10.189)

Substituting (10.189) into (10.186) and equating each of the coefﬁcients of ξ22 , ξ2 ξN2 , and ξN22 on both sides, we obtain Γ1 D

δ 13 δ 13 δ 13 , Γ2 D 2 , Γ3 D 2 2 2ω ω 6ω

(10.190)

It follows from (10.189) and (10.190) that δ 13 2 ξ C ξN22 6ξ2 ξN2 h 11 C hN 11 D 3ω 2 2

(10.191)

281

282

10 Systems with Quadratic and Cubic Nonlinearities

Substituting (10.191) into (10.188) yields @h 22 @h 22 N @h 22 @h 22 N ξ1 C ξ2 h 22 D µO 2 ξ2 ξN2 g2 C i ω ξ1 ξ2 @ξ1 @ξ2 @ ξN1 @ ξN2 2 i δ 13 2 N 2 N N ξ1 C ξN1 C ξ ξ C δ ξ C C ξ 6ξ C ξ (10.192) 2 2 23 2 2 2 2ω 3ω 2 2

Equating g 2 to the resonance terms in (10.192), we have g 2 D µO 2 ξ2 C

i ξ1 2ω

(10.193)

Then, substituting h 22 D Γ4 ξN2 C Γ5 ξN1 C Γ6 ξ22 C Γ7 ξ2 ξN2 C Γ8 ξN22

(10.194)

into (10.192), using (10.193), and equating each of the coefﬁcients of ξN2 , ξN1 , ξ22 , ξ2 ξN2 , and ξN22 on both sides, we obtain i µO 2 1 δ 23 δ 13 C , , Γ5 D 2 , Γ6 D 2 2ω 4ω 2ω 6ω 4 δ 23 δ 13 δ 23 δ 13 Γ7 D 2 C 4 , Γ8 D 2 ω ω 6ω 18ω 4

Γ4 D

(10.195)

Substituting (10.189)–(10.191) and (10.193)–(10.195) into (10.187) yields @h 12 @h 12 N @h 12 @h 12 N g1 C i ω ξ1 C ξ2 h 12 ξ1 ξ2 @ξ1 @ξ2 @ ξN1 @ ξN2 i D 2Γ1 ξ2 C Γ2 ξN2 µO 2 ξ2 C ξ1 2ω i N ξ1 Γ2 ξ2 C 2Γ3 ξN2 µO 2 ξN2 2ω 2 2δ 13 2 N N ξ ξ2 µO 1 ξ1 ξ1 C 3ω 2 2 i µO 2 1 i ξ1 C ξN1 C δ 13 ξ2 C ξN2 ξ2 ξN2 2 ω 2ω 4ω 2 δ δ 13 2 N 2 23 2 N N N ξ C ξ2 6ξ2 ξ2 C 4 ξ2 C ξ2 C 18ξ2 ξ2 C 9ω 3ω 2 2 3 i δ 12 ξ1 C ξN1 ξ2 C ξN2 C α 14 ξ2 C ξN2 C 2ω δ 12 δ 13 2 N 2 N N ξ2 C ξ2 6ξ2 ξ2 ξ2 C ξ2 (10.196) C 3ω 2 Since we are stopping at this order, we do not need to determine h 12 explicitly, and hence all that we need to do is to determine g 1 . Choosing g 1 to eliminate the resonance terms from (10.196), we have g 1 D µO 1 ξ1 C i α e ξ22 ξN2

(10.197)

10.8 A Nonsemisimple One-to-One Internal Resonance

where 1 10 5 38 2 3α 14 αe D δ 13 δ 23 δ 12 δ 13 C δ 2ω 3ω 2 3ω 2 9ω 4 13

(10.198)

Substituting for the g n and h m n in (10.183)–(10.185), we obtain η 1 D ξ1 C

δ 13 2 3ξ2 6ξ2 ξN2 ξN22 C 6ω 2

(10.199)

η 2 D ξ2 C

(10.200)

ξP1 D i ωξ1 1/2 µO 1 ξ1 C i 1/2 α e ξ22 ξN2 C

(10.201)

i 1/2 ξP2 D i ωξ2 1/2 µO 2 ξ2 C ξ1 C 2ω

(10.202)

10.8.2 The Method of Multiple Scales

Alternatively, we use the method of multiple scales to determine an approximation to the solution of (10.172) and (10.173). Following a procedure similar to that used above, we note that if u 1 D O(), then u 2 D O( 1/2 ) and µ n D O( 1/2 ). Hence, we introduce scaled variables deﬁned by u 1 D ν 1 ,

u 2 D 1/2 ν 2 ,

µ n D 1/2 µO n

(10.203)

in (10.172) and (10.173) and obtain νR 1 C ω 2 ν 1 C δ 13 ν 22 C 2 1/2 µO 1 νP 1 C 1/2 δ 12 ν 1 ν 2 C α 14 ν 32 C D 0 (10.204) νR 2 C ω 2 ν 2 C 2 1/2 µO 2 νP 2 C 1/2 ν 1 C δ 23 ν 22 C D 0

(10.205)

We seek an expansion of the solution of (10.204) and (10.205) valid up to O( 1/2 ) in the form ν 1 (tI ) D ν 10 (T0 , T1 ) C 1/2 ν 11 (T0 , T1 ) C

(10.206)

ν 2 (tI ) D ν 20 (T0 , T1 ) C 1/2 ν 21 (T0 , T1 ) C

(10.207)

where T0 D t and T1 D 1/2 t. In terms of T0 and T1 , the time derivatives become d D D0 C 1/2 D1 C dt

and

d2 D D02 C 2 1/2 D0 D1 C d t2

where D n D @/@Tn . Substituting (10.206) and (10.207) into (10.204) and (10.205) and equating coefﬁcients of like powers of , we obtain

283

284

10 Systems with Quadratic and Cubic Nonlinearities

Order (0 )

D02 ν 10 C ω 2 ν 10 D δ 13 ν 220

(10.208)

D02 ν 20 C ω 2 ν 20 D 0

(10.209)

Order ()

D02 ν 11 C ω 2 ν 11 D 2D0 D1 ν 10 2 µO 1 D0 ν 10 2δ 13 ν 20 ν 21 δ 12 ν 10 ν 20 α 14 ν 320

(10.210)

D02 ν 21 C ω 2 ν 21 D 2D0 D1 ν 20 2 µO 2 D0 ν 20 ν 10 δ 23 ν 220

(10.211)

The solution of (10.209) can be expressed as ν 20 D A 2 (T1 )e i ω T0 C cc

(10.212)

Then, (10.208) becomes D02 ν 10 C ω 2 ν 10 D δ 13 A22 e 2i ω T0 C A 2 AN 2 C cc whose solution can be expressed as ν 10 D A 1 (T1 )e i ω T0 C

δ 13 2 2i ω T0 δ 13 A e 2 A 2 AN 2 C cc 3ω 2 2 ω

(10.213)

Substituting (10.212) and (10.213) into (10.210) and (10.211) yields D02 ν 11 C ω 2 ν 11 D 2i ω A01 C µO 1 A 1 e i ω T0 5 A22 AN 2 e i ω T0 3α 14 δ δ 12 13 3ω 2 C cc C NST 2δ 13 ν 20 ν 21

(10.214)

δ 13 2 2i ω T0 D02 ν 21 C ω 2 ν 21 D 2i ω A02 C µO 2 A 2 e i ω T0 A 1 e i ω T0 A e 3ω 2 2 δ 13 C 2 A 2 AN 2 δ 23 A22 e 2i ω T0 C A 2 AN 2 C cc (10.215) ω Eliminating the term that produces secular terms from (10.215), we have 2i ω A02 C µO 2 A 2 C A 1 D 0 (10.216) Then, the solution of (10.215) can be expressed as δ 13 δ 23 δ 13 δ 23 2 2i ω T0 ν 21 D A A 2 AN 2 C cc C e C 2 9ω 4 3ω 2 ω4 ω2

(10.217)

Substituting (10.217) into (10.214) and eliminating the terms that produce secular terms, we obtain 2i ω A01 C µO 1 A 1 C 2ωα e A22 AN 2 D 0 (10.218) where α e is given by (10.198).

10.9 Exercises

Letting ξn D A n e i ω n t in (10.201) and (10.202) yields (10.216) and (10.218). Hence, the results obtained by using the method of normal forms are in full agreement with those found by using the method of multiple scales.

10.9 Exercises

10.9.1 Consider the system uR 1 C ω 21 u 1 D δ 1 uP 1 uP 2 uR 2 C ω 22 u 2 D δ 2 uP 21 Use the method of normal forms, the method of multiple scales, and the generalized method of averaging to determine a second-order approximation to the solution of this system when ω 2 2ω 1 . 10.9.2 Consider the system uR 1 C ω 21 u 1 D δ 1 u 1 uR 2 uR 2 C ω 22 u 2 D δ 2 u 1 uR 1 Use the method of normal forms, the method of multiple scales, and the generalized method of averaging to determine a second-order approximation to the solution of this system when ω 2 2ω 1 .

285

287

11 Retarded Systems Several approaches have been proposed in the literature to analyze the nature of Hopf bifurcations of retarded systems, including integral averaging, the Fredholm alternative, the implicit function theorem, the method of multiple scales, and center-manifold reduction. In this chapter, we follow Nayfeh (2008) and use two of these approaches, the method of multiple scales and center-manifold reduction, to analyze the nature of Hopf bifurcations in retarded systems modeled by nonlinear homogeneous ordinary-differential equations with discrete time delay. Such equations model the behavior of many physical systems arising in physiology, biology, population dynamics, machine-tool chatter, neural networks, and time-delayed feedback controlled systems. To describe the approaches without getting involved in the algebra, we use three simple systems, namely a scalar equation, a single-degree-of-freedom system, and a three-neuron model. We show that the normal forms obtained with all the approaches are the same. However, the method of multiple scales seems to be simpler. In fact, the method of multiple scales is directly applied to the retarded differential equations. In contrast, in the center-manifold approach, one needs to convert the retarded differential equations into operator equations in a Banach space, introduce a device that acts like an inner product because the Banach space does not have a natural inner product associated with its norm, deﬁne the adjoint associated with the linear part of the differential equations, perform the projection onto the center manifold, and calculate the normal form of the subsystem describing the dynamics on the center manifold. Finally, we consider a problem in which the retarded term appears as an acceleration and treat it using the method of multiple scales only.

11.1 A Scalar Equation

We start by a scalar equation with discrete time delay, which models a single neuron with a general activation function; that is, d x(t) O D x(t) O C g x(t) O β x(t O τ) dt

(11.1)

The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

288

11 Retarded Systems

where g( x) O is a general three-times continuously differentiable function, τ > 0, and β > 0. We let x denote an equilibrium of (11.1); that is, x D g[(1 β)x ] Then, we let x(t) O D x C x (t) in (11.1), expand the result in a Taylor series in x, keep up to cubic terms, and obtain x(t) P D (α 1 1)x (t) α 1 β x (t τ) α 2 [x (t) β x (t τ)]2 3 α 3 x (t) β x (t τ)

(11.2)

where α 1 D g 0 [(1 β)x ] ,

α 2 D 12 g 00 [(1 β)x ] ,

α 3 D 61 g 000 [(1 β)x ]

Linearizing (11.2) yields x(t) P D (α 1 1)x (t) α 1 β x (t τ)

(11.3)

which has solutions of the form x D x0 e (σCi ω)t

(11.4)

where σ is the growth or decay rate, ω is the frequency of oscillations, and x0 depends on the initial conditions. Substituting (11.4) into (11.3) yields the characteristic equation σ C i ω D α 1 1 α 1 β e (σCi ω)τ

(11.5)

For a given α 1 , β, and τ, the complex-valued (11.5) has inﬁnitely many solutions σ and ω. When σ > 0 the trivial solution is unstable; when σ < 0 the trivial solution is asymptotically stable; and σ D 0 deﬁnes the stability boundary. To locate this boundary, we let σ D 0 in (11.5) and obtain the complex-valued characteristic equation i ω D α 1 1 α 1 β e i ω τ which, upon separating its real and imaginary parts, yields ω D α 1 β sin(ωτ) α 1 1 D α 1 β cos(ωτ) Hence, ωD

q

α 21 β 2 (α 1 1)2

(11.6)

11.1 A Scalar Equation

For real frequencies, the coefﬁcient under the radical must be positive; that is, α 21 β 2 (α 1 1)2 > 0. In what follows, we take β to be the bifurcation parameter and denote its critical value that separates stable from unstable trivial solutions by β c and the corresponding value of ω by ω c . It follows from (11.5) that α 1 [cos(ω c τ) α 1 β τ] d(σ C i ω) D 6D 0 Real dβ [cos(ω c τ) α 1 β τ]2 C sin2 (ω c τ) at (β, ω) D (β c , ω c ) and hence the bifurcation is a Hopf bifurcation. Next, we construct the normal form of this bifurcation by using the method of multiple scales in the next section and center-manifold reduction in Section 11.1.2. 11.1.1 The Method of Multiple Scales

We seek a uniform second-order approximate solution of (11.2) in the form x (tI ) D x1 (T0 , T2 ) C 2 x2 (T0 , T2 ) C 3 x3 (T0 , T2 ) C

(11.7)

where T0 D t, T2 D 2 t, and is a nondimensional bookkeeping parameter. The solution does not depend on the slow scale T1 D t because secular terms ﬁrst appear at O( 3 ). The derivative with respect to t is transformed into d @ @ C 2 C D D0 C 2 D2 C D dt d T0 @T2

(11.8)

Moreover, we express x (t τ) in terms of the scales T0 and T2 as 3 X

x (t τI ) D

m x m T0 τ, T2 2 τ C

mD1

which upon expansion for small becomes 3 X

x (t τI ) D

m x m (T0 τ, T2 ) 3 τ D2 x1 (T0 τ, T2 ) C

(11.9)

mD1

Next, we introduce the detuning parameter δ to describe the nearness of β to the critical value β c deﬁned by β D βc C 2 δ Substituting (11.7)–(11.9) into (11.2) and equating coefﬁcients of like powers of , we obtain D0 x1 C (1 α 1 )x1 C α 1 β c x1τ D 0

(11.10)

D0 x2 C (1 α 1 )x2 C α 1 β c x2τ D α 2 (x1 β c x1τ )2

(11.11)

289

290

11 Retarded Systems

D0 x3 C (1 α 1 )x3 C α 1 β c x3τ D D2 x1 α 1 δ x1τ C α 1 β c τ D2 x1τ 2α 2 (x1 β c x1τ ) (x2 β c x2τ ) α 3 (x1 β c x1τ )3

(11.12)

where x i D x i (T0 , T2 ) and x i τ D x i (T0 τ, T2 ). The general solution of (11.10) can be expressed as N 2 )e i ω c T0 C x1 D A(T2 )e i ω c T0 C A(T C AN m (T2 )e (σ m i ω m )T0

i

1 h X

A m (T2 )e (σ m Ci ω m )T0

mD1

(11.13)

where ω c is the critical frequency corresponding to σ D 0 on the stability boundary and it is given by (11.6); the σ ˙ i ω m are the remaining roots of (11.5); and the function A(T2 ) is determined by eliminating the secular terms at O( 3 ). Near the stability boundary, all of the eigenvalues have negative real parts except the eigenvalue corresponding to ω c whose real part changes sign as the stability boundary is crossed. Hence, as time increases all of the terms under the summation in (11.13) decay with time, leaving only the ﬁrst two terms. Therefore, the long-time behavior of the system is given by N 2 )e i ω c T0 x1 D A(T2 )e i ω c T0 C A(T

(11.14)

Substituting (11.14) into (11.11) yields D0 x2 C (1 α 1 )x2 C α 1 β c x2τ D α 2 γ 2 A2 e 2i ω c T0 α 2 γ γN A AN C cc (11.15) whose solution can be expressed as x2 D α 2 γ 2 Γ e 2i ω c T0 α 2 γ γN Γ0 A AN C cc

(11.16)

and γ D β c e i ω c τ 1 , Γ D

1 , 2i ω c C 1 α 1 C α 1 β c e 2i ω c τ

Γ0 D

1 1 α1 C α1 βc

(11.17)

Substituting (11.14) and (11.16) into (11.12) and eliminating the terms that lead to secular terms, we obtain the complex-valued form of the normal form of the bifurcation α 1 δ e i ω c τ A 1 α 1 β c τ e i ω c τ γ 2 γN 3α 3 2α 22 Γ 1 e 2ω c τ C 2(1 β c )Γ0 A2 AN C 1 α 1 β c τ e i ω c τ

A0 D

(11.18)

11.1 A Scalar Equation

11.1.2 Center-Manifold Reduction

We start by converting (11.2) into an operator differential equation. We note that, whereas the solution of (11.2) when τ D 0 depends on a single value, the solution of (11.2) with τ ¤ 0 depends on an entire set of values of x in the interval τ t 0. Hence, the initial space is a function space. Consequently, the delay differential equation 11.2 can be expressed as an abstract evolution equation (Campbell et al., 1995; Hale, 1977; Kuang, 1993) on the Banach space B of continuously differentiable complex functions from [τ, 0] to C 2 of complex functions with the norm jjp jj D maxτθ 0 jp (θ )j; that is, xP t D A x t C F(x t )

(11.19)

where x t (θ ) B is deﬁned by the shift operator x t (θ ) D x (t C θ ) for

τθ 0

The linear operator A is deﬁned by 8 < d p (θ ) for A p (θ ) D d θ :(α 1)p (0) α β p (τ) for 1 1 c

(11.20)

τθ 0 θ D0

and the operator F can be written as ( 0 for τ θ 0 FD f for θ D 0

(11.21)

(11.22)

where 2 3 f D 2 α 1 δ x (t τ) α 2 x (t) β c x (t τ) α 3 x (t) β c x (t τ) (11.23) To carry out the analysis, we need an inner product and the adjoint operator associated with (11.21). In contrast with C 2 , the space B does not have a natural inner product associated with its norm. However, following Hale (1977), we introduce the following bilinear form that acts like an “inner product”: Z0 hq, p i D qN (0)p (0) C

qN (ξ C τ) (α 1 β c )p (ξ )d ξ

(11.24)

τ

where q 2 B , the space of continuously differentiable functions from [0, τ] to C 2 with the norm jjqjj D max0θ τ jq(θ )j. With this bilinear form, one can construct the following formal adjoint operator (Hale, 1977) associated with (11.21): 8 0, the trivial solution is unstable; when σ < 0, the trivial solution is asymptotically stable; and σ D 0 separates stable from unstable trivial solutions. Putting σ D 0 in (11.133) and separating real and imaginary parts, we obtain ω D α β sin(ωτ) α 1 D α β cos(ωτ)

305

306

11 Retarded Systems

Hence, for given α and τ, the critical value ω c of ω is given by ω c D (α 1 1) tan(ω c τ)

(11.135)

Differentiating (11.133) with respect to σ C i ω and β and evaluating the result at the critical values yields d(σ C i ω) i ωc C 1 α D dβ β c (i ω c C 1 α) τ whose real part is different from zero. Therefore, the trivial solution loses stability as a result of a pair of complex conjugate eigenvalues transversely crossing the imaginary axis as β exceeds the critical value β c , and hence the bifurcation is a Hopf bifurcation. Next, we construct the normal form of this bifurcation by using the method of multiple scales in the following section and by using center-manifold reduction in Section 11.3.2. 11.3.1 The Method of Multiple Scales

Because the nonlinearity is cubic, we seek a uniform second-order approximate solution of (11.128)–(11.130) in powers of 1/2 instead of powers of as done above for cases with quadratic and cubic nonlinearities. Thus, we let x(tI ) D 1/2 x 1 (T0 , T1 ) C 3/2 x 2 (T0 , T1 ) C

(11.136)

where T0 D t, T1 D t, and is a nondimensional bookkeeping parameter. The derivative with respect to t, in this case, is transformed into @ d @ C C D D0 C D1 C D dt d T0 @T1

(11.137)

Moreover, we express x(t τ) in terms of the scales T0 and T1 as x(t τI ) D 1/2 x 1 (T0 τ, T1 τ) C 3/2 x 2 (T0 τ, T1 τ) C which upon expansion for small becomes x(t τI ) D 1/2 x 1 (T0 τ, T1 ) C 3/2 x 2 (T0 τ, T1 ) 3/2 τ D1 x 1 (T0 τ, T1 ) C

(11.138)

Next, we introduce the detuning parameter δ to describe the nearness of β to the critical value β c deﬁned by β D β c C δ

(11.139)

11.3 A Three-Dimensional System

Substituting (11.136)–(11.139) into (11.128) and equating coefﬁcients of like powers of yields D0 x 1 (T0 , T1 ) Lx 1 (T0 , T1 ) C β c R x 1 (T0 τ, T1 ) D 0

(11.140)

D0 x 2 (T0 , T1 ) Lx 2 (T0 , T1 ) C β c R x 2 (T0 τ, T1 ) D δ R x 1 (T0 τ, T1 ) D1 x 1 (T0 , T1 ) C β c τ R D1 x 1 (T0 τ, T1 ) C f x 1 (T0 , T1 ), x 1 (T0 τ, T1 )

(11.141)

The nondecaying solution of (11.140) can be expressed as N 1 )Nc e i ω c T0 x 1 (T0 , T1 ) D A(T1 )c e i ω c T0 C A(T

(11.142)

where c is given by 2 3 1 c D 4c25 , c3

α 2 1 β c e i ω c τ c2 D , 1 C i ωc

and

c3 D

α1

1 C i ωc 1 β c e i ω c τ (11.143)

Substituting (11.142) into (11.141) yields D0 x 2 (T0 , T1 ) Lx 2 (T0 , T1 ) C β c R x 2 (T0 τ, T1 ) D δ R cAe i ω c τ e i ω c T0 I β c τ R e i ω c τ c A0 e i ω c T0 3γ 2 γN A2 AN Of e i ω c T0 C cc C NST (11.144) where 3 α 1 c 23 cN3 Of D 4 α 2 5 α 3 c 22 cN2 2

γ D β c e i ω c τ 1 ,

(11.145)

Because the homogeneous part of (11.144) has nontrivial solutions, the nonhomogeneous equation has a solution only if a solvability condition is satisﬁed. To determine this solvability condition, we seek a particular solution of (11.144) in the form x 2 (T0 , T1 ) D φ(T1 )e i ω c T0 and obtain L β c R e i ω c τ i ω c I φ

D I β c τ R e i ω c τ c A0 C δ R cAe i ω c τ C 3γ 2 γN Of A2 AN

(11.146)

We note that the problem of ﬁnding the solvability condition for the system of differential equations (11.144) has been transformed into ﬁnding the solvability

307

308

11 Retarded Systems

condition of the system of algebraic equations (11.146). Again, because i ω c is an eigenvalue of the homogeneous part of (11.146), the nonhomogeneous equation has solutions if and only if a solvability condition is satisﬁed. This condition demands that the right-hand side of (11.146) be orthogonal to every solution of the adjoint homogeneous problem. In this case, the adjoint is governed by L β c R e i ωc τ C i ωc I b D 0 (11.147) We note that b is not unique, and to make it unique we impose the condition b .c D 1

(11.148)

Solving (11.147) and (11.148) and using (11.145), we obtain the unique adjoint

bD

1 3

2 3 1 4 b 25 , b3

b2 D

1 i ωc , α 2 (1 β c e i ω c τ )

and

b3 D

α 1 (1 β c e i ω c τ ) 1 i ωc (11.149)

Taking the “inner product” of the right-hand side of (11.146) with b yields the solvability condition, normal form, A0 D δ Λ 1 A C Λ 2 A2 AN where

(11.150)

α 1 c 2 c 23 C α 2 c 3 C α 3 c 22 e i ω c τ 3c 2 c 3 β c τ α 1 c 2 c 23 C α 2 c 3 C α 3 c 22 3 α 1 c 2 c 33 cN3 C α 2 c 3 C α 3 c 32 cN 2 γ 2 γN Λ2 D 3c 2 c 3 β c τ α 1 c 2 c 23 C α 2 c 3 C α 3 c 22

Λ1 D

(11.151)

11.3.2 Center-Manifold Reduction

Again, we start by representing (11.128) as an operator differential equation; that is, xP t D A x t C F(x t )

(11.152)

where x t (θ ) B is deﬁned by the shift operator x t (θ ) D x(t C θ ) for

τθ 0

the linear operator A is deﬁned by 8 < d p (θ ) for A p (θ ) D d θ : L p(0) β R p (τ) for c

τθ 0 θ D0

(11.153)

11.3 A Three-Dimensional System

and the operator F can be written as ( 0 for τ θ 0 FD f δ R x(t τ) for θ D 0 The adjoint operator associated with (11.153) is deﬁned by 8 0. Determine the range of gain k and time delay τ for which the limit cycles are quenched. 11.5.3 Consider the system uR C ω 2 u C 2µ uP C α u3 C k u(t τ) D f cos Ω t Determine the resonance frequency for the linear system. Then, determine an approximation to the nonlinear system response for values of Ω near the resonance frequency.

313

315

References Abed, E.H. (1994) Bifurcation–theoretic issues in the control of voltage collapse, in Proceedings of the IMA Workshop on System and Control Theory for Power Systems (eds J.H. Chow, P.V. Kokotovic, and R.J. Thomas), Springer-Verlag, New York. Andronov, A.A., Vitt, E.A., and Khaiken, S.E. (1966) Theory of Oscillations, Pergamon Press, Oxford. Arnold, V.I. (1978) Mathematical Methods of Classical Mechanics, Springer-Verlag, Berlin. Arnold, V.I. (1983) Geometrical Methods in the Theory of Ordinary Differential Equations, Springer-Verlag, New York. Arnold, V.I. (1987) Ordinary Differential Equations, MIT Press, Cambridge, Massachusetts. Arnold, V.I. (1988) Dynamical Systems, Springer-Verlag, New York. Babcock, K.L. and Westervelt, R.M. (1987) Dynamics of simple electronic neural networks. Physica D, 28, 305–316. Bogoliubov, N.N. and Mitropolsky, Y.A. (1961) Asymptotic Methods in the Theory of Nonlinear Oscillations, Gordon & Breach, New York. Campbell, S.A., Bélair, J., Ohira, T., and Milton, J. (1995) Limit cycles, tori and complex dynamics in a second order differential equation with delayed negative feedback. J. Dyn. Diff. Eqs., 7, 213–236. Carr, J. (1981) Applications of Centre Manifold Theory, Springer-Verlag, New York. Coddington, E.A. and Levinson, N. (1955) Theory of Ordinary Differential Equations, McGraw-Hill, New York.

Cole, D. (1951) On a quasilinear parabolic equation occurring in aerodynamics. Q. Appl. Math., 9, 225–236. Drazin, P.G. (1992) Nonlinear Systems, Cambridge University Press, Cambridge, England. Dumortier, F. (1977) Singularities of vector ﬁelds on the plane, J. Diff. Eqs., 23, 53–106. Fallside, F. and Patel, M.R. (1965) Stepresponse behaviour of a speed-control system with a back-e.m.f. nonlinearity. Proc. IEE, 112, 1979–1984. Gilsinn, D. (2002) Estimating critical Hopf bifurcation parameters for a second-order delay differential equation with application to machine tool. Nonlinear Dyn., 30, 103– 154. Golubitsky, M. and Schaeffer, D.G. (1985) Singularities and Groups in Bifurcation Theory, Vol. 1, Springer-Verlag, New York. Gopalsamy, K. and Leung, I. (1996) Delay induced periodicity in a neural network of excitation and inhibition. Physica D, 89, 395–426. Guckenheimer, J. and Holmes, P.J. (1983) Nonlinear Oscillations, Dynamical Systems and Bifurcation of Vectorﬁelds, SpringerVerlag, New York. Hale, J.K. (1977) Theory of Functional Differential Equations, Applied Mathematical Sciences, Vol. 3, Springer-Verlag, New York. Hale, J. (1980) Ordinary Differential Equations, Krieger, Malabar, Florida. Hanna, N.H. and Tobias, S.A. (1974) A theory of nonlinear regenerative chatter. ASME J. Eng. Indust., 96, 247–255. Hirsch, M. W. and Smale, S. (1974) Differential Equations, Dynamical Systems, and Linear Algebra, Academic Press, New York.

The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

316

References Hopf, E. (1950) The partial differential equation u t C uu x D µu x x . Comm. Pure Appl. Math., 3, 201–230. Hopﬁeld, J. (1984) Neurons with graded response have collective computational properties like those of two-state neurons. Proc. Nat. Acad. Sci. USA, 81, 3088–3092. Iooss, G. (1979) Bifurcation of Maps and Applications, North Holland, Amsterdam. Kalmár-Nagy, T., Stépán, G., and Moon, F.C. (2001) Subcritical Hopf bifurcation in the delay equation model for machine tool vibrations. Nonlinear Dyn., 26, 121–142. Kamel, A.A. (1970) Perturbation method in the theory of nonlinear oscillations. Celestial Mech. Dyn. Astron., 3, 90–106. Kuang, Y. (1993) Delay Differential Equations with Applications in Population Dynamics, Academic Press, Boston, Massachusetts. Liao, X., Guo, S., and Li, C. (2007) Stability and bifurcation analysis in tri-neuron model with time delay. Nonlinear Dyn., 49, 319–345. Liao, X., Wong, K.W., and Wu, Z. (2001) Bifurcation analysis on a two-neuron system with distributed delays. Physica D, 149, 123–141. Lorenz, E.N. (1963) Deterministic nonperiodic ﬂow. J. Atmos. Sci., 20, 130–141. Marsden, J.E. and McCracken, M. (1976) The Hopf Bifurcation and Its Applications, Springer-Verlag, New York. Mingori, D.L. and Harrison, J.A. (1974) Circularly constrained particle motion in spinning and coning bodies. AIAA Journal, 12, 1553–1558. Namachchivaya, N.S. and Malhotra, N. (1992) Parametrically excited Hopf bifurcation with non-semisimple 1 : 1 resonance, in Nonlinear Vibrations (eds R.A. Ibrahim, N.S. Namachchivaya, and A.K. Bajaj), ASME DE, 50, 29–46. Nayfeh, A.H. (1968) Forced oscillations of van der Pol oscillator with delayed amplitude limiting. IEEE Trans., CT–15, 192–200. Nayfeh, A.H. (1973) Perturbation Methods, Wiley-Interscience, New York. Nayfeh, A.H. (1981) Introduction to Perturbation Techniques, Wiley-Interscience, New York. Nayfeh, A.H. (1984) The response of single degree of freedom systems with quadratic

and cubic non-linearities to a subharmonic excitation. J. Sound Vib., 89, 457–470. Nayfeh, A.H. (1985) Problems in Perturbation, Wiley-Interscience, New York. Nayfeh, A.H. (1986) Perturbation methods in nonlinear dynamics, in Nonlinear Dynamics Aspects of Particle Accelerators, Lecture Notes in Physics, No. 247, Springer-Verlag, New York, 238–314. Nayfeh, A.H. (2008) Order reduction of retarded nonlinear systems – the method of multiple scales versus center-manifold reduction. Nonlinear Dyn., 51, 483–500. Nayfeh, A.H. and Balachandran, B. (1995) Applied Nonlinear Dynamics, WileyInterscience, New York. Nayfeh, A.H., Chin, C.M., and Pratt, J. (1997) Perturbation methods in nonlinear dynamics – Applications to machining dynamics. J. Manuf. Sci. Eng., 119, 485–493. Nayfeh, A.H. and Jebril, A.E.S. (1987) The response of two-degree-of-freedom systems with quadratic and cubic non-linearities to multifrequency parametric excitations. J. Sound Vib., 115, 83–101. Nayfeh, A.H. and Mook, D.T. (1979) Nonlinear Oscillations, Wiley-Interscience, New York. Nayfeh, N.A. (2006) Local and Global Stability and Dynamics of a Class of Nonlinear TimeDelayed One-Degree-of-Freedom Systems, PhD Dissertation, Virginia Polytechnic Institute and State University, Blacksburg, Virginia. Nayfeh, T.A., Asrar, W., and Nayfeh, A.H. (1992) Three-mode interactions in harmonically excited systems with quadratic nonlinearities, Nonlinear Dyn., 3, 385–410. Poincaré, H. (1899) Les Méthodes Nouvelles de la Mécanique Céleste, 3 Vols., Gauthier-Villars, Paris. Rössler, O.E. (1976a) An equation for continuous chaos. Phys. Lett. A, 57, 397–398. Shaw, S.W. and Rand, R.H. (1989) The transition to chaos in a simple mechanical system. Int. J. Nonlinear Mech., 24, 41–56. Tezak, E.G., Nayfeh, A.H., and Mook, D.T. (1982) Parametrically excited nonlinear multi-degree-of-freedom systems with repeated frequencies. J. Sound Vib., 85, 459– 472. Thom, R. (1975) Structural Stability and Morphogenesis, Benjamin/Cummings, Menlo Park, California.

References Van Dyke, M. (1964) Perturbation Methods in Fluid Mechanics, Academic Press, New York. Annotated version (1975), Parabolic Press, Palo Alto, California. Walter, W. (1976) Gewohnliche Differentialgleichungen, Springer-Verlag, New York. Whitham, G.B. (1974) Linear and Nonlinear Waves, Wiley-Interscience, New York.

Wiggins, S. (1988) Global Bifurcations and Chaos: Analytical Methods, Springer-Verlag, New York. Wiggins, S. (1990) Introduction to Applied Nonlinear Dynamical Systems and Chaos, Springer-Verlag, New York.

317

319

Further Reading Abraham, R.H. and Shaw, C.D. (1982) Dynamics – The Geometry of Behavior, Vols. 1–4, Aerial Press, Santa Cruz, California. Ariaratnam, S.T. and Namachchivaya, S.N. (1984) Degenerate Hopf bifurcation, in Proceedings of the IEEE International Symposium on Circuits and Systems, Vol. 3, Montreal, Canada, pp. 1343–1348. Arrowsmith, D.K. and Place, C.M. (1982) Ordinary Differential Equations, Chapman and Hall, New York. Arrowsmith, D.K. and Place, C.M. (1990) An Introduction to Dynamical Systems, Cambridge University Press, Cambridge, United Kingdom. Ashkenazi, M. and Chow, S.-N. (1988) Normal forms near critical points for differential equations and maps. IEEE Trans. Circuits and Sys., 35, 850–862. Baider, A. (1989) Unique normal forms for vector ﬁelds and Hamiltonians. J. Diff. Eqs. 78, 33–52. Baider, A. and Churchill, R.C. (1988) Uniqueness and non-uniqueness of normal forms for vector ﬁelds. Proc. R. Soc. Edinburgh Sec. A, 108, 27–33. Bajaj, A.K. (1986) Resonant parametric perturbations of the Hopf bifurcation. J. Math. Anal. Appl., 115, 214–224. Baker, G.L. and Gollub, J.P. (1990) Chaotic Dynamics, Cambridge University Press, Cambridge, United Kingdom. van der Beek, C.G.A. (1989) Normal forms and periodic solutions in the theory of nonlinear oscillations. Existence and asymptotic theory. Int. J. Nonlinear Mech., 24, 263– 279. van der Beek, C.G.A. and van der Burgh, A.H.P. (1987) On the periodic windinduced vibrations of an oscillator with two

degrees of freedom. Nieuw Arch. Wisk, 5, 207–225. Belitskii, G.R. (1978) Equivalence and normal forms of germs of smooth mappings. Russ. Math. Surv., 33, 107–177. Belitskii, G.R. (1979) Normal forms with respect to the ﬁltering action of a group. Russ. Math. Surv., 40, 3–46. Belitskii, G.R. (1986) The smooth equivalence of germs of vector ﬁelds with one zero or a pair of pure imaginary eigenvalues. Funct. Anal. Appl., 20, 253–259. Bensoussan, A. (1988) Perturbation Methods in Optimal Control, Wiley-Interscience, New York. Bibikov, Y.N. (1971) On the reducibility of a system of two different equations to normal form. J. Diff. Eqs., 7, 1438–1441. Birkhoff, G.D. (1927) Dynamical Systems, American Mathematics Society Publications, Vol. 9, Providence, Rhode Island. Birkhoff, G.D. (1935) Nouvelles Recherches sur les systémes dynamiques. Mem. Pont. Acad. Sci. Novi. Lynacaei, 1, 85–216. Birkhoff, G.D. (1975) Versal deformations of a singular point on the plane in the case of zero eigenvalues. Funct. Anal. Appl., 9, 144–145. Bogdanov, R.I. (1976) Reduction to orbital normal form of a vector ﬁeld on the plane. Funct. Anal. Appl., 10, 61–62. Bogdanov, R.I. (1981) Versal deformation of a singularity of a vector ﬁeld on the plane in the case of zero eigenvalues. Sel. Math. Sov., 1, 389–421. Braun, M. (1983) Differential Equations and Their Applications, Springer-Verlag, New York.

The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

320

Further Reading Bruno, A.D. (1971) Analytical form of differential equations, I. Trans. Moscow Math. Soc., 25, 132–198. Bruno, A.D. (1972a) Analytical forms of differential equations, II. Trans. Moscow Math. Soc., 25, 199–299. Bruno, A.D. (1972b) Local methods in nonlinear resonances. Arch. Math. (Brno), 4, 177– 178 (Russian). Bruno, A.D. (1973) Local invariants of differential equations. Math. Notes, 14, 844–848. Bruno, A.D. (1974) The normal form of differential equations with a small parameter. Math. Notes, 16, 832–836. Bruno, A.D. (1975) The normal form of real differential equations. Math. Notes, 18, 722–731. Bruno, A.D. (1976) The normal form and averaging methods. Sov. Math. Dokl., 17, 1268– 1273. Bruno, A.D. (1979) Normal forms in perturbation theory, in Proc. VIII Int. Conf. Nonlinear Oscil., 1, 177–182, Prague, Czechoslovakia. Bruno, A.D. (1989) Local Methods in Nonlinear Differential Equations, Springer-Verlag, New York. Caprino, S., Maffei, C., and Negrini, P. (1984) Hopf bifurcation with 1 : 1 resonance. Nonlinear Anal. Theor. Math. Appl., 8, 1011– 1032. Chen, K.T. (1963) Equivalence and decomposition of vector ﬁelds about an elementary critical point. Am. J. Math., 85, 693–722. Chen, K.T. (1965) Local diffeomorphisms: C 1 realizations of formal properties. Am. J. Math., 87, 140–157. Chow, S.-N. and Hale, J.K. (1982) Methods of Bifurcation Theory, Springer-Verlag, New York. Chow, S.-N. and Wang, D. (1986) Normal forms of bifurcating periodic orbits, in Multi-Parameter Bifurcation Theory (eds M. Golubitsky and J. Guckenheimer), 9–18. Coullet, P. and Spiegel, E.A. (1983) Amplitude equations for systems with competing instabilities. SIAM J. Appl. Math., 43, 774– 819. Chua, L.O. and Lin, P.M. (1975) ComputerAided Analysis of Electronic Circuits, Prentice-Hall, Englewood Cliffs, New Jersey. Cushman, R. (1984) Normal form for Hamiltonian vector ﬁelds with periodic ﬂows,

in Differential Geometric Methods in Mathematical Physics (ed. S. Sternberg), Reidel, Dordrecht, The Netherlands, pp. 125–144. Cushman, R., Deprit, A., and Mosak, R. (1983) Normal form and representation theory. J. Math. Phys., 24, 2103–2116. Cushman, R. and Sanders, J.A. (1986) Nilpotent normal forms and representation theory of sl(2,R), in Multi-Parameter Bifurcation Theory (eds M. Golubitsky and J. Guckenheimer), pp. 31–51. Cvitanovic, P. (1984) Universality in Chaos, Adam Hilger, New York. Deprit, A. (1969) Canonical transformations depending on a small parameter. Celestial Mech. Dyn. Astron., 1, 12–32. Deprit, A. (1982) Delaunay normalizations. Celestial Mech. Dyn. Astron., 26, 9–21. Deprit, A., Henrard, J., Price, F., and Rom, A. (1969) Birkhoff’s normalization. Celestial Mech. Dyn. Astron., 1, 225–251. Devaney, R.L. (1985) An Introduction to Chaotic Dynamical Systems, Benjamin/Cummings, Menlo Park, California. Dowell, E. (1989) A Modern Course in Aeroelasticity, Kluwer, Dordrecht, The Netherlands. Dowell, E.H. and Ilgamova, M. (1988) Studies in Nonlinear Aeroelasticity, Springer-Verlag, New York. Duistermaat, J.J. (1984) Bifurcation of periodic solutions near equilibrium points of Hamiltonian systems, in Bifurcation Theory and Applications, Lecture Notes in Mathematics, Springer-Verlag, New York, Vol. 1057, pp. 57–105. Elphick, C., Tirapegui, E., Brachet, M.E., Coullet, P., and Iooss, G. (1987) A simple global characterization for normal forms of singular vector ﬁelds. Physica D, 29, 95–127. Evan-Iwanowski, R.M. (1976) Resonance Oscillations in Mechanical Systems, Elsevier, New York. Feng, Z.C. and Sethna, P.R. (1990) Global bifurcation and chaos in parametrically forced systems with one-one resonance. Dyn. Stab. Sys., 5, 201–225. Franks, J.M. (1982) Homology and dynamical systems, in CBMS Regional Conference Series in Mathematics, American Mathematics Society, Providence, Rhode Island, 49. Galin, D.M. (1982) Versal deformations of linear Hamiltonian systems. Am. Math. Soc. Trans., 118, 1–12.

Further Reading Gamero, E., Freire, E., and Ponce, E. (1991) Normal forms for planar systems with nilpotent linear part. Int. Series Num. Math., 97, 123–127. Gaponov-Grekhov, A.V. and Rabinovich, M.I. (1991) Nonlinear Physics, Oscillations, Chaos, Structures, Springer-Verlag, New York. van Gils, S.A., Krupa, M., and Langford, W. (1990) Hopf bifurcation with nonsemisimple resonance. Nonlinearity, 3, 825–850. Giorgilli, A. and Galgani, L. (1978) Formal integrals for an autonomous Hamiltonian system near an equilibrium. Celestial Mech. Dyn. Astron., 17, 267–280. Golubitsky, M. and Langford, W.F. (1981) Classiﬁcation and unfoldings of degenerate Hopf bifurcations. J. Diff. Eqs., 41, 375– 415. Golubitsky, M., Stewart, I., and Schaeffer, D.G. (1988) Singularities and Groups in Bifurcation Theory, Vol. 2, Springer-Verlag, New York. Hagedorn, P. (1981) Non-Linear Oscillations, Clarendon, Oxford, United Kingdom. Haken, H. (1983) Advanced Synergetics, Springer-Verlag, New York. Hale, J.K. (1963) Oscillations in Nonlinear Systems, McGraw-Hill, New York. Hartman, P. (1963) On the local linearization of differential equations. Proc. Am. Math. Soc., 14, 568–573. Hassard, B.D., Kazarinoff, N.D., and Wan, Y.-H. (1980) Theory and Applications of the Hopf Bifurcation, Cambridge University Press, Cambridge, United Kingdom. Hassard, B. and Wan, Y.H. (1978) Bifurcation formulae derived from center manifold theory. J. Math. Anal. Appl., 63, 297– 312. Hayachi, C. (1964) Nonlinear Oscillations in Physical Systems, McGraw-Hill, New York. Hinch, E.J. (1991) Perturbation Methods, Cambridge University Press, Cambridge, United Kingdom. Hirsch, M.W. (1976) Differential Topology, Springer-Verlag, New York. Hirsch, M.W., Pugh, C.C., and Shub, M. (1977) Invariant manifolds, in Springer Lecture Notes in Mathematics, Springer-Verlag, New York, p. 583. Holmes, P.J. (1981) Center manifolds, normal forms, and bifurcation of vector ﬁelds with

application to coupling between periodic and steady motions. Physica D, 2, 449–481. Hsu, L. and Favretto, F. (1984) Recursive bifurcation formulae for normal form and center manifold theory. J. Math. Anal. Appl., 101, 562–574. Hubbard, J. and West, B. (1991) Differential Equations – A Dynamical Systems Approach Part I: Ordinary Differential Equations, Springer-Verlag, New York. Ibrahim, R.A. (1985) Parametric Random Vibration, Wiley-Interscience, New York. Iooss, G. and Joseph, D.D. (1980) Elementary Bifurcation and Stability Theory, SpringerVerlag, New York. Jackson, E.A. (1989) Perspectives of Nonlinear Dynamics, Vols. I and II, Cambridge University Press, Cambridge, United Kingdom. Jezequel, L. and Lamarque, C.H. (1994) Analysis of non-linear dynamical systems by the normal form theory. J. Sound Vib., 149, 429–459. Jordan, D.W. and Smith, P. (1977) Nonlinear Ordinary Differential Equations, Oxford University Press, Oxford, United Kingdom. Kahn, P.B. (1990) Mathematical Methods for Scientists and Engineers: Linear and Nonlinear Systems, Wiley-Interscience, New York. Kevorkian, J. and Cole, J.D. (1981) Perturbation Methods in Applied Mechanics, SpringerVerlag, New York. Kim, J.H. and Stringer, J. (1992) Applied Chaos, Wiley-Interscience, New York. Kocak, H. (1984) Normal forms and versal deformations of linear Hamiltonian systems. J. Diff. Eqs., 51, 359–407. Kubicek, M. and Hlavacek, V. (1983) Numerical Solution of Nonlinear Boundary Value Problems with Applications, Prentice-Hall, Englewood Cliffs, New Jersey. Kubicek, M. and Marek, M. (1983) Computational Methods in Bifurcation Theory and Dissipative Structures, Springer-Verlag, New York. Kukulin, V.L., Krasnopolsky, V.M., and Horacek, J. (1989) Theory of Resonances: Principles and Applications, Kluwer, Dordrecht, The Netherlands. Kummer, M. (1971) How to avoid “secular” terms in classical and quantum mechanics. Nuovo Cimento B, 123–148.

321

322

Further Reading Lam, L. and Morris, H.C. (1990) Nonlinear Structures in Physical Systems Pattern Formulations, Chaos and Waves, SpringerVerlag, New York. Langford, W.F. (1979) Periodic and steadystate mode interactions leading to tori. SIAM J. Appl. Math., 37, 22–48. Lictenberg, A.J. and Lieberman, M.A. (1983) Regular and Stochastic Motion, SpringerVerlag, New York. Marchenko, V.A. (1987) Nonlinear Equations and Operator Algebras, Kluwer, Dordrecht, The Netherlands. Marek, M. and Schreiber, I. (1991) Chaotic Behavior of Deterministic Dissipative Systems, Cambridge University Press, Cambridge, United Kingdom. van der Meer, J.C. (1982) Nonsemisimple 1 : 1 resonance at an equilibrium. Celestial Mech. Dyn. Astron., 27, 131–149. van der Meer, J.C. (1986) The Hamiltonian Hopf Bifurcation, Springer-Verlag, New York. Meyer, K.R. (1974) Normal forms for Hamiltonian systems. Celestial Mech. Dyn. Astron., 9, 517–522. Meyer, K.R. and Schmidt, D.S. (1977) Entrainment domans. Funkcialaj Ekvacioj, 20, 171– 192. Mikhailov, A.S. (1994) Foundations of Synergetics I Distributed Active Systems, SpringerVerlag, New York. Minorsky, N. (1947) Introduction to NonLinear Mechanics, J.W. Edwards, Ann Arbor, Michigan. Moon, F.C. (1987) Chaotic Vibrations: An Introduction for Applied Scientists and Engineers, Wiley-Interscience, New York. Moon, F.C. (1992) Chaotic and Fractal Dynamics, Wiley-Interscience, New York. Morse, M. and Hedlund, G.A. (1938) Symbolic dynamics. Am. J. Math., 60, 815–866. Moser, J. (1973) Stable and Random Motion in Dynamical Systems, Hermann Weyl Lectures, Princeton, New Jersey. Murdock, J.A. (1991) Perturbations Theory and Methods, Wiley-Interscience, New York. Murdock, J. and Robinson, C. (1980) Qualitative dynamics from asymptotic expansions: Local theory. J. Diff. Eqs., 36, 425–441. Namachchivaya, N.S. (1989) Non-linear dynamics of supported pipe conveying pulsating ﬂuid-I. Subharmonic resonance. Int. J. Nonlinear Mech., 24, 185–196.

Namachchivaya, N.S. and Ariaratnam, S.T. (1986) Periodically perturbed non-linear gyroscopic systems, J. Struct. Mech., 14, 153–175. Namachchivaya, N.S. and Ariaratnam, S.T. (1987) Periodically perturbed Hopf bifurcation. SIAM J. Appl. Math., 47, 15–39. Namachchivaya, N.S. and Van Roessel, H.J. (1986) Unfolding of degenerate Hopf bifurcation for supersonic ﬂow past a pitching wedge. J. Guid. Cont. Dyn., 9, 413–418. Nayfeh, A.H. and Chin, C.-M. (1999a) Perturbation Methods with Mathematica, Dynamics Press, Virginia. http://www.esm.vt.edu/~anayfeh/ Nayfeh, A.H. and Chin, C.-M. (1999b) Perturbation Methods with Maple, Dynamics Press, Virginia. http://www.esm.vt.edu/~anayfeh/ Nitecki, Z. (1971) Differentiable Dynamics, MIT Press, Cambridge, Massachusetts. Palis, J. and de Melo, W. (1982) Geometric Theory of Dynamical Systems: An Introduction, Springer-Verlag, New York. Parker, T.S. and Chua, L.O. (1989) Practical Numerical Algorithms for Chaotic Systems, Springer-Verlag, New York. Peixoto, M. (1973) Dynamical Systems, Academic Press, New York. Perko, L.M. (1991) Differential Equations and Dynamical Systems, Springer-Verlag, New York. Poincaré, H. (1892) Les Méthodes Nouvelles de la Mécanique Céleste, Vol. 1, Gauthier-Villars, Paris. Poincaré, H. (1929) Sur les propriétés des functions déﬁnies par les équations aux différences partielles, in Oeuvres, GauthierVillars, Paris. Poston, T. and Stewart, I. (1978) Catastrophe Theory and Its Applications, Pitman, Aulander, North Carolina. Rabinowitz, M.I. (1987) Periodic Solutions of Hamiltonian Systems and Related Topics, Kluwer, Dordrecht, The Netherlands. Rabinowitz, M.I. and Trubetskov, D.I. (1989) Oscillations and Waves in Linear and Nonlinear Systems, Kluwer, Dordrecht, The Netherlands. Rand, R.H. (1984) Computer Algebra in Applied Mathematics: An Introduction to Macsyma, Research Notes in Mathematics, Pitman, Aulander, North Carolina.

Further Reading Rand, R.H. and Armbruster, D. (1987) Perturbation Methods, Bifurcation Theory, and Computer Algebra, Springer-Verlag, New York. Rasband, S.N. (1990) Chaotic Dynamics of Nonlinear Systems, Wiley-Interscience, New York. Rössler, O.E. (1976b) Different types of chaos in two simple differential equations. Z. Naturforsch., 31a, 1664–1670. Ruelle, D. (1989) Elements of Differential Dynamics and Bifurcation Theory, Academic Press, New York. Samovol, V.S. (1972) Linearization of a system of differential equations in the neighborhood of a singular point. Sov. Math. Dokl., 13, 1255–1259. Samovol, V.S. (1979) Linearization of systems of differential equations in a neighborhood of a singular point. Trans. Moscow Math. Soc., 38, 190–225. Samovol, V.S. (1982) Equivalence of systems of differential equations in a neighborhood of a singular point. Trans. Moscow Math. Soc., 44, 217–237. Sanders, J. and Verhulst, F. (1985) Averaging Methods in Nonlinear Dynamical Systems, Springer-Verlag, New York. Schmidt, G. and Tondl, A. (1986) Non-Linear Vibrations, Cambridge University Press, Cambridge, United Kingdom. Schuster, H. (1984) Deterministic Chaos, Physik Velag, Weinheim, Germany. Sell, G.R. (1971) Topological Dynamics and Differential Equations, Van Nostrand Reinhold, London, United Kingdom. Sell, G.R. (1978) The structure of a ﬂow in the vicinity of an almost periodic motion. J. Diff. Eqs., 27, 359–393. Sell, G.R. (1979) Bifurcation of higher dimensional tori. Arch. Rat. Mech. Anal., 69, 199– 230. Seydel, R. (1988) From Equilibrium to Chaos – Practical Bifurcation and Stability Analysis, Elsevier, New York. Shaw, S.W. and Pierre, C. (1992) On nonlinear normal mode, in Nonlinear Vibrations (eds R.A. Ibrahim, N.S. Namachchivaya, and A.K. Bajaj), ASME DE, Vol. 50, pp. 1– 5. Shub, M. (1987) Global Stability of Dynamical Systems, Springer-Verlag, New York. Siegel, C.L. (1952) Über die Normalform analytischer Differentialgleichungen in der

Nähe einer Gleichgewichtslösung. Nachr. Akad. Wiss. Goettingen Math. Phys. Kl. Math. Phys. Chem. Abt, 21–30. Sparrow, C. (1982) The Lorenz Equations: Bifurcations, Chaos, and Strange Attractors, Springer-Verlag, New York. Starzhinskii, V.M. (1972) Normal forms of the fourth-order for nonlinear oscillations. Mech. Solids, 7, 1–7. Starzhinskii, V.M. (1973) A practical method for computing normal forms in nonlinear oscillation problems. J. Appl. Math. Mech., 37, 389–396. Starzhinskii, V.M. (1974) On oscillations in third-order systems when the linear approximation is neutral. Mech. Solids, 9, 18– 24. Starzhinskii, V.M. (1980) Applied Methods in the Theory of Nonlinear Oscillations, MIR Publishers, Moscow, Russia. Stauffer, D. and Stanley, H.E. (1990) From Newton to Mandelbrot – A Primer in Modern Theoretical Physics, Springer-Verlag, New York. Sternberg, S. (1958) On the structure of local homeomorphisms of Euclidean n-space. Am. J. Math., 80, 623–631. Sternberg, S. (1959) The structure of local homeomorphisms. Am. J. Math., 81, 578– 604. Sternberg, S. (1961) Finite Lie groups and the formal aspects of dynamical systems. J. Appl. Math. Mech., 10, 451–478. Szemplinska-Stupnicka, W. (1990) The Behavior of Nonlinear Vibrating Systems: Vol. I: Fundamental Concepts and Methods; Applications to Single Degree-of-Freedom Systems, Kluwer, Dordrecht, The Netherlands. Szemplinska-Stupnicka, W. (1990) The Behavior of Nonlinear Vibrating Systems: Vol. II: Advanced Concepts and Applications to Multi-Degree-of-Freedom Systems, Kluwer, Dordrecht, The Netherlands. Tabor, M. (1989) Chaos and Integrability in Nonlinear Dynamics: An Introduction, Wiley-Interscience, New York. Takens, F. (1974) Singularities of vector ﬁelds. Publ. Math. I.H.E.S., 43, 47–100. Takens, F. (1979) Forced oscillations and bifurcations. Comm. Math. Inst., 3, 1–59. Thompson, J.M.T. and Stewart, H.B. (1986) Nonlinear Dynamics and Chaos: Geometrical Methods for Engineers and Scientists, WileyInterscience, New York.

323

324

Further Reading Tokarev, S.P. (1977) Smooth equivalence of differential equation systems in the plane in the case of a noncoarse focus. J. Diff. Eqs., 613–617. Troger, H. and Steindl, A. (1991) Nonlinear Stability and Bifurcation Theory, SpringerVerlag, New York. Urabe, U. (1967) Nonlinear Autonomous Oscillations, Academic Press, New York. Ushiki, S. (1984) Normal forms for singularities of vector ﬁelds. Japan J. Appl. Math., 1, 1–37.

Verhulst, F. (1990) Nonlinear Differential Equations and Dynamical Systems, SpringerVerlag, New York. Weinstein, A. (1973) Normal modes for nonlinear Hamiltonian system. Inv. Math., 20, 47–57. Yoshizawa, T. (1975) Stability Theory and Existence of Periodic Solutions and Almost Periodic Solutions, Springer-Verlag, New York.

325

Index a activation function 287 adjoint 46, 273 Airy equation 5 amplitude, unique deﬁnition of 20 Andronov 115 aperiodic 87 attractive subsystem 292 attractor 101 autoparametric resonance 219, 220, see also combination internal resonance, one-to-one internal resonance, three-to-one internal resonance, two-to-one internal resonance – simultaneous 194, 220, 223 averaging, method of – generalized 6, 26, 276 b Banach space 287 basis, natural 36 Bessel’s equation 1 bifurcation 76, 78, 107 – degenerate 115 – diagram 78 – nondegenerate 82 Blasius problem 3 bookkeeping device 12 boundary layer 3 bracket, Lie 6, 33, 41, 45 Burger’s equation 3 c canonical form see Jordan form Cartesian 44 center 102 center manifold reduction 85, 142 – for continous systems 103–107 – for Hopf bifurcation 141–144 – for maps 72–75

– for retarded 3-D systems 308–310 – for retarded scalar equation 291–295 – for retarded SDoF equation 299–304 – for static bifurcation 126–132 center-manifold theorem 73, 103, 108 characteristics 3 circle map 87 codimension 33 combination autoparametric resonance 219 – in systems with quadratic nonlinearities 222 combination parametric resonance 194, 207 – in linear gyroscopic systems 208 – in linear nongyroscopic systems 191–192 – in systems with repeated frequencies 198, 199 combination resonance 228, 236, 253 complementary subspace 36, 38, 46 constant solution see also ﬁxed point control, speed 151 – time-delayed feedback 287 crane 311 critical point see ﬁxed point cutting force 295 cycle see limit cycle d degeneracy 33 degenerate – bifurcation 150 – form 199 detuning parameter 162 diagonal matrix 191 distinguished limit 199 divisor see small divisor dominant 4 Dufﬁng equation 9, 15 – forced oscillations of 161–172

The Method of Normal Forms, Second Edition. Ali Hasan Nayfeh © 2011 WILEY-VCH Verlag GmbH & Co. KGaA. Published 2011 by WILEY-VCH Verlag GmbH & Co. KGaA

326

Index – free oscillation 9–13 e eigenvalues – distinct 62 – purely imaginary 42, 52 – repeated 64 – zero 32, 48, 54 – zero and two purely imaginary 52 eigenvector 61 – generalized 61, 62, 65, 66, 97, 99 – of matrix 32 equilibrium solution see ﬁxed point evolution equation 291 excitation see also multiplicative, parametric – harmonic 161 – multifrequency 257 f fast variation 11 ﬁxed point 6, 31, 66, 93, 97, 100 – hyperbolic 32, 72, 101, 107 – neutrally stable 102 – nonhyperbolic 32, 72, 101, 107 – nonstable 102 ﬂip bifurcation see period-doubling bifurcation ﬂutter 115, 196 focus 101 fold 76, see also saddle-node bifurcation form see Jordan form function space 291 fundamental parametric resonance 188, 245 – in Mathieu equation 187 – in nonlinear SDoF systems 208 – in systems with repeated frequencies 245–249 g galloping 115 generalized eigenvector 32, 61, 62, 65 generating vector 6 generic bifurcation 109, 117 gyroscopic systems – externally excited 225–228, 249–255 – parametrically excited 205–208 h harmonic excitation 161 Hartman–Grobman theorem 72, 102 heat transfer equation 3 Hénon map 93 heteroclinic connection 150, 151 homeomorphism 103

homology equations 21, 67 hoop 152 Hopf bifurcation 2, 108, 115, 116 – conditions for 76, 115 – in maps 85–88 – in retarded systems 287–312 – normal form of 2, 115–117, 137–146 hyperbolic ﬁxed point 32, 72, 108 i image 36 incompressible ﬂow 3 inevitable resonance 71 inﬂection point 119 inner product 291 internal resonance 220, see also combination autoparametric resonance, one-to-one autoparametric resonance, three-to-one autoparametric resonance, two-to-one autoparametric resonance, see also autoparametric resonance – absence of 232, 236, 250, 262 – in gyroscopic systems 225–228, 249–255 – in systems with cubic nonlinearities 238–243 – in systems with quadratic and cubic nonlinearities 263–277 – in systems with quadratic nonlinearities 220–225 – in systems with repeated frequencies 238–243, 279–285 invariant – circle 87 – set 66, 100 – time 11, 14 inverted pendulum 151 irrational number 87 j Jacobian 33, 101 – matrix 77, 86, 101, 102 Jordan canonical form 32, 98 k Krylov–Bogoliubov–Mitropolsky technique 170 l Lagrangian 263–265, 267, 270, 274 lathe 295–304 Lie 6, 33, 41 – bracket 41, 45 – transform 6

Index limit cycle 6 linear normal form 35 Liouville equation 4 logistic map 83 long-period terms 6 Lorenz equations 158 lunar orbital dynamics 115 m machine tool 295–304 main, resonance see primary manifold 73, see also center manifold reduction maps 61–95 Mathieu Equation 54, 187 – treated by method of multiple scales 56–57 – treated by method of normal forms 54–56, 187–188 modal-damping 191 monomials 41 multifrequency excitation 257 multiple scales, method of – applied to 2-DoF systems with quadratic nonlinearities 267–276 – applied to Dufﬁng equation 12 – applied to Hopf bifurcation 138–141 – applied to Mathieu equations 56 – applied to Rayleigh equation 15 – applied to retarded 3-DoF systems 306–308 – applied to retarded scalar equation 289–290 – applied to retarded SDoF systems 296–299 – applied to static bifurcation 117–126 – applied to systems with purely imaginary eigenvalues 42, 47 – applied to systems with quadratic and cubic nonlinearities 17 – applied to systems with repeated frequencies 283–285 – applied to time-delay crane control 311–312 – applied to van der Pol equation 19 multiplicity 32, 39 n natural basis 36 near-identity transformation 8 near-resonance 21, 35, 67–70 Neimark–Sacker bifurcation 76, 85 neuron 304 node 101

nonbifurcation 108, 110 nongyroscopic 191, 217 nonhyperbolic ﬁxed point 32, 72, 101 nonsemisimple 196, 240, 244, 279 nonsingular 21 – matrix 31 – transformation 4 nonstable ﬁxed point 102 nonuniform expansion 10 normalization 8, 13, 31 NST 48 null – matrix 33 – space 46, 49 o one-to-one autoparametric resonance 251 – in systems with cubic nonlinearities 239 – in systems with quadratic and cubic nonlinearities 264, 279 – in systems with repeated frequencies 195–205, 240–249, 279–285 orbit – of maps 66 – period-two 83 – quasiperiodic 88 ordering scheme 12 orthogonal 46 p parametric excitation see also Mathieu equation, 187–216 – in gyroscopic systems 205–208 – in single-degree-of-freedom nonlinear systems 208–212 – in systems with distinct frequencies 194 – in systems with repeated frequencies 196–205 – multifrequency 257 parametric resonance – simultaneous 194, see combination parametric resonance, fundamental parametric resonance, principal parametric resonance partial differential equation 1–3 pendulum 29, 151 period-doubling bifurcation 81, 85 pitchfork bifurcation 80, 108, 114 – in continuous systems 113–114 – in maps 80 Poincaré 115 Poisson bracket 33, 41

327

328

Index polar 44 – coordinates 44 – form 11, 19, 24 potential 148 primary resonance 161 – in 2-DOF systems with quadratic nonlinearities 176 – in Dufﬁng equation 161 – in gyroscopic systems 226, 250 – in systems with cubic nonlinearities 239 – in systems with quadratic and cubic nonlinearities 176, 223 principal parametric resonance 188, 190, 244 – in 2-DoF systems 244 – in gyroscopic systems 208 – in Mathieu equation 190 – in nongyroscopic systems 194, 197 – in nonlinear SDoF systems 210 – in systems with repeated frequencies 197 projection method – for Hopf bifurcation 144 – for static bifurcation 132 purely imaginary eigenvalues 7, 33, 42, 52, 107, 115 q quasiperiodic orbit 88, 90 r range 46 rational number 87, 90 Rayleigh equation 13, 15 reconstitution, method of 27, 213, 269 regenerative model 295 regular 22 repeated frequency 195 repellor 102 resonance, deﬁnition of – terms see also autoparametric resonance, combination resonance, fundamental parametric resonance, principal parametric resonance, subharmonic resonance, superharmonic resonance resonance, deﬁnition of 10 – condition for continous systems 37 – condition for maps 68, 69 – conditions of continous systems 35 – inevitable 71 – strong 71 – terms 10, 21, 33, 35, 42, 67–69 retarded systems 287–313

reverse pitchfork bifurcation 80, 114 Reynolds number 3 Rössler equation 157 rotating – circular hoop 152 – particle 29 – phase 6 s saddle 101, 102 saddle-node bifurcation 76, 77, see also static bifurcation, 108, 109 – in continuous systems 108–109 – in maps 76–78 secondary resonance see combination resonance, subharmonic resonance, superharmonic resonance secular terms 10, 20, 47 semisimple 39 sensitivity to initial conditions 88 set, invariant 66, 100 shift operator 291 Shoshitaishvili theorem 103 similarity transformation 3 simultaneous resonance 194, 220, 223 singular 68 – point 100 sink 101 slow variation 11 small divisor 21, 35, 68 solvability conditions 118, 273 source 101, 102 spindle 295 spiral 87, 90 stability – interchange of 79 – local 101 state-space form 271 static bifurcation see pitchfork bifurcation, saddle– node bifurcation, transcritical bifurcation, see ﬁxed point – normal form of 117–137 stationary solution see ﬁxed point straightforward expansion 10, 43 stream function 3 strong resonance 71, 86 subcritical 117 subcritical bifurcation 85, 114, 116 – Hopf bifurcation 116, 117 – period-doubling bifurcation 83–84 – pitchfork bifurcation 83, 85, 114, 120 subharmonic resonance 161, 164, 178 – in Dufﬁng equation 164–167, 171 – in gyroscopic systems 228, 253

Index – in nonlinear SDoF systems 178–180, 182–185 – in systems with quadratic nonlinearities 220 subspace 36, 38, 41 successive transformations 17, 66 supercritical bifurcation 114 – Hopf bifurcation 116, 117 – period-doubling bifurcation 83–84 – pitchfork bifurcation 80, 114, 120 superharmonic resonance 161, 180 – in Dufﬁng equation 167, 172 – in gyroscopic systems 253 – in MDoF systems 220, 228 – in SDoF systems 180–184 t tangent bifurcation 76, see also saddle-node bifurcation three-to-one autoparametric resonance 236 – in gyroscopic systems 251, 255 – in systems with cubic nonlinearities 238, 251 – in systems with quadratic and cubic nonlinearities 263

time delay 287–313 time invariant 11 transcritical bifurcation 79, 108, 112 – in continuous systems 111 – in maps 79 transversality condition 115 troublesome terms 21 turning point 5 two-to-one autoparametric resonance 219 – in gyroscopic systems 225–228 – in systems with quadratic and cubic nonlinearities 266–276 – in systems with quadratic nonlinearities 220, 226–228, 231 – simultaneous 223 – treated with generalized method of averaging 276–278 v van der Pol equation 15, 24, 152 variation of parameters 5, 8 w wanders 88

329

Our partners will collect data and use cookies for ad personalization and measurement. Learn how we and our ad partner Google, collect and use data. Agree & close