Documente Academic
Documente Profesional
Documente Cultură
Series Editors
Vol. 51: Besdo, D., Heimann, B., Klüppel, M., Vol. 39: Lucchesi, M., Padovani, C., Pasquinelli, G., Zani, N.
Kröger, M., Wriggers, P., Nackenhorst, U. Masonry Constructions: Mechanical
Elastomere Friction Models and Numerical Applications
249 p. 2010 [978-3-642-10656-9] 176 p. 2008 [978-3-540-79110-2]
Valery N. Pilipchuk
123
Valery N. Pilipchuk
Professor (Research)
Mechanical Engineering
Wayne State University
5050 Anthony Wayne Dr., 2140
Detroit, Michigan 48202
USA
E-mail: pilipchuk@wayne.edu
e-ISSN 1860-0816
This work is subject to copyright. All rights are reserved, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilm or in any other ways, and storage in data banks.
Duplication of this publication or parts thereof is permitted only under the provisions of the German
Copyright Law of September 9, 1965, in its current version, and permission for use must always be
obtained from Springer. Violations are liable for prosecution under the German Copyright Law.
The use of general descriptive names, registered names, trademarks, etc. in this publication does not
imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
Typeset & Cover Design: Scientific Publishing Services Pvt. Ltd., Chennai, India.
9876543210
springer.com
Preface
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Brief Literature Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Asymptotic Meaning of the Approach . . . . . . . . . . . . . . . . . . . . 4
1.2.1 Two Simple Limits of Lyapunov Oscillator . . . . . . . . . . 4
1.2.2 Oscillating Time and Hyperbolic Numbers,
Standard and Idempotent Basis . . . . . . . . . . . . . . . . . . . 6
1.3 Quick ‘Tutorial’ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3.1 Remarks on the Basic Functions . . . . . . . . . . . . . . . . . . . 9
1.3.2 Viscous Dynamics under the Sawtooth Forcing . . . . . . 9
1.3.3 The Rectangular Cosine Input . . . . . . . . . . . . . . . . . . . . 11
1.3.4 Oscillatory Pipe Flow Model . . . . . . . . . . . . . . . . . . . . . . 12
1.3.5 Periodic Impulsive Loading . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.6 Strongly Nonlinear Oscillator . . . . . . . . . . . . . . . . . . . . . 15
1.4 Geometrical Views on Nonlinearity . . . . . . . . . . . . . . . . . . . . . . 17
1.4.1 Geometrical Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.4.2 Nonlinear Equations and Nonlinear Phenomena . . . . . 19
1.4.3 Rigid-Body Motions and Linear Systems . . . . . . . . . . . 21
1.4.4 Remarks on the Multi-dimensional Case . . . . . . . . . . . . 23
1.4.5 Elementary Nonlinearities . . . . . . . . . . . . . . . . . . . . . . . . 24
1.4.6 Example of Simplification in Nonsmooth Limit . . . . . . 25
1.4.7 Non-smooth Time Arguments . . . . . . . . . . . . . . . . . . . . . 26
1.4.8 Further Examples and Discussion . . . . . . . . . . . . . . . . . . 28
1.4.9 Differential Equations of Motion and
Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.5 Non-smooth Coordinate Transformations . . . . . . . . . . . . . . . . . 33
1.5.1 Caratheodory Substitution . . . . . . . . . . . . . . . . . . . . . . . 33
1.5.2 Transformation of Positional Variables . . . . . . . . . . . . . 33
1.5.3 Transformation of State Variables . . . . . . . . . . . . . . . . . 35
VIII Contents
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
Appendices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
Chapter 1
Introduction
describe essentially nonlinear phenomena out of the scope of the weakly non-
linear methods.
So the key notion of the present work suggests possible recipes for selecting
such models among so-called non-smooth systems while keeping the class of
smooth motions still within the range of applicability.
Note that different non-smooth cases have been also considered for several
decades by practical and theoretical reasons [81], [39], [17], [42], [127], [72],
[196], [82], [204], [172], [45], [65], [204], [169], [29], [14], [93], [30], [180], [51],
[206], [94], [44], [173], [25], [162], [55], [185], [85], [130], [129], [128], [1], [68],
[69], [37]. On the physical point of view, this kind of modeling essentially em-
ploys the idea of perfect spatio-temporal localization of strong nonlinearities
or impulsive loadings. For instance, sudden jumps of restoring force charac-
teristics are represented by absolutely stiff constraints under the assumption
that the dynamics in between the constraints is smooth and simple enough to
describe. As a result, the system dynamics is discretized in terms of mappings
and matchings different pieces of solutions.
The present work however is rather close to another group of methods deal-
ing with the differential equations of motion on entire time intervals despite
of discontinuity and/or nonsmoothness points. Such methods are developed
to satisfy the matching conditions automatically by means of specific coordi-
nate transformations on preliminary stages of study. To some extent, these
can be qualified as adaptations of the differential equations of motion for fur-
ther studies by another methods. Among such kind of transformations, the
Caratheodory substitution [45] can be mentioned first. This linear substitu-
tion, which includes the unit-step Heaviside function, eliminates Dirac delta-
functions participating as summands in differential equations. Much later,
non-smooth nonlinear coordinate transformations were suggested in [199],
[204] for the class of impact systems. This strongly nonlinear transforma-
tion effectively eliminates stiff barriers by unfolding the system configuration
space in a mirror-wise manner with respect to the barrier planes. A similar
idea was implemented later regarding the system phase space [72] in order
to resolve certain problems related to non-elastic impacts. Some technical
details and discussions on these methods are included below in Section 5 for
comparison reason.
In contrast, the present approach employs time histories of impact systems
as new time arguments. Originally such a nonsmooth temporal transforma-
tion (NSTT), or a sawtooth oscillating time1 , was introduced for strongly
nonlinear but smooth periodic motions with certain temporal symmetries
[132]. Then, it was shown that such an approach still works for general
cases by generating specific algebraic structures in terms of the coordinates
[134], [144]. The occurrence of such algebraic structures seems to be essential
1
Here and further, the term ‘sawtooth’ is used for periodic piecewise-linear func-
tions regardless symmetries of their teeth; in most cases, it is the triangular sine
wave.
1.1 Brief Literature Overview 3
.. ..
x x2 n1 0 x x2 n1 0
n1 n
1 1 Τt
sin t
t t
Π 2Π 2 4
cos t et
1 1
Fig. 1.1 Two asymptotic limits described by elementary functions associated with
subgroups of rigid-body motions.
Fig. 1.2 Complex and hyperbolic planes are shown on the left and right, respec-
tively; in contrast to the circle, each of the hyperbola branches is covered exactly
once as the hyperbolic angle is varying in the infinite interval.
x = X (τ ) + Y (τ ) τ̇ (1.1)
On the right-hand side, the functions X and Y are easily expressed through
the original function x(t), if this function is known; see Chapter 4 for further
details.
In case x(t) is an unknown periodic motion of some dynamical system,
equations for X and Y components are obtained by substituting (1.1) into
the corresponding differential equation of motion. Then either analytical or
numerical procedures can be applied. For instance, one may seek solutions
in the form of power series with respect to the ‘oscillating time’ τ . Therefore
expression (1.1) can be qualified as non-smooth time transformation, t → τ ,
on the manifold of periodic motions.
Note that the structure of hyperbolic numbers has been known for quite
a long time mostly as a formal extension of the regular complex numbers
with no relation to either vibrating systems or nonsmooth functions; see the
references below. In the mathematical literature, such an extension (which
is often regarded to as a simple particular case of Clifford’s algebras5) is
introduced as follows. Whereas the algebraic equation p2 = 1 has the real
number solutions p = ±1, the existence of a unipotent u is assumed such
that u = +1 and u = −1 but u2 = 1. Then by considering the elements
{1, u} as a standard basis, any hyperbolic number w ∈ H can be written in
the form w = x + uy, where x and y are real numbers [176]. √ The hyperbolic
conjugate of w is defined by w− = x−uy so that |w|H = ww− is the norm of
w. Fig. 1.2 illustrates the difference between complex and hyperbolic planes.
More details and some, rather abstract, applications of this algebraic theory
can be found in references [90], [176], [7], [187]. As to applied areas, the same
kind of algebraic structures occurred in hydrodynamics in connection with
characteristics of partial differential equations [90].
In our case, the unipotent is not a number but the discontinuous function
of certain physical nature i.e. the rectangular cosine wave e(t). Indeed since
t is running then there is no unique choice for the magnitude of e, whereas
always6 e2 = 1. Therefore, identity (1.1) generates the hyperbolic structure
from the very general properties of periodic processes.
Note that both terms on the right-hand side of (1.1) are essential as those
responsible for components with different temporal symmetries. For instance,
5
William Kingdon Clifford (1845-1879), English mathematician who, in particular,
developed the idea that space may not be independent of time.
6
More precisely, for almost any t.
8 1 Introduction
Fig. 1.3 Geometrical interpretation of the particular case with a sine-wave tem-
poral symmetry: observing the coordinate x does not allow to conclude which of
the two temporal variables, τ or t, is actually ‘running.’
x = X + Y e = X(e+ + e− ) + Y (e+ − e− )
= (X + Y )e+ + (X − Y )e−
≡ X+ (τ )e+ + X− (τ )e−
τ̇ = e
∞
ė = 2 [δ (t + 1 − 4k) − δ (t − 1 − 4k)] (1.4)
k=−∞
e2 = 1
where x(0) is an arbitrary initial position, and a part of the problem, which
is calculating the integral, still persists.
Now let us represent the particular (periodic) solution in the form
x = X(τ ) + Y (τ )e (1.7)
Y (τ ) − qτ + X (τ )e + Y (τ )ė = 0 (1.8)
Y (τ ) = qτ , X (τ ) = 0, Y (±1) = 0 (1.9)
where the boundary condition provides zero factor for all δ−functions of the
derivative ė.
In contrast to (1.5), the integration of equation (1.9) is strightforward since
the variable of integration is τ . So substituting obvious solution of problem
(1.9) into (1.7), gives general solution of equation (1.5)
q
x = C + (τ 2 − 1)e (1.10)
2
where C is an arbitrary constant.
Then taking into account that τ (0) = 0 and e(0) = 1, gives x(0) = C − q/2
and therefore solution (1.10) takes the form
q q
x = x(0) + + (τ 2 − 1)e (1.11)
2 2
As compared to the direct approach (1.6), the NSTT allowed for conducting
the integration of the differential equation of motion with no dealing with the
piece-wise structure of the integrand. Moreover, comparing solutions (1.6)
and (1.11), gives the result of direct integration (1.6) in the form
t
1
τ (ϕ)dϕ = {1 + [τ 2 (t) − 1]e(t)} (1.12)
0 2
1.3 Quick ‘Tutorial’ 11
n
ẋ(t) = qk τ 2k−1 (t) (1.13)
k=1
Y + (X − p)e + Y ė = 0
Y (τ ) = 0, X (τ ) = p, Y (±1) = 0 (1.16)
In this simple case, the above solution could be written though based on the
definition of e(t) (1.3). As a generalization of the right-hand side of equation
(1.15), let us consider equation
m
ẋ(t) = pi τ i (t) e(t) (1.18)
i=0
does exist for both odd and even exponents of the polynomial in (1.18).
Finally, combining the right-hand sides of equations (1.13) and (1.18) and
considering the corresponding infinite series, gives
∞
∞
f (t) = qk τ 2k−1
(t) + pi τ (t) e(t) ≡ Q(τ ) + P (τ )e
i
(1.20)
k=1 i=0
Y + ak(X 2 + Y 2 ) = ap0
X + 2akXY = ap1 (1.25)
Y (±1) = 0
U + akU 2 = aF
V − akV 2 = −aG (1.26)
U (±1) − V (±1) = 0
U (1, C1 ) − V (1, C2 ) = 0
U (−1, C1 ) − V (−1, C2 ) = 0 (1.28)
10 T8
9 T4
Q 8
Fig. 1.4 shows what happens to the steady state flow profile as the period
of pressure wave becomes twice longer. The model parameters are k = 0.03,
p0 = 2.0, and p1 = 1.5. In cases T = 4 and T = 8, the arbitrary constants are
C1 = 8.2248, C2 = −0.3125 and C1 = 11.7200, C2 = −0.2651, respectively.
Note that the average flow for the period T = 4 is < Q >= 8.12292, whereas
the longer period T = 8 gives somewhat smaller average, < Q >= 8.02996.
1.3 Quick ‘Tutorial’ 15
where λ and p are a coefficient of viscosity and a half of the pulse amplitude
per unit mass, respectively.
Equation (1.31) has a simple form involving the distribution. Such kind
of problem is usually solved by applying either the generalized Fourier series
or Laplace transform, or by considering the equation between the pulses by
matching different pieces of the solution at the pulse times. Alternatively, the
solution can be obtained in few quick steps by using identity (1.1).
Indeed, substituting (1.1) in (1.31) and taking into account (1.4), gives
(Y + λX) + (X + λY ) e + (Y − p)ė = 0
or
Y + λX = 0, X + λY = 0, Y |τ =±1 = p (1.32)
Solving boundary value problem (1.32), gives the closed form particular
solution
p
v=− (sinh λτ − e cosh λτ ) (1.33)
cosh λ
The entire general solution is obtained by adding the term with an arbitrary
constant, C exp(−λt), due to linearity of equation (1.31).
ẍ + x2n−1 = 0 (1.34)
Now the idea is to take advantage of the fact that the new temporal argument
is periodic and bounded, −1 ≤ τ ≤ 1. So, in order to design an analogous
of the quasi harmonic approaches, successive iterations can be applied. This
tool requires no small parameter to be present explicitly. As a result, one gen-
erally must rely on convergence of the procedure. Although the convergence
is usually slow7 , the physical basis of the convergence can always be deter-
mined whenever the problem has indeed some physical content. Note that the
harmonic balance method also uses no explicit small parameter by assuming
however that temporal mode shapes are close to harmonic. In other words,
high frequency terms just correct but not exceed the principal harmonic. In
the present case, solution is approximated by the triangular sine, which is
corrected by higher powers of the same triangular sine. On the physical point
of view, the model under consideration has to be close to the impact oscillator
rather then the harmonic one. In terms of the new time variable τ , such an
assumption simply means that the right-hand side of the differential equation
of motion (1.37) is small enough to justify the following generating system
X0 = 0 (1.38)
Indeed, this equation describes a family of impact oscillators with the trian-
gular sine wave time histories
X0 = Aτ (t/a) (1.39)
Fig. 1.5 Rigid-body rotation that generates both sine and triangular waves within
the same class of elementary functions.
Suppose now that α → 1 − 0. In this case, (1.44) takes the triangular sine
wave shape
π 2ϕ
y→ τ (1.45)
2 π
Interestingly enough, both types of observation of the above rigid-body model
are associated with different elastic oscillators. Namely, on one hand, as a
function of time, expression (1.43) satisfies equation
Ÿ + ω 2 Y = 0 (1.46)
On the other hand, function y(t) (1.44) satisfies the differential equation of
motion of conservative oscillator
tan y
ÿ + =0 (1.47)
cos2 y
under condition,
α2 + ω −2 = 1 (1.48)
In particular, condition (1.48) implies that the frequency ω depends upon
the disc radius monotonically in such a way that 1 ≤ ω < ∞ as 0 ≤ α <
1. Therefore, nonsmooth limit (1.45) is reached under the infinitely large
frequency ω, when the total energy of oscillator (1.47) also becomes infinitely
large,
ẏ 2 tan2 y
H= + →∞ (1.49)
2 2
The reasons for considering oscillator (1.47) as a generating model in non-
linear dynamics will be described later. At this stage, let us bring attention
to the fact that both equations (1.46) and (1.47) actually describe the same
rigid-body rotation but in different coordinate systems.
the familiar linear region and enters the more complex world of nonlinear
effects: rivers become turbulent, amplifiers overload and distort, chemicals
explode, machines go into uncontrollable oscillations, plates buckle, metals
fracture, and structures collapse.” The author goes even further by saying
that “within such an approach, mind may no longer appear as an alien stuff
in a mechanical universe; rather the operation of mind will have resonances
to the transformations of matter, and indeed, the two will be found to emerge
from a deeper ground.”
However, mathematical approaches can still be helpful based on the fol-
lowing logical identity
However, both sets of equations (1.52) and (1.53) describe the same mechan-
ical system in polar and Cartesian coordinates, respectively. On the plane of
configurations, the corresponding coordinate transformation is given by
x1 = ρ cos ϕ, x2 = ρ sin ϕ
In this example, it is clear that the system is linear from the physical point
of view; indeed nonlinearity of equations (1.52) is due to specific choice for
coordinates. Therefore a ‘physical definition’ for linear systems must specify
the type of reference coordinate system, for instance, as follows:
A mechanical system is linear if its differential equations of motions in
Cartesian coordinates are linear 8 .
This definition involves already some physical principles since the Cartesian
coordinates are uniquely associated with general properties of the physical
space. Note that the above definition still involves the mathematical notion
of coordinate systems. As known from observations however, some dynamic
phenomena are perceived as nonlinear with no explicit system coordinates.
So next two subsections make an attempt to clarify why such a perception
of nonlinearity is actually possible.
z z̄ = z z̄ (1.55)
8
This definition was suggested by V.Ph.Zhuravlev during a private discusion at
International Conference “Nonlinear Phenomena,” Moscow, 1989.
22 1 Introduction
where the over bar means complex conjugate, and the prime denotes any new
position of the body.
Expression (1.55) recalls mathematical objects generated by the Galilean
group of rotations. The corresponding operator G is introduced through re-
lationship
z = Gz (1.56)
where G must depend on some parameter, say ϕ, characterizing the transition
z → z.
Substituting (1.56) in (1.55), gives
2
GḠ = |G| = 1 =⇒ G = exp(iϕ), i2 = −1 (1.57)
Since ϕ is the only parameter of the only one ‘process’ then ϕ must be
qualified as time with possibly some scaling factor. Introducing an arbitrary
scaling factor, say ω, gives then
Therefore both operators, G and Ḡ, satisfy the differential equation of har-
monic oscillator associated with the operator
d d d2
− iω + iω = 2 + ω 2 (1.59)
dt dt dt
By representing the scaling factor in the form ω = k/m, one can interpret
then m and k as, respectively, mass and stiffness of a mass-spring model
associated with the ‘product’ of clockwise and counter-clockwise rotations of
the disc.
Let us consider now n rotating discs with the angular frequencies
{ω1 , ..., ωn } by generalizing product (1.59) as follows
n
2
d d d d d
− iω1 + iω1 ... − iωn + iωn = + ω 2
j
dt dt dt dt j=1
dt2
(1.60)
Operator (1.60) represents an arbitrary n-degrees-of-freedom linear elastic
system. Replacing d/dt → λ, gives the corresponding characteristic equation
in the form
n
2
λ + ωj2 = 0 (1.61)
j=1
1.4 Geometrical Views on Nonlinearity 23
πλ
sinh =0 (1.62)
a
as n → ∞.
Equation (1.62) associates with a linearly elastic string of the length l = π
with two fixed ends,
∂ 2 u (t, x) ∂ 2 u (t, x)
2
− a2 = 0; u (t, 0) = u (t, π) (1.63)
∂t ∂x2
Therefore, it is shown that basic linearly elastic dynamic models can be logi-
cally obtained from the kinematics of freely rotating discs or even from a more
fundamental concept given by expression (1.55). To some extent, this may
explain why nonlinearity is perceptible as a physical phenomenon. Indeed the
linear dynamics is simply matching with the usual perception (experience) of
the space. All other motions give therefore an impression of unusual, in other
words, nonlinear phenomena.
f (v) − f (w)
=
v − w
(1.64)
f (v)
=
v
=
−v
=
f (−v)
and
f (v) − f (−v)
=
v− (−v)
= 2
v
hold due to (1.64), and thus f (v) and f (−v) are antipodal: f (−v) = −f (v).
24 1 Introduction
Second, taking into account the latter result and definition (1.64), trans-
forms the polarization identity as follows
1 2 2 2
v·w =
v + w
−
v
−
w
2
1
=
f (v) − f (−w)
2 −
f (v)
2 −
f (−w)
2
2
1 2 2 2
=
f (v) + f (w)
−
f (v)
−
f (w)
2
= f (v) · f (w)
Therefore, the inner product is preserved. Now, if {ui }is orthogonal basis in
Rn , then {f (ui )} is another orthogonal basis. Taking the dot product of the
both sides of identity (1.65) with the arbitrary basis vector f (ui ), gives
or
(av + bw) · ui ≡ av · ui + bw · ui
The above relation proves identity (1.65) in terms of the coordinates associ-
ated with the basis {f (ui )}.
All linear isometries of Rn are denoted by the symbol O (n). By choosing
a basis for Rn , one can represent every element of O (n) as a matrix. It can
be shown that O (n) consists of all n × n matrixes such that A−1 = AT :
1 = det I = det AA−1 = det AAT
= det A det AT = (det A)2 (1.66)
t = ± (t − a) (1.68)
Now, combining different branches of (1.68) for t < a and t > a, gives
t = s (t) = |t − a| (1.69)
Function (1.69) also satisfies equation (1.67), for all t except may be single
point t = a, where the classic derivative of s (t) has no certain value. Namely,
equality holds for almost all t except may be t = a.
Nevertheless, solution (1.69) admits an obvious physical interpretation
since it describes a free material point moving in the space splitted into two
subspaces by a perfectly stiff plane, were the normal to the plane velocity is
scaled by
ṡ2 = 1 (1.70)
Since the perception of empty space rejects any built in stiff planes then a
sudden V -turn of the particle will represent unusual, in other words, strongly
non-linear event. This is obviously the most elementary nonlinear event
whose simplicity nevertheless will be employed further in less trivial cases.
The nonlinear force on the right-hand side gives the ‘rigid body’ limit as the
parameter approaches zero. Let us assume that ẋ = −1 and x → ∞ as
t → −∞. In this case, equation (1.71) has exact solution
t−a
x = 2 ln cosh 2 (1.72)
where a is an arbitrary parameter such that ẋ(a) = 0; see Fig. 1.6, where
a = 1.
Now let us consider the asymptotic limit → 0. Taking into account
evenness of cosh-function, brings solution (1.72) to the form
2 |t − a| 2 s
x = ln cosh = ln cosh (1.73)
2 2
x
2.0
Ε0.1
1.5 Ε0.5
Ε1.0
1.0
0.5
t
1 1 2 3
Fig. 1.6 Time histories of the particle when the barrier approaches rigid body
limit, → 0.
x → s = |t − a| (1.75)
as → 0.
The above example shows that (1.69) can be viewed as a natural time
associated with the temporal symmetry of the V-turn. Note however that
the new temporal argument is nonsmooth and non-invertible with respect to
the original time. Therefore, using (1.69) for temporal substitutions in the
differential equations of motion has certain specifics as discussed below.
Now relation (1.70) reveals that combination (1.76) represents specific al-
gebraic structure with the basis {1, ṡ}. In other words, time belongs to the
algebra of hyperbolic ‘complex numbers’ with the table of products gener-
ated by (1.70). For instance, taking (1.76) squared, gives another hyperbolic
‘number’
t2 = a2 + s2 + 2asṡ (1.77)
Moreover, it is easy to prove for any function x(t) that
where
1
X (s) = [x (a + s) + x (a − s)]
2
1
Y (s) = [x (a + s) − x (a − s)]
2
For instance, setting a = 0 in the case x(t) = exp t, gives
1 1 (a − sṡ) a s
= = = 2 − 2 ṡ
t a + sṡ (a + sṡ) (a − sṡ) a − s2 a − s2
Fig. 1.7 Mechanical model generating the saw-tooth sine and rectangular cosine,
τ (t) and τ̇ (t)
planes with no energy loss as shown in Fig. 1.7, where spatio-temporal coor-
dinates are normalized so that
dt2 = dτ 2 ⇐⇒ e2 = τ̇ 2 = 1 (1.80)
Note that relationship (1.80) is a periodic version of (1.70) whereas, for any
periodic function of the period T = 4, representation (1.1) is a periodic
version of (1.78). In other words, any periodic motion uniquely associates
with the standard impact vibration, for instance as follows
πt πt πτ πτ
x (t) = A sin + B cos = A sin + B cos e (1.81)
2 2 2 2
where τ = τ (t), e = τ̇ (t), and A and B are arbitrary constants.
Exercise 1. By taking sequentially derivatives, show that differentiation
keeps the result within the set of hyperbolic numbers provided that the cor-
responding process is smooth enough. In particular,
πτ πτ π 2n−1 πτ πτ
d2n−1 A sin + B cos e /dt2n−1 = −B sin + A cos e
2 2 2 2 2
πτ πτ
π 2n πτ πτ
d2n A sin + B cos e /dt2n = A sin + B cos e
2 2 2 2 2
The well known convenience of Fourier series for handling partial and or-
dinary differential equations is due to the fact that exp(iωt) is an eigen
function of the time derivative, in other words, d exp(iωt)/dt = iω exp(iωt).
This important advantage is unfortunately missing when using the set
{τ (ωk t), e(ωk t)} as a nonsmooth basis for Fourier expansions.
The alternative way is based on power series expansions. Indeed, in terms
of the oscillating time τ , polynomial approximations can be applied with no
periodicity loss. For example,
πt πτ πτ 1 πτ 3 1 πτ 5
sin = sin = − + − ...
2 2 2 3! 2 5! 2
πt πτ
1 πτ 2
1 πτ 4
cos = cos e= 1− + − ... e (1.83)
2 2 2! 2 4! 2
πt πτ π2 2
cos = cos e = [1 − τ + 1 −
2
τ − τ4
2 2 8
π2 π4 4
+ 1− + τ − τ 6 + ...]e
8 384
Fig. 1.9 illustrates both convergence and smoothness of the transformed series
as compared with the diagrams in Fig.1.8. Moreover, it will be shown that
the power series of τ can be re-ordered in such a manner that their particular
sums become as smooth as necessary. Still the convenience of such kind of
series is that they can accurately approximate non-smooth or close to them
processes just by first few terms.
Finally, there are many physical processes easily described by very
simple combinations of the triangular sine and rectangular cosine waves,
which otherwise would require quite long Fourier sums; see Fig. 1.10, for
examples.
30 1 Introduction
Fig. 1.8 Truncated Maclaurin NSTT expansions with respect to τ for sin(πt/2)
and cos(πt/2) including one, two, and three first terms of the expansions as indi-
cated on the curves.
Fig. 1.9 First term of the modified NSTT series with respect to τ for sin(πt/2)
and cos(πt/2); 0 - original functions, 1 - first term of the series.
t2
(ẍ + Px (x)) δxdt = 0 (1.86)
t1
or
t2
1 2
−δ ẋ − P (x) dt = −δS = 0 (1.87)
2
t1
Fig. 1.10 Sample temporal shapes of periodic processes easily described by com-
binations of the triangular sine and rectangular cosine
1.5 Non-smooth Coordinate Transformations 33
ẋ = kx3 + qδ (t − t1 ) (1.88)
−1≤x≤1 (1.92)
x = τ (s) (1.93)
ẍ + ω 2 x = 0, −1≤x≤1 (1.96)
9
Note that both notations and normalization for the period for the triangular
wave function differ from the original work [200].
1.5 Non-smooth Coordinate Transformations 35
to the form
s̈ + ω 2 τ (s)e(s) = 0, −∞<s<∞ (1.97)
Note that, as a side effect of the elimination of constraints, the differential
equation lost its linearity. In contrast to (1.97) however, the linear differential
equation in (1.96) does not describe the entire system, which is strongly non
linear from the very beginning due to impact interactions with constraints.
Finally, results of comparison between NSCT and NSTT can be summa-
rized as follows [143]:
Coordinate transformation Time transformation
Unfolds the space with no effect on Folds the time, and generates the
the time variable hyperbolic algebraic structures in
space
Targets rigid barriers (constraints) Barriers do no matter
Dynamic regime independent Assumes certain temporal symme-
tries of motion
Applies to a function (image) Transforms an argument (pre-
image)
Essentially changes the ODE struc- Preserves most structural properties
ture, for instance, linear on strongly of ODEs
nonlinear
As follows from the table, NSCT is quite opposite to NSTT from the very
different viewpoints.
where x = [x1 (t), x2 (t)]T is the state vector such that x2 = ẋ1 , and
0 1
A= (1.99)
−ω 2 0
x = Sy (1.100)
In the component-wise form, expressions (1.100) and (1.102) are written as,
respectively,
x1 = x1 (s, v) ≡ ssgn(s)
x2 = x2 (s, v) ≡ sgn(s)[1 − ksgn(sv)]v (1.103)
and
ṡ = [1 − ksgn(sv)]v
v̇ = −ω 2 s[1 + ksgn(sv)]/(1 − k 2 ) (1.104)
Now both unknown components of the state vector are continuous, whereas
specifics of non-elastic collisions are captured by transformation (1.103).
Finally, consider the general case of one-degree-of-freedom nonlinear
oscillator
ẋ1 = x2
ẋ2 = −f (x1 , x2 , t) (1.105)
ṡ = [1 − ksgn(sv)]v
v̇ = −f (x1 (s, v), x2 (s, v), t)sgn(s)[1 + ksgn(sv)]/(1 − k 2 ) (1.106)
the sum of smooth functions, such as sines, can represent a non-smooth shape
of the string. These discussions were finalized by the Fourier theorem.
Let us reproduce the result for a periodic function of time f (t) of the
period T in the complex form
∞
f (t) = ck exp(iωk t) (2.1)
k=−∞
T /2
1 2π
ck = f (t) exp(−iωk t)dt, ωk = k
T −T /2 T
The perturbation on the right-hand side of equation (2.3) changes the prin-
cipal frequency of the oscillator so that
ϕ = ωt (2.6)
as follows
ε
ȧ = − < f (a cos ϕ, −aω0 sin ϕ) sin ϕ >ϕ
ω0
ε
ϕ̇ = ω0 − < f (a cos ϕ, −aω0 sin ϕ) cos ϕ >ϕ (2.10)
ω0 a
Solutions of system (2.10) are considered then as approximate solutions of
the original system (2.9).
This method was essentially generalized in thirties [28] by incorporating
the Lindstedt-Poincare and Van-der-Pol’s ideas as follows.
Let us consider the general system with one fast phase
ẋ = εX(x, y)
ẏ = ω(x) + εY (x, y) (2.11)
x = q + εu(q, ψ) + O(ε2 )
y = ψ + εv(q, ψ) + O(ε2 ) (2.12)
which eliminates the fast phase entirely from the system by bringing equation
(2.11) to the form
q̇ = εA(q) + O(ε2 )
ψ̇ = ω0 (q) + εω1 (q) + O(ε2 ) (2.13)
2.2 A Brief Overview of Smooth Methods 41
l˙ = μR(μl, s, ϕ, θ)
ṡ = μ2 S(μl, s, ϕ, θ)
θ̇ = μQ(l, s) + μ2 Θ(μl, s, ϕ, θ) (2.14)
ϕ̇ = Ω(μl, s) + μ2 G(μl, s, ϕ, θ)
Following the idea of averaging, one eliminates the second term on the right-
hand side by means of the coordinate transformation
l = q + μf (q, s, ϕ, θ) (2.17)
42 2 Smooth Oscillating Processes
Now the fast phase ϕ is eliminated from the equation in first order of μ by
taking into account (2.14) and imposing condition
∂f
Ω(0, s) = R(0, s, ϕ, θ)− < R(0, s, ϕ, θ) >ϕ (2.19)
∂ϕ
Since the fast phase ϕ is eliminated from the terms of order μ then the
averaging procedure is applied to the terms of order μ2 analogously to the first
stage of the method. As follows from (2.19), < ∂f /∂ϕ >ϕ =< ∂f /∂θ >ϕ = 0,
therefore after the averaging, equation (2.20) takes the form
q̇ = μ < R(0, s, ϕ, θ) >ϕ +μ2 < qRμl (0, s, ϕ, θ) >ϕ +O(μ3 )
or
q̇ = μ < R(μq, s, ϕ, θ) >ϕ +O(μ3 ) (2.21)
The second approximation therefore is obtained by applying the operator of
averaging to original equation (2.14). Note however that the meaning of the
coordinate q is now different. Namely, the original coordinate l is expressed
through the new coordinate q by relationship (2.17), which due to (2.19)
takes the form
ϕ
μ
l=q+ (R(μq, s, ϕ, θ)− < R(μq, s, ϕ, θ) >ϕ )dϕ + O(μ2 ) (2.22)
Ω(μq, s)
0
In this expression, the variable μq was put back into the expression R in-
stead of zero. Although such a manipulation has no effect on the order of
approximation, the new form is in a better match with the form of equation
(2.21).
Expressions (2.21) and (2.22) summarize the averaging procedure in second
order of μ.
2.2 A Brief Overview of Smooth Methods 43
Note that the case of multiple fast phases turns out to be more complicated
in many respects due to the well known problem of small denominators.
dA dĀ
exp(iω0 t) + exp(−iω0 t) = 0 (2.26)
dt dt
By solving equations (2.23) and (2.24) with respect to A one obtains
1
A= exp(−iω0 t)(ẋ + iω0 x) (2.27)
iω0
Similar kind of complex amplitudes is used in both physics [89] and nonlinear
mechanics [102].
Now equation (2.3) gives
dA ε
= exp(−iω0 t)f (2.28)
dt iω0
where f = f (x, ẋ) is expressed trough (2.23) and (2.24).
Equation (2.28) is still exactly equivalent to (2.3). If the parameter ε is
small then the amplitude A is slow, and one can apply the averaging
44 2 Smooth Oscillating Processes
0
2π/ω
dA ε
= exp(−iω0 t)f dt (2.29)
dt 2πi
0
dz
= T (z, ε) , z |ε=0 = y
dε
where the choice for vector-function T (z, ε) depends upon desired properties
of the transformed system.
1
Marius Sophus Lie ( 1842-1899 ), Norwegian mathematician; different mathe-
matical objects are named after him, for instance, groups, operators, algebras,
and series.
2.2 A Brief Overview of Smooth Methods 45
q̈k + ωk2 qk = εFk (q1 , ..., qn+1 , q̇1 , ..., q̇n+1 ); k = 1, ..., n + 1 (2.30)
1 1
qk = (yk − ȳk ), q̇k = (yk + ȳk ) (2.32)
2iωk 2
Introducing the transformation (2.31) into the equations of motion (2.30),
gives
d
q̈k + ωk2 qk = (q̇k + iωk qk ) − iωk (q̇k + iωk qk ) =
dt
dyk
= − iωk yk = εFk (y1 , ..., yn+1 ; ȳ1 , ..., ȳn+1 )
dt
or
ẏk = iωk yk + εFk (y1 , ..., yn+1 ; ȳ1 , ..., ȳn+1 ) (2.33)
and the corresponding complex conjugate (cc) set of equations, where the
functions Fk (y1 , ..., yn+1 ; ȳ1 , ..., ȳn+1 ) are obtained by substituting (2.32) in
the right hand side of equation (2.30). These terms can be represented in the
polynomial form
46 2 Smooth Oscillating Processes
m l
Fk = Fkσ y1m1 · · · yn+1 ȳ1 · · · ȳn+1
n+1 l1 n+1
(2.34)
|σ|=2,3,...
1 ∂ |σ| Fk
Fkσ = m
|
ln+1 y=0
σ! ∂y1m1 · · · ∂yn+1
n+1
∂ ȳ1l1 · · · ∂ ȳn+1
Equations (2.33) correspond to the standard form, which is ready for analysis
in terms of Lie group operators.
To apply the theory of the Lie groups we rewrite equations (2.33) in the
form
ẏ = Ay, A = A0 + εA1 (2.35)
where y = (y1 , ..., yn+1 ; ȳ1 , ..., ȳn+1 )T , and
n+1
∂
n+1
∂
A0 = iωk yk + cc and A1 = Fk + cc (2.36)
∂yk ∂yk
k=1 k=1
ε2 2
y = e−εU z = z−εU z+ U z − ... (2.37)
2!
T
where z = (z1 , ..., zn+1 ; z̄1 , ..., z̄n+1 ) , and the operator of transformation U
is represented in the power series form with respect to the small parameter ε
U = U0 + εU1 + · · · (2.38)
where
Tj,k = Tj,k (z1 , ..., zn+1 ; z̄1 , ..., z̄n+1 ) (2.40)
are unknown functions to be determined.
2.2 A Brief Overview of Smooth Methods 47
One of the advantages of this process is that the inverse coordinate trans-
formation to the form (2.37) can be easily written as
eεU y = z (2.41)
ż = Bz (2.42)
ε2
B = A + ε [A, U ] + [[A, U ] , U ] + ... (2.43)
2!
where [A, U ] = AU − U A is the commutator of operators A and U .
An optimized iterative algorithm for high-order solutions of equation (2.43)
was suggested in [202] and [204]. In order to illustrate just the leading order
terms of asymptotic expansions, let us follow the direct procedure though.
Substituting the power series expansions for A and U given by relations (2.35)
and (2.38) into (2.43) gives
n+1
∂
B= {iωk zk + ε [Fk + (A0 − iωk ) T0,k ]} + O ε2 + cc (2.45)
∂zk
k=1
ẏ = Ay → ż = Bz
where
Equations (2.50) represent the normal form of the system, where the sum-
mation is much simpler than that in the original set (2.48). Namely, the
summation in (2.50) contains only those terms that give rise to resonance
while the first term on the right hand side stands for the fast component of the
2.2 A Brief Overview of Smooth Methods 49
Substituting (2.51) into (2.50) and taking into account the resonance condi-
tion, Δσk = 0, gives
mn+1 l1 ln+1 2
ȧk = ε Fkσ am 1
1 ...an+1 ā1 ...ān+1 + O ε (2.52)
|σ|=2,3,...
Δσk =0
ẍ + x2n−1 = 0 (3.1)
can be expressed in terms of special Lyapunov’s function [96], [77], [56] such
as snθ and csθ as defined by expressions1
snθ
1−2n
θ= 1 − nz 2 2n
dz, cs2n θ + n sn2 θ = 1
0
1 √ 1
√ dx π Γ 2n
T =4 n √ =2
1 − x2n n Γ n+1
2n
0
So the family of oscillators (3.1) includes the two quite simple cases asso-
ciated with the boundaries of the interval 1 ≤ n < ∞. Respectively, one has
the two couples of periodic functions
and
{x, ẋ} → {τ (t), τ̇ (t)}, if n → ∞ (3.5)
where τ̇ (t) is a generalized derivative of the sawtooth sine and will be named
as a rectangular cosine.
Earlier, the power-form characteristics with integer exponents were em-
ployed for phenomenological modeling amplitude limiters of vibrating elastic
structures [191] and illustrations of impact asymptotics [132], [137]. It should
be noted that such phenomenological approaches to impact modeling are de-
signed to capture the integral effect of interaction with physical constraints
bypassing local details of the dynamics near constraints. Such details first
of all depend upon both the vibrating body and amplitude limiter physical
properties. In many cases, Hertz model of interaction may be adequate to
describe the local dynamics near constraint surfaces [64]. Note that direct
replacement of the characteristic x2n−1 by the Hertzian restoring force kx3/2
in (3.1) gives no oscillator. The equation,
ẍ + kx3/2 = 0 (3.6)
ẍ + ksgn(x)|x|3/2 = 0 (3.7)
However, oscillator (3.7) essentially differs from oscillator (3.1) since equation
(3.7) describes no gap (clearance) between the left and right constraint sur-
faces. In other words models (3.1) and (3.7) still represent physically different
situations. The gap 2Δ with its center at the origin x = 0 can be introduced
in equation (3.7) as follows
p2 tan2 x tan x
H= + ⇒ ẍ + =0 (3.9)
2 2 cos2 x
and
p2 tanh2 x tanh x
H= + ⇒ ẍ + =0 (3.10)
2 2 cosh2 x
where p = ẋ is the linear momentum of the Hamiltonian H.
3.2 Nonlinear Oscillators Solvable in Elementary Functions 55
tan x cos2 x
100
50
Π Π Π Π
x
2 4 4 2
50
100
Fig. 3.1 Hardening characteristic.
tanh x cosh2 x
0.5
Π Π Π Π
x
2 4 4 2
0.5
Fig. 3.2 Softening characteristic.
Notice that oscillators (3.9) and (3.10) complement each other as those
with stiff and soft characteristics represented in Figs. 3.1 and 3.2, respectively.
These oscillators can be represented also in the form
1.0
Α0.001
0.5
Α0.999
Ns 0.0
0.5
1.0
0 1 2 3 4 5 6
Fig. 3.3 Normalized temporal mode shapes of the stiff oscillator, N s(ϕ, α) =
arcsin(α sin ϕ)/ arcsin α.
The original coordinate and the velocity are expressed by the action-angle
variables as follows [158]
√ √
2I + I 2 (1 + I) 2I + I 2 cos ϕ
x = arcsin sin ϕ , p = (3.17)
1+I 1 + (2I + I 2 ) cos2 ϕ
As it is seen, the differential equation of the oscillator (3.9) takes the linear
form with respect to the action-angle coordinates, and thus possess the exact
general solution
I = I0 , ϕ = (1 + I0 ) t + ϕ0 (3.20)
where I0 > 0 and ϕ0 are arbitrary constants. By substituting (3.20) in (3.17),
one can express the solution via the original coordinates. The meaning of the
initial action is clear from the energy relationship
58 3 Nonsmooth Processes as Asymptotic Limits
1 1
E = I0 + I02 = tan2 A (3.21)
2 2
Note that the linearity of the Hamiltonian equations is due to the specific
strongly non-linear form of the coordinate transformation (3.17). In other
words, the system nonlinearity has been ‘absorbed’ in a purely geometric way
by the nonlinear coordinate transformation.
As mentioned at the beginning, simplicity of the transformed system and
that of the corresponding solution can be essentially employed for the purpose
of perturbation analysis. Let us consider, however, the differential equation
of motion in the Newtonian form
tan x
ẍ + = εf (x, ẋ) (3.22)
cos2 x
were ε is a small parameter.
This system is weakly non-hamiltonian. However, it is still possible to
consider expressions (3.17) as a change of the coordinates {x, p} −→ {I, ϕ}
by imposing the compatibility condition dx/dt = p, where x and p must be
taken from (3.17). This gives
εf (x, p) sin ϕ
ϕ̇ = 1 + I −
(1 + I) (2I + I 2 ) [1 + (2I + I 2 ) cos2 ϕ]
√
εf (x, p) 2I + I 2 cos ϕ
I˙ = (3.23)
1 + (2I + I 2 ) cos2 ϕ
where the function f (x, p) must be expressed through the action-angle coor-
dinates by means of (3.17).
For instance, in the case of linear damping, f (x, p) ≡ −p, one obtains
ε cos ϕ sin ϕ
ϕ̇ = 1 + I + (3.24)
1 + (2I + I 2 ) cos2 ϕ
ε (1 + I) 2I + I 2 cos2 ϕ
I=−
˙
1 + (2I + I 2 ) cos2 ϕ
At this stage, let us implement just one step of the procedure and evaluate its
effectiveness. Applying the operator of averaging with respect to the phase
variable gives the corresponding first-order averaged system in the linear form
ϕ̇ = 1 + I, I˙ = −εI (3.25)
1.5 1.5
1.0 1.0
0.5 0.5
x 0.0 x 0.0
0.5 0.5
1.0 1.0
1.5 1.5
0 10 20 30 40 50 0 10 20 30 40 50
t t
10 10
5 5
v 0 v 0
5 5
10
1.5 1.0 0.5 0.0 0.5 1.0 1.5 1.5 1.0 0.5 0.0 0.5 1.0 1.5
x x
Fig. 3.4 The dynamics of the linearly damped stiff oscillator under the initial
conditions I0 = 10 and ϕ0 = 0, and two different damping parameters: ε = 0.2 (on
the left,) and ε = 0.8 (on the right.) Analytical and numerical solutions show a
slight mismatch only on the top right diagram.
ϕ̇ = 1 + I, I˙ = −εI 2
and thus
1 I0
ϕ=t+ ln (1 + εI0 t) + ϕ0 , I=
ε 1 + εI0 t
Using the coordinate transformation (3.17), gives solution
I0 (2 + I0 + 2εI0 t) ln(1 + εI0 t)
x = arcsin sin t + + ϕ0 (3.28)
1 + I0 + εI0 t ε
As Fig. 3.5 shows, the high-energy vibration shape approaches the rectangular
wave and thus essentially differs of that observed in the stiff case.
Based on solution (3.29), the action-angle coordinates are introduced by
means of expressions
√ √
2I − I 2 (1 − I) 2I − I 2 cos ϕ
x = arc sinh sin ϕ , p = (3.30)
1−I 1 − (2I − I 2 ) cos2 ϕ
where
1
I =1− (3.31)
cosh A
Nevertheless, all the analytical manipulations are analogous to those in the
stiff case. For instance, taking into account (3.31), gives the total energy as
a function of the action coordinate
1 1
E= tanh2 A = I − I 2 (3.32)
2 2
3.3 Nonsmoothness Hiden in Smooth Processes 61
1.0
Α106
0.5
Α0.001
Nh 0.0
0.5
1.0
0 1 2 3 4 5 6
ϕ̇ = 1 − I, I˙ = −εI (3.34)
0.8 0.6
0.6
0.4 0.4
0.2
x 0.0 x 0.2
0.2 0.0
0.4
0.6 0.2
0 10 20 30 40 50 0 10 20 30 40 50
t t
1.0 1.0
0.5
0.5
v v
0.0
0.0
0.5
0.5
0.60.40.2 0.0 0.2 0.4 0.6 0.8 0.2 0.0 0.2 0.4 0.6
x x
Fig. 3.6 The dynamics of the linearly damped softening oscillator under the initial
conditions I0 = 0.5, ϕ0 = 0, and two different damping parameters: ε = 0.2 (on
the left,) and ε = 0.8 (on the right.) Analytical and numerical curves practically
coincide.
from different viewpoints of physics and nonlinear dynamics [84], [59], [99],
[188], [88]. Interestingly enough, phase variables of interacting oscillators with
close natural frequencies may show non-smoothness of temporal behavior
during the beating [59], for instance similar to that of a vibro-impact process
[101], [104].
dAj dĀj
exp(iΩt) + exp(−iΩt) = 0 (3.38)
dt dt
dAj 1
= exp(−iΩt)fj (3.39)
dt iΩ
Assuming that the coupling and nonlinearity parameters are sufficiently small
and averaging the right-hand side with respect to Ωt, gives
βi 3α 2
Ȧ1 = A1 − A2 + A1 Ā1 (3.40)
2Ω 4β
βi 3α 2
Ȧ2 = −A1 + A2 + A Ā2
2Ω 4β 2
K = ψ 1 ψ̄ 1 + ψ 2 ψ̄ 2 (3.43)
64 3 Nonsmooth Processes as Asymptotic Limits
3α 4 4
G = ψ 1 ψ̄ 2 + ψ 2 ψ̄ 1 − |ψ 1 | + |ψ 2 | (3.44)
8β
Integral (3.43) admits substitution
√ 1 π
ψ 1 = K cos θ(t) + exp[i δ1 (t)] (3.45)
2 4
√ 1 π
ψ 2 = K sin θ(t) + exp[i δ 2 (t)]
2 4
3 K2 α
G = −K cos Δ cos θ − (3 − cos 2 θ) = const. (3.46)
32 β
β 3K α
Δ̇ = − cos Δ tan θ + sin θ (3.47)
Ω 8Ω
β
θ̇ = sin Δ (3.48)
Ω
where Δ = δ 2 − δ 1 + π.
It will be shown below that the phase variable θ, which determines the
process of energy flow between the oscillators, is described by the oscillator
(3.22).
Equations (3.47)-(3.48) are similar to those obtained in [101], [104], where
it was noticed that temporal shapes of the phase variables θ and Δ may resem-
ble the behavior of the state variables of impact oscillator. This observation
seems to be important since it is hard to expect any “impact oscillators” in
weakly nonlinear systems of type (3.36).
Below, the corresponding ‘conservative oscillator’ admitting the impact
limit will be explicitly obtained and analyzed by using the methodology de-
scribed in the previous section. However, the approach below deals with a
general class of nonlinear restoring force characteristics admitting power-
series expansions. It will be shown also that equations (3.47)-(3.48) can be
derived by introducing the standard set of amplitude-phase variables and
applying then the traditional one fast phase averaging technique.
where b is the coupling stiffness per unit mass, and p(u) is the restoring force
characteristic, which is assumed to be an analytic function that admits a
power series expansion.
Assuming that the system has equilibrium at zero and introducing the
notation
Ω2 = b + k
ε = b/Ω 2 = b/(b + k) (3.50)
f (u) = [(b + k)/b][p(u) − ku]
u̇1 = v1
u̇2 = v2
v̇1 = −Ω 2 u1 + ε[Ω 2 u2 − f (u1 )] (3.51)
v̇2 = −Ω u2 + ε[Ω u1 − f (u2 )]
2 2
Now, in order to track the oscillator energies during the vibration process,
let us use quantities
1 2 1 2
E1 = (v + Ω 2 u21 ) = Ω K cos2 ϕ
2 1 2
1 1 2
E2 = (v22 + Ω 2 u22 ) = Ω K sin2 ϕ (3.53)
2 2
and
1 2
E0 = E1 + E2 = Ω K (3.54)
2
Besides, expressions (3.53) and (3.54) clarify the physical meaning of the
variables K and ϕ participating in transformation (3.52), where other two
variables, δ1 and δ2 , are phases of the vibrating oscillators. In particular, K
is proportional to the total energy of the decoupled and linearized oscillators,
whereas the phase ϕ characterizes the energy split between the oscillators. In
case ε = 0, the energy parameter K will have small temporal fluctuations due
to coupling and nonlinear terms in (3.51). Nevertheless expressions (3.53) and
(3.54) still will be used as the energy related quantities for characterization
of the energy exchange process between the oscillators.
In order to conduct the transition to the new variables, let us substitute
(3.52) in (3.51) and then solve the set of equations with respect to the deriva-
tives as follows
√
2ε K
K̇ = −εKΩ sin 2ϕ sin(2Ωt + δ1 + δ2 ) +
√ Ω
×{f [ K cos ϕ cos(Ωt + δ1 )] cos ϕ sin(Ωt + δ1 )
√
+f [ K sin ϕ cos(Ωt + δ2 )] sin ϕ sin(Ωt + δ2 )}
ε ε
ϕ̇ = Ω[sin(δ1 − δ2 ) − cos 2ϕ sin(2Ωt + δ1 + δ2 )] − √
2 KΩ
√
×{f [ K cos ϕ cos(Ωt + δ1 )] sin ϕ sin(Ωt + δ1 )
√
−f [ K sin ϕ cos(Ωt + δ2 )] cos ϕ sin(Ωt + δ2 )} (3.55)
δ̇ 1 = −εΩ cos(Ωt + δ1 ) cos(Ωt + δ2 ) tan ϕ
ε √
+√ cos(Ωt + δ1 ) sec ϕf [ K cos ϕ cos(Ωt + δ1 )]
KΩ
δ̇ 2 = −εΩ cos(Ωt + δ2 ) cos(Ωt + δ1 ) cot ϕ
ε √
+√ cos(Ωt + δ2 ) csc ϕf [ K sin ϕ cos(Ωt + δ2 )]
KΩ
System (3.55) is still an exact equivalent to system (3.49) and represents a
standard dynamic system with a single fast phase, ψ = Ωt. As a next natural
stage, let us apply the direct averaging to the right-hand side of (3.55) with
respect to the fast phase ψ:
3.4 Nonlinear Beat Dynamics: The Standard Averaging Approach 67
K̇ = 0
ε
ϕ̇ = Ω sin(δ1 − δ2 )
2
ε ε
δ̇ 1 = − Ω cos(δ1 − δ2 ) tan ϕ + F1 (K, ϕ) (3.56)
2 Ω
ε ε
δ̇ 2 = − Ω cos(δ1 − δ2 ) cot ϕ + F2 (K, ϕ)
2 Ω
where the residue theorem has been applied so that
2π √
1 1
F1 (K, ϕ) = √ f ( K cos ϕ cos ψ) cos ψdψ
K cos ϕ 2π 0
1 √ 1 1 1 1
= √ Res{f [ K cos ϕ (z + )] (z + 2 )}
K cos ϕ 2 z 2 z
2π √
1 1
F2 (K, ϕ) = √ f ( K sin ϕ cos ψ) cos ψdψ
K sin ϕ 2π 0
1 √ 1 1 1 1
= √ Res{f [ K sin ϕ (z + )] (z + 2 )}
K sin ϕ 2 z 2 z
δ2 = δ1 + Δ − π
1 1
ϕ= θ+ π (3.57)
2 4
Substituting (3.57) in (3.56) and introducing the slow time parameter t1 =
εΩt, gives
dθ
= sin Δ
dt1
dΔ
= − cos Δ tan θ + F (θ) (3.58)
dt1
where
1 1 1 1 1
F (θ) = [F2 (K, θ + π) − F1 (K, θ + π)]
Ω2 2 4 2 4
It can be shown that system (3.58) has the integral
where
h(θ) = − F (θ) cos θdθ (3.60)
Now let us show that system (3.58) can be reduced to a single strongly
nonlinear oscillator with respect to the coordinate θ. Taking time derivative
of both sides of the first equation in (3.58) and eliminating from the result
dΔ/dt1 and cos Δ by means of the second equation in (3.58) and the integral
of motion, (3.59), respectively, gives
d2 θ tan θ
+ = R(θ) (3.61)
dp2 cos2 θ
f (x) = α3 x3 + α5 x5 (3.64)
K(6α3 + 5Kα5 )
h= cos2 θ (3.65)
32Ω 2
whereas equation (3.61) takes the form
d2 θ tan θ
2
+ = μ sin 2θ (3.66)
dp cos2 θ
3.4 Nonlinear Beat Dynamics: The Standard Averaging Approach 69
where
K 2 (6α3 + 5Kα5 )2
μ= (3.67)
2048G2 Ω 4
Substituting (3.57) in (3.53), gives the corresponding energy values versus
phase θ:
1
E1 = KΩ 2 (1 − sin θ)
4
1
E2 = KΩ 2 (1 + sin θ) (3.68)
4
d2 θ 4
+ (1 − 2μ)θ + (1 + μ)θ3 = 0 (3.69)
dp2 3
1
dΘdp
1
2
3
1.5 1.0 0.5 0.0 0.5 1.0 1.5
Θ
1
dΘdp
1
2
3
1.5 1.0 0.5 0.0 0.5 1.0 1.5
Θ
one can select δ1 (0) = 0. Then, taking into account expressions (3.57), (3.53)
and (3.54), gives
√ π θ(0)
u1 (0) = K cos +
4 2
v1 (0) = 0
√ π θ(0)
u2 (0) = − K cos Δ(0) sin + (3.71)
4 2
√ π θ(0)
v2 (0) = KΩ sin Δ(0) sin +
4 2
3.4 Nonlinear Beat Dynamics: The Standard Averaging Approach 71
dΘdp 1
1
2
3
1.5 1.0 0.5 0.0 0.5 1.0 1.5
Θ
and
E1 (0) − E2 (0)
θ(0) = − arcsin
E1 (0) + E2 (0)
v2 (0)
Δ(0) = − arctan
Ωu2 (0)
2
K= [E1 (0) + E2 (0)] (3.72)
Ω2
K(6α3 + 5Kα5 )
G= − cos Δ(0) cos θ(0) + cos2 θ(0)
32Ω 2
where
1 2
E1 (0) = v1 (0) + Ω 2 u21 (0)
2
1 2
E2 (0) = v2 (0) + Ω 2 u22 (0)
2
Since the initial velocity of the first oscillator is fixed v1 (0) = 0 then the
system initial state is determined by the three quantities K, θ(0), and Δ(0);
see (3.71). Alternatively, one can specify u1 (0), u2 (0) and v2 (0) and then find
K, θ(0), and Δ(0) from (3.72).
dI I(2 + I)
=μ sin 2φ
dp (1 + I)2
dφ 2μ
= 1+I − sin2 φ (3.74)
dp (1 + I)3
Note that the coordinate transformation (3.73) still would be valid for general
case (3.61) although with different to (3.74) result. In contrast to (3.66),
system (3.74) is weakly nonlinear with a very simple solution at μ = 0.
In [158], the direct averaging was applied to the right-hand side of (3.74)
in order to obtain the first-order solution. The idea of averaging can be also
implemented as asymptotic integration of system (3.74) by means of the
coordinate transformation
J(2 + J)
I = J −μ cos 2ψ + O(μ2 )
2(1 + J)3
(J 2 + 2J − 2)
φ = ψ−μ sin 2ψ + O(μ2 ) (3.75)
4(1 + J)4
where J = const.
3.4 Nonlinear Beat Dynamics: The Standard Averaging Approach 73
Taking into account equation (3.58), dθ/dp = |G|−1 sin Δ, and calculating
the left - hand side of (3.81) at p = 0, gives
ζ[cos Δ(0) − ν 1 − ζ]2 + sin2 Δ(0) < ζν 2 (3.82)
Note that both conditions (3.80) and (3.82) include the same set of param-
eters, such as the initial phase shift from the out-of-phase mode, Δ(0), the
parameter characterizing the initial energy distinction between the oscilla-
tors, ζ, and the parameter characterizing the total energy and thus strength
of nonlinearity, ν.
Fig. 3.10 The mechanical model admitting both normal and local mode motions;
all the springs have hardening restoring force characteristics.
∂2w ∂4w
ρA 2
+ EI 4 = f1 (t)δ(y − y1 ) + f2 (t)δ(y − y2 ) (3.83)
∂t ∂y
and
∂ 2 w(t, y)
w(t, y)|y=0,l = 0, |y=0,l = 0 (3.84)
∂y 2
where
∂w(t, yi ) ∂ 2 w(t, yi )
fi (t) = −f [w(t, yi )] − c −m ; i = 1, 2 (3.85)
∂t ∂t2
are transverse forces applied to the beam from masses attached at the two
points y = y1,2 .
It will be assumed that the model is perfectly symmetric with respect to
y = l/2 so that the springs are attached at points
76 3 Nonsmooth Processes as Asymptotic Limits
where
3c π 4 EI
ζ̄ = , λ2 = 3 (3.89)
3m + Alρ l (3m + Alρ)
and √ √
3 3
F (z) = f z (3.90)
3m + Alρ 2
are constant parameters and a re-scaled restoring force function, respectively.
3.7 Normal and Local Mode Coordinates 77
Equations (3.88) are decoupled in the linear terms related to the elastic
beam centre line, whereas the modal coupling is due to the spring nonlinear-
ities included in F (z).
Local mode coordinates. Alternatively, the model can be discretized by
introducing the ‘local mode’ coordinates determined by the spring locations
Taking into account (3.86) and (3.87), and substituting in (3.91) reveals sim-
ple links between the normal and local coordinates as
√ √
3 3
u1 = (W1 + W2 ), u2 = (W1 − W2 ) (3.92)
2 2
or, inversely, √ √
3 3
W1 = (u1 + u2 ), W2 = (u1 − u2 ) (3.93)
3 3
Substituting (3.93) in (3.87), gives the ‘local mode expansion’ for the beam’s
centre line πy πy
w(t, y) = u1 (t)ψ1 + u2 (t)ψ2 (3.94)
l l
where the local mode shape functions are
√
ψ1 (x) 3 11 sin x
= (3.95)
ψ2 (x) 3 1 −1 sin 2x
Both normal and local mode shape functions are shown in Figs. 3.11 and 3.12,
respectively. Transformation (3.95) can be generalized for a greater number of
modes. Note that functions (3.95) satisfy the following ortogonality condition
π
π
ψi (x) ψj (x) dx = δij (3.96)
0 3
1.0
sin x
0.5
sin 2
x
0.0
0.5
1.0
Fig. 3.11 Normal mode shape functions; here and below dashed lines correspond
to second mode.
1.0
1 x 2 x
0.5
0.0
0.5
1.0
u̇1 = v1
u̇2 = v2
v̇1 = −ω 2 u1 + ε[ω 2 u2 − ζv1 − p(u1 )] (3.98)
v̇2 = −ω 2 u2 + ε[ω 2 u1 − ζv2 − p(u2 )]
where
2
17 15 λ ζ̄
ω= 2k + λ2 , k = F (0), ε = , ζ=
2 2 ω ε
3.7 Normal and Local Mode Coordinates 79
and √ √ √
3 2 3 2 3
p(ui ) = F ui − kui = βu3i
ε 3 3
are new constant parameters and the nonlinear component of the spring
characteristic; it is assumed that the damping coefficient and the nonlinear
component are small enough to provide the order of magnitude ζ = O(1) and
p(ui ) = O(1).
For calculation purposes, the spring characteristic is taken in the form
F (u) = u + (4/3)u3 which brings the nonlinearity parameter β to the form
64ω 2
β= (3.99)
135λ2
Equations (3.97), and analogously (3.98), possess advantages for transient
analysis because the corresponding linearized system has the same natu-
ral frequencies and the nonlinear components are decoupled. As a result,
the one-frequency perturbation tool becomes applicable. The corresponding
amplitude-phase variables are introduced as follows
α1 = −s(t) + ρ(t)
α2 = s(t) + ρ(t) (3.102)
δ2 = δ1 + Δ(t)
80 3 Nonsmooth Processes as Asymptotic Limits
In the case |ζ|
1, equation (3.103) describes an oscillator with a slow vary-
ing frequency. Making the frequency ‘frozen’ enables one of the determining
roots of the corresponding ‘characteristic equation’
1 3 2
k1,2 = ε − ζ ± i ω − βs
2 (3.105)
2 4
1 4ω 2
t∗ = ln (3.106)
ε|ζ| 3βs20
In order to provide numerical evidence for the dynamic transition from nor-
mal to local mode vibrations, let us introduce an indicator of the energy
partition calculated as
⎧
E1 − E2 ⎨ −1 if E1 = 0 and E2 = 0
P = = 0 if E1 = E2 (3.107)
E1 + E2 ⎩
1 if E1 = 0 and E2 = 0
where Ei = (vi2 + ω 2 u2i )/2 is the total energy of i-th oscillator under the
condition ε = 0.
3.8 Local Mode Interaction Dynamics 81
0.0
0.2
0.4
P
0.6
0.8
t
1.0
0 1000 2000 3000 4000 5000
t
Fig. 3.13 ‘Sudden’ transition from normal to local mode vibration as the system
energy has reached its critical value.
Δ = δ2 − δ1 + π (3.109)
Further, considering the local mode total energies Ei under no interaction
condition, and taking into account (3.100), (3.108) and (3.110), gives
1 2
Ei = (v + ω 2 u2i ); i = 1, 2 (3.110)
2 i
82 3 Nonsmooth Processes as Asymptotic Limits
1 2 2 1
E = E1 + E2 = ω (α1 + α22 ) = ω 2 K (3.111)
2 2
1 1
ΔE = E1 − E2 = ω 2 (α21 − α22 ) = − ω 2 K sin θ (3.112)
2 2
The variable K therefore is proportional to the total energy of the degen-
erated system, whereas the phase angle θ characterizes the energy partition
between the local modes (3.107) as follows
ΔE
P = = − sin θ (3.113)
E
The third variable (3.109) describes the phase shift in the high-frequency
vibrations between the local modes so that Δ = 0 corresponds to the out-
phase motions of the masses attached to the beam; note the difference with
(3.102).
Differentiating (3.109), (3.111) and (3.112), and enforcing equations (3.101),
gives
dκ ζ
=− κ
dt1 ω
dθ
= sin Δ (3.114)
dt1
dΔ
= − cos Δ tan θ + κ sin θ
dt1
where t1 = εωt is a new temporal argument, and
3β
κ= K (3.115)
8ω 2
In the conservative case, ζ = 0, the first equation in (3.114) gives the energy
integral κ = const, whereas another two equations admit the integral
1
G = − cos Δ cos θ + κ cos 2θ = const (3.116)
4
This particular case matches the results obtained for a linearly coupled set of
Duffing’s oscillators in [101], and later reproduced in [158] however by means
of different complex variable approaches. In particular, it was shown in [158]
that the last two equations in (3.114) are equivalent to a strongly nonlinear
conservative oscillator
d2 θ tan θ
+ G2 =0 (3.117)
dt21 cos2 θ
as κ → 0.
3.8 Local Mode Interaction Dynamics 83
Under condition (3.119), the factor κ in the third equation of (3.114) can
be viewed as a slowly growing quasi constant. In this case, making κ ‘frozen’
and linearizing the last two equations in (3.114) near the equilibrium (θ, Δ) =
(0, 0), gives
d2 θ
+ (1 − κ)θ = 0 (3.120)
dt21
Note that small θ and Δ bring the original system close to the out-of-phase
vibration mode as follows from (3.113) and (3.109). When the growing en-
ergy parameter κ passes the critical point κ = 1, the type of equilibrium
is changing from a focus to a saddle point and thus the variable θ becomes
exponentially growing. Practically, however, the exponential growth will be
suppressed by the nonlinearity. As a result two limit phase trajectories (sep-
aratrix loops) occur around two new stable equilibrium points. These two
points represent two new stable modes of the original system - local modes.
The phase plane diagrams for sub- and super-critical energy levels are shown
in Figs. 3.14 and 3.15, respectively. Note that the equilibrium subjected to
such qualitative change corresponds to the out-of-phase vibration mode of
the model, whereas another two equilibrium points, (θ, Δ) = (0, ±π), corre-
spond to the same in-phase mode and remain stable. Therefore, out-of-phase
vibrations appear to be less favorable to energy equipartition as the energy
reaches its critical level. Note that links between localization and specifics
of phase trajectories was discussed also in [101] based on the system of two
coupled Duffing oscillators. In particular, the limit phase trajectories were
interpreted as nonlinear beats of infinitely long period, keeping the energy
near one of the two oscillators.
84 3 Nonsmooth Processes as Asymptotic Limits
4
I
0 O
2
4
0 L L
2
4
3β ∗ ∗ 1 8ω 2
K 0 exp (ε |ζ| t ) = 1 =⇒ t = ln (3.121)
8ω 2 ε |ζ| 3βK0
As follows from (3.102) and (3.111), K0 = 2(s20 +ρ20 ) therefore, under the
condition ρ20
1, expressions (3.106) and (3.121) give the same result.
Note that the developed analytical approach, describing the local mode
interaction in terms of the energy and phase variables, appears to be inde-
pendent of the individual features of the illustrating model and this can be
used in other similar cases.
Compared to publications [101] and [158] introducing the same set of de-
scriptive variables, K, θ and Δ, current approach has distinctive features as
follows:
1) Instead of general mass-spring models, the elastic beam supported by
nonlinear springs is considered in this work. This provides clear geometrical
interpretations for both the normal and local modes through the correspond-
ing shape functions of the beam centre line.
2) Instead of using a quite complicated system reduction in terms of com-
plex coordinates, it is shown that the same result can be achieved by means of
the traditional set of amplitude-phase variables and one-frequency averaging
procedure.
3) The non-conservative case is considered in order to describe qualitative
changes in the dynamics as the total energy of the system adiabatically in-
creases or decreases. Based on such a generalization, new quantitative and
qualitative results are obtained.
In particular, explicit expressions have been obtained for the critical time
at which onset of the localization occurs. The phenomenon is explained in
terms of the related phase-plane diagram subjected to a qualitative change
(center-saddle transition) as the total energy of the system reaches its critical
level.
iβ
Ȧj = − (Aj−1 − 2Aj + Aj+1 )+
2Ω
αΩ 4 2 2
2
2 2
2
+ |Aj | − |Aj−1 | |Aj−1 | − |Aj+1 | − |Aj | |Aj+1 | Aj (3.125)
8
Let us consider first, the simplest model of two coupled oscillators (N = 2).
In this case, system (3.125) is reduced to
iβ αΩ 4
Ȧ1 = − (A2 − 2A1 ) + |A1 |2 − |A2 |2 |A2 |2 A1 (3.126)
2Ω 8
iβ αΩ 4
Ȧ2 = − (A1 − 2A2 ) + |A2 |2 − |A1 |2 |A1 |2 A2
2Ω 8
Despite of the presence velocities ẋj in the original equations (3.122), sys-
tem (3.126) still has the integral
2 2
K = |A1 | + |A2 | = 2(E1 + E2 )/Ω 2 = const.
where the angle ψ determines the energy distribution between the oscillators
as follows
|A2 | E2
tan ψ = = (3.128)
|A1 | E1
0 ≤ ψ < π/2
Substituting (3.127) into (3.126) and considering separately real and imagi-
nary parts, gives
β β
ϕ̇1 = − tan ψ cos (ϕ2 − ϕ1 )
Ω 2Ω
β β
ϕ̇2 = − cot ψ cos (ϕ2 − ϕ1 ) (3.129)
Ω 2Ω
β 1
ψ̇ = − sin (ϕ2 − ϕ1 ) − αK 2 Ω 4 sin 4ψ
2Ω 32
Introducing the phase shift Δ = ϕ2 −ϕ1 and new temporal variable p = Ωt/β,
gives
dΔ
= − cot 2ψ cos Δ (3.130)
dp
dψ 1
= − (sin Δ + λ sin 4ψ)
dp 2
88 3 Nonsmooth Processes as Asymptotic Limits
Π
2
Π
4
0
Π Π
2 0 2
Fig. 3.16 Low energy transition to the nonsmooth limit cycle; numerical solution
obtained for the following system parameter and initial conditions: λ = 0.2; Δ(0) =
0.0, Ψ (0) = π/4 + 0.1.
As follows from (3.127) and (3.128), at point (3.133), both oscillators vibrate
in-phase with the same energy, E1 = E2 . Linearized (near (3.133)) system
(3.130) has the following couple of roots of characteristic equation
r1,2 = λ ± i 1 − λ2 (3.134)
3.9 Auto-localized Modes in Nonlinear Coupled Oscillators 89
Π
2
Π
4
0
0 25 50
p
Π
2
Π
2
0 25 50
p
Fig. 3.17 Low energy transition to the “impact” limit cycle of phase variables at
λ = 0.2.
Expression (3.134) determines the ‘low energy’ interval 0 < λ < 1 with a
qualitatively similar system behavior. Equilibrium point (3.133) is unstable
for positive λ while no other equilibrium points exist within the rectangular
(3.132). As a result, the system trajectory is eventually attracted to the
boundary of rectangular R0 (3.132) as shown in Figs. 3.16 and 3.18. This is
a periodic limit cycle whose period is found in a closed form,
π/2 0
dψ dψ 2π
P =2 −2 = √ (3.135)
1 − λ sin 4ψ 1 + λ sin 4ψ 1 − λ2
0 π/2
where the horizontal pieces of the boundary ∂R0 have zero contribution as
those passed momentarily by the system (3.130). This is confirmed also by
the diagrams in Figs. 3.17 and 3.19 showing step-wise jumps of the variable
Δ(p) in steady state limits.
90 3 Nonsmooth Processes as Asymptotic Limits
Π
2
Π
4
0
Π Π
2 0 2
Fig. 3.18 Transition to the nonsmooth limit cycle under the energy level approach-
ing its critical value; the numerical solution obtained for the following system pa-
rameter and initial conditions: λ = 0.8; Δ(0) = 0.0, Ψ (0) = π/4 + 0.1
2
As already mentioned, the possibility of ‘vibro-impact dynamics’ of phase vari-
ables was noticed later in [101] when considering another model of nonlinear
beats.
3.9 Auto-localized Modes in Nonlinear Coupled Oscillators 91
Π
2
Π
4
0
0 15 30
p
Π
2
Π
2
0 15 30
p
Fig. 3.19 Transition to the “impact” limit cycle of phase variables at λ = 0.8.
Chapter 4
Nonsmooth Temporal Transformations
(NSTT)
Example 1. Combining identities (4.4) and the power series expansion of the
exponential function, gives
. |t|2 |t|3
exp (t) ≡ exp (|t| |t| ) = 1 + + ... + |t| + + ... |t|.
2! 3!
4.1 Non-smooth Time Transformations 95
or
exp (t) = cosh (|t|) + sinh (|t|) |t|. (4.5)
Expression (4.5) represents the exponential function as a sum of the even and
odd components.
Let us take a general function, x (t), and assume that
. .
x (t) = x (|t| |t| ) = X (|t|) + Y (|t|) |t| (4.6)
This gives the following set of equations for the components, X (|t|) and
Y (|t|),
.
x (|t|) = X (|t|) + Y (|t|) for |t| = 1
.
x (− |t|) = X (|t|) − Y (|t|) for |t| = −1
and, finally,
1
X (|t|) = [x (|t|) + x (− |t|)] (4.7)
2
1
Y (|t|) = [x (|t|) − x (− |t|)]
2
Therefore, identity (4.6) holds under condition (4.7). It is seen that the right-
hand side of expression (4.6) represents an element of the algebra with the
.
basis {1, |t| }. The corresponding ‘table of products’ is generated by the re-
. 2
lationship (|t| ) = 1. This leads to other useful algebraic relationships, for
instance,
.
X + Y |t| = 0 ⇐⇒ {X = 0, Y = 0}
and
. .
f (X + Y |t| ) = Rf (X, Y ) + If (X, Y ) |t| (4.8)
where
1
Rf (X, Y ) = [f (X + Y ) + f (X − Y )]
2
1
If (X, Y ) = [f (X + Y ) − f (X − Y )] (4.9)
2
Therefore, introducing the nonsmooth positive time |t|, imposes specific com-
plexification on the system coordinate
s = s (t) = |t − a| (4.11)
and thus,
1
X (s) = [x (a + s) + x (a − s)] (4.14)
2
1
Y (s) = [x (a + s) − x (a − s)]
2
Therefore, identity (4.12) holds for any x (t) under condition (4.14).
t = A (τ ) + B (τ ) τ̇ (4.16)
where
1 1 1
A= + τ and B = − τ
v1 v2 v1 v2
4.1 Non-smooth Time Transformations 97
and
τ̇ 2 = −v1 v2 + (v1 + v2 ) τ̇ (4.17)
Further, applying some function x to both sides of equality (4.16), gives
x (t) = X (τ ) + Y (τ ) τ̇ (4.18)
τ̇ 2 = 1 (4.24)
tn = An (τ ) + Bn (τ )τ̇ (4.28)
Now, let us assume that expression (4.29) holds even though the func-
tion x(t) is not analytic. In this case, one must show that the X- and Y -
components can be determined with no power series expansions. Indeed, since
either τ̇ = 1 or τ̇ = −1 on the entire interval −1 < t < 3, except may be the
point t = 1, then expression (4.29) gives two equations, x (τ ) = X (τ ) + Y (τ )
and x (2 − τ ) = X (τ ) − Y (τ ). Solving these equations for X and Y and
substituting the solution in (4.29), gives the identity,
1 1
x(t) = [x (τ ) + x (2 − τ )] + [x (τ ) − x (2 − τ )] τ̇ (4.30)
2 2
which obviously holds on the interval −1 < t < 3, except may be t = 1.
If the function x(t) is continuous at t = 1 then identity (4.30) is also true
at t = 1 because the step-wise discontinuity of the rectangular cosine τ̇ (t) at
t = 1 is suppressed by the factor
2Y (τ ) = x (τ ) − x (2 − τ ) → x (1 − 0) − x (1 + 0) → 0 (4.31)
f (X + Y τ̇ ) = Rf + If τ̇ (4.34)
1
Rf = [f (X + Y ) + f (X − Y )]
2
1
If = [f (X + Y ) − f (X − Y )]
2
provided that f (X ± Y ) are defined.
The right-hand sides of the above expressions appear to have the same hy-
perbolic structure of the element X + Y τ̇ .
Another example resembles Euler formula in the complex analysis, however
hyperbolic functions are involved as follows
Example 2. Introduce the sawtooth time into the functions sin t and cos t.
4.1 Non-smooth Time Transformations 101
Solution. In order to deal with the period normalized to four, let us introduce
new argument ϕ = 2t/π as
π
π 2t
sin t = sin = sin ϕ
2 π 2
where τ and e still depend on the same argument 2t/π. The right-hand sides
of these equalities can be viewed as elements of a more complicated algebra,
z = α + βe + γi + δei, with the basis elements {1, e, i, ei}. The corresponding
table of products is given by
× 1 e i ei
1 1 e i ei
e e 1 ei i
i i ei −1 −e
ei ei i −e −1
ẋ = Y + X τ̇ + Y τ̈ (4.39)
ẍ = X + Y τ̇ (4.40)
X |τ =±1 = 0 (4.41)
4.1 Non-smooth Time Transformations 103
provided that
Y (τ ) |τ =±1 = 0
X (τ ) |τ =±1 = 0
... (4.43)
Y (2k−2)
(τ ) |τ =±1 =0
if
Y (τ ) |τ =±1 = 0
X (τ ) |τ =±1 = 0
... (4.45)
X (2k−1)
(τ ) |τ =±1 = 0
1
X(τ )dτ = 0 (4.47)
−1
x2 = (X + Y e)2 = X 2 + Y 2 + 2XY e
Τ
1
t
1 2 3 4 5 6 7 8
1
e
1 Γ1
t
1 2 3 4 5 6 7 8
1
1 Γ
In this case, the inverse transformation of time on the period has the form
1 − γ2 1 1
t= [(1−γ) τ ] +e +[2−(1+γ) τ ] −e (4.54)
2 1+γ 1−γ
(−1 + γ ≤ t ≤ 3 + γ)
1 1 1
X= x [(1 − γ) τ ] + x [2 − (1 + γ) τ ]
2α 1 + γ 1−γ
1
Y = {x [(1 − γ) τ ] − x [2 − (1 + γ) τ ]} (4.56)
2α
e2 = α + βe (4.57)
where α = 1/ 1 − γ 2 and β = 2γα.
As compared to (4.36), relationship (4.57) somewhat complicates algebraic
and differential operations so that, respectively,
1 1 1
Rf (X, Y ) = f (Z+ ) + f (Z− )
2α 1 + γ 1−γ
1
If (X, Y ) = [f (Z+ ) − f (Z− )]
2α
Y
Z± = X ±
1∓γ
and
∂e
ẋ = αY + (X + βY ) e + Y (4.59)
∂t
where f (Z+ ) and f (Z− ) are defined, primes indicate differentiation with
respect to τ , and
∞
∂e
= 2α [δ (t + 1 − γ − 4k) − δ (t − 1 + γ − 4k)] (4.60)
∂t
k=−∞
ẋ = αY + (X + βY )e
ẍ = αX + αβY + [βX + (α + β 2 )Y ]e
Y |τ =±1 = 0
(X + βY )|τ =±1 = 0
Note that rules (4.61) are still valid in the symmetric case γ = 0, when the
differentiation matrix operator takes the form
01 d
D=
1 0 dτ
where
× e0 e1 e2 e3
e0 e0 e1 e2 e3
e1 e1 e0 e3 e2 (4.66)
e2 e2 e3 e0 e1
e3 e3 e2 e1 e0
where the coefficients at the right-hand side are defined by the system of
linear equations
1 = e+ + e− (4.70)
e = e+ − e−
Therefore,
x = X + Y e = X(e+ + e− ) + Y (e+ − e− )
= (X + Y )e+ + (X − Y )e−
X+ = X + Y = x(τ )
X− = X − Y = x(2 − τ ) (4.72)
The elements e+ and e− are mutually annihilating, and they are called idem-
potents because
e+ e− = 0
e2− = e− (4.73)
e2+ = e+
110 4 Nonsmooth Temporal Transformations (NSTT)
Fig. 4.2 The standard basis and idempotent basis (on the left and right, respec-
tively.)
As follows from (4.70), ee+ = e+ and ee− = −e− . Obviously, the matrix
representation of (4.71) is
X+ 0
X+ e+ + X− e− ←→
0 X−
Properties (4.73) make the idempotent basis very convenient to use in differ-
ent calculations. For instance, for any real number α,
(X+ e+ + X− e− )α = X+
α α
e+ + X− e− (4.74)
t = 1 + (τ − 1) e = (e+ + e− ) + (τ − 1) (e+ − e− )
= τ e+ + (2 − τ )e− (−1 < t < 3) (4.76)
t2 = τ 2 e+ + (2 − τ )2 e−
tα = τ α e+ + (2 − τ )α e− (4.77)
(α > 0)
Note that expressions (4.77) provide the direct proof of representation (4.71),
at least, in the class of analytical periodic functions x(t).
1
(X+ e+ − X− e− ) = f (X+ e+ + X− e− ) = f (X+ )e+ + f (X− )e−
a
or
X+ = af (X+ )
X− = −af (X− ) (4.82)
1 = e+ + e− (4.84)
1 1
e= e+ − e−
1−γ 1+γ
Fig. 4.3 The standard basis and idempotent basis (on the left and right, respec-
tively) generated by the asymmetric triangular wave with γ = 0.3.
t = (1 − γ) τ e+ + [2 − (1 + γ) τ ] e− (4.85)
1 ∂X+ 1 ∂X− 1 ∂e
ẋ = e+ − e− + (1 − γ 2 )(X+ − X− ) (4.92)
1 − γ ∂τ 1 + γ ∂τ 2 ∂t
These conditions eliminate the singular term ∂e/∂t from expression (4.92),
which, as a result, takes the form
1 ∂X+ 1 ∂X−
ẋ = e+ − e− (4.94)
1 − γ ∂τ 1 + γ ∂τ
1 ∂ 2 X+ 1 ∂ 2 X−
ẍ = e+ + e−
(1 − γ) ∂τ
2 2 (1 + γ) ∂τ 2
2
1 1 ∂X+ 1 ∂X− ∂e
+ (1 − γ 2 ) + (4.96)
2 1 − γ ∂τ 1 + γ ∂τ ∂t
As a result, the second derivative ẍ takes the form of expansion with respect
to the basis {e+ , e− }. Such a process can be continued as soon as derivatives
remain continuous in Λγ . Otherwise, time derivatives of the function e(t, γ)
acquire non-formal meaning and must be present in the entire expression.
1 ∂ 2 X+
+ f (X+ ) = p (4.99)
(1 − γ)2 a2 ∂τ 2
1 ∂ 2 X−
+ f (X− ) = q (4.100)
(1 + γ)2 a2 ∂τ 2
since algorithms for solving equations are often more complicated than those
applied to the boundary conditions.
where
τ
P+ (τ, γ) = (1 − γ) X+ (z, γ)dz + C
−1
τ
P− (τ, γ) = −(1 + γ) X− (z, γ)dz + C (4.102)
−1
Note that both expressions in (4.102) have the same arbitrary constant of
integration C. As a result, substituting (4.102) in (4.101) and taking into
account (4.84), gives a single constant of integration as follows: Ce+ + Ce−
= C(e+ + e− ) = C.
Proof of (4.102) is obtained by taking time derivative of both sides of
equality (4.101). Under the condition of continuity of the right-hand side of
(4.101), the differentiation leads to the boundary value problem
1 ∂P+ (τ, γ)
= X+ (τ, γ)
1−γ ∂τ
1 ∂P− (τ, γ)
− = X− (τ, γ) (4.103)
1+γ ∂τ
(P+ − P− )|τ =±1 = 0
This boundary value problem has solution (4.102) under the following con-
dition though
1 1
(1 − γ) X+ (τ, γ)dτ + (1 + γ) X− (τ, γ)dτ = 0 (4.104)
−1 −1
1 1
x[(1 − γ)τ ]d[(1 − γ)τ ] + x[2 − (1 + γ)τ ]d[(1 + γ)τ ]
−1 −1
(1−γ)
(1−γ)
3+γ T
= x(z)dz − x(z)dz = x(z)dz = x(t)dt = 0
−(1−γ) 2+(1+γ) −1+γ 0
In other words, condition (4.104) requires zero mean value for x(t) on the
period. Otherwise, the result of integration would appear to be out of the
class of periodic functions by making the algebraic structure of right-hand
side of (4.101) invalid.
d2 x
+ f (x) = 0 (4.105)
dt2
Suppose that another oscillating time parameter, say g, is introduced as
x(t) = X[g (t)], where g (t) is a periodic function, which is not necessarily
sawtooth. This kind of substitution brings the differential equation of motion
to the form
2 2
dg d X d2 g dX
+ + f (X) = 0 (4.106)
dt dg 2 dt2 dg
A side effect of such substitution is that, generally speaking, the system has
lost its original Newtonian form. In order to keep the differential equation
of motion in its original Newtonian form, the following conditions must be
imposed on equation (4.106)
2
dg d2 g dX
=1 and =0 (4.107)
dt dt2 dg
g =α+t or g = β − t (4.108)
second equality holds for all t under the condition (4.41). As a result, the
new equation (4.106) takes the same form as (4.105).
Therefore, the triangular wave time substitution τ (t) possesses the unique
property among all periodic time substitutions, namely, it preserves the form
of differential equations of conservative oscillators.
where the functions q (t), f (x, ẋ) and ∂f (x, ẋ) /∂ ẋ must be continuous ac-
cording to the Cauchy theorem.
Let the system be subjected to an impulsive external loading applied at
t = t0 . In this case, the external forcing function can be expressed by the
Dirac δ-function2
ẍ + f (x, ẋ) = Iδ (t − t0 ) (4.110)
where I is a constant linear momentum per unit mass.
The differential equation of motion (4.110) cannot be treated within the
classic theory of differential equations. It is also impossible to find a solution
in the class of twice differentiable functions. However, a physically meaningful
interpretation of equation (4.110) is suggested by the theory of distributions.
For example, in terms of function-theoretic approaches, distributions are not
classical functions but linear functionals acting on manifolds of appropriate
differentiable functions {ϕ (t)}. Further, equalities are thought to be integral
2
The Dirac δ-function belongs to the class of so-called generalized functions intro-
duced by Sergei Sobolev in 1935 and re-introduced by Laurent Schwartz in the
late 1940s, who developed a theory of distributions.
4.4 Discussions, Remarks and Justifications 119
"∞
Therefore, x (t) generates a linear functional of the form (...) x (t) dt and
−∞
thus can be qualified as a generalized solution3 .
In nonlinear cases, the concept of generalized solution does not work, but
equation (4.112) still holds and enables one to use the concept of ‘weak so-
lution’ [46]. This concept is involved to justify manipulations with the nons-
mooth functions at intermediate steps of the approach.
Now, let us the right-hand side of the equation includes the δ-impulse as a
summand
ẋ = f (t, x) + νδ (t − a) (4.113)
where a and v are constant parameters.
In this case, the right-hand side does not satisfy any more the Caratheodory
conditions. However, changing the variable
where the Dirac’s function has been eliminated due to the equality Ḣ (t − a) =
δ (t − a); as a result the right-hand side, F (t, y), satisfies the Caratheodory
conditions.
Note that, due to the presence of discontinuous function H (t − a), the
right hand side of the transformed equation, F (t, y), is generally piecewise
continuous in t, even though the function f (t, x) may be continuous.
Relationships (4.113) through (4.115) admit direct extensions on the vector
space so that many mechanical and physical models can be represented in
the form (4.113).
d4 W
+ γ (ξ) W = qδ (ξ − a) (4.117)
dξ 4
Let us show now that introducing the nonsmooth argument (4.11) eliminates
the delta-pulse in equation (4.113) in such a way that no step-wise disconti-
nuity occurs in the new equations. Indeed, substituting (4.12) and (4.13) into
equation (4.113) and taking into account expression s̈ = 2δ (t − a), gives
where expressions
1
Rf = [f (a + s, X + Y ) + f (a − s, X − Y )]
2
1
If = [f (a + s, X + Y ) − f (a − s, X − Y )]
2
are obtained analogously to (4.34).
Eliminating the delta-pulse in (4.118) and equating separately both com-
ponents of the remaining hyperbolic element to zero, gives the boundary-value
problem
X + If (s, X, Y ) = 0 (4.119)
Y + Rf (s, X, Y ) = 0
v
Y |s=0 =
2
122 4 Nonsmooth Temporal Transformations (NSTT)
ẍ + (a + b cos 2t) x = 0
which admits periodic solutions in terms of the power series with respect to
the new temporal argument |τ | ≤ 1 [195].
Transformation (4.120) with the power series methods were employed for
non-linear vibrating systems as well [198], [110], [163]. Non-harmonic time
transformations dealing with Jacobian functions can be also found in the
literature [21]. As it is known, however, such transformations of time are
restricted by special cases and cannot be applied to any periodic motion.
From the mathematical point of view, it is caused by the fact that an inverse
transformation does not exist on the whole period.
In this section, different versions of periodic time are introduced in such a
way that the corresponding transformations are valid for any periodic motion.
This is reached by a special complexification of the coordinates.
Let us start with a generalization of substitution (4.120). For the sake of
compactness, notations
Proof. By collecting separately terms with odd and even wave numbers in
the corresponding Fourier series, one obtains
∞
A0
x (ϕ) = + [A2n cos 2nϕ + A2n−1 cos (2n − 1) ϕ]
2 n=1
∞
+ [B2n sin 2nϕ + B2n−1 sin (2n − 1) ϕ] (4.123)
n=1
Then, introducing notations (4.121) into the tabulated expressions [52] (For-
mulas No. 1.332), gives
n
n
cos 2nϕ = a2i τ 2i , cos[(2n − 1) ϕ] = a2i−1 τ 2i−2 e
i=0 i=1
n
n
sin 2nϕ = b2i−2 τ 2i−1 e, sin[(2n − 1) ϕ] = b2i−1 τ 2i−1(4.124)
i=1 i=1
and
e2 = 1 − τ 2 (4.126)
Note also that the components of representation (4.122) apparently are lin-
early independent and thus the whole combination is zero if and only if its
both components are zero.
124 4 Nonsmooth Temporal Transformations (NSTT)
Let x = τ (ϕ) be the system coordinate determined implicitly from the energy
integral
τ(ϕ)
ds
=ϕ (4.136)
1 − P (s)
0
e2 = 1 − P (τ ) (4.137)
where
1
e (ϕ) = τ (ϕ) and e (ϕ) = − P (τ ) (4.138)
2
Now let us formulate (without proof)
Proposition 5. Any periodic function x (ϕ) whose period is normalized to
1
ds
T =4
1 − P (s)
0
can be represented in the form (4.122), where the functions τ (ϕ) and e (ϕ)
are given by (4.136) and (4.138).
For example, one can transform equation (4.127) based on the potential
energy function P (x) = x2n . Then the boundary value problems (4.128)
through (4.131) and (4.132) through (4.134) can be derived as particular
cases n = 1 and n −→ ∞, respectively.
N
x(t) = A + Bn cn2n (ωt + αn , k 2 ) (4.139)
n=1
E(m)
x(t) = − + dn2 [2K(m)t] (4.140)
K(m)
where K(m) and E(m) are compete elliptic integrals of first kind and sec-
ond kind, respectively, and m = k 2 is the parameter (Jacobi), the period is
normalized to unity.
If the parameter m is small enough, function (4.140) takes almost harmonic
shapes, however, when m is getting larger then x(t) describes so-called cnoidal
waves as shown in Fig. 4.4.
From the physical point of view, function (4.140) with specific scaling
factors exactly describes temporal behavior of the interaction force between
particles of the Toda lattice
π(e+ + e− ) π2
xn,p (t) = − + (4.144)
2K(m)K(1 − m) [2K(1 − m)]2
n
p+1
×{ cosh−2 [λ(m)(τ − 4l)]e+ + cosh−2 [λ(m)(τ − 2 + 4l)]e− }
l=−n l=−p
πK(m)
λ(m) =
4K(1 − m)
128 4 Nonsmooth Temporal Transformations (NSTT)
Fig. 4.5 NSTT of the periodic cnoidal wave at different Jacobi parameters m:
3-three terms truncation (n = 0, p = 0), 5-five terms truncation (n = 1, p = 0),
and ∞ - exact expression.
where the infinite limits of summation have been replaced by the integers
n > 0 and p > 0; the oscillating triangular wave time, τ = τ (4t), and the
idempotent basis e+ = [1 + e(4t)]/2 and e− = [1 − e(4t)]/2 are introduced;
substitution 1 = e+ + e− has been made in order to emphasize that the
entire expression (4.144) has the form of hyperbolic number represented in
the idempotent basis.
4.4 Discussions, Remarks and Justifications 129
Obviously, x∞,∞ (t) = x(t), so that, in this case, (4.144) becomes exact
equivalent to (4.143). However, convergence properties of series (4.144) are
significantly improved as compared to (4.143). Indeed, as time t is growing
in (4.143), one must switch from one term to another to keep sufficient preci-
sion of approximation so that on the infinite time interval, −∞ < t < ∞, the
entire series (4.143) is needed. In contrast, the temporal argument of series
(4.144) is always bounded by the standard interval −1 ≤ τ ≤ 1, whereas the
periodicity of wave is captured by the basis functions rather than multiple
terms of the series. As a result, terms of series (4.144) are exponentially de-
caying as the summation index increases. As follows from the diagrams in
Fig. 4.5, for the range of Jacobi parameter m > 0.2 just five terms in (4.144)
are enough to capture both quasi-harmonic and cnoidal wave shapes with a
very good precision. For greater parameters m, when the wave becomes essen-
tially cnoidal, only three terms provide quite a perfect match with the exact
shape. As seen from the upper fragment in Fig. 4.5, the three term approx-
imation gives essential discontinuous error for very small Jacobi parameter
m = 0.08. However, in this almost harmonic range, it is more effective to
use few or even one term Fourier series (4.142) rather than series (4.143).
Another essential advantage of series (4.144) is that, in the cnoidal range,
different algebraic manipulations with truncated series are essentially eased
due to the idempotent properties.
Chapter 5
Sawtooth Power Series
2N
X (i) (0)
X (τ ) = τ i + O τ 2N +1 (5.1)
i=0
i!
Obviously, the power terms τ i and, generally speaking, finite sums in (5.1)
are non-differentiable at the set of of points Λ = {t : τ (t) = ±1}. In contrast,
the binomials
τi τ i+2
φi (τ ) = − ; i = 1, 2, ... (5.2)
i i+2
are twice differentiable with respect to t on Λ that can be verified by taking
derivatives with respect to t. For instance, first two generalized derivatives
are
dφi (τ )
= τ i−1 (1 − τ 2 )τ̇ (t)
dt
d2 φi (τ )
= (i − 1) τ i−2 − (i + 1) τ i
dt2
where the term τ i−1 (1 − τ 2 )τ̈ has been removed from the second derivative
due to the presence factor 1 − τ 2 that takes zero value whenever τ̈ = 0.
It is seen therefore that functions (5.2) are twice continuously differentiable
with respect to t even though each of the functions is combined of two non-
differentiable terms. This is achieved by the specific choice for the coefficients
and signs in (5.2) such that the power terms have same jumps of slopes but
with opposite signs.
Now, considering (5.2) as equations with respect to different powers of τ ,
gives, respectively odd and even powers, as
τ 2N +3
τ = φ1 (τ ) + ... + φ2N +1 (τ ) +
2N + 3
τ 2N +3
τ 3 = 3 φ3 (τ ) + ... + φ2N +1 (τ ) + (5.3)
2N + 3
2N +3
τ
τ 2N +1 = (2N + 1) φ2N +1 (τ ) +
2N + 3
and
2 τ 2N +2
τ = 2 φ2 (τ ) + ... + φ2N +2 (τ ) +
2N + 2
4 τ 2N +2
τ = 4 φ4 (τ ) + ... + φ2N +2 (τ ) + (5.4)
2N + 2
2N τ 2N +3
τ = 2N φ2N (τ ) +
2N + 2
These polynomials provide continuity for the term ψi [τ (t)]τ̇ (t) as well as its
first derivative
d[ψi (τ )τ̇ ]
= ψi (τ ) τ̇ 2 = iτ i−1 − (i + 2) τ i+1
dt
where the term ψi (τ ) τ̈ has been eliminated due to the boundary conditions
ψi (±1) = 0.
However, second derivative appears to be a step-wise discontinuous func-
tion at every time t ∈ Λ,
d2 [ψi (τ )τ̇ ]
2
= i (i − 1) τ i−2 − (i + 2) (i + 1) τ i τ̇
dt
Therefore, particular sums of the re-built series
i (2k−1)
∞
Y (0) Y (2k) (0)
Y (τ ) τ̇ = ψ2i−1 (τ ) + ψ2i (τ ) τ̇ (5.7)
i=0 k=0
(2k − 1)! (2k)!
are less smooth than those obtained for the X-component (5.5).
Combining (5.5) and (5.7) and considering an arbitrary period, T = 4a,
gives
∞
∞
τi τ i+2
x (t) = X (0) + KX (i) − +e KY (i) τ i − τ i+2 (5.8)
i=1
i i+2 i=0
where τ =τ (t/a) and e =τ (t/a), and the coefficients are calculated as follows
i
X (2k−s) (0)
KX (2i − s) =
(2k − 1 − s)!
k=1
i
Y (2k−s) (0)
KY (2i − s) = (5.9)
(2k − s)!
k=0
(s = 0, 1)
134 5 Sawtooth Power Series
Example 9
ln 3 π 2 τ2 τ4 π2 π4 τ4 τ6
ln (2 + cos t) = + − + − − + ...
2 12 2 4 12 48 4 6
ln 3 ln 3 π 2 2
+ 1 − τ2 + − τ − τ4
2 2 12
ln 3 π 2 π4 4
+ − + τ − τ 6 + ... e
2 12 192
φi (τ ) φi+2 (τ )
ϕi (τ ) = (3)
− (3)
∈ C4 (5.10)
φi (τ ) |τ =1 φi+2 (τ ) |τ =1
and,
ψi (τ ) τ̇ = τ i − τ i+2 τ̇ ∈ C 1
ψi (τ ) ψi+2 (τ )
χi (τ ) τ̇ = (2)
− (2)
τ̇ ∈ C 3 (5.11)
ψi (τ ) |τ =1 ψi+2 (τ ) |τ =1
ẋ = v
v̇ = −f (x, v, t) (5.13)
ṫ = 1
Then, the dynamics of system (5.13) can be locally described by the Lie series
[70]
1
x = exp[(t − t0 )G]x0 ≡ 1 + (t − t0 )G + (t − t0 )2 G2 + · · · x0 (5.14)
2!
∂ ∂ ∂
G = v0 · − f (x0 , v0 , t0 ) · + (5.15)
∂x0 ∂v0 ∂t0
where G is Lie operator associated with system (5.13), and {x0 , v0 , t0 } is
some initial point in the system’ phase space.
The dot between two quantities indicates dot products. For example,
∂ ∂ ∂
v0 · ≡ v01 + · · · + v0n
∂x0 ∂x01 ∂x0n
Series (5.14) is simply Taylor series whose coefficients are calculated by en-
forcing equations (5.13). Unfortunately, this general idea is still of little use for
oscillatory processes probably due to locality of expansion (5.14). In other
words, even entire expansion (5.14) does not explicitly reveal such global
characteristics of oscillations as their amplitude and period. Moreover, the
136 5 Sawtooth Power Series
Theorem 1. [145] Assume that system (5.13) admits a periodic solution x(t)
of the period T = 4a so that x(t + 4a) = x(t) for any t, and some point
{x0 , v0 , t0 } belongs to this solution. Then such a solution can be expressed in
the form
where τ and e are saw-tooth sine and rectangular cosine, whose periods are
normalized to four and amplitudes are normalized to unity as
and,
e(ϕ) = sgn[ cos(πϕ/2)] (5.18)
respectively, and ϕ = (t−t0 )/a is a re-scaled time. If, in addition, the solution
is odd with respect to one half of the period, x(t+2a) = −x(t), then expression
(5.16) simplifies to
e2 = 1 (5.21)
at almost every time instance1 . In order to prove the particular case (5.19),
one should keep in mind that exp(2aG)x0 = x(t0 + 2a) = −x0 , as it follows
from (5.14), and the oddness condition assumed.
Note that τ and e are indeed quite simple piece-wise linear functions; the
above analytical expressions (5.17) and (5.18) just define them in the unit-
form which enables one to avoid conditioning of computation in the original
temporal scale, t0 ≤ t < ∞. This possibility becomes essential when the
dynamics includes some evolutionary component.
1
The set of isolated points {ϕ : τ (ϕ) = ±1}appears to have no effect on the results
[144].
5.2 Sawtooth Series for Normal Modes 137
cosh(aG)x0 = 0
cosh(aG)v0 = 0 (5.22)
ẍ + f (x) = 0, x ∈ Rn (5.23)
where f (−x) = −f (x), and the initial conditions are x|t=0 = x0 = 0 and
ẋ|t=0 = v0 .
The normal mode solutions of system (5.23) are obtained as a particular
case of (5.19) and (5.22)
∂ ∂
G = v0 · − f (x0 ) ·
∂x0 ∂v0
have been applied.
Relationship (5.24) can be interpreted as a parametric equation of normal
mode trajectories of the system with the parameter interval −1 ≤ τ ≤ 1.
Let us illustrate relationships (5.24) and (5.25) based on the linear sys-
tem so that the result could be compared with the well known conventional
solution.
Example 10. Suppose that f (x) = Kx, where K is positively defined sym-
metric n × n-matrix with eigen-system {v0 , ω 2 } so that Kv0 = ω 2 v0 . In
this case, by applying the operator G twice, one obtains that v0 is also an
eigen-vector of the operator G2 , namely, G2 v0 = −ω 2 v0 . Then, keeping in
mind the power series form of expressions (5.24) and (5.25) as those in (5.19)
and sequentially applying the operator G2 , gives x = (v0 /ω) sin(aωτ ) and
cos(aω) = 0, respectively. Notably, the last equation shows that there ex-
ist an infinite number of roots {a} related to the same eigen-frequency ω!
However, it is easily to find that all the roots produce the same solution in
terms of the original time t. The minimal quarter of the period is a = π/(2ω),
therefore x = (v0 /ω) sin(πτ /2), and τ = (2/π) arcsin sin ωt.
Nonlinear cases and the related problems dealing with truncated expansions
of (5.25) will be further discussed in a full-length paper.
X + a2 Rf (X, Y, X , Y , τ ) = 0 (5.26)
Y + a2 If (X, Y, X , Y , τ ) = 0
5.3 Lie Series of Transformed Systems 139
The idea of Lie series enables one of representing solution of the system (5.26)
in the power series form with respect to the sawtooth argument τ under the
initial conditions at P0 (X0 , Y0 , X0 , Y0 , τ0 ); it is assumed that τ0 = 0. Then,
the corresponding solution of the original equation is represented in the form
ẍ + f (x, t) = 0 (5.32)
where the vector-function f (x, t) is odd with respect to the positional vector
x and even with respect to the quarter of the period t = a. In this case,
boundary-value problem (5.26) and (5.27) reduces to
140 5 Sawtooth Power Series
X + a2 f (X, aτ ) = 0
X |τ =1 = 0 (5.33)
Y ≡0
∂ 3 X(X0 , τ )
|τ =1 = 0
∂τ 3
Example 11. Consider now a two degree of freedom oscillating system
d2 x1 ζ (2x1 − x2 ) + x31 − P sin Ωt 0
+ =
dt2 x2 ζ (2x2 − x1 ) + x32 0
In this case, the parameter a = π/(2Ω) is the quarter of the period of the
external forcing function. The operator Lie of the system is
∂ ∂ πτ ∂
G = X1 + X2 − a2 ζ (2X1 − X2 ) + X13 − P sin
∂X1 ∂X2 2 ∂X1
∂ ∂
−a2 [ζ (2X2 − X1 ) + X23 ] +
∂X2 ∂τ
and
3
τ3 τ τ5 1
X2 = τ − A2 + − A2 + (a2 ζ A1 − 2 a2 ζ A2 )
3 3 5 2
5.3 Lie Series of Transformed Systems 141
τ5 τ7 1
+ − A2 + (a2 ζ A1 − 2 a2 ζ A2 )
5 7 2
1 4
+ [a P π ζ − 8 a4 ζ 2 A1 + 2A2 5 a4 ζ 2 − 6 a2 A2 2 ]
48
In this case, equations (5.31) give two scalar equations for A1 and A2 ,
∂ 3 X1 (A1 , A2 , τ ) ∂ 3 X2 (A1 , A2 , τ )
|τ =1 = 0, |τ =1 = 0 (5.34)
∂τ 3 ∂τ 3
where the first and the second equations guarantee smoothness of the ac-
celeration components ẍ1 (t) and ẍ2 (t), respectively. Intersections of curves
(5.34) on the Cartesian plane A1 A2 allow to locate the roots, as it is
shown√in Figs. 5.1 and 5.2. For instance, setting ζ = 1.0, P = 0.2 and
Ω = 1.01, gives (A1 , A2 ) = (0.7711, 0.6677). This solution corresponds to
a weakly nonlinear perturbation of the in-phase √ linear mode. Similarly, in
the neighborhood of the second frequency Ω = 3.01, one finds the solution
(A1 , A2 ) = (0.5971, −0.8181), which is close to the out-of-phase linear mode;
see Fig. 5.1. Interestingly enough, Fig. 5.2 shows that there are also two
roots, (2.3315, 2.2864) and (−2.2065, −2.2548), corresponding to in-phase vi-
brations even though the frequency of the external forcing function is close
to that of the out-phase linear mode. This effect may take place due to large
amplitudes such that the in-phase nonlinear frequency becomes close to the
out-of-phase linear frequency.
Since the initial conditions, corresponding to the periodic regimes are known,
one can integrate the differential equations of motion numerically in order to
check the analytical solutions. In the latter example, one would obtain that
both results are perfectly matching. It must be noted, however, that some of
the periodic solutions may appear to be unstable. As a result, even a very
small imperfection in the initial conditions will lead to a significant divergence
of the results. Moreover, the ‘tails’ of polynomial expansions may give roots
which do not correspond to any solution. Nevertheless, the above approaches
seems to make sense as compared to those based on the direct Fourier expan-
sions. Indeed, the number of algebraic equations in (5.22), (5.30) or (5.31) is
independent on the number of terms in the series.
Fig. 5.1 The curves of smoothness for√ the accelerations ẍ1 (t) and ẍ2 (t), respec-
tively thin and solid lines, when Ω = 1.01.
Fig. 5.2 The curves of smoothness for√ the accelerations ẍ1 (t) and ẍ2 (t), respec-
tively thin and solid lines, when Ω = 3.01.
5.3 Lie Series of Transformed Systems 143
motion are still derivable from the transformed functions in the standard way
of the analytical dynamics.
Consider first Lagrangian that depends on time t periodically with the
period T = 4a,
L = L(x, ẋ, t) (5.35)
On the manifold of smooth periodic motions of the same period T , the vector-
function x(t) is represented as x (t) = X (τ ) + Y (τ ) e, where τ = τ (t/a) and
e = e(t/a). As a result, Lagrangian (5.35) takes the form
H = H(p, q, t) (5.37)
which may be periodic with respect to time with the period T = 4a.
On the manifold of periodic motions of the period T , the coordinates and
the linear momenta are represented as
q = X (τ ) + Y (τ ) e and p = U (τ ) + V (τ ) e
where
RaH a
= [H (U + V, X + Y, aτ ) ± H (U − V, X − Y, 2a − aτ )]
IaH 2
144 5 Sawtooth Power Series
N1 N2
i j
τ1 τ1i+2 τ2 τ2j+2
X (τ1 , τ2 ) = X (0, 0) + KX (i, j) − − (5.38)
i=1 i=1
i i+2 j j+2
ẍ + ω02 x = 0 (6.1)
First, let us obtain exact general solution of the oscillator (6.1) in terms of
the sawtooth temporal argument by using the substitution
x = X (τ ) + Y (τ ) e (6.2)
where τ = τ (t/a) and e = e (t/a) are the standard triangular and rectangular
wave functions, respectively.
Substituting (6.2) in (6.1), gives the boundary value problem
Tj = 4aj = jT (6.8)
where A and B are arbitrary constants, and x(t) is the same harmonic wave
regardless the number j.
In this section, the free linear oscillator was considered for illustrating
purposes. Of course, there is no other pragmatic reason for introducing the
sawtooth time into equation (6.1). The situation drastically changes however
in non-autonomous cases of non-smooth or discontinuous inputs. It is shown
below that, in such cases, the sawtooth time variable can help to facilitate
determining particular solutions. The effect of ‘temporal quantization’ repre-
sented by expression (6.9), which seems to be just identical transformation
in the autonomous case, acquires helpful meaning at the presence of external
6.2 Non-autonomous Case 147
(U − V )|τ =±1 = 0
(U + V )|τ =±1 = 0 (6.21)
6.3 Systems under Periodic Pulsed Excitation 149
All the coefficients and right-hand sides of both equations in (6.20) are con-
stant, and the equations are decoupled. As a result, solution of boundary
value problem (6.20) and (6.21) is easily obtained in the form
p 2p exp(−ατ )
U (τ ) = 2
− 2
(6.22)
ω βω (cos 2β + cosh 2α)
×[cos β cosh α(β cos βτ + α sin βτ ) + sin β sinh α(α cos βτ − β sin βτ )]
p 2p exp(ατ )
V (τ ) = − + (6.23)
ω2 βω 2 (cos 2β + cosh 2α)
×[cos β cosh α(β cos βτ − α sin βτ ) + sin β sinh α(α cos βτ + β sin βτ )]
where α = ωaζ and β = ωa 1 − ζ 2 .
Substituting (6.22) and (6.23) in (6.19), gives closed form particular so-
lution of original equation (6.17). Transition to the original temporal vari-
able is given by the functions τ (ϕ) = (2/π) arcsin[sin(πt/2)] and e(ϕ) =
sgn[cos(πt/2)]. Since the system under consideration is linear, the general
solution of equation (6.17) can be obtained by adding general equation of the
corresponding equation with zero right-hand side. Finally, note that neither
trigonometric expansions nor any integral transforms were involved into the
solution procedure.
its derivatives. For instance, let us seek a particular solution of the first order
differential equation
∞
v̇ + λv = μ [δ (t + 1 − 4k) − δ (t − 1 − 4k)] (6.24)
k=−∞
Fig. 6.1 If the particle’ mass is very small compared to the total mass of the
tank then the inertia force applied to the particle inside the tank has the periodic
pulse-wise character.
First, note that the right-hand side of equation (6.24) can be expressed
through the generalized derivative of the rectangular wave function as follows
μ
v̇ + λv = ė (t) (6.25)
2
Now let us represent the particular solution in the form
μ
Y + λX = 0, X + λY = 0, Y |τ =±1 = (6.28)
2
In contrast to equation (6.24) or (6.25), boundary value problem (6.28) in-
cludes no discontinuities whereas the new independent variable belongs to
the standard interval, −1 ≤ τ ≤ 1.
Solving the boundary value problem (6.28) and taking into account sub-
stitution (6.26), gives periodic solution of equation (6.24) in the form
μ
v = X +Ye = (− sinh λτ + e cosh λτ )
2 cosh λ
or
μ
v= exp [−λτ (t) e (t)] e (t) (6.29)
2 cosh λ
Fig. 6.2 illustrates solution (6.29) for μ = 0.2 and different magnitudes of λ.
0.2
Λ 0.4
0.1
v 0.0
Λ 1.9
0.1
0.2
0 2 4 6 8
t
Fig. 6.2 The family of discontinuos periodic solutions.
de (ωt)
ẍ + ω02 x = p (6.31)
d (ωt)
where Y ≡ 0.
Solution (6.35) is continuous, but nonsmooth at those times t where
τ (ωt) = ±1. All possible resonances are given by
2 ω0
ω= ; k = 1, 3, 5, ... (6.36)
π k
where the factor 2/π is due to different normalization of the periods for
trigonometric and sawtooth sines.
Now let us consider the case of viscous damping described by the differen-
tial equation of motion
de (ωt)
ẍ + 2ζ ẋ + ω02 x = p (6.37)
d (ωt)
0.2
0.1
x 0.0
0.1
0.2
0 2 4 6 8
t
Fig. 6.3 Response of the damped harmonic oscillator under the periodic impulsive
excitation for p = 0.1, ζ = 0.5, ω0 = 4 and ω = 0.2 (low-frequency pulses.)
Figs. 6.3 through 6.5 illustrate qualitatively different responses of the sys-
tem when varying the input frequency. In different proportions, the responses
combine properties of the harmonic damped motion and the non-smooth mo-
tion due to the impulsive loading. For instance, when ω >> ω0 and ω >> ζ,
the system is near the limit of a free particle under the periodic impulsive
force. In this case, the boundary value problem is reduced to
p
X = 0, Y = 0; X |τ =±1 = , Y |τ =±1 = 0 (6.40)
ω2
154 6 NSTT for Linear and Piecewise-Linear Systems
0.06
0.04
0.02
x 0.00
0.02
0.04
0.06
0 2 4 6 8
t
Fig. 6.4 System response on ‘resonance’ pulses ω = (2/π)ω0 = 2.5465.
0.003
0.002
0.001
x 0.000
0.001
0.002
0.003
0 2 4 6 8
t
Fig. 6.5 Response on high-frequency pulses; ω = 6.
de (ωt)
M ÿ + Ky = p (6.41)
d (ωt)
Let {e1 , ..., eN } and ω1 ,...,ωN be the normal mode basis vectors and the
corresponding natural frequencies, respectively, such that
N
y= xj (t) ej (6.42)
j=1
N
(eTj p)ej sin [(ωj /ω) τ (ωt)]
y= (6.44)
j=1
ωωj cos (ωj /ω)
∂ 2 τ (ωt, γ) ∂e (ωt, γ)
v̇ + λv = p 2 =p (6.45)
∂ (ωt) ∂ (ωt)
∞
2p
= [δ (ωt + 1 − γ − 4k) − δ (ωt − 1 + γ − 4k)]
1 − γ2
k=−∞
156 6 NSTT for Linear and Piecewise-Linear Systems
v = X (τ ) + Y (τ ) e (6.46)
where τ = τ (ωt, γ) and e = e (ωt, γ); see Fig. 6.6 for graphic illustrations.
Substituting (6.46) in equation (6.45), gives
∂e(ωt, γ)
ωαY + λX + [ω (X + βY ) + λY ] e + (ωY − p) =0 (6.47)
∂(ωt)
where α = 1/ 1 − γ 2 , β = 2γα, and the identity e2 = α + βe has been taken
into account.
Equation (6.47) is equivalent to the boundary-value problem
ω (X + βY ) = −λY
ωαY = −λX (6.48)
ωY |τ =±1 = p
where the last term, that consists of the periodic sequence of δ-functions,
must be excluded by imposing the boundary condition for Y -component
Y |τ =±1 = 0 (6.53)
(X + βY ) |τ =±1 = 0 (6.55)
Substituting (6.51) and (6.54) in the differential equation of motion (6.50)
and taking into account the algebraic properties, brings the left-hand side of
the equation to the algebraic form {· · ·} + {· · ·}e. Then, setting separately
each of the two algebraic components to zero, gives the set of differential
equations for X (τ ) and Y (τ ) in the following matrix form
α αβ X 1 αε X
+ r2 =0 (6.56)
β α + β2 Y ε 1 + βε Y
where r = ω0 /ω.
Further, any particular solution of linear differential equations with con-
stant coefficients (6.56) can be represented in the exponential form
X 1
=B exp (λτ ) (6.57)
Y μ
1 αk 2 − r2 1 αl2 − r2
μ1 = − and μ2 = −
α βk 2 − εr2 α βl2 − εr2
Substituting (6.60) in boundary conditions (6.53) and (6.55), gives the ho-
mogeneous set of four linear algebraic equations with respect to the arbitrary
6.4 Parametric Excitation 159
10
6
10Ε
0
0 2 4 6 8 10
r
Fig. 6.7 Instability zones for piecewise constant parametric excitation when
γ = 0.7
One the parameter plane, ε − r, equation (6.61) describes the family of curves
separating stability and instability zones as shown in Fig. 6.7, where the
instability zones are shadowed.
2
2 2r2 1 − γ 2 sin2 2r
p = (6.63)
cos 4r − cos 4γr
1.0
0.8
0.6
p
0.4
0.2
0.0
0 2 4 6 8 10
r
Fig. 6.8 ‘Collapse’ of the instability zones at γ = 1/5: each fifth zone is missing;
here and below, only the upper half-plane is shown due to the symmetry.
1.0
0.8
0.6
p
0.4
0.2
0.0
0 2 4 6 8 10
r
ω 2 (αX + αβY ) + QX + αP Y = 0
ω 2 α + β 2 Y + βX + P X + QY + βP Y = 0 (6.65)
where, in the case of fixed sign of impulses, the matrix p should be provided
with the factor sgn(τ ).
Together with (6.53), relations (6.65) and (6.66) represent a boundary-
value problem for determining the vector functions X and Y and the corre-
sponding conditions for existence of periodic solutions.
Note that substitution (6.51) in equation (6.64) generates the specific term
e∂e/∂ϕ. Let us show that, within the theory of distributions, this terms can
be interpreted as follows
∂e 1 ∂e
e = β (6.67)
∂ϕ 2 ∂ϕ
First, note that, at this point, the relationship (6.67) is a result of formal
differentiation of both sides of the relation e2 = α + βe with respect to the
phase ϕ. To justify (6.67), let us assume that ω = 1 so that ϕ ≡ t and consider
expression (6.53) locally, near the point t = 1 − γ, which is a typical point of
the entire set of discontinuity points Λ = {t : τ (t) = ±1}.
162 6 NSTT for Linear and Piecewise-Linear Systems
for all positive ε, and δε (t) = 0 outside the interval −ε < t < ε.
Therefore, in terms of weak limits, δε (t) → δ (t) as ε → 0.
Now, sequences of smooth functions approximating e and ∂e/∂t in the
neighborhood of point t = 1 − γ can be chosen as, respectively,
1 β ∂eε β
eε = − θε (t − 1 + γ) and = − δε (t − 1 + γ) (6.69)
1−γ γ ∂t γ
"t
where θε (t) = δε (ξ) dξ and −1 + γ < t < 3 + γ.
−∞
Based on the above definitions for eε and ∂eε /∂t, one has eε → e and
∂eε /∂t → ∂e/∂t as ε → 0 in the interval −1 + γ < t < 3 + γ.
Substituting (6.69) in equality (6.67) instead of e and ∂e/∂ϕ, reduces the
problem to the proof of identity
1
θε δ ε = δε (6.70)
2
as ε → 0.
For simplicity reason, let us move the origin to the point t = 1 − γ and
show that the left-hand side of (6.70) gives δ (t) /2 as ε → 0 in the sense of
weak limit.
First, the area bounded by θε δε is
ε ε
dθε 1 1
θε δε dt = θε dt = θε2 |ε−ε =
−ε −ε dt 2 2
Then, let φ (t) belongs to the class of continuous testing functions, which
is usually considered in the theory of distributions. By definition, in some
ε-neighborhood of the point t = 0, one has | φ (t) − φ (0) |< 2η, where η is as
small as needed whenever ε is sufficiently small. Therefore,
ε ε
1
| θε (t) δε (t) φ (t) dt − φ (0) |≤ θε (t) δε (t) | φ (t) − φ (0) | dt ≤ η
−ε 2 −ε
6.5 Input-Output Systems 163
In other words, ε
1
θε (t) δε (t) φ (t) dt → φ (0)
−ε 2
as ε → 0.
This completes the proof.
d4 x1 d3 x1 m1 d2 x1 c1 dx1 k1 k2
m1 4
+ c1 3 + (k1 + k2 + k2 ) 2 + k2 + x1
dt dt m2 dt m2 dt m2
d2 F1 k2
= 2
+ F1 (6.72)
dt m2
System (6.72) is a particular case of (6.71), where n = 4 and m = 2.
Let us consider the step-wise discontinuous periodic function F1 (t) =
u(t) = e(ωt) and represent equation (6.72) in the form
164 6 NSTT for Linear and Piecewise-Linear Systems
d4 y dy
a4 4
+ ... + a1 + a0 y = b2 ω 2 e + b1 ωe + b0 e (6.73)
dt dt
where ≡ d/d(ωt), and all the coefficients and variables are identified by
comparing (6.72) to (6.73).
The right-hand side of equation (6.73) contains discontinuous and singular
functions, therefore equation (6.73) must be treated in terms of distributions.
Nevertheless, let us show that, on the manifold of periodic solutions, equation
(6.73) is equivalent to some classic boundary-value problem.
Let us represent the output in the form
d3 y
= (Y + X e + Y e )ω 3 (6.76)
dt3
d4 y
= (X (4) + Y (4) e + X e + Y e )ω 4
dt4
The fourth-order derivative in (6.76) takes into account the equality ee = 0,
which easily follows from (6.53) in the symmetric case β = 0.
Substituting (6.75) and (6.76) in (6.73), and considering {1, e, e, e } as a
linearly independent basis, gives equations
a4 ω 4 X IV + a3 ω 3 Y + a2 ω 2 X + a1 ωY + a0 X = 0 (6.77)
a4 ω 4 Y IV + a3 ω 3 X + a2 ω 2 Y + a1 ωX + a0 Y = b0
Y = 0, X = 0 (6.78)
b2 1 a3
ω 2 Y = , ω 3 X = b1 − b2
a4 a4 a4
In contrast to equation (6.73), the boundary value problem (6.77) and (6.78)
does not include discontinuous terms any more.
6.6 Piecewise-Linear Oscillators with Asymmetric Characteristics 165
2 α
α
j βj j βj
X= Aj cosh τ sin τ + Bj sinh τ cos τ (6.79)
j=1
ω ω ω ω
2 α
α
j βj j βj b0
Y = Aj sinh τ sin τ + Bj cosh τ cos τ +
j=1
ω ω ω ω a 0
a4 p4 + ... + a1 p + a0 = 0 (6.80)
The assumption that both of the roots are complex reflects the physical mean-
ing of the example, however other cases would lead to even less complicated
expressions.
Finally, substituting (6.79) in (6.78) gives a linear algebraic set of four
independent equations with respect to four constants: A1 , A2 , B1 and B2 .
Although the corresponding analytical solution is easy to obtain by using the
R
standard Mathematica commands, the result is somewhat complicated for
reproduction. Practically, it may be reasonable to determine the constants by
setting the system parameters to their numerical values moreover that only
numerical solution are often possible for characteristic equations.
where H(q) is Heaviside unit-step function, m and k are mass and stiffness
parameters, respectively, and |ε|
1.
Therefore, k− = k and k+ = k(1 − ε) are elastic stiffness of the oscillator
for q < 0 and q > 0, respectively.
The exact general solution of oscillator (6.81) can be obtained by satisfying
the continuity conditions for q and q̇ at the matching point q = 0, where the
characteristic has a break. Such approaches are often facing quite challenging
algebraic problems, however, as the number of degrees of freedom increases
or external forces are involved. This is mainly due to the fact that times of
crossing the point q = 0 are a priory unknown.
In this section, it will be shown that, applying a combination of asymp-
totic expansions with respect to ε and nonsmooth temporal transformations,
gives a unit-form solution for oscillator (6.81) with a possibility of general-
ization on the normal mode motions of multiple degrees-of-freedom systems.
In particular, the nonsmooth temporal transformation:
1) provides an automatic matching the motions from different subspaces
of constant stiffness, and
2) justifies quasi-linear asymptotic solutions for the specific nonsmooth
case of piece-wise linear characteristics.
Let us clarify the above two remarks. Introducing the notation ω 2 = k/m,
brings equation (6.81) to the standard form of a weakly non-linear oscillator
q̈ + ω 2 q = εω 2 H(q)q (6.82)
q = A cos ϕ
q̇ = −ωA sin ϕ (6.83)
The right-hand sides of equations (6.85) are 2π-periodic with respect to the
phase variable, ϕ. Therefore, nonsmooth transformation of the phase variable
applies through the couple of functions
Substituting (6.90) in (6.88) and taking into account (6.89), gives boundary-
value problem
(X − Y ) = 0
(X + Y ) επ sin πτ
=− (6.91)
X +Y 4 1 − ε cos2 πτ
2
Y |τ =±1 = 0 (6.92)
where ≡ d/dτ .
Solution of the boundary value problem, (6.91) and (6.92), is obtained by
elementary integration. Then representation (6.90) gives
where functions Fi (φ) are such that the new phase variable also has a constant
temporal rate even though ε = 0.
In other words, transformation (6.98) should bring equation (6.97) to the
form
φ̇ = ω(1 + εγ1 + ε2 γ2 + ...) (6.99)
where γi are constant coefficients to be determined together with Fi (φ) during
the asymptotic procedure.
Note that the procedure, which is described below, has several specific
features due to the presence of nonsmooth periodic functions. In particu-
lar, high-order approximations require a non-conventional interpretation for
power series expansions; see the next subsection for the related remarks.
Other modifications occur already in the leading order approximation.
Substituting (6.98) into equation (6.97), then enforcing equation (6.99)
and collecting the terms of order ε, gives
where the triangular and rectangular waves depend now on the new phase
variable φ as τ = τ (2φ/π) and e = e(2φ/π), respectively.
According to the conventional averaging procedure, the constant γ1 is se-
lected to achieve zero mean on the right-hand side of equation (6.100) and
thus provide periodicity of solution, F1 (φ). In the algorithm below, the peri-
odicity is due to the form of representation for periodic solutions, whereas the
operator of averaging occurs automatically from the corresponding conditions
of smoothness that is boundary conditions for the solution components.
170 6 NSTT for Linear and Piecewise-Linear Systems
Substituting (6.101) in (6.100) and following the NSTT procedure, gives the
boundary-value problem
π
U1 (τ ) = M (τ )
2
π
V1 (τ ) = [G(τ ) − γ1 ] (6.102)
2
V1 (±1) = 0
Note that there are two conditions on the function V1 (τ ) described by the
first-order differential equation in (6.102). However, there is a choice for γ1 ,
which is to satisfy one of the two conditions. As a result, solution of boundary-
value problem (6.102) is obtained by integration in the form
π τ
U1 (τ ) = M (z)dz
2 0
π τ
V1 (τ ) = [G(z) − γ1 ]dz (6.103)
2 −1
1 1
γ1 = G(τ )dτ
2 −1
where
2
M2 (τ ) = [U1 (τ )G (τ ) + V1 (τ )M (τ )] − M (τ )γ1
π
2
G2 (τ ) = U1 (τ )M (τ ) − G(τ )γ1 + γ12 (6.105)
π
2
P2 (τ ) = U1 (τ )M (τ )
π
e ≡ de(2φ/π)/d(2φ/π)
Now, we return to the illustrating model. In particular case (6.89), one has
1 πτ
G(τ ) ≡ M (τ ) ≡ − cos2 (6.109)
2 2
and
G(±1) = M (±1) = 0
G (±1) = M (±1) = 0 (6.110)
G (±1) = M (±1) = −π 2 /4
where ≡ d/dτ .
First two lines of conditions (6.110) provide continuity for the right hand
side of (6.97) and its first derivative at those ϕ where τ (2ϕ/π) = ±1. As
follows from (6.105), for this class of smoothness one has P2 (±1) = 0 and
thus no singular terms occur in the first two steps of asymptotic procedure.
Finally, taking into account (6.109) and (6.110) and conducting integration
in (6.103) and (6.108), brings solution (6.98) to the form (6.94) and (6.95).
172 6 NSTT for Linear and Piecewise-Linear Systems
q
0
-1
-2
0 10 20 30 40
t
Fig. 6.11 Second-order asymptotic and numerical solutions shown by solid and
dashed lines, respectively
6.7 Multiple Degrees-of-Freedom Case 173
Fig. 6.11 compares analytical solution (6.83), (6.93) and (6.94) shown by
the solid line and numerical solution shown by the dashed line. As expected,
the amplitude show the perfect match, whereas some phase shift develops
after several cycles.
where ωj and Aj are arbitrary eigen frequency and eigen vector (normal
mode) of the linearized system:
Substituting (6.114) in (6.113), taking into account (6.87) and (6.115) in the
first order of ε, gives
d2 x(1) 1 1 πτ
ωj2 M 2
+ Kx (1)
= BAj + ( BAj + γ (1)
KAj )e cos (6.116)
dϕ 2 2 2
and taking into account M -orthogonality of the set of eigen vectors, gives
2
2ωj πτ
Xi + ωi2 Xi = βij cos , Xi |τ =±1 = 0 (6.121)
π 2
2
2ωj πτ
Yi + ωi2 Yi = (βij + γ (1) κij ) cos , Yi |τ =±1 = 0 (6.122)
π 2
where
1 Ai BAj Ai KAj
βij = , κij = (6.123)
2 Ai M Ai Ai M Ai
are dimensionless coefficients.
Note that despite of the similar representation for solution (6.114),
there is a noticeable difference between the classic Poincare-Lindshtedt
method and current procedure due to (6.117). Namely, according to the
6.7 Multiple Degrees-of-Freedom Case 175
In the particular case i = j, problem (6.121) still has a solution, but problem
(6.122) generally speaking does not. Fortunately, in this case, we have κjj = 0
and thus the problem is set to have the trivial solution by condition
βjj
γ (1) = − (6.126)
κjj
Therefore,
πβjj πτ 2 πτ
Xj = τ sin + cos (6.127)
4ωj2 2 π 2
Yj = 0 (6.128)
In this case, the first-order asymptotic solution for the in-phase (j = 1) and
out-of-phase (j = 2) takes the form, respectively,
176 6 NSTT for Linear and Piecewise-Linear Systems
πτ επ 2 πτ πτ
x1 = e cos + cos + τ sin + O(ε2 )
2 16 π 2 2
πτ εk1 πτ πτ
x2 = e cos − e cos + cos (6.130)
2 8k2 2 2
−1/2
k2 k2 πτ k2 π
− 1+2 cos 1+2 / sin 1+2 + O(ε2 )
k1 k1 2 k1 2
k1 ε
ϕ= 1 − + O(ε2 )t
m 4
and
πτ εk1 πτ πτ k2
x1 = −e cos + e cos + cos − 1+2
2 8k2 2 2 k1
πτ k2 π k2
× cos / 1+2 / sin / 1+2 + O(ε2 ) (6.131)
2 k1 2 k1
πτ εk1 π 2 πτ πτ
x2 = e cos + cos + τ sin + O(ε2 )
2 16(k1 + 2k2 ) π 2 2
*
k1 + 2k2 εk1
ϕ= 1− + O(ε2 )t
m 4(k1 + 2k2 )
where it is assumed that m1 = m2 = m. Solutions (6.130) and (6.100) show
that a bi-linearity may have quite different effect on different modes. In par-
ticular, solution (6.130) reveals the possibility of internal resonances, when
πω2 ω2 k2
sin = 0, = 1+2 (6.132)
2ω1 ω1 k1
If, for instance, the system is close to the frequency ratio ω2 /ω1 = 2 then the
in phase mode may be affected significantly by a crack even under very small
magnitudes of the parameter ε. In contrary, solution (6.100) has the denom-
inator sin[(π/2)ω1 /ω2 ], which is never close to zero because 0 < ω1 /ω2 < 1.
Therefore, in current asymptotic approximation, the influence of crack on the
out-of phase mode is always of order ε provided that k2 /k1 = O(1).
The influence of the bilinear stiffness on inphase mode trajectories in the
closed to internal resonance case is seen from Fig. 6.13, where both analytical
and numerical solutions are shown for comparison reasons. The frequency ra-
tio ω2 /ω1 = 2.0025 is achieved by conditioning the spring stiffness parameters
as follows k2 = (3/2)k1 + 0.005.
6.8 The Amplitude-Phase Problem in the Idempotent Basis 177
1.0
0.5
Ε 0.01 Ε0
x2 0.0
0.5
1.0
1.0 0.5 0.0 0.5 1.0
x1
Fig. 6.13 The influence of a small “crack” ε = 0.01 on the in-phase mode tra-
jectory near the frequency ratio ω2 /ω1 = 2; the dashed line shows the numerical
solution, and the thin solid line corresponds to the perfect (linear) case ε = 0.
or
πτ π
(1 − ε cos2 )X+ = − εX+ sin πτ
2 4
X− =0 (6.136)
de (t)
v̇ + εv 3 = p (7.1)
dt
under the initial condition
v |t=0 = v 0 (7.2)
Substituting (7.6) into (7.4) and (7.5) and matching coefficients of the same
powers of ε, gives a sequence of simplified boundary value problems. In par-
ticular, zero order problem is
∂X0 ∂Y0
= 0, =0
∂τ ∂τ
7.1 Nonsmooth Two Variables Method 181
Y0 |τ =±1 = p
∂X1
= − 3pA20 + p3
∂τ
∂Y1 dA0
=− 3
+ A0 + 3A0 p 2
(7.8)
∂τ dt0
Y1 | τ =±1 = 0
dA0
+ A30 + 3A0 p2 = 0 (7.10)
dt0
Therefore Y1 ≡ 0, whereas function A1 t0 in the expression for X1 remains
unknown.
Equation (7.10) admits exact solution in the following explicit form
1 2 0 −1/2
A0 = − 2 + const · exp 6p t
3p
Finally, the leading order asymptotic solution of the Caushy problem is writ-
ten as
1 1 0 −2 2 −1/2
v(t) = ± − 2 + + v − p exp 6p εt + pe (t) + O (ε)
3p 3p2
(7.11)
where plus or minus sign are taken if v 0 − p > 0 or v 0 − p < 0, respectively.
Substituting asymptotic solution (7.11) in equation (7.1), leads to the er-
ror εR(εt)e (t), where R(εt) is a function of slow time scale. Although the
error is of the same order of magnitude as the perturbation, it relates to the
singular component of the equation with a scaling factor of order one. In
other words, for such kind of systems, regular and singular error components
182 7 Periodic and Transient Nonlinear Dynamics under Discontinuous Loading
0.8
0.6
v
0.4
0.2
0.0
0 10 20 30 40
t
Fig. 7.1 Velocity of the nonlinearly damped material point under the pulsed pe-
riodic excitation.
should be considered separately. Note that the first term in expression (7.11)
is a transient response of the system that vanishes as time increases; see
Fig. 7.1 for illustration.
In conclusion, it is seen from solution (7.11) that the amplitude of impulses
may have a strong effect on the time scale of transient response of the system.
Y |τ =±1 = 0 (7.16)
one obtains
∂Y ∂X ∂X ∂Y
ẋ = ω +ε 0 + ω +ε 0 e (7.17)
∂τ ∂t ∂τ ∂t
Then, substituting (7.15) and (7.17) in (7.14), eventually gives the boundary
value problem (7.16),
∂X
ω2 |τ =±1 = εp (7.18)
∂τ
and
∂2X ∂2Y 2 nπ 2
2∂ X
ω 2 2 + 2εω + ε + ω X
∂τ
∂τ ∂t0 ∂t02 2
= −ε σX + X + 3Y X
3 2
∂2Y ∂2X 2 nπ 2
2∂ Y
ω 2 2 + 2εω + ε + ω Y (7.19)
∂τ
∂τ ∂t0 ∂t02 2
= −ε σY + Y 3 + 3X 2 Y
Note that, due to substitution (7.15), the boundary value problem, (7.16),
(7.18) and (7.19), includes no singular functions. Therefore, further averaging
procedures can be correctly applied.
Following the formalism of two variable expansions, we seek asymptotic
solutions in the form of power series
X = X0 τ, t0 + εX1 τ, t0 + ε2 X2 τ, t0 + ...
Y = Y0 τ, t0 + εY1 τ, t0 + ε2 Y2 τ, t0 + ... (7.20)
∂ 2 X0 nπ 2 ∂X0
+ X0 = 0, |τ =±1 = 0
∂τ 2 2 ∂τ
∂ 2 Y0 nπ 2
+ Y0 = 0, Y0 |τ =±1 = 0 (7.21)
∂τ 2 2
These equations and the boundary conditions are satisfies at any t0 by
solution
184 7 Periodic and Transient Nonlinear Dynamics under Discontinuous Loading
nπ nπ
X0 = A0 t0 sin τ, Y0 = D0 t0 cos τ (7.22)
2 2
where A0 t0 and D0 t0 are arbitrary functions of the slow time scale to
be determined on the next step of iteration.
So, collecting all terms of order ε, gives the boundary value problem of the
next asymptotic order as
2 nπ 2
2 ∂ X1 ∂ 2 Y0
ω + X1 = − 2ω 3 2
+ σX0 + X0 + 3Y0 X0
∂τ 2 2 ∂τ ∂t0
∂X1
ω2 |τ =±1 = p (7.23)
∂τ
∂ 2 Y1 nπ 2 ∂ 2 X0
ω2 + Y1 = − 2ω + σY0 + Y0
3
+ 3X 2
0 Y0
∂τ 2 2 ∂τ ∂t0
Y1 |τ =±1 = 0
and
A0 3 2 2
2p nπ
D0 = σ+ A0 + D0 + sin (7.26)
nπω 4 nπω 2
Then, the boundary condition for Y1 gives
C1 t0 ≡ 0
and
D0 3 2
A0 = − σ+ A0 + D02 (7.27)
nπω 4
In current approximation, another
two arbitrary
functions of the slow time
scale can be chosen as A1 t0 ≡ 0 and D1 t0 ≡ 0, at least, because the
same type of terms are included already into the generating solution (7.22).
Substituting derivatives from (7.26) and (7.27) in (8.24) and (7.25) and taking
into account (7.22), gives
nπτ ε nπτ
X = A0 sin 1 − 2 2 2 A20 − 3D02 cos2
2 2n π ω 2
2εp nπ nπτ nπτ
− 2 2 2 sin nπτ cos − sin + O(ε2 ) (7.28)
n π ω 2 2 2
nπτ ε nπτ
Y = D0 cos 1 + 2 2 2 3A20 − D02 sin2 + O(ε2 ) (7.29)
2 2n π ω 2
Substituting (7.28) and (8.87) in (7.15), finally gives the closed form approx-
imate solution x(t), despite of the discontinuous impulsive loading.
Figs. 7.2, 7.3, and 7.4 illustrate the results of calculations under the fol-
lowing parameters and initial conditions: n = 1, ω = 1.0, σ = 0.1, p = 1.0,
ε = 0.1, A0 (0) = 0.1, D0 (0) = 0.0. The coordinate-velocity diagram shows
two velocity jumps per one cycle due to the periodic impulsive excitation.
Let us consider the case of strongly nonlinear exactly solvable oscillator de-
scribed in Chapter 3 by adding periodic impulsive loading on the right-hand
side as follows
F de (t/a)
ẍ + tan x + tan3 x =
a d (t/a)
∞
= 2F (−1)k δ[t − (2k − 1)a] (7.30)
k=−∞
186 7 Periodic and Transient Nonlinear Dynamics under Discontinuous Loading
1.5
1.0
B0
0.5
0.0
0.5
A0
1.0
1.5
2.0
0 5 10 15 20 25
0
t
Fig. 7.2 The amplitudes A0 (t0 ) and D0 (t0 ) in the slow time scale, t0 = εt, shown
by the gray and black lines, respectively.
x 0
1
2
0 50 100 150 200 250
t
Fig. 7.3 The time history of the oscillator near the principal resonance.
Here 2F > 0 and 4a = T > 0 characterize the amplitude and period of the
impulsive loading.
Making substitutions x = X(τ ) and τ = τ (t/a) in equation (7.30), gives
d2 X dX de (t/a)
a−2 2 + tan X + tan3 X = a−2 aF −
dτ dτ d (t/a)
7.3 Strongly Nonlinear Oscillator under Periodic Pulses 187
v 0
1
2
3
2 1 0 1 2
x
Fig. 7.4 The coordinate-velocity plane trajectory on the interval 0 < t < 40.
or
d2 X
a−2 + tan X + tan3 X = 0 (7.31)
dτ 2
under the following continuity condition for x(t),
dX
|τ =±1 = aF (7.32)
dτ
Boundary value problem (7.31) and (7.32) describe the class of steady state
periodic motions of the period T .
With reference to Chapter 3, equation (7.31) has exact solution of the form
35
30
25
20
T
15
k3
10
k2
5 k1
0 k0
1.1 1.2 1.3 1.4 1.5
A
Fig. 7.5 First few branches of the period versus amplitude diagram under the fixed
pulse amplitudes, F = 1.8.
It can be shown that solutions (7.34) and (7.35) exist in the interval
1 π
arccos √ = Amin < A < Amax = (7.36)
1+F 2 2
Fig. 7.5 illustrates the sequence of branches of solutions (7.34) and (7.35) at
different numbers k, and the parameter F = 1.8. The diagram gives such
combinations of the period and amplitude at which oscillator (7.30) can have
periodic motions with the period of external impulses, T = 4a. The upper and
7.4 Impact Oscillators under Impulsive Loading 189
lower branches of each loop correspond to plus and minus signs in expression
(7.34), respectively. Solution (7.35) corresponds to the number k = 0 and
has the only upper branch. For the selected magnitude of F , the minimal
amplitude is found to be Amin = 1.0637, which corresponds to the left edges
of the amplitude-period loops in Fig. 7.5. As a result, further slight increase
of the amplitude is accompanied by bifurcation of the solutions as shown
in Fig.7.6. The diagram includes only first three couples of new solutions
(k = 1, 2, 3) from the infinite set of solutions. Further evolution of temporal
shapes of solutions, related to the amplitude increase, is illustrated by Figs.
7.7 and 7.8. As follows from these diagrams, the influence of external pulses
on the temporal shapes is decreasing as the amplitude grows. This is the
result of dominating the restoring force over the external pulses. When the
amplitude becomes close to its maximum Amax = π/2, the oscillator itself
generates high-frequency impacts; see Chapter 3 for details.
where S (l) ≡ |l| in case (7.38) or S (l) ≡ τ (l) in case (7.39), so that the
new coordinate l belongs to the entire infinite interval, l (t) ∈ (−∞, ∞), in
contrast to (7.38) or (7.39).
Taking into account the identity [S (l)]2 = 1, gives the final result of
transformation (7.40) in the form
d2 l dl de (ωt)
2
+ 2ζ + S (l) f [S (l) , ωt] = pS (l) (7.41)
dt dt d (ωt)
where, as mentioned above, the new coordinate l (t) is free of any constraints.
190 7 Periodic and Transient Nonlinear Dynamics under Discontinuous Loading
x x
1.0 k1 1.0
0.5 0.5
t t
1 2 3 4 5 1 2 3 4 5 6
0.5 0.5
1.0 1.0
x k2 x
1.0 1.0
0.5 0.5
t t
2 4 6 8 10 12 2 4 6 8 10 12
0.5 0.5
1.0 1.0
x k3
x
1.0 1.0
0.5 0.5
t t
5 10 15 5 10 15
0.5 0.5
1.0 1.0
Fig. 7.6 Periodic solutions corresponding to the first three closed loops of the
diagram shown in Fig. 7.5 at A = 1.0647; the left and right columns relate to lower
and upper branches of the loops, respectively.
Note that the right-hand side of equation (7.41) makes sense within the
distribution theory under the condition that times of interaction with con-
straints (7.38) or (7.39) never coincide with the external pulse times.
Consider now the class of periodic motions of the period of external load-
ing, T = 4/ω. As follows from Chapter 6 and the previous sections of this
chapter, within such class of motions, the external pulses are eliminated by
introducing the new time argument as follows
t −→ τ : τ = τ (ωt) , l = X (τ ) + Y (τ ) e (7.42)
7.4 Impact Oscillators under Impulsive Loading 191
x x
1.0 1.0
k1
0.5 0.5
t t
0.5 1.0 1.5 2.0 2.5 3.0 1 2 3 4 5 6 7
0.5 0.5
1.0 1.0
x x
k2
1.0 1.0
0.5 0.5
t t
2 4 6 8 2 4 6 8 10 12
0.5 0.5
1.0 1.0
x x
k3
1.0 1.0
0.5 0.5
t t
2 4 6 8 10 12 5 10 15
0.5 0.5
1.0 1.0
ω 2 X (τ ) + 2ωζY (τ ) + R (X, Y, τ )
+ ω 2 Y (τ ) + 2ωζX (τ ) + I (X, Y, τ ) e (7.43)
de (ωt)
= pRS (X, Y ) + pIS (X, Y ) e − ω 2 X (τ )
d (ωt)
where
R 1
= {S (X + Y ) f [S (X + Y ) , τ ] ± S (X − Y ) f [S (X − Y ) , 2 − τ ]}
I 2
192 7 Periodic and Transient Nonlinear Dynamics under Discontinuous Loading
x x
k1
1.0 1.0
0.5 0.5
t t
0.5 1.0 1.5 1 2 3 4
0.5 0.5
1.0 1.0
x x
k2
1.0 1.0
0.5 0.5
t t
1 2 3 4 5 2 4 6 8
0.5 0.5
1.0 1.0
x x
k3
1.0 1.0
0.5 0.5
t t
2 4 6 8 2 4 6 8 10
0.5 0.5
1.0 1.0
and
RS 1
= {S (X + Y ) ± S (X − Y )}
IS 2
under the necessary condition of continuity for l (t),
Y (±1) = 0 (7.44)
ω 2 X (τ ) + 2ωζY (τ ) + R (X, Y, τ ) = 0
ω 2 Y (τ ) + 2ωζX (τ ) + I (X, Y, τ ) = 0 (7.45)
x (t) = S (X (τ ) + Y (τ ) e) (7.47)
ẋ = f (x) (8.1)
where f (x) is a continuous vector-function, and the over dot indicates time
derivative.
We consider the class of periodic motions of the period T = 4a. Note that,
in the autonomous case, the period is a priori unknown. Periodic solutions
usually require specific, a priory unknown, initial conditions. Practically, how-
ever, such kind of the specific initial conditions are determined in a backward
way, after some periodic family of solutions is obtained under the assumption
of periodicity. In our case, the assumption of periodicity is imposed automat-
ically by the form of representation for periodic solutions
x = X(τ ) + Y (τ )e (8.2)
where τ = τ (t/a) and e = e(t/a) are the standard triangular sine and rect-
angular cosine, respectively, and X(τ ) and Y (τ ) are unknown components of
the solution.
Substituting (8.2) in (8.1), gives
(Y − aRf ) + (X − aIf )e + Y e = 0
where
1
Rf = Rf (X, Y ) = [f (X + Y ) + f (X − Y )]
2
1
If = If (X, Y ) = [f (X + Y ) − f (X − Y )]
2
Eliminating the periodic singular term e = de(t/a)/d(t/a) by means of the
boundary condition for Y (τ ), gives the non-linear boundary value problem
on the standard interval, −1 ≤ τ ≤ 1,
Y = aRf (X, Y )
X = aIf (X, Y ) (8.3)
Y |τ =±1 = 0
U (τ ) = X(τ ) + Y (τ )
V (τ ) = X(τ ) − Y (τ )
brings the boundary value problem, (8.7) and (8.8), to the form, in which
the differential equations are decoupled at cost of coupling the boundary
conditions though,
U + a2 f (U, U /a, 2a − aτ ) = 0
V + a2 f (V, − V /a, 2a − aτ ) = 0 (8.9)
U + V |τ =±1 = 0
U − V |τ =±1 = 0
In case when analytical methods are applied, the differential equations (8.9)
for U (τ ) and V (τ ) can be usually solved in a similar way. Nevertheless, the
previous boundary value problem, (8.7) and (8.8), may appear to have some
advantages for analyses. For instance, the problem may admit families of
solutions with either Y (τ ) ≡ 0 or X(τ ) ≡ 0.
Secondly, due to substitution (8.2), the major qualitative property of solu-
tions, such as periodicity, is a priory captured by the new argument, τ . As a
result, the following simplified system can be employed as a generating model
for analytical algorithms of successive approximations
X = 0 (8.10)
Y =0
ẍ + f (x) = 0 (8.12)
Based on the conditions assumed, the boundary value problem (8.7) and (8.8)
is reduced to the following one
X + a2 f (X) = 0 (8.14)
X |τ =1 = 0
8.3 Periodic Solutions of Conservative Systems 199
We seek solutions of the boundary value problem (8.14) in the form of series
of successive approximations
X = X 0 (τ ) + X 1 (τ ) + X 2 (τ ) + ... (8.15)
a2 = h0 (1 + γ1 + γ2 + ...) (8.16)
In order to organize the corresponding iterative procedure, it is assumed that
O(
X i
) O(
X i+1
)
O(γi+1 ) O(γi+2 ) (8.17)
(i = 0, 1, 2, ...)
X 0 = 0 (8.18)
X 0 = A0 τ (8.23)
A0T A0
h0 = (8.26)
"1
A0T f (A0 τ )dτ
0
1
2
A = h0 [γ1 f (A0 τ ) + fx (A0 τ )X 1 (τ )]dτ (8.28)
0
A0T A2 = 0 (8.29)
This condition means that the vector A2 must be orthogonal to the corre-
sponding vector of the generating solution A0 in order to keep the amplitude
fixed.
Substituting (8.28) in (8.29), gives
"1
A0T fx (A0 τ )X 1 dτ
0
γ1 = − (8.30)
"1
A0T f (A0 τ )dτ
0
This completes the second step of successive approximations. All the further
steps can be passed in the same way.
In general terms, convergence properties of the above procedure are due
to the following integral operator
⎧ 1 ⎫
⎨ τ ⎬
F [X] ≡ a2 τ f (X(ξ))dξ + ξf (X(ξ))dξ (8.31)
⎩ ⎭
τ 0
where
"1
A0T f (A0 τ )dτ
0
a2 = h 0 (8.32)
"1
A0T f (X)dτ
0
Based on definition (8.31), the original boundary value problem (8.14) admits
representation in the form X = F [X]. Therefore, the convergence condition
is
FX [X 0 ]δX
<1 (8.33)
δX
In the case of linearized system, condition (8.33) leads to the set of in-
equalities ωi /ωj < 1 for all i = j, where ωj is the eigen frequency of the
linear normal mode, which is chosen to be a generating solution.
X = X0 (τ ) + X1 (τ ) + X2 (τ ) + ...
a2 = h0 + h1 + h2 + ... (8.34)
X0 = Aτ (8.35)
τ
X1 = −h0 (τ − ξ)f (Aξ)dξ
0
1
h0 = A/ f (Aξ)dξ (8.36)
0
All the next steps of the procedure are passed then in a similar way based
on relationships
i τ
Xi = − hj−1 (τ − ξ)Ri−j dξ
j=1 0
i−1
hi−1 = − αi−j hj−1 (8.37)
j=1
(i = 2, 3, ...)
8.3 Periodic Solutions of Conservative Systems 203
where the coefficients and integrands are generated by means of the formal
auxiliary parameter, ε, as follows
1 1
αi = Ri dξ/ R0 dξ (8.38)
0 0
1 di f (X 0 + εX 1 + ε2 X 2 + · · ·)
Ri = |ε=0
i! dεi
(i = 0, 1, 2, . . .)
ẍ + xm = 0
Fig. 8.1 The first three terms of iteration (thin lines) and their sum (solid line)
for the temporal mode shape of the oscillator ẍ + x5 = 0.
5
Period, T
harmonic balance
4
exact solution
3
sawtooth
0 10 20 30 40 50
Degree of nonlinearity, m
Fig. 8.2 The period of the oscillator with power characteristic at different expo-
nents m obtained by three different methods under the total energy E = 2.
period than the harmonic balance as the exponent m increases. Note that the
entire series (8.39) and (8.40) are not asymptotic with respect to m or 1/m
in the sense of Poincare, however the series perfectly capture the asymptotic
of impact oscillator as m → ∞.
Here, m is mass, k is the linear stiffness of each spring, l is the length of each
spring at the equilibrium position at which the springs are horizontal, and w
is the particle vertical coordinate.
In terms of the dimensionless coordinate x = w/l and phase ϕ =
(2k/m)1/2 t, the corresponding differential equation of motion takes the form
d2 x x
+x− √ =0 (8.42)
dϕ2 1 + x2
Then, applying substitution (8.13) as x = X(τ ) and τ = τ (ϕ/a), leads to the
boundary problem
X
X = −h X − √ ≡ −hf (X)
1 + X2
X |τ =1 = 0 (8.43)
where h = a2 .
wt
l l
Fig. 8.3 The system which becomes weakly non-linear at large amplitudes.
206 8 Strongly Nonlinear Vibrations
X0 = Aτ
√ −1
1 1 + A2 − 1
h0 = −
2 A2
and
h0 1
X1 (τ ) = − 2 (Aτ ) + 2Aτ − Aτ 1 + (Aτ ) − arcsinh(Aτ ) (8.44)
3 2
2A 3
h0 6A + A3 9A − A3 1
γ1 = 3 + √ − 1+ √ arcsinh(A)
A 12 6 1 + A2 1 + A2
Interestingly enough, this model is essentially nonlinear at small amplitudes,
but it becomes linear as the amplitudes are infinitely large. Indeed, taking
the corresponding limits, shows that
X1 1 3
→ − τ 5 , h0 A2 → 8, γ1 → as A → 0 (8.45)
A 5 10
and
X1 1 1
→ − τ 3 , h0 → 2, γ1 → as A → ∞ (8.46)
A 3 6
The asymptotic (8.45) obviously corresponds the nonlinear oscillator, whereas
the limit case (8.46) associates with the harmonic oscillator. Nevertheless,
solution (8.44) is valid for both large and small amplitudes. Both limit cases
follow from equation (8.42). In the case of small amplitudes, |x|
1, by using
the estimate (1 + x2 )−1/2 ∼ 1 − x2 /2 in equation (8.42), we obtain
d2 x/dϕ2 + x3 /2 = 0 (8.47)
In the case of large amplitudes, it follows even from Fig. 8.3 that the distance
between the spring fixed ends becomes negligible if as compared to l. As
a result, the mechanical model becomes effectively close to a mass-spring
oscillator of mass m with a single spring of stiffness 2k. In terms of the
differential equations of motion, we also obtain the corresponding limit from
exact equation (8.42) by√assuming that, during ‘most of the time of vibration
cycle,’ |x| 1 and thus 1 + x2 ∼ |x|. As a result, equation (8.42) is replaced
by
d2 x/dϕ2 + x − sgn(x) = 0 (8.48)
where the term sgn(x) has to be neglected due to the same condition |x| 1
that gives the standard linear oscillator.
Alternatively, the discontinuous term in equation (8.48) can be saved and
then considered as a perturbation of the harmonic oscillator. Note that equa-
tion (8.48) admits another form as follows
8.3 Periodic Solutions of Conservative Systems 207
Although force (8.50) has no certain value at the point x = 0, this still can
play the role of equilibrium position. From the physical standpoint, this is
equilibrium of a small bead at the bottom of V -shaped potential well. The
local dynamics in a small neighborhood of such type of equilibria is considered
later in this chapter; see the text to Fig. 8.13.
It follows from (8.50) that f (−x) = −f (x). Therefore, periodic motions of
the corresponding oscillator can be described by the function x =
X(τ ), where X(−τ ) = −X(τ ) and τ = τ (t/a). In terms of these NSTT
variables, the oscillator of a unit mass is described by the boundary value
problem
where h = a2 .
This problem is exactly solvable, and solution that satisfies the continuity
of state condition at τ = 0 has the form
X± (τ ) ≡ Aτ ± 2 ln (1 + exp(−A)) (8.52)
h
∓2 ln 1 + exp(±Aτ − A)
2h0
and
1 + exp(−A)
X = sgn(τ ) A|τ | + 2 ln (8.55)
1 + (h/(2h0 )) exp(A|τ | − A)
X |τ =1 = 0 (8.56)
or
−1
h h
1− 1+ =0 (8.57)
h0 2h0
This exactly solvable case can play the role of a majorant for evaluation of
convergence properties of successive approximations. For that reason, we in-
troduce a formal ‘small’ parameter, ε = 1, and represent solution of equation
(8.57) in the form
ε −1
h = 2h0 = εh0 1 − (8.58)
2
Taking into account (8.58), brings (8.55) to the form
1 − (ε/2) (1 − exp(−A))
X = sgn(τ ) A|τ | + 2 ln (8.59)
1 − (ε/2) (1 − exp(A|τ | − A))
εh0
= εh0 (1 + εγ1 + ε2 γ2 − ...) (8.61)
1 − ελ1 − ε2 λ2 − ...
This is equivalent to
Taking the product of series on the left-hand side of (8.62) and consider-
ing different orders of ε, generates a sequence of equations for the coefficients
8.3 Periodic Solutions of Conservative Systems 209
εh0
h= (8.63)
1 − εγ1 − ε2 (γ2 − γ12 ) − ...
In many cases, expansion (8.63) appears to be more effective than (8.16). Note
that it is also possible to organize the successive approximations procedure
by using expansion (8.60) instead of (8.16).
ẍ + f (x) = 0 (8.64)
It was shown in this chapter that periodic solutions of the oscillator (8.64)
admit the form x = X(τ ), where
Based on the transition from equation (8.64) to equation (8.67) and the
general symmetry properties (8.65) and (8.66), we introduce the following
representation for periodic solutions of equation (8.67)
where τ = τ (t/a) and the expansion (8.40) for a2 requires only the replace-
ment m → α. Expansion (8.69) represents an approximate solution of the
equation
ẍ + sgn(x)|x|α = 0 (8.70)
where the notation α substitutes m in order to emphasize that the new expo-
nent can take any positive real value, such as even, odd, rational or irrational.
Figs. 8.4 and 8.5 illustrate solution (8.69) compared to numerical solution for
two different exponents α and the same parameter A = 1. As both figures
show, the analytical and numerical solutions are in a better match under the
large exponent α due to the influence of vibroimpact asymptotic, α → ∞.
0.6
3 2
0.4
0.2
x 0.0
0.2
0.4
0.6
0 1 2 3 4 5 6 7
t
Fig. 8.4 Analytical and numerical solutions of the modified oscillator shown by
continuous and dashed lines respectively.
8.4 Periodic Motions Close to Separatrix Loop 211
1.0
2010
0.5
x 0.0
0.5
1.0
0 5 10 15 20 25 30 35
t
Fig. 8.5 Analytical and numerical solutions of the generalized oscillator under the
large exponent α.
The common feature of the algorithms and examples of this section is that
generating solutions for successive approximations are represented by trian-
gular sine waves of proper amplitudes and periods. Such generating solutions
belong to the ‘real’ component of the hyperbolic ‘number,’ x = X + Y e. In
contrast, the next section introduces algorithms of successive approximations
based on the ‘imaginary’ component. It will be seen that these two approaches
have different physical contents.
ẍ + sin x = 0 (8.71)
It is assumed that the pendulum oscillates inside the separatrix loop around
the stable equilibrium (x, ẋ) = (0, 0) in between two physically identical
unstable saddle-points (x, ẋ) = (±π, 0). The separatrix loop also represents
trajectory of the system with the total energy
π
Es ≡ sin xdx = 2 (8.72)
0
212 8 Strongly Nonlinear Vibrations
The pendulum remains close to the separatrix loop in the area of periodic
motions if the total energy E is within the range
E
0<1− << 1 (8.73)
Es
Let us show that, under condition (8.73), a successive approximation solution
can be derived from the particular case of boundary value problem (8.7) and
(8.8). Such particular case is given by setting X ≡ 0 so that
d2 Y1
− p2 Y1 = 0 (8.79)
dτ 2
d2 Y3 1 3
− p Y3 = p λ2 Y1 − Y1
2 2
(8.80)
dτ 2 6
d2 Y5 2 λ2 3 1 5 1 2
− p 2
Y5 = p 2
λ2 + λ4 Y1 − Y + Y + λ Y
2 3 − Y Y3 (8.81)
dτ 2 6 1 120 1 2 1
...
8.4 Periodic Motions Close to Separatrix Loop 213
A2
λ2 = (8.83)
8 cosh2 p
A3 cosh 3pτ
Y3 = (8.84)
192 cosh3 p
√
where τ = τ (t/a) and a = h.
214 8 Strongly Nonlinear Vibrations
x 0
1
2
0 5 10 15
t
Fig. 8.6 Analytical and numerical solutions of the period T=8.5 (p=2) shown by
solid and thin lines, respectively.
Note that the equation (8.71) admits the group of time translations. As
a result, another arbitrary parameter, say t0 , is introduced by substitution
t → t + t0 .
Figures 8.6 and 8.7 show that the analytical and numerical solutions are
matching better for larger periods as the system trajectory becomes closer to
the separatrix loop.
x 0
1
2
3
0 10 20 30 40
t
Fig. 8.7 Analytical and numerical solutions of the period T=24.5 (p=6) shown by
the solid and thin lines, respectively.
where f (x) and g(x) are analytic functions, such that (Lienard’ conditions)
"x
a) G(x) = g(u)du is an odd function such that G(0) = G(±μ) = 0 for some
0
μ > 0,
b) G(x) → ∞ if x → ∞, and G(x) is a monotonously increasing function
for x > μ,
c) f (x) is an odd function such that f (x) > 0 for x > 0.
The above conditions guarantee that system (8.90) has a single stable limit
cycle. In this case, the boundary-value problem (8.7) and (8.8) takes the form
X = X0 (τ ) + εX1 (τ ) + ε2 X2 (τ ) + · · · (8.92)
Y = Y0 (τ ) + εY1 (τ ) + ε Y2 (τ ) + · · ·
2
a = q0 + εq1 + ε2 q2 + · · · (8.93)
216 8 Strongly Nonlinear Vibrations
Further solution procedure can be simplified by taking into account the sym-
metry properties X(−τ ) ≡ −X(τ ) and Y (−τ ) ≡ Y (τ ) due to the above
conditions (a) through (c). Substituting (8.92) and (8.93) into (8.91) and
matching the coefficients of the respective powers of ε, gives the following
sequence of boundary value problems
X0 = 0 (8.94)
Y0 = 0, Y0 |τ =1 = 0
Xi+1 = −FX,i , Xi+1 |τ =1 = 0
Yi+1 = −FY,i , Yi+1 |τ =1 = 0 (8.96)
(i = 1, 2, ...)
where
1 di FX 1 di FY
FX,i = |ε=0 , FY,i = |ε=0 (8.97)
i! dεi i! dεi
Note that zero-order and first-order approximations are coupled through the
boundary condition for X-component in (8.95), whereas no boundary condi-
tion is imposed on X0 in (8.94). This specific represents a formalization of
the physical assumption regarding the dominating component in the tempo-
ral mode shape of vibration, which is assumed to be close to the sawtooth
sine rather than rectangular cosine. As a result, the generating system (8.94)
gives solution
X0 = Aτ , Y0 ≡ 0 (8.98)
where A is an arbitrary constant.
Substituting (8.98) in the right-hand side of equations (8.95) and integrat-
ing, yields
τ τ
X1 = −q02 (τ − ξ)f (Aξ)dξ, Y1 = −Aq0 (τ − ξ)g(Aξ)dξ (8.99)
0 0
1 1
q02 f (Aξ)dξ = A, (1 − ξ)g(Aξ)dξ = 0 (8.100)
0 0
8.5 Self-excited Oscillator 217
Relationships (8.98) through (8.100) complete first basic steps of the iterative
procedure. Further iterations are organized then in a similar way as follows
τ τ ζ
Xi+1 = − (τ − ξ)FX,i (ξ)dξ, Yi+1 = − dζ FY,i (ξ)dξ (8.101)
0 1 0
1
FX,i (ξ)dξ = 0; i = 1, 2, .... (8.102)
0
Note that the boundary conditions for Yi+1 are satisfied automatically due
to the lower limit of the outer integral in (8.101), whereas the boundary con-
ditions for Xi+1 generates equations (8.102) for determining the coefficients
of series (8.93). Practically, high-order approximations can be obtained by
using computer systems of symbolic manipulations.
Example 15. Consider the self-excited oscillator with the power form stiffness
of the degree m = 3,
ẍ + (bx2 − 1)ẋ + x3 = 0
In this case, g(x) = bx2 − 1 and f (x) = x3 . Conducting elementary integra-
tions in (8.100), gives the algebraic system
1 2 3 1
q A = A, q0 A 6 − bA2 = 0
4 0 12
with non-trivial solution
2b 6
q0 = , A=
3 b
As a result, integrating (8.99), yields
6 τ5
X1 = − , Y1 = τ 2 − τ 4
b 5
All further steps of the procedure are conducted according to the same scheme
(8.101) and (8.102). For instance, first two steps of the procedure give ap-
proximate solution
6 τ5 1
x= {τ − + [105τ 9 + 900τ 7 b − 21τ 5 (70b + 9) + 350τ 3 b]}
b 5 3150
1
+(1 − τ 2 ){τ 2 − [20 − 43τ 2 + 20τ 4 + 216τ 6 ]}e
420
and the period
b 3
T = 4a = 8 1+
6 20
218 8 Strongly Nonlinear Vibrations
Two more steps of the procedure correct the above expression for the period
as follows
b 3 2960b + 2121 7367360b2 + 4554992b + 8659035
T = 4a = 8 1+ + +
6 20 50400 605404800
Figs.8.8 and 8.9 show limit cycle trajectories described by the analytical so-
lutions in one and two iterations, respectively. For comparison, the numerical
solution for transition to the limit cycle is also presented. Then, Fig. 8.10
illustrates dependence of the quarter of period parameter, a = T /4, versus
the quantity b−1/2 , which can be viewed as an estimate for the amplitude of
limit cycle.
v
3 b 1.0
m3
2
0 x
1 analytical, 1 iteration
numerical
2
3
2 1 0 1 2
Fig. 8.8 Trajectories of numerical solution and the one iteration analytical limit-
cycle solution.
v
3 b 1.0
m3
0 x
analytical, 2 iterations
1 numerical
2
3
Fig. 8.9 Trajectories of numerical solution and approximate (two iterations) ana-
lytical limit-cycle solution.
a
analytical, four iterations
2
numerical estimates
1
0
0 2 4 6 8
12
b
Fig. 8.10 Illustration of convergence of the iterative procedure on the parameter
plane.
220 8 Strongly Nonlinear Vibrations
where f (x) is an odd function such that xf (x) ≥ 0, and 0 < μ << 1.
The idea of two variable expansions will be employed below in combination
with the sawtooth time substitution. Let us assume that τ (ϕ) is a ‘fast’ time
scale, whose phase ϕ depends on the ‘slow’ time scale t0 = μt according to
the following differential equation
ϕ̇ = ω(t0 ) (8.104)
∂2X ∂Y
ω2 2
= −Rf − μH − μ2 LX
∂τ ∂τ
∂2Y ∂X
ω 2 2 = −If − μH − μ2 LY (8.107)
∂τ ∂τ
where, as follows from (8.5) and (8.6),
1
Rf = Rf (X, Y ) = [f (X + Y ) + f (X − Y )]
2
1
If = If (X, Y ) = [f (X + Y ) − f (X − Y )]
2
and two linear differential operators are introduced
∂ dω
H ≡ 2ω 1 + 0 + 0 (8.108)
∂t dt
∂2 ∂
L ≡ 02 + 2 0
∂t ∂t
Note that the fast and slow temporal scales are associated with different
physical processes developed in the system. The slow energy dissipation pro-
cess is represented by the explicit small parameter μ, but there is no explicit
parameter associated with perturbations of the triangular sine wave, which
is supposed to be a generating solution of the iterative process. However, as
8.6 Strongly Nonlinear Oscillator with Viscous Damping 221
and
ε(1 − εμ2 )
ω2 =
4 arcsin2 ε/2
where C is an arbitrary constant, and another arbitrary constant can be
introduced through the arbitrary time shift.
Note that solution (8.109) includes no Y -component because, at every
stage of the iterative process, it appears to be possible to satisfy condition
H(∂X/∂τ ) = 0 (8.110)
Example 16. Consider the weakly damped oscillator of the m degree power
form restoring force characteristic
222 8 Strongly Nonlinear Vibrations
ẍ + 2μẋ + xm = 0 (8.111)
At this stage, the exponent m is an odd positive number. (It will be shown
later that a broader class of power characteristics can also be considered.)
Note that, for this kind of oscillators, whether or not the damping is small
depends on the level of amplitude and the exponent m. This is due to the
fact that, under small enough amplitudes, the elastic force becomes negli-
gible regardless the magnitude of damping. By assuming that the influence
of damping is negligible during one cycle of vibration, one can use expres-
sions (8.40) for estimations of magnitudes of damping and elastic forces. As
a result, the condition of ‘small damping’ derives in the form
1
μ2
(m + 1)Am−1 (8.112)
4
One step of the procedure gives approximate solution [137]
−4μt τ m+2
x = Cexp τ− (8.113)
m+3 m+2
where τ = τ (ϕ) and the phase variable is approximated by
m−1
ϕ = ϕ∞ 1 − exp −2μ t (8.114)
m+3
1 m + 3 C (m−1)/2
ϕ∞ =
2μ m − 1 2(m + 1)
Interestingly enough, the above approximate solution predicts that the oscil-
lator makes only a finite number of waves as m > 1 and t → ∞.
Figs. 8.11 and 8.12 illustrate damped responses of the oscillator with two
different degrees of nonlinearity, m = 3 and m = 7, respectively. As follows
from the diagrams, the approximate analytical solution and numerical one
are matching relatively well, especially at higher exponent, m = 7. In partic-
ular, this justifies the idea of using the sawtooth wave in strongly nonlinear
cases, when the oscillator becomes close to the standard vibroimpact model.
Fig. 8.11 Approximate analytical and numerical solutions of the damped oscillator
with cubic power form characteristic.
Fig. 8.12 Approximate analytical and numerical solutions of the damped oscillator
with the seven-th degree power form characteristic.
224 8 Strongly Nonlinear Vibrations
and
∂ 2 X0 (0) ∂X0
ω02 = −Rf (X0 , Y0 ), |τ =±1 = 0
∂τ 2 ∂τ
2
∂ Y0 (0)
ω02 = −If (X0 , Y0 ), Y0 |τ =±1 = 0 (8.117)
∂τ 2
and
∂ 2 X1 (1) ∂ 2 X0 ∂Y0
ω02 = −Rf (X 0 , Y0 , X 1 , Y1 ) − 2ω 0 ω 1 − H0
∂τ 2 ∂τ 2 ∂τ
∂X1
|τ =±1 = 0
∂τ
∂ 2 Y1 (1) ∂ 2 Y0 ∂X0
ω02 2
= −If (X0 , Y0 , X1 , Y1 ) − 2ω0 ω1 2
− H0 (8.118)
∂τ ∂τ ∂τ
Y1 |τ =±1 = 0
(0) (1) (0) (1)
where Rf , Rf , ... and If ,If , ... are determined by the expansions
By taking into account the assumptions on f (x) in equation (8.103), one can
represent solution of problem (8.117) in the following general form
X0 = X0 (τ, A, ω0 ), Y0 ≡ 0 (8.119)
∂X0 (τ, A, ω0 )
|τ =1 = 0 (8.120)
∂τ
In general, this relationship determines the implicit function ω0 = ω0 (t0 ).
Now substituting solution (8.119) in (8.118), gives equations
∂ 2 X1 ∂ 2 X0
ω02 + f (X 0 )X 1 = −2ω 0 ω 1 (8.121)
∂τ 2 ∂τ 2
8.6 Strongly Nonlinear Oscillator with Viscous Damping 225
and
∂ 2 Y1 ∂X0
ω022
+ f (X0 )Y1 = −H0 (8.122)
∂τ ∂τ
Let us consider equation (8.122). The best choice would be achieved by setting
the right-hand side to zero and therefore making possible the solution Y1 ≡ 0
which is consistent with zero-order solution (8.119) and provides a better
smoothness property of the corresponding solution at this stage; see condition
(8.110). Note that the right-hand side cannot be always made zero for any
τ unless the zero-order solution admits separation of the variables t0 and
τ. However, it is still possible to ‘minimize’ the right-hand side of equation
(8.122) by making it orthogonal to solution of the corresponding homogeneous
equation, ∂X0 /∂τ , in other words,
1 . /
1 ∂X0 ∂X0 ∂X0 ∂X0
H0 dτ ≡ H0 =0 (8.123)
2 ∂τ ∂τ ∂τ ∂τ
−1
Taking into account the expression H0 = 2ω0 1 + ∂/∂t0 + dω0 /dt0 and
condition (8.123), gives
0
2 1
∂X0
ω0 = C exp(−2t0 ) (8.124)
∂τ
where the functions A(t0 ) and ω0 (t0 ) are coupled by relation (8.40) as follows
−1/2
1 A(α−1)/2 α (α + 1)2
ω0 = = √ 1+ 1+ (8.127)
a α+1 2(α + 2) (α + 2)(2α + 3)
It follows from expressions (8.128) and (8.129) that the linear system α = 1
plays the role of a boundary between the two strongly nonlinear areas
1/(α+1)
v02 (α + 1) 7α3 + 31α2 + 47α + 24
C=
2(α + 2)2 (2α + 3)
x = τ (ϕ) (8.131)
v0
ϕ= [1 − exp (−2μt)]
2μ
Solution (8.131) exactly describes the system motion in the square potential
well.
2) When α → 0, expressions (8.126) through (8.129) are reduced to
4 |τ (ϕ)|
x = v0 exp − μt τ (ϕ) 1 −
2
(8.132)
3 2
3 2
ϕ= exp μt − 1
2μv0 3
where the identity sgn[τ (ϕ)]|τ (ϕ)| ≡ τ (ϕ) has been taken into account.
If, in addition μ = 0, then solution (8.132) also exactly describes the
system dynamics with another nonsmooth limit of the potential energy, |x|,
as shown in Fig. 8.13. However, if μ = 0 then substituting solution (8.132)
into the differential equation of motion gives an error O(μ2 ). In terms of first-
order asymptotic solutions, the error of order μ occurs on the time period of
order 1/μ. Therefore, solution (8.132) exactly captures the carrying shape of
the vibration, but gives only asymptotic estimate for the exponential decay.
Fig. 8.13 Potential energy representation for the two limit oscillators.
228 8 Strongly Nonlinear Vibrations
Fig. 8.14 Temporal mode shape of the vibration for α ∈ N∞ , C = 1.5 and μ =
0.04; here and below, the dashed line represents numerical solutions.
v 0
2
4
1.0 0.5 0.0 0.5 1.0
x
Note that the error of solution (8.126) is due to the error of the iterative
procedure for elastic vibrations and the error of asymptotic for energy dissi-
pation. As shown above, the error of successive approximations vanishes as
either α → ∞ or α → 0, but the error of asymptotic vanishes only as α → ∞.
Finally, let us discuss the qualitative difference of the dynamics in the pa-
rameter intervals N0 and N∞ . As follows from equation (8.128), for α ∈ N0 ,
the phase of vibration and the corresponding frequency are exponentially in-
creasing in the slow time scale μt . The physical meaning of this phenomenon
8.6 Strongly Nonlinear Oscillator with Viscous Damping 229
1.0
Α 1 14
analytical, numerical
0.5
x 0.0
0.5
1.0
0 20 40 60 80 100 120 140
t
Fig. 8.16 Temporal mode shape of the vibration for α ∈ N0 , C = 2.5 and μ = 0.04.
becomes most clear in the limit case α = 0. In this case, according to solution
(8.132), the amplitude and frequency are, respectively
v02 4 1 2 1
A(μt) = exp − μt and ϕ̇ = exp μt = (8.133)
2 3 v0 3 2A(μt)
number of cycles such that the phase ϕ remains bounded for any time t.
Again, the most clear interpretation is obtained in the limit case α → ∞,
when, as follows from (8.131), the phase variable ϕ(t) represents the total
distance passed by the particle by time t, and ϕ̇ = v is the absolute value
of the velocity. Since the barriers are perfectly elastic the particle reflects
with no energy loss, the velocity v (t) remains continuous function of time
described by the linear differential equation v̇ + 2μv = 0 or ϕ̈ + 2μϕ̇ = 0.
Under the initial conditions ϕ(0) = 0 and ϕ̇(0) = v0 , one obtains exactly
solution (8.131).
In conclusion, the explicit analytical solution for a class of strongly nonlin-
ear oscillators with viscous damping is introduced. Two different nonsmooth
functions involved into the solution are associated with two different nons-
mooth limits of the oscillator. As a result, the solution is drastically simplified
to give the best match with numerical tests if approaching any of the two
limits.
where ż− and ż+ are velocities right before and immediately after the impact,
respectively.
Let h be a natural spatial scale of the system. This can be, for instance,
the maximal height that has been reached by the ball during the very first
cycle. Introducing
the coordinate transformation z = h|x| and re-scaling the
time as t = h/gp, brings equations (8.134) to the form [204]
d2 x dx
+ 2μ + sgnx = 0 (8.135)
dp2 dp
dx dx dx
− = (1 − κ)
dp + dp − dp −
where *
1 c h
μ=
2m g
8.7 Bouncing Ball 231
ż = u
c
u̇ = −g − u (8.137)
m
where u+ = −κu− whenever z = 0.
Further, introduce new unknown state variables {s, v} according to [73]
z = ssgn(s)
u = sgn(s)[1 − ksgn(sv)]v (8.138)
1−κ
k=
1+κ
Applying (8.138) to (8.137), gives1
ṡ = [1 − ksgn(sv)]v (8.139)
c g
v̇ = − v − [sgn(s) + ksgn(v)]
m 1 − k2
As compared to (8.137), system (8.138) automatically accounts for the veloc-
ity jump condition at z = 0. In other words, transformation (8.138) makes the
strong nonlinearity (due to non-elastic impacts) explicitly present in (8.139).
From the standpoint of numerical procedures, such a transformation enables
one of using built-in solvers of different packages with no impact conditioning
at z = 0.
Note that, in the particular case κ = 1 (k = 0), system (8.139) becomes
equivalent to (8.135). In this particular case, the direct numerical integration
R
of equations (8.139), using the NDSolve procedure built in Mathematica
1
Note that differentiation of sgn-functions will produce Dirac’s delta-functions
with effectively zero factors however.
232 8 Strongly Nonlinear Vibrations
0.05
0.04
0.03
analytical, numerical
z
0.02
0.01
0.00
0 1 2 3 4 5 6
t
Fig. 8.18 NSTT analytical and direct numerical solutions for the height of bounc-
ing ball under the linear dissipation condition; m = 1.0, c = 0.5, z(0) = 0 and
ż(0) = 1.0.
Assuming that the damping is small enough and using the classical
parabolic approximation for the ball height during one cycle, z(t) = −gt2 /2+
v0 t, yields
2v0 /g
2cef f v03 1
cef f ż 2 dt = = m(1 − κ2 )v02
0 3g 2
or
3mg
cef f = (1 − κ2 ) (8.140)
4v0
Expression (8.140) shows that, in this case, effective linear damping cannot
be introduced on the entire time interval since the “initial velocity” v0 de-
creases from cycle to cycle. Choosing v0 = ż(0) in (8.140), provides a good
enough match between analytical (8.136) and numerical solutions during first
few cycles of the motion, as follows from Fig. 8.19. Then, the divergence
between the curves accelerates. The result can be improved by using some
8.7 Bouncing Ball 233
0.05
0.04
0.03 analytical
z numerical
0.02
0.01
0.00
0 1 2 3 4 5 6
t
Fig. 8.19 NSTT analytical solution with effective damping corresponding to the
coefficient of restitution κ = 0.97, and direct numerical solutions for the height of
bouncing ball; the parameters are: m = 1.0, c = 0.0, z(0) = 0 and ż(0) = 1.0.
0.05
0.04
0.03 numerical
z analytical
0.02
0.01
0.00
0 1 2 3 4 5 6
t
Fig. 8.20 NSTT analytical solution with effective damping adjusted by “variable
initial velocity,” and direct numerical solutions for the height of bouncing ball; the
parameters are: m = 1.0, c = 0.0, κ = 0.97, z(0) = 0 and ż(0) = 1.0.
estimate for the “initial velocity” decay based on solution (8.136), where the
effective damping is still constant, however. This one-step iteration gives
c
ef f
c̃ef f = cef f exp t (8.141)
3m
Fig. 8.20 shows a better match between numerical and analytical solutions
achieved due to modification (8.141).
234 8 Strongly Nonlinear Vibrations
θ = Aτ (2t − 1) + B (8.149)
ẍ + F (x) = 0 (8.156)
236 8 Strongly Nonlinear Vibrations
X + hF (X) = 0 (8.157)
X |τ =±1 = 0 (8.158)
where, as usually, h = a2 = (T /4)2 .
Following the physical reasoning discussed in Introduction and the second
section of this chapter, we seek solution in the form of successive approxima-
tion series with no explicit small parameter
X = X0 (τ ) + X1 (τ ) + X2 (τ ) + ... (8.159)
h = h0 + h1 + ...
X0 (τ, τ1 , A0 ) = x1 + A0 (τ − τ1 ) (8.160)
τ
X1 (τ, τ1 , A0 , A1 ) = A1 (τ − τ1 ) − h0 (τ − ξ)F (X0 (ξ, τ1 , A0 ))dξ (8.161)
τ1
1
A0 + A1 − h0 F (X0 (τ, τ1 , A0 ))dτ = 0
τ1
τ1
A0 + A1 + h0 F (X0 (τ, τ1 , A0 ))dτ = 0 (8.163)
−1
1
F (X0 (τ, τ1 , A0 ))dτ = 0 (8.164)
−1
1
F (X0 (τ, τ1 , A0 ))dτ = h−1
0 (A0 + A1 ) (8.165)
τ1
τ
X2 = −h1 (τ − η)F (X0 (η, τ1 , A0 ))dη (8.166)
τ1
τ
−h0 (τ − η)F (X0 (η, τ1 , A0 ))X1 (η, τ1 , A0 , A1 )dη + A2 (τ − τ1 )
τ1
where zero lower limit of integration provides X2 (τ1 ) = 0, and the prime indi-
cates derivative with respect to the entire argument, F (X0 ) ≡ dF (X0 )/dX0 .
Note that the characteristic F (x) is, generally speaking, nonsmooth at the
matching point x1 so that the derivative F (x) may not exist at x = x1 . Al-
though integration of step-wise discontinuous functions is still possible, the
current procedure is designed to avoid calculating derivatives of the char-
acteristic at the point x1 . For that reason, the lower limit of integration in
(8.166) is associated with the non-smoothness point by expression (8.160).
In addition, as follows from (8.161), the uncertainty F (x1 ) in the integrand
is suppressed by zero factor X1 (τ1 ) = 0. Obviously, on the next step of iter-
ation, this factor will accompany the second derivative F (x1 ), whereas the
first derivative acquires the factor X2 (τ1 ).
Now, applying boundary conditions (8.158) to (8.166), yields
1
−h1 F (X0 (τ, τ1 , A0 ))dτ
τ1
1
− h0 F (X0 (τ, τ1 , A0 ))X1 (τ, τ1 , A0 , A1 )dτ + A2 = 0 (8.167)
τ1
τ1
h1 F (X0 (τ, τ1 , A0 ))dτ
−1
τ1
+ h0 F (X0 (τ, τ1 , A0 ))X1 (τ, τ1 , A0 , A1 )dτ + A2 = 0 (8.168)
−1
Substracting the both sides of equation (8.167) from (8.168) and taking into
account equation (8.164), gives
1
F (X0 (τ, τ1 , A0 ))X1 (τ, τ1 , A0 , A1 )dτ = 0 (8.169)
−1
8.9 Oscillators with Piece-Wise Nonlinear Restoring Force Characteristics 239
Then, taking into account (8.165), brings equation (8.167) to the form
1
F (X0 (τ, τ1 , A0 ))X1 (τ )dτ = −h1 h−2 −1
0 (A0 + A1 ) + h0 A2 (8.170)
τ1
u = u(θ, x, t) (9.1)
where θ = θ(x, t) is a phase variable, x and t are the coordinate and time,
respectively.
In the stationary case, the phase variable is usually the only argument of
the wave shape function u = u(θ), where θ = ωt − kx with temporal and
spatial wave numbers, ω and k, respectively, so that
∂θ ∂θ
=ω and = −k (9.2)
∂t ∂x
Function (9.1) includes also the variables x and t explicitly in order to describe
wave modulation effects. Therefore, we assume that the dependence upon
explicit x and t is slow. Such an assumption can be formalized by means of
the formal auxiliary parameter ε [194], [116]
u = u(θ, x0 , t0 ) (9.3)
where x0 = εx, t0 = εt, and relationships (9.2) take now the form
∂θ ∂θ
= ω(x0 , t0 ) and = −k(x0 , t0 ) (9.4)
∂t ∂x
Let the wave shape function u be periodic with respect to θ with the period
T = 4. In this case, the sawtooth phase argument can be introduced as follows
τ = τ (θ),
u = U (τ, x0 , t0 ) + V (τ, x0 , t0 )e (9.5)
where e = e(θ) = τ (θ) is the rectangular cosine of the phase variable.
∂2u ∂2u
− 2 + f (u) = 0 (9.6)
∂t2 ∂x
where f (u) is an odd function.
Substituting (9.5) in (9.6) and taking into account (9.4), gives
2 ∂2U ∂V
ω − k2 = −Rf (U, V ) − εH − ε2 LU
∂τ 2 ∂τ
2 ∂ 2V ∂U
ω − k2 2
= −If (U, V ) − εH − ε2 LV (9.7)
∂τ ∂τ
∂U
|τ =±1 = 0 (9.8)
∂τ
V |τ =±1 = 0
where
∂ ∂ ∂ω ∂k
H ≡ 2 ω 0 +k 0 + 0 + 0
∂t ∂x ∂t ∂x
∂2 ∂2
L ≡ 02 − 02
∂t ∂x
and
1
Rf = [f (U + V ) + f (U − V )]
2
1
If = [f (U + V ) − f (U − V )]
2
Analogously to the case of strongly nonlinear damped oscillator considered
in the previous chapter, we represent solution in the form of series
9.2 Klein-Gordon Equation 243
Substituting (9.9) into (9.7) and (9.9) and equating coefficients of like powers
of ε, gives in first two steps
2 ∂ 2 U0
ω − k2 = −f (U0 )
∂τ 2
V0 ≡ 0 (9.10)
∂U0
|τ =1 = 0 (9.11)
∂τ
where U0 (−τ ) = −U0 (τ ) due to oddness of the function f (U0 ), and
2 ∂ 2 U1
ω − k2 = −f (U0 )U1
∂τ 2
2 ∂ 2 V1 ∂U0
ω − k2 = −f (U0 )V1 − H (9.12)
∂τ 2 ∂τ
∂U1
|τ =±1 = 0
∂τ
V1 |τ =±1 = 0 (9.13)
The generating solution is obtained from (9.10) in the following general form
U0 = U0 (τ, A, ω 2 − k 2 ) (9.14)
∂2u ∂2u
− 2 + sgn(u)|u|α = 0 (9.17)
∂t2 ∂x
244 9 Strongly Nonlinear Waves
∂(ωA2 ) ∂(kA2 )
+ =0 (9.19)
∂t ∂x
Equations (9.18) and (9.19) have been written in terms of the original time
and coordinate by setting the auxiliary parameter to unity, ε = 1. However,
according to the basic assumption, A = A (t, x), ω = ω (t, x) and k = k (t, x)
still should be treated as slow functions as compared to τ (θ).
Chapter 10
Impact Modes and Parameter
Variations
well as the parameter variation and averaging methods for weakly nonlinear
cases.
Let us consider a one degree of freedom free harmonic oscillator between
two absolutely rigid barriers. A mechanical model of such an oscillator can
be represented as a mass-spring model with two-sided amplitude limiters as
shown in Fig. 10.1 (a).
Fig. 10.1 The oscillator with bilateral rigid barriers (on the left) is replaced by
the oscillator under the periodic series of external impulses (on the right).
Since the normal mode regimes are periodic by their definition then the re-
action of constraints can be treated as a periodic series of external impulses
acting on the masses of the system.
10.1 An Introductory Example 247
where δ (ξ) is the Dirac function, τ (ξ) is the triangular sine wave, 2p, ω, and
α will be interpreted as arbitrary parameters.
For further convenience, the right-hand side of equation (10.2) is expressed
through second-order generalized derivative of the triangular sine wave with
respect to the entire argument, ωt + α. The parameter ω will be called a
frequency parameter, although it differs by the factor π/2 from the regular
trigonometric frequency, Ω = (π/2) ω. Both parameters, ω and Ω, may be
used below.
In contrast to system (10.1), the auxiliary system (10.2) is linear but not
completely equivalent to the original one; see the analyses below.
Representing unknown steady-state periodic solution in the form,
x = X (τ ) , τ = τ (ωt + α) (10.3)
ω 2 X (τ ) + ω02 X (τ ) = 0 (10.4)
−2
X (τ ) |τ =±1 = pω
and the related solution is represented in the saw-tooth time form [139]
x = ±Δ if τ = ±1 ⇐⇒ τ = 0 (10.6)
2 a 2 b
1 1
x 0 x 0
1 1
2 2
0 5 10 15 20 0 5 10 15 20
t t
Fig. 10.2 Real (a) and “phantom” (b) solutions corresponding to the first (small-
est) and second roots respectively, (ω0 /ω)1 = 1.1502 and (ω0 /ω)2 = −1.1502 + π.
The total energy level is E = 1.2E∗ .
where the period and the related basic frequency are T = 2/ω and Ω = πω,
respectively.
ẋ = v
v̇ = −2ζω0 v − ω02 x (10.13)
sin(γτ (φ))
x=Δ (10.14)
sin γ
cos(γτ (φ))
v = ω0 Δ e(φ)
sin γ
250 10 Impact Modes and Parameter Variations
where τ = τ (φ) and e = τ (φ) depend upon the fast phase φ = φ(t), and
γ = γ(t) determines a relatively slow evolution of the temporal mode shape
of the vibration.
Substituting (10.14) in (10.13), gives still exact equations
γ = γ0 exp(2ζω0 t)
1
φ= [1 − exp(−2ζω0 t)] (10.17)
2ζγ0
1 ω02 Δ2
E(t) = (10.18)
2 sin2 γ(t)
The duration of the impact stage of the dynamics is estimated via solution
(10.17),
1 π
γ(tmax ) = π/2 =⇒ tmax = ln (10.19)
2ζω0 2γ0
where γ0 = γ(0).
As follows from Fig. 10.3, the above averaging procedure leads to practi-
cally no error of the time history record within the entire interval of validity
of the approach. However, the analytical solution based on the reduced model
10.2 Parameter Variation and Averaging 251
1.0
0.5
x 0.0
0.5
1.0
0 10 20 30 40
a t
1.0
numerical, averaging
0.5
x 0.0
0.5 analytical
1.0
80 82 84 86 88 90
b t
Fig. 10.3 The time history of the impact oscillator according to exact equations,
those after the averaging, and the analytical closed form solution; the parameters
are γ0 = 0.2, ζ = 0.01, ω0 = 1.0, and Δ = 1.0.
gives some deviation from the exact curve at the end of the impact stage of
the dynamics.
Notice that there is no impact interactions with the constraints
for t > tmax , where the model becomes harmonic oscillator whose ampli-
tude exponentially decays due to the energy dissipation. At this stage, the
252 10 Impact Modes and Parameter Variations
transformation (10.14) is not valid any more nor there is any need in trans-
formations. However, a question occurs about such solutions that would be
capable of describing both impact and non-impact stages within the same
unit-form expressions.
where ω02 = k/m, and the parameters ω and p must provide the following
condition
|x1 | ≤ Δ (10.21)
The impact mode solution is represented in the form
xn (t) = Xn (τ ) (10.22)
X1 |τ =±1 = pω −2
X2 |τ =±1 = 0 (10.24)
X1 |τ =±1 = ±Δ (10.26)
10.4 Averaging in the 2DOF System 253
This gives
√ √
p = 2ωω0 Δ[tan(ω0 /ω) + ( 3/3) tan( 3ω0 /ω)]−1 (10.27)
In order to insure that solution (10.25) and (10.27) describe real motions,
condition (10.7) must be verified. As follows from detailed considerations
below, condition (10.7) is satisfied at least in the frequency range (π/2) ω >
ω2 , where ω2 is the frequency of the out-of-phase mode of the related no
impact system.
ẋ1 = v1
v̇1 = −f1 (x1 , x2 , v1 , v2 ) + pe (φ)
ẋ2 = v2 (10.28)
v̇2 = −f2 (x1 , x2 , v1 , v2 )
are impact and linear force components per unit mass, β = k/m is
the pa-
rameter of coupling, ζ = c/(2Ω1 m) is the damping ratio, and Ωi = ki /m
(i = 1, 2).
Fig. 10.4 The two-degrees-of-freedom model with viscous damping in the impact
subcomponent.
254 10 Impact Modes and Parameter Variations
sin(γτ (φ))
x1 = Δ
sin γ
cos(γτ (φ))
v1 = Ω1 Δ e(φ)
sin γ
πτ (φ) πτ (φ)
x2 = A sin + B cos e(φ) (10.30)
2 2
πτ (φ) πτ (φ)
v2 = −Ω1 B sin + Ω1 A cos e(φ)
2 2
Further reduction of system (10.31) includes two major steps, such as averag-
ing with respect to the fast phase φ and applying the power series expansion
10.4 Averaging in the 2DOF System 255
γ̇ = 2ζΩ1 γ
Ω1
φ̇ =
γ
β + Ω12 + Ω22 π
Ȧ = −B − φ̇ (10.33)
2Ω1 2
2 2
β + Ω1 + Ω2 π 4βΔ
Ḃ = A − φ̇ − 2
2Ω1 2 π Ω1
Ρ averaging
0.00010
0.00005
0.00000
0 20 40 60 80
t
Fig. 10.5 The total energy of the second oscillator at relatively weak coupling,
β = 0.01
256 10 Impact Modes and Parameter Variations
0.007
numerical
0.006
0.005 averaging
0.004
Ρ
0.003
0.001
0.000
0 20 40 60 80
t
Fig. 10.6 The total energy of the second oscillator at stronger coupling, β = 0.05
two different strengths of the coupling β. The initial conditions and other
parameters are selected as follows: γ(0) = 0.2, φ(0) = 0.0, A(0) = −0.01,
B(0) = 0, ζ = 0.01, Ω1 = Ω2 = 1.0, Δ = 1.0. The diagrams show the energy
versus time of the second oscillator based on numerical solutions for three
different equations, such as exact equations (10.31), the equations obtained
by averaging (not described here), and the reduced set (10.33). The solutions
are in quite a good match most of the time interval, however, the solution
of truncated set (10.33) show some error near the end of the interval. This
happens because the parameter γ is slowly approaching its limit magnitude
π/2, at which the first oscillator stops interacting with the constraints and the
entire system becomes linear. Remind that equations (10.33) were obtained
by truncating the polynomial expansions in the neighborhood of γ = 0; as a
result, the accuracy of the equations is low near the point γ = π/2. However,
the precision can be significantly improved by keeping few more terms of the
power series with respect to γ.
Fig. 10.7 The mass-spring model of a discrete elastic system with displacement
limiters.
is limited by perfectly stiff elastic constraints, such that |xa | ≤ Δa or, in the
matrix form, T
Ia x ≤ Δa (10.34)
T
where Ia = 0, ..., 1, ..., 0 .
a
Inside the domain (10.34), the differential equations of motion are assumed
to be linear,
M ẍ + Kx = 0 (10.35)
where M and K are constant mass and stiffness N ×N -matrixes, respectively.
The form of matrix equation (10.35) however is general enough to describe
different models, not necessarily mass-spring chains, see the first example
below.
In order to obtain impact mode solutions, the system (10.34) and (10.35)
is replaced by the following impulsively forced linear system under no con-
straints condition
M ẍ + Kx = pIa τ (ωt + α) (10.36)
where p is a priory unknown ‘eigen-value,’ and ω and α are arbitrary constant
parameters.
A family of periodic solutions of the period T = 4/ω can be found as
a linear superposition of solutions (10.5) for each of the N linear modes of
system (10.35) with appropriate replacement of the parameters,
N
(eTj Ia )ej sin[(ωj /ω)τ (ωt + α)]
x(t) = p (10.37)
j=1
ωωj cos(ωj /ω)
where ej and ωj are the jth normal mode and the natural frequency of linear
system (10.35); it is assumed that ωi < ωj when i < j, and the linear normal
modes are normalized as
258 10 Impact Modes and Parameter Variations
where eTj Ia is the ath component of the jth linear mode vector.
Substitution (10.40) into (10.37) gives a two parameter family of the peri-
odic solutions for the impact modes. The parameter α is an arbitrary phase
shift, whereas the frequency parameter ω is subjected to some restrictions due
to condition (10.34). As shown in [156], condition (10.34) is satisfied when the
principal frequency of vibration, Ω = (π/2) ω, exceeds the highest frequency
of the linear spectrum, Ω > ωN . The corresponding impact mode represents
an extension of the highest linear normal mode, in which any two neighboring
masses vibrate out-of-phase. Such an impact mode becomes spatially local-
ized as Ω → ∞. This result was obtained also by qualitative methods in [197].
Condition (10.34) may be satisfied also when the frequency Ω is located in a
small enough right neighborhood of any frequency ωj and, in addition, ωi /ωj
is not an odd number for all i = j. The idea of the proof is to find such cases
when ITa x is a monotonic function of τ on the interval −1 ≤ τ ≤ 1, and hence
condition (10.39) at the boundaries guarantees that inequality (10.34) holds
inside the entire interval.
Generally, the impact modes appear to have a quite complicated spec-
tral structure. Therefore, a detailed investigation may be required in order
to formulate necessary and sufficient conditions of impact mode existence.
However, a sufficient condition of non-existence can be formulated by using
the physical meaning of the parameter p. Namely, if an impact mode exists
then the inequality p (ω) > 0 holds. Indeed, the parameter p (10.40) cannot
be negative for any real impact vibrating regime as a reaction of constraint,
because it cannot be directed to a barrier. Thus impact modes can not exist
when p (ω) < 0.
|ϕ1 | ≤ Δ1 1
Assuming that the angle ϕ2 is also small enough, brings the differential equa-
tions of motion between the barriers to the linear form
μ2 ϕ̈1 + ϕ̈2 + μ2 ϕ1 = 0
ϕ̈1 + ϕ̈2 + ϕ2 = 0
where the modal vectors satisfy the orthogonality condition, eTj M ei = δji ,
with respect to the mass matrix
2
μ 1
M=
1 1
where k and m are the stiffness of each spring and the mass of each particle,
respectively.
In this case, the corresponding linear modes and their frequencies are de-
scribed exactly by expressions
T
2 πj N πj
ej = sin , ..., sin (10.44)
N +1 N +1 N +1
πj
ωj = ωN +1 sin
2 (N + 1)
where notation2 ωN +1 = 2 k/m has been introduced, and the basis vectors
are normalized so that eTj ei = δji .
The ath component of the jth normal mode vector is therefore
T 2 aπj
ej Ia = sin (10.45)
N +1 N +1
Let us show that the impact mode periodic solution (10.37) and (10.40)
becomes localized as ω → ∞.
First, replacing the trigonometric functions by their asymptotic estimates
sin(ωj τ /ω) ∼ ωj τ /ω, cos(ωj /ω) ∼ 1, and tan(ωj /ω) ∼ ωj /ω, gives
2N T T
j=1 (ej Ia )(ej Ib )
xb ∼ Δa 2N τ (ωt + α) as ω→∞ (10.46)
T 2
j=1 (ej Ia )
In this particular example, the sums can easily be evaluated. So, taking into
account expression (10.45) and the standard trigonometric sums [52], gives
N
2
N
aπj bπj
(eTj Ia )(eTj Ib ) = sin sin = δab (10.47)
j=1
N + 1 j=1 N +1 N +1
N
1 2
(eTj Ia )(eTj Ib )ωj2 = ω (−δa,b−1 + 2δa,b − δa,b+1 ) (10.48)
j=1
2 N +1
Expression (10.49) shows that the ath particle of the chain vibrates accord-
ing to the saw-tooth temporal mode shape with the infinitely large frequency,
2
Note that this is just a suitable notation since the (N + 1)th frequency does not
physically exist.
262 10 Impact Modes and Parameter Variations
whereas all the other particles are at rest. Therefore, the impact mode be-
comes spatially localized as ω → ∞.
In this case, the impulsive excitation on the right-hand side of the auxiliary
equation must act on both ath and bth particles, so that the equation takes
the form
M ẍ + Kx = (pa Ia + pb Ib )τ (ωt + α) (10.51)
where pa and pb are parameters to be determined.
The related solution includes terms related to pa and pb , and can be rep-
resented in the form
N
pa (eTj Ia )ej + pb (eTj Ib )ej sin(ωj τ (ωt + α) /ω)
x(t) = (10.52)
j=1
ωωj cos(ωj /ω)
Following the idea of normal modes, let us assume that the impact mode
periodic motion is accompanied by synchronous impacts of both particles
with the constraints according to conditions
kaa pa + kab pb = Δa
kba pa + kbb pb = Δb (10.54)
where
N
(eTj Ia )(eTj Ib ) ωj
kab = tan (10.55)
j=1
ωωj ω
Expressions (10.52) and (10.54) give a formal impact mode solution, which
indeed describes an impact mode when the determinant of system (10.54)
is non-zero and condition (10.50) holds. Solution (10.52) can be viewed as a
strongly non-linear superposition of the two basic impact modes with a single
impacting mass.
10.6 Systems with Multiple Impacting Particles 263
Let us consider the higher frequency domain, (π/2) ω >> ωN , for the
above example of mass-spring system. In this case, tan(ωj /ω) > 0 for all
j = 1, ..., N , and therefore, the coefficients kab (10.55) create a Gram matrix
with non-zero determinant [23]. The asymptotic estimation below confirms
this conclusion.
Namely, assuming that (π/2) ω >> ωN , provides that all the ratios ωj /ω
are small enough to justify the following asymptotic estimate
sin[(ωj /ω)τ ] ωj τ 1 ωj 3 τ3
= + τ− + O ω −5 (10.56)
cos(ωj /ω) ω 2 ω 3
Kaa Pa + Kab Pb = Δa
Kba Pa + Kbb Pb = Δb (10.58)
where
1 ωN +1 2
Kab = ω 2 kab = δab − (δa,b−1 − 2δab + δa,b+1 ) + O ω −4 (10.59)
6 ω
As follows from (10.59), equations (10.58) have the solution Pa = Δa and
Pb = Δb as ω → ∞. In this limit, the vibration energy becomes localized on
the two particles vibrating between the barriers with the saw-tooth temporal
shape,
xc ∼ (Δa δac + Δb δbc ) τ (ωt + α) as ω → ∞ (10.60)
According to (10.60), the particles with numbers c = a and c = b are at rest.
However, they oscillate with amplitudes of different orders of ω −2 when the
parameter ω takes a large but finite value.
Expansion (10.57) shows that the temporal mode shapes of particles c = a
and c = b are non-smooth and getting close to the triangular sine wave as the
frequency ω increases. The temporal mode shapes of the nearest particles,
c = a ± 1 and c = b ± 1, have amplitudes of order ω −2 and appear to be twice
continuously differentiable with respect to time, t. In particular, calculating
directly first two derivatives, gives
264 10 Impact Modes and Parameter Variations
d τ3
τ− = ω 1 − τ 2 τ ∈ C 1 (R)
dt 3
d 2
τ3 2
τ − = ω 2
−2τ (τ ) + 1 − τ 2
τ = −ω 2 2τ ∈ C (R)
dt2 3
where prime denotes differentiation with respect to the whole argument of the
saw-tooth sine, τ ≡ dτ /d (ωt + α), and the term underlined is zero3 because
1 − τ 2 = 0 on the set of points {t : τ (ωt + α) = ±1}, where τ = 0.
3
Namely, it gives zero contribution into the related integrals of the theory of
distributions.
Chapter 11
Principal Trajectories of Forced
Vibrations
− ω 2 M q + Kq = μq (11.4)
Now, let es and ωs be conventional linear normal modes and natural frequen-
cies of the system. (The related eigen-value problem is obtained from (11.4)
by setting μ = 0.) As follows from the linear theory, the normal mode vectors
are orthogonal with respect to the mass matrix M so that the normalization
condition can be represented in the form
Using the normal mode basis for the above steady-state, gives
eTs Q
x= es sin ωt (11.8)
s
ωs2 − ω 2
Since the uniqueness theorem holds, expansions (11.6) and (11.8) must rep-
resent the same solution, and therefore,
vsT Q eTs Q
vs = es (11.9)
s
μs s
ωs2 − ω 2
268 11 Principal Trajectories of Forced Vibrations
Let the external force amplitude vector Q be directed along one of the prin-
cipal directions. Then, expansion (11.6) will include only one term, whereas
expansion (11.8) still includes all n terms.
Now, let us consider the case, when the mass matrix is equal to the identity
matrix, M = E. In this particular case, expression (11.4) takes the standard
form of the eigen-value problem for normal modes with respect to the eigen-
value parameter ω 2 + μ,
− ω 2 + μ Eq + Kq = 0 (11.10)
As follows from (11.10), the eigen-values of free and forced vibration are
coupled by expression
μ
m=− (11.14)
ω2
Indeed, substituting expression x (t) = q sin ωt in equation (11.13) and taking
into account expression (11.14), gives definition (11.2) in the form p =μx. In
contrast to linear case (11.2), however, definition (11.13) allows non-harmonic
temporal shapes.
Current definition itself does not imply that the system is weakly nonlinear.
However, if the parameter ε is small then explicit solutions can be obtained in
terms of conventional asymptotic expansions as described in the next section.
As mentioned, the notion of principal trajectories seems to relate to the
idea of ‘natural forcing functions’ introduced in [57] for the Duffing oscillator.
Let us consider now a multidimensional case from that point of view.
Applying definition (11.13) to the general nonlinear system
where the matrix of the transformation includes the effective mass parame-
ter m.
Relationship (11.16) can be viewed as a vector version of the concept of
natural forcing functions.
On the other hand, using the definition for principal trajectories and ex-
cluding the external forcing vector p (ωt) from the equation of motion, gives
an auxiliary free system described by the differential equation of motion
(M − mE) ẍ + F (x) = 0
x = X (τ ) (11.18)
X (τ ) |τ =±1 = 0 (11.20)
As mentioned above, the temporal and spatial variables generally are not
separable any more in nonlinear cases, therefore it is impossible to obtain an
exact nonlinear version of the eigenvector problem (11.4). As a result, both
temporal and spatial mode shapes must be corrected on each step of the
related asymptotic process as described below.
Remind that the differential operator L in equation (11.19) includes the
frequency parameter ω fixed, whereas the mass m is an eigen value to be
determined.
Let ma and ea (τ ) be the eigen value and eigen vector of the linearized
problem, ε = 0, respectively,
2
2ω
Lea = ma ea (11.21)
π
ea | τ =±1 = 0
where the index a = {s, j} consists of spatial and temporal mode shape
numbers, s = 1, ..., n and j = 1, ..., respectively.
The scalar product of any two vector-functions x = x (τ ) and y = y (τ )
will be defined as follows
1 1 T
x, y = x ydτ
2 −1
Let us represent solution of the weakly nonlinear eigen value problem (11.19)
and (11.20) in the following form of asymptotic expansions
11.5 Definition for Principal Modes of Continuous Systems 271
X(τ ) = Aea (τ ) + εX(1) (τ ) + O ε2 (11.22)
m = ma + εη1 + O ε2
Then substituting (11.22) in (11.19) and (11.20), and matching the coeffi-
cients of the first order of ε, gives equation
2
2
2ω 2ω
LX (1)
− (1)
ma X = −f (Aea ) + η1 Aea (11.23)
π π
(1)
where b = {r, i} is a double index, ab are yet unknown constant coefficients,
and boundary condition (11.24) is automatically satisfied.
Let us assume the following normalization condition for the eigen vector-
functions
0, b = a
ea (τ ), eb (τ ) = (11.26)
1, b=a
Substituting (11.25) in (11.23) and taking into account (11.26), determines
(1)
the coefficients ab and η1 . As a result, expansions (11.22) give first-order
asymptotic solution
π 2 e , f (Ae ) e
b a b
X = Aea + ε + O ε2 (11.27)
2ω mb − ma
b=a
π 2 e , f (Ae )
a a
m = ma − ε + O ε2
2ω A
As follows from the form of solution (11.27), all the coefficients are uniquely
determined under the condition that ma = mb for a = b. The possibility of
degeneration, namely ma = mb for a = b, depends on the inner properties
of the system and the frequency parameter ω. The related examples were
considered earlier [136], [141].
∂ 2 u(t, y)
σ = p(ωt, y) (11.31)
∂t2
will be called a principal mode of forced vibration.
In one-dimensional cases, σ is a priory unknown effective mass per unit length.
Substituting (11.31) in (11.28), gives the following partial differential equa-
tion for principal modes of forced vibrations
∂2u
Lu+εf [u] = σ (11.32)
∂t2
Introducing the triangular wave time substitution as τ = τ ((2ω/π)t) and
u(t, y) = U (τ, y), gives
2
2ω ∂2U
LU +εf (U ) = σ
π ∂τ 2
2
2ω ∂2 ∂2
L≡ ρ(y) − T (11.33)
π ∂τ 2 ∂y 2
The boundary conditions are formulated for both temporal and spatial vari-
ables as
U (τ, 0) = U (τ, l) = 0 (11.34)
and,
∂U (τ, l)
|τ =±1 = 0 (11.35)
∂τ
respectively.
11.5 Definition for Principal Modes of Continuous Systems 273
Periodic solutions and their bifurcation diagrams often reveal important qual-
itative features of the dynamics even though the system motion is not ex-
pected to be periodic. In particular, the number of periodic orbits, their
distribution and properties reveal the structure of chaotic orbits; see, for in-
stance, works [12], [53], [123], and references therein. Direct numerical tools
for detection and construction of periodic orbits based on the mapping ap-
proach can be found in the book [125] and paper [54]. Different formulations
in terms of boundary value problems for ordinary differential equations are
described in [10]. Theoretical and applied results regarding periodic motions,
bifurcations and chaos are reported in [13] and [196].
In this chapter, a special two-dimensional visualization of the shooting
method is introduced in order to incorporate the general two component
NSTT as a preliminary analytical stage [144]. Note that the same approach
using the one-component NSTT was suggested earlier in [154] and imple-
R
mented in Mathematica interface [18]. In particular, subharmonic orbits of
the forced pendulum and bifurcation diagrams were obtained by examining
the shooting curves and their zeros.
x → {X, Y } : x = X (τ ) + Y (τ ) τ (12.3)
whereτ = τ (t/a) the triangular sine wave of the period T = 4a, and
τ = dτ (t/a) /d (t/a) is its first generalized derivative, which is a step-wise
discontinuous function at the time instances
(12.1). In some cases, such a singular term can be employed though in order
to eliminate singularities from original equations; see below.
Despite of a relatively complicated form of equations (12.5) and (12.6),
the new formulation brings some advantages due to the fact that the new
temporal variable τ is bounded and automatically accounts for periodicity
of solutions regardless their temporal shapes. This property appears to be
important in those cases when the solutions do not represent a final stage of
investigation but must be used for further analyses. The dimension increase
is often compensated by an effective decrease of the temporal interval of
the problem, since the range −1 τ 1 is covered by the original time
domain −a t a, which is twice shorter than the whole period T =
4a. Moreover, there are many cases when the number of equations can be
reduced to that of the original system due to the symmetry of equations.
If, for instance, the vector-function F (ẍ, ẋ, x, t) is even with respect to the
velocity ẋ or includes no velocity at all, and the explicit dependence on time
t produces zero ‘imaginary component’ then boundary value problem (12.5)
through (12.7) admits a family of solutions on which
X X X
Y ≡ 0, F , , , aτ = 0 (12.8)
a2 a a
X |τ =±1 = 0 (12.9)
The particular case (12.8) and (12.9) was investigated numerically by the
shooting method in [154] and [177] based on a single- and multiple-degrees-of
freedom systems, respectively. It should be noted that no special require-
ments are imposed on numerical methods or packages for solving the above
boundary-value problems. However, the shooting algorithm in the Mathe-
R
matica interface provides a physically meaningful way of visualization of
periodic solutions due to the specific combination analytical and numerical
commands.
and
In equation (12.10), the function f (x, ẋ, t) may still include parametric terms
of the period T = 4a with possible step-wise discontinuities on Λ. The accel-
eration ẍ also participates as a summand, since it must have the same kind
of singularities as the external forcing function, q (t). According to the distri-
bution theory [165], p (τ (t/a)) must be at least continuous on Λ, otherwise
the ‘product’ p (τ ) τ cannot be treated as a distribution. Note, that behavior
of the function p(τ (t/a)) between the times Λ is arbitrary, since only values
p (−1) and p (1) contribute into the expression
The numbers p (−1) and p (1) control the ‘amplitudes’ and directions of the
δ−pulses. For example, all the pulses can be positively co-directed by setting
p (τ ) = − sign τ .
and
a−2 X |τ =±1 = p (±1) (12.17)
where
Rf 1 X + Y X − Y
= f X + Y, , aτ ± f X − Y, − , 2a − aτ
If 2 a a
(12.18)
Note that the singular term a−1 Y τ is eliminated from the velocity vector
ẋ (t) by imposing another boundary condition
Y |τ =±1 = 0 (12.19)
ẍ + ζ ẋ + x3 = B sin ωt (12.22)
In this case, the differential equations (12.15) and (12.16) take the form
πτ
a−2 X + ζa−1 Y + X 3 + 3XY 2 = B sin (12.23)
2
a−2 Y + ζa−1 X + Y 3 + 3X 2 Y = 0 (12.24)
By the idea of shooting method, the initial value problem (12.23), (12.24),
(12.26) and (12.27) must be iteratively solved multiple times at different g
and h until sufficient precision has been reached for boundary conditions
(12.25) at right end τ = 1,
∂X(τ ; g, h)
|τ =1 ≡ G(g, h) = 0 (12.29)
∂τ
Y (τ ; g, h)|τ =1 ≡ H(g, h) = 0
When dealing with the particular cases (12.20) or (12.21), such a procedure is
not difficult since one has only one equation with a singe unknown, G(g) = 0
or H(h) = 0. Multidimensional cases, such as (12.29), appear to be more
difficult and time consuming. From this point of view, the important feature
of Mathematica is that it is possible to program the functions G(g, h) and
H(g, h) ‘explicitly’ in such a way that the arguments g and h are included
into the numerical solver of differential equations. This can be done as follows.
First, the numerical solution is defined as a function of the arguments g and
h according to the command
20
10
d
c e
b
h 0
a
10
20
4 2 0 2 4
g
Fig. 12.1 The curves G(g, h) = 0 (continuous) and H(g, h) = 0 (dashed) and their
intersections for the Ueda oscillator under the sine-wave input and the following
parameters: ζ = 0.1, B = 12, and ω = 1 (0.1592 Hz).
A variation of the solution x(t), say u(t), is described by the linear differ-
ential equation with periodic coefficients
ü + q1 (t)u̇ + q2 (t)u = 0
where q1 (t) = ∂f (x, ẋ, t)/∂ ẋ is assumed to be independent on ẋ, and q2 (t) =
∂f (x, ẋ, t)/∂x.
Then substitution
⎛ ⎞
t
1
u = y(t) exp ⎝− q1 (z)dz ⎠
2
0
12.3 Sample Problems and Discussion 283
x x
6 6
a b
4 4
2 2
t t
1 2 3 4 5 6 1 2 3 4 5 6
2 2
4 4
6 6
x x
6 6
c d
4 4
2 2
t t
1 2 3 4 5 6 1 2 3 4 5 6
2 2
4 4
6 6
x
6
e
4
2
t
1 2 3 4 5 6
2
4
6
Fig. 12.2 The temporal mode shapes of periodic solutions.
gives
ÿ + p(t)y = 0 (12.30)
where
1 1
p(t) = q2 (t) − [q1 (t)]2 − q̇1 (t)
4 2
As known from the Floquet theory, stability of the solution x(t) is determined
by the Floquet multipliers
μ1,2 = A ± A2 − 1 (12.31)
284 12 NSTT and Shooting Method for Periodic Motions
v v
6
a
4 5 b
2
0 x 0 x
2
4 5
6
3 2 1 0 1 2 3 3 2 1 0 1 2 3
v v
5 c 5 d
0 x 0 x
5 5
3 2 1 0 1 2 3 3 2 1 0 1 2 3
v
6
e
4
2
0 x
2
4
6
3 2 1 0 1 2 3
Fig. 12.3 The projections of periodic trajectories on xv-planes.
where A = [y1 (T ) + ẏ2 (T )]/2, and y1 (t) and y2 (t) are two fundamental solu-
tions of equation (12.30) given by the initial conditions
0
x
−5
50 100 150 200 250 300
Time
2.5
2
Frequency
1.5
0.5
0
0 10 20 30 40 50
Time
Fig. 12.4 The time history record and its spectrogram (in Hz) for Ueda oscillator
after the direct numerical integration. The parameters are: ζ = 0.1 and B = 12.0.
where the index correspond to the type of periodic solution of the original
equation; see Figs. 12.1, 12.2, and 12.3. These numbers confirm that only
solution (e) is stable.
dτ (t/a)
ẍ + ζ ẋ + x3 = B (12.35)
d (t/a)
ẍ + ζ ẋ + x3 = pτ (12.38)
∞
t t
= 2p δ + 1 − 4k − δ − 1 − 4k
a a
k=−∞
40
20
h 0
a
c b
20
40
4 2 0 2 4
g
Fig. 12.5 The curves G(g, h) = 0 (continuous) and H(g, h) = 0 (dashed) for the
Ueda oscillator under the step-wise input and the following parameters: ζ = 0.05,
B = 7.4, and ω = 1.
In this case, both equations (12.36) and (12.37) should have zero right-
hand side, however, the non-homogeneous version of the boundary con-
dition (12.17) must be imposed in order to eliminate the pulses. The
second expression in (12.26) and the first one in (12.29) must be modified
as X (−1) =a2 B=[π/(2ω)]2 B and G (g, h)=a2 B=[π/(2ω)]2 B, respectively.
Therefore, the singular terms are eliminated from the system due to the saw-
tooth time, and the shooting procedure can be applied in the same fashion as
that under the smooth input. The shooting diagram and the corresponding pe-
riodic solutions are shown in Figures 12.9, and 12.10 and 12.11, respectively.
The projections of the phase trajectories show discontinuities of the veloc-
ity on the xv-plane caused by the external pulses. The last four projections,
288 12 NSTT and Shooting Method for Periodic Motions
x x
6 6
a b
4 4
2 2
t t
1 2 3 4 5 6 1 2 3 4 5 6
2 2
4 4
6 6
x x
6 6
c d
4 4
2 2
t t
1 2 3 4 5 6 1 2 3 4 5 6
2 2
4 4
6 6
x x
6 6
e f
4 4
2 2
t t
1 2 3 4 5 6 1 2 3 4 5 6
2 2
4 4
6 6
x
6
g
4
2
t
1 2 3 4 5 6
2
4
6
Fig. 12.6 The temporal mode shapes of periodic solutions og Ueda circuit under
the step-wise input.
v v
6
4
a 4
b
2 2
0 x 0 x
2 2
4
4
6
3 2 1 0 1 2 3 3 2 1 0 1 2 3
v v
10
6
4 5
c d
2
0 x 0 x
2
4 5
6
10
3 2 1 0 1 2 3 4 2 0 2 4
v v
10
20
5 10
e
0 x 0 f x
5 10
20
10
4 2 0 2 4 6 4 2 0 2 4 6
20
10
g
0 x
10
20
6 4 2 0 2 4 6
Fig. 12.7 The projections of periodic trajectories on xv-planes.
intersections are difficult to determine due to a very small angle between the
intersecting curves.
290 12 NSTT and Shooting Method for Periodic Motions
0
x
−5
50 100 150 200 250 300
Time
2.5
2
Frequency
1.5
0.5
0
0 10 20 30 40 50
Time
Fig. 12.8 The time history record and its spectrogram (in Hz) for evolution of the
solution (a) under the step-wise voltage of the amplitude B = 7.4.
30
j k
20
10
g
i d h
a b
h 0 c
e
f
10
20
30
6 4 2 0 2 4 6
g
Fig. 12.9 The curves G(g, h) = 0 (continuous) and H(g, h) = 0 (dashed) for Ueda
circuit under the impulsive input and parameters: ζ = 0.05, B = 1.4, and ω = 1.
x x x
a 2 b 2
1.5
1.0 c
1 1
0.5
t t t
0.5 1 2 3 4 5 6 1 2 3 4 5 6 1 2 3 4 5 6
1.0 1 1
1.5 2 2
x x x
d e
2 2 2 f
1 1 1
t t t
1 2 3 4 5 6 1 2 3 4 5 6 1 2 3 4 5 6
1 1 1
2 2 2
x x x
g 3 h 3 i
2
2 2
1 1 1
t t t
1 2 3 4 5 6 1 1 2 3 4 5 6 1 1 2 3 4 5 6
1
2 2
2 3 3
x x
6 6
4 j k
4
2 2
t t
1 2 3 4 5 6 1 2 3 4 5 6
2 2
4 4
6 6
Fig. 12.10 The temporal mode shapes of periodic solutions og Ueda circuit under
the impulsive input.
v v v
4 4
2
a 2 b c
1 2
0 x 0 x 0 x
1 2 2
2 4 4
1.5
1.0
0.50.00.51.01.5 2 1 0 1 2 2 1 0 1 2
v v v
4 4
e 4 f
2 d 2 2
0 x 0 x 0 x
2 2
2
4
4 4
6
2 1 0 1 2 2 1 0 1 2 2 1 0 1 2
v v v
6 10
4 g i
5 h 5
2
0 x 0 x 0 x
2 5 5
4
10
2 1 0 1 2 321 0 1 2 3 321 0 1 2 3
v v
20 20
j k
10 10
0 x 0 x
10 10
20 20
6 4 2 0 2 4 6 6 4 2 0 2 4 6
In this case, the shooting procedure should be based on the initial conditions
at τ = −1,
where the numbers g and h are determined to satisfy the boundary conditions
on the right end of the interval −1 ≤ τ ≤ 1,
∂X1 (τ ; g, h)
|τ =1 ≡ G(g, h) = 0
∂τ
∂X2 (τ ; g, h)
|τ =1 ≡ H(g, h) = 0 (12.44)
∂τ
In this case, every solution g and h of system (12.44) gives the initial position
on the configuration plane X1 X2 at which the system starts with zero velocity
its periodic motion of the period T = 4a.
The form of equations (12.45) is easier than (12.42), but the parameter a
becomes unknown. It is possible to avoid determining the parameter a by
considering it as a control parameter for tracking the evolution of shooting
diagrams. Alternatively, the parameter a can be considered as a shooting
parameter by imposing one constraint on the parameters g or h. Let us con-
sider, for instance, the system trajectories
in the configuration plane X1 X2 .
Introducing the amplitude A = g 2 + h2 in (12.43), gives
X1 (−1) = A cos ϕ
X2 (−1) = A sin ϕ
where the angle ϕ (0 ≤ ϕ < 2π) together with the parameter a can play the
role of a new unknowns to be determined by shooting whereas the amplitude
A is considered as a control parameter.
Chapter 13
Essentially Non-periodic Processes
Abstract. This chapter describes a possible physical basis for NSTT in case
of essentially non-periodic processes. The physical time is structurised to
match the one-dimensional dynamics of rigid-body chain of identical particles.
Namely, the continuos ‘global’ time is associated with the propagation of
linear momentum, whereas a sequence of non-smooth ‘local’ times describe
behavious of individual physical particles. Such an idea helps to incorporate
temporal symmetries of the dynamics into differential equations of motion
in many other cases of regular or irregular sequences of internal impacts
or external pulses. Since the local times are bounded, a much wider set of
analytical tools becomes possible, wereas matching conditions are generated
automatically by the corresponding time substitution.
s
1.0
0.8
0.6
0.4
0.2
t
1.0 0.5 0.5 1.0 1.5 2.0
Fig. 13.1 The unit slope ramp function at d = 1.0
s
1.0
0.8
0.6
0.4
0.2
t
1.0 0.5 0.5 1.0 1.5 2.0
Fig. 13.3 Physical meaning of the ramp function: s(t; 1) describes position of the
bead struck by another bead from the left and moving until it strikes the next bead
of the same mass.
Now, let us consider an infinite chain of the identical perfectly stiff beads
located regularly on a straight line at the points xi (i = 0, 1, ...). No energy
loss is assumed so that any currently moving bead has the same velocity. As
a result, the linear momentum is translated with the constant speed v = 1,
whereas the beads are interacting at the time instances ti = xi . Making the
temporal shift t → t − ti in function (13.1), gives
1
si (t) = s (t − ti , di ) = (di + |t − ti | − |t − ti+1 |) (13.2)
2
where di = ti+1 − ti .
13.1 Nonsmooth Time Decomposition and Pulse Propagation 297
Due to the unit velocity, function (13.2) can play the role of ‘local’ time for
the bead moving within the interval xi < x < xi+1 during the ‘global’ time
interval ti < t < ti+1 . The term “local” means that the temporal variable si
starts at zero when the “global” time, t, has reached the point t = ti .
In other words, the global temporal variable is associated with the linear
momentum, whereas all the local temporal variables are “attached” to the
physical bodies.
For any sequence of time instances, Λ={t0 , t1 , ...}, the global time, t ∈
[t0 , ∞), can be expressed through the sequence of local times, {si }, in the
form
∞
t= (ti + si ) ṡi (13.3)
i=0
where the relationship (ti+1 + si+1 (ti+1 ) − ti − si (ti+1 )) = 0 has been used
for calculations.
Therefore, for almost all t > t0 , expression (13.5) gives
∞
1= ṡi (13.6)
i=0
Equality (13.6) holds based on the definition for ṡi as illustrated by Fig. 13.2,
therefore expansion (13.3) also holds for almost all t > t0 at least, and its
time derivative includes no singular functions.
Note that the right-hand side of expansion (13.3) can be viewed as an
element of algebra with the basis {si } and multiplication rule (13.4). This
significantly eases different manipulations with the temporal variable (13.3),
for instance
298 13 Essentially Non-periodic Processes
∞
n n
t = (ti + si ) ṡi , n = 1, 2, ... (13.7)
i=0
or, generally,
∞
∞
x (t) = x (ti + si ) ṡi = Xi (si ) ṡi (13.8)
i=0 i=0
Since the right-hand sides of (13.7) and (13.8) have the same structure as the
argument t itself, then considering further a general function g, gives
∞
g (x) = g (Xi ) ṡi (13.9)
i=0
Now, differentiating (13.8) with respect to time t, and taking into account
that si (ti ) = 0 and si−1 (ti ) = di−1 , gives
∞
∞
ẋ (t) = Xi (si ) ṡi + Xi (si ) [δ (t − ti ) − δ (t − ti+1 )] (13.10)
i=0 i=0
∞ ∞
= Xi (si ) ṡi + [Xi (0) − Xi−1 (di−1 )] δ (t − ti )
i=0 i=0
Under condition (13.11), the derivative ẋ (t) has the same algebraic structure
as the function x (t) itself. As a result, transformation (13.3) can be applied to
a general class of dynamical systems. Moreover, in case of impulsively loaded
systems, the sequences of δ-pulses in (13.10) can be utilized for eliminating
the corresponding singularities from dynamical systems.
ẋ = f (x, t) (13.14)
x (0) = p0 (13.15)
Below, solution of the initial value problem (13.12) and (13.13) is introduced
in the specific form based on the operator Lie associated with dynamical
system (13.15)
∂ ∂
A = f (x, t) + (13.16)
∂x ∂t
∂ ∂ ∂
= f1 (x, t) + .... + fn (x, t) +
∂x1 ∂xn ∂t
It is known, for instance, that the exponent of operator (13.16) produces
temporal shifts as follows
Proposition 6. Solution of the initial value problem (13.12) and (13.13) can
be represented in the form
∞
x (t) = [ai−1 + pi + F (ai−1 + pi , ti , si (t))] ṡi (t) (13.18)
i=0
a−1 = 0 (13.19)
ai = ai−1 + pi + F (ai−1 + pi , ti , di ) ; i = 1, 2, ...
Note that the distances di between times Λ are not necessary small, how-
ever, the precision of the solution can be improved by increasing the number
of terms of the Lie series ezA f (x, t) with respect to z, rather than reducing
the distances di .
δ = |J − Jappr | /J (13.32)
302 13 Essentially Non-periodic Processes
0.030
0.025
0.020
Δ 0.015
0.010
0.005
0.000
0.0 0.2 0.4 0.6 0.8
d
Fig. 13.4 Relative error of the determinant based on the truncated Lie series
including terms of order O(z 2 ).
Fig. 13.5 Relative error of the determinant based on the truncated Lie series
including terms of order O(z 3 ).
where ρ is the mass density per unit length, T is tension, q(y/ε, y, t) is the
body force or external loading, which is assumed to be periodic in the ‘fast
scale’ y/ε (0 < ε
1) with the period normalized to four, and similar
assumption is made with respect to the transverse displacement of the string
u = u(y/ε, y, t); see Fig. 14.1 for illustration.
Note that equation (14.1) has no point-wise interpretation due to the pres-
ence of Dirac δ-function. Both sides of the equation therefore must be viewed
as distributions producing the same output if applied to the same testing
function. Correctness of such type of modeling was intensively discussed in
the literature; see, for instance, [46]. Omitting details, the series of δ-functions
in equation (14.1) has certain meaning if the function f (u(y/ε, y, t)) is at least
continuous at y = ε(1 + 2k) for every k. Such a continuity condition easily
follows from the very form of equation (14.1). Indeed, if the displacement u
were step-wise discontinuous at y = ε(1 + 2k) then the derivative ∂ 2 u/∂y 2
would produce uncompensated derivatives of δ-function. So the displacement
u is at least continuous function of the coordinate y that is in match with the
physical meaning of model. Moreover, it will be shown below that appropriate
nonsmooth substitution for the spatial argument y/ε eliminates the singu-
larities from equation (14.1). As a result, new equations can be considered
within the classical theory.
In equation (14.1), the co-directed set of localized forces generated by the
elastic springs can be expressed through second derivative of the triangular
wave τ = τ (y/ε) in the form
∞
y
2f (u) δ − 1 − 2k = −f (u)sgn(τ )τ (14.2)
ε
k=−∞
Both the transverse displacement and the external loading are expressed in
terms of the non-smooth variables as follows
Finally, substituting (14.2) and (14.3) in equation (14.1) and using the differ-
ential and algebraic rules of the non-smooth argument transformation, gives
the differential equations and boundary conditions as, respectively,
∂2U ∂2V 2
2 ρ ∂ U ∂2U Q(τ, y, t)
= −2ε +ε − − (14.4)
∂τ 2 ∂y∂τ T ∂t2 ∂y 2 T
∂2V ∂2U 2
2 ρ ∂ V ∂2V P (τ, y, t)
2
= −2ε + ε 2
− 2
− (14.5)
∂τ ∂y∂τ T ∂t ∂y T
and
V |τ =±1 = 0 (14.6)
∂U ε2
|τ =±1 = ∓ f (U )|τ =±1 (14.7)
∂τ T
where the form of boundary condition (14.7) was eased by taking into account
condition (14.6).
14.1 Periodic Nonsmooth Structures 307
Equations (14.9) and (14.10) are of the same structure, except opposite signs
of the partial derivatives.
The buckling of a circular ring loaded by a discrete regular set of concen-
trated compressive forces was considered in [189]; see Fig.14.1. Taking into
account identity (14.2), the differential equation of equilibrium of such ring
can be represented in the form
∞
t
ẏ(t) = f (y(t)) − λ δ − 1 − 2k (14.11)
a
k=−∞
λ
≡ f (y(t)) + sgn(τ )τ
2
0 ≤ t ≤ 2π
where t = s/R is the arc length of the undeformed ring axis per radius, y(t)
is a six-component vector-function characterizing elastic states of the ring,
308 14 Spatially-Oscillating Structures
Fig. 14.1 Linear string on the discrete regular set of nonlinearly-elastic springs.
τ = τ (t/a) is the triangular sine wave of the period T = 4a and prime means
its Schwartz derivative, a = ε/2 is a small parameter as compared to unity, λ
is a dimensionless parameter which is proportional to the localized forces P
applied to the ring, and conditions of periodicity are imposed on the elements
of vector-function y(t).
Let us represent the unknown vector-function in the form
1
Y |τ =±1 = ± λε (14.15)
4
where
1
R(X, Y ) = [f (X + Y ) + f (X − Y )]
2
1
I(X, Y ) = [f (X + Y ) − f (X − Y )]
2
14.1 Periodic Nonsmooth Structures 309
Fig. 14.2 The circular ring under discrete regular set of compressive radial forces.
The mechanical model, which is shown in Fig. 14.3, was considered in [150]
based on the generalized (asymmetric) version of the triangular wave. In
contrast to the model shown in Fig. 14.1, the sprigs are linearly elastic and
shifted in dipole-wise manner so that the differential equation of motion with
respect to the string deflection u = u(t, y) has the form
∂ 2u ∂ 2 u ku ∂ 2 τ (y/a, γ)
ρ − T − (1 − γ 2
)sgn(τ (y/a, γ)) =0 (14.16)
∂t2 ∂y 2 a ∂(y/a)2
−∞ < y < ∞
where the triangular wave with different positive and negative slopes is given
by
z/ (1 − γ) for − 1 + γ ≤ z ≤ 1 − γ
τ = τ (z, γ) = (14.17)
(−z + 2) / (1 + γ) for 1 − γ ≤ z ≤ 3 + γ
−1 < γ < 1
∞
∂ 2 τ (z, γ)
= 2α [δ (z + 1 − γ − 4k) − δ (z − 1 + γ − 4k)] (14.20)
∂z 2
k=−∞
where α = 1/ 1 − γ 2 and β = 2γα.
Let us represent the string deflection in the form
∂2U
ρ −T
∂t2
α ∂2 ∂2 αβ ∂ 2 2α ∂ 2
× + U+ + V =0 (14.26)
a2 ∂τ 2 ∂y 2 a2 ∂τ 2 a ∂τ ∂y
∂2V
ρ 2 −T
∂t
β ∂2 2 ∂2 α + β2 ∂2 2β ∂ 2 ∂2
× + U + + + V =0
a2 ∂τ 2 a ∂τ ∂y a2 ∂τ 2 a ∂τ ∂y ∂y 2
(14.27)
14.1 Periodic Nonsmooth Structures 311
Note that some terms have been eliminated from condition (14.28) by taking
into account condition (14.24).
Further analysis of the boundary value problem (14.24) and (14.26)
through (14.28) can be implemented by using the asymptotic approach [151]
considering a as a small parameter.
Fig. 14.3 Linear string on the discrete periodic set of linearly-elastic dipole-wise
shifted springs of the stiffness k.
λ∗ = λ(1 + lτ )
μ∗ = μ(1 + mτ ) (14.29)
τ = ∂τ (x/a, γ) /∂(x/a)
where U (i) = U (i) (τ, y, z), V (i) = V (i) (τ, y, z), and W (i) = W (i) (τ, y, z).
Then, both representations, (14.29) and (14.30), are substituted into the
static equations of three-dimensional elastic bodies with variable elastic
312 14 Spatially-Oscillating Structures
Here the spatial scale ϕ = y/ε associates with the structural periodicity, the
vector-function f (z, ϕ, y) ∈ Rn is continuous with respect to z and y, but
it is allowed to be step-wise discontinuous with respect to ϕ at the points
{ϕ : τ (ϕ) = ±1}, and p(y) ∈ Rn is a continuous vector-function describing
the amplitude modulation of the localized loading.
equation (14.31) describes an elastic rod whose cross sectional area is a piece-
wise constant periodic function of the longitudinal coordinate as it is shown
in Fig. 14.4 Indeed, substituting (14.32) in (14.31) and eliminating then v(y),
gives the second-order differential equation of equilibrium for the rod
du
q(y) y
d y
EF 1 + ατ = τ . (14.33)
dy ε dy 2ε ε
Fig. 14.4 An elastic rod with a periodic non-smoothly varying cross sectional area
and concentrated loading.
14.3 Two Variable Expansions 313
1 ∂ i Rf 1 ∂ i If
Rfi = |ε=0 , Ifi = |ε=0
i! ∂εi i! ∂εi
Substituting (14.40) in (14.38) and (14.39), and matching the coefficients
of the same powers of ε, gives the corresponding sequence of equations and
boundary conditions. In particular, zero-order problem takes the form
∂X 0 ∂Y 0
= 0, =0 (14.42)
∂τ ∂τ
and
Y 0 |τ =±1 = 0 (14.43)
As follows from (14.42), the generating solution is independent on the fast
oscillating scale τ . Therefore, taking into account (14.43), gives solution
X 0 = A0 (y 0 ), Y0 ≡0 (14.44)
∂X 1 ∂Y 0
= If0 − = If (A0 , 0, τ, y 0 ) (14.45)
∂τ ∂y 0
∂Y 1 ∂X 0 dA0
= Rf0 − = R f (A0
, 0, τ, y 0
) − (14.46)
∂τ ∂y 0 dy 0
and
Y 1 |τ =±1 = p(y 0 ) (14.47)
Integrating equations (14.45) and (14.46), gives first-order terms of the
asymptotic solution
τ
1
X = If0 dτ + A1 (y 0 ) (14.48)
0
τ
dA0
Y1 = Rf0 dτ − (τ + 1) + p(y 0 )
dy 0
−1
14.4 Second Order Equations 315
1
dA0 1
= Rf0 dτ ≡< Rf0 > (14.49)
dy 0 2
−1
Note that the ‘slow scale’ equation (14.49) was obtained by satisfying the
boundary condition in contrast to the conventional scheme of two variable
expansions in which such kind of equations are obtained by eliminating the
so-called ‘resonance terms.’
Enforcing now equation (14.49), brings the component Y 1 to the final form
τ
Y =1
(Rf0 − < Rf0 >)dτ + p(y 0 ) (14.50)
−1
where ∂Rf0 /∂A0 is the Jacobian matrix, and the vector-function F 1 is known.
Note, that equation (14.51) is linear. Moreover, on the next steps, equa-
tions for the vector-functions A2 , A3 ,... will be of the same linear structure,
including the same Jacobian matrix.
and
z(y) = X(τ (ϕ), y 0 ) + Y (τ (ϕ), y 0 )τ (ϕ)
where y 0 ≡ y represents the slow spatial scale.
This leads to the boundary value problem
2
∂2X ∂2Y ∂ X
= −2ε −ε 2
+ QX + P Y − G (14.54)
∂τ 2 ∂τ ∂y 0 ∂y 02
∂2Y ∂2X ∂2Y
= −2ε −ε 2
+ P X + QY − F (14.55)
∂τ 2 ∂τ ∂y 0 ∂y 02
and
∂X
|τ =±1 = ε2 [r(y) − p(y)X]|τ =±1 (14.56)
∂τ
Y |τ =±1 = 0
X 0 = B 0 (y 0 ), Y 0 ≡ 0
and
X 1 ≡ 0, Y 1 ≡ 0
where B 0 is an arbitrary function of the slow argument y 0 .
As a result, first two non-trivial steps of the averaging procedure give
Here, notation < • > means averaging with respect to τ as defined in (14.49),
the vector-function B 0 = B 0 (y 0 ) satisfies equation
d2 B 0
+ < Q(τ, y 0 ) > B 0 =< G(τ, y 0 ) > (14.60)
dy 02
Now, in equation (14.61), the term sgn(τ )τ (ϕ) generates δ-loads of the same
direction, whereas the boundary condition (14.56) for the X-component takes
the form
∂X
|τ =±1 = ∓ε2 [r(y) − p(y)X]|τ =±1 (14.62)
∂τ
The form of expressions (14.58) and (14.60) is modified as, respectively,
τ
2
X = (τ − ξ)[G(ξ, y 0 )− < G(τ, y 0 ) > (14.63)
−1
τ2
−(Q(ξ, y 0 )− < Q(τ, y 0 ) >)B 0 ]dξ − (r − pB 0 ) + B 2 (y 0 )
2
and,
d2 B 0
+ (< Q(τ, y 0 ) > +p)B 0 =< G(τ, y 0 ) > +r (14.64)
dy 02
and the component Y 2 is still described by (14.59).
d4 w cw x x
∞
D + δ − 1 − 2k = qload (14.65)
dx4 a a L
k=−∞
−∞ < x < ∞
cL4 L4
γ= , ψ(y) = qload (y)
aD aD
where ε = a/L << 1. As a result the above equation (14.65) for the beam’
center line takes the form
d4 W 1
4
− γsgn[τ (ϕ)]τ (ϕ) W = ψ(y) (14.66)
dy 2
The first equation in (14.68) gives B20 = d2 B10 /dy 2 , whereas the second equa-
tion after elimination of B20 takes the form
d4 B10 γ
4
+ B10 = ψ(y) (14.69)
dy 2
d2 w
M (x) = D (14.70)
dx2
a 2 x
πx
= −M0 1−γ τ2 sin
2πL a L
14.5 Acoustic Waves from Non-smooth Periodic Boundary Sources 319
1.0
0.5
MxMo
0.0
0.5
1.0
0 1 2 3 4 5 6
x
Fig. 14.5 Bending moment of the beam on the discrete elastic foundation; numer-
ical values of the parameters are as follows: L = π, a = 0.2, and γ = 1948.0
Finally, note that the homogenization procedure described in [24] gives the
averaged equation in the slow spatial scale and a so-called ‘cell problem’ in
the fast scale. In the above approach, the analog of cell problem associates
with the fast oscillating spatial scale given by the triangular wave function
τ (y/ε). As a result, solution for the cell problem automatically unfolds on
the entire structure so that the fast and slow components of elastic states are
eventually expressed though the same coordinate.
Let us describe acoustic waves by the linear wave equation in the standard
form
1 ∂2P ∂2P ∂ 2P ∂2P
= + + (14.71)
c2f ∂t2 ∂x2 ∂y 2 ∂z 2
where P is a pressure deviation from the static equilibrium pressure; x, y, z
and t are spatial coordinates and time, respectively, and cf is the speed of
sound in the media.
Further, the plane problem is considered when P = P (t, y, z), and there-
fore, ∂ 2 P/∂x2 = 0. Such an assumption can be justified by sufficiently
long piezoelectric rods whose characteristics are constant along the x-co-
ordinate. Suppose that the pressure generated by the rods near the bound-
ary is P0 = A sin ωt, where A and ω are constant amplitude and frequency,
respectively.
Let the boundary condition at z = 0 to have the form
P0 (t) for (4n − 1) a ≤ y ≤ (4n + 1) a
P (t, y, 0) = (14.72)
0 for (4n + 1) a ≤ y ≤ (4n + 3) a
n = 0, ±1, ±2, ...
Note that, based on what is actually known near the fluid-source interface,
the boundary condition can also be formulated for pressure derivatives. From
the mathematical standpoint, this do not affect much the solution procedure
though.
Let us seek the steady-state solution, which is periodic with respect to t
and y and remains bounded as z → ∞.
Since the boundary condition is periodic along y-coordinate with period
T = 4a, then, according to the idea of non-smooth argument transforma-
tion, the triangular wave periodic coordinate is introduced as y → τ (y/a).
As a result, the boundary condition (14.72) and yet unknown solution are
represented as, respectively,
1 1 y
P (t, y, z)|z=0 = P0 (t) + P0 (t) τ (14.73)
2 2 a
and
P (t, y, z) = P1 (t, τ (y/a), z) + P2 (t, τ (y/a), z)τ (y/a) (14.74)
where the components P1 and P2 are considered as new unknown functions.
14.5 Acoustic Waves from Non-smooth Periodic Boundary Sources 321
P2 |τ =±1 = 0 (14.76)
∂2P 1 ∂ 2 P1 1 ∂ 2 P2 y
= + τ (14.77)
∂y 2 a2 ∂τ 2 a2 ∂τ 2 a
under the condition
∂P1
|τ =±1 = 0 (14.78)
∂τ
Note that both derivatives, (14.75) and (14.77), as well as the original function
(14.74) appear to have the same algebraic structure of hyperbolic numbers.
Obviously, differentiation with respect to t and z preserve such a structure as
well. As a result, substituting the second derivatives into differential equation
(14.71) and collecting separately terms related to each of the basis elements
{1, τ }, gives two partial differential equations for the components of repre-
sentation (14.74)
1 ∂ 2 Pi 1 ∂ 2 Pi ∂ 2 Pi
= + (14.79)
c2f ∂t2 a2 ∂τ 2 ∂z 2
(i = 1, 2)
m
(−1)k−1 1 y
A sin ω (t − Kk z) cos k − πτ + (14.81)
(k − 1/2) π 2 a
k=1
∞
k−1
(−1) 1 y y
sin ωt exp (−χk z) cos k − πτ τ
(k − 1/2) π 2 a a
k=m+1
where
*
2 2
1 π 2
ω
Kk = − k− , k = 1, ..., m
cf
2 a
*
2
2
1 π 2 ω
χk = k− − , k = m + 1, ...
2 a cf
and m is the maximum number at which the expression under the first square
root is still positive.
2
6
P
0
2 4
1
z
0 2
1
y
2
0
3
Fig. 14.7 Acoustic wave surface for the set of parameters : cf = 10.0, a = 1,
ω = 172, t = 3, and A = 2.
Fig. 14.8 Contour plot of the wave field shown in Fig. 14.7.
that shorter waves are carrying the information about the discreetness of the
wave source for a longer distance from the source.
In conclusion, solution (14.81) could be of-course obtained in terms of the
standard trigonometric expansions by applying the method of separation of
variables directly to the original problem, (14.71) and (14.72). However, the
derivation of solution (14.81) implies no integration of discontinuous func-
tions, since all the discontinuities have been captured in advance by trans-
formation (14.74).
It is also worth to note that, the terms of series (14.81) are calculated
on the standard interval, −1 ≤ τ ≤ 1, which is covered by one half of the
total period, whereas the standard Fourier expansions must be built over the
entire period. This is due to the fact that representation (14.74) automatically
unfolds the half-period domain on the infinite spatial interval.
Fig. 14.9 Acoustic wave surface for the set of parameters: cf = 10.0, a = 1,
ω = 86, t = 3, and A = 2.
Fig. 14.10 Contour plot of the wave field shown in Fig. 14.9.
14.6 Spatio-temporal Periodicity 325
where the function f is periodic with temporal period Tt = 2π/ω and spatial
period Ty = 4a.
Introducing the triangular wave spatial argument, τy = τ (y/a), gives
where
1
F1 (t, τy ) = [f (t, aτy ) + f (t, 2a − aτy )]
2
1
F2 (t, τy ) = [f (t, aτy ) − f (t, 2a − aτy )] (14.84)
2
In a similar way, introducing the triangular wave temporal argument, τt =
τ (2ωt/π), into both of the components, F1 and F2 , gives eventually expression
of the form
f (t, y) = f0 (τt , τy )e0 + f1 (τt , τy )e1 + f2 (τt , τy )e2 + f3 (τt , τy )e3 (14.85)
e0 = 1
e1 = τ (2ωt/π)
e2 = τ (y/a) (14.86)
e3 = e1 e2
× e0 e1 e2 e3
e0 1 e1 e2 e3
e1 e1 1 e3 e2 (14.87)
e2 e2 e3 1 e1
e3 e3 e2 e1 1
Fig. 14.11 Fragment of the map of periodic elastic foundation; a = 1.0 and b = 2.0.
With reference to Fig. 14.11, function (14.92) is defined on the infinite plane,
such that
0 (x, y) ∈ any “dark field”
K(x, y) = (14.93)
k (x, y) ∈ any “light field”
In the same way, Fig. 14.12 provides maps for the elements of basis
e0 = 1
e1 = τ (x/a)
e2 = τ (y/b) (14.94)
e3 = e1 e2
The above table of products (14.87) still valid for basis (14.94). As a result,
function (14.92) takes eventually the form
k
K(x, y) = (e0 + e1 + e2 + e3 ) (14.95)
4
Now let us represent the membrane deflection in the form
Fig. 14.12 The standard basis map: each of the elements is equal to unity within
light domains and zero within dark domains.
k
Ku = (X + Y + Z + W )(e0 + e1 + e2 + e3 ) (14.97)
4
14.7 Membrane on a Two-Dimensional Periodic Foundation 329
Then, taking into account (14.87), (14.94), (14.98) and (14.99), gives first
derivatives of (14.96) in the form
∂u 1 ∂Y ∂X 1 ∂X ∂Y
= + e0 + + e1
∂x a ∂τx ∂x a ∂τx ∂x
1 ∂W ∂Z 1 ∂Z ∂W
+ + e2 + + e3 (14.100)
a ∂τx ∂x a ∂τx ∂x
de1 (x/a) 1
+(Y + W e2 )
d(x/a) a
∂u 1 ∂Z ∂X 1 ∂W ∂Y
= + e0 + + e1
∂y b ∂τy ∂y b ∂τy ∂y
1 ∂X ∂Z 1 ∂Y ∂W
+ + e2 + + e3 (14.101)
b ∂τy ∂y b ∂τy ∂y
de2 (y/b) 1
+(Z + W e1 )
d(y/b) b
Y |τx =±1 = 0
W |τx =±1 = 0 (14.102)
and
Z|τy =±1 = 0
W |τy =±1 = 0 (14.103)
ux = Dx u (14.105)
uy = Dy u
where
⎡ ⎤
∂/∂x a−1 ∂/∂τx 0 0
⎢ a−1 ∂/∂τx ∂/∂x 0 0 ⎥
Dx = ⎢
⎣0
⎥ (14.106)
0 ∂/∂x a−1 ∂/∂τx ⎦
0 0 a−1 ∂/∂τx ∂/∂x
and ⎡ ⎤
∂/∂y 0 b−1 ∂/∂τy 0
⎢0 ∂/∂y 0 b−1 ∂/∂τy ⎥
Dy = ⎢
⎣ b−1 ∂/∂τy
⎥
⎦ (14.107)
0 ∂/∂y 0
0 b−1 ∂/∂τy 0 ∂/∂y
These differential matrix operators automatically generate high-order deriva-
tives of combination (14.96) provided that necessary smoothness (boundary)
conditions hold. For instance, the components of expansion for Δu are given
by the elements of vector-column (D2x + D2y )u under conditions
1 ∂X ∂Y
+ |τx =±1 = 0
a ∂τx ∂x
1 ∂Z ∂W
+ |τx =±1 = 0 (14.108)
a ∂τx ∂x
1 ∂X ∂Z
+ |τy =±1 = 0
b ∂τy ∂y
1 ∂Y ∂W
+ |τy =±1 = 0 (14.109)
b ∂τy ∂y
Note that equations (14.110) have constant coefficients whereas the step-wise
discontinuities of foundation have been absorbed by the triangular wave argu-
ments τx and τy . The corresponding boundary conditions (14.112) generate a
so-called “cell problem” [76] within the rectangular domain {−1 ≤ τx ≤ 1,
−1 ≤ τy ≤ 1}. Therefore, the arguments τx and τy locally describe the fast
varying component of membrane shape. In contrast, the explicitly present co-
ordinates x and y describe the slow component in the infinite plane {−∞ <
x < ∞, −∞ < y < ∞}. Finally, the increase in number of equations (14.110)
of-course complicates solution procedures from the technical standpoint. How-
ever, the obvious symmetry of equations helps to ease the corresponding deriva-
tions. For instance, equations (14.110) can be decoupled by introducing new
unknown functions Ui = Ui (τx , τy , x, y) (i = 1, .., 4) as follows
U1 = X + Y + Z + W
U2 = X + Y − Z − W
U3 = X − Y + Z − W (14.113)
U4 = X − Y − Z + W
332 14 Spatially-Oscillating Structures
The main reason for using basis (14.115) is that its table of products has the
normalized diagonal form
ik in = δkn (14.117)
where δkn is the Kronecker symbol.
The geometrical meaning of property (14.117) follows from the maps in
Fig. 14.13.
14.8 The Idempotent Basis for Two-Dimensional Structures 333
Fig. 14.13 The map of idempotent basis: each of the elements is equal to unity
within light domains and zero within dark domains.
4
u(x, y) = Uk (τx , τy , x, y)ik (14.118)
k=1
As a result,
4
4
f( Uk ik ) = f (Uk )ik (14.119)
k=1 k=1
∂u(x, y)
(14.120)
∂x
4
1 ∂Uk (τx , τy , x, y) ∂Uk (τx , τy , x, y) ∂ik
= e1 ik + ik + Uk (τx , τy , x, y)
a ∂τx ∂x ∂x
k=1
and
∂i1 ∂e+
1 + 1 ∂e1 +
= e = e
∂x ∂x 2 2 ∂x 2
∂i2 ∂e+ 1 ∂e1 −
= 1 e− = e
∂x ∂x 2 2 ∂x 2
∂i3 ∂e− 1 ∂e1 +
= 1 e+ =− e
∂x ∂x 2 2 ∂x 2
∂i4 ∂e− 1 ∂e1 −
= 1 e− =− e
∂x ∂x 2 2 ∂x 2
As a result, derivative (14.120) takes the form
∂u 1 ∂U1 ∂U1 1 ∂U2 ∂U2
= + i1 + + i2
∂x a ∂τx ∂x a ∂τx ∂x
1 ∂U3 ∂U3 1 ∂U4 ∂U4
+ − + i3 + − + i4 (14.121)
a ∂τx ∂x a ∂τx ∂x
1 ∂e1 + 1 ∂e1 −
+ (U1 − U3 ) e + (U2 − U4 ) e
2 ∂x 2 2 ∂x 2
Analogously, one obtains
∂u 1 ∂U1 ∂U1 1 ∂U2 ∂U2
= + i1 + − + i2
∂y b ∂τy ∂y b ∂τy ∂y
1 ∂U3 ∂U3 1 ∂U4 ∂U4
+ + i3 + − + i4 (14.122)
b ∂τy ∂y b ∂τy ∂y
1 ∂e2 + 1 ∂e2 −
+ (U1 − U2 ) e + (U3 − U4 ) e
2 ∂y 1 2 ∂y 1
14.8 The Idempotent Basis for Two-Dimensional Structures 335
Let us introduce vector, associated with expansion (14.118), and the corre-
sponding differential matrix operators as, respectively,
⎡ ⎤
U1
⎢ U2 ⎥
u =⎢ ⎥
⎣ U3 ⎦ (14.123)
U4
and
⎡1 ∂ ∂ ⎤
a ∂τx + ∂x 0 0 0
⎢0 1 ∂
+ ∂
0 0 ⎥
Dx = ⎢
⎣0
a ∂τx ∂x ⎥
⎦ (14.124)
0 − a1 ∂τ∂x + ∂
∂x 0
0 0 0 − a1 ∂τ∂x + ∂
∂x
⎡1 ∂ ∂ ⎤
b ∂τy + ∂y 0 0 0
⎢0 − 1b ∂τ∂y + ∂ ⎥
⎢ ∂y 0 0 ⎥
Dy = ⎢ ⎥ (14.125)
⎣0 0 1 ∂
b ∂τy + ∂
∂y 0 ⎦
0 0 0 − 1b ∂τ∂y + ∂
∂y
Substituting (14.118) into the original equation (14.91), using the differenti-
ation rules for idempotent basis, and assuming that a = b = ε, gives
2
∂ U1 ∂ 2 U1
Δτ U1 + 2ε + + ε2 ΔU1 = ε2 [q(x, y) + kU1 ]
∂τx ∂x ∂τy ∂y
2
∂ U2 ∂ 2 U2
Δτ U2 + 2ε − + ε2 ΔU2 = ε2 q(x, y)
∂τx ∂x ∂τy ∂y
2
∂ U3 ∂ 2 U3
Δτ U3 − 2ε − + ε2 ΔU3 = ε2 q(x, y) (14.126)
∂τx ∂x ∂τy ∂y
2
∂ U4 ∂ 2 U4
Δτ U4 − 2ε + + ε2 ΔU4 = ε2 q(x, y)
∂τx ∂x ∂τy ∂y
where the notations Δτ and Δ have the same meaning as those in equations
(14.110).
Equations (14.126) are decoupled, at cost of coupling the boundary con-
ditions though
∂(U1 − U3 ) ∂(U1 − U2 )
|τx =±1 = 0, |τy =±1 = 0
∂τx ∂τy
(U1 − U3 )|τx =±1 = 0, (U1 − U2 )|τy =±1 = 0
∂(U2 − U4 ) ∂(U3 − U4 )
|τx =±1 = 0, |τy =±1 = 0 (14.127)
∂τx ∂τy
(U2 − U4 )|τx =±1 = 0, (U3 − U4 )|τy =±1 = 0
336 14 Spatially-Oscillating Structures
Note that both boundary value problems (14.110) through (14.112) and
(14.126) through (14.127) implement the transition from two to four spa-
tial arguments: {x, y} → {τx , τy , x, y}. The arguments τx and τy naturally
relate to cell problems and incorporate the corresponding elastic components
within the class of closed form solutions.
Fig. 14.14 The map of idempotent basis generated by the asymmetric triangular
waves with parameters: a = b = 1.0, γ1 = 0.2 and γ2 = 0.6; each of the elements is
equal to unity within light domains and zero within dark domains.
1
e+
i = [1 − γi + (1 − γi2 )ei ]
2
1
e−
i = [1 + γi − (1 − γi2 )ei ] (14.128)
2
(i = 1, 2)
1. Acary, V., Brogliato, B.: Numerical methods for nonsmooth dynamical sys-
tems: Applications in Mechanics and Electronics. Springer, Heidelberg (2008)
2. Andreaus, U., Casini, P., Vestroni, F.: Non-linear dynamics of a cracked can-
tilever beam under harmonic excitation. International Journal of Non-Linear
Mechanics 42(3), 566–575 (2007)
3. Andrianov, I.V.: Asymptotic solutions for nonlinear systems with high degrees
of nonlinearity. Prikl. Matem. Mekhanika (PMM) 57(5), 941–943 (1993)
4. Andrianov, I.V., Awrejcewicz, J.: Methods of small and large δ in the nonlinear
dynamics—a comparative analysis. Nonlinear Dynam. 23(1), 57–66 (2000)
5. Andrianov, I.V.: Asymptotics of nonlinear dynamical systems with a high
degree of nonlinearity. Doklady Mathematics 66(2), 270–273 (2002)
6. Andrianov, I.V., Awrejcewicz, J., Barantsev, R.G.: Asymptotic approaches
in mechanics: New parameters and procedures. Applied Mechanics Re-
views 56(1), 87–110 (2003)
7. Antonuccio, F.: Hyperbolic numbers and the Dirac spinor (1998),
http://arxiv.org/abs/hep-th/9812036v1
8. Arnol d, V.I.: Mathematical methods of classical mechanics. Springer, New
York (1978)
9. Arnol d, V.I.: Mathematical methods of classical mechanics. Springer, Heidel-
berg (1978)
10. Ascher, U.M., Mattheij, R.M.M., Russell, R.D.: Numerical solution of bound-
ary value problems for ordinary differential equations. Classics in Applied
Mathematics, vol. 13. Society for Industrial and Applied Mathematics (SIAM),
Philadelphia (1995); Corrected reprint of the 1988 original
11. Atkinson, C.P.: On the superposition method for determining frequencies of
nonlinear systems. In: ASME Proceedings of the 4th National Congress of
Applied Mechanics, pp. 57–62 (1962)
12. Auerbach, D., Cvitanovic, P., Eckmann, J.-P., Gunaratne, G., Procaccia, I.:
Exploring chaotic motion through periodic orbits. Phys. Rev. Lett. 58(23),
2387–2389 (1987)
13. Awrejcewicz, J., Bajaj, A.K., Lamarque, C.-H. (eds.): Nonlinearity, bifurca-
tion and chaos: the doors to the future. Part II. World Scientific Publishing
Co., Singapore (1999); Papers from the International Conference held in Do-
bieszków, September 16-18 (1996); Internat. J. Bifur. Chaos Appl. Sci. Engrg.
9(3) (1999)
340 References
14. Awrejcewicz, J., Lamarque, C.-H.: Bifurcation and Chaos in Nonsmooth Me-
chanical Systems. World Scientific, Singapore (2003)
15. Azeez, M.A.F., Vakakis, A.F., Manevitch, L.I.: Exact solutions of the problem
of vibroimpact oscillations of a discrete system with two degrees of freedom.
Prikl. Mat. Mekh. 63(4), 549–553 (1999)
16. Azeez, M.A.F., Vakakis, A.F., Manevich, L.I.: Exact solutions of the problem
of vibro-impact oscillations of a discrete system with two degrees of freedom.
Prikl. Mat. Mekh. 63(4), 549–553 (1999)
17. Babitsky, V.I.: Theory of Vibroimpact Systems and Applications. Springer,
Berlin (1998)
18. Bahler, T.B.: Mathematica for Scientists and Engineers. Addison-Wesley, New
York (1995)
19. Baker Jr., G.A., Graves-Morris, P.: Padè Approximants, 2nd edn. Encyclope-
dia of Mathematics and Its Applications, vol. 59. Cambridge University Press,
Cambridge (1987)
20. Balescu, R.: Statistical Dynamics, Matter out of Equilibrium. Imperial College
Press, Singapore (1997)
21. Bateman, H., Erdelyi, A.: Higher Transcendental Functions. McGraw-Hill,
New York (1955)
22. Belinfante, J.G.F., Kolman, B.: A survey of Lie groups and Lie algebras with
applications and computational methods. Society for Industrial and Applied
Mathematics (SIAM), Philadelphia (1989); Reprint of the 1972 original
23. Bellman, R.: Introduction to Matrix Analysis. McGraw-Hill Company, New
York (1960)
24. Bensoussan, A., Lions, J.-L., Papanicolaou, G.: Asymptotic analysis for peri-
odic structures. North-Holland Publishing Co., Amsterdam (1978)
25. Blazejczyk-Okolewska, B., Czolczynski, K., Kapitaniak, T., Wojewoda, J.:
Chaotic Mechanics in Systems with Impacts and Friction. World Scientific,
Singapore (1999)
26. Boettcher, S., Bender, C.M.: Nonperturbative square-well approximation to a
quantum theory. Journal of Mathematical Physics 31(11), 2579–2585 (1990)
27. Bogoliubov, N., Mitropollsky, Y.: Asymptotic Methods in the Theory of Non-
linear Oscillations. Gordon and Breach, New York (1961)
28. Bogoljubow, N.N., Mitropolski, J.A.: Asymptotische Methoden in der Theorie
der nichtlinearen Schwingungen. Akademie-Verlag, Berlin (1965)
29. Brogliato, B.: Nonsmooth Mechanics: Models, Dynamics and Control.
Springer, Berlin (1999)
30. Brogliato, B.: Impacts in Mechanical Systems: Analysis and Modelling.
Springer-Verlag, Berlin (2000)
31. Caughey, T.K., Vakakis, A.F.: A method for examining steady state solutions
of forced discrete systems with strong non-linearities. International Journal of
Non-Linear Mechanics 26(1), 89–103 (1966)
32. Chati, M., Rand, R., Mukherjee, S.: Modal analysis of a cracked beam. Journal
of Sound and Vibration 207, 249–270 (1997)
33. Chen, S., Shaw, S.W.: Normal modes for piecewise linear vibratory systems.
Nonlinear Dynamics 10, 135–163 (1996)
34. Cooper, K., Mickens, R.E.: Generalized harmonic balance/numerical method
for determining analytical approximations to the periodic solutions of the x4/3
potential. Journal of Sound and Vibration 250, 951–954 (2002)
References 341
35. Coppola, V.T., Rand, R.H.: Computer algebra implementation of Lie trans-
forms for hamiltonian systems: Application to the nonlinear stability of l4.
ZAMM 69(9), 275–284 (1989)
36. Cveticanin, L.: Oscillator with strong quadratic damping force. Publications
de L’institut Mathematique (Nouvelle serie) 85(99), 119–130 (2009)
37. Dankowicz, H., Paul, M.R.: Discontinuity-induced bifurcations in systems with
hysteretic force interactions. Journal of Computational and Nonlinear Dynam-
ics 4(Article 041009), 1–6 (2009)
38. Deprit, A.: Canonical transformations depending on a parameter. Celestial
mechanics (1), 1–31 (1969)
39. Dimentberg, M.F.: Statistical Dynamics of Nonlinear and Time-Varying Sys-
tems. John Wiley & Sons, New York (1988)
40. Dimentberg, M.F., Bratus, A.S.: Bounded parametric control of random vibra-
tions. R. Soc. Lond. Proc. Ser. A Math. Phys. Eng. Sci. 456(2002), 2351–2363
(2000)
41. Dimentberg, M.F., Iourtchenko, D.V., Bratus’, A.S.: Transition from planar
to whirling oscillations in a certain nonlinear system. Nonlinear Dynamics 23,
165–174 (2000)
42. Feeny, B.A., Guran, A., Hinrichs, N., Popp, K.: A historical review on dry
friction and stick-slip phenomena. ASME Applied Mechanics Reviews 51, 321–
341 (1998)
43. Ferrari, L., Boschi, C.D.E.: Nonautonomous and nonlinear effects in general-
ized classical oscillators: A boundedness theorem. Physical Review E 62(3),
R3039–R3042 (2000)
44. Fidlin, A.: Nonlinear Oscillations in Mechanical Engineering. Springer, Hei-
delberg (2005)
45. Filippov, A.F.: Differential equations with discontinuous righthand sides.
Kluwer Academic Publishers Group, Dordrecht (1988) (Translated from the
Russian)
46. Fucik, S., Kufner, A.: Nonlinear differential equations. Elsevier, Amsterdam
(1980); Studies in Applied Mechanics 2. Elsevier Scientific Publishing Com-
pany, Amsterdam
47. Gendelman, O., Manevitch, L.I., Vakakis, A.F., M’Closkey, R.: Energy pump-
ing in nonlinear mechanical oscillators. I. Dynamics of the underlying Hamil-
tonian systems. Trans. ASME J. Appl. Mech. 68(1), 34–41 (2001)
48. Gendelman, O.V.: Modeling of inelastic impacts with the help of smooth func-
tions. Chaos, Solitons and Fractals 28, 522–526 (2006)
49. Gendelman, O.V., Manevitch, L.I.: Discrete breathers in vibroimpact chains:
Analytic solutions. Physical Review E 78(026609) (2008)
50. Giacaglia, G.E.O.: Perturbation methods in non-linear systems. Springer, New
York (1972); Applied Mathematical Sciences, Vol. 8
51. Goldsmith, W.: Impact: The Theory and Physical Behaviour of Colliding.
Courier Dover Publications, North Chelmsford (2001)
52. Gradshteyn, I.S., Ryzhik, I.M.: Table of Integrals, Series, and Products, 5th
edn. Academic Press, Boston (1994)
53. Grebogi, C., Ott, E., Yorke, J.A.: Unstable periodic orbits and the dimensions
of multifractal chaotic attractorscontrolling chaos. Physical Review A 37(5),
1711–1724 (1988)
54. Guckenheimer, J., Meloon, B.: Computing periodic orbits and their bifur-
cations with automatic differentiation. SIAM J. Sci. Comput. 22(3), 951–
985(electronic) (2000)
342 References
55. Guran, A., Pfeiffer, F., Popp, K.: Dynamics with Friction: Modeling, Analysis
and Experiments. World Scientific, Singapore (2001)
56. Hahn, W.: Stability of motion. Springer Series in Nonlinear Dynamics.
Springer, New York (1967)
57. Harvey, T.J.: Natural forcing functions in nonlinear systems. ASME Journal
of Applied Mechanics 25, 352–356 (1958)
58. Hascoët, E., Herrmann, H.J., Loreto, V.: Shock propagation in a granular
chain. Phys. Rev. E 59(3), 3202–3206 (1999)
59. Holm, D.D., Lynch, P.: Stepwise precession of the resonant swinging spring.
SIAM J. Applied Dynamical Systems 1(1), 44–64 (2002)
60. Hong, J., Ji, J.-Y., Kim, H.: Power laws in nonlinear granular chain under
gravity. Phys. Rev. Lett. 82(15), 3058–3061 (1999)
61. Hori, G.: Theory of general perturbations with unspecified canonical variables.
Publ. Astron. Soc. Japan 18(4), 287–296 (1966)
62. Hori, G.: Mutual perturbations of 1: 1 commensurable small bodies with the
use of the canonical relative coordinates. I. In: Resonances in the motion of
planets, satellites and asteroids, pp. 53–66. Univ. São Paulo, São Paulo (1985)
63. Hu, H., Xiong, Z.-G.: Oscillations in an x(2m+2)/(2n+1) potential. Journal of
Sound and Vibration 259, 977–980 (2003)
64. Hunt, K.H., Crossley, F.R.E.: Coefficient of restitution interpreted as damping
in vibroimpact. ASME Journal of Applied Mechanics 97, 440–445 (1975)
65. Hutchins, C.M.: A history of violin research. J. Acoust. Soc. Am. 73(5), 1421–
1440 (1983)
66. Ibrahim, R.A., Pilipchuk, V.N., Ikeda, T.: Recent advances in liquid sloshing
dynamics. Applied Mechanics Reviews 54(2), 133–199 (2001)
67. Ibrahim, R.A.: Liquid Sloshing Dynamics. Cambridge University Press, New
York (2005)
68. Ibrahim, R.A.: Vibro-Impact Dynamics: Modeling, Mapping and Applications.
LNACM, vol. 43. Springer, Heidelberg (2009)
69. Ibrahim, R.A., Babitsky, V.I., Okuma, M. (eds.): Vibro-Impact Dynamics of
Ocean Systems and Related Problems. Springer, Heidelberg (2009)
70. Ince, E.L.: Ordinary Differential Equations. Dover, New York (1956)
71. Iomin, A., Fishman, S., Zaslavsky, G.M.: Quantum localization for a kicked
rotor with accelerator mode islands. Physical Review E 65(036215) (2002)
72. Ivanov, A.P.: Dynamics of Systems with Mechanical Collisions. International
Program of Education, Moscow (1997) (in Russian)
73. Ivanov, A.P.: Impact oscillations: linear theory of stability and bifurcations.
Journal of Sound and Vibration 178(3), 361–378 (1994)
74. Jackson, L.B.: Signals, Systems, and Transforms. Addison-Wesley Publishing
Company, New York (1991)
75. Jiang, D., Pierre, C., Shaw, S.W.: Large-amplitude non-linear normal modes of
piecewise linear systems. Journal of Sound and Vibration 272, 869–891 (2004)
76. Kalamkarov, A.L., Andrianov, I.V., Danishevskyy, V.V.: Asymptotic ho-
mogenization of composite materials and structures. Applied Mechanics Re-
views 62(030802), 1–20 (2009)
77. Kamenkov, G.V.: Izbrannye trudy v dvukh tomakh. Tom. I. Izdat, Nauka,
Moscow (1971); Ustoichivost dvizheniya. Kolebaniya. Aerodinamika. [Stability
of motion. Oscillations. Aerodynamics], With a biography of G. V. Kamenkov,
a survey article on his works by V. G. Veretennikov, A. S. Galiullin, S. A.
Gorbatenko and A. L. Kunicyn, and a bibliography, Edited by N. N. Krasovskiı̆
References 343
100. Manevich, L.I., Mikhlin, Y.V., Pilipchuk, V.N.: Metod normalnykh kolebanii
dlya sushchestvenno nelineinykh sistem, Nauka, Moscow (1989)
101. Manevich, L.I.: New approach to beating phenomenon in coupled nonlinear
oscillatory chains. Archive of Applied Mechanics 77, 301–312 (2007)
102. Manevitch, L.I.: The description of localized normal modes in a chain of non-
linear coupled oscillators using complex variables. Nonlinear Dynamics 25,
95–109 (2001)
103. Manevitch, L.I., Gendelman, O.V.: Oscillatory models of vibro-impact type
for essentially non-linear systems. Proceedings of the Institution of Mechanical
Engineers, Part C: Journal of Mechanical Engineering Science 222(10), 2007–
2043 (2008), doi: 10.1243/09544062JMES1057
104. Manevitch, L.I., Musienko, A.I.: Limiting phase trajectory and beating phe-
nomena in systems of coupled nonlinear oscillators. In: 2nd International Con-
ference on Nonlinear Normal Modes and Localization in Vibrating Systems,
Samos, Greece, June 19-23, pp. 25–26 (2006)
105. Marsden, J.E.: Basic complex analysis. Freeman, San Francisco (1973)
106. Maslov, V.P., Omel janov, G.A.: Asymptotic soliton-like solutions of equations
with small dispersion. Uspekhi Mat. Nauk. 36(3), 63–126 (1981)
107. Mickens, R.E.: Oscillations in an x4/3 potential. J. Sound Vibration 246, 375–
378 (2001)
108. Mikhlin, Y.V., Reshetnikova, S.N.: Dynamical interaction of an elastic system
and a vibro-impact absorber. Mathematical Problems in Engineering (Article
ID 37980), 15 (2006)
109. Mikhlin, Y.V., Volok, A.M.: Solitary transversal waves and vibro-impact mo-
tions in infinite chains and rods. International Journal of Solids and Struc-
tures 37, 3403–3420 (2000)
110. Mikhlin, Y.V., Zhupiev, A.L.: An application of the Ince algebraization to the
stability of the non-linear normal vibration modes. Internat. J. Non-Linear
Mech. 32(2), 393–409 (1997)
111. Minorsky, N.: Introduction to non-linear mechanics. J.W. Edwards, Ann Arbor
(1947)
112. Mitropl’sky, Y.A., Senik, P.M.: Construction of asymptotic solution of an au-
tonomouse system with strong nonlinearity. Doklady AN Ukr.SSR (Ukrainian
Academy of Sciences Reports) 6, 839–844 (1961)
113. Moon, F.C.: Chaotic Vibrations. John Willey & Sons, New York (1987)
114. Moser, J.: Recent developments in the theory of Hamiltonian systems. SIAM
Rev. 28(4), 459–485 (1986)
115. Moser, J.K.: Lectures on Hamiltonian systems. In: Mem. Amer. Math. Soc.
No. 81, p. 60. Amer. Math. Soc., Providence (1968)
116. Nayfeh, A.H.: Perturbation methods. John Wiley & Sons, New York (1973);
Pure and Applied Mathematics
117. Nayfeh, A.H.: Perturbation methods in nonlinear dynamics. In: Nonlinear
dynamics aspects of particle accelerators (Santa Margherita di Pula, 1985),
pp. 238–314. Springer, Berlin (1986)
118. Nayfeh, A.H.: Method of normal forms. John Wiley & Sons Inc., New York
(1993); A Wiley-Interscience Publication
119. Nayfeh, A.H.: Nonlinear interactions: analytical computational, and exper-
imental methods. John Wiley & Sons Inc., New York (2000); A Wiley-
Interscience Publication
References 345
120. Nayfeh, A.H., Balachandran, B.: Applied nonlinear dynamics. John Wiley
& Sons Inc, New York (1995) Analytical, computational, and experimental
methods. A Wiley-Interscience Publication
121. Nesterenko, V.F.: Dynamics of Heterogeneous Materials. Springer, New York
(2001)
122. Nesterov, S.V.: Examples of nonlinear Klein-Gordon equations, solvable in
terms of elementary functions. In: Proceedings of Moscow Institute of Power
Engineering, vol. 357, pp. 68–70 (1978)
123. Ott, E., Grebogi, C., Yorke, J.A.: Controlling chaos. Phys. Rev. Lett. 64(11),
1196–1199 (1990)
124. Ozorio de Almeida, A.M.: Hamiltonian systems: chaos and quantization. Cam-
bridge University Press, Cambridge (1988)
125. Parker, T.S., Chua, L.O.: Practical numerical algorithms for chaotic systems.
Springer, New York (1989)
126. Peat, F.D.: Synchronicity: the bridge between matter and mind. Bantam
Books, New York (1988)
127. Peterka, F.: Introduction to Oscillations of Mechanical Systems with Internal
Impacts (in Czech). Academia, Prague (1981)
128. Pfeiffer, F.: Mechanical system dynamics. Springer, Heidelberg (2008)
129. Pfeiffer, F., Glocker, C.: Multibody dynamics with unilateral contacts. Wiley,
New York (1996)
130. Pfeiffer, F., Kunert, A.: Rattling models from deterministic to stochastic pro-
cesses. Nonlinear Dynamics 1(1), 63–74 (1990)
131. Pierce, J.R.: Coupling of modes of propagation. Journal of Applied
Physics 25(2), 179–183 (1954)
132. Pilipchuk, V.N.: The calculation of strongly nonlinear systems close to vi-
broimpact systems. Journal of Applied Mathematics and Mechanics 49(5),
572–578 (1985)
133. Pilipchuk, V.N.: Transformation of oscillating systems by means of a pair of
nonsmooth periodic functions. Dokl. Akad. Nauk Ukrain. SSR Ser. A (4),
37–40, 87 (1988)
134. Pilipchuk, V.N.: Transformation of the vibratory-systems by means of a pair
of nonsmooth periodic-functions. Dopovidi Akademii Nauk Ukrainskoi Rsr
Seriya A-Fiziko-Matematichni Ta Technichni Nauki (in Ukrainian) 4, 36–38
(1988)
135. Pilipchuk, V.N.: On the computation of periodic processes in mechanical sys-
tems with the impulsive excitation. In: XXXI Sympozjon “Modelowanie w
Mechanice”, Zeszyty Naukowe Politechniki Slaskiej, Z.107, Gliwice (Poland),
pp. 335–342. Politechnica Slaska (1992)
136. Pilipchuk, V.N.: On special trajectories in configuration space of non - linear
vibrating systems. Mekhanika Tverdogo Tela (Mechanics of Solids) 3, 36–47
(1995)
137. Pilipchuk, V.N.: Analytical study of vibrating systems with strong non-
linearities by employing saw-tooth time transformations. J. Sound Vibra-
tion 192(1), 43–64 (1996)
138. Pilipchuk, V.N.: On the computation of mechanical systems with impulse
excitation. Prikl. Mat. Mekh. 60(2), 223–232 (1996)
139. Pilipchuk, V.N.: Application of special nonsmooth temporal transformations
to linear and nonlinear systems under discontinuous and impulsive excitation.
Nonlinear Dynam. 18(3), 203–234 (1999)
346 References
178. Stakgold, I.: Green’s Functions and Boundary Value Problems. Wiley Inter-
science, New York (1979)
179. Starushenko, G., Krulik, N., Tokarzewski, S.: Employment of non-symmetrical
saw-tooth argument transformation method in the elasticity theory for layered
composites. International Journal of Heat and Mass Transfer 45, 3055–3060
(2002)
180. Stronge, W.J.: Impact Mechanics. Cambridge University Press, Cambridge
(2000)
181. Thomsen, J.J., Fidlin, A.: Near-elastic vibro-impact analysis by discontinuous
transformations and averaging. Journal of Sound and Vibration 311, 386–407
(2008)
182. Timoshenko, S.P., Yang, D.H., Wiver, U.: Kolebaniya v inzhenernom dele.
Mashinostroeniye, Moscow (1985)
183. Tippetts, J.R.: Analysis of idealised oscillatory pipe flow. In: 2nd Interna-
tional Symposium on Fluid - Control, Measurement, Mechanics - and Flow
Visualisation, Sheffield, England, September 5-9 (1988)
184. Toda, M.: Nonlinear lattice and soliton theory. IEEE Transactions on Circuits
and Systems 30(8), 542–554 (1983)
185. Turner, J.D.: On the simulation of discontinuous functions. Journal of Applied
Mechanics 68, 751–757 (2001)
186. Ueda, Y.: Randomly transitional phenomena in the system governed by Duff-
ing’s equation. J. Statist. Phys. 20(2), 181–196 (1979)
187. Ulrych, S.: Relativistic quantum physics with hyperbolic numbers. Physics
Letters B 625, 313 (2005)
188. Uzunov, I.M., Muschall, R., Golles, M., Kivshar, Y.S., Malomed, B.A., Led-
erer, F.: Pulse switching in nonlinear fiber directional couplers. Phys. Rev.
E 51, 2527–2537 (1995)
189. Vakakis, A.F., Atanackovic, T.M.: Buckling of an elastic ring forced by a
periodic array of compressive loads. ASME Journal of Applied Mechanics 66,
361–367 (1999)
190. Vakakis, A.F., Manevitch, L.I., Mikhlin, Y.V., Pilipchuk, V.N., Zevin, A.A.:
Normal modes and localization in nonlinear systems. John Wiley & Sons Inc.,
New York (1996); A Wiley-Interscience Publication
191. Vedenova, E.G., Manevich, L.I., Pilipchuk, V.N.: Normal oscillations of a
string with concentrated masses on nonlinearly elastic supports. Prikl. Mat.
Mekh. 49(2), 203–211 (1985)
192. Vestroni, F., Luongo, A., Paolone, A.: A perturbation method for evaluating
nonlinear normal modes of a piecewise linear two-degrees-of-freedom system.
Nonlinear Dynamics 54(4), 379–393 (2008)
193. Waluya, S.B., van Horssen, W.T.: On the periodic solutions of a generalized
non-linear Van-der-Pol oscillator. Journal of Sound and Vibration 268, 209–
215 (2003)
194. Whitham, G.B.: Linear and nonlinear waves. John Wiley & Sons Inc., New
York (1999); Reprint of the 1974 original. A Wiley-Interscience Publication
195. Whittaker, E.T., Watson, G.N.: A Course of Modern Analysis, 4th edn. Cam-
bridge University Press, Cambridge (1986)
196. Wiercigroch, M., de Kraker, B. (eds.): Applied Nonlinear Dynamics ans Chaos
of Mechanical Systems with Discontinuities, vol. 28. World Scientific, Singa-
pore (2000)
References 349
n 3;
AnalyticalSolution STSRn, Α;
DoXi1 AnalyticalSolution1i, i, 1, n
1;
DoHi1 AnalyticalSolution2i, i, 1, n;
DoPrint"X", i, "", Xi
Simplify, i, 0, n
DoPrint"H", i, "", Hi
Simplify, i, 0, n 1
X0A Τ
A Τ2Α
X1
2Α
A2 Α Α Τ A Τ1Α AΑ 3 2 Α 2 Α Τ A ΤΑ
X2
2 2 Α2 3 2 Α
H0A1Α 1 Α
A1Α Α 1 Α
H1
2 2 Α
A1Α Α 1 Α3
H2
2 2 Α2 3 2 Α
i NLY2k
1
KY2i , i, 1, m
;
k0 Factorial2k
2m Τi Τi
2
Ifsmoothness, X RE
. Τ 0
KXi ,
i1 i i
2
2m 2m NLXkΤk1
Y KYiΤi Τi
2 , X ,
i0 k1 Factorialk 1
2m NLYkΤk1
Y
k1 Factorialk 1
2
T 4;
fnosmootht, 1
fnosmootht, 2
354 Appendix
fnosmootht, 3
fnosmootht, 4
fnosmootht, 5
Out[5]= 0
1
Out[6]= Π3 Τt3
8
1 1
Out[7]= Π3 Τt3 Π5 Τt5
8 64
1 1 13 Π7 Τt7
Out[8]= Π3 Τt3 Π5 Τt5
8 64 15 360
1 1 13 Π7 Τt7 41 Π9 Τt9
Out[9]= Π3 Τt3 Π5 Τt5
8 64 15 360 1 548 288
Out[11]= 0
3 Τt3 Τt5
Out[12]= Π3
8 3 5
3 Π3 5 Π5 91 Π7 Τt7 Τt9
8 64 15 360 7 9
;
Π 2
Π
eΞ : Sign Cos Ξ
;
2
Appendix 355
x
4
Out[17]=
t
2 4 6 8
2
4
1.5
1.0
0.5
Out[18]=
t
2 4 6 8
0.5
1.0
1.5
Appendix 357
1 1 1
eqX
Τ,Τ XΤ
pXΤ
YΤ,
Τ XΤ
Τ YΤ, a Τ
a2 2 a
1
pXΤ YΤ,
Τ XΤ
Τ YΤ, 2 a a Τ
0
Simplify;
a
1 1 1
eqY
Τ,Τ YΤ
pXΤ
YΤ,
Τ XΤ
Τ YΤ, a Τ
a2 2 a
1
pXΤ YΤ,
Τ XΤ
Τ YΤ, 2 a a Τ
B
Simplify;
a
Parameters:
Ω 1.0; Ζ 0.05; Ε 1.0; B 7.4;
15
10
0
h
-5
-10
-15
-20
-3 -2 -1 0 1 2 3
g
-4
-4.2
-4.4
-4.6
-4.8
0.113351, 4.51957
Check precision:
;
Π 2
Π
et : Sign Cos t
;
2
Cleary;
Tmax 4a; dirsol NDSolve
t,t yt
Ζ
t yt
Ε yt3 B et
a,
y0 x0, y '0 v0, y, t, 0, Tmax,
MaxSteps Infinity; y y
. dirsol1;
yyvv ParametricPlotEvaluateyt, y 't, t, 0, Tmax,
PlotRange All, AxesLabel "x", "v",
PlotStyle Thickness.001, TextStyle FontSize 14,
PlotPoints 500, Frame True, DisplayFunction Identity;
ShowGraphicsArrayxxtt, xxvv, yyvv;
360 Appendix
x
3
2
1
t
1 2 3 4 5 6
-1
-2
-3
4
2
0 x
-2
-4
-3 -2 -1 0 1 2 3
4
2
0 x
-2
-4
-3 -2 -1 0 1 2 3
Appendix 361
t,t x f x, t x, t 0.
EXAMPLES
X Τ Y Τ
Out[3]=
XΤ3 3 XΤ YΤ2
0, 3 XΤ2 YΤ YΤ3
0,
a2 a2
Y1
0, Y1
0, X 1
0, X 1
0
362 Appendix
X Τ X Τ
Out[4]=
X Τ3
0, X Τ3
0, X 1 X 1
0,
a2 a2
X 1 X 1
0, X 1 X 1
0, X 1 X 1
0
ΠΤ 4 X Τ ΠΤ 4 X Τ
Out[5]=
P Sin X Τ3
0,P Sin X Τ3
0,
2 Π2 2 Π2
X 1 X 1
0, X 1 X 1
0, X 1 X 1
0, X 1 X 1
0
ΠΤ 4 X Τ
Out[6]=
P Sin XΤ3 3 XΤ YΤ2
0, 3 XΤ2 YΤ
2 Π2
4 Y Τ
YΤ3
0, Y1
0, Y1
0, X 1
0, X 1
0
Π2
In[7]:= If present, the function ee4t
T must be shown with
no argument; derivatives de
dt are not allowed by this code,
but necessary generalizations are easy to implement
X Τ
Out[8]=
XΤ Β XΤ3 Α YΤ 3 Β XΤ YΤ2
0,
a2
Y Τ
Α XΤ YΤ 3 Β XΤ2 YΤ Β YΤ3
0, Y1
0,
a2
Y1
0, X 1
0, X 1
0
X Τ
Out[9]=
1 Α X Τ Β X Τ3
0, 1 Α X Τ Β X Τ3
a2
X Τ
0, X 1 X 1
0, X 1 X 1
0, X 1
a2
X 1
0, X 1 X 1
0
PROJECT: Using the idempotent basis, find T-periodic solution of the linear
oscillator under external and parametric rectangular wave periodic excitation
of the period T=4,
Appendix 363
1
t,t x 1 2 etΩ2 x pet.
T 4;
1
bvp TauTransformx ''t
Ω2 1
e xt p e 0, xt, t, T, 0
2
3 1
Out[11]=
p Ω2 X Τ X Τ
0, p Ω2 X Τ X Τ
0,
2 2
X 1 X 1
0, X 1 X 1
0, X 1 X 1
0,
X 1 X 1
0
sol DSolvebvp, X
Τ, X Τ, Τ
Simplify
Ω 3 3 3 Ω
2p 3 Cos Sin 2 Ω Cos 2 Ω 4 Cos 2 Τ Ω Sin
2 2
Out[12]=
X Τ ,
Ω 3 3 Ω
3 Ω2 3 Cos Sin 2 Ω Cos 2 Ω Sin
2 2
Ω 3 ΤΩ 3
X Τ 2 p 3 3 Cos Sin Ω 4 3 Cos Sin Ω
2 2 2 2
3 Ω Ω 3
3 Cos Ω Sin 3 Ω2 3 Cos Sin Ω
2 2 2 2
3 Ω
Cos Ω Sin
2 2
X
Τ X
Τ
. sol1;
X Τ X Τ
. sol1;
2 Π
Τt : ArcSin Sin t
;
Π 2
Π
et : Sign Cos t
;
2
In[19]:= Parameters and graphic output:
Ω 20.0;
p 1.0;
PlotEvaluatex, t, 0, 8, PlotRange All,
AxesLabel "t", "x ", PlotStyle Thickness.005
x
0.010
0.005
Out[21]=
2 4 6 8
t
0.005
0.010
0.005
Out[26]= x
2 4 6 8
0.005
0.010