On Nonlinear Vibrations of Systems with Many Degrees of Freedom

BY R. M . ROSENBERG

.
Department of Mechanical Engineering. Division of Applied Mechanics. University of
California. Berkeley Califwnia

Page
I. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
I1 . Admissible Systems and Motions . . . . . . . . . . . . . . . . . . . . 168
1. The General System . . . . . . . . . . . . . . . . . . . . . . . . 158
2. Admissible Systems . . . . . . . . . . . . . . . . . . . . . . . . 159
3. Admissible Motions . . . . . . . . . . . . . . . . . . . . . . . . . 162
I11 .
The Trajectories in Configuration Space . . . . . . . . . . . . . . . . 163
1 . Transformation and Trajectories . . . . . . . . . . . . . . . . . . . 163
2. The Restricted Principle of Least Action . . . . . . . . . . . . . . . 166
IV . General Description of the Geometrical Method . . . . . . . . . . . . 167
V. Trajectories of Admissible Motion . . . . . . . . . . . . . . . . . . 169
1. General Properties . . . . . . . . . . . . . . . . . . . . . . . . . 169
2 . Admissible Trajectories of Autonomous Systems . . . . . . . . . . . . 170
VI . Normal-Mode Vibrations of Nonlinear Systems . . . . . . . . . . . . 173
1. A New Definition of Normal Modes . . . . . . . . . . . . . . . . . 173
2. Straight Modal Lines . . . . . . . . . . . . . . . . . . . . . . . . 173
3. Interpretation of Modal Lines . . . . . . . . . . . . . . . . . . . . 176
VII . Properties of Motions with a Rest Point . . . . . . . . . . . . . . . 176
1. The Transversals (or P-curves) . . . . . . . . . . . . . . . . . . . 176
.
2 The 2'-curves . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
VIII . Special Autonomous Systems . . . . . . . . . . . . . . . . . . . . 187
1. Smooth Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 187
2. Uniform Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 188
3. Sequential Anchored Systems . . . . . . . . . . . . . . . . . . . . 189
4. Homogeneous Systems . . . . . . . . . . . . . . . . . . . . . . . 190
6. Symmetric Systems . . . . . . . . . . . . . . . . . . . . . . . . 190
IX . Theateb-Functions . . . . . . . . . . . . . . . . . . . . . . . . . 191
1. Their Origin and Importance . . . . . . . . . . . . . . . . . . . . 191
2. The Inversions . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
3. The Amp-Functions . . . . . . . . . . . . . . . . . . . . . . . . 196
4. The Sam- and Cam-Functions . . . . . . . . . . . . . . . . . . . . 197
X. Nonsimilar Normal-Mode Vibrations . . . . . . . . . . . . . . . . . 199
1. The Perturbation Potential . . . . . . . . . . . . . . . . . . . . . 199
2 . The Curved Modal Line . . . . . . . . . . . . . . . . . . . . . . . 200
3. Integrable Cases . . . . . . . . . . . . . . . . . . . . . . . . . . 201
4 . GeneralProperties . . . . . . . . . . . . . . . . . . . . . . . . . 202
XI . Exact Solutions to Steady-State Forced Vibrations . . . . . . . . . . . 203
1. Introductory Remarks . . . . . . . . . . . . . . . . . . . . . . . 203
2. Homogeneous Systems . . . . . . . . . . . . . . . . . . . . . . . 204
3. The Response Curves . . . . . . . . . . . . . . . . . . . . . . . . 210

166
166 R. M. ROSENBERG

XII. Steady-State Forced Vibrations by Approximate Methods . ....... 214

1. Description of the Method. ..................... 214
XIII. Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
1. Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
2. Stability of Normal-Mode Vibrations . . . . . . . . . . . . . . . . .220
3. Homogeneous Systems . . . . . . . . . . . . . . . . . . . . . . . 221
4. Symmetric System with Two Degrees of Freedom . . . ........ 225
5. Nonsimilar Normal-Mode Vibrations . . . . . . . . . . . . . . . . . 230
8. Forced Vibrations . . . . . . . . . . . . . . .
. . . ....... 230
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240

I. INTRODUCTION

The study of the vibrations of nonlinear systems with many degrees of

freedom is concerned with the search for some or all periodic solutions of
systems of nonlinear differential equations, and to deduce as many properties
of these solutions as the state of the applicable mathematical knowledge
permits. Unfortunately, this body of knowledge is limited and not unified;
in consequence, many and vaned disciplines within mathematics are com-
monly used to deduce partial results. In a general way, these results fall
into two broad categories: those which apply to systems that are “weakly
nonlinear,”* and those which apply when the systems are “strongly non-
linear.”
The results in the first category contain a good deal of detailed informa-
tion, and they resemble in many ways those familiar from linear theory.
Those in the latter category usually contain fewer details, being more con-
cerned with general questions of existence, uniqueness, boundedness and
stability of solutions.
The task of presenting a meaningful survey of the methods and results
concerning the vibrations of nonlinear multi-degree-of-freedom systems is
too ambitious for this relatively brief contribution. Moreover, several survey
articles and books, foremost among them the recent, admirable work by
Minorski [l] have served this purpose. Here, it is intended to display cer-
tain geometrical methods, and to summarize the results stemming from their
application.
The only systems considered are those whose mechanical models can be
constructed of masses and “massless” springs. One, several, or all masses
may also be acted on by time-dependent, periodic “exciting forces.” The
nonlinearities of these systems are those arising from “nonlinear springs.”
Within these restrictions, the system may be linear, “weakly” or “strongly”
nonlinear, or “nonlinearizable.” In the geometrical methods used here,
“weak” nonlinearity is of no advantage, and results for strongly nonlinear

All terms in quotation marks are defined precisely in the text.

 NONLINEAR VIBRATIONS 167

or nonlinearizable systems are frequently as easy to deduce as those of

almost linear systems.
It is plausible that general solutions (i.e. the class of all possible motions)
cannot be deduced for such nonlinear systems. However, if the class of
desired solutions is suitably restricted, one can frequently gain a great deal
of detailed knowledge regarding them, no matter how strongly nonlinear
the system. However, to be meaningful for the physical scientist, this
restricted class of desired solutions must be such as to explain and/or predict
the incidence of physically important phenomena. In linear systems, these
physically interesting solutions are the so-called “natural free vibrations,”
and the steady-state forced vibrations. It is precisely these types of motion
that are treated here for nonlinear systems.
The methods used to deduce these motions are not conventional (in the
sense of conventional present-day research in nonlinear vibrations), but they
are classical in the sense of theoretical mechanics that originated with
Hamilton, Jacobi and their contemporaries and that was described and ex-
tended relatively recently in a beautiful paper by Synge [2]. These methods
are especially useful in the search for “free vibrations in normal modes”
when the systems are strongly nonlinear. However, the concept of “normal
modes” and of eigenvalues is well defined only in linear systems. In fact, the
demonstration of the existence of eigenvalues has its roots in the theory of
quadratic forms, and its application to vibration problems requires that the
potential energy be a quadratic form. But, it is the very essence of non-
linearity, as defined here, that the potential energy is not a quadratic form.
Hence, the question of existence of normal-mode vibrations cannot be decided
in the conventional way of finding eigenvalues, and of attaching to each an
eigenfunction. Therefore, new definitions and methods are required when the
problems are nonlinear.
The usefulness of the concept of normal-mode vibration is, of course,
greatly impaired in nonlinear problems because the principle of superposition
fails. Nevertheless, the normal-mode vibration retains, even in nonlinear
systems, a position of central importance because, as will be shown, resonance
occurs in the neighborhood of normal-mode vibrations, whether the system is
linear or nonlinear. Hence, when the amplitude of the forcing function is
small, the solution of the normal-mode vibration problem furnishes the
“zero’th order” solution of the problem of steady-state forced vibration, as
demonstrated by Rauscher [3], for instance. As in the Duffing problem [a],
it is also necessary in problems with many degrees of freedom to examine the
stability of the solutions; this is done in the last section.
Although examples are given, the question of application has not been
treated in a general way. Suffice it to say that in our model a displacement
of any one mass has “an influence in depth”; i.e., its effect is not only felt
by the neighboring masses, but by all of them. Hence, it is likely that the
158 R. M. ROSENBERG

systems discussed here are applicable to crystal and molecule vibrations.

It is also quite possible that one can discuss, by the methods described here,
the mechanism of pulsating stars [S]. At any rate, these methods have been
examined in detail [S] by an astronomer having this purpose in mind.
No attempt has been made to cite in the bibliography all recent litera-
ture dealing with vibrations of nonlinear systems having many degrees of
freedom; it is, in fact, restricted to the material which was used in this
contribution,

AND MOTIONS
11. ADMISSIBLESYSTEMS

1. The General System

The general system considered here may be modelled on a great variety of

mechanical, electrical, or other devices, and its configuration may be deter-
mined by many different types of physical measurements. One such model

FIG.1. Mechanical model of spring-mass system.

is Figure 1. It consists of a chain of n + ..

1 mass points P,,(i = 0,. ,lr)
< <
and of s springs, n s n(lr 1)/2.+
<
The mass of P, is m,.(O< m, a).Any m, = a is a fixed point in
inertial space. We require that at least one of the mass points have infinitely
large mass, e.g. m, = 00.
The term “spring” is used to describe a massless one-dimensional device
that changes its length under the action of a force. In physical terms, a
spring is assumed to shorten under a compressive force of given magnitude
 NONLINEAR VIBRATIONS 159

precisely by the same amount as it lengthens under a tensile force of the

same magnitude, and it is capable of storing, but not of creating or destroying
energy. These conditions are fulfilled when
(2.1) o i j( w ) = oij(lwl),
and when

where O,(w) is the negative of the potential energy V, stored in the spring
Sij terminating on the ith and jth mass points, w is the length change of
Si, and Fii is the force which has produced the length change. Evidently,
combining (2.1) and (2.2), one has
(2.3) F-(m)
$1 = - F;i(- 20).

Each mass point is connected to one, several, or all other mass points
+
by a spring; hence, the maximum number of springs is n(n 1)/2. If one
extremity of a spring terminates on an infinitely large mass point it is called
an anchor spring; all others are coq5ling springs. I t is not essential that all
+
n(n 1)/2 springs be present in the system. However, it is required that,
in the absence of additional constraints, no mass point can be displaced
without giving rise to an elastic force acting on every mass point. For in-
stance if
(2.4) u*=o, ( i = O ,... I - l ) ,
)

the rth mass would be decoupled from all its neighbors on the left, and the
+
system of n 1 mass points would be separated into two spring-mass
systems with no elastic coupling between them. Such a separation into
several mutually uncoupled systems is excluded. In consequence the mini-
mum number of springs is n.
Each mass point has a single degree of translational freedom ui in the
direction of the chain. In view of mo = 00, one has uo E 0; hence, the
system cannot accelerate in the rigid body mode.
Each mass point may be acted upon by a time-dependent force, called
an exciting force and denoted by fi(t). Exciting forces act in the direction
of the chain of mass points. It is assumed that the spring forces Fii and the
exciting forces f, are the only forces acting on the mass points.
2. Admissible Systems
In order to fit into the framework of vibrations, linear or nonlinear,
systems must have certain properties whose physical description is :
(i) in the absence of exciting forces, the system must possess a single
equilibrium configuration, and it can execute “free vibrations” about that
equilibrium configuration ;
160 R. M. ROSENBERG

(ii) in the presence of exciting forces, the system can execute so-called
“steadystate forced vibrations.”
Some of these conditions can be insured by endowing the potential
function of the entire system with certain properties. This potential
function is, evidently,
n-1 I(

where uj - ui is the length change in the spring Sij.

We shall assume that, on any domain where it exists, 0 is smooth and
its first partial derivatives with respect to the ui are, a t least, piecewise
smooth. The second partial derivatives of 0 are thus assumed to exist
everywhere except possibly at isolated points of discontinuity of the a0/aui.
These assumptions are consistent with the physical properties of springs
in that they provide for the possibility of spring forces “with corners”;
yet they assure the existence of unique solutions under specified initial
conditions.
In addition, we require that the potential fulzction 0 be negative definite, or

O(0,.. .,O) = 0,

O(u,,. . . ,un)< 0 when the ul,. . , ,u, do not vanish simultaneously,

and the first partial derivatives of 0 vanish at the origin only, or

a .
-O(0,. .,O) = . .. = -a O(0,. . *,O) = 0.
8% aun

Finally, as a consequence of (2.1) and (2.6), 0 is symmetric with respect to

the origin of the zc,, ..
.,un-s#ace, or

Equations (2.6) and (2.7) insure that there exists a unique equilibrium
configuration, and the origin of the ui has been so chosen that every ui
vanishes in the equilibrium configuration. Equation (2.8) is a direct con-
sequence of the symmetry properties of the individual springs, described
in (2.3).
A potential function 0 which satisfies all of the above properties will
be called admissible. Moreover, when 0 is admissible, the azctonomous system
i s admissible.
 NONLINEAR VIBRATIONS 161

Exciting forces yi(t) are required to be a t least piecewise continuous,

and bounded for all t . In addition, they must satisfy

(2.10)

where T is a constant. Equation (2.9) says that the exciting forces are
periodic and, if there is more than one such force, they all have the same
period T.
Exciting forces /<(t)which satisfy the above conditions are called ad-
missible. Moreover, when both, 0 and the f,(t) are admissible, the nonauto-
nomous system is said to be admissible.
Among admissible systems, autonomous or not, we distinguish between
those that are linearizable, nonlinearizable, and weakly, or strongly non-
linear. In our system, these properties are associated with the spring forces;
hence, it may be best to define them in terms of these forces; i.e., in terms
of the partial derivatives of 0.
Because of the admissibility conditions on 0, these derivatives may
be written in the form
,. n
(2.11)

where the functions Pi vanish at the origin and are of degree higher than 1
and, because of (2.8),
(2.12)
-
aij = aii.

Let the determinant of the cZii be defined by

Then, a system is said to be linearizable if d # 0, and nonlinearizable if

d = 0. A system is said to be weakly nonlinear if

(2.14) Pi(u1,.. .,u,,) = E&u~,. . .,u,,), (i = 1,. . .,n)

where every leil << Id1 where d is defined in (2.13) and every Qi is bounded.
If (2.13) is not satisfied, the system is said to be strongly nonlinear.
162 R. M. ROSENBERG

3. Admissible Motions
From the definition of admissible systems, it is evident that these may
be strongly nonlinear, or even nonlinearizable. In consequence, it is not
likely that one can find the general solutions of the equations of motion or,
the class of all motions of which the system is capable. However, it is
possible to deduce certain types of motion and, in particular, the motions
which are the most important from the point of view of technical application.
These are the motions which, in linear systems, are called the “natural free
vibratiqns in normal modes” if the system is autommo%s, and the “steady-
state forced vibrations” if the system is nmutonomous. The latter include
the phenomena of “resonance” and of the tuned dynamic vibration absorber.
Both, the free vibrations in normal modes, and the steady-state forced
vibrations are so-called “vibrations-in-unison”. An admissible system is
said to execute a vibration-in-unison if its motion satisfies all of the fol-
lowing properties [7] :
(i) There exists a T = corzstant such that
(2.15) ui(t) = z ~ i ( b + T), (i = 1,. . .,%).
In words, the motions of all masses are equiperiodic.
(ii) If t, is any instant of time, there exists a single to in

tr < t < tr + TI2

such that
(2.16) u&) = 0, (i = 1,. . .,rz).
In words, during any time interval of a half period, the system passes
precisely once through its equilibrium configuration.
(iii) There exists a single 4 # to in
t, < t < 4 + T/2
such that
(2.17) ti&) = 0.

In words, during any time interval of a half period, the velocities of all
masses vanish precisely once or, all masses attain their maximum displace-
ment (in absolute value) precisely once during any interval of a half
period.
(iv) Let r (fixed) be any one of the i = 1,. . . ,n. Then every %&) and
ur(t) may be written, for a.ll t, in the form
(2.18)
 NONLINEAR VIBRATIONS 163

in which t is a parameter. It is now required that every ~~(21,) be single-

valued. In other words, the displacement of any one mass at any instant
of time determines uniquely that of every other mass a t the same instant
of time.
These are well-known properties of normal-mode vibrations, and of
steady-state forced vibrations, of the linear system. We shall show that
these same types of motion also exist in admissible nonlinear systems. Any
motion of an admissible system which satisfies (2.15) to (2.18) is said to
be an admissible motion.
Among admissible motions, it is necessary to distinguish between those
which are “similar” and those which are not. An admissible motion is said
to be similar [8] if
(2.19) =
uj(t)/zli(t) cii = const.

This terminology corresponds to common usage because, when (2.19) is

satisfied by an admissible motion, the wave-shapes of the time-displacement
curves are all geometrically similar. When (2.19) is not satisfied by an
admissible motion, that motion is said to be nonsimilar.
The normal-mode motions of linear systems, and their steady-state
forced vibrations under simple harmonic excitation are of the form
(2.20) aci=Qicosol. (i= l,...,n)
where the Qi are constants. Hence, these motions satisfy (2.19); i.e., they
are similar. We shall show that the linear system is not the only one having
similar vibrations-in-unison. In fact, there exist strongly nonlinear, and
nonlinearizable systems whose vibrations-in-unison are also similar.

111. THE TRAJECTORIES

IN CONFIGURATIONSPACE

1. Transformation and Trajectories

The equations of motion of the physical system are

a
(34
.
mjii, = -0(z+, . .,us) f#),
a@, + (i = 1,. , . ,n)

where dots denote differentiation with respect to time. It is now convenient

to introduce the transformations
(3-2) xi = F ui. (i = I , . ...n).
Then the system (3.1) transforms into the system

(3-3) H: Ei = U,(x,,. ..,zs)+ fj(t), (i = 1,. ..

$8)
164 R. M. ROSENBERG

where

and the subscript xi denotes partial differentiation with respect to xi.

But (3.3) may be regarded as the equations of motion of a mass point
having unit mass that moves in the n-dimensional space of Cartesian co-
ordinates xi,(i = 1,. . .,n), called the configuration space. This unit mass
point is being accelerated by two forces: One is the conservative force whose
components in the xi-directions are the UZi,and the other is a time-dependent
force whose components in the xi-directions are the fi(t). The unit mass
point and the forces acting on it constitute what may be called the pseudo-
system [9].
When moving in the configuration space, the unit mass point traces out
a trajectory. Here, we derive the equations of this trajectory. The solutions
of (3.3) are of the form

(3.4) xi = xi@), (i = 1,. . ,,n).

If one eliminates the parameter t between them, one obtains the equation
of the trajectory in the form

(3.6) xi = x i ( x l ) , (i= 2,. . .,n)

where x, = x1 has been arbitrarily chosen as the independent variable in
accordance with (2.18). One finds from (3.6) by direct differentiation
(3.6) i j= X i t i . , , Xj = XilZl + Xj)’il8
where primes denote differentiation with respect to xl. Moreover, if one
regards the velocities as functions of the displacement, the equations of
motion H become

Now, in view of (2.7),every xi takes on its extreme value Xi at t = t,. More-

over, in view of the continuity and differentiability of strictly* Newtonian

* A system is said to be “strictly Newtonian” if the forces, and hence the

accelerations, are bounded for every t. Thus, infinite forces of bounded impulse
and discontinuities in velocities do not arise in them.
 NONLINEAR VIBRATIONS 166

systems, there aIways exists some domain [ x j , X j ] ,(i = 1,. . .,n),of the
configuration space in which every xi@) has an inverse

(3.8) t = ti( q) I

Hence, summing the equations (3.7) over N, and integrating them once over
the domain [xi,Xi] one finds
"
(3.9)
1 "
-2
2 i-1
=
i<* . .JJ + 12 + i=
u(x~, 21Fi(xjJX<)
where h is a constant of integration. and

(3.10)
i
F ~ ( x < , X=~ )fi(tj(xi))dxi.
xi
It is evident that (3.9) is an equation of the energy balance of the system,
and the quantities Fi(xi,XI),(i = 1,. . .,rc) represent the work done on the
unit mass p i n t by the force components fi(t) in a displacement [xi,Xi].
When these forces vanish identically, (3.9) reduces to the energy integral
of the autonomous system. The idea of utilizing the inversions (3.8) is also
employed in Rauscher's method [3].
The substitution of the first of (3.6) in (3.9) results in an equation in
xla which may be explicitly solved for that quantity. That equation together
with the system H and the second of (3.6) is sufficient to eliminate the
parameter t. Hence, the resulting equation is that of the trajectory of the
unit mass point. It turns out to be
r 1

(3.11)
where we have omitted the self-evident subscripts on the 6. The corresponding
equation for the autonomous system is

(3.12)
The equation (3.12), but restricted to two degrees of freedom, was
probably first given by Kauderer [lo]. An equation, equivalent to (3.12)
has been given earlier [ll];in it, the arc length s was used as a parameter.
166 R. M. ROSENBERG

The elimination of s which has resulted in the compact forms (3.11) and
(3.12) was first given by Kinney [12].
The systems of equations (3.11) and (3.12) are denoted as the systems M
because they are derivable from a principle akin to the principle of Mau-
pertuis, given below. Similarly, the notation H in (3.3) may recall Hamilton's
principle.

2. The Restricted Principle of Least Action

It can be readily shown that the system M of the autonomous system;

i.e., (3.12) constitutes the Euler-Lagrange equations of the variational
principle
(3.13) 6A=O
where the action

(3.14)
'I

and

(3.16)

is the square of the line element in the configuration space. Hence, equa-
tions (3.12) follow directly from the principle of Maupertuis.
Now, it is well known that the principle of Maupertuis is only applicable
to scleronomous conservative systems. However, if one defines an action

$1

one can show that (3.11) is immediately obtained from the variational
principle
(3.17) SA=O
and from (3.10). Thus one has the [13].
Restricted Princi~lcof Least Action: The natural motion of any system
.
with equations of motion (3.3), 0% a%y interval [xi,Xi], (i = 1,. .#n)in which
the imersions (3.8) exist, i s such that the action A, defined in (3.16), i s
a mircimzlm.
 NONLINEAR VIBRATIONS 167

While it has only been shown here that the action is stationary, the
usual proof that the stationary value is minimum also succeeds here.
The above principle is restricted to the domain [xi,Xi] where the in-
versions (3.8) exist. In problems of dynamics in general, this domain is
not known u priori. However, because of (2.16), (2.17) and (2.18), that
domain is, for admissible motions, sufficiently large to contain at least the
closed intervals lying between the maximum amplitudes and the equilibrium
position, and the corresponding time interval is sufficiently large to cover
at least one quarter period. Since the systems M are used directly in the
methods to be presented, the restricted principle of least action is of con-
siderable usefulness in problems where admissible motions of admissible
systems are examined.

IV. GENERALDESCRIPTION METHOD

OF THE GEOMETRICAL

The problem in hand has now found a dynamical representation by means

of the system H of equations of motion (3.3), and a geometrical one by
means of the system M of equations of the trajectories (3.11) and (3.12).
It was shown by Darboux [la], for instance, that the correspondence between
these for corresponding initial conditions is one-to-one. The essentials of
the methods to be used here consist in utilizing the system M to deduce the
trajectories and, subsequently, to utilize these trajectories in order to deduce
the behavior of the system in the the-displacement domain. This is called
“the geometrical method”, a term borrowed from Synge’s paper [2].
It is interesting to note that the equations M of the trajectory are
intrinsicalIy nonlinear, even when the springs are linear. In fact, there
exist no admissible systems, autonomous or not, for which the equations of
the trajectory become linear. The assumption of dimear springs implies
merely that all Uxi are of the form
U

(44 u, = 2
i- 1
uijxj, (a = 1,. . ..n)
where the aii are either constants or functions oft. Under it, equations (3.3)be-
come linear while (3.11) and (3.12) remain nonlinear. The intrinsic nonlinearity
of the system M will perhaps be better appreciated if on recalls that its
integrals are Lissajous curves. These may have many points of self-crossing
and may even be compact on some domain of the configuration space. One
may conclude that admissible systems with linear springs are more easily
examined by means of the system H of equations of motion than by M.
The fortunate circumstance that the superposition theorem holds in this
case can, then, always be utilized to find the general motions and, if one
wants to work very hard, their trajectories in the configuration space [XI.
168 R. M. ROSENBERG

While linearity of the physical system has no essential simplifying ef-

fects on the system M , nonlinearity has no fearful, complicating effects
on MI either. Hence, if these equations can be used to deduce solutions
of the physicaZZy linear problem, they can frequently be used with equal
ease to deduce those of physically nonlinear problems [MI.
Now, it is quite possible that a system of very difficult, nonlinear dif-
ferential equations have certain simple integrals. In fact, we shall demon-
strate that the trajectories of admissible motions are such simple integrals.
They are of the form
(44 xi = x&J, (i = 2,. . .,%), (9 = 1,2,. . .)
where the index p denotes the mode of vibration. Say, the trajectory
corresponding to the mth mode of admissible motion is given by the functions
(4.3) xi = xim(X1), (i = 2,. . .$2)
that satisfy the system M, and which are supposedly known. Then, their
substitution in the first equation of the system H,for instance, results in

(4.4)
.. = UX,(Xl*%n(&
x1 * * * %&l)) + fl(&

an equation in x1 only.
If the system is autonomous, fl(t) 3 0, and (4.4) is of the form
(4.6) 21 = g(x1).
Then, for initial conditions (for instance)
(44 x1(0) = x,, ZJO) = 0 ,
and with the definition

(4.7)

the integral of (4.4) becomes

and the period of the motion is

0

(4.9) T m = 4 112 [G(v,XJ - G ( X ~ , X ~ ) ] } - W V .

XI
 NONLINEAR VIBRATIONS 169

If the system is nonautonomous, and the amplitude of the forcing function

is small; i.e.,
(4.10) fl(4 = &!At), (El small,
the steady-state forced vibration problem can be shown to become a pertur-
bation problem on that of free vibrations in admissible motion. It is evident,
then, that the central task in either problem (autonomous, or forced)
consists in determining the trajectories of admissible motions of autonomous,
admissible systems.

V. TRAJECTORIES MOTION
OF ADMISSIBLE

1. General Properties

We shall assume that, after formal reduction to a first-order system,

the right-hand side of (3.3) satisfy a Lipschitz condition. Trajectories then
exist and all are continuous. Because of (2.16), every admissible trajectory
passes through the origin of the configuration space. In view of (2.17), the
equation (3.9) reduces at t = t, to

This equation defines a surface in configuration space called the L-surface,

or the bounding surface. Then, it follows from (2.16) that every admissible
trajectory intercepts the L-surface. The equation (2.18) states that the xj(x,)
are single-valued for every i and any r. Hence, for instance, x k ( x I )is single-
valued, and the same is true for its inverse x I ( x k ) . But this requires that
each be monotonic. We denote this property by saying that every admissible
trajectory is strictly monotonic. Finally, since trajectories satisfy the Euler-
Lagrange equations of (3.17), existence of real trajectories requires that
the quantity under the square root of (3.16) be non-negative. But, i t is
seen from (5.1), that this quantity vanishes on the L-surface. Then, it
follows from the properties of U and the F , that no admissible trajectory
can pierce the L-surface. These properties are summarized in
Theorem V-I Every continuous, strictly monotonic curve, defined by
solzltions of (3.11) or (3.12), which passes through the origin of the configuration
space and which terminates on the L-surface (5.1) is that of a vibration-in-
unison as defined in (2.15) to (2.18).
From the preceding remarks it follows that the unit mass point of the
pseudo-system oscillates back and forth along this trajectory, and the
period of this oscillation is the same as that of the vibration-in-unison of
the physical system.
170 R. M. ROSENBERG

If the vibrations-in-unison are similar as defined in (2.19), the correspon-

ding trajectories are defined by
(6.2) xi(xl) = c i l x l r (i = 2,. . .,n)
where the cil are constants. In that case, (3.11) reduces to

(5.3) xj' +
[ux, f i ( t ( 4 ) I = uxj + f j ( t ( x j ) ) s (I' = 2,. - 1%)

because

But, (6.3) may be recast in the form

It follows that the differential equations of similar vibrations-in-unison

satisfy the system (6.3) or (6.4).
All properties of admissible trajectories, given above, hold whether the
system .is autonomous or not. Additional properties of admissible trajec-
tones can be deduced when the system is autonomous.

2. Admissible Trajectories of Azltonomozcs Systems

The equations of autonomous systems are obtained from those above

by putting
(6.6) f i ( t ) = Fi(Zi,Xi) 0, (i = 1,. . .,?z),
In that case, the energy integral

(6.6) T(R1,.. .'Xu) - U(X1,...,xu) = h

is seen from (3.9) to exist where

T(21,.. .,A?,,)= - 2
1
Xi'
2 i=l

is the kinetic energy of the pseudo-system, and h 0 is the energy level

of a given motion. The system M of the autonomous case is given in (3.12),
and the corresponding action integral is (3.14).
For this case, the bounding surface is defined by

(6.8) U(Xl,. . .,X") + h = 0,

 NONLINEAR VIBRATIONS 171

and this is also the maximum equipotential surface because (6.8) coincides
with (6.6) when the kinetic energy vanishes. In order to call attention to
the fact that (5.8) defines the bounding surface when the system is auto-
nomous, it will be denoted as the La-surface.
From (2.5), (2.6), (2.7), and (5.8) it follows that:
(i) the La-surface is a closed, smooth surface surrounding the origin
of the configuration space, and star-shaped with respect to it [9].
From (2.8) one sees that
(ii) the La-surface is symmetric with respect to the origin of the configura-
tion space.
From the admissibility condition (2.6) and (6.8) it is evident that
+
(iii) the quantity U h > 0 a t the origin of the configuration space.
This quantity does not change sign in the open, finite domain D, bounded
by the La-surface, it vanishes on L,, and it is negative (at least near the
La-surface) in the infinite open domain bounded by the La-surface.
These properties lead immediately to
Theorem V-I1 : Every trajectory, admissible or not, defined by solutions
x I . ( x I ) ,(i = 2 , . . .,n) of (3.12), lies i n the closed, finite domain D, of the con-
figuration space, bounded by the La-surface defined i n (5.8).
It is not necessary that all trajectories actually attain the La-surJace.
However, if they do, this occurs necessarily at an instant t, when all a,(t,)
vanish. Trajectories which attain the L,-surface will be called trajectories
of motion with a rest point [19]. Not all motions with a rest point need be
admissible because the corresponding trajectories need not pass through
the origin. However, all admissible motions are motions with a rest point.
Next we state a property, first observed by Mawhin [6]. I t is
Theorem V-I11 : If 8 i s any trajectory, admissible or not, of an admissible,
autonomous system, the trajectory 8 ,symmetric to 8 with respect to the origiB,
is also a trajectory.
The proof follows immediately from (2.8) and its consequence

Of importance to admissible trajectories is [S] the

Corollary V-IV: Every trajectory, admissible or not, (of an admissible,
autonomous system) which passes through the origin is symmetric with respect
to it.
172 R. M. ROSENBERG

Finally, we have

Theorem V-V: If a trajectory, admissible or not, of a n admissible, auto-

it intercepts thut surface orthogonally.
nomous system attains the L,,-surface.
To prove it we observe that the system M of autonomous systems, given in
(3.12), reduces on the La-surface to

(6.10) xjrUx,= U x j , ( j = 2 , . . .,n),

and this may be cast in the form

(6.11)

But equations (6.11) are precisely the conditions for orthogonal intersec-
tion.
Special additional properties hold for similar vibrations-in-unison of
autonomous systems. They are given in [9].

Theorem V-VI : Every trajectory of a similar uibration-in-unison of a n

admissible autonomous system intersects every equipotential surface ortho-
gonally, and [9]

Theorem V-VII: If there exists a straight line which intersects every equi-
potential szcrface of an admissible, autonomous system orthogonally, it i s a
trajectory.

To prove these theorems, we observe that equipotential surfaces are

defined by

(6.12) . +
U(X1,. .,xn) h* = 0. (O< h* < h)
and, in view of the properties of admissible potential functions, the equi-
potential surfaces are simple, closed, non-intersecting surfaces surrounding
the origin, and compact on D,.
It follows from (6.2) that, in the case of similar motion,

(6.13) xl’(xl) E 0, (i= 2,. , . ,n).

Hence, the system (3.12) reduces throughout D, to (6.10) or (6.11). Now,
+
the total differential of every equipotential surface d(U h*) is independent
of h* which proves the first of the theorems. It is clear that the second theo-
rem also holds because (6.11) is merely one way of writing the system M
when equations (5.13) are satisfied.
 NONLINEAR VIBRATIONS 173

VI. NORMAL-MODE
VIBRATIONS SYSTEMS
OF NONLINEAR

1 . A Neu Definition of Normal Modes

It was observed in connection with the definition of admissible motions
that the normal mode vibrations of linear, autonomous systems are in fact
vibrations-in-unison. They are usually deduced from the dynamical system
by means of eigenvalue theory, and in the proof of their existence the fact
is utilized that the potential function U is a quadratic form [17]. Now,
it is the very essence of nonlinearity (as defined here) that the potential
function is not a quadratic form : hence, the application of conventional
methods for deducing normal mode vibrations is difficult to generalize to
nonlinear systems.
We shall show that the normal mode vibrations of linear systems may
be defined in terms of admissible trajectories, and that these are deducible
from the system M ,rather than H. In fact, the normal mode vibrations of
autonomous systems are its only vibrations-in-unison.
Furthermore, it will be demonstrated that nonlinear admissible systems
also possess such vibrations-in-unison. The existence proof is given in the
appendix. These observations permit a new definition of normal mode
vibrations which is wide enough to include the well-known normal-mode
vibrations of linear systems as a special case.

Defhition I ; Vibrations-in-unison of an admissible, aertolsomous system

are the normal-mode vibrations of that system.

We shall also use

Defirtition I1 : Trajectories in the configuration space corresponding to

normal-mode vibrations of admissible, autonomous systems are called modal
lines.
Then, one sees from Theorem V-I that modal lines are strictly monotonic
curves passing through the origin, and terminating on the La-surface.

2. Straight Modal Lines

In (2.20), it was noted that the vibrations-in-unison of the linear. auto-

nomous system are similar or, that the modal lines of that system are
straight. Since we wish to deduce these vibrations by the geometrical
method, we derive first the conditions, necessary and sufficient for the
existence of straight modal lines, and we demonstrate subsequently that
the linear, autonomous, admissible system satisfies these conditions. The
derivation of the n.a.s. conditions is conveniently done by transforming
174 R. M. ROSENBERG

from Cartesian to spherical coordinates in n-space. These transformations

are of the form [ll]

(6.1) xi = rf& .. ..
(i = 1,. ,n)
and, specifically, they are [9]

(6.2) xi = Y sin en+l--i n

n--i

j- 1
cos ej, e,, = n/2, (i= 1,. ..,n).

When the equation of the L,surface is transformed into spherical coordinates,

it is of the form
(6.3) O(r,el, . . + h =0
Then, one has [9] the
Theorem VI-I : The conditions, necessary and szcfficient for the existence
of at least one mode of similar normal-mode vibration of an admissible, azcto-
nomow system are that every partial derivative of 0 with respect to the Oi be
ofthe fovm

(6.4) 0, = si,(r,el,.. .,en-,) - @,,(el,.. . (i = 1,. . .,n - 1),

and that the eqwations
(6.6) 8ie(81,. .. = 0, (i = 1,. ..,n - 1)
have at least one system of real roots &*.
To sketch the proof, we observe that, in spherical coordinates, the equi-
potential surfaces are given by

(6.6) O(r,e,,. ..,e,,-l) + h* = 0, o< h* < h.

Now,radius vectors r to points on equipotential surfaces having stationary
distances from the origin with respect to neighboring points on these same
surfaces satisfy
(6-7) dr = 0,
and these radius vectors intersect the equipotential surfaces orthogonally.
From (6.3)

or in view of (6.7)
(6.9) 0, = 0, (i= 1,. . .,n - 1).
 NONLINEAR VIBRATIONS 176

The locus of these points of stationary distance on the equipotential surfaces

is a straight line if, and only if, the system (6.9) has roots independent of
Y which requires that the Oei be of the form (6.4).
Theorem VI-I is useful not only because it may be used readily to test
whether straight modallinescanexist. but the direction-cosines of the straight
modal lines, if they exist, are furnished by the roots of (6.6). Hence, the
theorem provides also a method for computing the straight modal lines.
It can be shown that the new Definition I of normal mode vibrations
yields only the normal modes, and all of the normal modes of admissible,
autonomous, linear systems [S].

3. Interpretation of Modal Lines [ll]

The modal lines have a very simple, geometrical significance which will
be illustrated here by means of an admissible, autonomous system having
the three degrees of freedom
(6.10) XI = x, xg = y , xg=z.

i
Y

X
- y=ct ; x

FIG. 2. Modal line of linear system with three degrees of freedom.

Suppose the system is linear. Then, the equipotential surfaces are given
by the concentric three-dimensional ellipsoids (with center at the origin)
(6.11) t(x2 + y2 + %a! = h*, o< h* < h,
and the vibration in the pth normal mode is given by
(6.12) x = Xp cos apt, JJ= Yp cos opt, z = 2, cos apt
where X,, Y,,Z p are constants.
It follows from these equations that the pth normal mode satisfies

(6.13) ylx E YpIXp = c$, alx =Z,lX, = c?#.

176 R. M. ROSENBERG

Then, the modal line in the pth mode is the intersection of the planes

(6.14) P
y = cy+x, z = l$X.

Evidently, this intersection is a straight line passing through the origin.

From the known geometric properties of ellipsoids and from Theorem V-VII
it follows that there exist three such modal lines, and they are the principal

(a1 (b)
FIG. 3. Modal line of nonlinear system with three degrees of freedom.

axes of the ellipsoid. The construction of a modal line, and its location
inside the bounding ellipsoid is shown in Figures 2(a) and (b).
If the problem is nonlinear, but the normal mode vibrations are similar,
equations (6.14) still apply, but, in that case, the bounding ovaloid is no
longer an ellipsoid.
If the problem is nonlinear] and the normal-mode vibrations are non-
similar, equations (6.14) are replaced by equations of the form

(6.16) y = qJ(X), = $(4

where y and $ are nonlinear, monotonic functions of x . Each is a cylinder

in xyz-space containing the origin, as shown in Figure 3(a). Their intersection
is a modal line, and this modal line inside the bounding ovaloid is shown
in Figure 3(b).

VII. PROPERTIES
OF MOTIONS WITH A REST POINT[19, 241

1. The Transversals (or P-cwves)

Motions with a rest point have been defined as those in which there
exists a time t, at which the velocity of all mass points of the physical system
vanishes. If the system is autonomous, the trajectories corresponding to
motions with a rest point are those which attain the Lasurface, and ad-
 NONLINEAR VIBRATIONS 177

missible motions belong to this class. The equipotential surfaces of these

systems are given by (5.12), and the transversals are defined in terms of
these equipotential surfaces. Transversals, or P-curves, are the lines ortho-
gonal to the equipotential surfaces. From their definition, one has imme-
diately
..
Theorem VII-I : At a point (xl,. ,xn)of the configuration space where
the slope of a transversal is well-defined, it coincides with the direction of the
force which acts, at that point, on the w i t mass point of any admissible, auto-
mmous pseudo-system.
Hence, the P-curves have an important dynamical significance, and i t
is of interest to determine their properties. One observes that, when n > 1.
the transversals constitute a family of curves which pass through the origin
of the n-space, which are compact on D,, and which are nonintersecting,
except a t the origin. It would be interesting to determine their geometrical
properties in the entire domain 0,.However, this is a difficult task when
the number of degrees of freedom exceeds two; therefore, our attention will
be restricted to the case n = 2.
Consider an admissible, autonomous system having the degrees of freedom
(7.1) u,= u, u2= v .

We admit linearizable and certain nonlinearizable systems by writing

a
-O(U,V) = - auk - 6 V k + P(z4.V).
au
(7.2)
-o
a ( ~=, -~ ) tUk - avk + Q(.,.)
av
where k is a positive, odd integer, and P and Q are of degree higher than k.
The determinant

(7.3)

With these definitions, the system is linearizable if k = 1, and nonlinear-

izable if k > 1.
The equations of motion of the physical system are

.. -
m2v =
a O(u,v),
av
178 R. M. ROSENBERG

those of the pseudo-system are

(74

and the equation of the trajectories is

(7.6) 2 [u(X,y)+ h]Y" + (1 f y'*) Ly'UJW') - Uy(%y)I= 0

where priqes denote differentiation with respect to x.
The system possesses the energy integral

(7.7) 7-(2,$)- V ( x , y )= h.

The bounding surface (here a curve) is the locus of the points P(X,Y) for
which the kinetic energy vanishes when the energy level of the motion is k,
and the equation of the La-curve is

(7.8) + h = 0.
U(X,Y)
The equipotential curves are the functions

(7.9) Y = &)
which satisfy the equation
(7.10) +
U(x,qJ) k* = 0, OC h' <h
and they will be called E-crcrves. By differentiating (7.10), one finds that
E-curves satisfy the differential equation

(7.11)

It follows that the transversals, or P-curves, are the functions

(7.12) Y =e(4
which satisfy the differential equation

(7.13)

We observe that (7.13) is singular only at the origin, and the equation
of the trajectories (7.6) is singular only on the La-curve. Consequently, we
define as regular p o d s of the xy-plane all points, except the origin, which
lie in the open, finite domain D, bounded by the La-curve.
 NONLINEAR VIBRATIONS 179

To begin with, we prove

Theorem VII-I1 : The origin i s always a node for the P-curves of admissible,
autonomous systems having two degrees of freedom.

Suppose the system is linearizable [19]. or k = 1 in (7.2). Then, in the

neighborhood of the ongin,

(7.14)
u, = XU,(O,O) + y ~ x y ( 0 , O+) . -
* I

u, = X V , y ( O , O ) + yU,,(O,O) + . . .*

The substitution of these in the equation (7.11) of the E-curves results in

(7.16) *=
dx
- xu, - 'pu,
XU, + pU,, i-
+ . ..
. ..
where the partial derivatives are evaluated at the origin. But, the integrals
of (7,11), and hence (7.16), are known; they are E-curves and they form a
continuum of simple, closed, nonintersecting curves surrounding the origin.
Then, it follows that the origin is a center for the E-curves and this requires,
by the Poincart! theory of singular points [l], that

(7.16) u,,u, - uz,> 0.

The substitution of (7.14) in the equation (7.13) of the P-curves results
in

(7.17)

Application of the PoincarC theory to this equation has as immediate con-

sequences

(7.18)

These are two of the three conditions which are necessary for the origin to
be a node. But, in view of (7.16).

(7.19) U,Uee - Ute > 0.

This is the third necessary condition, and (7.18) and (7.19) are also sufficient.
Hence, Theorem VII-I1 is satisfied in the lilrearizable case.
180 R. M. ROSENBERG

To prove the theorem for nonlinearizable systems [21], we note that,

in the neighborhood of the origin, the potential function of the pseudo-
system is

(7.20)

where a,b,c > 0 are constants.

The derivatives of this function (with y replaced by 0) are now sub-
stituted in the equation of the P-curves (7.13), and the transformations

(7.21)

are introduced in the resulting equation. Then, one finds, instead of (7.17)
in the neighborhood of the origin

(7.22) _
dr] - - cvk + b(E - q ) k + . . .
at - - atk - b ( t - q)b + .. . *

The singularity at the origin of this equation may be discussed by the method
of Argemi and Sideriadks [22, 231. These authors consider the equation

(7.23)

where X and Y are homogeneous functions in x and y of the same degree

k. They introduce a transformation
(7.24) y =d ( x )
under which the functions go over into
(7.26) x = a+/(& Y = x*g(A)
and the differential equation (7.23) becomes

(7.26)

where k(A) = g(1) - it/(& Now, this last equation has simple singular
points on the it-axis at the zeros Ri of k(it), and their discussion is accessible
to the Poincarh theory. Expansion of the numerator and denominator
of the right-hand side of (7.26) near the singular points gives
 NONLINEAR VIBRATIONS 181

_ - h’(li)l + . . .
dl -
(7.27)
dx f(&)x + ,. ’
,

The authors show that only saddles and nodes can occur, and one has
a node if f(li)h’(&)> 0,
(7.28)
a saddle if f(A,)h’(l$)< 0.
The quantity t(&)h’(&)does not vanish, in general, because the zeros of
/ ( A ) do not coincide with those of h(1) when X # Y,and h’(1) does not
vanish at the zeros of h(1). Moreover, when the 1, are ordered, according

(a 1 (b)
FIG.4. Saddles and nodes in xl-plane and mapping on configuration space.

to their magnitudes (in ascending or descending order), nodes and saddles

occur alternately. Clearly, the inverse of the transformation (7.24) maps
all singular points of (7.27) into the origin of the xy-plane, and the zeros
li of h(1) are the slopes of the trajectories at the origin. This proves The-
orem VII-I1 for the case of nonlinearizable systems.
It may be interesting to follow the construction in greater detail, as
.
done in Figure 4. Let A,,&,, . . . give the nodes in the xl-plane, and &,,A,, . .
the saddles. Next, construct the straight lines that pass through the origin
.
with slopes &,(i = 1,2,. .) and assume that 4 > 0; i.e., the line having
this slope lies in the first and third quadrants. Next consider the trajectories
lying in the sector enclosed by the lines of slopes 1, and 3.8 (starting in the
182 R. M. ROSENBERG

first quadrant and proceeding in a counterclockwise direction). This sector

contains a continuum of trajectories that pass with slope 4 through the
origin, and they tend to lines parallel to the line of slope 4. The latter is
a degenerate trajectory of this class. The trajectories in the next sector
between the lines of slopes & a n d & pass through the origin with slope &,
and tend to lines parallel to the line of slope 4, that line being also a degener-
ate trajectory of this class. This process is continued until the xy-plane
is filled in the neighborhood of the origin with trajectories. IndFigure 4(a),
the trajectories and singular points of the xl-plane are shown, and in Figure
4(b), the corresponding picture in the xy-plane is shown.
Suppose, as a special example, that k = 3 in (7.20) and (7.22). Then,
[24] a simple computation shows that

(7.29)
/(A) = - b[(2 - 8) + (1 - 4 s 1 v
h(A) = - b[A4 - aAs + P A - 13
where
C
a=2--
b'
(7.30)

It is now necessary to determine first the zeros & of h(l) and, next, ,the
sign of the product fh' at these zeros. Writing h(A) = 0 in the form
(7.31) A4 - 1 = (als - 8)l
one sees that the left-hand side is even in A and has two real zeros at 1 = f 1,
and the right-hand side is odd and has one or three real zeros. Hence, these
two functions have always at least two real intersections, or h(A) has always
at least two real zeros. One may assume without loss of generality*
that c >, a. Now, it follows from the second of (7.29) and from the assump-
tions on a, b, and c, that
h ( w ) = h(- 00) = - 00,
h(0) = b,
(7.32)
h(1) = - b(- a + 8) = - (C - a ) < 0.
h(- 1) = - h(1) = c - a 2 0.

* If, on the contrary G < a, we simply denote the displacement of mII by u and
that of m1 by u which restores the assumed relation.
 NONLINEAR VIBRATIONS 183

Hence, h(A) looks as shown in Figure 6 (where no attention is to be paid to

the location of the maximum). It follows from (7.32)and Figure 6 that

hc. 6. The curve h(A).

one root, say 4, is positive and lies in the interval O < A,< 1. and & is
negative. Moreover, the slopes are

(7.33)

Next, we examine f(A). It follows from the first of (7.29) that

(7.34)

and that /(A) has only a single real zero a t A = 1 - v2 - B. To evaluate

the /(Ai), it is convenient to return to the physical constants, so that

Since b > 0, a/b > 0, and 0 < 4 < 1, and since 4 < 0, one has from (7.36)
f(4)< 01
(7.36)
f(U< 0.
Combining (7.33) and (7.36),

(7.37)
184 R. M. ROSENBERG

Hence, the xl-plane has, at least, two singular points on the A-axis. One
lies on the positive branch between 0 and 1, and it is a node. The other
lies on the negative branch, and it is a saddle.
We shall assume that Al,z are the only zergs* of h ( l ) . Under this assump-
tion we have now demonstrated
Theorem VII-11: When k = 1.3 and h ( l ) has only two zeros, all trans-
versals of admissible, autonomous systems pass through the origin of the xy-
plane with common slope OOp.
To determine the properties of P-curves not near the origin, we determine
the locus of their inflection points, if any. By differentiating (7.13), one has

This locus will be called the F-curve:

It follows from (7.39) that F-curves pass through, and are symmetric with
respect to, the origin.
Next we determine the locus of points such that the tangents to P-
curves at these points pass through the origin. Evidently, these are the
points for which

(7.40)

Hence, the locus of these points, called the G-curve, is given by

(7.41) G(x,Y)= Y U , - X U , = 0.
G-curves also pass through the origin and are symmetric with respect to it.
Obviously, F and G-curves have the slope 0; at the origin.
Now, transversals are seen from (7.13) to be themselves symmetric with
respect to the origin. Hence, the field of transversalslooks as shown in Figure 6.
In that diagram, P,(Xo,Yo) is the point where the straight line, passing
with slope 0; through the origin, intersects the La-curve. P1(X,,Y,) is that
where the C-curve intersect the La-curve, and P,(X,,Yp) is that where the
F-curve intersects the La-curve.
We show now, that a t least one F-curve and G-curve always exist and
intersect the La-curve. Since every equipotential curve has, in the first
quadrant for instance, always at least one point of stationary distance

It can happen that k(L) has four real roots when R = 3. This has been shown
to be an exceptional case and is not treated here. [MI.
 NONLINEAR VIBRATIONS 186

from the origin, that quadrant contains at least one G-curve. Consider
now the transversal issuing from P I . I t points initially towards the origin.
If Po and P I do not coincide, this particular transversal must have an
inflection point prior to its amval at the origin with slope 6;. In that
case, the transversal issuing from Po points initially above the origin, and

F I G . 6 . The directed transversals in nonlinear system with two degree of freedom.

it must also have a point of inflection before arriving with slope 6; at the
origin. Hence, in general, there exist transversals with points of inflection,
and there exists at least one F-curve in the first quadrant. The transversals
will be endowed with a sense of direction by considering the points P ( X , Y )
which compose the Ldcurve as their points of issue. These directed trans-
versals in Figure 6 are supplied with arrow heads pointing toward the origin.
In this way, the transversals give not only the direction, but also the sense,
of the forces acting on the unit mass point of the pseudo-system.
2. The T-curves [19]
The trajectories corresponding to motions (at energy level h) of admissible,
autonomous systems fall into two general classes: those which attain the
La-surface, and those which do not. By definition, T-curves are those integral
curves of (3.12) which do intercept the La-surface. Hence, the T-curves
constitute the class of all trajectories which correspond to motion with a
186 R. M. ROSENBERG

rest-point of admissible, autonomous systems. Note that T-curves need

not be admissible in the sense of Section V.
We shall establish here certain general properties of T-curves of systems
having two degrees of freedom. All T-curves intercept the L,-curve, and
all P-curves intersect it. The P-curve which issues from the same point
P(X,Y) from which a T-curve issues will be called “the transversal associated
with that T-curve,” or simply the associated P-came.
Some of the properties of T-curves are easily established by means of
the equation of the trajectories (7.6), written in the form

(7.42)

it results immediately from a substitution of (7.13) in (7.6).

Because of the definition of P-curves and T-curves and Theorem V-V,
one has:
Properly 1 : Every T-curve is tangent to its associated P-curve on the
L@-curve.
Less obvious is:
Property 2: The curvature K ~ ( X , Yof) a T-curve at P(X,Y) has the
same sign, but is less in magnitude than the curvature K ~ ( X , Yof) its asso-
ciated P-curve.
To prove it, one observes that the second derivative y“ in (7.6) is in-
determinate on the La-curve. Evaluating it by 1’Hospital’s rule

(7.43)
1 Uxy
y”(X,Y) = -
3 1
But, by Property 1 and (7.13),
+ y‘(Uyy -
ux
UXZ

1-
- Y’UZY)
x.*

(74 y y x , ~=
) e y x , y ) = u,(x,Y)/u,(x,Y).
Consequently,

and, comparing (7.38) and (7.45)

(7.46) KT(X,Y)= iKp(X,Y),
which proves Property 2.
Next, we have:
Property 3: A T-curve which is Not a straight line cannot coincide
everywhere with its associated P-curve.
The proof is evident from (7.42). From that equation, one can also
deduce immediately:
 NONLINEAR VIBRATIONS 187

Property 4: If any trajectory (and, hence, any T-curve) is tangent to

a P-curve at any regular point, the trajectory has zero curvature at that
point.
This property of trajectories was earlier noted by Kauderer [lo].
Property 6 : At any regular point, every curved trajectory presents
its convex side to the incoming, directed P-curves.
This property is a consequence of Theorem VII-I. It merely states that
the trajectory of the unit mass point is being turned, by the forces acting
on it, in such a direction as to yield to these forces.
Prq5erty 6: Every T-curve that intersects its associated P-curve at a
regular point Q(x,y), must have passed through a point of zero curvature
on the arc between P(X,Y) and the point of intersection Q(x,y).This prop-
erty follows from the observation that the P-curves constitute a regular
field, and the T-curves are smooth. Hence, there exists a point on the arc
lying between P(X,Y) and Q(x,y) where the T-curve is tangent to some
P-curve. By Property 4, this is a point of zero curvature of the T-curve.
Finally, we have :
Property 7: Every trajectory (and, hence, every T-curve) either passes
through the origin of the xy-plane or, else it must have at least one point
of tangency with at least one E-curve (see the appendix).
To prove it, we observe that the E-curves constitute a regular field of
simple, closed curves surrounding the origin. Moreover, trajectories are
smooth, and by Theorem V-11, they are confined to the closed domain 0..
It follows from these observations that Property 7 is true.
It is clear that one may consider the origin as the (degenerate) equi-
potential line of zero energy. In that case, Property 7 might be interpreted
as stating that every T-curve is tangent at least once to an E-curve, and
T-curves which are tangent to the E-curve of zero energy are necessarily
simple trajectories and, perhaps, modal lines.
It seems probable, that the above properties of T-curves could also be
established for systems whose number of degrees of freedom exceeds two.

VIII. SPECIALAUTONOMOUS
SYSTEMS

1. Smooth Systems
It is evident from the .equations of motion (3.3) of the pseudo-system
that all differences between given, autonomous systems consist in differences
between their potential functions.
An autonomous system is said to be smooth if its potential function is
of the form
188 R. M. ROSENBERG

where the a$) are constants. Then, the spring force of a spring Sij between
the masses mi and mi is

where

(8.3) w=u.- "j

is the length-change. Hence, all derivatives of every spring force with

respect to the length-change exist, which is the reason for the terminol-
ogy "smooth". While all spring forces of smooth systems are assumed to be
finite polynomials, it is not required that these polynomials be all of the
same degree. Thus, while the restriction to smooth systems may appear
severe from the mathematical point.of view, it is mild from the standpoint
of physics.

2. Uniform Systems [9]

A system is said to be uniform if its potential function is of the form [9]

One sees from (8.4) that uniform systems are those in which all springs are
equal, and all masses are equal (in which case one may, without loss of
generality, put the masses equal to unity).
For uniform systems, one can readily demonstrate an interesting and
useful property. Let us denote as the potential function of the associated,
linear system that whose potential function is [9]

where the values of a(l) are the same in (8.4) and (8.6). Then, one has
Theorem VIII-I : The modal lines of an autonomous, uniform system are
straight. Their directions coincide with those of the associated linear system.
One proof of this theorem is due to Mawhin [6] who shows that the modal
lines of (8.4) are straight, and that their direction is independent of m.
The importance of this theorem is evident. The modal lines, when U = 0,
are the principal axes of the n-dimensional ellipsoids

(8.6) O + h* = 0, (0< A* < h)

 NONLINEAR VIBRATIONS 189

and the directions of these axes are readily found by means of linear eigen-
value theory. The time-historyof the motions can, then, be found subsequently
by simple quadratures.
The geometrical interpretation of this result is that the equations

(8.7) U + h+ = 0, (0< h* < h)

are ovaloids which are “distorted ellipsoids.” However, these distortions
are such that the ovaloids have the same symmetries with respect to the
modal lines as the ellipsoids of (8.6).

3. Sequential Anchored Systems

With Haughton [27]. we call a system seqwntial if it consists of a chain
of n finite masses in which each mass is coupled by a “smooth” spring (as
above) to only its neighbors. It is said to be anchored if the first and nth
masses are also coupled to a fixed point by “smooth” springs. This is in
fact, the model that is usually employed in the theory of linear vibrations
of multi-degree-of-freedom systems. The potential function of anchored,
sequential systems is of the form

+
where i = i 1.
Haughton [27], using methods similar to those of Mawhin [S]proves the
following remarkable
Theorem VIII-I1 : If an admissible. anchored, seqrcential system has straight
modal lines, their directims are the same as those of the associated, linear
system of potential function

(8.9)
If, in an anchored, sequential system, all masses are equal and all springs
are equal it is said to be anchored, sequential and d i m . In that case,
one can readily show that straight modal lines do, indeed, exist; the prob-
lem of normal mode vibrations can, then, be completely solved by con-
sidering a linear system and by performing one additional quadrature.
190 R. M. ROSENBERG

4. Homogeneous Systems [28]

An admissible, autonomous system is said to be himogmcoscs if its po-

tential function is of the form
n u
(8.10)

where k is a real number in O C k < do. I t derives its name from the fact
that the potential function is homogeneous in the xi of degree k 1.+
It is interesting to forni the expression for the spring forces. These
forces are given by

This equation shows that the spring forces are odd functions, and proportional
to the kth power, of the length-changes. Clearly, when one puts k = 1,
one recovers the linear system.
Physically, neither masses nor springs need be equal in homogeneous
systems. Instead, the springs are nonlinear “in the same way”; i.e., all
obey the same simple power law.
It is not difficult to construct a physical system for which k = 3 [21].
Consider an arrangement of masses and springs as in Figure 1. However, in
the present case, the translations ui are not in the direction of the chain,
but t t o d to it. Moreover, every spring is assumed l i w and its free
length is exactly equal to the distance between the masses interconnected
by that spring when the system is in the equilibrium configuration. Clearly,
the spring forces are always restoring, or 0 is negative, definite; moreover,
the spring forces reverse sign when the displacements do, or 0 is symmetric
with respect to the origin. However, infinitesimal displacements I(~ which
are small of the first order produce length-changes in the springs which
are small of the second order. Hence, for sufficiently small displacements,
k = 3.
By applying the criterion (6.3) one can readily prove [S]
Theorem VIII-111: The modal lines of admissible, autolumtous, homo-
g m u s systems are straight. Their direction i s defined by the sysiems of roots
of the transcedental eqwations (6.4).

6. Symmetric Systems

With Mawhin [6] we call a system symmetric if its potential function is

of the form
 NONLINEAR VIBRATIONS 191

The physical properties of symmetric systems are the following: All masses
are equal (hence, assumed equal to unity, without loss of generality), all
spring forces are smooth (in the above sense), all anchor springs are equal,
and the spring forces in all coupling springs are polynomials of thesame degree.
This is a generalization of the “symmetric two-degree-of-freedom system”
1291 which was called “symmetric” because the system had literally physical
symmetry about its centerline. (It should be noted that Mawhin denotes
the spring between m, and m, as So, that between m, and mi as S j , and
that between m, and m,, as S,.)
If a symmetric system has only two degrees of freedom, one can readily
show [29] that straight modal lines exist, and they have the same inclination
as those of the associated linear system.

IX. THE ATEB-FUNCTIONS

[20]

1. Their Origin and Importance

As the name indicates, the ateb-functions are inversions of certain beta
functions, much in the same way as elliptic functions are inversions of
elliptic integrals. Similar to elliptic functions which satisfy the autonomous
Duffing equation, the ateb-functions solve the problem of normal-mode
vibrations sf homogeneous systems. In common with elliptic functions,
the ateb-functions depend on a parameter n as well as an argument. I t is
interesting to note that Gauss [30] was the first to invert a beta-
function for the case n = 2. His problem arose in the rectification of the
lemniscate, and his lemniscate functions were, in fact, the first functions
to be defined by the inversion of an integral. Later, Legendre showed [30]
that the period of the lemniscate functions was expressible in terms of
gamma functions-an observation that has been repeated independently
by many authors in recent times [31, 32, 10, 331.
We have shown that the modal lines of homogeneous systems are straight.
Hence, they are expressible in the form
(9.1) xi = ciPxl, (i = 2,. . .,n; p = 1 , . . .,m)
where m is the number of modal Lines. If one substitutes (9.1) into the
first of the equations of motion
192 R. M. ROSENBERG

of a homogeneous system, one finds

(9.3) 2, + C9x11x1~)-1 = 0, ( p = 1,. * .,m)

where

(9.4)

pj" = mil-.
We shall show that (9.3) has periodic solutions of period Tp,and it follows
from (9.1) that, then all x&), (i = 2,. , ,,n)are also periodic of the same
period. Hence, the resulting motion is a normal mode vibration.
We shall integrate (9.3) for the Cases I and 11 where, for

Case I and 11: %l =0 when x, = XI> 0,

(9.5) Case I: x,= 0 when t = 0,

Case 11: xl = X I when t-0.

The conditions for Case I1 are ilzitial conditions because x, and are speci-
fied at the same instant of time. However, in Case I the velocity and dis-
placement are prescribed at different instants of time. Hence, in Case I,
it must be shown a posteriori that, if x,(O) = 0, then there exists an %,(O)
such that %,(to) = X,, (X,,t, > 0).
First, we change (9.3) into the canonical form by means of the trans-
formations [20]
(9.6) z = (c&8)l/2Xl"-1t; x, = t X 1 ; 1( = (k + 1)/2
and, because of the definition of n and the bounds on k,

(9.7) n=l when k = l ; 1 / 2 < ~ < 00.

Under (9.6), the equation (9.3) goes over into

and equations (Q.6)become for

Case I and 11: c=O when E= 1,
(9.9) case I: €=O when v=O.
Case 11: €= 1 when r=0.
 NONLINEAR VIBRATIONS 193

A first integral of (9.8) satisfying the first of (9.9) is

(9.10) E' = f (1 - I[p)'le

and a second integral becomes, for Case I

c
(9.11) 7 = kj(1 - IYlen)-w,
0

and for Case I1

< <
with 0 [ 1. The sign ambiguities in (9.11) and (9.12) are readily re-
solved with the result that the +
sign must be chosen in (9.11) and the
- sign in (9.12) [W].
The question whether there exists a real to>0 such that E ( T ~ )= 1
is now easily answered by observing that (9.11) exists when E = 1. In
fact, as observed by many authors 131, 32, 10, 331 under the change of
variable
(9.13) r& --s

the integral (9.11) with upper limit 6=1 becomes

is the completebeta function B(P.q) of parametersp = 1/2%,

where B(1/(2%),+)
q = 4. Similarly,

(9.16)

where B&q) is the incomplete beta function+ of argument [* and par-

ameters p = 1 / 2 ~ q, =+.
t We use here Pearson's notation [84].
194 R. M. ROSENBERG

2. The Inversions
From the above results it follows that the solutions of (9.8) are, for
case I

(9.16) z=
1 ’
(- )
1
Bp 2n’T ’
1

and for Case I1

(9.17) t= &[B 1 1
-

with
(9.18) [* = €11”.
But, if € ( t is
) periodic, the solutions (9.16) and (9.17) cannot be single-
valued on an interval exceeding one-half period. Hence, it is desirable to
invert these solutions. The definitions used in the inversions, and the trans-
formations are summarized in the table below. Cumbersome sign distinction
are avoided if “n is regarded to behave like an odd integer.” This phrase
+
means that negative quantities, raised to the power fi p , where 9 is either
zero or an even integer, remain negative, and negative quantities, raised
+
to the power n q where q is an odd integer, become positive. The
absolute values of these powers are given by
(9.19) ,*+P = l,P+P

with a similar expression for powers n + q.

Case I Case I1
 NONLINEAR VIBRATIONS 196

0 < t1& < 1;

It is seen that the inversions lead to four ateb-functions: Those called
a m p n ~ are
~ , ~inversions of C19(~1,2)r
where GIm2are beta functions, and
those called Sam (nul) and cam (nuz) are respectively the inversions of
Flr(tl,z)which are also beta functions.
A certain number of properties of ateb-functions have been determined
[20] and are summarized below.

3. Thc Am9-Functions

The amp-functions are odd, or

(9.20) amp - n ~ amp

~= - , nu1.2.
~

When n = 1,
(9.21) amp %I+ = '%,a.
Particular values are given by

amp nu* = 4 2 ,
(9.22)
amp0 = 0,
where

(9.23)

They satisfy the relation (also satisfied by the elliptic am-function)

where fi is a positive odd integer.

Their first derivatives are
d
.- (amp mI)= N sam* - (nu1),
clrc,
(9.26)
d
-(amp nt4z)= n cam" - (n94).
dua
196 R. M. ROSENBERG

Special values of these derivatives are

du,
=1 when It-1, for u,=O
=oo when n < 1,)

for %=y+.

(9.26)

I
d
-(ampn%) = 0, for %= 0,
du,
d
-(ampn%)=O
dust
=1 for %=%+
=do when n t 1.
,
(9.27)

“amp”-functions look as shown in Figure 7(a) and @).

amp nul amp nu2

I I

FIG.7. The amp-functions.

 NONLINEAR VIBRATIONS 197

4 . The Sam- and Cam-Functions

The sam- and cam-functions are periodic of the same period; i.e.,

(9.28)
+
sam (ml)= sam (nu1 4nu*),

cam (nus)= cam (1221, + 4nu*),

where M* is defined in (9.23).
One is odd, the other even, or

sam (-a%) = - sam (n%),

(9.29)
cam (- nup) = cam (nHn).
When n = 1, they become circular functions, and when n = 2, they become
elliptic functions, or
sam (ul) = sin ul,

cam (113 = cos %,

sam (2%) = sn (2-'/*.y),

cam ( 2 ~ =
~ cn
) (2-1/2,~a).

If ulo and U s o are those values of u1 and ugfor which amp nulo = amp nuaO,

(9.31) +
Sam2*(nulo) cam& (nuso)= 1,
thus generalizing a well-known identity of trigonometric functions. The

I
derivatives of these functions are

d
-
dU,O
[sam (nulo)]= cam" (nuno),
(9.32)
d
K~[cam (nuno)]= - sam" (nulo),

I
-
d4
[sam (aul)]= - n sam%- 1 ( n y ) ,
(9.33)
da
7[cam (*US)] = - n cam*"-' (w).
The sam- and cam-functions look as shown in Figure 8 and Figure 9 for
a variety of values of n in 9 n < <00.
198 R. M. ROSENBERG

FIG. 8. The sam-function.

For n = 1, the curves are the trigonometric sine and cosine curve,
respectively. For n = 4, they are parabolas, and for B = 00, they consist
of segments of straight lines.

0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 16 1.8 2.0

u2

FIG. B. The cam-function.

It is evident that the solutions of (9.8)are for Case I and 11, respectively,
.f = Sam (m),
(9.34)
.f = cam (n7),
as one can readily verify by direct substitutions, and by making use of (9.33).
 NONLINEAR VIBRATIONS 199

X. NONSIMILAR VIBRATIONS[8, 181

NORMAL-MODE

1. The Pertwrbation Potential

The trajectories of nonsimilar normal-mode vibrations of autonomous,
admissible systems are strictly homogeneous curves passing through the
origin of the configuration space and intercepting the La-surface, and they
are integral curves of the system M defined in (3.12).
At present. there exists no hope of finding the integrals of M in the
general case, considering the nonlinear character of the system (3.12).
However, as shown above, there exist many strongly nonlinear systems
whose modal lines are straight. Systems "neighboring on these" may have
curved modal lines; hence, their normal-mode vibrations may be nonsimilar..
Here, we shall show how such modal lines may be found by perturb-
ation methods. The results, reported here, are due to Kuo [MI.They are
generalizations of the results found earlier in the case of two degrees of
freedom [S]. Let
(10.1) Uo= Uo(xl, . . ., x , , ; q , . . ..m,,;al,. . .,at)
be the admissible potential function of an autonomous system having at
least one straight modal line, defined by
(10.2) xi* = cix,, (i = 2,.. ,,n).
In (lO,l), the mi, (i = 1,. . . , f i ) are the n masses and the u j , ( j = 1,...,2)
are parameters defining the properties of the springs.
We shall examine systems whose potential function neighbors on U,,;
hence, we call Uo the potential function of the parent system.
We shall say, "a system neighbors on the parent system" if its poten-
tial function is of the form
(10.3) 0 = u, + ewe, 1.1 << p o l ,
where 0 is admissible; consequently, wo must also satisfy the admissibility
conditions of potential functions. We call 0 the potential function of the
perturbed system, and w,, the perturbation potential. It is evident from
(10.1)that the mi and uf are the only parameters subject to perturbations.
Therefore, the most general form of the perturbation potential is

(10.4)

where the masses and spring parameters of the perturbed system are

(10.6)
*i = mi + &a& (i = 1,.. .,?$),
Gj = uj + && .
( j = 1,. . J ) .
While some of the ai and Pi may be zero, all are fixed, once chosen.
200 R. M. ROSENBERG

2. The Curved Modal Line

Kuo [8] demonstrates

Theorem X-I: If the parent system
(10.6) zi = avolaxi, ..
(i = 1,. ,n)

possesses, at some energy level h, a straight modal line, then the fierturbed system

(10.7) Yi = a 0 / a x , , (i = 1,. . .,%)

moving at the same energy level h, possesses a wnique modal line in an &-neigh-
borhood of that of the parent system.
The proof of this theorem is lengthy and will not be reproduced here.
In view of this theorem, the modal line of the perturbed system may be
written as
(10.8) x&) = cixl + E&~(X,) + e*&(xl) + . . ., (i = 2 , . . .,n)
and this is valid everywhere except, possibly, on the La-surface where the
system M is singular.
Then, substituting (10.8) in (3.12); i.e., in the system MI and expanding
in powers of E , it turns out that must satisfy the system of linear, inhomo-
geneous, second-order equations

(10.9)
where

(10.10)

and a sirn..ar notation is used for oo* and for the partia. derivatives of
these quantities. Moreover, it turns out that hi must satisfy a system
of equations in which the left-hand side is identical to (10.9) while the right-
hand side becomes a function the tlj. Consequently, one can find &, if one
can solve (10.9), and we need concern ourselves only with (10.9); i.e., with
first-order perturbations.
 NONLINEAR VIBRATIONS 201

3. Integrable Cases

There exist certain cases in which the solutions of (10.9) is reducible

to quadratures [8, 181. Where this cannot be done, (10.9) must be integrated
by numerical methods.
One of the integrable cases is that of weak coupling where the forces,
exerted by the coupling springs under finite displacements, are of order E ;
i.e.,

(10.11)

Under the assumption of weak coupling, the system (10.9) reduces to

where the subscript 1 on the tj1has been deleted.

Defining the function
(10.13) uo* + h = G(Xi),
it can be shown that the system (10.12) becomes the self-adjoint system

(10.14) 2(G[i')' - (G't;)' = - C&(Z~), (i = 2 , . . .,n)

where

(10.16) =5
a@,) h= 1
ch2 5
XI

0
(cj awe*
ax,
- w)
axj
dx,, (i = 2,. . .,%).

The general solution of (10.14) is

where A j and Bi are constants of integration.

Inasmuch as the modal line of the perturbed system must pass through
the origin, one can readily show that Bd = 0. (i = 2,. ..,n).However, the
only other condition on modal lines-that they intercept the &-surface-
does not fix the values of the Ad. Nevertheless, it has been shown for systems
m2 R. M. ROSENBERG

with two degrees of freedom [8], and by Kuo [18] for those with any finite
number of degrees of freedom, that the Ai = 0, (i = 2,. . .,n)as well, when
coupling is weak. The arguments leading to this result are lengthy and
not simple, even in systems with only two degrees of freedom, because the
system of equations M is singular on the La-surface. Therefore, we shall
not reproduce them here. However, the result leads directly to
Theorem X-11: The slope at the origin of the modal line of the Perturbed
system i s the same as that of its weakly coupled parent system, on which it
neighbors.
It follows that the modal line of the perturbed system is given, to first
order in E , by
(10.17) xj(xl) = cix, + e&(xl), (i = 2,. ..,n)
where

(10.18)

and the. 0,are defined in (10.16).

I t is evident that the modal line of the perturbed system is curved in
the case considered here because the modal line of the perturbed system
has, at the origin, the same slope as has the straight modal line of the parent
system, but everywhere else, these modal lines do not coincide. Hence,
the corresponding normal-mode vibration is nonsimilar.
A number of examples, involving perturbations of mass .and/or spring
parameters have been computed [8, 181. In one of these [8], results of
first-order perturbation theory have been compared with the actual behavior
of the physical system. Even though, in that particular case, IeI is much
larger than contemplated in perturbation theory, the agreement is excellent.

4. General Properties

It was observed in Theorem V-VI that every modal line, straight or

curved, intercepts the L,-surface orthogonally. One can also show readily
by means of (3.12)that a curved modal line intersects no equipotential
surface orthogonally, other than the La-surface. Moreover, Theorem V-V
states that straight modal lines intersect every equipotential surface ortho-
gonally. From these observations it follows that a straight modal line
is invariant under changes of energy level h, but a curved modal line is
not. In other words, if the modal line in a given mode of motion is curved,
a new modal line must be computed for each value of h.
 NONLINEAR VIBRATIONS 203

Consider now (4.4) with f,(t) 0; i.e., the equation determining the
period of a normal-mode vibration. It is well known that the period T,
given in (4.9), is independent of X, if (4.4)is linear, but it does depend
on X, if (4.4) is nonlinear. Summarizing these results leads to an interesting
gradation of normal-mode vibrations. In it, we replace the words “modal
line” by “mode.”
(i) In linear systems, the normal-mode vibrations are always similar;
the normal-mode coordinates decouple the equations of motion for arbitrary
motions; mode and period of the normal-mode vibration are independent
of the energy level of the motion. Linear combinations of solutions are also
s o htions .
(ii) If the system is nonlinear, the normal-mode coordinates decouple
the equations of motion for that mode only. If the normal-mode vibration
is similar, the mode is independent of the energy level, but the period is
not. Linear combinations of solutions are no longer solutions.
(iii) If the system is nonlinear and the normal-mode vibration is non-
similar, mode as well as period depend on the energy level; otherwise, the
conclusions are the same as in (ii) above.

XI. EXACTSOLUTIONSTO STEADY-STATE VIBRATIONS[36]

FORCED
1. Introductory Remarks
Steady-state forced vibrations can only arise in nonautonomous systems,
They are periodic motions of certain physical systems that are acted upon
by time-dependent, equiperiodic forces. We employ
Definition 111: Steady-state forced vibrations of an admissible, nonautono-
mous system are vibrations-in-mison having as their least period that of the
exciting forces.
In all that follows, we shall assume (as is usually done in the linear
problem) that only one of the masses, say m,,is subjected to an admissible
forcing function. Hence, the system H is

(11.1)

and the system M is

where primes denote differentiation with respect to x,.

204 R. M. ROSENBERG

Among the systems represented by (11.1) and (11.2) there exists an

entire class whose steady-state forced vibrations can be determined without
approximation. It is clear that the linear system, for instance, belongs to
that class. We shall show that exact solutions of the steady-state forced
vibration problem can be found for the entire class of homogeneous systems
under suitable excitation; among them is the linear system under simple
harmonic excitation.
The technically most interesting information regarding steady-state
forced vibrations is contained in the so-called “resonance curves” well known
from linear theory. These are the families of curves in the frequency-
amplitude planes which relate, for prescribed amplitudes of the forcing
functions, the amplitudes Xi of the motions in the various degrees of freedom
to the frequency o of the exciting force. In the steady-state vibrations
.
of nonlinear systems these response curves Xd(o), (i = 1.. . ,w) play a
similar role. We shall show that these curves are integrals of equations
of the form [35]

(11.3)

and that the Xiw-planes contain certain singular points. The number,
location, and character of the singular points of (11.3) can then be used to
determine the geometrical character of the response curves. In particular,
one can ascertain where (if a t all) resonance occurs, and at what frequencies
(if a t all) the phenomenon of “tuned dynamic vibration absorption” takes
place.

2. Homogeneom Systems

The steady-state response of the linear system to periodic excitation

is normally discussed for the case when the excitation is simple harmonic
[36, 371. In fact, the response curves cease to have meaning under general
periodic excitation. Since the steady-state response of the linear system
to simple harmonic excitation is, again, simple harmonic, it turns out that
in that case,

(11.4) xi(t)/f(t) = const., .

(i = 1,. . ,n).

We shall show that the linear system under simple harmonic excitation
is merely a special case of the homogeneous system under the excitation
V1(t)lkwhere
(11.6) f i ( t ) = P,cam (Ha),
 NONLINEAR VIBRATIONS 206

and the steady-state response of this class is of the form

(11.6) x&) = X i cam (%at), (i = 1 , . . . ,n);
hence, (11.6) and (11.6) are such that (11.4) is satisfied for all k in 1 k < w <
<
(or 1 n < 00). We shall make this demonstration only for a system having
two degrees of freedom [35], but we indicate the results of generalizing
to any number of degrees of freedom [12].
Let the degrees of freedom be
(11.7) u, = u, 21, = v,

and write the forcing function in the form

(11.8)

where
(11.9) D = (cX)~-*,
and c is a constant. Form (11.8) has a certain advantage over (11.5). In
(11.8), the quantity I plays a role similar to that of the circular frequency
of simple harmonic excitation, but it is not the frequency of excitation when
k > 1. Hence, I will be called the generalized circular frequency.
Then, the equations of motion of the physical system are

(11.10)
m,ii + aluk + a& - v ) k = P, camk (n~ / ( D A ~t )/ ,N )
m,i + a3vk- a2(u- v ) =~ 0,
those of the pseudo-system are

(11.11)
x+-=
.* ;L( Vm1
m, 7(
+=-=-=
m1 Vm1 VY_,)*=+camh(nEt),

and the equation of the trajectories is

208 R. M. ROSENBERG

( 11.12)
Simple substitution of

(11.13)
y = cx

into (11.11) and (11.12) (and making use of the formulas (9.33) for the
derivatives of cam-functions) shows that (11.13) are indeed the solutions,
and the constant c in the second of (11.13) is the same as that in (11.9).
But it follows from (11.13) that, in the case considered, x(t) and y(t)
are cam-functions of the same argument. Hence, the steady-state vibrations
are similar, or the trajectories satisfying (11.12) are straight lines.
If we define
(11.14) Y = cx.

the equations (11.11) and (11.12). with (11.13) and (11.14) substituted in
them, are three equations in the three unknowns X,Y and c, and they can
be solved explicitly for these quantities.
One finds
c = aJl*/{a,'Ik + (a, - ~ J h + W J ~ ) 1 / b } ,
(11.16) Xk = { P o m ~ / 2 [ a , + * - ?%a(~+~l/*A*)1P]k)/R,
~ / (as

Yk = { Poa,m,RI,}/R
where
 NONLINEAR VIBRATIONS 207

(11.17)

and substitution shows that

(11.18)
x1 = x,cam (n YFt)
Xj(t)/Z,(t) cj = X j / X , , (i = 2 , . . .,m)
satisfy this system as well as the equations of the trajectory which are not
reproduced here. Now, there are m quantities Xi and m - 1 quantities
cj which constitute the unknowns, and there are m equations of motion
and m - 1 equations of the trajectory. Hence, the problem is determinate
WI.
It is now advantageous, as done in the linear forced vibration problem,
to introduce non-dimensional quantities as follows:

(11.19)

y = P/[a,/m,(k+1)q
where it should be remembered that X and P are nondimensional ampli-
tudes, and y is a nondimensional generalized frequency ratio. With the
introduction of these quantities, (11.15) become

where
208 R. M. ROSENBERG

One must now find the locus of the endpoints of the trajectories, i.e.,
the L-surface, which is, here, a curve. The L-curves are found as follows:
the first of (11.20)is solved for y as a function of E. This is substituted in
the second, thus eliminating y , and resulting in an equation of the form
(11.22) xk = qE).
If one substitutes in that equation
(11.23) E= PIR,
one finds an equation in x
and p ; this is the equation of one of the loci
of end-points of the trajectories. The second locus is found by substituting
x
- for 19; then, the equations of the two loci become [36]
(11.24) x k y + aa1(8- p ) k ( p+ p:2x\) - a z l a , , p l ~ ~ ~ *p = 0.
2

From these equations one can readily deduce the following properties of
the L-curves :

(i) they intersect the y-axis at y = 0 and at 9 = f a i l ” ;

(ii) they intersect the &axis at the origin only;
(iii) in the linear system ( k = l), they are hyperbolas;
(iv) as 1x1
and (91+00,718- constants equal to the slope of the modal
lines of the autonomous system [36].
The last property proves the following
Theorem XI-I: A s the generalized freqzcency ratio y + y1,2, where yl,2
are the generalized frequency ratios of normal-mode vibrations of the autonomozcs
system, the nondimensional amplitudes IxI,IP1 -* w . Hence, the system goes
into resonance in the neighborhood of free normal-mode vibrations.
The L-curves and trajectories of steady-state vibrations have been
computed for the nonlinear problem

2
(11.26) ple = 2, aB1= 112, as2= 3, k =3
and for the associated linear problem in which all parameters have the
values given in (11.25),except that k = 1. The results in the 27-plane are
shown for these two examples in Figures lO(a) and (b), respectively.
For an understanding of the general theory, to be presented in the next
section, it is helpful to interpret these diagrams. In general, the trajec-
tories of steady-state forced vibrations are, here, segments of straight lines
 NONLINEAR VIBRATIONS 209

( b)

FIG. 10. Trajectory of steady-state forced vibration in (a) nonlinear and (b) corre-
sponding linear systems.
210 R. M. ROSENBERG

that pass through the origin and terminate on the L-curves. Their slope
is a function of the frequency y and, in fact, their slope E(y) is given in the
first of (11.20).
Assume that y, i.e., essentially the frequency of excitation, is small,
so that E is less than the slope of the modal line in the first and third quadrant
(shown dotted). As y is increased, the trajectory begins to rotate in a counter-
clockwise direction and becomes longer, tending to infinity as the slope
E(y) approaches that of the modal line. As y is further increased, the trajec-
tory continues to rotate, but it becomes now shorter until E = 00. This
occurs where G ( y ) = 0 and, at this frequency, there is bounded 9-motion,
but no 3-motion. In other words, the +motion vanishes identically because
m2 acts as a tuned dynamic vibration absorber for m,. A further increase
in y will further rotate the trajectory, and the motion becomes out-of-phase
because the slope ?< 0. At first, the trajectory will lengthen until the mo-
tion becomes, again, unbounded when y = yz. However, further increases

these amplitudes tend to zero as y -.

in y will reduce the amplitudes of the steady-state forced response, and
m.

3 . The Response Curves [36]

The differential equations, satisfied by the response curves, are found

from the second and third of (11.20); they are

(11.26)

where
(11.27) F ( y )= ~ G ’ o v w- G(YW’(Y),
and G and H are defined in (11.21).
Both of (11.26) are seen to have singular points on the y-axes a t the
zeros of H ( y ) . The first of them has additional singular points on the y-axis
at the zeros of G ( y ) .
Now, the generalized frequency ratios of normal-mode vibrations of the
autonomous system are, in general, functions of the amplitudes R and P,
and they tend to certain limits as the amplitudes tend to zero. It can be
shown [12, 131 that their limiting values coincide with the zeros of H ( y ) .
In other words, the equation

(11.28) H(Y) = 0
 NONLINEAR VIBRATIONS 211

can be shown to play the part of the so-called “frequency equation” of the
autonomous system. Hence, both equations (11.26) have singular points
at the natural frequencies of normal-mode vibrations at vanishingly small
amplitudes.
Denote the zeros of H(y) by y,,. Then expanding the first of (11.26)
in the neighborhood of (K= 0, y = y,,) one finds

(11.29)

because
(11.30) fi(yn) = - G(y*)H’(yn).
Expanding the second of (11.26), one has directly

(11.31)

The integrals of (11.29) and (11.31) are

(11.32) .pm” = const., Py,,k = const.

These are equations of hyperbolas; hence, the singular points on the

y-axes at the zeros of H ( y ) are for both (11.26) saddles.
Let the zeros of G(y) be denoted by ym. Then, expanding the first of
(11.26) in the neighborhoods of (8= 0, y = ym) one finds

(11.33)

because

(11.34) fi(yu) = kG‘(yu)H(yu).

The integrals of (11.33) are the straight lines

(11.36) Rly,,, = const.

Hence, the singular points of the first of (11.26) on the y-axis at the zeros
of G(y) are star points (nodes).
Now, it follows from the definition (11.21) that Gly) has only one real
zero at
(11.36) yu = y* = 1 + Q.2.

Hence, we have [36]

212

FIG. 11. Frequency-amplitude curves in (a) &-plane and @) ?y-plane.

 NONLINEAR VIBRATIONS 218

(a) ( b)

FIG. 12. Frequency-amplitude curves in physical %&plane (a) and l b p l e n e (b).

<
Theorem XI-11: The system (11.11)i s szcch that, for any k ifi 1 k < 00,
there exists a single nondimensional generalized fveqzrency ratio y* of the ex-
citing force swh that the mass m, acts as a tuned dynamic vibation absorber
for the mass ml that i s bcilrg excited.

Response curves in the 8 y and Py-planes for the values of the parameter
given in (11.26) are shown in Figures ll(a) and @), respectively.
It remains to map the Xy and Py-planes on the KQ and PQ-planes,
where the frequency Q of the forcing function is defined through (11.6).
It follows from (9.23) and (11.19) that this mapping is given by [%I

and by a similar formula involving P which is found by replacing 8 in

(11.37) by l’/E The results of this mapping for the example of Figure 11
is shown in Figures 12(a) and (b).
214 R. M. ROSENBERG

XII. STEADY-STATE FORCEDVIBRATIONS BY APPROXIMATEMETHODS

1. Descriptkorr of the Method [12]

The problems of steady-state forced vibrations, considered here, are those
of systems that neighbor on systems of known vibrations-in-unison. Two
cases may arise : either the vibrations-in-unison of the unperturbed (parent)
system are themselves steady-state forced vibrations of the parent system,
or the vibrations-in-unison of the parent system are normal-mode vibrations
of that system. It is clear that in the first case the amplitude of the forcing
function need not be small, but in the second case, that amplitude must be
small in absolute value because the nonautonomous system now neighbors
on the autonomous one. In either of these two cases, the problem of finding
steady-state forced vibrations can be reduced to the perturbation problem
discussed in the section on nonsimilar normal-mode vibrations. The results
reported here are, largely, due to Kinney [12].
We consider a parent system of potential function
(12.1) u = U(X,, . . ,xn)
I

for which at least one vibration-in-unison is known. Let the trajectory in

configuration space of this vibration-in-unison be given by the (known)
functions
(12.2) xj = X j * ( X , ) , (i= 2,. . .,n).
Now, the only vibrations-in-unison that have been deduced are either
straight lines, or they are curved in the sense treated in the section on non-
similar vibrations. Hence, they are always given by
(12.3) Xf*(X1) = cp, + O(a,)
where is small, and when the trajectory is straight, the terms of order
E~ are absent in (12.3).
The time-history of x1 of the known vibration-in-unison is the (known)
function
(12.4) x1 = x,*(t),
and the time-histories of the other xi are found by substituting (12.4) in
(12.3). Since the motion is a vibration-in-unison, (12.4) possesses on the
interval LO,X,] (where X, is defined by %l(Xl)= 0) the inverse
(12.6) t = t*(x,).
We shall, first, suppose that the parent system is nonautonomous, or
the unit mass of the unperturbed pseudo-system is subjected to a force
in the /,-direction given by
(12.6) fl = f(t) = f(t + T)
 NONLINEAR VIBRATIONS 215

where T is the only period of f . Hence, the circular frequency of the periodic
force is
(12.7) w = 2n/T.

We consider a neighboring, or perturbed system of potential function

(12.8) 0 = U(x,, .. .,xn) + &+(XI, * * -,%&)

where #(xl,. . .J,,) is a given perturbation potential which satisfies the

admissibility conditions, and
(12.9) &* = O(&)
everywhere where t,4 is defined. Then, the trajectories of the perturbed
system are the xi(xl),(j= 2,. . .,n)which satisfy
"

i.e., t * ( x l ) is the zeroth order approximation of t ( x l ) .

If one now uses an iteration scheme with respect to t(xl), combined
with conventional perturbation techniques for the xi(xl), the problem
of determining the steady-state forced vibrations reduces to that of (10.9),
treated in the section on nonsimilar normal-mode vibrations. The procedure
consists in replacing t ( x l ) by to(x,) in (12.10) and then using conventional
perturbation methods ; these result in a linear nonhomogeneous system
of equations which the &(xl) must satisfy. Now,the solutions of that system
are functions of to(%,),and this will be indicated by using an appropriate
superscript on the ti. Using the notation

(12.13) Q*(xJ = Q ( x 1 ~ 2 . 1 t - scnxi)

for U and t,4 and their partial derivatives, and using the definition of F(x,,X,)
given in (3.10), the equations of the tj turn out to be
216 R. M. ROSENBERG

(12.14)

But, this system of equations is of the form (lo.$), and questions of its
integration are discussed in the section on nonsimilar normal-mode vibrations,
and also by Kinney [12] for different examples.
Once the integrals #o)(x,) have been found the functions (12.11)
(12.16) xj(x1) = cjxl + E#')(x~), (i = 2,. . 1%)

are substituted in the first of the equations of motion; that equation can
then be integrated by quadratures for the conditions
3, = 0 when x1 = X,,
(12.16)
xl = XI when t=0.

This results in a function

(12.17) x, = X,(t,X,)
and its inverse
(12.18) t = tl(xl).

This latter is the first approximation of t ( x l ) (as compared to the zeroth

approximation), and it may now be substituted in (12.10), and the entire
procedure is repeated. Kinney [12] shows that it is more convenient to
introduce the transformation
(12.19) t = wt,

and he has computed a number of interesting examples, including that of

the tuned dynamic vibration absorber of weakly nonlinear two-degree-of-
freedom systems.
If the parent system is autonomous, one simply deletes the terms in
..
(12.10) involving f ( t ( x l ) ) and F(x,,X,), and one replaces y5(x1,. J,,)by
g(t(x,)) where g(t) is periodic of period T. In view of (12.9), this latter step
implies that the forcing function has small amplitude] and the result will
be that of steady-state forced vibrations in the neighborhood of the vibrations-
in-unison of the autonomous nonlinear system. It may be interesting to
interpret the results of forced steady-state vibrations neighboring on the
 NONLINEAR VIBRATIONS 21 7

vibrations-in-unison of a parent system. This interpretation will be made

in terms of trajectories in the configuration space, and of response curves
in the frequency-amplitude planes. As an example, we consider a parent
system that is linearizable within the meaning of (2.11).
Consider the n-space and an equipotential surface indicated in Figure 13
together with the modal lines; in this diagram, the modal lines are sur-
rounded by &-tubes. Then, the desired trajectories of the steady-state

FIG. 13. Trajectories of free and forced vibrations in nonlinear n-degree-of-freedom

system.

forced vibration pass through the origin and are monotone because they
are those of a vibration-in-unison; moreover, they must lie wholly in the
&-tubesbecause of (12.11). Consequently, the L-surface which is the locus
of the endpoints of these trajectories, also lies inside the &-tube.
The direction with which the trajectories pass through the origin is a
function of the frequency w of the exciting force as explained in connection
with Figure 10. If that frequency lies near that of a normal-mode vibration,
the trajectory lies near a modal line and is long. If not, the trajectory is,
generally, short, and the amplitudes of the forced vibration are of O(E).
In this way, one understands readily the occurrence of resonance.
It is also instructive to translate this information into the frequency-
amplitude planes, shown in Figure 14. The dotted lines, composed of the
w-axes and the backbone curves [29, 381, are those of every vibration-in-
unison of the autonomous system. In other words, if the autonomous system
does not movein normal-mode vibrations, it is at rest with respect to admissible
motions. (Because of Definition I, every vibration-in-unison of the auto-
nomous system is a normal-mode vibration.)
Hence, we may surround the w-axes and the backbone curves by &-tubes
as shown in the diagram, and the forced steady-state vibrations of the non-
autonomous neighboring system must be represented by response curves
lying inside these &-tubes. One such response curve, for a prescribed, small
force-amplitude is shown. It is evident, then, that the intersections of
218 R. M. ROSENBERG

w-axes and backbone curves are saddle points. In addition to these, there
may be star points as well, if the steady-state forced vibrations are similar.
If they are not, one can easily show that the only singular points in the

FIG. 14. Frequency-amplitude curves in nonlinear rP-degree-of-freedom systems.

frequency-amplitude planes occur at the intersection of backbone curves

with the w-axes, and all are saddles.

XIII. STABILITY

1. Definitions
Two definitions seem particularly useful in the examination of the
stability of vibrations-in-unison : “stability in the sense of Liapunov,”
or L-stability and “stability in the sense of Poincarb” or P-stability [39].
To define the former, let
x; = % i +. (.t ) , I
(13.1) (i = 1,. ..,n),
x;*(O) = xi, &*(O) = 0,
 NONLINEAR VIBRATIONS 219

denote the set of solutions of a vibration-in-unison, satisfying the system

H. Further, let
(13.2) xi = Zi(t), (i = 1,. . .,n),
denote any set of solutions of the system H. Then, the set x,*(t) is called
L-stable if, for every E , with O < e << X i , (i = 1,. . .,n),there exists a S(e),
<
with O < S(E) E , such that
(13.3) IZi(t) - x;*(t) I < E
whenever
(13.4) IXi(0) - S*(O)l< b(E).

To define P-stability, let

1
xi = .j*(X1),
(13.5) (i = 2, . . .,N),
Xi(&) = xi,

denote the set of solutions of a trajectory corresponding to a vibration-in-

unison, and satisfying the system M. Further, let

(13.6) xi = X i ( % , ) , (i = 2 , . . .,n)

denote any set of solutions of the system M. Then, the set xi*(xl) is called
P-stable if, for every E with 0 < E << X i , (i = 1,. . . ,n),there exists a 6(e)
<
with O < S(E) E such that
(13.7) Ixi(x1) - xi*(xI) I < E
whenever
(13.8) Ixi(x,) - Xi*(X,) I< 44.
The interpretations of these two definitions of stability are clear. Let
.
the space having coordinates (x,,. .,x,,,t) be called the event-space. Then
..
x,*(t), (i = 1,. ,N) is a curve in the event-space. If any solution xi@),
having a single point inside an &-tubeabout xi*(t) remains within it every-
where, the solution is L-stable. Similarly, xi*(xl) is a curve in the con-
figuration space. If any solution xi(xl), having a single point inside an
e-tube about xi*(xl) remains within it everywhere, the solution is P-stable.*
It is evident, therefore, that every L-stable solution is also P-stable, but
the converse need not be true. Moreover, every solution that is P-unstable,
is also L-unstable but, again, the converse need not be true.

This type of stability is frequently called orbital stability [a].

220 R. M. ROSENBERG

In both definitions of stability it is required that I E ~ be small. Consequently

it is sufficient to consider the equations of the first variation of H with
respect to xi*($) ; i.e.,

(13.9) &= jzl

I
as
tjjgjg [V(x,*(t),.. .,x,,*(t))], .
(i = 1,. .,n).

Clearly, this is a system of linear equations with periodic coefficients, and

stability in any sense requires that all characteristic exponents of (13.9)
have real negative parts. Now, the number of characteristic exponents
depends on whether the system H is an explicit function of t, or not; i.e.,
whether the system H is autonomous or not [26]. Hence, the two cases of
normal-mode vibrations and of steady-state forced vibrations must be
discussed separately.

2. Stability of Normal-Mode Vibrations

I t is simple to deduce [21]

Theorem XIII-I: If an admissible potential function U is not a quadratic
form, the normal-mode vibrations are never L-stable.
This result is an immediate consequence of the well-known fact that the
periodic solutions (if they exist) of autonomous, nonlinear second-order
equations are not isochronous. It follows from this theorem that normal-
mode vibrations of nonlinear systems are, at most, P-stable, and their
stability properties depend in general, not only on the magnitudes of the
mass and spring parameters, but on the energy level of the motion as well.
The problem of discussing the behavior of the ti in (13.9)is greatly
simplified when that system of coupled equations can be decoupled. If the
system (13.9)is written in the matrix form

(13.10) i'+B(t)E= 0
it turns out that the matrix B(t) is of the form

(13.11)

where B, and the Bi are square matrices with constant elements, and the
fi(t) are periodic scalar functions of time. Then, if a transformation matrix
T exists such that
(13.12) T-'B(t)T = d ( t )
 NONLINEAR VIBRATIONS 221

where A ( t ) is a diagonal matrix, the transformed system (13.9) is decoupled.

Hsu has shown [M] that the conditions, necessary and sufficient for the
existence of such a transformation matrix T is, that the Bi, (i = 0,. ..,m)
be all commutative and that at least one of them have distinct eigenvalues.

3. Homogeneozcs Systems
It is simple to demonstrate [7]
Theorem XIII-11: The system of variatioml equations of admissible,
homogmeous systems with respect to any normal-mode vibratioH can dways
be decwpled.
To prove it, one shows that, in homogeneous systems, (13.11) can always
be reduced to

(13.13)

where A is a constant, square matrix. Thus, Hsu’s criterion is trivially

satisfied.
The normal-mode vibrations of homogeneous systems have the further
(unusual) property that their stability does not depend on the energy level
of the motion. In this respect, they are similar to linear systems, the latter
being stable at all amplitudes. If a normal-mode vibration of a homoge-
neous system is stable (unstable) at any one amplitude, it is stable (unstable)
at all amplitudes [9]. Whether a given normal mode vibration of a given
homogeneous system is stable or not depends only on the mass and spring
parameters of the system, and on the particular normal mode examined,
but not on the energy level.
As an example, consider a homogeneous system having the degrees of
freedom zc and v . Then, the equations of motion may be written in the
form suggested by Loude [41]:

(13.14)
%ii = - b#(v) + A#(% - v )
where
(13.16) $(w) = wlwl*-l.
If the trajectory of a given normal mode is

(13.16) v = cu;
222 R. M. ROSENBERG

(13.17)

one can show that the variational equations have solutions that satisfy [41]

where B takes on one or the other of the two values

Bl = C,
(13.19)
B, = - ml/cm,.
The variational equations reduce to

(13.20) € +4 s $ ( q ) € =0

where b,,*are constants defined by

D 1 --- a
+ -((IA - c)Il - c(k--1,
m1 m1
(13.21)

It is easy to show that these equations are independent of the am-

plitude X. To do this, we note that the motion in normal modes is an
ateb-function. (The arguments which follow are the same, whether we deal
with Sam- or cam-functions; thus, only one of these need be considered.)
From (9.6) and (9.34),
x = X cam (nt),
(13.22)
T = (c/n)"2X"- ' t .

But, the independent variable in (13.20) is t, not t. Now, one has from
(13.22)

(13.23)

where E is a constant not containing X, and from the definition (13.16)

of the #-function together with (13.22),

(13.24) lp(gl) = Xb-'*(cam (nt)).

 NONLINEAR VIBRATIONS 223

Hence, (13.20) in terms of t, becomes

(13.26)

where Dlg = &/E. It follows that the two equations (13.26) do not
contain the amplitude X.
The discussion of the characteristic exponents of the equations (13.26)
is difficult because they are Hill equations in which the periodic coefficients
axe cam-functions. The detailed stability properties of such Hill equations
are not known.

-1 0 1 2 3 4

FIG.16. Strutt chart.

However, one can gain some insight into the stability of normal-mode
vibrations of homogeneous systems by replacing the cam-function in (13.25)
by a cosine-function. Proceeding in this manner is, in fact, equivalent to
expanding the periodic coefficient into Fourier series and retaining only the
first term of the expansion. This is usually done in examining the details
of stability of vibratory motions of nonlinear systems [4]. It has the con-
sequence that equations (13.26) become Mathieu equations whose stability
properties are well known [a, 421.
224 R. M. ROSENBERG

One finds, in place of (13.25), two Mathieu equations of the form

(13.26) €” + (81.2+ &1,2cos t)€= 0

and for stability of the normal-mode vibration. it is necessary, that both

these equations “indicate stability.“
The stability study is carried out by means of the Strutt chart [4].
This chart consists of a set of curves in the &-plane separating stable regions
(shaded) from unstable ones; it is shown in Figure 15. If the values P1(6,,el)
and P , ( ~ , , Edefine
~ ) two points PI and P, in the &&planewhich lie in the
interior, or on the boundary, of a stable region, the motion is stable. If
one or both of these points lie in the interior of an unstable region, the
motion is unstable. It turns out that the values of the parameters e,b in
one of the equations (13.26) is such that one of the points P ( E , ~always
)
lies on a boundary between stable and unstable regions. This is known to
lead, at most, to P-stability [4], thus verifying Theorem XIII-I. The other
point may lie anywhere in the &&plane.
The results of the computation for
(13.27) B=B1=c
are given below for that one of the two Mathieu equations for which P(e,S)
does not automatically lie on a boundary.
That equation is
(13.28) F’+ (6 + & C O S t ) 6 = 0
where [9]

(13.29)

k
B = - [a + A l l - cIR-1(1 - c)].
 NONLINEAR VIBRATIONS 226

For any given system parameters] the formulas (13.29) can be utilized to
determine the location of P(S,&). Similar results are found, when
(13.30) B=B,=-m 11cm8'

In either case, it turns out that the point P(e,S) lies on a straight line in
the &plane which passes through the origin, which lies always in the right
half-plane, and whose slope is a function of k only. When K = 1, this slope
is zero so that the point lies then on the (positive) S-axis which consists
entirely of stable points. Thus, linear systems are always stable. When
k > 1, the slope of the straight line is positive, and when O < K < 1. the
slope is negative. The location of P(8,e) on the straight line is a function
of the system parameters [28].

4. Symmetric System with Two Degrees of Freedom [9]

As a second example permitting decoupling of the variational equations,
consider the symmetric two-degree-of-freedom system (not homogeneous)
[9]. The equations of the pseudo-system are

(13.31)

and the trajectories of the normal modes are

(13.32) y = Yi*(X) = x ; y =yo*(%) = - x
where the subscripts i and o indicate an in-phase mode and an out-of-phase
mode, respectively. Using the notation

(13.33) e(x*(o*r*(w= Q*
for U and its derivatives, it is easily shown that, in symmetric systems,
one has always
(13.34)
*
uxx = uy,
*
for both, the i and o-modes.
The coupled variational equations are
.. * *
E - uxxt - u x y q = 0,
(13.36) * *
q - UXYE- U,,q = 0.
Under the transformations
tl = 'I+ 5,
(13.36)
c, = q - E,
226 R. M. ROSENBERG

the equations (13.35) become, in virtue of (13.34),

(13.311

and these are, evidently, decoupled.

As a specific example, consider the system+

(13.38) u = u ( L ) + U(N)
where

(13.39)

We now introduce the following notation, already indicated, in part, in

(13.33): when U or its partial derivatives are evaluated along a modal
line, that quantity is marked with *. When, its value along such a modal
line is computed at the values X , y ( X ) = Y, the quantity is marked with **.
If the modal line in question is that of the in-phase mode, the quantity is
supplied with an i-subscript, if it is that of the out-of-phase mode, the
quantity has an o-subscript. If no subscript is supplied the quantity is the
same in the two modes.
With this notation we define:

(13.40)

Note that the quantities arising from the linear part of the potential function
are the only ones independent of the mode.
It will be remembered from the earlier example that the nonlinear
vibrations x*(t),y*(t) will be replaced, in the vuriatiod eqaruhzs, by their
zeroth approximation; i.e., by simple harmonic functions. It is entirely
consistent with this approximation, to replace the frequencies of the normal-
mode vibrations by their Duffing approximations [4]. These are, in terms
of the quantities defined in (13.40)

t The superscripts (L)and (N)indicate the terms in U which give rise, respectively,
to linear and to nonlinear terms in the equations of motion.
 NONLINEAR VIBRATIONS 227

If one utilizes

(13.42)

the equations of the first variation become

(13.43) cY,~+ (&,\+ &\ cos ti,o)Ci,a = 0

where primes denote differentiation with respect to qO,corresponding
super- and subscripts must be used, and

or, combining them,

(13.46)

In (13.44) and (13.46), the + signs are used with the subscript 1, and the
- signs with the subscript 2.
Evidently then, (13.43) represents four equations, those for two modes,
and two equations for each mode. As in the earlier example, one of the
two equations in each mode defines points in the Strutt chart which lie always
on the boundary between stable and unstable regions of the &plane. Hence,
it is the other two equations which determine the stability; these are the
combinations (1,o) and (2,i).
In applying the foregoing formulas to the example of potential function
(13.38) and (13.39), it is convenient to introduce the nondimensional quanti-
ties
228 R. M. ROSENBERG

In terms of these quantities, the parameters become, for the i-mode,

(13.47)

and a similar result in the o-mode. In both cases, the parameters &,8depend
on E ; i.e., on the amplitude of the motion.

FIG. 16. Stability of in-phase mode of nonlinear, symmetric system with two
degrees of freedom.

The results for the i-mode are shown in Figure 16 and are as follows:
(i) The point on the Strutt chart, deciding on the stability of the i-
mode lies on a straight line passing through the point (C#, and intersecting
the positive &axis at a point that depends on oc, only. This straight line
is called a stability chracteyistic.
(ii) The portion of the stability characteristic lying in the upper half-
plane applies yhen u8 > 0, that in the lower half-plane when a, < 0.
 NONLINEAR VIBRATIONS 229

FIG. 17. Stability of out-of-phase mode of nonlinear, symmetric system with two
degrees of freedom.

The results for the o-mode are shown in Figure 17. They differ in some
respects from those for the i-mode.
(i) There exist, again, straight stability characteristics. These terminate
on a point lying on a straight line of unit slope, passing through the origin.
The point on this line. at which the stability characteristics terminate.
depends on a, only. Further, the stability characteristics intersect the
positive &axis at a point whose value depends on a, only. Thus, know-
ledge of a, and ol, determines the stability characteristic and the amplitude
of the motion determines the point on the stability characteristic. Items
(ii) and (iii), listed above for the i-mode also apply to the 0-mode. From
these results, knowledge of El, a,, a3 and the amplitude of the motion permits
the determination of its stability.
230 R. M. ROSENBERG

6. Nomimilar Normal-Mode Vibratiom [S].

We shall consider here the stability of a nonsimilar normal-mode vi-

bration of an autonomous system that neighbors on a parent system of
potential function
(13 -48) u = V(Xl,. . .,xn)
of equations of motion

(13.49) Zi = U,(xl,. . .,xn). .

(i = 1,. .,n)

and having a similar normal-mode vibration

(13.60) xi = Xi*(t), (i = 1,. . .,n).
Let the perturbed system have the admissible potential function

(13.61) 0 = V ( X ~ .,xn)
, e + E O ( X I , . . -,xn)j (14 IUl).
The equations of motion of the perturbed system are
(13.62) . + .
Y i = U,(X,,. .,xn) ~ o , ~ ( z ~.,x,)),
,. (i = 1,. ..,a),

and it has been shown in a previous section that this system has a normal-
mode vibration (in general nonsimilar) of the form
(13.63) g6(t)= xi*(t) + e&(t) + .. ., (i = 1,. ..#n),
Then, one can easily demonstrate [8].
Theorem XIII-111: The stability of the nomttal-mode Vibration (13.53)
of the Perturbed system i s the same as tlte stability of the similar n o d m o d e
vibratiolz (13.60) of the pareNt system.
For the proof, one forms the equations of the first variation of (13.62)
with respect to the solution (13.53). It turns out that these are precisely
the same equations as those of the first variation of the equations of motion
of the parent system with respect to the solutions (13.60). which proves the
theorem. Hence, if the variational equations of the parent system can be
decoupled, the stability problem of normal-mode vibrations of neighboring
systems is solved.
6. Forced Vibratiows

In general, the variational equations of the equations of motion with

respect to a vibration-in-unison cannot be decoupled, even when the tra-
jectories of the motion in configuration space are straight. In these cases,
Hsu has developed criteria for determining the stability [43, 4-41; his results
 NONLINEAR VIBRATIONS 231

are based on the application of the method of Strubel [39] in a slightly

modified form. A comprehensible summary of Hsu’s methods and results
is too lengthy to be reproduced here. However, we shall apply his method
to a particular case of steady-state forced vibrations in which the variational
equations cannot be decoupled; this example is due to Kinney [12].
We consider the nonautonomous, admissible system

21 = Ox,(%# - pxn) + f(@,

(13.64)
%j = Oxj(%l,....xn) ; 0’ = 2,. ..,%I,
where
(13.66) 0 u(+e- - - +n) + -$(%p - - -A),
and
(13.56) . . .,%”)
u = U(%,,
is a negative quadratic form. In other words, the parent system is a linear
admissible, nonautonomous system in which the mass “r, is subjected to
periodic forcing of period
(13.67) T=2 ~ 1 ~ .
We inquire whether a vibration-in-unison of the perturbed system,
given by
(13.68) xi = Xi*(& (i = 1,. ..,%)
is stable.
The equations of the first variation of (13.64) with respect to (13.58)
are

where use was made of notation (13.33) with respect to the second deriv-
atives of U and 4. Since ZJ is a quadratic form, the a W / a x i a x , are constants
for all i and k.
We Write (13.69) in the matrix form
(13.60) 2 + [B, + sB(l)]z= 0
where z is a column matrix, B, is a constant square matrix,
S
(13.81) B(1) =
8-
2 .. B,cosswt,
I$,.

and the B, are constant square matrices.

232 R. M. ROSENBERG

It will be noticed that, here, we have not made the assumption that the
solutions (13.68) are to be treated as simple harmonic for purposes of a
stability analysis. In fact, that assumption is unnecessary here because the
approach through Mathieu equations is not used.
It is now supposed that B, and the B, do not satisfy Hsu's criterion
[40] and that, in consequence, (13.69) cannot be decoupled.
We introduce the transformation
(13.62) z = TE

which is such that T is a constant matrix that diagonalizes B,. Under

it, (13.60) goes over into a matrix equation of the form

g+(n+EzD,COSSd 1
s
(13.63) l=o
s-1

where A
2 is a diagonal matrix with elements
(13.64) ~ 1 < wg' < . . . < wn2,
2

and the
(13.66) D,= [d$'], (S = 1,. . . ,S)
are constant matrices.
Then, Hsu's results show that instability occurs when, and only when,
the following inequalities are satisfied :

(13.66)

Now, the functional relation between the frequency w of the exciting func-
tion and the amplitude X, of x,*(t) is known. Hence, the inequalities (13.66)
determine the amplitudes X , (if any) for which the motion is unstable.
Kinney [12] has applied this theory to the weakly nonlinear problem
of the dynamic vibration absorber. The analysis is too detailed and tedious
to be reproduced here because of the involved form of the matrix elements
dii'. His result for that problem is:
The portions of the response cwves (in the freqwncy-am#Littde planes)
which have negative slope cmespond to unstable motion. All other portions
of these cwves correspolzd to stable motion.
 NONLINEAR VIBRATIONS 233

On the Existence of Simple Trajectories ila Admissible n-Degree-of-Freedom

systems *
In this appendix we employ
Definition A-I: An integral of the autonomous system H (3.3) or of the
system M (3.12)is said to be a simple trajectory in configwation space if
and only if:
(i) it passes through the origin;
(ii) it reaches the bounding, or maximum equipotential surface (6.8);
(iii) it has no tangency with any equipotential surface (6.12)except at
the origin and the bounding surface.
The last restriction was not imposed by Kuo [18] who introduced the
concept of the simple trajectory. But the lack of tangency with any equi-
potential surfaces U + h* = 0, 0 < h* < h is a property of modal lines.
In fact, every modal line is a simple trajectory, and every simple trajectory
which satisfies (i) to (iii) as well as
ax;
-# 0, (i = 1, . . ,n)
I

dx,
for all xl, is a modal line.
With the above definition we prove in this appendix the central
Theorem A-I: Every admissible azctonomotrs system possesses at kast
one simple trajectory.
Before presenting the proof of this theorem it may be helpful to sketch
its essential ideas; these are very similar to those used in the two-dimensional
problem [191.
We consider trajectories in configuration space which pass through (or
issue from) a point Po on the bounding surface; such trajectories are called
T-curves. We then demonstrate that every T-curve passes through an
infinity of isolated distinct points Qi(i = 1,2,.. .), corresponding to con-
figurations of stationary potential energy with respect to neighboring
points on T. The corresponding levels of potential energy are denoted by
hi, and the transit times through the Qj by t i . We show, further, that the
ti and hi are uniquely determined by the initial point P o , and both are
differentiable with respect to P o . Next, we demonstrate that there exists
at least one T = T*, issuing from a point Po* for which the ordered
Qi(i = 1,2,. . ,) lie alternately at the origin and at Po*. Finally, we show
that T* satisfies condition (iii) above ; this completes the proof.

* This appendix is based in large part on work by J. K. KUO [IS] and by

C. H.PAK.
234 R. M. ROSENBERG

We use the following notation;

q = (ql,. . . ,qn) is an n-dimensional physical displacement vector;

q = (g,, . . . ,Qn) is an dimensional physical velocity vector;
0
K = 1/2CmiQia is the kinetic energy of the physical system;
i=l
-0= - #(q) is the potential energy stored in the springs;
h =K 0 is the total energy of the physical system;
VU = grad U is an n-dimensional vector.
We assume that 0 satisfies the following conditions, called “admissibility
conditions” :
(i) U(q) = 0(- q) is of class P ;
(ii) - 0 is positive definite;
(iii) PO = 0 only at q = 0.

(iv) Trajectories are the vectors q(t) which satisfy the equations of
motion,
a
miqi = - U(q).
%i
(i = 1,. .. ,n)
where the mi are real, positive constants. The equations of motion transform
under
-
(A4 x; = vmiqi
into
.. = -
X;
a U ( X )= U,i
ax,
or, in vector notation,
(A.3) P = VU.
With Kuo [18], we transform (A.3) into the system of 2n first-order equations

(A.9 ji = fib)
where
Yi = xi, yi+u = 2i I
fi=Yn+i, f,+n= UIi.
 NONLINEAR VIBRATIONS 236

Since the system is admissible, the functions f j are of class C1, or the Jacobian
J is continuous, where

..

In consequence, (A.4) has a unique solution under specified initial conditions :

y = Y when t = t; that solution is
(A4 y =y(t,t,Y), - =< t< + =.
Conversely, if the initial condition Y = y at t = t is given, one has uniquely
(A.7) Y = Y(t,t,y), - oo<z< + m.
Finally, since J is of class C1. the functions y and Y in (A.6) and (A.7) are
once differentiable with respect to all arguments. If one regards (A.6) as
a mapping which, for fixed t , maps the initial point Y onto y , and (A.7)
as the inverse of this mapping, one notes that this mapping and its inverse
are continuous, or (A.4) is a topological mapping in the phase space. Clearly,
the energy integral of (A.4) exists and is
(A4 - U + 4 IlkllS = h = const
where / [ 11 denotes the Euclidean distance function, h 2 0 is the total
energy of the system, and it is bounded. The configuration of maximum
potential energy is
(A.9) U+h=O.
This is the maximum equipotential surface and it is the bounding surface
for all possible trajectories of admissible systems (Theorem V-11).
All trajectories satisfy
Lemma A-I: No integral curve of (A.4) termirsates at a point.
Lemma A-11: If an integral curve of (A.4) passes through ( X , 8 ) at t = 0
a d it reaches the bounding surface at t = t o , then it passes through ( X , - 2)
at t = 2t0 .
The proofs of these lemmas are simple and will only be indicated. To
prove the first, one supposes to the contrary that a trajectory does terminate
.
at a point x when t = 'cO, Then, for all t > t o ,x = E = . . = 0. But,
when x = 0, the configuration is that of maximum potential energy, and
when j ; = 0, x is a t the origin. Since the origin does not lie on the bounding
surface under our admissibility condition, Lemma A-I is true.
236 R. M. ROSENBERG

For proof of the second lemma one need only observe that trajectories
which attain the bounding surface intercept that surface orthogonally
(Theorem V-V). Hence, the initial conditions (position and slope) are
defined; then, in view of the uniqueness of (A.6), the trajectory must retrace
itself.
I t is now convenient to rewrite the energy integral in terms of the initial
energy when the initial point is not on the bounding surface:
(A.lO) - + tl141s= - U X ) + i l l q *
W)
where x and as functions of the initial values X and d are given in'(A.6)
because y and Y correspond, respectively, to ( x , i ) and (X,ft).
Differentia-
ting (A.lO) with respect to X and ft, we have:

-ax,
__ ... ax, _an,_...- an,,'
ax, ax, ax, ax,
ax,
- ax,
(A.ll)
ax,, axu
ax, ax, aa, an"
_-
ad, ax, a f t , aft,

We shall denote by A(t) the 2n x 2n matrix on the left-hand side of (A.lI), or

and we shall demonstrate the existence and regularity of A(t), because these
are essential for the proof of the central theorem. For instance, the transpose
A* may be found from

(A.13)

with the initial condition A*(O) = I where I is the identity matrix. Since
J is of class C1, the existence, uniqueness, and continuity of A(t) is assured.
But J is a function of the configuration x ; hence the solution of (A.13)
depends on the initial configuration X, and we write it as
 NONLINEAR VIBRATIONS 237

(A.14) A = A(t,X).

Let us consider A as a mapping in the phase space which, for fixed t and X
maps the point (- VU,$ onto the initial point (- Po U , x ) where voU
denotes grad U at X . Then, A has the important property, formulated in
Lemma A-111: For fixed t and X , the mapping A i s volume-preserving;
i.e., the determinant of A is unity. Conseq.uently, a unique inverse of A exists.
To prove this lemma, we make use of the Jacobi-Liouville formula

(A.16)
I
det IA*(t)I = exp tr J(s)ds
0

where J is defined in (A.6), and tr J denotes the trace of J . Now, the elements
in the main diagonal of J are all zero. Therefore, det A is unity and the
inverse A-l exists. Thus, we may write (A.ll) as

(A.16) [-xvu]= A-W [-8vou1’

and A-1 is a one-to-one mapping in the phase space which maps the point
(- r o U , f t ) , for fixed t and X , onto a point (- V U , i ) .
In the following, we shall ogly consider trajectories originating from an
initial point Po lying on the bounding surface; i.e., at x = X , 5 = 0. These
are called T-curves (see Section VII.2) ; they correspond to motions with
a rest-point. We shall now prove

Theorem A-11: Along every T-curve there is a countable infinity of

isolated points Q1 ,Qz,. . . , at which the configuration is one of stationary
potential energy with respect to points on T neighboring on Q1,Qs,. , . .
The proof of this theorem is lengthy, and we shall only sketch it. At
every Qi ,(i = 1,2,. . .) one has

(A.17)
au - 0
--
at

and the theorem holds if (A.17) has a countable infinity of isolated t i .

.
( j = 1,2,. .).
First, we note that (A.17) may be rewritten in the form

(A.18)
238 R. M. ROSENBERG

which shows that the gradient VU is orthogonal to the velocity vector 2

at all Qjor, at the Qi, the velocity vector lies in a tangent plane (actually
the intersection of the tangent surfaces) to an equipotential surface.
Now, T-curves are either tangent to equipotential surfaces at finite times
ti or, if they are not, they tend to an equipotential surface as t + 00. To prove
Theorem A-11, we need only show that the latter cannot arise. We suppose
that a trajectory (not necessarily a T-curve) lies in an equipotential surface.
Such a trajectory either is a closed orbit, or it tends asymptotically and
monotonically to a closed orbit. This latter is true because the transversals
(see VII-1) are, by definition, orthogonal to the equipotential surfaces;
thus, no trajectory, lying in an equipotential surface, can have curvature
components in a tangent plane (as above) to that surface.
Next, we observe that, if there exists a closed trajectory in an equi-
potential surface, it must lie in a two-dimensional plane which contains
the origin. This is true because such an orbit must be symmetric with
respect to the origin by virtue of Theorem V-111. Thus, if a T-curve tends
to a closed orbit which lies in an equipotential surface it will, after a finite
time, have a position and slope which differs from that of the closed orbit
by as little as we please. Since that T-curve satisfies the system M of (3.12).
the closed orbit must satisfy (3.12)within O(E). But it can be shown that
the only two-dimensional orbit lying in an equipotential surface which
satisfies (3.12)is a circle. That demonstration is based on the observation
that [ (211s= const for any trajectory in an equipotential surface. Finally,
we show that equipotential surfaces of admissible systems do not contain
circles in a plane containing the origin, because such equipotential surfaces
occur only in systems which are not elastically coupled throughout; i.e.,
which satisfy (2.4).Thus, no trajectory satisfying (3.12)can be asymptotic
to an equipotential surface; consequently, we have proved that aUlat
vanishes for bounded 4 . By the same arguments as above, no arc of finite
length of a trajectory can coincide with an equipotential surface because,
if it did, the arc would have to be that of a circle about the origin, and no
equipotential surface contains such arcs. This proves Theorem A-11.
Let us denote by hj the energy levels of the Qi,and let us write
(A.19) ti tj(X)
because it has been shown that the ti depend on the initial point X.
We wish to show that every hi is continuous and differentiable with
respect to the initial point X. The differential of U is
n
(A.20) dU = Utdt + 2 UxadXa
a=l

for variations in transit time through a given energy level, and for variations
in the initial point. The existence of this differential is assured by the prop-
 NONLINEAR VIBRATIONS 239

erties of U and the uniqueness of (A.6). Consider now one of the hj , say h, .
Its differential is the same as (A.20) except that U,= 0 by definition.
Hence,

(A.21)

Thus, the hi are continuous and differentiable with respect to the initial
point Po(X).
To prove the existence of a simple trajectory we extremize h j ( X ) subject
to the equation of constraint
(A.22) U(X) + h = 0.
In other words, we determine the stationary values of the potential energy
for a trajectory and we insure by the constraint equation (A.22) that this
trajectory is a T-curve. Thus, we define a function
(A.23) g ( X ) = hj(X) - ilU(X)
where il is a Lagrangian multiplier. Since we wish to deal here with the
phase space we may write, instead of (A.23)

with = 0. Then, the necessary conditions for the extremum are

(A.26)

and, in view of (A.21), we have at t = ti

(A.26)

Then, we may rewrite (A.26) as

(A.27)

where A ( t , X ) has been defined in (A.12). But, if we write (A.16) at t = 4

we have

(A.28)
240 R. M. ROSENBERG

or, combining the last two,

(A.29) A(t, ,X) ['"-"' 7

R = 0.

But, since det A = 1, the only possible solutions to (A.27)are

The first of these states that U,= 0 at the origh since VU = 0 at the origin
only, and the second, that Ut = 0 on the bounding surface because that
surface is defined by R = 0. Hence, a T-curve T* exists which connects
an initial point on the bounding surface with the origin. To prove Theorem
A-I, it remains to show that the T*-curve is a simple trajectory.
Let Po* be the initial point of T* on the bounding surface. Let its points
of first, second, etc. tangency with equipotential surfaces be Q1,Q2,. .. .
Now, suppose that Q1 is not a t the origin. Then, in view of the above
demonstration, Q1 is on the bounding surface. In that case, T* moves at
first toward configurations of lower potential energy and then toward
configurations of higher potential energy. Thus, there is a point R between
Po* and Q1 at which U,= 0. But this contradicts the fact that Q1 is the
first point along T* where Ut= 0. Hence, Q1 is a t the origin. This com-
pletes the proof.

ACKNOWLEDGMENT

This work was supported, in part, by a grant from the National Science
Foundation.

Retmcnces

1. MINORSKI,N.. "Nonlinear Oscillations," Van Nostrand, N. J.. 1962.

2. SYNGE,On the geometry of dynamics, Phil. Trans. A, 226, 31-106 (1926).
3. RAUSCHER. M., Steady oscillations of systems with nonlinear and unsymmetrical
elasticity, Journ. A@. Mech. 6, A169-Al77 (1938).
4. STOKER, J. J., "Nonlinear Vibrations," Interscience Publ., New York. 1950.
5. LEDOUX, P.,Personal communication, Lihge, 1964.
6. B~AWHIN.J., Oscillations en modes normaux de systhmes dynamiques nonlintbires
B plusieurs degrC de liberth, Bull. SOC. Royule Sciences Lilge, 9-10. 540-557
(1964).
7. ROSENBERG, H. M., On normal mode vibrations, PYOC.Cumb. Phil. Soc.. 80, 595-611
(1964).
 NONLINEAR VIBRATIONS 241

R. M. and Kuo, J .
8. ROSENBERG, K.,Nonsimilar normal mode vibrations of non-
linear systems having two degrees of freedom, Trans. ASME. Journ. Awl.
Mech., 81, E, 2, 283-290 (1964).
9. ROSENBERG, R. M. and Hsu. C. S., On the geometrization of normal vibrations
of nonlinear systems having many degrees of freedom, in Proc. IUTAM Symp.
on Nonl. Vibr.. Ukrainian Academy of Sciences, Kiev, 1. 380-415 (1963).
10. KAUDERER, H., Nichtlineare Mechanik, Springer, Gtjttingen, 1968.
11. ROSENBERG, R. M., The normal modes of nonlinear n-degree-of-freedom systems,
Trans. ASME. Journ. A w l . Mech., 29, E, 1, 7-14 (1962).
12. KINNEY,W. D., On the geometrization of the forced oscillations of nonlinear
systems having many degrees of freedom, Ph. D. Thesis, Univ. of Calif., Berkeley,
1905.
13. KINNEY.W. D. and ROSENBERG,
R. M., On steady-state forced vibrations in
strongly nonlinear systems having two degrees of freedom (Les vibrations
forcbs dans les systbmes nonlin6aires). Proc. CNRS I s t . Coll. 148, 351-373
(1965).
14. DARBOUX, G.. “Thbrie a n b r a l e des Surfaces,” vol. 2, pp. 438-509. Gauthier-Villar,
Paris, 1889.
16. J ., “Oscillatory Motions” (Translation of “Les Mouvements Vibratoires”),
HAAG,
vol. 1, pp. 36-39, Wadsworth Publ. Co., 1962.
16. ROSENBERG, R. M., La g6ometrie de la dynamique et les vibrations des systbmes
non-linbaires. L’Onde Ebctrique 48, 433, 425-434 (1963).
17. COURANT, R. and HILBERT,D., ,,Methoden der mathematischen Physik‘, vol. 1,
pp. 24&245. Springer, Berlin, 1931.
18. Kuo, J . K., On non-similar normal mode vibrations of nonlinear systems having
many degrees of freedom, Ph. D. Thesis, Univ. of Calif., Berkeley, 1964.
19. ROSENBERG, R. M., On the existence of normal mode vibrations of nonlinear systems
with two degrees of freedom, Quart. Appl. Math., 22, 3, 217-234 (1964).
20. ROSENBERG, R. M., The ateb(h)-functions and their properties, Quart. Appl. Math.
el, 6, 37-47 (1963).
21. ROSENBERG. R. M., “Les vibrations non-lineaires de systbmes B plusieurs degrbs
de libert6.” Centre de Rech. Phys., Marseille, 1984.
22. SIDERIAD~S, L. and ARGBMI,J.. Sur la singularit6 multiple des systhmes dyna-
miques plans, Comp. Rend., Acad. Scie. 268, 2037-2039, Gauthier-Villar, Paris
(1961).
23. ARGEMI,J. and SIDERIADkS, L., Sur la singularitd multiple dbfinie par les formes
homogbnes des systkmes dynamiques plans, Comp. Rend. 864, 4135-4136.
Paris (1962).
24. ROSENBERG, R. M., On motions with a rest point and the existence of normal
mode vibrations in nonlinear systems, in “Proc. Eleventh. Int. Congr. Appl.
Mech., Munich” 1964, (to appear).
25. MAHAREM, N., On homogeneous systems and the stability of their normal modes,
Ph. D. Thesis, Univ. of Calif., Berkeley, 1965.
26. CODDINGTON, E. A. and LEVINSON, N., “Theory of Ordinary Differential Equations,”
McGraw-Hill, New York, 1966.
27. HAUGHTON. K. E., Similar motion of multi-degree-of-freedom vibrating systems
with nonsymmetric springs, Ph. D. Thesis, Univ. of Calif., Berkeley, 1964.
28. ROSENBERG, R. M., On normal vibrations of a general class of nonlinear dual-mode
systems, Trans. ASME. Joum. Awl. Mech.. S8, E, 2, 275-283 (1961).
30. ROSBNBERG, R. M. and ATKINSON.C. P., On the natural modes and their stability
in nonlinear two-degree-of-freedom systems, Trans. ASME, Journ. Appl. Mech.,
B6, E, 3, 377-386 (1959).
242 R. M. ROSENBERG

E. T. and WATSON,G. N., “Modern Analysis,” Cambridge Univ. Press,

30. WHITTAKER,
London and New York, p. 624, 1927.
31. MIISHKES, A. D., On the exactness of approximate methods of analyzing small
nonlinear vibrations of one degree of freedom, in “Dynamic Problems” (coll.
papers) vol. 1, p. 139, publishers, Riga, 1963 (in Russian).
32. MCDUFF. J. N. and CURRERI,J. R., “Vibration Control,” McGraw-Hill, New York,
p. 72, 1968.
83. LANDAU, L. D. and LIFSHITZ,E. M., “Mechanics,” Pergamon Press and Addison-
Wesley, London, Reading, Mass., p. 21, 1900.
34. PBARSON, K., “Tables of Incomplete Beta Functions,” Cambridge Univ. Press.
London and New York, 1927.
36. ROSBNBERG, R. M.,On a geometrical method in nonlinear vibrations (Les vibrations
forcQs dans les systhmes non-lin6aires), Proc. CNRS In#. ColE.. 148, 69-79
(1906).
86. TIMOSHENKO, $3.. “Vibration Problems in Engineering,” Van Nostrand, Prince-
ton, N. J., 1956.
37. DENHARTOG, J. P., “Mechanical Vibrations,” Third ed., McGraw-Hill, New York,
1947.
38. KLOTTER,K., Steady-state vibrations in systems having arbitrary restoring and
arbitrary damping forces, in “Proc. Symp. Nonl. Circuit Anal. 234-267, Poly.
Inst. Brooklyn” (1963).
39. STRUBLE, R. A., “Nonlinear Differential Equations,” McGraw-Hill, New York.
1962.
40. Hsu. C. S., On a restricted class of coupled Hill’s equations and some applications,
Trans. A S M E , Journ. A+@. Mcch., 28, E. 4, 651-666 (1961).
41. LOUDE,W., Personal communication (1961).
42. MACLACHLAN. N. W., “Ordinary Non-Linear Differential Equations in Engineering
and Physical Sciences,” Clarendon Press, Oxford, 1960.
43. Hsu. C. S., On the parametric excitation of a dynamic system having multiple
degrees of freedom, Trans. A S M E , Joum. Appl. Mech., 80, E , 3, 307-372, (1963).
44. Hsu, C. S.. On the stability of periodic solutions of nonlinear dynamical systems
under forcing, Proc. CNRS Int. CoR., (Les vibrations forc6es dans les systhmes
non-lindaires). 148, 125-141 (1966).