Vibrations of Continuous Systems

INTERNATIONAL CENTRE FOR MECHANICAL SCIENCES
EBERHARD BROMMUNDT
TECHNICAL UNIVERSITY OF DARMSTADT
VIBRATIONS OF CONTINUOUS SYSTEMS

THEORY AND APPLICATIONS
COURSE HELD AT THE DEPARTMENT

FOR MECHANICS OF DEFORMABLE BODIES
SEPTEMBER - OCTOBER 1969
UDINE 1969
CO U R S E SAN D L E C T U RES - N. 1
ISBN 978-3-211-81305-8 ISBN 978-3-7091-2918-0 (eBook)
DOI 10.1007/978-3-7091-2918-0
Copyri~ht 1970 by Springer-Verlag Wien
Originally published by Springer Vienna in 1970
First Reprint.
PREFACE
This booklet contains the notes of my lec-

tupes on vibpations of (solid) continuous systems de-
liveped at CISM in Fall of 1969. The lectures were
presented to an auditory of engineers and physicists
interested in vapious bpanches of mechanics.
Stapting from vibpations of conservative,
linear syst.ems I tried to give an introduction to some
problems, methods of solution, and phenomena of non-
conservative and nonlineap systems. The examples cho-
sen to demonstpate the different notions and proce-
dures are very simple to avoid lengthy calculations
which might hide the basic ideas.
I would like to express my sincere thanks to
the authorities of CISM, in particular to professors
W.Olszak and L.Sobrero, for their kind invitation and
continued interest.
E.Brommundt
Udine, October 1969.

Introduction 5
o. Introduction
Objectives of vibrational investigations are tech-
nical (physical, chemical), biological, economic, etc. systems.
Purposes of such investigations:
to "comprehend" phenomena observed (experimentally) in ac-
tual systems;
to "predict" the behavior, qualitatively as well as quantitative-
ly, of systems not yet (experimentally) tested,and of systems
which are only projected as in engineering design.
The procedures of these investigations are al-
ways similar, see Fig. O. 1.
There is no way to compare mathematically the
results obtained for the model with the behavior of the real
system.
In these lectures we shall restrict ourselves to

(ct.p.8)
0'
Multiple interaction with

environment
Boundaries of the system only

vaguely known
simplification)
Isolated, idealized system Choice of the state
variable s to be
(model, physical model)
measured; connec-'
tion with instru-
Definition of state Well defined
variables{coor- interaction
dinates), applica- with environment; Interaction
tion of "basic" ® well defined bound- with environ-
laws, constitutive aries; list of the ment partial-
equations etc. characteristicsof the ly suppressed
I-t
various elements
constituting the model
a
..,
o
g-
<"I-
for ....oo
I;j
1:)
t "i
o
model (')
Mathematical (1)
(equations of motion) §"

"i
(1)
@ ~MeasL:l'ernents til
i rMathen"latiCS o
....,
® H(analytical, numerical, I-<
iletc. methods) ~
(1)
til
M-
.....
(JQ
?l
Mathematical results Experirnental res.Its M-
.....
solption s
o
::s
i
I COD:lparison of
L_ -1_ _- - - 1
the re St-'lts
/
~'"
satisfactory / '""" Junsa tisfactory
t
agreement "<l agreemen
'",
The mathematical model "hnprovernents "
""
serves for further calcu- at @) (l)) ®) 0, (f)
lations ..J
FIG. O. I
8 Introduction
mechanical system. We shall be concerned with the steps CD

and, mainly, 0 of Fig. O. 1.
Lit A. G. J. Macfarlane, Engineering systems analysis,
G. G. Harrap & Co, London 1964 (German trans!. BI-

Taschenbuch Nr. 81, Bibliogr. Institut, Mannheim
1967).
R. H. Cannon, jr., Dynamics of physical systems,

Mc Graw-Hill, New York 1967.
Coo rdinate s
9
I. Continuous systems
J. 1 Coordinates
1. 11 Reference coordinates
v = Volume
S = Surface
Fig. 1. 1
The particle P is denoted by the Lagrangian coor-

dinates
which may be interpreted as its{curvilinear}coordinates in a

certain reference configuration of the body. (Unique notation,
continuously differentiable ---. no cracks, etc.)
1. 12 Position (spatial, Eulerian) coordinates
The position of the particle P at the time t

is given by
10 Continuous Systems
(x(~.t),Y(K.t).z(~.t))=
( X1 (~ • t ) I XZ (~ • t ) I X3 (~ .l)) ..
~ (~. t)
The systems of coordinates
need not coincide.
I. 2 State of the body

1. 21 Classical continuum
The state is completely known far t ~ to if
! = ~(~. t) is known for all P E V and t ~ to
1.22 Modern extensions
1. 221 Thermodynamical state variables are taken in-

to account.
1.222 Microstructure of the material (grains, cryst-
als, complex molecules) are taken into account by associating
"directions" etc. with the particles ~ multipolar media.
From a formal point of view both extensions mean
that the number of the state variables xi (K.t) is increased,
i = 1, ••• , N j N> 3 ; (x (C. t) , ... , xN (~. t))=~ (~ . t)

1
State of the Body 11
Lit. : C. Truesdell, The elements of continuum mechanics,
Springer, Berlin 1966

A. C. Eringen, Mechanics of continua, J. Wiley, New
York 1967
W. Jaunzemis, Continuum mechanics, Macmillan, New
York 1967
1. 3 Problem
Find ! ". 1 (!, • f) and, maybe, some quantities
derived from X
N
e. g. I stres ses, strains etc.
12 Classical Linear Systems
2. Some classical, conservative, linear systems

2. I Longitudinal vibrations of a rod
2. 11 The mechanical model
--- X,X,u.
t = length
Fig. 2. 1
Rod originally straight, reference configuration as shown

in Fig. 2. 1
Problem: Investigate the longitudinal vibrations.
Simplifying as sumption : All originally plane cros s - sect ional
areas x = const. remain plane and perpendicular to the
x- axis.
Displacement lA. (~.t) = u(X,t) = X{X.t) -X.

Longitudinal Vibrations 13
2. 12 Equation of motion
2. 121 Deformation
Strain: e ::
2. 122 Constitutive equation

Hooke's law cr=Cl;
cr- stress, E .. E (X) - modulus of elasticity
2.123 Force
F=A·cr
A (X) cross-sectional area
2.124 Equilibrium
F(X.t) F(X+dX,t) ~p.u..=F'
X,X,lL () ()
=-
x
:--
at
p. :: (l A - mass density per unit length

e - mass density
2.125 Equation of motion

Elimination of F yields
uJ'
,u. i.i. - [A E =0
).J. , A, E > 0 , sufficiently smooth
2. 13 Boundary conditions
2. 131 Homogeneous boundary conditions
x= L
c
x~o
Bounda ry conditions for X= 0 End:
2.1311 F u.(O.t)::O fixed
2.1312 c:=:::J F (O.t):: 0 t-..U: (O.t)-=O free

2. 1313 ~ F(O,t):Ku.(O,t),k-spring constant fixed by a
~ AE u' ( 0, t) - Ku ( 0, t) = 0 spring
mu,(O,t)= F(O.t)' m-ma~s

2. 1314 attached to a
~ rnLi.(O.t) - AEu:(O.t)=O rigid mass
Transversal Vibrations 15
Similar conditions hold for X = z.
2. 132 Nonhomogeneous boundary condition
~ x=L
~~--------------~~ p
2. 1321 F (t,t) :; P (t) P (t) given force
c.... AE u.' (l,t) :: P (t)
2. 1322 ,,(t,t) = f (t), f(t} given displace-

ment
2. 2 further conservative. linear systems

2. 21 String
P - constant force
)J. (X) - mass per unit length
,uti - PU" = 0
2.212 Boundary conditions (homogeneous)
u, (O.t) = 0 fixed
U:(O/t) = 0 sliding
etc.
2. 22 Euler-Bernoulli beam
EI - bending stiffness
IJ. - mass per unit length
(may depend on X)
2.222 Boundary conditions (homogeneous)
2.2221
x=o u. ( 0 ,t) =0, u." ( 0 .t) :: 0 supported
~=
2.2222
~ clamped
Timo shenko Beam 17
2.2223 lL" ( 0, t) = 0) [E I LL~'/ = 0 free

x=o
2.2224 point rna ss
at X" 0
2.23 Timoshenko beam
lL1 - late ral di splacement

LLz - angular displacement
EI - bending diffne s s may
GA'f> - shear stiffness depend
fl1 - mass per unit length on X
)1-2 - rotatory inertia per unit length

U.2 fl1U1= -5'
M~~M
#2. Ul = M' - 5
EI U,2' =M
M - bending mon'lent
GAs {u" - u. n '" 5
5 - shear for ce ~l tili.ilT[GA~ (Ur U;)], ~ 0
j.tz til + GAs (u'2- u;)-(EIu;)'=O
Matrix notation
where state vector
,u. _(pi
"" - 0
0 )
}J.z
- inertia matrix
L ::: (-6A!o 0) at (-(GA,;,)' GAs) () (0

(GAs)')
'" 0 -£1 ()X~ + - GAs -(EI) , ()X + 0 GA":.
-linear differential operator (matrix)
2.232 Boundary conditions

}(:Q
2.2321 M(O.t):::O S(O,t)=O free
2.2322 fixed
2.2323 guided
M(O,t) = 0 mu.1(O.t)+S{O.t)::O point

I mass
2. 24 Plate, transverse vibrations

Plate Vibrations 19
h - thickness (uniform)
u. - deflection
jJ. - mass per unit area. J.l. .::. qh
K Eh 3 /12 (1 - 1)2) bending stiffness
:It
'U - Pois son I s ratio
Laplacian operator

For X =0 y. t arbitrary
()2.u.
U. :: 0 ()2.u. + '" - - = 0 supported
I () Xz () y2
F
X
u. : a Ou. 0 clamped
2.2422 ::
oX
1
2.2423
r X
02u.
() x a
+'U -
()2u.
() ya
=0
,
a'"
free
~
() x () Xli +(2-U) () y2 v·g.O
Lit. : W. FlUgge, Handbook of engineering mechanics,

McGraw-Hill, New York 1962, Sect. 61
2.25 Isotropic continuous body (three-dimensional)
.!;t = {IJ.1,IJ.2,U 3} displacement vector

Q - density
,,- Poisson's ratio
6 - shear modulus
cJ.iv LL = () u" + U t.l2 + () U3 =e - dilatation

... '0 X1 () Xt () X~
grad a/v lL = V2 U. + curl cu.rl!J. .

'" '"
Special motions
curLu. '" 0: only dilatation

'"
{no distortion}
Isotropic Continuous Body 21
divlJ,
... =0 only distortion
(no dilatation)
Zero stresses, zero displacements etc., d. Fllig-
ge, Handbook (cited above).

22 Wave Solutions
3. Wave solutions
3. I Longitudinal waves in an uniform rod (d. 2.1)
).J., A , E - ~::L stant
Equation of motion (cf. 2.125)
C 2 U" - il = 0 J C2 = I
A E p. (*')
3. 11 Travelling waves in an infinite rod
General solution of (*):
U= f (x-ct) + g(xt-c,t); f,g 0rbitraryfunctions

sufficiently srnooth
f and g represent travelling waves

U
x
Fig. 3. 1
The arrows indicate the direction of the propagation,
C - wave velocity.
No distortion (dispersion) of the waves.

Longitudinal Waves 23
3. III Initial value problem
Gi ven: LL (X 0)
I = cp (X) I U (X,D) :! " " (X) j
find f and 9
D'Alembert's solution:
+[cp D+ 2 e J'" (~) d ~

X+d
u ( XJ t) := (X - c t) + <p (X + c: t i
X.c.t
3.12 Waves in a finite rod
Equation of motion:
c 2 u"- i.i = 0
Chosen boundary conditions
U(Olt) ::: 0
lL'(Lt) = 0
fixed
free }
Initial conditions
u (X,D) = <p (X) } for 0 ~ X~ L

u (X,D) = 'I' (X)
24 Wave Solutions
Gene ral solution of (*), d. 3. 11 :
u = f (X - c t,) + 9 (X + ct) .
= derivative with resp. to

the argument
+
From (B) we obtain
;
f ( !;) = cp (~) - 21c J 1j1 ( ~) d ~
forO~1;~L
9 (~) t = 'P ( ~) + / ] " en d. ~

o
Continuation by means of the equations (A) :
9(L .. I;)=f(L-~)+g(L)-F(L) reflection at X= L,
f (-~) = - g (t;) reflection at X = o.

9, f
F -.-.
Fig. 3.2
L -, 21 ~
-4-- "
g~ .....
Waves in a Timoshenko Beam 25
3. 2 Waves in an infinite, uniform Timoshenko beam
d. 2. 23; P.1 , .Pz 1 [I 1 6 As - con stant.
Equations of motion (matrix notation, cf. sect. 2.231)
)Lu,-
IV
"
...,
(GA~
.
0
O\u,,,+/O 6As)u. +(O
El ) - ~GAs 0 ~
I O)u:O.(*)
0 GAs ".
No solutions with arbitrary f
3.21 Special waves
Assumption: !d- = ~ e Lx (X-ct) 1 i, = '{T. (* *)
(ll'dlf) in (*):
{- ~ x'c' + (G;. f~) lC',UA, 66,) L" + (g ~A') } 2 ~ 0

'-homogeneous system of equations for i = [a 1 , at j .
-)1.1 ,,2C 2 + GA~ ,,~ GAs Lx

0==.1 (x,d = clet { ..} =
-GAsLIt - flz ")(/c 2 + EI ·x} l' GA s
26 Wave Solutions
Wave length two waves,
dispersion: CJlll depend on )G •
GAs c!
ae-- 00:
2
Cx 00
)1.1
- shear wave
d. FlUgge
below
£1
C~I 00 =
pC!
- cC!B -bending weve
Euler - Bernoulli beam: )).2. - 0 Cg ~oo
(~ parabolic diff. equ. )
Lit. : W. FlUgge, Die Ausbreitung von Biegungswellen in Sta-
ben, ZAMM 22 (1942) 312-318
3. 3 Waves in a three - dimensional continuum
3.31 Infinite continuum
Equations of motion d. 2.251
curL u. : 0 2(1-U)G!(f-2lJ) e
waves of dilatation
plane wave - longitudinal, no dispersion
Waves in an Isotropic Continuum 27
div
-
LL = 0 :
waves of distortion
plane wave- transversal, no dispersion
3. 32 Bounded continuum
3. 321 Half space
Rayleigh surface waves; no dispersion
d. A. E. Love, A treatis'e on the mathematical theory of
elasticity, Dover Publicat., New York 1944;

Sect 64 (by E. E. Zajac) in FHigge I s Handbook.
,>, 322 Vibrations of a circular cylinder
Pochcharnrner, Chree waves;

rorsiona1, longitudinal and transversal vibrations j
d. Love and Zajac cited above.

28 Other Fo rms of the Equation of Motion
4. Other forms of the equations of motion for linear, finite

liystems
4. I Vafiational problem
Example: Euler-Bernoulli-bearn
Notation see sect. 2. 22
Kineti c energy:
Fig. 4. I
EltLt\ = +f ,au
a
l
l dX
TJEI
L
Potential energy: E pot = (u:·) 2 d. X
o
Hamilton's principle
ta.
tf
J( E/(Ln
tz.
- Epot) d t = ext rernum,) or
OJ(E kLn Epot) clt = a

t1
L '" EKll'\
" - Epot - Lagrangian function
4. 2 Lagrangian equations
Lagrangian Equations 29
{Euler-Bernoulli beam}
~ 1.
Kinetic energy density e k~n. = T)J. LL 2
d. 4. 1
Potential energy density 1:. pot = i £1 (u.")'
Lagrangian density (A)
Hamilton 1 S principle:
of f 1.. dX cit =o.

tz l
t1 0 8 't
From (A) 6 't. = 6 u. l) Ii. + 61.. (j u."
(5 u."
Putting (B) into (*) and integrating by parts we
J f {~:.
L tz. L L
Ia
obtain tl.
~: lludX + I iJLL / - (d~ ~:,) iJLL

,a v
t 1.I t1 \. 0
y
J
<D ®
+J [if.
L
d. ';(2
'O:t -
() u." ~ VlJ dX} dt =0
cit () u.
5u.
a \. y" J
@
(D vanishes because ou/:Ou./.O -
(chosen)
tz t1
® Natural (dynamical) boundary conditions:
30 Other Forms of the Equation of Motion
1.
OU /
I
- ( dXA~)8u/L=O
vU: 0
o
( 0 u. and" u,1 have to satisfy the geometrical boundary condi-
tions pres cribed for u. and u. 1
o
).
Lagrange equation (= Eulerian equation of the variational
principle):
d dZ () ';I.. = a c:.... equation of motion

dt aU. dXz () u"
(Ouarbitraxy). Partial derivatives mean only differentiation
with respect to explicit dependence, total derivatives nlean
differentiation with respect to implicit dependence too.

Lit. : H. Goldstein, Classical mechanics, Addison-Wesley,
Cambridge, Mas s., 1951
4. 3 Integral equation
Example: String
~
P - constant
p
F = stati c load
F
Fig. 4.2
Integral Equation 31
(or 0 ~ X~ ~ }
u. = I«X} ~)F
for ~E X ~ t f
Green's function
x (Influence function)
Symmetry:
K K(X.~): K(~JX)
(Maxwell's reciprocity)
K is symmetric if the
problem is self-adjoint;
d. sect. 5.4 and Col-
latz's book cited in sec-
tion 5. 5
Fig. 4.3
Distributed load: F (~) c:..- f (~) d. 5

(IU) is linear e-.. superposition
1.
~ u(x)= JK(X,~)f(~)d.~ (***)
o
Fig. 4. 4
Vibration e.-- dynamic load: f (~);;:: - i.i(~.t)p. (~)
Jf( (X, ~),u(~) U. (~.t)

L
u.(X,t) = - d. ~.
o
32 Eigenvalue Problems
5. Separation of variables; boundary value problem;

eigenvalue problem
Example: Longitudinal vibrations of a rod
Equation of motion, d. 2.125,
Boundary conditions u, (O,t),. 0 i U' (L,t) .. o. (•• )

~x.o xat
Fig. 5. 1
5. I Separation of the variables
u. (X,t) = U(X) T(t)

substituted in (*) and variables separated
A. - pa rameter of sepa ration.

We obtain two ordinary, second orde r diffe rential equation s
..
T+)"T::O (A)
and
[A [UJ' +).,,)1. == 0 (B)
From boundary conditions (**) : U(O):. OJ U'(Z} = o. (C)

Boundary Value Problems 33
5. 2 Boundary value problem - eigenvalue problem
and Uil (X,A-) be two linearly
independent solutions of (B). General solution of (B)
should satisfy the boundary conditions(C) :
Cz Uz (O,,,,) + ell Uji (O,A,):: a } homogeneous system of

(D) equations for C r , C n
C1U;(l,A.) + cEuir (t)..) = 0
A(Ao]:= det {Ur(O,A.)

U;(L)..)
Characteristic equation :
a ()..) =0 ; roots A.1, 1..2.)'"
'--- eigenvalues,
characteristic values.
For A.: A. 1'\ ' - - CI = (C I) n ' C11 :II ( CE)n c:.......,.. s 01 uti on U =Un'
( eigenfunction)
(B) and (C) are satisfied by A.n I Un :
[AEUJ'+ A.np Un = 0, (E)
Un(O):::O, U~(L)=O.
((B) and (C) constitute a Sturm-Liouville eigenvalue pr.oblem).
In general
fi(A.) =0 - trascendental equation ~infiLitely many roots
A.n., 0.=1,2, ... , -denume rable

- integral function of ),. c:...-.- no finite cluster
points.
Lit. : R. Courant, D. Hilbert, Methods of mathematical phy-

sics, Vol. 1, Interscience Publ., N. Y. 1953
E. A. Coddington, N. Levinson, Theory of ordinary
differential equations, Mc Graw-Hill, New York 1955
M. A. Neumark, Lineare Differentialoperatoren, Aka-
demie-Verlag, Berlin 1963

Eigenfunction 35
5. 3 Example: Uniform rod
A , £ }P. - constant
c 2 U" + /I. U = a (B ')
General solution of (B ')
u = CI sin. w X + C!! (as w X

c c
Boundary conditions (C)
cos ~L + -w
w L =0
W
c
C1 c c.
ell sin
c
Characteristic equation
Ll (w) = W
C05 W L '" 0
C C
Roots: w '" 0 and wn::: Tc Jt 1 ) ,n.:: 1 ,2 ,

( n,- -2
'---y----.J
A trivial sd. (negative n
'---" no new
solutions)
(e ii ) n. ::: 0 , (Cr)n = en. arbLtrory

36 Eigenvalue Problem
Solution of the eigenvalue problem

Un (X) =- en s~n. wn X
~
eigenfunction I
normal mode
Corresponding Tn from (A), p. 32,
Special solutions of (*), (*.), d. 5.1,
t.(.n (X,t) ::: (An. SLn wnt + Bn cos runt) sin CcJn
C
X
All = an en , Bn::: bn en - arbi.trary j n=1,2 ...•
( An , Bn. have to be calculated from the initial conditions, see

sect. 5. 5).
x
-1
Fig. 5. 2
Adjoint Eigenvalue Problem 37
5. 4 Adjoint and self-adjoint eigenvalue problem
5.41 Inner product orthogonality, norm
Inner product;
L
(u, V) :: JUV dX , v-complex conjug. of V j
o
sometimes:
l
Jw (X) u V dX ; w-weight function j
a
for vectors
(y,y)= UT - transposed vector.
(v, V) is the projection of V onto V, or vice versa. Two func-

tions V, V a re called orthogonal if
(V, V) = o.
Norm:
/I U /I = (u) u) i/z ~ 0 ,
V, V sufficiently smooth.
38 Eigenvalue Problem
Lit. N. I. Achieser, I. M. Glasmann, Theorie der liearen 0-

peratoren im Hilbert-Rau m, Akademie- Verlag, Berlin
1958.
Any text on functional analysis
5.42 Adjoint operator, adjoint boundary values, adjoint and

self-adjoint eigenvalue problem
Equation (B) from 5. 1 :
Operator form
LU = "A.p. U.
L - linear differential operator.
I
L
f [A f u'J' Ii d X.
( L U , V) :: -
o
Integration by parts yields
l
( _/L _,/L
) =-AEU'V+AEUV
LU,V
o 0
-Jo [A Eii']'dX
U
~------~vr--------~J '------,v,....----~
bOlJndary term~ (b. t.)

Adjoint Eigenvalue Problem 39
L+ is the adjoint operator of L.
Here L = L+ c..-., L is a self-adjoint operator
Rounda ry terms:
b. t. =
\
-AEU'(t)V(t) - AEU(O)
v
v'a /
+
vanish because of (C), p. 32
+AE-U'(O)V(O) + AEU(L) V'(L)

\ ~ ~
vanish for arbitrary U'(O) and U( t)
V(0):: 0'- V (0)= O} boundary conditions

( C+) { adjoint to the boundary
V'(L);: O,--V'(t);O conditions (C) on page3Z.
Here:"adjoint boundary conditions" = "original boundary

conditions" ____ self-adjoint boundary conditions.
Adjoint eigenvalue problem:

+
Adjoint differential equation L +U = A.jJ. U
{
adjoint bounda ry conditions (C ), see above.
Self-adjoint eigenvalue problem
+
{ L = L,
(C +) = (C),
thus, see p. 38,

(LU, V) = (U, LV).
5.43 Eigenvalues of self-adjoint eigenvalue problems are real
L et U and A be an eigenfunction and an eigenvalue, resp.,

1
( LU, U) - (U , L U) (A-i) 1).£ u[j d ><
.
II
\ I '----v--' ,0
'V
~
0 •0 r "'--....."
>Obecause).L>O for all X
(positive definite)
Since L is real ~ U may be assumed to be real.
5.44 Eigenfunctions belonging to different eigenvalues of self-

adioint eigenvalue problems are orthogonal
U ,U - eigenfunctions
n m
A. ,P Am. - corresponding eigenvalues, A.1l +- A. m
L
(LU n , Um ) - (Un ,lUm) = (A,,-A m) !JLUnU m cJ.X
~~_ _ _ _""'y~_ _ _ _~1 ~ l~O~_~y~__~1
=0 +0 ;:01
U J U are orthogonal with respect to the weight function
n m l
...u(><) ; in general fUn Um dX =p O.
o
Initial Value Problem 41
5.45 Normalized eigenfunctions
Frequently it is convenient to normalize the eigen-
functions in such a way that

L
f,u U~ ciX: 1
o
:hen _{01Lfn.m.
Lf n*rn
i~'or the example 5. 3 we obtain because of )l;; 1:
CA.> 1\ X.
(,
5. 5 Initial value problem
The rod shown in fig. 5. 1 has initially. at t~ 0

the displacement u( X,O) .. cp(X) } d. 3. 111
the velocity u(X ,0) = 11' (X)
(*), (**), d. p. 32 , has a solution of the form

cf. p. J6 j let the Un (X) be normalized, cf. above.
Substituting (A) into the initial conditions we obtain
(B)
(B) represents two infinite systems of linear equations for

An and BTl .11.=1,2, .•.•
Generalized Fourier coefficients
To solve (B) we multiply the equations by )J.(X) UIl1 (X)

and integrate over O::!S X ~ L . Bec", use of
L
1)1. Un Um ciX = llnm ,
we obtain 0
f
t
8m ~ fj> (X))l (X) Um (X) cLX
o
(c)
and L
f lJ1(X)p(X)Um (X)dX.
a
Am CUm and 8m • given by (C). represent the generalized
Fourier-coefficients of "" (X) and cP (X) • respectively. with

re~T)prt to the system of functions Un (X) .
Initial Value Problem 43
The series (A),with the coefficients Am Bm,con-

I
verges and satisfies the initial conditions if cp(X)and 'P (X) arlo
sufficiently smooth and if the system Un (X) is complete
("expansion theorem").
General investigations
Courant, Hilbert, Vol. 1

cf. p.34
Coddington, Levinson,
Neumark
Achieser, Glasmann, d. p.38
L. Collatz, Eigenwertemfgaben mit technischen
Anwendungen, Akadernische Verlagsgesellschaft,
Leipzig 1949.
44 Forced Vibrations
6. Forced vibrations
Example: Longitudinal vibrations of a rod.
6. I Excitation at one end (of a rod)
Fig. 6. 1
Equation of motion!
B ounda ry C ondi tion s :
u. (O,t) ::: 0
AEu:(t,t)- P(t) o - nOll homogeneous

} (*-~)
'
Procedure: Supe rposition
where up is il partie',ILlr solntion which s,ttisfies (:t), (**),
and un is the solution of the corresponding system. with
homogeneous boundary conditions, d. (*), (**) on p. 32.
In general it is impossible to find in closed
forrn.
T ri cl~
Discrete Excitation 4-5
tL 1 satisfies the (nonhomogeneous) boundary conditions (**)

but not the differential equation (*),
lLK satisfies the differential equation (*), disturbed by 'U z
(see below), and the homogeneous boundary conditions
tljf (0, t)
u.'! ( L, t)
= OJ
= O.
} ... ( )
We choose
Ui .. P(t)X Et = E (t) i At ~ A(t)

- £t At
(**) is satisfied. Putting
u.p = -p- X ~ ttj (A)

fL Al
yields
..
[ AElLjj'J' =-p.--X-tAE-
-
p ()' P
EtA t
[tAL
(B)
This is a nonhom.ogeneous differential equation of the form

).t ii - [A E u.] I:: p ( X It)
which corresponds to a system with distributed excitation, cf.
section 6.2.
6. 2 Distributed excitation
p(x,t) X- L
Fig. 6.2
Equation of motion;
~i.i. - [AELL'J' = p (X,t),

Homogeneous boundary conditions:
u.(O,t) = 0
} (HI
u' ( L , t) = 0
ll.21 Normal coordinates
A solution of (*), C.*) is sought in the form
(A)
Un - eigenfunctions of the homogeneous problem conoes-
ponding to (*), (**), d. 5.2.

Distributed Excitation 47
d 1'\ - normal coordinates - to be sought
Putting (A) into (*) we obtain
Application of the relation (E) from p. 34:,
yields
Fourier expansion, d. 5.45, 5. 5, leads to
an. + w~an. :: bn. , rl.= 1,2, ... ) (B)
where
L
bn.(t):: Jp (X ,t) Un(X) dX. (C)
o
(B) is an decoupled system of infinitely many ordinary non-
homogeneous diffe rential equations with constant coefficients.
Solutions of (B) :
(D)
t
+ ..L
Wn
f 5~n. 6)n (t _.t") bnC t') d't
~--------~v---------~ ,D v-----------~
d np
An. , Bn. from initial conditions, d. sect. S. S'.
6. 22 Resonance
b n (t) periodic, d. (C); for instance

bn. "fon. si.n nt.
From (D)
a np = 5Ln. !l.t.
Large amplitude - resonance- for .0. .... Wn if

fin:F o. Dange rous region: IW n -.0.1 small.
Technically important in rotating and reciprocating
machines.
Calculation of Eigenvalues
7. Methods to calculate eigenvalues of conservative systems
Separation of variables, d. sect. 5. 1, transforms

I-he equations of motion of section 2 into
LU = A.)J.U,
where land ).J. may be matrices and U may be a vec-
tor, e. g., in sect. 2.23.
The -homogeneous - boundary conditions have the

form
1 U= 0
7. t Analytical method
An anal yti cal solution of the diffe rential equation
(*) and a subsequent analytical or numerical solution of the

cha racteristic equation are only possible for some very
special systems. Realistic models have to be investigated by
approximate methods.
7. 2 Raylelgh·s quotient
7. Z 1 Formal introduction
50 Calculation of Eigenvalues
Multiplication of (*) by U and integration over X
yield
J(
L t
L U) U d,X = A. f Up. UdX
o o
This is a scalar equation;

L
f U)J. U eX X > 0 Lf U¢ 0 and)J. > 0
o
(positive definite).
We obtain Rayleigh I s quotient

L
o!(LU)UcLX
A = ~--~------ (A)
o
f U)J. U dX
~
1£ we substitute the eigenfunction U 1\ for U, we obtain by(A)

the eigenvalue A. = A. 1\. If we choose any function U (X) which
satisfies the boundary conditions (**),we get from (A) an
approximate value, A. ~ , for the fi rst eigenvalue, A1

A. ~ is always greater than or equal to ).., , A.~ ~ A."
Tf we put a function U into (A),which satisfies the boundary
conditions and is orthogonal to the first (n-1) eigenfunctions,

l
jU"p.UdX =0 for k= 1) ... , n-1,
o
Rayleigh's Quotient 51
we obtain from (A) an approximation, , for the
eigenvalue ~n j
..
A. 1\ ~ A. n.
Lit. : Colla.tz, Eigenwertaufgaben, d. p.43.
J. P. Den Hartog, Mechanical vibrations, 4. ed.,
McGraw-Hill, New York 1956

R. ZurmUhl, Praktische Mathematik, Springer,
Berlin 1965
7.22 Introduction of Rayleigh's quotient by energy considera-
tions
Conservative system
E Kin + Epot :: E tot ::: constant;

d. 4. 1.
Assuming an "in phase" sinusoidal motion of the whole system
we have
Epot /m~x = E tot at ma.ximal deflection,

(zero velocity)
at maximal velocity,
(zero deflection),
or E pot /max = EKin/max .

Example. Euler-Bernoulli beam, d. 2.22
1
1
!U,.uUdX,
2 o
l
.,
EKin 1m." :a T ~ JUp U dX ,
W2 o
., .,
fo V"
L
::11-
2 .. - .5£n' c...>t £1 U· dX
2
l
fo U" £I U" olX
Equating these expressions for ElCin/max and E.pot-jmu,
cf. above, we obtain
").., = fy"_EI_U"_dX
__
---.:o~
(B)
l
jUp.U dX
o
The denominator of this quotient has the fo rm of the denomina-
tor in se ction 7. 2l. The numerator diffe rs from the nume ra-
Rayleigh-Ritz Method 53
tur in 7.21 only by some partial integrations (if the boundary
conditions are self-adjoint). So there are only formal differ-

ences between (A) and (B).
In section 5.43 we showed that self-adjoint eigen-

value problems have real eigenvalues. Looking at the numera-
tor and the denominator of (B), we see, because of their
mechanical rre aning, that hoth of them must be positive
definite (they are ~ 0 for any choice of real functions U );
thus, A. must be positive (w real, w 2 =).. ); d. sect. 8.
7.23 Rayleigh-Ritz method
Because of A. ~ ~ A 1 for all admis sible func-

tions U ,\,ve may conclude
)"1= m~n
(C).
all dd.mi~!> U
Ritz: Introduce a function which de-
pends on some parameters a" ... , a 11 and satisfies

the boundary conditions for arbitrary am 1m':. 1 ) .. , M. By
(A) or (B) we obtain A~ = }...~ (o.1}''') ttM) ~ Ai .
The best approximation for A. of we find from
This is an ordinary minimum problem. Further investigations

and simplifications see in Zurmuhl, Praktische Mathematik,
d. p. S1.
Error estimates H. Schellhaas, TIn Verfahren znr Berech-
nung von Eigenwertschranken niit Anwendun-
gen auf das Beulen von Rechteckplatten, Ing:-

Arch. 37{ 1968) 243-250
S. G. Michlin, Variationsmethoden der ma-
thematischen Physik, Akademie Verlag,
Berlin 1962
7. 3 Iteration process
Procedure: Choose ze ro approximation U (0) (X)

and calculate Uc·f)(X) from
L U (m+1) =)-l U (m) 1 m = OJ 1 J •••
(D)
Iteration Px-ocess 55
d. (*) and (**) p. 49.

If Green's function K (X.~) is known, cf. sect.
4. 3, the procedure (D) can be replaced by
Z
u(m+,n(X) =JK(Xl~)).J.(~) u(rn)(l;) d~ J rn=O,1, ....
o
which is deduced from the integral equation corresponding to
The procedure conve rge s to the lowe st eigenfunction, U 1 •
The respective eigenvalue can be expressed by Rayleigh's
quotient.
Higher order eigenfunctions, Un In> 1 , and
eigenvalue s, )... n , can be calculated if U (0) and, subse-
quently, because of the unavoidable inaccuracies in the calcu-

lations 1 the U (rn) are orthogonalized with respect to the low-
e r eigenfunction s.
A semigraphical algorithm was introduced by
Stodala, see Den Hartog, p. 51. Numerical procedures are
given in ZurmUhl' s book, d p. 51.
Variations of this procedure are possible{e. g. ,
Grammel's procedure, d. ZurmUhl).

7. 4 Variational methods
7.41 Ritz's method
In section 4. I we replaced the equations of motion -
for conservative systems - by Hamilton's principle,
tz
f( £"~,, - Epot) d. t = extremum,
t1
a variational principle.
Choosing the example "Euler-Bernoulli beam" from
section 7.22 we have
t. f1- L{,~Lllt -
tz L,.
£I (UU)z} dX dt= extremum.
Assuming u. = U (X) !lin w t, t1 = 0 t z " 2 1£/ w
I
we obtain o){ ')..)J. U2 - E1 (UI~Z.} d X = extremum, A" (a)J.
We solve this problem by Ritz's direct method:
Ii
U (X) = E am U!{Y)} (E)
",·1
the U,: have to satisfy only the geometric boundary conditions
and are linearly independent.
From ~ 1.o{..u A. U2 - E1 (UII)JdX = 0
r
M
and 8 U " 8 am U,: loam - arbifrary 7
m·1
Ritz's Method 57
we obtain a set of linear homogeneous equations for ct m
a,
A11 Aiz '"
AZi
A1H
/ BI1 Biz •.• Sllo!
BZ1 az
=0, (F)
A.
AHi AI1H
\ BH1 BI1H aM I
,I
l
A~k :: jftUtU/(-dX AKr ,
o
z
SjX = JE1 Uj' U; dX = BK~ .
o
For nontrivial solutions (am '* 0 } m 1} ... ) M) of
'·quation (F),
4\" (A) = cl e t { ...} :: 0 (G)
must vanish. (G) is an algebric approximation of the charac-
teristic equation. Its roots, )..~, ... } "'*M are approx-
imations for the eigenvalues Ai",,}).. M

(The approx
inlations for the lower eigenvalues are better than those for
the higher ones). The solution vectors ( : : ) of (F). intro-
duced into (E). yield approximate eigenfunctions.
The same set of equations, (F), (G), can be obtained
by the Rayleigh-Ritz method, d. 7.23, if (E) is introduced
into (C). p . .53 ; d. ZurmUhl, Praktische Mathematik.
7. 42 Galerkin I s method
Galerkin IS method starts from the differential equa-
tion (*). p. 49 . in the form
LU - J...}L U = o. (H).
If U. A is a solution of (H) which satisfies the bonl1dary
conditions (**),then the projection, d. 5.41, of (H) onto an
arbitrary function V (X) must vanish:
(LU - A.)1.U, V) = 0 (I)

(For the Euler-Bernoulli beam: LU = E I U IV)
ff we introduce an approximation of the forn1 (E),
U(X)=EamU!(X) (E ')
m=1
where the u*
m have to satisfy all boundary conditions(**),
and if we choose v = U,: (X) , m = 1 I ... ) M ,we obtain
Galerkin's Method 59
the equations (F) ,p. 57, again (with the same coefficit:!llts
Ai#( I Bil( if the U: (X) in (E) and (E') are the same). Thus,
the equations (F) are sometimes called Ritz-Galerkin equa-
tions. Subsequent calculations d. 7.41.
o
'"
7.43 Differences between Ritz's and Galerkin's method
R i t z Galerkin I
----.
~ifferential equation
~ a sis Variational principle principle of Vl rtual work)
I
I
iApproxim ate solution the geometrical boundary all boundary conditions
Ff. (E), (E I), has to satisfy conditions (can be weakened)
,
cfU (Ritz) must be derived from the can be chosen arbitrarily

(')
approximate solution (have to satisfy only the III
V (Gale rkin) .......
geometrical boundary g
.......
conditions if the principle III
<'+
.....
of virtual work is used) o
~
o
......
A. ppli ca-",le to problems which can be to problems which are CD
.....
C1Q
expres sed as variational governed by differential CD
~
problems equations or the principle <
III
of virtual work ~
CD
------ ~---- CIl
Transfer Matrices 61
Lit. : L. W. Kantorovits ch, W. 1. Krylow,

Ntlherungsmethoden der hBheren Analysis, VEB
Deutscher Verlag der Wissenschaften,
Berlin 1956
7. 5 Transfer matrices (Example: Euler - Bernoulli beam)
7.51 Basic idea

Example Euler-Bernoulli beam
sections "fields"
1 2 N-1 N / '
:c
~ ~~
[ Jl-----L..---L_ I ]
o 2. N-1 N
Y. N-1 !tN
state vectors
Fig. 7. I
Linear relation
operator
J. 52 Basic eguations x
Fig. 7.2
U,i lateral displacement u~ Ll.z
u,2. - angular di splacement

U2 - M U.3
IT IT
U.3 = !,,1 - bending monlcnt
u'3 s = U"
u" .. S - shear force
..
£I - bending stiffne s s u.'4 )J, U1
p. - mass per unit length Assumption

LL}=: U·i ~in wt
Ui' ::. Uz
1 U
U~ = 3
[I
U~ UL,
U~ :; - A).L Vi
Matrix notation t.. ~ w2..
u'= ~ (A.X)!d,
where
st~te vector
a 1 0 0
A ::: a o -1/EI 0
a a 0 1
-A)J.. a 0 a
Field n. with EI I J.1- - constant, initial point X = 0 , te rm

inal point X = Ln
or Fn - field matt'
'"
(K)
Successive applications of (K) yield
7. 53, Boundary conditions

Let us choose the following boundary conditions
(d. Fig. 7.3)
Fit!. 7.3
7.54 Characteristic equation

Introducing (L) into the boundary conditions (M)
we obtain
I t
§o
Yo = 0 (N)
§ N f N f N-1 • • • £2 E1
V" .
hypermatnx M
r
t1 ( A.) Yo:: O. (N 1)
(N) is a lineal' hOlnogcneous systelu of cquiltions for Yo

We obtain the c:laracteristic equation
L\ ( A.) '" de t M ( A.) = a (p)

'"
Equation (p) C<1I1 be solved numcricillly on a digital conlputer.
A value A* is chosen, the field 111atricl's f" (A~) Zll'e calcu
~ated, d. ('qu. (K). Finally Ll ( A.. *) is calCt~lzIted by equa -
tions (N), (p) and plotted, cf. Fig. 7. 4.Thc zeroes of L1(A.),AkJ
are eigenvalues of the systenl. There Zlre rnany luodifica -
tions of this procedure to gain numerical advantages and to
avoid numerical difficulties (differences of almost equal l1U111-

~ (A)
Fig. 7.4
General considerations and extensions:

E. C. Pe.stel,. F. A. Leckie, Matrix tnethods in elastotnechan .
les, McGraw-Hill, New York 1963

W. C. Hurty, M. F. Rubinstein, Dynatnics of structures,
Prentice - Hall, Englewood Cliffs, N. J., 1964

~utnerical difficulties: K. Marguerre, R. Uhrig,
ZAMM 44 (1964), pp. 1-21 and 349-360
66 Relations between Stability and Vibrations
8. Relations between stability and vibrations of conservative

systems (Example: Compressed bar)
Example: Vibrations of a compressed bar hinged at
its ends, d. Fig. 8. 1
8. I Equations of motion; boundary conditions
Fig. 8. 1
From Fig. 8.2

M(X,t):z Pu.(X,t)1-V(l.-X)
JL
+ P i.i ( ; - X) d. ~.
X
Fig. 8.2 Moment curvature relation:

M = - £I u"
Elimination of M yields the equation of motion:
noundary conditions:
u(O,t):o u(t.t):O 1
M (0, t) = M ( L. t) ,,0 c... u." ( 0, t) = u." ( L. t): 0 J (. *)
Discussion of an Example 67
Trivial solution of (*), (**): u. (X, t) =. 0 . (***)

8. 2 Eigenfunctions and natural frequencies
Separation of the variables, u. (X. t) .. U(X)e 'it,

leads to the eigenvalue problem
EIU 'V + PU·1 - )...J.LU= 0, (')..=-t' ,.w2., w- LY) I (C)
u (0) = u(t):: a ,
u"(o):2 U"(t) = 0 I
(self-adjoint eigenvalue problem).
} (D)
Solutions of (C), (D) :

·
E Igen f '
unctlons: u
n · f\ 1C X
-L- (not normalized)
[1- £Ipp
SLn
=:
·
E Igenva 1ues : 1\.1\
" I
<">1\ • - YI\a = -1 EIn'1r 4 ]
=:
).L Lit nZj{2.
8. 3 Discussion of the results
Distribution of the eigenvalues Yn (= - i W n ) see Fig. 8. 3/4.

The eigenvalues ITlove as indicated by the arrONH
if P P / EI is increased.
For PL 2 / E1 > 1 ¥1 ~ real, positive,the
first normal mode, U1 (X) ,increases exponentially c......, the
trivial solution becomes unstable.

68 Relations between Stability and Vibrations
Imy/\. 1m Yn
'(J
paO ~ P>O
'/1
'11 l.
-'11
-'/a
Reyn. "I RlIYn.
-t. t
Fig. 8. 3 Fig. 8.4
Borde r line between stable and unstable
trivial solutions: y.... 0 ; the eigenvalue problem reduces to:
EI U IV + P U" =0 (C')
U (0) = u (t) '" 0
UI/(O) = U'(L)= 0 } (D')
P - parameter {eigenvalue}.
Solutions Un / Pn of (C'), {D'} represent

nontrivial static deflections of the rod.
Static stability criterion: The trivial solu-
tion (state of equilibrium) of the system becomes unstable for
, where P (riot is the lowest value of

the parameter (load) for which exists a nontrivial static de-
flection in the vi cinity of the trivial solution ( P cnt is the
lowest eigenvalue of (C'), (D')).
The Principle of Virtual Work 69
9. Equations of motion for (nonlinear) nonconservative

systems
9. I Differential equations; boundary conditions
Differential equations and boundary conditions can
be established as in section 2. 1. Nonlinear terms arise from
geometrical nonlinearities and from nonlinear constitutional
equations.
9. 2 Principle of virtual work
(A)
- virtual work of the intertia forces (D'Alernbert's
forces)
- virtual work of the "working" forces ("applied"
forces in Goldstein's book cited in sect. 4.2).
The (infinitesimally small) virtual displacements
have to satisfy the geometrical constraints (boundary condi-
tions ).
70 Equations of Motion
Example: Extensible (linear) viscoelastic string (plane

vibrations)
Fig. 9. 1
L - length of the unextended string
(B)
(C)
A - cross-sectional area
p, - mass per unit length
E 1 , E2 - constants
Virtual work of the intertia forces

Z
OW r =f )J. (ii,1 0(,(,1 + i.i 2 Ou,2.) d X (D)
o
Virtual work of the working forces
1
oWw=-J F of dX J (E)
o
where F is taken from equation (C), and
ol = (1 + u,;) J u,'; + U,2 01.(.2 , d. equation (B); (F)

VC1 + u,;)'I. + (lLz) 2
The Principle of Virtual Work 71
=1 (ou') i. = 1,2 - partial derivatives

()X "
of the virtual displacements
The virtual displacements = are
arbitrary functions which need not depend on the time, t )
especially,the dU..j, are not related to the actual displacements,
U,.: ( X, t,) ; but both the actual displacements, ,u4, ,and
the virtual displacements, OlA.i ' are subjected to the geome-

trical boundary conditions (d. Fig. 9. 1) :
U1 (0, t) =
Instead of the double symbol fill. =Ju.(X) we

could write, for instance, V (X) . But the J in front of
the U , etc. indicates and reminds that the virtual quantities
have the characte r of diHerentials; the relations between the
virtual quantities are relations between diHerenfials, d. the
relation (F) between SI.) OUi) OU2 above.

The principle of vi rtual work, (A), can be used as
basic equation for Galerkin's method, d. sect. 7. 42. The

equation (A) is always linear with respect to the virtual dis-
placements and can be interpreted as a projection which has
to vanish for arbitrary, sufficiently smooth, functionsd"u {=vir-

tual displacements}. If (A) is used for Galerkin's method, both
the approximate solution and the "projection" functions (in sec-
tion 7.42 denoted by U ard V J respectively) have to satisfy only
the geometrical conditions (d. table in section 7.43).
A general expression for the virtual work of the
stresses in a (classical) continuous body is given in
A. E. Green, W. Zerna, Theoretical elasticity,
Clarendon Press, Oxford 1963; (formula 2.6.7).
9. 3 The general form of Hamilton' s principle
We obtain H<.lmilton's principle from the principle
of vi rtual work by ir:troch,cing time -dependent virtu:>l displace-

ments, e. g.) Ju (X It) , and integrating (A) over a time intcr-
val t1 :!: t ~ tz
tz
!(dWr + dWIV ) dt =0 (II)

t1
Dy partial integration 'with respect to the time
ann sHitable as sumntion s about the vi rtual displa cem('nts at
The General Hamilton's Principle 73
it is always possible to express the first
term of (H) by means of the kinetic energy:
t2
: lufkin dt :: (I)
t1
d. section 4. 1.
If the system is conservative , 0' Ww tou can
b~ written as a variation of the potential energy. £ pot
J Ww = - a£ pot J (J)
and we obtain immediately Hamilton's principle in one of its
usual forms which are given in section 4. 1.

But the usual form of Hamilton's principle can be
valid for special nOl-,conservative systems too: If 0 W", can
be considered as the variation of a time-dependent force func-
tion UF ( 1.L1 r.'J t) ,
then Hamilton's principle can be written in one of the forms

t2
dJ( Ek~n + U F) dt =: a, (K)
t,
or
74 Equation of Motion
t2
/(E,," + U,) elt • extremum. (K')
t1
Such a system is not conservative but (K) and (K') hold.
Lanczos calls such systems "monogenic" (the
forces are generated by a single function); d. C. Lanczos,
The Variational Principles of Mechanics, 3. ed., University
of Toronto Press, Toronto 1966.
Example for a monogenic system The beam loaded by a
pulsating force, d. Fig. 9. 2
pC t) - verLcdL
~.J=+V,
EI
x
Fig. 9.2
E I - bending stiffness {elastic beam}, no longittldin~l defor-
matiol"!;srnall deflections uJX ,t). l

Ela sti c potential ene rgy. Epot = i jf I ( u."/d X
a
Let v~ denote the vertical displacement <it X = L
Then
The General Hamilton's Principle 75
(d. the signs in (K), above) and in (*), section 4. 1}.

For small u.' (X It) we have
=il(U,)2 dX .
L
vl
The kinetic energy is the same as In section 4. 1.
We obtain
;/f {u ",2 - E I (,-,-")2 + P{ t) ('-'-'J2) dX dt~ edr. (Ll

[1 0
(L) might be solved by Ritz's dire ct method.
Geometric constraints (boundary conditions):
u(o,t):::O, U'(O,t)=O
Equation (H), again, can be used as basis for Ga-
le r kin's method.
Goldstein's form of Hamilton I s principle for non-
conservative systems (d. pp. 38/39 in the book cited in sec-

tion 4.2) is misleading because he defines a work function
{vector notation}
and
JW=EF- ·Jr-
'Vi-
- rv /..
/,
even for F i 'V

which may depend on r-
~~ )
r·
",L
, etc.
M. Levinson (Application of the Galerkin and
Ritz methods to nonconservative problems of elastic stabili-
ty, ZAMP!l (1966),431-442) who tries to save Ritz's

method for nonconservative systems starts from Goldstein's
form of Hamilton's principle. I consider his investigations
to be somewhat arbitrary and, mathematically, not straight
forward.
Cf. H. Leipholz, Uber die Befreiung der Ansatz-
funktionen des Ritzschen und Galerkinschen Verfahrens vor!

den Randbedingungen, lng. -Arch. 36 (1967), 251-261.
Linear External Damping 77
10. Damping (linear)

10. I Linear .external damping
To take "some" damping effects into account, the
equations of motion, d. sect. 2, are frequently supplemented
by a term b u. ; b - damping coefficient,
Lu+bu,+pii ::0,
(*) is still linear but/in general,in this form it is not separa-

ble (see section 11), By assuming
(A)
u=U(~)T(t)
we obtain from (*)
(B)
Separation impossible if band/or p. depend ~n the space
variables.
78 Linear Damping
10. II Special separable case: b is proportional to )J.
In this case ft - constant I (C)
and from (B) we obtain
L U = A.fl U (D)
T+fot+AT ==0. (E)
Additionally we have the boundary conditions (which we assu-
me to be separable). If the eigenvalue problem constituted
by (D) and the boundary conditions has the set of (positive)
eigenvalues An., 11. =1,2, ... , then (E) lias the solutions
Tn. ::" e -JH/2 (An stn. Wn.. t + Bn. cOoS Wn t) }
where
d. section 8.2/8. 3.
If /l is large, some of th() lower may be
negative real).
The set of the is shifted
into the negative imaginary half-

plane of the c::lmplex plane
(cf. Fig. 8. 3 and Fig. 10. 1).
Fig. 10.1
Further separable cases exist for special systems
10.12 The general case
The as sumption
transforms (*) and the boundary conditions into
(F)
and boundary conditions for u which equally may include
the eigenvah1.e t . Such "generalized" eigenvalue problems

are sometimes called "nonlinear" because r occurs non-
linear in equation (F) and/or the bounda ry conditions.
Little general investigations of such problems
are kno\vn (for instance the notion of adjoint operators,cf.sect.
5, is not helpful in this case; d. sect. 11). Numerical deter-
Dlination of eigenvalue's, 'In ,and eigenfunctions, Un

is possible but tedious (d. section 13).
In gene ral. both the tn and the Un. are complex.

The conlputations can be done with complex numbers, or -
after splitting '( I U, the equation (F), the boundary con-

80 Linear Damping
aitions, and the characteristic equation into real and imagina-
ry parts- with real numbers. The result
is, in general, a complex function of X and

..... t
and the boundary conditions(for u. ) are real,then
U II (~, t) (H)
:: eot fuR. cos wt - VI s/nCJIl J
(I)
= erJt CUR Sin wt + VI Cos wt1 ,
are two linearly independent, real solutiolls. (The subscript, 11 ,
is dropped in these equations. )

The solution (G) can be interpreted visl'.ally as shown
in Fig. 10.2.
logarithmic spiral
increasing if 0 > 0
decreasing if 0 <: 0
circle =
if tS 0
U - space curve
(rotating on logarithmic
spirals about the X-axis)
Reu Imu
Fig. 10. 2
From equations (H), (r),and from figure 10.2 it

is easily seen that there do not exist fixed nodal points (d. sect
5.3) if a (real) separation of the variables is impossible (cf.
section 1 1). The nodal points - the ze roe s of lL R , U. I -
move along the X - axi s.
If the equation of motion of the

beam continuous part of the system is
separable but the boundary condi
dash pot tions are not separable(with iden-
X(damper) tically the same parameter of
separation) then there do not
exist separable (real) svlutions
,///,,///' of the problem (d. the system
sketched in Fig. 10.3). The so-
Fig. 10.3 lutions have the form (G).
82 Lineal' Damping
10.13 Investigation of a damped system by Lyapunov's direct
method
~xample Eule r - Be rnol111i beam
Undamped (flu")" + pii. '"' (J)

Damped: ( EI 1.1.")" + b u + )1. (K)
~ 0====__::1 Bounda ry conditions('see Fig.
}(L)
10. 4)
u"\ I] • t J' (O.t) = 0
: U'
Fig. 10.4
Total energy 1'0, el1(' , •• ,d;\nlped
U,,;' l
systcnl
\
r\
: - - '" (E I u,") =0
X= l
'I
fi
L
Etot =: ~ I (u"r dX
,--Q.____ -y-_ ___ -----I
+ ~ 1P
_Q_ --y-_______ J
'-. -
u,2 ci X
(M)
positive definite positive definite

'---- --- ._------------------v-- ------- _____ . ______ ::...J
positive definite
L.... Etot > 0 if U~O and U=:O if Etot=O, (N)
u sufE ciently smooth.
Etot = V (u) is used as Lyapul10v s fun ction for the damped

I
~nd the undamped system.
Differentiation of V (U) with respect to the time

Lyapunov's Direct Method 83
yields
Io I
t t
V= EI u." 11" d. X + )J. u. ii d. X. (0)
0
By partial integration and substitution of (K) we obtain from
(0)
J
t
V = - 6 (l·ci X ...:: a)
o
but u. == 0 I u.~ 0 is no solution of (K). thus, V --.. 0 and
u. (X, t ) --- 0, all disturbances (oscillations) fade away

asymptotic stability (d. sect. 12). (P)
Similar arguments hold if in (K) the term b U. is
replaced by other. e. g., nonlinear, damping terms.
For the stability problem rre ntioned in the section
8. 1 the energy Etot is not positive definite. therefore a

conclusion similar to (p) is not possible. Cf. section 2. 3 in
H. Leipholz. Stabilitlitstheorie, Teubner, Stuttgart 1968.
Literature on Lyapunov's method:

W. Hahn, Theorie und Anwendung der direkten Methode von
Lyapunov, Springer, Berlin 1959
V.1. Zubov, Methods of A. M. Lyapunov and their application,
N oordhoff, Groningen 1964.

84 Linear Damping
10. 14 Forced vibrations
Forced vibrations can be studied in a similar way as
'll chapter 6. Since the natural frequencies, wn. , are re-
placed by the complex values yn. a resonance catastrophe
with deflections of arbitrarily large amplitudes can not occur,
because the real frequencies .n.. of the exciting forces can
never coincide with '(n" tin. T L w n. • Comparative-
ly large amplitudes occur if In - wn. I and Ion. I are
small.
This subject is treated in many books. Here we men-
tion only E. Sknrll"7.yk's book {written from the acoustical
point of view} where much more literature is cited:

E. Skudrzyk, Simple and complex vibratory systems,
The Pennsylvania State University Press, 1968.
10. 2 linear viscoelastic internal damping
The internal damping of the material is frequently
taken into account by a viscoelastic stress strain relation:
(A)
Constitutive equations of this form are usually vi-
sualized by "models II composed of dashpots and springs as
shown in Figs. 10.5,10.6.

Viscoelastic Damping 85
Voigt-Kel yin Maxwell

( ~ equ. (A)) ( 1\ equ. (C))
Fig.10.5 Fig. 10.6
Various combinations of these elements are possible. They re-

present constitutive equations of the form
If (B) contains time derivatives of ($ the momentary stress

depends on the strain-history.
Example : The Maxwell model (d. Fig. 10.6)
Stress strain relation:
Elimination of C1 I 1:.2 leads to

(j •
+ - =t, (C)
£2.
86 Linear Damping
We consider e(t)to be a known function of t and (C) to be

an ordinary differential equation for 6 (t) We obtain the
solution
-00 -00
Lit. : Short summary: Chapter 53 (by E. H. Lee) in W. FHigge!s'

Handbook cited in section 2.242
Extension to three-dimensional states-of stress: S. FHig-
ge, Handbuch der Physik, BeL 6, Elastizittlt und Plas-
tizittlt, Springer, Berlin 1958
10. 21 Distribution of eigenvalues
Example: Euler-Bernoulli beam with stress strain

relation (A).
~~II,E71£2 - constant
(S x J~ 3- Equation of motion :
E1 I E2 I U. IV a ()
" l // ~ U
IV
.,.
"
+,.1.1. U " E
Boundary 'conditions :
Fig. 1 O. 7 LL ( O. t) =:; u (l~ lr~ 0
M(0. t) -= Et I u." (0. t) +- E2 I u O. t) " 0 /I (
r-'/ (L, t) ,~ [1 [ LL" ( L, t) + [2 r li I! ( () • t) ~ U

Both the equation of motion and the boundary conditions are
separable.
Solutions
U n. (X • t) S ,. n n. i'C X e Yn t
L
J'
=
Fig. 10.8 and

y~ =- 7C 4 n" j!J/2 :!: Vn8n.8J.//4 - rr.I.n,t.
The couples of cOITlplex conjugate values t~ lie on the

circle shown in Fig. 10.8, the real values y: lie in the
interval (-oo,-1Ift) of the real axis, -1/fo is a cluster
point.
Remark concerning the cluster point ( d. section 5.2):

88 Linear Damping
Assuming a solution of the form
u. (X It) = U ( X) e At
for (E) we obtain the differential equation
U'~
J..Ct
.,. --=----- u =0 (F)
E.1 .,. Ez1 A
for U (X ,f...) . Obviously. the term with A. in (F) is not an

integral function of ").. , so the characteristic determinant.
a (A.) ,formed by means of four linea rly independent so-
1utions Ui (X, A) I i =1 J ••• , 4 ) of (F) will not be an integral

[unction of A. , therefore, the roots t.. n. of the cha racte r-
istic equation, AlA)" a, may have a cluster point.

Mechanically, the appearance of the cluster point in
our example can be inte rpreted as follows : Let us as sume

the beam has initially a kink, see Fig. 10. 9. Since the mate-
rial is vis coelastic the
kink
kink will disappear slow-
ly. Thus, to represent
thi s behavior by a Fou-
riel' expansion with
respect to the eigen-

Fig. 10.9
functions, d. se ct. 5. 5,
we need high order eigenfunctions - to be able to represent the

kink -whi.ch belong to eigenvalues with comparatively small
negative real parts. Such functions are provided by the eigen-
values in the vicinity of the cluster point.
10. 22 Self-exdted vibrations of a rotating shaft caused b y

internal damping
Internal damping can be the cause of self-excit-
ed vibrations of a rota.ting shaft.
Lit. : W. Kellenberger, Die StabiliWt rotierender Wellen in-

folge innerer und ~usserer D~mpfung, lng. -Arch. ~ (1963)
p. 323
G. F. Schirmer, Zur Stabilit~t der Schwingungen von
Turbinenwellen, Dissertation, Techn. Hochschule Darmstadt,
1969.
Survey article: R. E. D. Bishop, G. Parkinson, Vibra-
tion and balancing of flexible shafts, Appl. Mech. Rev. Q, No.
5 (1968).
90 Linear Damping
10. 3 "Complex modulus" damping
The hysterctic loop of a Voigt-Kelvin material(cf.
Fig. 10. 5 and equation (A) in sect. 10. Z))ej(tended sinusoidal-
ly} is an ellipse, see Fig. I O. 10. Metals show such
Fig. 10.10 Fig. 10. 11
a behavior only at low stresses. For intermediate stresses

metals show a rate-independent damping and a hysteretic
loop of the forITl indicated in Fig. 10. 11 (nonlinear!).
For practical probleITls the rate -independence of the
d:l.ITlping is often approxiITlated by the linear relation
(G)
""here E* is the cOITlplex ITlodull.ls of elasticity
[ "-E'
- + L.£" }
E I - storage ITlodull.ls (elastic ITlodull.ls, real ITlodl.lll.ls)

Complex Modulus 91
E" - loss modulus (dissipation modulus)

E" IE I = T) - loss coefficient (loss factor, dampi ng factor)
In the relation (G) (j and c are assume d to be

complex and to be represented by complex Fourier series or
Fourier integrals. (So the actual values are ReG" and Ret
or Irno and ImE. ). Correspondingly, equations of motion in
which (G) is involved must be inte rpreted and solved in an ade-
quate way.
*
Frequently, for special systems, E is obtained by
sinusoidal excitation of the system and given in one of the
forms E * (w) or E *(w ,A) where W is the frequency
and A is a characteristic amplitude. The application of such

an E * to non-sinusoidal motions, they might even be periodic,
is not possible from the logical standpoint.
Lit. : R. Plunkett, Damping definitions and classifica-
tions, Lecture at the 1968 meeting of the Acous-

tical Society of America in Cleveland, Ohio.
Survey articles in "Applied mechanics surveys",
ed. by Abramson, Liebovitz, Crowley, Juhasz,
Spartan Books, Washington, D. C., 1966:
R. Plunkett, Vibration Damping, pp.691-702,
B. J. Lazan, Damping properties of materials,

members and composites, pp. 703-716.
92 Non-Sel£-Adjoint Eigenvalue Problems
11. Some remarks on the separability of damped systems

and non-self-adjoint eigenvalue problems
II. I Transformation of a "damped equation of motion.. Into a

separable form
We introduce
(A)
into the equation (..) of section 10. I and obtain
(B)
where
1= (~l ~b) ~ = (~
1 :).
Equation (B) can be separated formally:
~(~)t)= ~(~)T(t), (C)
(D)
.
T = YT . (E)
In general the eigenvalue problem constituted by
(D) and a proper set of boundary values will not be self-ad-
joint (the eigenvalues Yn need not be real, d. sect. 10. 11).

Non-Self-Adjoint Eigenvalue Problems 93
11. 2 Non-self-adjoint eigenvalue problems

Example Damped longitudinal vibrations of a rod, d. sec-
tion 2. 1 and Fig. 11. 1.
Equation of motion :
- (A Eu' Y+ bu. + ,U ii = o. (F)

Fig. 11. 1
Boundary conditions x=o

(G)
x= t :
The transformation (A) does not yield a form suitable for the
further investigation. Therefore, we replace (F) by the first
order system
Lu.
.
.......... = J,Z u
(H)
where
U. 1 == U. ) U. = (~;) )
1)o ~oX _(b 0)

a 0 ,~-
_(p.a
The boundary conditions (G) are equivalent to
(I)
(Some general considerations about the equivalence of a
system of differential eqalations and a single differential equ-
ation see chapt. 1, § 2 in vol. 2 of Courant, Hilbert's book
cited in sect. 5.2).

94 Non-Self-Adjoint Eigenvalue Problems
The separation (C) leads to the eigenvalue problem
hY=Y~!d, i~=O. (J)
'" u
t ...., = 0 denote s the boundary conditions
(K)
By the procedure outlined in section 5.42 we obtain the ad-
joint eigenvalue problem
= "~Y.. , !+ Y = 0 ,
where AE)
o ax
~ _(b (AE)') + l
0 a '"
The eigenvalue problem (J) is not self-adjoint.

Neumark (cited in sect. 5.2) proves: If '{ is
an eigenvalue of (J) then u =t (complex conjugate) is an

+
eigenvalue of (J ). But, in general, if ~ is an eigenfunc-
- +
tion of (J), V = ~ need not be an eigenfunction of (J ).
11.21 Orthogonality of eigenfunctions

Let U 1 A ,and V) u be eigenfunctions
~ '"
+
and the correlated eigenvalues of (J) and (J ), respectively.
Similarly as in section 5.44 we obtain(for the one dimensional
case)
Non-Self-Adjoint Eigenvalue Problems 95
ciX •
~--------~v~--------~
=0
-
U and V
1.
are 0 rtho gonal if '( =1= '\J
f
0
y T ~ T ~ d.X =0 (L)
If the eigenfunctions
known,(L) can be applied in the same way as the correspond-
U and
'"
"
.... are
ing relation in section 5.44 to solve the initial value problem,

d. section 5. 5.
The notions mentioned in this section are closely related to

corresponding notions in the theory of integral relations, cf.
Courant, Hilbert, vol. I;
W. A. Smirnov, Lehrgang der h~heren Mathematik, Bd. IV,
Deutsch. VerI. d. Wissensch., Berlin.
96 Lyapunov's Stability
t 2. Lyapunov's definition of stability
Let
G [u] =0 (A)
be an equation of motion which governs the motion of a given

system completely (differential equation with proper boundary
conditions, Hamilton's principle in its general form, etc. i cf.
sect. 9). In general u. will be an N -dimensional vector, cf.
~(~/t) in section 1.2.

Let
u = u·(~.t)
be a solution of (A). We define a new state variable v (X.t)

.... by
v = u - u. ~ (B)
Introducing V into (A) we obtain
G*[V] = 0,
an equation for V ( 3 ' t)
In ordinary differential equations the -nonlinear-equa-
tion *
(A) is sometimes called the variational equation - belong-
ing to the solution u.. - (cf. L. Cesari, Asymptotic behavior
and stability problems in ordinary differential equations,
Variational Equations 97
Springer, Berlin 1963). But in general the linearization of
(A*) with respect to V
(A l. *)
is called variational equation. (Cesari calls (A L *) linearized
variational equation}.
/.Lilt is called the undisturbed motion of the
given system and V is the disturbance. (A *) and (A 1. *) have
"lways the trivial solution v .. :: a which corresponds to the
*
solution u. of (A).
Let
IIt.LlI~O (c)
be a norm for all LL which are elements of a proper function

space (d. !full in sect. 5.41), and
II u II : : : 0 if and only if (n)

(Some more relations which Hull has to satisfy and some
generalizations see Zubovrs book cited in sect. 10.13).

Example Scalar U (X t)
J I 0 ~ X~ L
II u II = [J~ u2. c1 X] 11< L2~ norm.

2-
a
12. I Lyapunov'8 (weak) stability
The trivial solution *

of (A ) - and, cor-
respondingly, the solution u" of (A) - is cailled (weakly)

stable if for any t :> a there exists a ~ :> 0 such that
II V (.5 . t ) II < f, for
if
I Vo II <: v) 0 = 0 (f,) I
where Vo = V (! to)
I is the initial state {disturbance}at
12. 2 Lyapunov's asymptotIc stability
The trivial solution is called asymptotically stable
if it is stable and if
II V (~, t ) II --- 0 for t -- 00.
12. 3 Lyapunov's unstabHJty
The trivial solution is called unstable if it is not
(weakly) stable, d. 12.1.

Stability Investigation 99
12. 4 Remarks
The stability definitions given here concern dis-
turbances of the initial conditions (12. 1 :"The disturbed solu-

tion v (!. t) stays in an Eo -neighborhood of the trivial solution,
v" == 0 ,if the initial disturbance I Vo ,is small enough'~)

Extensions to disturbances of parameters see Hahn's book
cited in section 10. 13.
There exist many more stability definitions, see,
for instance, J. LaSalle, S. Lefschetz, Stability by Lyapunoys
second method with applications, Academic Press, New Yor!':
1.961
'v Bogusz, Lyapunov's stability in engineering, pp. 61-67 in
Zagadnienia Drgan Nieliniowych (Nonlinear Vibration Pro -
blems), Vol. 9, Polish Academy of Sciences, Warszawa
1968.
12. 5 Stability Investigation by means of the (linear) variational equation
In many cases the stability of a given system is
investigated on the basis of the linear equation ( A ~ ) be-

cause its solutions are much easier to discuss than the solu-
tions of the nonlinear equation (A ). *

For ordinary differential equations and equations of
motion, , which can be split into
(E)
where G: [V] is of second or higher order with respect to
v, Lyapunov proved the following theorems:
12. 51 If the trivial solution of the (linearized) equation
( At) is asymptotically stable then the trivial solu-
*
tion of the nonlinear equation (A ) will be asymptotic;l-
ly stable.(For sufficiently small disturbances}.
12.52 If the trivial solution of the (linearized) equation (A t)

is unstable then the trivial solution of the nonlinear
*
equation (A ) will be unstable.
12.53 If the trivial solution of the (linearized) equation ( A~ )

is only weakly stable then the stability of the trivial
*
solution of the nonlinear equation (A ) depends on the
higher order terms, it may be either stable or un-
stable. (This is the so-called "critical case").
To my knowledge there do not exist the correspond-
ing theorems, e. g., for partial differential equations (con-

Stability Investigation 101
tinuous systems), but I feel sure that similar relations hold in
that field too.
Let us reflect upon the common stability problems

of civil engineering from this point of view: Most stability
problems civil engineers investigate concern linear con-
servative systems. In the case of stability the eigenvalues,
y ,are distributed along the imaginary axis, d. sect.
8. If some damping} which is always found in an actual sys-
tern, is taken into account, the eigenvalues are shifted into
the left half of the cOITlplex plane, cf. sects. 10. 11,10.21, the
trivial solution becoITles aSYITlptotically stable. Hence, sITlall
nonlinear terms can not change the type of stability.
102 Kinetic Stability of Autonomous Systems
13. Kinetic stability of autonomous systems
A system is called autonomous if its equation of mo-
tion does not depend explicitly on the time, otherwise ic is
called non-autonomous.
A mechanical system is in the state of static equi-
librium under a given constant load. At the time t:: 0 the
system is disturbed {e. g., displaced slightly from its equi-
librium state}. What happens to the system for t :> a ? The
figures 13. 1-13.6 shew six different typl'S of response.

u. u. u.
-1--+-____--===..... ___ \.La

-------11_
...
t t
Fig. 13. 1 Fig.13.3 Fig. 13. 5
i~
a,~7 t
Fig. 1 3. 2 Fig. 13.4 Fig. 13.6
By It we denote the deviation from the equilibrium
position, lto is a small initial distrnbance.
In Fig. 13. 1 and Fig. 13.2 the disturbances disap-
pear "exponentially" and "oscillatory", respectively, the

Kinetic Stability Criterion 103
system returns to its original equilibrium position. In Fig.
l3. 3 the disturbance causes an os dIlation with amplitudes
which have the magnitude of u'o . In Fig. 13.4 the "disturb-
ed" position is a new equilibrium position. In Fig. 13. 5 and
Fig. 13.6 the initial disturbance causes deviations which in-

crease exponentially and oscillatory,respectively.
Corresponding to Lyapunov's definition we have

Fig. 13. 1 I 13. 2 13. 3 .. 1 13.4 13. 5
I 13.6
s tali lity asymptotic we ak unstable
Fig. 13. 7
We are interested in the unstable cases shown in
Fig. 13.5 and Fig. 13.6.

The Fig. 13. 5 represents, for instance, the
buckling of a column, d. section 8. The instability starts
when the fi rst eigenvalue, '{ 1 , m.oves through the origin,
'( == a , onto the positive real axis, d. Fig. 8.4. The
critical load, Pcrit in section 8, can be found by static
methods.
The unstable motion shown in Fig. 3.6 starts when
one of the eigenvalues, 'In· , say, moves into the positive
real half-plane, Re 'In· > 0 , because the load para-

meter, say P again, is increased over a critical value/Pcri.t
So we have the ("kinetic") stability criterion
Asymptotic stability if Re tn. < a for all '{n..
Instability if Re 'In. ;::. 0 for at least one Yn.

The case of weak stability corresponds to
the border case P = PCri.t and Re Yn- = 0; Re yn. < 0

for Yo, =1= Yn..
Imy Pcri.t
In Fig. 13.8 the arrows indicah:
/ two different ways in which an
eigenvalue, Yn * , can enter
Ret the positive real half -plane v.hell
P is increased over the critic-

" Pc.rLt
al value, PCI'"~t
Fif. 13. 8
We see: To find Pua for such systems
:.he eigenvalues Yn. themselves must be investigated. Thus
we have to study the motion, i. e., the "kinetic" (or"dynamic")
behavior of the system. That is the reason why such problems
are called "kinetic" (or "dynamic") stability problems.

Follower Forces 105
In aerodynamicswhere many of these pro"-
lems originate the unstable motions from Fig. 13. 5 and 13.6
are called "divergence" and "flutter", respectively. We shall
use these notations. (Cf. G. Herrmann, R. W. Bungay, On the
the stability of elastic systems subjected to nonconservative
forces, J. App!. Mech . .l..L (1964) 435-440).
J3. I Follower forces
Example: Euler-Bernoulli beam loaded as
shown in Fig. 13.9; P - follower force of constant magnitude,
direction: 1p:: (1 - rn) a . (see Fig. 13.9), m -pa ioamete r
rn. = 1-vertical P, m= 0 - tangential P

"'l' Equation of motion (in a
x non-dimensional form) :
, u." +
LL'V + 20 ii "" 0 (:\ )
E r,p
u = wiL, ~ X/l, 1'= tVEI/p.Z4)
==
2 P = PtZjEI.
Fig. 13.9 Boundary conditions
LL(o,r) = LL'(O,T)=O}(B)
LL" (1,1') = 0 ,uur(1, r) -r 2prnu.'(1,l')~ 0
The equation (A) is self-adjoint but the boundary con-
ditions, (B), are not if m =1= 1
13. 11 Static investigation
The eigenvalue problem reduces to
u.'1! + 2p u," = 0,
and the boundary conditions (B)~ p is the ejgcnvalue parame-
ter. We obtain the characteristic equation.
m (co~ V2P - 1) +1 = 0 (c)

(C) has real solutions, Pn.' only for rn ~ 0,5. The lowest
value, p cr~drnJ , is
shown (for m ~ 0,5)
in Fig. 13.10. Con'cs-
Fig. 13.10 ponding to the static
stability inve stigation
the re would be no sta-
bility loss for arbitra-

.-
-2. ·1 o O,!I 2 rily large loads in the
Follower Forces 107
range m .c:: 0,5.
13. 12 Kinetic investigations
Assuming solutions of the form
we obtain from (A), (B) the eigenvalue problem
UIV + Z P U /I + AU:: 0 I A = Y 2.,
The characteristic equation can be established in an
"~·n1i.cit form (cf. H. Kenig, Die Knickkraft beim einseitig ein·
~(;spannten Stab unter nichtrichtungstreuer Kraftwirkung, Der

Stahlbau 29 (l960) 150-154).
We have the stability condition
Re Yn. <: a for all )'n.

The critical load, p,rLt , is obtained from
Re y1 = 0
assuming that the lowest normal mode becomes instable first.
Unfortunately, the calculations have to be done numerically
(d. H. Kenig cited above). The curve pcrd (m) is represented

in Fig. 13. 10 by a dashed line (marked fo = 0 ). Surpris-
ing1y there is a jump in P'r'~t at m= 0.5 . (Cf. similar Jump:.;

in Hermann &: Bungayls paper cited above).
13. 13 Influence of damping
KCJnig showed that external viscid damping, d. sec-
tion 10.1, changes the kinetic stability behavior just slight-
lY.. and the static stability behavior not at all.
I myself investigated the behavior of a viscoelastic
beam loaded as shown in Fig. 13.9 (E. Brommundt, Uber den
Einflus s von Werkstoffdlimpfung auf die StabilitHtsgrenze von

Stliben mit Folgelasten, ZAMM 46 (1966) T 117-119):
Equation of motion
U II/ + J.> uI" + 2. Pu," + u. = 0 I (D)

fi - damping coefficient.
Boundary conditions
u, ( 0 , 1:' )::: u' (0,;;)= U" ( 1, T ) + fo u." ( 1, -r) := (E)
(,t'll (1, r) -I- J~ U, "I (1. 'r) + 2 p m u,' = o.
The characteristic equation can be established ex-
plicitly ag-ain, Putt has to be calculated numerically. The
results are shown in Fig. 13.10 for some values}) > o. The
curve fi-... 0 is located slightly below the curve 13 = 0,001.
Follower Forces 109
Vie see:
1. Damping can decrease p c.r~ t in the range of the kinetic
stability criterion (d. H. Ziegler, Die Stabilit~tskriterien
der Elastomechanik, lng. -Arch. 20 (1952) 49-56).

2. Pcr~t for p,--O and PcrLt for f> = 0 need not coincide.
3. In the kinetic range P crLt depends on the type of the
damping.
4. The jump of P crLt at m = 0.5 disappears.
In the range of the kinetic stability criterion the mass
di stribution, too, can influence the stability behavior of the
system (d. Leipholz's book cited in section 10.13). If the locrl
acting on the end of the beam is distributed in a more general

,vav,different types of flutter are possihle : see S. Nemat-Nas-
ser, G. Herrmann, Torsional instability of cantilevered bars
subjected to nonconservative loading, J. Appl. Mech. (1966)
102-104.
13.14 Movements of the eigenvalues .::..fo:c..::..r_~ffi-=--=~O.

Imy
r-Ret
'Pc.r.t
Fig.13.11 shows how the in move,
when p is increased, in the static
range, cf.Fig. 8.3

Fig. 13.11
1m)'
Fig. 13. 12 shows how
the '( n move in the
kinetic range.
Rey
(How do the '(n move

Fig. 13.12 for fl ~ 0 ?)
J3. 2 Pipes conveying fluid
Sxample Uniform pipe, d. Fig. 13. 13, length ~
£ I,fl ,.l.l b
bean'l density per unit
')'!,,-
~::;:::;:~G1--~ length
ftf - fluid density per unit
length
V - fluid velocity
EI.fo - d. section 13.1
Fig. 13. 13 (all coefficients constant)
EC[uation of nlotion
u.. 'V + jlJ 1L 'V + 2 P u.." + Cj u.. I + ii == 0

(A)
U,j3, etc. d. section 13.1, 2.P=+)J.fv2L~/EI.
g== 2 V)1 f L/ Vfl £ I } /u = A b ~ ).4 f .

<\dditionally, we have fOllr boundary conditions.
Pipes Conveying Fluid III
In general,this system is not conservative
since energy can be exchanged between the transversal
oscillations of the pipe and the longitudinal motion of the fluid
Therefore, the kinetic criterion must be applied.
Some results, for special boundary condi-
tions, are given,e. g., in the paper S. Ne mat-Nasser, S. N.
Prasad, G. Herrmann, Destabilizing effect of velocity-depen-
dent forces in nonconservatice continuous systems, North-
western University, Evanston, Ill., Techn. Report No. 65-4,
and in- my paper mentioned above.
13.21 Cases for which a static stability investigation leadsto
the prope r re s ult s
In section 13.3 we shall refer to Leipholz's
more systematic investigations of the problem mentioned in
the title. Here we shall derive just a special result obtainable
by Lyapunov's method.
Let us define for (A) a Lya punov function
v (u) = E K,n + E pot I
where
112 Kinetic Stability of Autonomous Syst~ms
Differentiating V with respect to 1: and eliminating U. by
means of (A) we get after some calculations
fa (
1
V = - fl lk"? d s+ B I
where
V is not positive if the boundary terms, B ,vanish. Then
we have stability if V itself is positive definite (d. Zubov, p.

38). E Ki.n. is always positive definite, E pot IS positive de-
finite if P < Pc:ri.t. ~tdt ,where p crit. ~tat is the statil
critical load. Thl!s, for B = a the static stability investiga-
tion leads to the proper reslllts. (The condition B = 0 is only
sufficient but need not bE' nccessary for the admissibility of
the static criterion). Follower loads may be included in (A).

13. 3 Galerkin"s method
Example Cantilever pipe, conveying fluid (no damping).
Equation of motion{cf.sect.13.2):
l u.'V+ 2pu." + qu' + ii, = 0 (A)
Fig. 13.14
Bounda ry conditions:
u, (0. l) = u ' (0, t) :: u." ( 1 , t) =: U'" (1, L ) = O. (B)
Introduction of (B) into the expression for the boundary terrns.
B, in the preceding section does not show that we are allowed
to apply the static stability criterion.

By
u."u(~)eyr
we obtain from (A). (B) the eigenvalue problem
U IV + 2 P U 1/ T t qU' + Y 2 U :: a) ( C)
U (0) = U'(o) = U" (1) :: UII ' (1) '" O. (D)

We apply Galerkin's method to solve (C), (D).
As approximate functions we use the first N normal modes

of the cantilever beam shown in
t
Fig. 13.15. Its equation of motion
is
u,v + U. = 0,
Fig. 13.15
the boundary conditions are the same as in Fig. 13. 14, i. e. ,
the conditions (B). We denote the normal modes of the canti-
1eve r beam by Vn.

Introducing the approximation
(E)
into (C) and applying the Galerkin projection method (d. sect.
7) we obtain a set of N linear homogeneous equations for tht:·
coefficients an.. In matrix notation we have
(F)
c3 is the column vector ~ = { a1 I ••• ) d N } •
~ ~ are N x N matrices having the elements

A }K = (Vr of- 2pVj', Vx )

d. sect. 7.4 (G)
B JK = (V; , VI<)
In general)
A kX * AK t J B tk =i= BK~'
I is the identity matrix (the Vn. are ortho-
gona1 and assumed to be normali7.ed).
The approxirn::tte expression, (E), satisfies
the boundary conditions (D) since the boundary conditions are
the same for the systems shown in Fig. 13.14 and Fig. 13.15.
(If for a given problem no functions, Vn. , can be fonnd which

satisfy all boundary conditions, Ga1erkin's method may be
replaced by the expressions of the principle of virtual work,
d. sect. 9. Then the geometrical boundary conditions only
have to be satisfied).
(F) constitutes a matrix eigenvahle problem which has, say,
the eigell\'-dlucs Yi ; and the eigenvectors ,e} (see R. Zur -

mUh1, Matrizen, Springer, Berlin).
Tntroducing ~} into (E) we obtain an approxirrn. te
('igenfu]"~ction Uj (;) for (C), (D) which is correlated to the

approximate eigenvalue fJ.

H. Leipholz (StabiliHitstheorie, Teubner, Stuttgart
1968) investigated the convergence of this procedure for

some classes of nonconservative problems. Furthermore, he
discus sed the applicability of the static stability criterion
depending on the structure of the matrices A and B

~
in
equation (F).
Lit. see survey article: G. Herrmann, Stability of equilibrium

of elastic systems sUbjected to nonconservative forces, Appl.
Mech. Reviews, ~, No. 2 (1967).

Stability of a Compressed Bar 117
14. Kinetic stability of non - autonomous systems

14. 1 A simple f'xample (compressed bar)
Incompressible column (Euler-Bernoulli beam), see Fig.

14. 1, loaded axially by a periodic force P (t) = Po + P1 cos n t
1- 1
~f::=/==~\=E=JJ=}J.==~~7/m 'P (t)
Fig. 14. 1
Equation of motion:
u.. IV + P ("C) u.. \I + U. = 0 (A)
\\'he re
w=n/VET!p..t'-,cf. sect. 13.1

Bounda ry conditions :
(B)
Assuming solutions of the form
Lll'\= al'\(r) sLn n1t1; ) an.-norm.al coordinates, (C)

118 Stability of Non-Autonomous Systems
we satisfy the boundary conditions. (B). and obtain from (A)
n= ~,." (D)
(D) is an uncoupled infinite system of linear ordinary second
order differential equations with time dependent coefficients.
These equations are of Hill's type if pC'"') is periodic. they
a re Mathieu-equations if
as in our example.
14.11 Some results on Mathieu eguations
(See, e. g., .T. Meixner, F. W. Sch~fke, Mathicusche

und Sph~roidfunktionen ... , Springer, Berlin 1954; Bolotin IS
book cited below).
We write (D) in the form
d 2y ( 2 )
(E)
-- + A. - h co~ 2X y = 0
dxz
By Floquet's theorem (E) has two linea rl y independ-
ent solutions of the form
YF = Cf (x.) eVx (F)
(there is an e ..... c('ntional C'1SC'). q (x) is ? (cmnplex)pcriodic

function
q (x+ 'Jt) = q(x).
v - characteristic exponent {complex!}.
The Floquet solutions) 'J F ) increase ex-
ponentially with X (the trivial solution 'f :5 0 of (F) is
unstable) if 'Rev> 0 , they do not increase, we have {weak}
stability, if 'R ev IE o. V depends on A. and h2 , the para-
meters of (F). Fig. 14.2 shows stable and unstable parameter
regions. (The chart is symmetrical with respect to the A. -axis).

(weak) s ta bili ty
Fig. 14.2
On the calculation of such stability charts
see the references cited in the literature mentioned above.
14. 12. Stability discussion for the given problem
Comparing (D) with (E) we obtain

where is the reduced natur-
al frequency 0f the statically compressed beam. (AssumptiOl
Po « PcrLt, stat. L
From Fig. 14.2 we see that, for small P1 ' we hav(
unstable regions C'resonance ll ) in the vicinity of
=m m = 1,2,3 .... ( G)
m IS called the order of the resonance, m = 1 -rn::\ln
resonance).
We sketch a stability chart
(weak) st<lbility
m = 3 2 1
0,5 2
Fig. 14.3
Danlping has two effe cts
The weakly stable regions become asyn,ptotically stable.

2.. The stable regions grow, threshold values, P 1th )
come into existence, i. e., the instability regions in Fig.
14. 3 are elevated.
Cf. sect. 62 (by E.Mettler) in W. Fltigge's Handbook cited in
sect. 2. 24
14. 2 A non-separable problem (compressed bar)
In section 14. 1 we were able to separate the
variables, d. equation (C). Let us consider now an example
where no separation is possible: Tre column built in at both
its ends.
t
Fig. 14.4
The equation of motion, pet) etc., is the same
as in section ].~.l. The boundary conditions·J·E'
U.(O,T)= U'(O,L)= u(1,-c)= u'(1,1:)= 0
Again we apply the eigenfunctions, Vn (d.
13.3 and 14. I) , of the corresponding conservative system, i.
e., the clamped-clamped beam compressed by po , to ex-
press U. (~)--r) . But here, different from 14.1, the systems
of the functions
V II a re not simultaneously
, n 01'-
thogona 1. Thus we can not obtain a decoupled system of ordi-
nary differential equations for the normal coordinates an,

cf. equation (D) in the preceding section.
Introducing
A Non-Separable Problem 123
N
= ~ an(-r) Vn(~) (A)
n:1
which satisfies the boundary conditions, into the equations of
motion we obtain by Galerkin 1 s method
a = { d-j } ••. , aN} is the column vector of the an, ~ I '§ I £

are N x N matrices. ~ and B are diagonal matrices, £ is
a symmetric matrix, (d. F. Weidenhammer,
Del' eingespannte, achsial pulsierend belastete Stab als Sta-
bilit~tsproblem, Ing. -Arch. l1. (1951), 162-191).

Equation (B) represents a finite system of linear
n 1-dinary differential equations with periodic coefficients.
Again Floquet's theorem is valid(for first order systems see
Coddington, Levinson's book cited in sect. 5.2, special re-
snIts for second order systems see Bolotin's book cited be-
1,:;\':). Therefore,(B) has, except for some special cases, N
linearly independent Floquet solutions
( C)
',,'v ne re
(D)
is periodic and V is the characteristic exponent as in sectim-·
14. 11.
The stability criteria are the same as in sec-
tion 14. 11. The calculation of stability regions is very trouble-
some and intricate. For small P1 perturbation techniques are
suitable (see, e. g., E. Mettler, Allgemeine Theorie der Sta-
bi1it~t erzwungener Schwingungen elastischer K~rper, lng.
Arch . .!1. (1949), 418-449). For special, numerically given,
larger parameters the stability can be investigated by a numer-
ical integration of the differential equation (B).
14.21 Qualitative discussion of some results
Mettle r showed, d. the paper cited in the

preceding section, that in the general case instability regions
start not only from the points w = 2w h /m of the (U-
aX'is, d. section 14.12, but also from the points
W = (E)
m
CD n W k - natural frequencies of the (conservative}auto-
nomous system.
The instability regions starting from
W = (CU n + Wk)/ m J CU n ~ OOk,are called instability regions
of the second kind. (The instability regions of the fi rst kind

A Non-Separable Problem 125
aT'.d those of the second kind coincide for the example 14. 1).
For the beam shown in Fig. 14.4 results can be found in F.
Weidenhammer, Der eingespannte, achsial pulsierend belaste-
te Stab als StabiliHltsproblem, lng. -Arch. 19(1951), 162-191.

The effect of damping terms depends on the as-
sumptions about the magnitude of the damping coefficients. If
the damping coefficients are introduced proportional to P1 '

d. equation (B), then damping may increase the angle of the
wedge-shaped region of instability, d. Fig. 14.5. If, however

the damping coefficients a re independent of P1 ,then a
threshold value P1 th ' similar as in Fig. 14.3, appears but
the vvidth of the inst<l.bility region can be increased, d. Fig.
14. 6 (G. Schmidt, F. Weidenhammer, lnstabilitliten gedlimpf-
ter rheolinearer Schwingungen, Math. Nachr. Q (1961), 301 ).
Pl b- P1 P1 b= t()n~t
A. A
"\ '\
b=O /
/ \ b=O
\ ~
"<-, Xv
P1th
W 1+W2 W W1+ W 2. W
Fig. 14. 5 Fig. 14.6

14.22 Two remarks
If solutions of the form (A) converge for
N - 00 , then there exist Floquet solutions of the linear
periodic partial differential equation which describes the mo-
tion of this system (d. equation (A) in sect. 14.1),
(F)
where
(G)
To illy knowledge for partial differential
equations a Floquet theorem is still not proved.
As we have seen, the investigations In the
preceding sections are based on Floquet's theorem. Thus, no
extension of these methods to arbitrary time functions p (1:')

is possible (d. section 14.3).
14. 3 A l yapunov-type approach
We consider some results recently publish-
(:d : J. R. Di cke 1'son, T. K. Caughey, Stability of continuous
dynamic systems with parillnetric excitation, J. Appl. Mcch.

A Lyapunov-Type Approach 127
liF: (1969). 212-216. (We sketch here a simplified "lineariz-

eu" version of the theorem A of that paper).
For the system shown in Fig. 14.4 we have th\.;
equation of motion (cf. section 14.1)
ii + b u, + u"v + p (-t) u," =0 (A)
and the boundary conditions
Damping included, b> a - damoi ng 1 ~

coefficient.
(A) is a special case of the general equation (we call the va-
riables t and X again)
L1 ) Lz are constant (time independent!) differential

operators with rcspect to X or coefficients (they might depend
on X), L3 (t) may depend on the time, too.
(In (A) L1 =b, LZ =U 4 /()X 4 , L3= p(t)u 2 /uX2.)
We introduce an inne r product :
(u, ]v) = (v ) u.) = f u. ( X, t) V (X It) d Xj u.. v - rea1. )

(D)
o
128 Stability of Ncn-Autonomous Systems
?nd define a Lyapunov function
(E)
We assume
(AI)
which is easily verified for our example. Because of (AI) the

Lyapunov function V (u,) is positive definite:
Furthermore we assmne
(A2)
which, again, is true for (A).
From (E) we obtain by differentiation with
respect to t .llld elimination of ii by meClns of (C), taking (A2)

into account,
(G)
A Lyapunov- Type Approach 129
We demand
( u., L 1 U ) ;t A (u-. J u.)) A. > 0 ) (A '»)
(is satisfied for (A) if b > 0 ) apply this inequality to the
right hand side of (G) and put those terms into a suitable
order:
~(u.' + -l3LL) , C' l~U.)~

cLV
~ - At.. - U. + - -
cLt At At
(H)
The ten-..1.S with t.. are introduced to have some

flexibility for some estimates which follow below.
dV / d.t shall be estimated by a relation of
th,' form
~~ ~ - r,V + lj(t)V (I)
t;) 11 (t) - positive. (1) is integrable:
JTi(-r) d-c }
t
Vet) ~ Vo exp { - r, t + (J)

o
To convert (H) into the form (I) we drop the
first line on the right hand side of (H), (since that term is
negative, the inequality is not violated). The second line and
the third line of (H) are estimated in the following way:
(K)
C1nd
(L)
To simplify M1 in (K) "ve estimate V by
which is obtained from (E) by using
Thus ,we obtain from (K) and (L)
M1(E.) [( 1 - E.) A ( u. ,u.) + (L 1 LA., Ll U. ) ] ~

(A4)
[ 3 (L 11L I L1 u.) /4 + ( lL I L 2 u.) + 2 (Ii ) u.)]
l.nd
(A4) and (AS) can be read as definitions (or conditions) for
M 1 and M2 ' respectively.

Now, the second and the third line of (H) are
limited by -V/M 1 and V M2 ' respectively. We have, d.

~H) and (r),
1
~ = (M)
From (J) we obtain
wherE'
et = ~1 - ~up
t
+ t
!Mz(t.,-r)d-r
0
(0)
The trivial solution, Ll == 0 , of (C) is asympto-
tically stable if the assl1.mptions (Al)-(A5) are satisfied and if

a > O. Because of (F) we have
(p)
We apply (N),(O) to the example (A) :
The assumptions (A 1) - (A 3) a re satisfied. (A4) has the form
M1 [(1 - t) b ( u. , u. ) + ( b u" u,IV ) ] ~

(Q)
[3 (b u, , b u, ) / 4 + (u" u,'V) + 2 (u., u. )]
(Q) is valid if
(R)
and
(S)
1'0 estimate (u.,u.) by ( Ll., u,'V) = ( lL II IlL II )

we start f rom
f J
x X
lL = u,' d. ~ u.' = u," d.~

o o
d. the boundary conditions (B), and obtain by Schwarz's ine-
quality
fo (u."/ d~
)C
u.12 ~ X
and
( u, •u.) ~ (\lit. u.") = (u. I U. IV) •
Thus,(R) and (S) are satisfied by
M1 = max [ 2
(1-t) b ].
From (AS) we obtain, p = p (t) ,
Ml [(bu., bu.) / 4 i' (u. u. IV )]

J l!:
(T)
[(Pu" Pu,)/ A.t. + I (bu., Pu.)O·
(T) is satisfied by
i'herefore,we have asymptotic stability if
f [ Pb2(~)
t
5U.p 1t I
+ b p Ct) I] d:! < b (u)
o L. ma)( [_2_
1-t
, "5 bj 4+ 1J
2
For I p ("C) I :> 1 no stability is guaranteed by (U). The
static critical load is Pcr~t. !:otat = 4 'Jt: 2.
14. 4 Concluding remarks
We have restricted ourselves to linear in-
vestigations. Some nonlinear effects will be discussed later.
There exist many investigations "on the para-
metric response of structures". In a survey article of that
title by R. M. Evan -Iwanowski (Appl. Mech. Reviews JJL ' No.
9 (1965)) some literature can be found. Special prominence has

to be given to the book: V. V. Bolotin, The dyn<lrnic stability
of elastic systems, Holden-Day, San Francisco 1964. A series
of interesting papers is contained in the proceedings of a con-
ference held in 1965 : G. Herrmann (ed.), Dynam.ic stability
of structures, Pergamon Press, Oxford 1967.

Torsional Vibrations l35
15. Nonlinear vibrations of autonomous systems
15. 1 Non linear torsional vibrations of a cylinder

(d. H. Kauderer, Nichtlineare Mechanik, Springer, Berlin
1958).
Fig. 13. I
(I' - angle of twist

MT - torque
I p = 1tr4j 2 - polar moment of inertia
Q - density
G - shea r modulus
Acceleration of an element
,\lonlinear (cubic}stress-strain relation leads to

136 Nonlinear Autonomous Systems
Elimination of M T and introduction of non-

dimensional quantities yield the equation of motion
""
1J U. = U. -
•• U.II ["1 + ""
Jol U.1 J = O.
2
u. ,..., cp J ~,..., ()C ; we denote the nondimensional independent

variables again by X and t , u.. u.(X,t); D is a non linear dif-
Lerential operator.
Boundary conditions
u. (0 , t) = u.' ( 1 , t) = 0 .
The equation of motion if nonlincar, the boundzll'Y conditions

are linear, no superposition is possible. Th(' syst('rn is con-
servative
J5.11 Solution of the linear problem
For ~ =0 we have a linear problem. Its
solutions (eig(~nfunctions, eigcnvalues, cf. sect. 5)

;1.re
Reduction to Ordinary Differential Equations 137
The system of the eigenfunctions, , is complete.
15. 12 Reduction of (*) to a system of ordinary differential
~quations
We assume that (*) has an approximate solution
of the form
N
lLA(X,t) = n=1
~an(t)Un(X). ( C)
Since the Una re the eigenfunctions of the linea r problem,
(C) satisfies the boundary conditions (*~) for arbitrary an(t).

By Gale rkin' s Inethod \ve obtain
The f n (a 1 ) ••. ) a N) are cubic functions of

their arguments. We abbreviate (D) in matrix notation by
..a +C a +,., i (~) =0 I (E)
where £ = d ~a <3 (U) ~) ) n =1, ... , N

For N= 2 we have
(F)
fi· = f.l1t4/ 64 (below we write fl instead of ,. .. ).
15. 2 First order approximation N = 1
For N = 1 we obtain from equation(F) of the
preceding section ( a1 = a )
(A)
This is Duffing's equation. E-nonlinear operator.
The· solutions of equations of the type
a + h(a) = 0 ) a h (a) > 0 for a ~ 0 ) I a I ~ I a Imax, (B)
are periodic and can always be rcduced to ,lll integration (be-
cause the conservation of energy represents a first integral
of (13)). For Duffing's cquation, (r\), we obtain

.dt)
t - to = f da
Va?(t o) - w~ a 2- fia 4/ 2
a(t o )
a (t) can be expressed by elliptic functions (cf. Kauderer's
book cited above).

Fourier Approximation 139
In general. it is impossible to find analytic solu-
tions for higher order approximations. N > 1 .
b.2l. One term Fourier approximation
As a rough approximate solution of (A) we choose
a ( t) = A c.o!. 00 t . (C)
Putting (C) into (A) and expanding E [A 5Ln ootJ into a Fou-
rier series we get
E[A ~in wtJ = [- Aul+ Acu~ + 3}~ A~/4J C05 wt + A3 /4 c.o~ 3 wt (D)
~----..-v --' \." J
l 0 om~tted.
We obtain
(E)
See Fig. 15.2 ( f.> > 0 : strain hardening; /J < 0 strain
softening).
A A
j) > 0 ,.,<0
4----------4r--------- 2
00
Fig. 15. 2
The frequency, W , depends on the ampli-
tude,A, of the oscillation.
Similar simple assumptions, an = Anco!l oot,

can be introduced into higher order approximations, N > 1,
d. equations (n) and (F) in section 15. 12. But, in general, it
may be neces sary to introduce solutions of the form
Since the system is autonomous, the origin of the tim.(' axis
~:an be shifted in such a way that, say, BI vanish('s. Bya pro-
'edure similar to (D) we obtain 2 N (nonlinear) equations for
.he 2 N unknowns AI' A Z ' B Z ' ... , AN' BN and W , cf.
section 15.4.
If the system of ordina ry differential equa-
tions (D) in section IS. 12 has solutions of this type, Cal and
the coefficients AJ, B~ a re functions of ,., and of one of the
coeffici('nts A~., say j UA ' d. (C) in sect. 15.12, is
Deriodi c.
Ldus assume that laCt)! is small. Intro-
duction of
Perturbation Technique 141
a = f.a ) f.-small parameter
into (A) yields
~ + oo! a ... t..

~
2 ~ a 3 = 0 (F)
small
(F) corresponds to equation (A) with f.l small.
So we have to investigate a (t J fl)'or small values fi

We want to expand a (t J fJ) with respect to fJ
Looking at (C) and (E) we can expect to obtain terms of the
following kind in that expansion:
a (t) = A cO~ ( 00 1 + ~ (~ )) t + ...

= A co~ Wit + A ~(Jl)t !)~n 001t+ ••.
Even if this expansion would converge we would not be able to

interprete the terrns in the second line because of the secular
terms, t t 2 , •••• Especially, we would not be able to

determine the period of the solution
a(t+T,)l) = aCt,))) (G)
To overcome this difficulty we introduce a new time scale
(Lindstedt, 1883).
(H)
a ("'C, p,) is 2n - periodic with respect to "'C . From (A) we

eJoJtain
(I)
.,
(We denote a ).
Now we exprtnd a and w 2 into series with respect to .r..
(J)
Putting (.1) into (1) <lnd cOH1paring tern,s \vith E'ql1al powers of
fi we obt;1in
£.,2.1
"""
a" (0) + ,.VJ
.... 12 a(O) -_ 0 ) genera t'Ing equatlon
,
(K)
Initial conditions:
a (0) =A (L)
(L) IS satisfied if
Perturbation Technique 143
a (0) ( 0) =A ) a (~)( 0) =0 I ( = i, 2) .•.
(M)
a(~) (0) = 0 , t = 0, ~ J 2 , ...
Furthermore all a(~) (,t) must be 2Jt-pe-
iodi c.
From the fi l'st equation (K) we obtain
a (0) = A c.os , gene rating solution,
--
~
a (1: , ,l) a (0) for j) - - 0

Putting
a (0) in the second equation (K) we get
Since a(1) shall be 21C-periodic, we have to choose
3 l (0)
"1= -;; A.
The 21C -periodic solution a(1)of (N) which satisfies the
initial conditions given above is
and so on.
In the first approximation we have, d. (J), (0).
This expression coincides with equation (E) which was found

by the one term Fourier approximation.
The preceding calculations represent a ra-

ther spacial outline of a general technique. Mathematical in-
vestigations of these procedures were originated by Poincare.
General theory, see: I. G. Malkin,
Some problems in the theory of non1in(';~ r
oscillations, transl. from Russian, AEC-tr-3766 (book 1/2),
US Department of Commerce, Institute for Applied Technology.
In general a better ('onvcr~{,lH.:e is achieved

if (1) is devided by W 2 ,
..
a + ,
and 1/ CJ.)2. is expanded into a series,
'lee Malkin's book cited above and Yu. A. Ryabov, pp. 425 -
445 in vol. 1 of the Proceeding s of the Inte rnational Sympo-

sion on N onlinea r Vibrations, Published by the Academy of
Sciences of the USSR, Kiev 1963.
Higher Order Approximations 145
If the nonlinear terms are continuously differ-
entiable but not analytic,a series expansion is impossible but
a successive approximation technique yields results, d. Mal-
kin's book.
t 5. 3 Higher order approximations N > 1

15. 31 Periodic solutions; perturbation technique.
Example (F) from section 15. 12
(A)
Here we have already transfonued t into 1:
co t = 1: (d. (H) in sect. 15.22).

Let us look for 21t-periodic functions Cl 1 (1: ) fo) , d2 (1: ) fl)
and an (U(~) which for fo - 0 tend towards
a 1 ("t,))) --- A (OS 1:
a2 ('t ,~ ) -- 0 (B)
(0 (Jl) --- 00 1
(That corresponds to the first normal mode, d. (A), (C) in
sect. 15. 11).
Introducing
(C)
into (A) we obtain
I
Wi
2. .. (0)
a1 + u/.a
1
(0)
1 =0
generating equations (D)
2. .. (Oi 2.
=0
(0)
00 1 a2. + 9(U1 aZ
(Note that w 2 = 3 Wi).

Because of (B) the (tentative) generating solutions of (D) are
}
a (0)
1 = COS 1:
(E)
(All solutions A2. c.o~ 3't + B2 ~i.n 3't of the sc cond eC"ju<ltion
(D) are 2TC-periodic. By the restriction (B) we eliminated
this arbitrariness).
Taking (E) into account we get from (A) and
(C) the following equations for a (1)

1 ,
Higher Order Approximation 147
(F)
(G)
In (F) resonance can be avoided by choosing
')C~ = .i
4
f'
III (G) internal resonance ~ not be prevented. The solutions
(E) , and the assumptions (B), do not lead to practicable equa-
lions.
We replace (E) by
a,(0), = A c.o!> "t

A=A" } (H)
Bya proper choice of A2 (A)and "1 (A) resonance can be p:-e-
vented in both equations.

General procedure see Malkin's book cited above.
We see:
1. In general, the generating solution can not be chosen arbi-
trarily.
2. A nonlinear periodic solution u.p ,here apr>roximated by U,A,
does not, in general, branch off from a linear normal more..

In Fig. 15. 3 such a behavior is shown qualitatively.
Fig. 15.3
"Reasonable" behavior occu rs, (B) is pos sible etc., if the

natural frequencies are not commensurable.
15. 32 Non -pc riodic solutions; some rem a rks
Let us assume that we know initial conditiolS
u.(X,O) = ",(X),
(1)
u. X, 0 = '" (X)
for the problem given in section 15. 1. Similar as in section
5.5, we can derive initial conditions from (I) for the coordi-
nate functions an (t) :
} n ' i , ... ,N. (J)
The equation (E) from section 15.12,
a+
N
C a + j3f(a) = 0 )
1'01 .....
(K)
together with (J) establish an initial value problem for a
system of ordinary differential equations.
Such a problem can be solved numerically for

fini te time in te rval s . If fJ is a small parameter, pertur-
bation techniques can yield results (assuming that we are only
interested in a finite time interval! even secular terms would
not matter). Some special cases may be investigated by asymp-
totic methods (slowly varying phase and amplitude)which we
shall discuss briefly in section 16; see N. N. Bogoliubov, Y. A.
Mitropolsky, Asymptotic methods in the theory of nonlinear
oscillations, Gordon & Breach, New York 1961; Y. A. Mitro-
polsky, B. 1. Moseenkov, Lectures on the application of asymp-

totic methods to the solution of partial differential equations (in
150 Nonlinear Nonautonomous Systems
Rus sia.n), Ukrainian Academy of Sciences, Kiev 1968.
15. 4 Numerical computation of periodic solutions
High order approximations for periodic so-
lutions of nonlinear differential equations can be calculated on-
ly by numerical methods. M. Urabe and A. Reiter (Numerical
computation of nonlinear forced oscillations by Galerkin's pro-
cedure, J. Math. Anal. Appl. 11- (l966) 107-140)describe a
procedure which is applicable to ordinary differential equa -
tions. Here we shall extend this procedure to partial differ-
entials equ;].tions.
Periodic solutions of (E),sect. 15.12, can
be found directly by Urabe's method (the restriction to forced
'librations in the paper cited above is not essential, see below).
But in general it is very cumbersome to establish the set of
equations (E). So lets sta rt right from the original diffe rentiil
equation (*).
15.41 Description of the method
Fi 1'5t we introduce wt = 1; into (*) and ob-
Lain
Numerical Fourier Approximation 151
(A)
(Afte r the time transformation we write t instead of 1;
We are looking for a 2 Jt -periodic solution of

(A),
u eX , t + 2'It) = u (X , t) . (B)
Again we start with
(c)
d. (C) in section 15.12 (here we drop the subscript A). The
Un (X) are the natural modes of the linear problem, d.

section 15.11, or some other appropriate linearly independent
functions which satisfy individually the boundary conditions(**).
ft is a numerically given parameter.

Here the an (t) are assumed to be 2'Jt-perio-
dic. We shall represent them by a truncated Fourier series.
For simplicity let us assume that the an (t) have the form
an(t) = ~anmco~(Zm-1)t ) n=1, ... )N. (D)

m: ~
We introduce
(E)
and
U. k (X ) t) = Un (X) CO~ (Z m- i) t . (F)

u. , d. (C)is represented by
N·H
U. = ~ bkLLk= Sb (G)
k=1
where the symbol ~ stands for "Fourie r synthesis 11 and b is
the vector
The ope rator is linear,
8 C.~1 + Et) = .s.e1 + 8,92

The gradient of u. with respect to ~ , denoted by U b '
has the form
u'b = ~! , (H)
where I denotes the (N M) x (N M) identity matrix.
We define the inner product
JJ
1 211
r k= (u. ) U. k) = u. U. II. cl X d t) k = 1 I ••• ) NM. (I)

o 0
rK - (generalized) Fourier coefficient.

By
r = .Au. (J)
·,'.'e symbolize the generalized "Fourier analysis". The operat-
or .A. is linear.
.A ( u. ... v) = .Au ... .A v . (K)
In the forms given here 8 and are not inverse oper-
ators,
because some factors in (I) are not chosen properly. {These
f<lctors depend on the Un' a proper choice could be intro-

duced but it is not essential for our investigations}.
Let us assume that we know a set of numbers b

which represent an approximate solution u. of ]) u. . In
gene ral D 8 b will not vanish identi cally. We requi re
(L)
The residunnt vector r is a nonlinear function of b
The vector equation
(M)
represents a set of NM nonlinear scalar equations for the ele-
ments bk ) k = i , ... I NM, of b . (We shall show how to

oS olve the equations in the next section)
For numerically given values b the resi-

'"
duum vector !: (E) can be obtained by numerical methods;
u. and r have to be computed corresponding to (G}and
(1). (Attention: For the numerical integration the subdivisions

of the X -inte rval and of the t -inte rval mu st be small enough.
In general a fraction of 1/ Nand 2 'Jt / M , say 1/ 3 ,
is necessary).
15.42 Solution of the eguations
We apply Newton's method to improve the

chosen b Let A~ denote the correction. From
'"
we obtain
(N)
and by linea rizing with respect to

we get
(0)
! (£) is the re siduum vector calculated for the chosen b ;

'"
Du. denotes the (Frechet) derivative (linear variation) of D,
for instanceJin the case (*) :
(p)
f is the rm trix
(Q)
F can be calculated nurne rically corresponding to equa-
Lion (Q). In gene ral, ,...,F depends on b d. the te rrns
lL = Sb in , equation (P).
For cLet E :F a we obtain from (o)
(R)
ilnd the improved Fourier coefficients
b * ... b+L\b.
'"
Iterative application of this procedure leads (within the limits
.
v f llumerlca I accuracy )t 0 a so It'
u Ion p*·of(L)I·fdet-:"O
~ .~ .,.
'" the neighbourhood of u. = 5 b· * and the initial vector b

is chosen close enoug h to b **
The re will be one diffi culty
we have to overcome. In general, the Fou-
'-ier coefficients will change very rapidly
.. hen the frcCluencv W is chan.u:ed (d. Fig. 15 . ..f

Fig. 15.4 and Fig. 15.2. ( W is parameter in our calcula -
tions). Very frequently the iteration procedure outlined ab.)ve
will converge to b·' = 0 , the trivial solution. This can

'"
be prevented by interchanging W (or W 2) and one of the
elements b k ' say b k' , in Newton's iteration process
b k. is introduced as parameter and used to calculate £(2) as

before, d. equation (L). (A tentative value 00 is estimated
and introduced into that computation). For Newton's procedure
we have A b k" = 0 and A. W (or A. (0 2 ) is put into the place

of A. b k. in A E F is appropriately modified. By (R)
we obtain the corrections and Aw

(or L\ w 2). Since bk.:F 0 this procedure can not converge to
the trivial solution b

""'
=0
In general,it is advisable to start the compu-
tation from parameters ~o,( b k.) 0 ' say, for which good
approltimate values bk , W can be guessed,and to apply
b ••
'" 0 ) W 0•• ,the numerically obtained solutions for p,. 0 ,
J<J
( b k *)0 ' as starting approximations for ~ 1 , (b k .. ) 1

which differ slightly from jl 0,( bk*)o ; etc.
15.43 Concluding rema rks
In general,in equation (D) a more complete

Damped Vibrations 157
Fourier series has to be taken into account. The computational
work can be decreased considerably if the formal calculations
included in are done beforehand analytically.
Known special characteristics of the solutions should be taken
into account.
The expression J+ D g ~ in equation (L) can be
replaced by the general form of Hamilton's principle, see
equation (H) in section 9. (The au. would be the functions
u. k ' cf. equ. (F), for the example discussed above). Ha-
milton I s principle has the advantage that nonlinear dynamic
boundary conditions are included. Nonlinear geometric bound-
ary conditions may be considered as constraints which can be
treated, in connection with Hamilton I s principle, by the methxl.
of Lagrangian multipliers. In that case the multiplier A. (t) ,

say. and the constraint must be expanded into Fourier series,
too. Similar considerations hold for distributed constraints.
(In connection with perturbation techniques nonlinear boundary
conditions can be expanded into series with respect to the
small parameters; d. Mitropolsky. Moseenkov cited in sec-
tion 15.32).
15. 5 Damped nonlinear vibrations
The equations of motion can be reduced to a

svstem of ordinary differential equations as outlined in sec-
tion 15.12. (Nonlinear boundary condition can frequ(nt...y trea
ed by some special tricks). If the system is dissipative (pas
sive), the oscillations will decay, no periodic solution will
exist. Small parameter expansions, especially, the asymptotic
methods of Krylov, Bogoliubov, Mitropolsky are applicable.

If the system is not dissipative (if it is ac-
tive), periodic solutions (may) exist. For instance, a rotating

shaft with internal friction can oscillate (periodically) beyond
the stable parameter regions if nonlinearities are taken into
account. Those periodic motions C;!ll be calculated numerical
ly as described in section 15.4. Tr,lllsicnts m.ight be investi-

gatcd by asymptotic methods.
15. 6 Stability
The (periodic) solutions of a problem (obtClD
('d by one of tlw nll'thods described ;!bove) need not be stable.
The stability corresponding to LY<l]mnov's definitions, d. sec
lion 12. has to be inv('stigatcd separately. Tn most cases it
will be inlpossible to find a Lyapunov's function. Approximate
illvestigations by n1l':IIlS of a (line;! ri"cd) va rii1tional cquation
of the co r rc S pOllding 0 rdin;t ry diffe rential eqllation, d. (E) in
scdioll 15.12 .. Vi;1 FloCjuct solutions, cf. s('ction 14.2, arc

Stability 159
very cumbersome. Again asymptotic methods are applicable.

16. Nonlinear vibrations of nonautonomous systems

16. I Nonlinear transversal vibrations of a rock
We assume an uniform
Eule r -Be rnoulli beam

P (X It) (originally plane cross-
~ljj~
,~/;;; ~
sectional areas remain
plain) and a nonlinear

Fig, 16,1
stress-strain relation
(j = E£ (1 +C!t,2)- consC'l"vdtive
(j' - "fl'(' S S
c -, s L l' <l i J1
i\Her sorne calculdtion we obtain tlw equation of lnotion, In
nondirnensionaJ V:I riables,
]J u. = i.i +
ru.IV lJ., II.+ Z U. Iii 2iJ Ll II
u. IV +)J(l,. L - P ( X I t) = 0
(d. r<:cndererls book cited ill :-;edion l S. ]). TIlt' houndary COli"
ditions ,Ire, d. Fig. 16.1,

Reduction to Ordinary Differential Equations 161
16. 11 Reduction of (.) to a system of ordinary differential
equations
Again we can solve the corresponding linearize(l
(homogeneous) problem and obtain the eigenfunctions Un (X)

and the natural frequencies W n , d. section 15. 11. Since we
have linear boundary conditions an approximate solution of the
form
N
= ~ anCt) Un(X) (A)

n:1
satisfies those boundary conditions and by Galerkin's method
we find
in vector notation:
~ + S~ + j) f (~) - ~ (t) = 0 . (c)
Except for the forcing terms qn (t) these equations coincide

with the equations (D) and (E) from section 15.12. In general,
it may happen that the (nonlinear) function f will depend
on t explicitly,
~ +f~ +j)£(~lt) - 9,(t) = 0 > (D)

for instance in the case of the parametrically excited systems
discussed in section 14 ( f
~
may include linear terms a ).
~
16. 2 First order approximation N =1
A first order approximation
u. = a (t) s~n 'ltX

'-or our special example leads to a _Duffing equation with a
forcing term:
a + a + fia 3 = q (t) (E)
a , t , q are multiplied by appropriate constants to

simplify the form of (E)).
No gene ral solutions of (E) are available.

Naturally, "solutions·· a (t) can be chosen and the corresporrl-
ing q (t) can be calculated from (E). Sometimes q (t) =
= ta(t)+ 71a~(t) is.d.ntroducedinto(E),where ~ ,11 are
constants. This case can be reduced to the autonomous pro-
blem, see section 15. I, and solved by means of elliptic func-
tions. (In fact this problem is autonomous and does not show
the phenomena which occur for IIrealistic ll forcing functions
q (t) )). (E) can be solved numerically for finite time
inte'rvals. Approximate procedures, series expansions

One Term Fourier Approximation 163
etc. are possible.
Of special interest are periodic solutions, aCt),

for periodic forcing functions, q (t). We discuss the widely
treated case q(t) = qo co~ Qt :
Ea == a + 0a + a + ,r., a ~ - q0 c.o~ .n t = 0 (F)
(To show some phenomena we include a linear damping term
in (F), 8 -coefficient). Differential equations which are pe-
riodic with respect to the explicitely contained time are call-
ed "periodic differential equations ll •
16.21 One term Fourier approximation
As a rough approximate solution of (F) we choose
a = A (O~(nt - 'f)
By Fourier expansion we obtain
(h. o. t. -higher order terms).
Omitting the h. o. t. we get fr011.1 F15 = 0 , F 1C = 0
tan <p =
no
q~ (G)
A'
A
(backbone turve)
6 >0
un!:ttab\e
Fig. 16.2
Fig. 16.2 shows the response curves. We see
1. No resonance (infinite A).
2. Two stable solutions in certain n regions.
3. Jump phenomena in the vicinity of n = 1 when n is
increased or decreased slowly.
For some more careful investigations on
jump phenomena and for response curves of other differen-
tial equations see Bogoliubov, Mitropolsky's book cited in sec-

tion 15. 32.
The following investigations will show us

30me of the shortcomings of the one term Fourier approxima-
tion.
Perturbation Techniques 165
16. 22 Series expansion; perturbation techniques for periodic

equations
For simplicity we assume a =0
(H)
" is a small parameter; (H) is a quasi linear differential

equation.
16.221 Non-resonance case, I Q - 1 I 4= 1
Series expansion with respect to j)
leads to
,,0 : a (0) + a (0) = q 0 cos n. t

A.1 •• ··(1)
1.1 a + a
(1)
= _Jl (a (0)) 3 •
.
Solutions:
Clo
1_.Q2.
cos nt
a (0)
We see: The series ex-
I~.B=O (\~near)
I pansion shows resonanc~:
-~ phenomena at
at
n =1
Q = 1/3. (Higher
and
1/3 1 Jl=O _ {}
~.; order approximations
,I~ show resonance for
n = 1 /n J n=1,3,5, ... ).
Fig. 16.3
16.222 Resonance case; In -1 I <:< 1
An expansion similar to equations (1), (J) in

section 16. 221 is impos sible. At least the term qo co~ n t must
be shifted into the second equation of (J): To achieve that we
assume jj > 0 and introduce
a = a/ f
into the equation (H), where
We obtain
-3
:.!
a +
-
a + ca = f.qo c.o~Qt (K)
A.n expansion simi la r to (T) leads to

Resonance Case 167
t- 0 :
:..!
a
(0)
+ a
- (0)
=0
.
A ssumlng -a(O) -_ A c.o~'" t as generating solution we car
not prevent resonance in the second equation. To avoid this
difficulty we apply Lindstedt's idea. We introduce 't = Q t

into (K) and obtain
Q 2~ + a + E a3 = f.. q 0 c.o~ 't . (L)
( -a·· = d,2 a- / d, ....." 2.). We are looking for 2 'Jt -periodic solutions
a ( 't + 2 Jt )= a ('t). A series expansion, d. (J) in section 15.22,
Q2= 1+X1t+'X2t.2+ ••. )

a = a-(0) + ca-(1) + t. 2-(2)
a + .•.
leads to
t. 0
:..!(o) -(0)
: a + a = 0 (M)
-(0»)3 :..!(O)}
- (a
:..!(1) -(1)
C1 : a + a = q0 c.o~ 1; -')(~a
As generating solution we choose

-a (0) = A c.oc;, 't .
For the higher order terms,

k > 0 , we as sume that the
;a(k) ('t) do not contain terms co~ 1: , ~in 't . (The term
- (0)
A co~ 't in a represents the cOlnplete first Fourier
term of a (1: )). A different choice might be

:.! (0)
Note that every equation k > 0 of (M) contains a term 'Kk a .
Thus, resonance can be prevented by a proper choice of "k .
We obtain from (M)
a (OJ = A co!' 1: ,
a (1) =K
32
co~ 3 't , ...
"1 = i.
4
A2 _ qo } ...
A
1/-
A f, - con!»t •
0 backbone
c.urve
- ,," Fig. 16. 4 shows
-- ....,...~",""
Q2(l.,qo,A) ,
cf. Fig. 16. 2
Fig. 16.4
If we characterize the 211: -periodic (odd harm2.

nic) solutions a ('t) C. , Q , q c) of (L) by their initial values
a (0 , f, , Q , '10)) a(0, t ) Q) '10) = 0 ) we obtain by the pro-
cedure outlined abovE': a parameter representation:
- A~
a (0 I£.' A) = A + c 32 +... ,
Q 2 (f. , qo ) A) = 1 + c [~ A2 - ~ J + ...
where A is the parameter.

Superharmonic Resonance 169
16.223 Superharmonic resonance; 1nO. -11« 1, n = 3,5 ....
Similar expansions as in section 16. 222 can be

used to investigate the solutions a (t)of (H) in the vicinity of
o.=1/n,n=-3,5,•••• In the neighbourhood of n = 1/3, say, the
Fourier series for a (t) contains a very large term A3 co~ 30 t
which may dominate all the other terms of the Fourier series.
Therefore such solutions are frequently looked upon as "supe.!:,
harmonic" response of the system (the period of the response,
T re~p = 21t / ~O, is a fraction of the period of the forcing
function, Tforc . = 21t/O).
But a differential equation of the form
F (a:. a ,do) = q (t) (N)
where F is an analytic function of its arguments and
q (t + Tfore) = q ( t) , T fore - least period,

can not possess solutions a(t)which have a period, T relop.)
a (t + T resp.) = a ( t) )
(0)
wllich is less than Tresp.
o < Tre!>p. < T fore. )

since substitution of (0) into (N) leads to a contradiction.
Example for a differentia~ equation which has a superharmonic
solution:
has the solutions
a = A co~ (3 Q t + <p)
A I <p arbitrary.
However, (N) may have solutions for which
T resp. = n T fore. ' n-integer. We sh~ll construct an example
for such a "subharrnonic" solution in the next section.
16.224 Subharmonic solutions
We start from the equation (H) <lnd introduc('
Qt = 1:
Q 2 a +a +jla 3 = qo c.o~ L ) a = d. 2a /dL 2. (P)

Now, we expand Q2 and a with respect to f>
Q2.=Q;+X1ftl +?<2fi 2 + .•. )
a = a(O)+fia(1) +jl2 a (2) + } (Q)
Putting (Q) 111 (P) we obtain
j) 0 Q2 • '(0) a (0)
: ~oa + = qo co~ L
a(1 ) = _ (a(0)')3_ .. (0) } (R)

,I) 1 : Q2 "(1)
.0 a + x~ a
Subharmonic Solutions 171
We choose the generating solution
a(o) = qo co~ 1: + A co!> 1

"C.
1-0~ 0 0
For 0 0 =2 we get
a(O)= - ~ co~ 1: + A co~ ..!..
3 2 .
liN 0 re s onance 11 In the se cond equation of (R) leads to
" 1
= 3 A2 + 1...
3
(12
'0'
Fig. 16.5 shows the corresponding response curves. At
a (0) = - q 0 / 3 + ." and Q2 = 4 + 2 fi q ~ / 3 + ••

a 41t -periodic solution of (P) - subharmonic solution of the
second order - branches off frorn the original 21t -periodic

solution. Higher order expansions show that "super-subhar-
moni c " solutions branch off from the original 2 'It -periodic
solution in the vicinity of all rational values Q 0 = n 1m,
n ,m -integers. If damping is included most of these (but
not all!) bifurcation points disappear. (That need not mean
that the corresponding "branched off" solutions must disap-
pear, the solutions may just separate from each other). It
can be proved that (F) possesses an infinite number of period
ic solutions, but if damping is taken into account there exists
only a finite number of periodic solutions.

a(O)
backbone curve.
A=O A:f:O
oint (A =0)
Fig. 16. '"
16.23 Some periodic solutions obtained numerically for Duf-
fing I S equation
Fig. 16. 6 shows some initial values
a (O),(.i(O)=O),for 21t/O-periodic solutions of
a + a + }!la 3 = q 0 c.o~ Q t
[or 1> = 1 and qo = 0,2 . At Q = 1 / n I n = 3 , 5) ... )
we find the resonances predicted by the series expansion, d.
section 16.221. In the vicinity of Q =1/n, n=2,4, ... , the
response curves have unstable regions. (High order series
expansions would be necessary to show this phenomenon ana-
lytically). At the points Band C, see Fig. 16.6,

Numerical Results 173
~.------------------------r--------------------~ __
a(O)
1+-----~--~------------~---,~----------------~
backbone
O,S curve (qo=O)
"rough" appro
o +-~~~~--~----------~-----------+----------~
Q 1 2
\ l weakly) stable
solution
-o,S
\ - - - unstable sol.
'\ V unstable region

-1 +-----~--~r_----------~----~----------------~
__ 0-- ____
0-0--
-2
..............
_ ~~.______________________- L________________. ___
- - -~.~
-
Fig. 16.6
21t/O periodic solutions aI(t) I an(t), am(t), a!! (t)

branch off. These solutions have stable regions. a I) a li ,which
branch off from B satisfy an(t) = - aI(t + tt/O)lthey are evei
with respect to t = 0 and t = 1(/ Q . am ,aN branch of from C,
satisfy aN (t) = - am (t+ 1C /Q), and are odd with respect to
t = 1C /ZQ and t = 3 Jt /20. . (It is possible that there exist
more unstable regions).
174 Nonlinear Nonautonomous System s
If (a smaU)damping is taken into account we

have to represent the periodic solutions by a(O)and a(O) =FO .
In Fig. 16.7 such a response curve is sketched qualitatively
( a = 2.10- 4 , P, = 1 ) qo = 1 ). There are no "holes" at the

l'"esonance frequencies but "loops". But the unstable region
a(O)
a (0)
Fig. Ib.7
lI1 the vicinity of Q = 1/2 is preserved. Band C arc still bi-
furcation point.s. (There 111ay exist SOlne n10rc very I1il.rrow
Illlst.ah1e regions).
16.24 Ahnost pCI'iodie eqllations
T" s('ction l(). 2.2 we investi~<It.('cl Twriodic

Almost Periodic Equations 175
forced vibrations, d. equations (E) and (F). Frequently, the
forcing functions are not periodic but can be represented by a
finite or infinite sum of Fourier terms, e. g. ,
(5 )
If the ratio
-=
0>2 m
is rational ( n , m are relatively prime integers), then (5) is
periodic; the least period of q(t) is
T = 211: n = 2'lt m
U)1 U)2.
If 00 1 /00 2 is irrational.(5) represents an almost periodic
function.
16.241 Almost periodic functions (a. p. f. )
Harald Bohr defined:

Ld q(t) be a continuous function defined for all values t from
-oo<t<oo Then q (t) is called an a. p. f. if for any arbi
trarily small positive number t a positive number 1. (c)
C<1n be found such that within each intervall of the length t at
least one number "t(L) can be found for which for all t the
inequality
I q(t +1;) - q(t)\ < c
is satisfied.
Some basic properties of a. p. f.
1. Every a. p. f. is bounded on - Q) < t < 00
2. Every a. p. f. is uniformly continuous on - 00 <t < 00 •
3. A finite sum of a. p. £. is an a. p. f.
4. The product of two a. p. f. is an a. p. f.
5. If g(t) is ana.p.f., andifs(t» a for -oo<t<oo, then 1/g.

is an a. p. f.
6. The limit ~(t) of a uniformly converging sequence of a. p. f.
~1(t)) ~2(t)) ... , is an a.p.£.
A periodic function is a special case of an

a. p. f. Thus, the finite sum
J
'l(t) = Ao + ~(A~ co~ w~t +BJ.5in U)~t)
~:1
is an a. p. f. for arbitrary (real) numbers CO k •
Any a. p. f. can be expanded into a (gene r<ll-
i zed) F au ri e r s e ri c s
QQ
q(t) = Ao+ ~(Atco~w}t+B~si.nU)~t). (T)

p1
(d. A. S. Desicovich, Almost periodic functions. Cambrid,ge

1938; reprinted by Dover, New York).
Almost Periodic Functions 177
Let Q (t 1 , tz )' .. , t K) be 2 'It -periodic with

respect to each of its arguments,
a. (t 1, t 2J ... ) t K)= Q(t 1 +21t,t 2 , ••• ,t k )= Q(t1,tZ +21t, ... t\l.)= ....
The special a. p. f.
(u)
is a quasi periodic function. The (real) numbers "k , k = i ) ... ,K)
are called basis frequencies. The w ~ in the Fourier ex-
pansion (T) of (U) are linear combinations of the V k '
oo} = mj.1"1 + mp"2 + ••. + mkk"Y.)

with integer numbers m t Ie. (In general, an a. p. f. has an
infinite basis).
16.242 Break down of the perturbation technique for a. p.

differential equations
The differential equation (D), section 16.11, is
called almost periodic if i (~ ,t) and/or 9,(t) are almost

periodic with respect to t
We try to find an a. p. solution of the special
quasiperiodic differential equation
(v)
by a series expansion with respect to f.»

(d. Malkin's book cited in section 15.22, and N. Minorsky,

l~onlinear oscillations, Van Nostrand, New York 1962).
Introducing
a - a(O)+ j'la(~) + fi2. a(,) +
into (V) we obtain
a (0) + a(o)= 0
a(1) + a(1)= q1 :,tn t + q, ~~n oot _ (a(O)) 3

~(2) + a(Z)= _ 3a(1) (a(O))Z
As generating solution we choose

a (0) = "Bo ~Ln t
To avoid resonance in a (1) we requi re
We obtain
B~ ~Ln
a(1)= - 32 . 2.t + 1 q2 2 ~~n cut + B 1 ~in t.
~
-(0
T o aVOI.d resonance In
. (2.)
a B1 = - B ~ / 48 is chosen, we
obtain
a'2.): _ 3 B~q1 ~Ln '3t + 3 &~q1 sin 5 t _ 3 B~q2 !>>In wt

1024 2048 2(1-002)2
3B~q2sin(w-2)t
+ ------~~------~-
2 (1 - 00 2) ~ - (w - 2) '2.]
,,·te.
This is a formal series expansion. Higher

a (k) will contain terms
Succes sive Approximation 179
SlLn (moo - nt)

[1 - (moo - n)2]
Since (U is irrational it may happen that 11 - (moo - n)21 « 1
for large integers n J m . Thus, it may happen that the higher
terms contain "small c:ivisors", the corresponding a(k) be-
come very large, the series for aCt) does not converge.
Special expansion techniques, due to Krylov and
Bogoliubov, are described in the books of Malkin and Mitro-
polsky.
16.243 Solution of a. p. differential equations by the method
of successive approximations
16.2431 Linear a. p. differential equations
The linear a. p. differential equation
a + oa + a = q (t) , 0<0<1/2,
where q (t) is an a. p. f., has the solution
t
a(t)= ~I e-S(t-0')/2:,~n'\l(t_0") q(<r)dO', (W)
-00
The right hand side of (W) is an a. p. f. :

la(t+1:)-a(t)1 =
I ~ Je -6(t+1:-o-)/2 5ln vet +1: -0-) q (o-)dO" - J

t+'t t
~ e -cS(t-o-)/2 sLnv Ct-o-)9(O-)d<r!
_~ -00
t
= ~ IJ[q(O"+'t) - q(O")] e-6(t-0')/2s~nv(t-(j")d(j"1 < f.
-00
if Iq(O"+1:)-q(<r)I<Zf.v/O.
Furthermore,
I a (t) I < 2 M/ ov ,
where
M = ma~ Iq(t) I
t
1 C. 24 '32 N onhnca r a. p. diffe rentia1 equations
Quasi linea r diffe rential equations:
a +oa+a +j)f(a,t)= q(t),
where 0<0 < 1/2 ) q (t) is an a. p. f. The function f (a. t) i

almost periodic with respect to t (uniformly with respect
to a ), furthermore Ha,t) satisfies a Lipschitz condition
with respect to a
If(a 1 ,t)-f(a z1 t)l< Lla 1 -a z l,

L does not depend on t
Successive approximations:
t
a(O) = ~f e- S (t-0')/2 s Ln'V(t-0')q(cr)dO" I
-00
t
a(k+1)= ~f e- i5 (t-O')/z si.n 'Vet - O"} [q(o-)- JH (a(k\cr) ,cr)] dO" )
-aD
k=O,1, ...•
All a(\<)(t) are a.p.£. The a(k) converge to an a.p. solution,

a (t) ,if j) is small enough.
16. 3 Higher order approximations N >1
The methods outlined in 16. 2 for quasilinear
periodic and almost periodic differential equations are appli-

cable to systems of such equations, too. In many problems
difficulties arise from internal resonances and small divisors.
) 6. 4 Numerical computation of special solutions
Pe riodic solutions of periodic equations of motion
can be calculated by Urabe 1 s version of Galerkin 1 s procedure
as discussed in section 15.4.

16. 5 Stability
'The stability of the periodic, almost periodic,
etc. solutions has to be investigated. For periodic solutions

the procedure mentioned in section 15. 6 is applicable. The
stability of almost periodic solutions of quasilinear differential
equations can be investigated by analytic methods (see Mal -
kin's book cited above).
Slowly Varying Amplitudes 183
17. Asymptotic methods
In this section we shall discuss some methods,

originally used in celestial mechanics, which were applied to
special technical problems first by van der Pol. Krylov and

Bogoliubov, and many others, extended and generalized these
methods. (See the survey article by Y. A. Mitropolsky, Aver
aging method in nonlinear mechanics, Internat. J. of Nonlin.
Mech. f. (1967) 69 - 95).
17. I The method of the slowly varying amplitudes
We resume the example of the hinged-hinged bar which is load

ed axially by a pulsating force, d. section 14. l, but this time
we take some nonlinearities into account. (This problem was
solved by F. Weidenhammer, Das Stabilitatsverhalten der
nichtlinearen Biegeschwingungen des axial pulsiernd belaste-
ten Stabes, lng. -Arch. 24 (1956) 53-68).
J 7. 1 i Equation of Motio n
Uniform EllIe r-Be rn~)il11i beam; ,u.) EI,A-constant (A- cross - se_c
tional area). The longitudinal extension is taken into account.
184 Asymptotic Methods
The strain of the cen-

x ter line of the beam is
approximated by (d.
Kauderer's book cited
Fig. 17. 1 insect.15.1):
Co = Vi + 2 u. 1' + 12 12
U 1 + U.2 -
1 ~ u. 11 of- 2"
i 12
U,2
The longitudinal strain of an element which has the distance Y
from the centerline is assumed to be
tx - t.o - YU. 2 \I
•
Introducing Hooke's law, neglecting the inertia terms iii> ap-
proximating the longitudinal mo'tion u. i (X, t) bel X U1 (t) ,

which is allowed for excitations far below the first longitudinal
natural freqency, and taking one term LL 2 (X,t)= a(t)SLn7tX/ t
into account, Weidenhammer obtains a single nonlinear second
order ordinary differential equation for aCt):
(A)
(A) is wri tten in a nondimensional form. The fi rst natu ral fre-
quency is normalized to 1, to is the (nondimensional) frequency
of the pulsating force, B a represents a linear external damping,

Pi is proportional to P1 (see Fig. 17.1).
Assuming () and P1 to be proportional to a
small pa rameter t , t () >6 , t Pi ~ Pi ,
Van der Pol's Method 185
and introducing a 'I{i for a (t) we find from (A)
W
2a.. + a = t
( .
- ocoa + Pi CO!! t a - a 3) (B)
17.12 Van der Pol's method
(B) has the trivial solution a (t) =0 . In sec-

tion 14.12 we saw that a (t) !II 0 becomes unstable in the
vicinity of w = l (= 2 001 !:Ii.nee Wi = 1) .
In that neighborhood we expect to find non-trivial periodic so-
lutions of (B) which have the period 4'It. Correspondingly we

write
aCt) = ACt) co~ t/2 + B(t)~i.n t/2 ( C)
or
aCt) = a(t) cos (t/2. + <pet)) (D)
(C) and (D) are equivalent:
A = Q (,05 <p B =- Q ~tn cp • (E)
A(t)and B(t) are introduced as time dependent func
tions to take into account the deviation of a(t)from a sinusoidal
motion ( Q Ct) and <pCt) might serve for the same purpose).
Hopefully, the te rms co~ t /2 and sin t /2 will catch the" fast!'
parts of the motion and the amplitudes A (t), B(t) will vary slowly
(in CD) the amplitude Qet) and the phase <p(t)would vary slowly).
Since we introduce by (C) two arbitrary
functions into (B) we are allowed to choose one arbitrary addi-
tiona1 condition. We assume
Aco~ t/2 +:B cos t /2. = a (F)
Because of (F) we obtain from (C)
a = - A/2 si.n t /2 + B/2 co~ t /2 )

a = - A/ 2 SLn t / z+ B /2 cos t /2 (G)
- A/ 4 cos t / 2 - B /4 5 i.n t /2 .
St 1 bstituting (G) into (B) and solving the resulting equation. to-
. .
geLher with (F), for A and B we obtain two first order differen-
tial equations for A (t) and B(t):
~z A = _A (~2 -1) cos ~ si.n i _ (~2

B -1) sin' ~
- c r (A, B , t) si.n ~ = RA (A , ~ ,t) I
j
(H)
-w B
2
2
• = A ( -cu~
4
- 1) co~ 2 -t + B
2
( -w 2 -1 ) SLn
4
. -t
2
co~ -t
Z
+cF'(A,B,t) = RsCA,B,t)
where
- (A tO~ i2
+ B~in 1)3
2'
This set of equations is still "exactly" equivalent to the equa-
tion (B).
Since (j) is assumed to be close to 2,lw2./4-11

will be of the order E- ,
Thus, the right hand side of (H) is of the order e , A (t) and
B (t) vary slowly.

Van der Pol's argument: Tl~ rapidly changing
terms on the right hand side of (H), which depend explicitely

on the time, do not contribute much to A and B. By taking the
"average" of those terms over their period 4 'It ,
41t
CJ)
2
A/2 = 41Tt~RA(A,B,t) dt
41t
A, B -" c.on~tant " , (I)
uJ'B/2 = 41'Jt ~RB(A B,t) I dt
we obtain the set of two first order nonlinear autonomous differ
ential equations
(J)
From (J) the time can be eliminated:
dA B [1 - CO '/ 4 + t Pi / 2 + 3 t (A 2 + 'B?) / 4] - l, {j 00 AI 2
-::a (K)
ciB -taooB/2 -A[1-oo,/4 - tP1/2 + 3e(A'+~2)/4J
(K) is a first order differential equation for A(B). It describes

a directional field in an A-, B- plane ("phase plane " ; cf. e. g.,
Bogoliubov, Mitropolsky' s book cited in section 15. 32) .
17. 13 Stationary solutions
Stationary solutions, A = Ao-constant, B = ~o

constant, are obtained from (J) by assuming A = 0 , B= 0 ~
Vanishing damping: We get three different sets of solutions:
1: Ao = o, ~o= 0, aCt):: 0
2: Bo = o, A~- -4[1-u//4-tP1!2] /3f., aCt):: ±Ao,o~t
3: Ao = o , B~ = - 4[1 - (l)'/ 4 - t P1/Z] I 3t a(t) = ± B0 5i n
I t
Non-vanishing damping:
Ao= 0 I Bo= 0 is again a solution. For Ao J Bo =1= 0

we obtain
In Fig. 17. 2 some results are represented
unstable region
from Fig.14.3
J
//
stable //
8>0//
unstable I //
V///
IV /
-+-------
bifuration
points ' "
,,I'
I~/
.L""r_____
2
Fig. 17.2
for a fixed value L p 1 . In the upper half of that figure the unsta-
190 Asym.ptotic Methods
ble region from Fig.14. 3 is sketched, below some response
cn rves are drawn. (The stationary solutions Ao, Bo are the si:,
Q'ular points of the equation (K). The stability of these solutio.1.s
depends on the character of the singular points. )
17.14 Curves A(B) in an A-, B- plane
The behavior of the system under investiga
tion for non-constant A, B can be represented by curves A(B)
-nhase-curves-in an A-, B- plane. These curves can be oe-
tained from equation (K) by integration. After A(B) has been
found, B (t) can be gained from the second equation (J) by a

quad rature.
17.141 Phase-curves for 0 = 0
Fo r 0 = a the equation (K) can be solve(~
explicitely. We obtain the phase-curves shown in Fig. 17.3-
17.6 (d. Wejdenharnrner's paper cited ahove). co<D' etc. see

Fig. 17.2.
A A
Ao
s~para~rix
B B
-Ail
Fig. 17. 3
Fig. 17.4
A A
separatrices
separatrix
8 8
Fig. 17. 5 Fig. 17. 6

The closed phase-curves (that are not separatrices) represent
periodic solutions A(t), 'B(t)with a period T different from 4n.
Thus, in general, the approximate solutions of the form (C)

are quasi periodic.
17. 142 Phase-curves for 8 > 0
For 8 > 0 the equation (K) must be inte-
grated approximately, e. g., by the methods of isoclines.

Figs. 17. 7 and 17.8 show some results of Weidenhammer:
A (&) =: (a)® A OO-OO®
/
/
I
I
\
\
\
9 9
\
,
\
\
I
I
I
/
~eparatricn
Fig. 17.7 transients

Fig. 17.8
By these figures the transients of our system can be studied.

Generalizations 193
17. 2 Generalizations of van der Pol's method
Van der Pol's method was extended into two direc

tions.
First, higher order approximations for (J) were
developed which permit to formulate autonomous equations of
the form (J) up to an accuracy of the order Eo m: The right hand
sides of the nonautonomous equations of the form (H) are expand
ed by special procedures with respect to Eo in such a way that
... + LmFm(A,B) + Lm+1rA(AJ~lt))}

(L)
... + t, III Gm(A "B)
I + t. m+ ~ r ~ (A I B , t ).
Then r A and rf, are neglected; the truncated equations are autox.:
omous. Mostl y, (L) is formulated for slowly varying phases and
amplitudes, d. equation (D).
By "up to an accuracy of the order m" is meant.
that the truncated equations satisfy a relation of the following
form
hm , m - fi.xed. . (M)
c---o
Expansions which satisfy a relation of the form (M) are called
asymptotic expansions. For details see Bogoliubov, Mitropol-

:31,:,;/IS book cited above.
Since the solutions of two first order auton-
:)lnous equations can always be interpreted graphically, as
shown above, some procedures were developed to reduce the
investigation of special (restricted) motions of multiple degree
of freedom systems to the study of the solutions of such diffe r
ential equations.
A second extension of Van der Polls method
went into the direction of multiple degree of freedom systems,
i. e., to a general method of averaging.
17. 3 The method of averaging
(Cf. Bogoliubov, Mitropolskyls book cited

in section 15.32).
17. 31 Standard form
A quasilinear systems of equations which
corresponds to the equation (D) of sectionI6. 11 has the form

Method of Averaging 195
~ + £e = f.g.(~, ~ ,t), (N)

'"
where C is assumed to be a positive diagonal matrix .and e

N
is a small parameter. (N) can be replaced by the first order

system
(0)
where
For f- == 0 the equation (0) has the solution
tD
y=e ~X,
~ N
where ~ = X(0). (Since f is a positive diagonal matrix,

exp (t~) is a quasiperiodic matrix. ) Variation of ~ leads to
~ = ce-t~ b(et~~. t) = cf(~, t).
X = f-f(X, t).
'" 'V
(P)
The form (P) of the equation of motion is called "standard form
i (~ It) is assumed to be quasiperiodic or, at least, al-

most periodic.
17. ~2 Some formal transformations
Let ! (!) t) be almost periodic with re-
spect to t and permit a representation of the form
(The 'V's are real numbers)
We define
T
!o(~) = ~ {f (~, t)} = Hm ~!I (~, t)d.t ) ~ fixed,

T--oo 0
and
We find
a!
() t
= F-
'"
Mt{.t} .
17.33 Averaged equation; first approximation
Applying the transfo rmations 17. 32 to the

equation (P) we obtain
x=
'"
f, fo (X)+ small sinusoidal terms.
'"
We neglect the small oscillating terms and get the autono-

mous equation
(Q)
The solution; (t) is an approximation to the so-

'"
lution ~ (t) of equation (P).
Let X and ~ satisfy the same initial conditions,
'" ,..,
then it can be proved that the error I~ - ~I can be made
arbitrarily small on an arbitrarily large, but finite, time in-
terval if t is chosen small enough and if certain restrictions
are satisfied.
In general, the solutions of the equation (Q) are
more easily obtained and investigated than the solutions of the
original equation (P). Frequently, (Q) has a stationary solu-
tion ~ = ~ 0 • In that case the (linear) variational equation

which may serve to investigate the stability of ~ 0 ' d. sec-
tion 12.5, is 2. differential equation with constant coefficients,

. (R)
L\~ = fo!; (~o) ~~
'"
fo! is the matrix of the partial derivatives(()fo~ / () ~ I< .)
The stability of the trivial solution of (R) can be checked by
the Hurwitz criteria.

17. 34 Order of the first approximation
To show that (Q) is an approximation of
the first order, d. section 17. 2, let us define ~

N
(t) implic-
itly by
x(t)= ~(t) +E.f(t,~) , d. section 17.32. (S)

fV "'" - "V
Differentiating (S) with respect to t , taking the last relation

of section 17. 32 into account, we obtain
Putting this into equation (p) and applying the expansions
and
.... () f ....
f (; +tf
"'.... '" , t) = f (;, t) + f. -=::.. f + ...
........ vi N
we get
By similar transformations higher order approximations can

be obtained. (Attention: the relation (8) does not mean that
I~- 5 I = 0 (t) is valid on an infinite time interval!)

200 Bifurcation Points
18. Investigation of bifurcation points
In sections 16 and 17 we encountered bifur-
cation points. In diagrams which represent the solutions de-
pending on some parameter e , say, there exist special val-

ues of this parameter, e5 ,where solutions coincide which
for e =1= es are of different character. Looked at e = es

from another point of view we might say that at e = e5 a new
solution branches off from the old, known, one; see Fig. 18. 1.
Thus, the investigation
of bifurcation points of a
b·t new known solution may serve
.>. . ~it",.tlon
solu.t~on
to dete ct new, unknown

solutions (new phenome-
point
original na). We shall outline
':Iolution
here a procedure which
is applicable to periodic
e
solutions and is connect-
Fig. 18.1 ed to the nume ri cal pro-
cedure described in section 15.4. In general, these calcula-

nons have to be done partly analytical, partly numerical. Fre
quently, the effor,t needed to do such investigations is tremcn-

Lo cation of Bifurcation Points 201
dous because numerical difficulties arise.
There exists much literature on the investigation
of bifurcation points. We follow mainly E. Schmidt, Ueber die
AufWsung der nichtlinearen Integra1g1eichungen und die Ver-

zweigungen ihrer LClsungen, Math. Annal. 65 (1908) 370-399,
and R. G. Bartle, Singular points of functional equations,
Trans. Amer. Math. Soc. ~ (1953) 366-384.
18. I Location of bifurcation points
In section 15.4 we established the equation
r = r (b)
""'" t"'<J,...,
= J.l.D e3 b = 0
'"
(A)
from whi ch solutions Q = £... were obtained by Newton I s
method. ( ! and bare MN = L -vectors, d. section 15.4)

'"
Let us introduce the parameter e into (A),
!o~ ,e)=O, (B)
and let b = b (e) be a solution of (B)-the asterisks are

'" rV
dropped. By the implicit functions theorem (B) has unique sol,!
tions £ (e) in the neighbo rhood of a fixed value e if the
Jacobian matrix of (B) at that point is nonsingular,
det £ =1= 0) (C)

where
or'"
ICe) =
ab )
15.42 (D)
'"
Thus, we can expect to find bifurcation points only at those

points e = es where F is singular,
'"
d.e t I (e s) = 0 )
the corresponding solution 2 (es) we denote by bs .

'"
To find all solutions which coincide at e = es we have to in-
vestigate the manifold of solutions ....b (e) in the vicinity of
18. 2 Some transformations
Let f, be a small parameter
e = es + t. (E)
and
(F)
we have to find ~ ( t) . Putting (E) and (F) into (B) we ob-

tain
1: (Es+~ leg + t)= 0,

Transformations 203
which we abbreviate by
! s (~ , E,) = 0 , (G)
! 5 (0 0) I = 0 (H)
Let us assume that I (es) = Is has the rank l - 1 . Let p

....,
and q...., denote the right hand and the left hand eigenvector
of ! s, respectively,
}
Is P
"'"
=0 J
(1)
Since the rank of Es is assumed to be L - 1 , it is possible

to no rrnalize ~ and p ,
'"
(J)
We define a new matrix H

...., by
(K)
H ... -Fs+p·qT
'" '" )
- '"
,
where f ; 9. T denotes the dyadic product of the vectors p
'"
and q T
The matrix H
rv
is nonsingular. We have
and
} (L)
We split off those terms of !s (~ ) c) which are linear
with respect to j):

'"
(M)
Because of (H) we have
~ (0,0) =0
Let us assume that b (~ , t )is sufficiently smooth to permit

the following expansions. We add t' 11. on both sides of (G)
and obtain
~ & = tf & + ! s (~ ) t) .
Applying (M) and (K) we get
and because of (K) we have
(N)
Bifurcation Equation 205
This is just another form of the equation (G) which is an equa-
tion for ~ (c.). But from (N) we shall be able to derive a bi-
furcation equation whi ch yields the information about the mani
fold of the solutions in the vicinity of e = e 5.
18. 3 Bifurcation equation
We introduce an (artificial) parameter )C ,
into (M):
(P)
(p) is an implicit equation for 12 (X It) . For small values

)C ,t the vector 12 ('X , t.) can be obtained from (P) by
the method of successive approximations:
(Q)
Let f.J" ()C , t)

N
be an (approximate) solution of (P), e. g. ,
given in the form of a series expansion with respect to X and
t Putting p. into (p) and multiplying that equation by 9. T
we obtain, since)C has to satisfy the relation (0),
(R)
or
B(x,t) == fb(~*(",t),t.) = O.
We have to solve B ()(, t.) = 0 fo r ?C (t) Substi-

tuting x (t) into fi... *(x t)
J we get
~Ct) = ~·(')(t) ,t) . (5)
The equation
B(x,t) =0 (T)
has certainly the solution ,,= 0 I t = 0 But, in general,

the solutions xC£.) of (T) are not unique in the neighborhood
of (0.0), ~(f.) can have multiple (real!!) branches to each of
which belongs a ~(t) = ~·(x (£.) , x) Thus, the solu-

tions of B (,)(, t) = a yield the required information a-
bout the bifurcations, (T) is t('rnwd bifurcation ('quation.
18. 4 Numerical procedures

18. 41 Computation of e.s
1n general, it is cliffi cult to compute the val
\Ie e -es forwhichdct fee) = 0 sin<"eNl~wton'smethod

(to find g (e) , d. 15.42) does not work for e~es . Thus, the
Num.erical Procedure 207
simplified Newton's method or a generalized form of the regu-
la falsi must be applied to get E (e) . If ~ (e) is known)

e 6 can be computed from det ! (e s ) = 0 rather easily.
In special cases no serious difficulties arise. Let
a system have odd harmonic solutions, say, within a certain

parameter range (odd harmonic: the Fourier series contains
only terms w, 3 00 J 5 fa) , . . . ). Then .eo (e) can be calcu
lated, det 10 (e) =I- 0 ,say. (The subscript 0 indicates
that the vector go contains only elements which belong to odd
harmonic Fourier terms. )

Let us assume that we are interested in a singu-
lar point e = es where a non-odd harmonic solution branches
off from the known solution. We introduce a more general vec-
tor 2 (i. which contains the elements of the odd harmonic vec
tor !2 0 but additionally elements for even harmonic Fourier
terms l2 E '
Along the original solution B Go (e) is known
QG (e) = (Qo~e)), 0 -zero vector.

Thus, e = e 5 can be computed from d..el!G (e 5 ) = O.
~8. 42 Computation of .e, ~ and ~
e and 9. have to be calculated from (I)
and to be normalized corresponding to (J). If Is is a symmet-
ric matrix then f = ~ . Numerical difficulties may arise
since es is not known accurately and det ! (es) will differ sligh~
ly from zero. Such a trouble can be avoided in the following
way Sil1ce det I is computed by means of a Gauss algorithm
F will be transformed into a triangular form in any case.
Thus for e = es the (single) term in the last line of the trian-
gular form has to vanish. Taking that into account £ and <i can
be computed from the triangular form of I (es) quite accu-
rately.
If rand <J are known, ~ can be corn put

ed from (K).
18. 43 Generation of the bifurcation equation
To obtain the bifu rcation equation (T) we
have to cal etIlate 12 *' ()( , t) from equation (Q). Fi rst )we have
to expand the operator D from section 15.4, d. equation (A)

Numerical Procedures 209
in section IS. H into a series with respect to u. and e (e. g. ,

e - w 2. ). Substituting a few terms of that expansion into
{Q} , d. equation (A), it is possible to obtain numerically a few
terms
The ~ ~ \( represent numerical vectors. Putting (U) into (R)

we obtain
(v)
where the A pc, are real numbers.
IS. 44 Solution of the bifurcation eguation
Equation (V) represents a truncated series for

To find the various real solutions 'X (t.) of
the algebrai c relation (V) we can apply a graphical procedure.
"Newton's polygon", outlined, e. g., in A. R. Forsyth, Theory
of functions of a complex variable, 3. ed., Cambridge Univer-
sity Press, Cambridge 1915. On the other hand it is possible
to choose numerical values t, , say, and to solve the resulting
algebraic equation on a digital computer. (Remark: If "01 ¢ 0,
(V) has a unique solution c ('X) , some of the b j. (e) have a
vertical tangent at e - es , d. Fig. 18. 1, there exist no bi-

furcations).
Afte r the various ox (l.) are known, the
corresponding j.) (to) can be calculated from (U), we obtain
I'V
"line elements" of the branching off solutions at the singular

point which may be used as starting approximations for the nu
merical procedure outlined in section 15.4.
211
Contents
Preface 3
o. Introduction 5
1. Continuous systems 9
1. 1 Coo rrunates 9
1. 2 State of the body 10
1. 3 Problem 11
2. Some classical, conservative, linear systems 12

2. 1 Longitudinal vibrations of a rod 12
2. 2 Further conservative, linear systems 15
2.21 String 15
2. 22 Euler-Bernoulli beam 16
2.23 Timoshenko beam 17
2. 24 Plate, transverse vibrations 18
2.25 Isotropic continuous body 20
3. Wave solutions 22
3. 1 Longitudinal waves in an uniform rod 22
3. II Travelling waves in an infinite rod 22
3. 12 Waves in a finite rod 23
3.2 Waves in an infinite, uniform Timoshenko

beam 25
21~ Contents
3.3 Waves in a three-dimensional continuum 26
4. Other forms of the equations of motion

for linear, finite systems 28
4. I Variational problem 28
4. 2 Lagrangian equations 29
4. 3 Integral equation 30
5. Separation of variables; boundary value problem;
eigenvalue problem 32
5. I Separation of the variables 32
5.2 Boundary value problem-eigenvalue problem 33
5. 3 Example: Uniform rod 35
5.4 Adjoint and self-adjoint eigenvalue problem 37
5.41 Inner product; orthogonality, norm 37
5. 42 Adjoint operator, adjoint boundary values,
adjoint and self-adjoint eigenvalue problem 38
5.43 Eigenvalues of self-adjoint eigenvalue
problems are real 40
5. 44 Eigenfunctions belonging to different
eigenvalues of self-adjoint eigenvalue
problems are orthogonal 40
5.45 Normalized eigenfunctions 41
5.5 Initial value problem 41
Contents 213
6. Forced vibrations 44
6. 1 Excitation at one end (of a rod) 44
6.2 Distributed excitation 46
6.21 Normal coordinates 46
6.22 Resonance 48
7. Methods to calculate eigenvalues

of conservative systems 49
7.1 Analytical method 49
7.2 Rayleigh's quotient 49
7.21 Formal introduction 49
7. 22 Introduction of Rayleigh's quotient
by energy considerations 51
7. 23 Rayleigh-Ritz method 53
7. 3 Iteration process 54
7.4 Variational methods 56
7.41 Ritz's method 56
7.42 Galerkin's method 58
7.43 Differences between Ritz's and
Galerkin's methods 60
7.5 Transfer m"trices (Example:Euler-Bernou1li

beam) 61
7. 51 Basic idea 61
214 Contents
7. 52 Ba.sic equations 61
7. 53 Boundary conditions 63
7. 54 Characteristic equation 64
8. Relations between stability and vibrations
of conservative systems (Example: Compressed

bar) 66
8. 1 Equations of motion; boundary conditions 66
8.2 Eigenfunctions and natural freqooncies 67

8. 3 Discussion of the results 67
9. Equations of motion for (nonlinear)
nonconservative systems 69
9.1 Differential equations; boundary conditions 69
9.2 Principle of virtual work 69
9.3 The general form of Hamilton's principle 72
10. Damping {linear} 77

10. 1 Linear external damping 77
10. 11 Special separable case 78
10.12 The general case 79
10. 13 Investigation of a damped system by

Lyapunov's direct method 82
10. 14 Forced vibrations 84
10. 2 Linear viscoelastic internal damping 84
Contents 215
10. 21 Distribution of eigenvalues 86

10. 22 Self-excited vibrations of a rotating
shaft caused by internal damping 89
10.3 "Complex modulus" damping 90
11. Some remarks on the separability of damped

systems and non-self-adjoint eigenvalue problems 92
11. 1 Transformation of a "damped equation of
motion" into a separable form 92
11. 2 Non-self-adjoint eigenvalue problems 93
11. 21 Orthogonality of eigenfunctions 94
12. Lyapunov's definition of stability 96

12. 1 Lyapunov's (weak) stability 98
12.2 Lyapunov's asymptotic stability 98
12.3 Lyapunov's unstability 98
12. 4 Remarks 99
12. 5 Stability investigation by means of the
(linear) variational equation 99
13. Kinetic stability of autonomous systems 102
13. 1 Follower forces 105
13. 11 Static investigation 106
13. 12 Kinetic investigation 107
1 3. 13 Influence of damping 108
13. 14 Movements of the eigenvalues 109

216 Contents
13. 2 Pipes conveying fluid 110

13. 21 Cases for which the static stability
investigation leads to the proper results III
1 3. 3 Gale rkin I s method 11 3
14. Kinetic stability of non-autonomous systems 117

14. 1 A simple example (compres sed bar) 117
14. 11 Some results on Mathieu equations 118
14. 12 Stability discussion for the
given problem 119
14.2 A non-separable problem(compressed bar} 122
14. 21 Qualitative discussion of some results 124
14.22 Two remarks 126
14. 3 A Lyapunov-type approach 126
14.4 Concluding remarks 134
15. Nonlinear vibrations of autonomous systems 135

15. 1 Nonlinear torsional vibrations of a cylinder 135
15. 11 Solution of the linear problem 136
15. 12 Reduction to a system of ordinary
differential equations 137
15. 2 First order approximation 138
15. 21 One term Fourier approximation 139
15. 22 Series expansion, perturbation technique 140
Contents 217
15. 3 Higher order approximations 145

15.31 Periodic solutions;perturbation technique 145
15. 32 Non-periodic solutions; some remarks 148
15.4 Numerical computation of periodic solutions 150
15.41 Description of the method 150
15.42 Solution of the equations 154
15.43 Concluding remarks 156
15. 5 Damped nonlinear vibrations 157
15. 6 Stability 158
16. Nonlinear vibrations of nonautonomous systems 160

16. 1 Nonlinear transversal vibrations of a rock 160
16.11 Reduction to a system of ordinary
differential equations 161
16.2 First order approximation 162
16. 21 One term Fourier approximation 163
16. 22 Series expansion; perturbation
techniques for periodic equations 165
16.221 Non-resonance case 165
16. 222 Resonance case 166

16. 223 Superharmonic resonance 169
16. 224 Subharmonic solutions 170
16. 23 Some periodic solutions obtained

numerically fo r Duffing I s equation 172
218 Contents
16. 24 A1mo st periodic equations 174

16. 241 Almost periodic functions 175
16. 242 Breakdown of the perturbation
technique for almost periodic

differential equations 1,7
16. 243 Solution of a. p. differential
equations by the method of

successive approximations 179
16. 3 Higher order approximations 181

16.4 Numerical computation of special solutions 181
16. 5 Stability 182
17. Asymptotic methods 183

17. 1 The method of the slowly varying amplitudes 183
17. 11 Equation of motion 183
17.12 Van del' Pol's method 185
17. 13 Stationary solutions 188
17.14 Phase curves A (B) 190
17.2 Geller<llizations of van der Pol's method 193
17. 3 The method of averaging 194

17. 31 Standard fo nn 194
17. 12 Some formal transformations 196
17.31 Averaged cquation;first approximation 196
17. 34 Order of the first approximation 198
Contents 219
18. Investigation of bifurcation points 200
18. 1 Lo cation of bifurcation points 201

18.2 Some transformations 202
18. 3 Bifurcation equation 205

18.4 Numerical procedures 206
Contents 211

Vibrations of Continuous Systems - Theory and Applications PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Vibrations of Continuous Systems - Theory and Applications PDF

Uploaded by

Copyright:

Available Formats

INTERNATIONAL CENTRE FOR MECHANICAL SCIENCES

COURSE HELD AT THE DEPARTMENT

SEPTEMBER - OCTOBER 1969

This booklet contains the notes of my lec-

Udine, October 1969.

Objectives of vibrational investigations are tech-

nical (physical, chemical), biological, economic, etc. systems.

Purposes of such investigations:

to "comprehend" phenomena observed (experimentally) in ac-

to "predict" the behavior, qualitatively as well as quantitative-

ly, of systems not yet (experimentally) tested,and of systems

which are only projected as in engineering design.

The procedures of these investigations are al-

ways similar, see Fig. O. 1.

There is no way to compare mathematically the

In these lectures we shall restrict ourselves to

Multiple interaction with

Boundaries of the system only

(equations of motion) §"

mechanical system. We shall be concerned with the steps CD

Lit A. G. J. Macfarlane, Engineering systems analysis,

G. G. Harrap & Co, London 1964 (German trans!. BI-

R. H. Cannon, jr., Dynamics of physical systems,

The particle P is denoted by the Lagrangian coor-

which may be interpreted as its{curvilinear}coordinates in a

1. 12 Position (spatial, Eulerian) coordinates

The position of the particle P at the time t

need not coincide.

I. 2 State of the body

The state is completely known far t ~ to if

! = ~(~. t) is known for all P E V and t ~ to

1.22 Modern extensions

1. 221 Thermodynamical state variables are taken in-

als, complex molecules) are taken into account by associating

"directions" etc. with the particles ~ multipolar media.

From a formal point of view both extensions mean

that the number of the state variables xi (K.t) is increased,

i = 1, ••• , N j N> 3 ; (x (C. t) , ... , xN (~. t))=~ (~ . t)

Lit. : C. Truesdell, The elements of continuum mechanics,

Springer, Berlin 1966

Find ! ". 1 (!, • f) and, maybe, some quantities

2. Some classical, conservative, linear systems

Rod originally straight, reference configuration as shown

Problem: Investigate the longitudinal vibrations.

Simplifying as sumption : All originally plane cros s - sect ional

areas x = const. remain plane and perpendicular to the

Displacement lA. (~.t) = u(X,t) = X{X.t) -X.

2. 122 Constitutive equation

cr- stress, E .. E (X) - modulus of elasticity

F(X.t) F(X+dX,t) ~p.u..=F'

p. :: (l A - mass density per unit length

2.125 Equation of motion

Bounda ry conditions for X= 0 End:

2.1311 F u.(O.t)::O fixed

2.1312 c:=:::J F (O.t):: 0 t-..U: (O.t)-=O free

mu,(O,t)= F(O.t)' m-ma~s

Similar conditions hold for X = z.

2. 132 Nonhomogeneous boundary condition

2. 1322 ,,(t,t) = f (t), f(t} given displace-

2. 2 further conservative. linear systems

)J. (X) - mass per unit length

2. 211 Equation of motion